xref: /freebsd/sys/netinet6/mld6.c (revision 2b8331622f0b212cf3bb4fc4914a501e5321d506)
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/sysctl.h>
79 #include <sys/kernel.h>
80 #include <sys/callout.h>
81 #include <sys/malloc.h>
82 #include <sys/module.h>
83 #include <sys/ktr.h>
84 
85 #include <net/if.h>
86 #include <net/if_var.h>
87 #include <net/if_private.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 void	mli_delete_locked(struct ifnet *);
108 static void	mld_dispatch_packet(struct mbuf *);
109 static void	mld_dispatch_queue(struct mbufq *, int);
110 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
111 static void	mld_fasttimo_vnet(struct in6_multi_head *inmh);
112 static int	mld_handle_state_change(struct in6_multi *,
113 		    struct mld_ifsoftc *);
114 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
115 		    const int);
116 #ifdef KTR
117 static char *	mld_rec_type_to_str(const int);
118 #endif
119 static void	mld_set_version(struct mld_ifsoftc *, const int);
120 static void	mld_slowtimo_vnet(void);
121 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
122 		    /*const*/ struct mld_hdr *);
123 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static void	mld_v1_process_group_timer(struct in6_multi_head *,
126 		    struct in6_multi *);
127 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
128 static int	mld_v1_transmit_report(struct in6_multi *, const int);
129 static void	mld_v1_update_group(struct in6_multi *, const int);
130 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
131 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
132 static struct mbuf *
133 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
134 static int	mld_v2_enqueue_filter_change(struct mbufq *,
135 		    struct in6_multi *);
136 static int	mld_v2_enqueue_group_record(struct mbufq *,
137 		    struct in6_multi *, const int, const int, const int,
138 		    const int);
139 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
140 		    struct mbuf *, struct mldv2_query *, const int, const int);
141 static int	mld_v2_merge_state_changes(struct in6_multi *,
142 		    struct mbufq *);
143 static void	mld_v2_process_group_timers(struct in6_multi_head *,
144 		    struct mbufq *, struct mbufq *,
145 		    struct in6_multi *, const int);
146 static int	mld_v2_process_group_query(struct in6_multi *,
147 		    struct mld_ifsoftc *mli, int, struct mbuf *,
148 		    struct mldv2_query *, const int);
149 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
150 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
151 
152 /*
153  * Normative references: RFC 2710, RFC 3590, RFC 3810.
154  *
155  * Locking:
156  *  * The MLD subsystem lock ends up being system-wide for the moment,
157  *    but could be per-VIMAGE later on.
158  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
159  *    Any may be taken independently; if any are held at the same
160  *    time, the above lock order must be followed.
161  *  * IN6_MULTI_LOCK covers in_multi.
162  *  * MLD_LOCK covers per-link state and any global variables in this file.
163  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
164  *    per-link state iterators.
165  *
166  *  XXX LOR PREVENTION
167  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
168  *  will not accept an ifp; it wants an embedded scope ID, unlike
169  *  ip_output(), which happily takes the ifp given to it. The embedded
170  *  scope ID is only used by MLD to select the outgoing interface.
171  *
172  *  During interface attach and detach, MLD will take MLD_LOCK *after*
173  *  the IF_AFDATA_LOCK.
174  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
175  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
176  *  dispatch could work around this, but we'd rather not do that, as it
177  *  can introduce other races.
178  *
179  *  As such, we exploit the fact that the scope ID is just the interface
180  *  index, and embed it in the IPv6 destination address accordingly.
181  *  This is potentially NOT VALID for MLDv1 reports, as they
182  *  are always sent to the multicast group itself; as MLDv2
183  *  reports are always sent to ff02::16, this is not an issue
184  *  when MLDv2 is in use.
185  *
186  *  This does not however eliminate the LOR when ip6_output() itself
187  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
188  *  trigger a LOR warning in WITNESS when the ifnet is detached.
189  *
190  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
191  *  how it's used across the network stack. Here we're simply exploiting
192  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
193  *
194  * VIMAGE:
195  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
196  *    to a vnet in ifp->if_vnet.
197  */
198 static struct mtx		 mld_mtx;
199 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
200 
201 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
202 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
203 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
204 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
205 
206 /*
207  * VIMAGE-wide globals.
208  */
209 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
210 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
211 VNET_DEFINE_STATIC(int, interface_timers_running6);
212 VNET_DEFINE_STATIC(int, state_change_timers_running6);
213 VNET_DEFINE_STATIC(int, current_state_timers_running6);
214 
215 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
216 #define	V_mli_head			VNET(mli_head)
217 #define	V_interface_timers_running6	VNET(interface_timers_running6)
218 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
219 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
220 
221 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
222 
223 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
224     "IPv6 Multicast Listener Discovery");
225 
226 /*
227  * Virtualized sysctls.
228  */
229 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
230     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
231     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
232     "Rate limit for MLDv2 Group-and-Source queries in seconds");
233 
234 /*
235  * Non-virtualized sysctls.
236  */
237 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
238     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
239     "Per-interface MLDv2 state");
240 
241 static int	mld_v1enable = 1;
242 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
243     &mld_v1enable, 0, "Enable fallback to MLDv1");
244 
245 static int	mld_v2enable = 1;
246 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
247     &mld_v2enable, 0, "Enable MLDv2");
248 
249 static int	mld_use_allow = 1;
250 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
251     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
252 
253 /*
254  * Packed Router Alert option structure declaration.
255  */
256 struct mld_raopt {
257 	struct ip6_hbh		hbh;
258 	struct ip6_opt		pad;
259 	struct ip6_opt_router	ra;
260 } __packed;
261 
262 /*
263  * Router Alert hop-by-hop option header.
264  */
265 static struct mld_raopt mld_ra = {
266 	.hbh = { 0, 0 },
267 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
268 	.ra = {
269 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
270 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
271 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
272 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
273 	}
274 };
275 static struct ip6_pktopts mld_po;
276 
277 static __inline void
278 mld_save_context(struct mbuf *m, struct ifnet *ifp)
279 {
280 
281 #ifdef VIMAGE
282 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
283 #endif /* VIMAGE */
284 	m->m_pkthdr.rcvif = ifp;
285 	m->m_pkthdr.flowid = ifp->if_index;
286 }
287 
288 static __inline void
289 mld_scrub_context(struct mbuf *m)
290 {
291 
292 	m->m_pkthdr.PH_loc.ptr = NULL;
293 	m->m_pkthdr.flowid = 0;
294 }
295 
296 /*
297  * Restore context from a queued output chain.
298  * Return saved ifindex.
299  *
300  * VIMAGE: The assertion is there to make sure that we
301  * actually called CURVNET_SET() with what's in the mbuf chain.
302  */
303 static __inline uint32_t
304 mld_restore_context(struct mbuf *m)
305 {
306 
307 #if defined(VIMAGE) && defined(INVARIANTS)
308 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
309 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
310 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
311 #endif
312 	return (m->m_pkthdr.flowid);
313 }
314 
315 /*
316  * Retrieve or set threshold between group-source queries in seconds.
317  *
318  * VIMAGE: Assume curvnet set by caller.
319  * SMPng: NOTE: Serialized by MLD lock.
320  */
321 static int
322 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
323 {
324 	int error;
325 	int i;
326 
327 	error = sysctl_wire_old_buffer(req, sizeof(int));
328 	if (error)
329 		return (error);
330 
331 	MLD_LOCK();
332 
333 	i = V_mld_gsrdelay.tv_sec;
334 
335 	error = sysctl_handle_int(oidp, &i, 0, req);
336 	if (error || !req->newptr)
337 		goto out_locked;
338 
339 	if (i < -1 || i >= 60) {
340 		error = EINVAL;
341 		goto out_locked;
342 	}
343 
344 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
345 	     V_mld_gsrdelay.tv_sec, i);
346 	V_mld_gsrdelay.tv_sec = i;
347 
348 out_locked:
349 	MLD_UNLOCK();
350 	return (error);
351 }
352 
353 /*
354  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
355  * For use by ifmcstat(8).
356  *
357  * VIMAGE: Assume curvnet set by caller. The node handler itself
358  * is not directly virtualized.
359  */
360 static int
361 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
362 {
363 	struct epoch_tracker	 et;
364 	int			*name;
365 	int			 error;
366 	u_int			 namelen;
367 	struct ifnet		*ifp;
368 	struct mld_ifsoftc	*mli;
369 
370 	name = (int *)arg1;
371 	namelen = arg2;
372 
373 	if (req->newptr != NULL)
374 		return (EPERM);
375 
376 	if (namelen != 1)
377 		return (EINVAL);
378 
379 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
380 	if (error)
381 		return (error);
382 
383 	IN6_MULTI_LOCK();
384 	IN6_MULTI_LIST_LOCK();
385 	MLD_LOCK();
386 	NET_EPOCH_ENTER(et);
387 
388 	error = ENOENT;
389 	ifp = ifnet_byindex(name[0]);
390 	if (ifp == NULL)
391 		goto out_locked;
392 
393 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
394 		if (ifp == mli->mli_ifp) {
395 			struct mld_ifinfo info;
396 
397 			info.mli_version = mli->mli_version;
398 			info.mli_v1_timer = mli->mli_v1_timer;
399 			info.mli_v2_timer = mli->mli_v2_timer;
400 			info.mli_flags = mli->mli_flags;
401 			info.mli_rv = mli->mli_rv;
402 			info.mli_qi = mli->mli_qi;
403 			info.mli_qri = mli->mli_qri;
404 			info.mli_uri = mli->mli_uri;
405 			error = SYSCTL_OUT(req, &info, sizeof(info));
406 			break;
407 		}
408 	}
409 
410 out_locked:
411 	NET_EPOCH_EXIT(et);
412 	MLD_UNLOCK();
413 	IN6_MULTI_LIST_UNLOCK();
414 	IN6_MULTI_UNLOCK();
415 	return (error);
416 }
417 
418 /*
419  * Dispatch an entire queue of pending packet chains.
420  * VIMAGE: Assumes the vnet pointer has been set.
421  */
422 static void
423 mld_dispatch_queue(struct mbufq *mq, int limit)
424 {
425 	struct mbuf *m;
426 
427 	while ((m = mbufq_dequeue(mq)) != NULL) {
428 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
429 		mld_dispatch_packet(m);
430 		if (--limit == 0)
431 			break;
432 	}
433 }
434 
435 /*
436  * Filter outgoing MLD report state by group.
437  *
438  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
439  * and node-local addresses. However, kernel and socket consumers
440  * always embed the KAME scope ID in the address provided, so strip it
441  * when performing comparison.
442  * Note: This is not the same as the *multicast* scope.
443  *
444  * Return zero if the given group is one for which MLD reports
445  * should be suppressed, or non-zero if reports should be issued.
446  */
447 static __inline int
448 mld_is_addr_reported(const struct in6_addr *addr)
449 {
450 
451 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
452 
453 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
454 		return (0);
455 
456 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
457 		struct in6_addr tmp = *addr;
458 		in6_clearscope(&tmp);
459 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
460 			return (0);
461 	}
462 
463 	return (1);
464 }
465 
466 /*
467  * Attach MLD when PF_INET6 is attached to an interface.  Assumes that the
468  * current VNET is set by the caller.
469  */
470 struct mld_ifsoftc *
471 mld_domifattach(struct ifnet *ifp)
472 {
473 	struct mld_ifsoftc *mli;
474 
475 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp));
476 
477 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_WAITOK | M_ZERO);
478 	mli->mli_ifp = ifp;
479 	mli->mli_version = MLD_VERSION_2;
480 	mli->mli_flags = 0;
481 	mli->mli_rv = MLD_RV_INIT;
482 	mli->mli_qi = MLD_QI_INIT;
483 	mli->mli_qri = MLD_QRI_INIT;
484 	mli->mli_uri = MLD_URI_INIT;
485 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
486 	if ((ifp->if_flags & IFF_MULTICAST) == 0)
487 		mli->mli_flags |= MLIF_SILENT;
488 	if (mld_use_allow)
489 		mli->mli_flags |= MLIF_USEALLOW;
490 
491 	MLD_LOCK();
492 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
493 	MLD_UNLOCK();
494 
495 	return (mli);
496 }
497 
498 /*
499  * Hook for ifdetach.
500  *
501  * NOTE: Some finalization tasks need to run before the protocol domain
502  * is detached, but also before the link layer does its cleanup.
503  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
504  *
505  * SMPng: Caller must hold IN6_MULTI_LOCK().
506  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
507  * XXX This routine is also bitten by unlocked ifma_protospec access.
508  */
509 void
510 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
511 {
512 	struct epoch_tracker     et;
513 	struct mld_ifsoftc	*mli;
514 	struct ifmultiaddr	*ifma;
515 	struct in6_multi	*inm;
516 
517 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
518 	    if_name(ifp));
519 
520 	IN6_MULTI_LIST_LOCK_ASSERT();
521 	MLD_LOCK();
522 
523 	mli = MLD_IFINFO(ifp);
524 	IF_ADDR_WLOCK(ifp);
525 	/*
526 	 * Extract list of in6_multi associated with the detaching ifp
527 	 * which the PF_INET6 layer is about to release.
528 	 */
529 	NET_EPOCH_ENTER(et);
530 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
531 		inm = in6m_ifmultiaddr_get_inm(ifma);
532 		if (inm == NULL)
533 			continue;
534 		in6m_disconnect_locked(inmh, inm);
535 
536 		if (mli->mli_version == MLD_VERSION_2) {
537 			in6m_clear_recorded(inm);
538 
539 			/*
540 			 * We need to release the final reference held
541 			 * for issuing the INCLUDE {}.
542 			 */
543 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
544 				inm->in6m_state = MLD_NOT_MEMBER;
545 				in6m_rele_locked(inmh, inm);
546 			}
547 		}
548 	}
549 	NET_EPOCH_EXIT(et);
550 	IF_ADDR_WUNLOCK(ifp);
551 	MLD_UNLOCK();
552 }
553 
554 /*
555  * Hook for domifdetach.
556  * Runs after link-layer cleanup; free MLD state.
557  *
558  * SMPng: Normally called with IF_AFDATA_LOCK held.
559  */
560 void
561 mld_domifdetach(struct ifnet *ifp)
562 {
563 
564 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
565 	    __func__, ifp, if_name(ifp));
566 
567 	MLD_LOCK();
568 	mli_delete_locked(ifp);
569 	MLD_UNLOCK();
570 }
571 
572 static void
573 mli_delete_locked(struct ifnet *ifp)
574 {
575 	struct mld_ifsoftc *mli, *tmli;
576 
577 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
578 	    __func__, ifp, if_name(ifp));
579 
580 	MLD_LOCK_ASSERT();
581 
582 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
583 		if (mli->mli_ifp == ifp) {
584 			/*
585 			 * Free deferred General Query responses.
586 			 */
587 			mbufq_drain(&mli->mli_gq);
588 
589 			LIST_REMOVE(mli, mli_link);
590 
591 			free(mli, M_MLD);
592 			return;
593 		}
594 	}
595 }
596 
597 /*
598  * Process a received MLDv1 general or address-specific query.
599  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
600  *
601  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
602  * mld_addr. This is OK as we own the mbuf chain.
603  */
604 static int
605 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
606     /*const*/ struct mld_hdr *mld)
607 {
608 	struct ifmultiaddr	*ifma;
609 	struct mld_ifsoftc	*mli;
610 	struct in6_multi	*inm;
611 	int			 is_general_query;
612 	uint16_t		 timer;
613 #ifdef KTR
614 	char			 ip6tbuf[INET6_ADDRSTRLEN];
615 #endif
616 
617 	NET_EPOCH_ASSERT();
618 
619 	is_general_query = 0;
620 
621 	if (!mld_v1enable) {
622 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
623 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
624 		    ifp, if_name(ifp));
625 		return (0);
626 	}
627 
628 	/*
629 	 * RFC3810 Section 6.2: MLD queries must originate from
630 	 * a router's link-local address.
631 	 */
632 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
633 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
634 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
635 		    ifp, if_name(ifp));
636 		return (0);
637 	}
638 
639 	/*
640 	 * Do address field validation upfront before we accept
641 	 * the query.
642 	 */
643 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
644 		/*
645 		 * MLDv1 General Query.
646 		 * If this was not sent to the all-nodes group, ignore it.
647 		 */
648 		struct in6_addr		 dst;
649 
650 		dst = ip6->ip6_dst;
651 		in6_clearscope(&dst);
652 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
653 			return (EINVAL);
654 		is_general_query = 1;
655 	} else {
656 		/*
657 		 * Embed scope ID of receiving interface in MLD query for
658 		 * lookup whilst we don't hold other locks.
659 		 */
660 		in6_setscope(&mld->mld_addr, ifp, NULL);
661 	}
662 
663 	IN6_MULTI_LIST_LOCK();
664 	MLD_LOCK();
665 
666 	/*
667 	 * Switch to MLDv1 host compatibility mode.
668 	 */
669 	mli = MLD_IFINFO(ifp);
670 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
671 	mld_set_version(mli, MLD_VERSION_1);
672 
673 	timer = (ntohs(mld->mld_maxdelay) * MLD_FASTHZ) / MLD_TIMER_SCALE;
674 	if (timer == 0)
675 		timer = 1;
676 
677 	if (is_general_query) {
678 		/*
679 		 * For each reporting group joined on this
680 		 * interface, kick the report timer.
681 		 */
682 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
683 			 ifp, if_name(ifp));
684 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
685 			inm = in6m_ifmultiaddr_get_inm(ifma);
686 			if (inm == NULL)
687 				continue;
688 			mld_v1_update_group(inm, timer);
689 		}
690 	} else {
691 		/*
692 		 * MLDv1 Group-Specific Query.
693 		 * If this is a group-specific MLDv1 query, we need only
694 		 * look up the single group to process it.
695 		 */
696 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
697 		if (inm != NULL) {
698 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
699 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
700 			    ifp, if_name(ifp));
701 			mld_v1_update_group(inm, timer);
702 		}
703 		/* XXX Clear embedded scope ID as userland won't expect it. */
704 		in6_clearscope(&mld->mld_addr);
705 	}
706 
707 	MLD_UNLOCK();
708 	IN6_MULTI_LIST_UNLOCK();
709 
710 	return (0);
711 }
712 
713 /*
714  * Update the report timer on a group in response to an MLDv1 query.
715  *
716  * If we are becoming the reporting member for this group, start the timer.
717  * If we already are the reporting member for this group, and timer is
718  * below the threshold, reset it.
719  *
720  * We may be updating the group for the first time since we switched
721  * to MLDv2. If we are, then we must clear any recorded source lists,
722  * and transition to REPORTING state; the group timer is overloaded
723  * for group and group-source query responses.
724  *
725  * Unlike MLDv2, the delay per group should be jittered
726  * to avoid bursts of MLDv1 reports.
727  */
728 static void
729 mld_v1_update_group(struct in6_multi *inm, const int timer)
730 {
731 #ifdef KTR
732 	char			 ip6tbuf[INET6_ADDRSTRLEN];
733 #endif
734 
735 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
736 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
737 	    if_name(inm->in6m_ifp), timer);
738 
739 	IN6_MULTI_LIST_LOCK_ASSERT();
740 
741 	switch (inm->in6m_state) {
742 	case MLD_NOT_MEMBER:
743 	case MLD_SILENT_MEMBER:
744 		break;
745 	case MLD_REPORTING_MEMBER:
746 		if (inm->in6m_timer != 0 &&
747 		    inm->in6m_timer <= timer) {
748 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
749 			    "skipping.", __func__);
750 			break;
751 		}
752 		/* FALLTHROUGH */
753 	case MLD_SG_QUERY_PENDING_MEMBER:
754 	case MLD_G_QUERY_PENDING_MEMBER:
755 	case MLD_IDLE_MEMBER:
756 	case MLD_LAZY_MEMBER:
757 	case MLD_AWAKENING_MEMBER:
758 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
759 		inm->in6m_state = MLD_REPORTING_MEMBER;
760 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
761 		V_current_state_timers_running6 = 1;
762 		break;
763 	case MLD_SLEEPING_MEMBER:
764 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
765 		inm->in6m_state = MLD_AWAKENING_MEMBER;
766 		break;
767 	case MLD_LEAVING_MEMBER:
768 		break;
769 	}
770 }
771 
772 /*
773  * Process a received MLDv2 general, group-specific or
774  * group-and-source-specific query.
775  *
776  * Assumes that mld points to a struct mldv2_query which is stored in
777  * contiguous memory.
778  *
779  * Return 0 if successful, otherwise an appropriate error code is returned.
780  */
781 static int
782 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
783     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
784 {
785 	struct mld_ifsoftc	*mli;
786 	struct in6_multi	*inm;
787 	uint32_t		 maxdelay, nsrc, qqi;
788 	int			 is_general_query;
789 	uint16_t		 timer;
790 	uint8_t			 qrv;
791 #ifdef KTR
792 	char			 ip6tbuf[INET6_ADDRSTRLEN];
793 #endif
794 
795 	NET_EPOCH_ASSERT();
796 
797 	if (!mld_v2enable) {
798 		CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
799 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
800 		    ifp, if_name(ifp));
801 		return (0);
802 	}
803 
804 	/*
805 	 * RFC3810 Section 6.2: MLD queries must originate from
806 	 * a router's link-local address.
807 	 */
808 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
809 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
810 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
811 		    ifp, if_name(ifp));
812 		return (0);
813 	}
814 
815 	is_general_query = 0;
816 
817 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
818 
819 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
820 	if (maxdelay >= 32768) {
821 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
822 			   (MLD_MRC_EXP(maxdelay) + 3);
823 	}
824 	timer = (maxdelay * MLD_FASTHZ) / MLD_TIMER_SCALE;
825 	if (timer == 0)
826 		timer = 1;
827 
828 	qrv = MLD_QRV(mld->mld_misc);
829 	if (qrv < 2) {
830 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
831 		    qrv, MLD_RV_INIT);
832 		qrv = MLD_RV_INIT;
833 	}
834 
835 	qqi = mld->mld_qqi;
836 	if (qqi >= 128) {
837 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
838 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
839 	}
840 
841 	nsrc = ntohs(mld->mld_numsrc);
842 	if (nsrc > MLD_MAX_GS_SOURCES)
843 		return (EMSGSIZE);
844 	if (icmp6len < sizeof(struct mldv2_query) +
845 	    (nsrc * sizeof(struct in6_addr)))
846 		return (EMSGSIZE);
847 
848 	/*
849 	 * Do further input validation upfront to avoid resetting timers
850 	 * should we need to discard this query.
851 	 */
852 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
853 		/*
854 		 * A general query with a source list has undefined
855 		 * behaviour; discard it.
856 		 */
857 		if (nsrc > 0)
858 			return (EINVAL);
859 		is_general_query = 1;
860 	} else {
861 		/*
862 		 * Embed scope ID of receiving interface in MLD query for
863 		 * lookup whilst we don't hold other locks (due to KAME
864 		 * locking lameness). We own this mbuf chain just now.
865 		 */
866 		in6_setscope(&mld->mld_addr, ifp, NULL);
867 	}
868 
869 	IN6_MULTI_LIST_LOCK();
870 	MLD_LOCK();
871 
872 	mli = MLD_IFINFO(ifp);
873 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
874 
875 	/*
876 	 * Discard the v2 query if we're in Compatibility Mode.
877 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
878 	 * until the Old Version Querier Present timer expires.
879 	 */
880 	if (mli->mli_version != MLD_VERSION_2)
881 		goto out_locked;
882 
883 	mld_set_version(mli, MLD_VERSION_2);
884 	mli->mli_rv = qrv;
885 	mli->mli_qi = qqi;
886 	mli->mli_qri = maxdelay;
887 
888 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
889 	    maxdelay);
890 
891 	if (is_general_query) {
892 		/*
893 		 * MLDv2 General Query.
894 		 *
895 		 * Schedule a current-state report on this ifp for
896 		 * all groups, possibly containing source lists.
897 		 *
898 		 * If there is a pending General Query response
899 		 * scheduled earlier than the selected delay, do
900 		 * not schedule any other reports.
901 		 * Otherwise, reset the interface timer.
902 		 */
903 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
904 		    ifp, if_name(ifp));
905 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
906 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
907 			V_interface_timers_running6 = 1;
908 		}
909 	} else {
910 		/*
911 		 * MLDv2 Group-specific or Group-and-source-specific Query.
912 		 *
913 		 * Group-source-specific queries are throttled on
914 		 * a per-group basis to defeat denial-of-service attempts.
915 		 * Queries for groups we are not a member of on this
916 		 * link are simply ignored.
917 		 */
918 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
919 		if (inm == NULL)
920 			goto out_locked;
921 		if (nsrc > 0) {
922 			if (!ratecheck(&inm->in6m_lastgsrtv,
923 			    &V_mld_gsrdelay)) {
924 				CTR1(KTR_MLD, "%s: GS query throttled.",
925 				    __func__);
926 				goto out_locked;
927 			}
928 		}
929 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
930 		     ifp, if_name(ifp));
931 		/*
932 		 * If there is a pending General Query response
933 		 * scheduled sooner than the selected delay, no
934 		 * further report need be scheduled.
935 		 * Otherwise, prepare to respond to the
936 		 * group-specific or group-and-source query.
937 		 */
938 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
939 			mld_v2_process_group_query(inm, mli, timer, m, mld, off);
940 
941 		/* XXX Clear embedded scope ID as userland won't expect it. */
942 		in6_clearscope(&mld->mld_addr);
943 	}
944 
945 out_locked:
946 	MLD_UNLOCK();
947 	IN6_MULTI_LIST_UNLOCK();
948 
949 	return (0);
950 }
951 
952 /*
953  * Process a received MLDv2 group-specific or group-and-source-specific
954  * query.
955  * Return <0 if any error occurred. Currently this is ignored.
956  */
957 static int
958 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
959     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
960 {
961 	int			 retval;
962 	uint16_t		 nsrc;
963 
964 	IN6_MULTI_LIST_LOCK_ASSERT();
965 	MLD_LOCK_ASSERT();
966 
967 	retval = 0;
968 
969 	switch (inm->in6m_state) {
970 	case MLD_NOT_MEMBER:
971 	case MLD_SILENT_MEMBER:
972 	case MLD_SLEEPING_MEMBER:
973 	case MLD_LAZY_MEMBER:
974 	case MLD_AWAKENING_MEMBER:
975 	case MLD_IDLE_MEMBER:
976 	case MLD_LEAVING_MEMBER:
977 		return (retval);
978 		break;
979 	case MLD_REPORTING_MEMBER:
980 	case MLD_G_QUERY_PENDING_MEMBER:
981 	case MLD_SG_QUERY_PENDING_MEMBER:
982 		break;
983 	}
984 
985 	nsrc = ntohs(mld->mld_numsrc);
986 
987 	/* Length should be checked by calling function. */
988 	KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
989 	    m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
990 	    nsrc * sizeof(struct in6_addr),
991 	    ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
992 	    m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
993 	    nsrc * sizeof(struct in6_addr), m0));
994 
995 	/*
996 	 * Deal with group-specific queries upfront.
997 	 * If any group query is already pending, purge any recorded
998 	 * source-list state if it exists, and schedule a query response
999 	 * for this group-specific query.
1000 	 */
1001 	if (nsrc == 0) {
1002 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1003 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1004 			in6m_clear_recorded(inm);
1005 			timer = min(inm->in6m_timer, timer);
1006 		}
1007 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1008 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1009 		V_current_state_timers_running6 = 1;
1010 		return (retval);
1011 	}
1012 
1013 	/*
1014 	 * Deal with the case where a group-and-source-specific query has
1015 	 * been received but a group-specific query is already pending.
1016 	 */
1017 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1018 		timer = min(inm->in6m_timer, timer);
1019 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1020 		V_current_state_timers_running6 = 1;
1021 		return (retval);
1022 	}
1023 
1024 	/*
1025 	 * Finally, deal with the case where a group-and-source-specific
1026 	 * query has been received, where a response to a previous g-s-r
1027 	 * query exists, or none exists.
1028 	 * In this case, we need to parse the source-list which the Querier
1029 	 * has provided us with and check if we have any source list filter
1030 	 * entries at T1 for these sources. If we do not, there is no need
1031 	 * schedule a report and the query may be dropped.
1032 	 * If we do, we must record them and schedule a current-state
1033 	 * report for those sources.
1034 	 */
1035 	if (inm->in6m_nsrc > 0) {
1036 		struct in6_addr		 srcaddr;
1037 		int			 i, nrecorded;
1038 		int			 soff;
1039 
1040 		soff = off + sizeof(struct mldv2_query);
1041 		nrecorded = 0;
1042 		for (i = 0; i < nsrc; i++) {
1043 			m_copydata(m0, soff, sizeof(struct in6_addr),
1044 			    (caddr_t)&srcaddr);
1045 			retval = in6m_record_source(inm, &srcaddr);
1046 			if (retval < 0)
1047 				break;
1048 			nrecorded += retval;
1049 			soff += sizeof(struct in6_addr);
1050 		}
1051 		if (nrecorded > 0) {
1052 			CTR1(KTR_MLD,
1053 			    "%s: schedule response to SG query", __func__);
1054 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1055 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1056 			V_current_state_timers_running6 = 1;
1057 		}
1058 	}
1059 
1060 	return (retval);
1061 }
1062 
1063 /*
1064  * Process a received MLDv1 host membership report.
1065  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1066  *
1067  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1068  * mld_addr. This is OK as we own the mbuf chain.
1069  */
1070 static int
1071 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1072     /*const*/ struct mld_hdr *mld)
1073 {
1074 	struct in6_addr		 src, dst;
1075 	struct in6_ifaddr	*ia;
1076 	struct in6_multi	*inm;
1077 #ifdef KTR
1078 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1079 #endif
1080 
1081 	NET_EPOCH_ASSERT();
1082 
1083 	if (!mld_v1enable) {
1084 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1085 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1086 		    ifp, if_name(ifp));
1087 		return (0);
1088 	}
1089 
1090 	if (ifp->if_flags & IFF_LOOPBACK)
1091 		return (0);
1092 
1093 	/*
1094 	 * MLDv1 reports must originate from a host's link-local address,
1095 	 * or the unspecified address (when booting).
1096 	 */
1097 	src = ip6->ip6_src;
1098 	in6_clearscope(&src);
1099 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1100 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1101 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1102 		    ifp, if_name(ifp));
1103 		return (EINVAL);
1104 	}
1105 
1106 	/*
1107 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1108 	 * group, and must be directed to the group itself.
1109 	 */
1110 	dst = ip6->ip6_dst;
1111 	in6_clearscope(&dst);
1112 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1113 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1114 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1115 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1116 		    ifp, if_name(ifp));
1117 		return (EINVAL);
1118 	}
1119 
1120 	/*
1121 	 * Make sure we don't hear our own membership report, as fast
1122 	 * leave requires knowing that we are the only member of a
1123 	 * group. Assume we used the link-local address if available,
1124 	 * otherwise look for ::.
1125 	 *
1126 	 * XXX Note that scope ID comparison is needed for the address
1127 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1128 	 * performed for the on-wire address.
1129 	 */
1130 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1131 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1132 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1133 		if (ia != NULL)
1134 			ifa_free(&ia->ia_ifa);
1135 		return (0);
1136 	}
1137 	if (ia != NULL)
1138 		ifa_free(&ia->ia_ifa);
1139 
1140 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1141 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1142 
1143 	/*
1144 	 * Embed scope ID of receiving interface in MLD query for lookup
1145 	 * whilst we don't hold other locks (due to KAME locking lameness).
1146 	 */
1147 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1148 		in6_setscope(&mld->mld_addr, ifp, NULL);
1149 
1150 	IN6_MULTI_LIST_LOCK();
1151 	MLD_LOCK();
1152 
1153 	/*
1154 	 * MLDv1 report suppression.
1155 	 * If we are a member of this group, and our membership should be
1156 	 * reported, and our group timer is pending or about to be reset,
1157 	 * stop our group timer by transitioning to the 'lazy' state.
1158 	 */
1159 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1160 	if (inm != NULL) {
1161 		struct mld_ifsoftc *mli;
1162 
1163 		mli = inm->in6m_mli;
1164 		KASSERT(mli != NULL,
1165 		    ("%s: no mli for ifp %p", __func__, ifp));
1166 
1167 		/*
1168 		 * If we are in MLDv2 host mode, do not allow the
1169 		 * other host's MLDv1 report to suppress our reports.
1170 		 */
1171 		if (mli->mli_version == MLD_VERSION_2)
1172 			goto out_locked;
1173 
1174 		inm->in6m_timer = 0;
1175 
1176 		switch (inm->in6m_state) {
1177 		case MLD_NOT_MEMBER:
1178 		case MLD_SILENT_MEMBER:
1179 		case MLD_SLEEPING_MEMBER:
1180 			break;
1181 		case MLD_REPORTING_MEMBER:
1182 		case MLD_IDLE_MEMBER:
1183 		case MLD_AWAKENING_MEMBER:
1184 			CTR3(KTR_MLD,
1185 			    "report suppressed for %s on ifp %p(%s)",
1186 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1187 			    ifp, if_name(ifp));
1188 		case MLD_LAZY_MEMBER:
1189 			inm->in6m_state = MLD_LAZY_MEMBER;
1190 			break;
1191 		case MLD_G_QUERY_PENDING_MEMBER:
1192 		case MLD_SG_QUERY_PENDING_MEMBER:
1193 		case MLD_LEAVING_MEMBER:
1194 			break;
1195 		}
1196 	}
1197 
1198 out_locked:
1199 	MLD_UNLOCK();
1200 	IN6_MULTI_LIST_UNLOCK();
1201 
1202 	/* XXX Clear embedded scope ID as userland won't expect it. */
1203 	in6_clearscope(&mld->mld_addr);
1204 
1205 	return (0);
1206 }
1207 
1208 /*
1209  * MLD input path.
1210  *
1211  * Assume query messages which fit in a single ICMPv6 message header
1212  * have been pulled up.
1213  * Assume that userland will want to see the message, even if it
1214  * otherwise fails kernel input validation; do not free it.
1215  * Pullup may however free the mbuf chain m if it fails.
1216  *
1217  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1218  */
1219 int
1220 mld_input(struct mbuf **mp, int off, int icmp6len)
1221 {
1222 	struct ifnet	*ifp;
1223 	struct ip6_hdr	*ip6;
1224 	struct mbuf	*m;
1225 	struct mld_hdr	*mld;
1226 	int		 mldlen;
1227 
1228 	m = *mp;
1229 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1230 
1231 	ifp = m->m_pkthdr.rcvif;
1232 
1233 	/* Pullup to appropriate size. */
1234 	if (m->m_len < off + sizeof(*mld)) {
1235 		m = m_pullup(m, off + sizeof(*mld));
1236 		if (m == NULL) {
1237 			ICMP6STAT_INC(icp6s_badlen);
1238 			return (IPPROTO_DONE);
1239 		}
1240 	}
1241 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1242 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1243 	    icmp6len >= sizeof(struct mldv2_query)) {
1244 		mldlen = sizeof(struct mldv2_query);
1245 	} else {
1246 		mldlen = sizeof(struct mld_hdr);
1247 	}
1248 	if (m->m_len < off + mldlen) {
1249 		m = m_pullup(m, off + mldlen);
1250 		if (m == NULL) {
1251 			ICMP6STAT_INC(icp6s_badlen);
1252 			return (IPPROTO_DONE);
1253 		}
1254 	}
1255 	*mp = m;
1256 	ip6 = mtod(m, struct ip6_hdr *);
1257 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1258 
1259 	/*
1260 	 * Userland needs to see all of this traffic for implementing
1261 	 * the endpoint discovery portion of multicast routing.
1262 	 */
1263 	switch (mld->mld_type) {
1264 	case MLD_LISTENER_QUERY:
1265 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1266 		if (icmp6len == sizeof(struct mld_hdr)) {
1267 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1268 				return (0);
1269 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1270 			if (mld_v2_input_query(ifp, ip6, m,
1271 			    (struct mldv2_query *)mld, off, icmp6len) != 0)
1272 				return (0);
1273 		}
1274 		break;
1275 	case MLD_LISTENER_REPORT:
1276 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1277 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1278 			return (0);
1279 		break;
1280 	case MLDV2_LISTENER_REPORT:
1281 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1282 		break;
1283 	case MLD_LISTENER_DONE:
1284 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1285 		break;
1286 	default:
1287 		break;
1288 	}
1289 
1290 	return (0);
1291 }
1292 
1293 /*
1294  * Fast timeout handler (global).
1295  * VIMAGE: Timeout handlers are expected to service all vimages.
1296  */
1297 static struct callout mldfast_callout;
1298 static void
1299 mld_fasttimo(void *arg __unused)
1300 {
1301 	struct epoch_tracker et;
1302 	struct in6_multi_head inmh;
1303 	VNET_ITERATOR_DECL(vnet_iter);
1304 
1305 	SLIST_INIT(&inmh);
1306 
1307 	NET_EPOCH_ENTER(et);
1308 	VNET_LIST_RLOCK_NOSLEEP();
1309 	VNET_FOREACH(vnet_iter) {
1310 		CURVNET_SET(vnet_iter);
1311 		mld_fasttimo_vnet(&inmh);
1312 		CURVNET_RESTORE();
1313 	}
1314 	VNET_LIST_RUNLOCK_NOSLEEP();
1315 	NET_EPOCH_EXIT(et);
1316 	in6m_release_list_deferred(&inmh);
1317 
1318 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
1319 }
1320 
1321 /*
1322  * Fast timeout handler (per-vnet).
1323  *
1324  * VIMAGE: Assume caller has set up our curvnet.
1325  */
1326 static void
1327 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1328 {
1329 	struct mbufq		 scq;	/* State-change packets */
1330 	struct mbufq		 qrq;	/* Query response packets */
1331 	struct ifnet		*ifp;
1332 	struct mld_ifsoftc	*mli;
1333 	struct ifmultiaddr	*ifma;
1334 	struct in6_multi	*inm;
1335 	int			 uri_fasthz;
1336 
1337 	uri_fasthz = 0;
1338 
1339 	/*
1340 	 * Quick check to see if any work needs to be done, in order to
1341 	 * minimize the overhead of fasttimo processing.
1342 	 * SMPng: XXX Unlocked reads.
1343 	 */
1344 	if (!V_current_state_timers_running6 &&
1345 	    !V_interface_timers_running6 &&
1346 	    !V_state_change_timers_running6)
1347 		return;
1348 
1349 	IN6_MULTI_LIST_LOCK();
1350 	MLD_LOCK();
1351 
1352 	/*
1353 	 * MLDv2 General Query response timer processing.
1354 	 */
1355 	if (V_interface_timers_running6) {
1356 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1357 
1358 		V_interface_timers_running6 = 0;
1359 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1360 			if (mli->mli_v2_timer == 0) {
1361 				/* Do nothing. */
1362 			} else if (--mli->mli_v2_timer == 0) {
1363 				mld_v2_dispatch_general_query(mli);
1364 			} else {
1365 				V_interface_timers_running6 = 1;
1366 			}
1367 		}
1368 	}
1369 
1370 	if (!V_current_state_timers_running6 &&
1371 	    !V_state_change_timers_running6)
1372 		goto out_locked;
1373 
1374 	V_current_state_timers_running6 = 0;
1375 	V_state_change_timers_running6 = 0;
1376 
1377 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1378 
1379 	/*
1380 	 * MLD host report and state-change timer processing.
1381 	 * Note: Processing a v2 group timer may remove a node.
1382 	 */
1383 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1384 		ifp = mli->mli_ifp;
1385 
1386 		if (mli->mli_version == MLD_VERSION_2) {
1387 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1388 			    MLD_FASTHZ);
1389 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1390 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1391 		}
1392 
1393 		IF_ADDR_WLOCK(ifp);
1394 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1395 			inm = in6m_ifmultiaddr_get_inm(ifma);
1396 			if (inm == NULL)
1397 				continue;
1398 			switch (mli->mli_version) {
1399 			case MLD_VERSION_1:
1400 				mld_v1_process_group_timer(inmh, inm);
1401 				break;
1402 			case MLD_VERSION_2:
1403 				mld_v2_process_group_timers(inmh, &qrq,
1404 				    &scq, inm, uri_fasthz);
1405 				break;
1406 			}
1407 		}
1408 		IF_ADDR_WUNLOCK(ifp);
1409 
1410 		switch (mli->mli_version) {
1411 		case MLD_VERSION_1:
1412 			/*
1413 			 * Transmit reports for this lifecycle.  This
1414 			 * is done while not holding IF_ADDR_LOCK
1415 			 * since this can call
1416 			 * in6ifa_ifpforlinklocal() which locks
1417 			 * IF_ADDR_LOCK internally as well as
1418 			 * ip6_output() to transmit a packet.
1419 			 */
1420 			while ((inm = SLIST_FIRST(inmh)) != NULL) {
1421 				SLIST_REMOVE_HEAD(inmh, in6m_defer);
1422 				(void)mld_v1_transmit_report(inm,
1423 				    MLD_LISTENER_REPORT);
1424 			}
1425 			break;
1426 		case MLD_VERSION_2:
1427 			mld_dispatch_queue(&qrq, 0);
1428 			mld_dispatch_queue(&scq, 0);
1429 			break;
1430 		}
1431 	}
1432 
1433 out_locked:
1434 	MLD_UNLOCK();
1435 	IN6_MULTI_LIST_UNLOCK();
1436 }
1437 
1438 /*
1439  * Update host report group timer.
1440  * Will update the global pending timer flags.
1441  */
1442 static void
1443 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1444 {
1445 	int report_timer_expired;
1446 
1447 	IN6_MULTI_LIST_LOCK_ASSERT();
1448 	MLD_LOCK_ASSERT();
1449 
1450 	if (inm->in6m_timer == 0) {
1451 		report_timer_expired = 0;
1452 	} else if (--inm->in6m_timer == 0) {
1453 		report_timer_expired = 1;
1454 	} else {
1455 		V_current_state_timers_running6 = 1;
1456 		return;
1457 	}
1458 
1459 	switch (inm->in6m_state) {
1460 	case MLD_NOT_MEMBER:
1461 	case MLD_SILENT_MEMBER:
1462 	case MLD_IDLE_MEMBER:
1463 	case MLD_LAZY_MEMBER:
1464 	case MLD_SLEEPING_MEMBER:
1465 	case MLD_AWAKENING_MEMBER:
1466 		break;
1467 	case MLD_REPORTING_MEMBER:
1468 		if (report_timer_expired) {
1469 			inm->in6m_state = MLD_IDLE_MEMBER;
1470 			SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1471 		}
1472 		break;
1473 	case MLD_G_QUERY_PENDING_MEMBER:
1474 	case MLD_SG_QUERY_PENDING_MEMBER:
1475 	case MLD_LEAVING_MEMBER:
1476 		break;
1477 	}
1478 }
1479 
1480 /*
1481  * Update a group's timers for MLDv2.
1482  * Will update the global pending timer flags.
1483  * Note: Unlocked read from mli.
1484  */
1485 static void
1486 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1487     struct mbufq *qrq, struct mbufq *scq,
1488     struct in6_multi *inm, const int uri_fasthz)
1489 {
1490 	int query_response_timer_expired;
1491 	int state_change_retransmit_timer_expired;
1492 #ifdef KTR
1493 	char ip6tbuf[INET6_ADDRSTRLEN];
1494 #endif
1495 
1496 	IN6_MULTI_LIST_LOCK_ASSERT();
1497 	MLD_LOCK_ASSERT();
1498 
1499 	query_response_timer_expired = 0;
1500 	state_change_retransmit_timer_expired = 0;
1501 
1502 	/*
1503 	 * During a transition from compatibility mode back to MLDv2,
1504 	 * a group record in REPORTING state may still have its group
1505 	 * timer active. This is a no-op in this function; it is easier
1506 	 * to deal with it here than to complicate the slow-timeout path.
1507 	 */
1508 	if (inm->in6m_timer == 0) {
1509 		query_response_timer_expired = 0;
1510 	} else if (--inm->in6m_timer == 0) {
1511 		query_response_timer_expired = 1;
1512 	} else {
1513 		V_current_state_timers_running6 = 1;
1514 	}
1515 
1516 	if (inm->in6m_sctimer == 0) {
1517 		state_change_retransmit_timer_expired = 0;
1518 	} else if (--inm->in6m_sctimer == 0) {
1519 		state_change_retransmit_timer_expired = 1;
1520 	} else {
1521 		V_state_change_timers_running6 = 1;
1522 	}
1523 
1524 	/* We are in fasttimo, so be quick about it. */
1525 	if (!state_change_retransmit_timer_expired &&
1526 	    !query_response_timer_expired)
1527 		return;
1528 
1529 	switch (inm->in6m_state) {
1530 	case MLD_NOT_MEMBER:
1531 	case MLD_SILENT_MEMBER:
1532 	case MLD_SLEEPING_MEMBER:
1533 	case MLD_LAZY_MEMBER:
1534 	case MLD_AWAKENING_MEMBER:
1535 	case MLD_IDLE_MEMBER:
1536 		break;
1537 	case MLD_G_QUERY_PENDING_MEMBER:
1538 	case MLD_SG_QUERY_PENDING_MEMBER:
1539 		/*
1540 		 * Respond to a previously pending Group-Specific
1541 		 * or Group-and-Source-Specific query by enqueueing
1542 		 * the appropriate Current-State report for
1543 		 * immediate transmission.
1544 		 */
1545 		if (query_response_timer_expired) {
1546 			int retval __unused;
1547 
1548 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1549 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1550 			    0);
1551 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1552 			    __func__, retval);
1553 			inm->in6m_state = MLD_REPORTING_MEMBER;
1554 			in6m_clear_recorded(inm);
1555 		}
1556 		/* FALLTHROUGH */
1557 	case MLD_REPORTING_MEMBER:
1558 	case MLD_LEAVING_MEMBER:
1559 		if (state_change_retransmit_timer_expired) {
1560 			/*
1561 			 * State-change retransmission timer fired.
1562 			 * If there are any further pending retransmissions,
1563 			 * set the global pending state-change flag, and
1564 			 * reset the timer.
1565 			 */
1566 			if (--inm->in6m_scrv > 0) {
1567 				inm->in6m_sctimer = uri_fasthz;
1568 				V_state_change_timers_running6 = 1;
1569 			}
1570 			/*
1571 			 * Retransmit the previously computed state-change
1572 			 * report. If there are no further pending
1573 			 * retransmissions, the mbuf queue will be consumed.
1574 			 * Update T0 state to T1 as we have now sent
1575 			 * a state-change.
1576 			 */
1577 			(void)mld_v2_merge_state_changes(inm, scq);
1578 
1579 			in6m_commit(inm);
1580 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1581 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1582 			    if_name(inm->in6m_ifp));
1583 
1584 			/*
1585 			 * If we are leaving the group for good, make sure
1586 			 * we release MLD's reference to it.
1587 			 * This release must be deferred using a SLIST,
1588 			 * as we are called from a loop which traverses
1589 			 * the in_ifmultiaddr TAILQ.
1590 			 */
1591 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1592 			    inm->in6m_scrv == 0) {
1593 				inm->in6m_state = MLD_NOT_MEMBER;
1594 				in6m_disconnect_locked(inmh, inm);
1595 				in6m_rele_locked(inmh, inm);
1596 			}
1597 		}
1598 		break;
1599 	}
1600 }
1601 
1602 /*
1603  * Switch to a different version on the given interface,
1604  * as per Section 9.12.
1605  */
1606 static void
1607 mld_set_version(struct mld_ifsoftc *mli, const int version)
1608 {
1609 	int old_version_timer;
1610 
1611 	MLD_LOCK_ASSERT();
1612 
1613 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1614 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1615 
1616 	if (version == MLD_VERSION_1) {
1617 		/*
1618 		 * Compute the "Older Version Querier Present" timer as per
1619 		 * Section 9.12.
1620 		 */
1621 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1622 		old_version_timer *= MLD_SLOWHZ;
1623 		mli->mli_v1_timer = old_version_timer;
1624 	}
1625 
1626 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1627 		mli->mli_version = MLD_VERSION_1;
1628 		mld_v2_cancel_link_timers(mli);
1629 	}
1630 }
1631 
1632 /*
1633  * Cancel pending MLDv2 timers for the given link and all groups
1634  * joined on it; state-change, general-query, and group-query timers.
1635  */
1636 static void
1637 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1638 {
1639 	struct epoch_tracker	 et;
1640 	struct in6_multi_head	 inmh;
1641 	struct ifmultiaddr	*ifma;
1642 	struct ifnet		*ifp;
1643 	struct in6_multi	*inm;
1644 
1645 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1646 	    mli->mli_ifp, if_name(mli->mli_ifp));
1647 
1648 	SLIST_INIT(&inmh);
1649 	IN6_MULTI_LIST_LOCK_ASSERT();
1650 	MLD_LOCK_ASSERT();
1651 
1652 	/*
1653 	 * Fast-track this potentially expensive operation
1654 	 * by checking all the global 'timer pending' flags.
1655 	 */
1656 	if (!V_interface_timers_running6 &&
1657 	    !V_state_change_timers_running6 &&
1658 	    !V_current_state_timers_running6)
1659 		return;
1660 
1661 	mli->mli_v2_timer = 0;
1662 
1663 	ifp = mli->mli_ifp;
1664 
1665 	IF_ADDR_WLOCK(ifp);
1666 	NET_EPOCH_ENTER(et);
1667 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1668 		inm = in6m_ifmultiaddr_get_inm(ifma);
1669 		if (inm == NULL)
1670 			continue;
1671 		switch (inm->in6m_state) {
1672 		case MLD_NOT_MEMBER:
1673 		case MLD_SILENT_MEMBER:
1674 		case MLD_IDLE_MEMBER:
1675 		case MLD_LAZY_MEMBER:
1676 		case MLD_SLEEPING_MEMBER:
1677 		case MLD_AWAKENING_MEMBER:
1678 			break;
1679 		case MLD_LEAVING_MEMBER:
1680 			/*
1681 			 * If we are leaving the group and switching
1682 			 * version, we need to release the final
1683 			 * reference held for issuing the INCLUDE {}.
1684 			 */
1685 			if (inm->in6m_refcount == 1)
1686 				in6m_disconnect_locked(&inmh, inm);
1687 			in6m_rele_locked(&inmh, inm);
1688 			/* FALLTHROUGH */
1689 		case MLD_G_QUERY_PENDING_MEMBER:
1690 		case MLD_SG_QUERY_PENDING_MEMBER:
1691 			in6m_clear_recorded(inm);
1692 			/* FALLTHROUGH */
1693 		case MLD_REPORTING_MEMBER:
1694 			inm->in6m_sctimer = 0;
1695 			inm->in6m_timer = 0;
1696 			inm->in6m_state = MLD_REPORTING_MEMBER;
1697 			/*
1698 			 * Free any pending MLDv2 state-change records.
1699 			 */
1700 			mbufq_drain(&inm->in6m_scq);
1701 			break;
1702 		}
1703 	}
1704 	NET_EPOCH_EXIT(et);
1705 	IF_ADDR_WUNLOCK(ifp);
1706 	in6m_release_list_deferred(&inmh);
1707 }
1708 
1709 /*
1710  * Global slowtimo handler.
1711  * VIMAGE: Timeout handlers are expected to service all vimages.
1712  */
1713 static struct callout mldslow_callout;
1714 static void
1715 mld_slowtimo(void *arg __unused)
1716 {
1717 	VNET_ITERATOR_DECL(vnet_iter);
1718 
1719 	VNET_LIST_RLOCK_NOSLEEP();
1720 	VNET_FOREACH(vnet_iter) {
1721 		CURVNET_SET(vnet_iter);
1722 		mld_slowtimo_vnet();
1723 		CURVNET_RESTORE();
1724 	}
1725 	VNET_LIST_RUNLOCK_NOSLEEP();
1726 
1727 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
1728 }
1729 
1730 /*
1731  * Per-vnet slowtimo handler.
1732  */
1733 static void
1734 mld_slowtimo_vnet(void)
1735 {
1736 	struct mld_ifsoftc *mli;
1737 
1738 	MLD_LOCK();
1739 
1740 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1741 		mld_v1_process_querier_timers(mli);
1742 	}
1743 
1744 	MLD_UNLOCK();
1745 }
1746 
1747 /*
1748  * Update the Older Version Querier Present timers for a link.
1749  * See Section 9.12 of RFC 3810.
1750  */
1751 static void
1752 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1753 {
1754 
1755 	MLD_LOCK_ASSERT();
1756 
1757 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1758 		/*
1759 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1760 		 */
1761 		CTR5(KTR_MLD,
1762 		    "%s: transition from v%d -> v%d on %p(%s)",
1763 		    __func__, mli->mli_version, MLD_VERSION_2,
1764 		    mli->mli_ifp, if_name(mli->mli_ifp));
1765 		mli->mli_version = MLD_VERSION_2;
1766 	}
1767 }
1768 
1769 /*
1770  * Transmit an MLDv1 report immediately.
1771  */
1772 static int
1773 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1774 {
1775 	struct ifnet		*ifp;
1776 	struct in6_ifaddr	*ia;
1777 	struct ip6_hdr		*ip6;
1778 	struct mbuf		*mh, *md;
1779 	struct mld_hdr		*mld;
1780 
1781 	NET_EPOCH_ASSERT();
1782 	IN6_MULTI_LIST_LOCK_ASSERT();
1783 	MLD_LOCK_ASSERT();
1784 
1785 	ifp = in6m->in6m_ifp;
1786 	/* in process of being freed */
1787 	if (ifp == NULL)
1788 		return (0);
1789 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1790 	/* ia may be NULL if link-local address is tentative. */
1791 
1792 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1793 	if (mh == NULL) {
1794 		if (ia != NULL)
1795 			ifa_free(&ia->ia_ifa);
1796 		return (ENOMEM);
1797 	}
1798 	md = m_get(M_NOWAIT, MT_DATA);
1799 	if (md == NULL) {
1800 		m_free(mh);
1801 		if (ia != NULL)
1802 			ifa_free(&ia->ia_ifa);
1803 		return (ENOMEM);
1804 	}
1805 	mh->m_next = md;
1806 
1807 	/*
1808 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1809 	 * that ether_output() does not need to allocate another mbuf
1810 	 * for the header in the most common case.
1811 	 */
1812 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1813 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1814 	mh->m_len = sizeof(struct ip6_hdr);
1815 
1816 	ip6 = mtod(mh, struct ip6_hdr *);
1817 	ip6->ip6_flow = 0;
1818 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1819 	ip6->ip6_vfc |= IPV6_VERSION;
1820 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1821 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1822 	ip6->ip6_dst = in6m->in6m_addr;
1823 
1824 	md->m_len = sizeof(struct mld_hdr);
1825 	mld = mtod(md, struct mld_hdr *);
1826 	mld->mld_type = type;
1827 	mld->mld_code = 0;
1828 	mld->mld_cksum = 0;
1829 	mld->mld_maxdelay = 0;
1830 	mld->mld_reserved = 0;
1831 	mld->mld_addr = in6m->in6m_addr;
1832 	in6_clearscope(&mld->mld_addr);
1833 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1834 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1835 
1836 	mld_save_context(mh, ifp);
1837 	mh->m_flags |= M_MLDV1;
1838 
1839 	mld_dispatch_packet(mh);
1840 
1841 	if (ia != NULL)
1842 		ifa_free(&ia->ia_ifa);
1843 	return (0);
1844 }
1845 
1846 /*
1847  * Process a state change from the upper layer for the given IPv6 group.
1848  *
1849  * Each socket holds a reference on the in_multi in its own ip_moptions.
1850  * The socket layer will have made the necessary updates to.the group
1851  * state, it is now up to MLD to issue a state change report if there
1852  * has been any change between T0 (when the last state-change was issued)
1853  * and T1 (now).
1854  *
1855  * We use the MLDv2 state machine at group level. The MLd module
1856  * however makes the decision as to which MLD protocol version to speak.
1857  * A state change *from* INCLUDE {} always means an initial join.
1858  * A state change *to* INCLUDE {} always means a final leave.
1859  *
1860  * If delay is non-zero, and the state change is an initial multicast
1861  * join, the state change report will be delayed by 'delay' ticks
1862  * in units of MLD_FASTHZ if MLDv1 is active on the link; otherwise
1863  * the initial MLDv2 state change report will be delayed by whichever
1864  * is sooner, a pending state-change timer or delay itself.
1865  *
1866  * VIMAGE: curvnet should have been set by caller, as this routine
1867  * is called from the socket option handlers.
1868  */
1869 int
1870 mld_change_state(struct in6_multi *inm, const int delay)
1871 {
1872 	struct mld_ifsoftc *mli;
1873 	struct ifnet *ifp;
1874 	int error;
1875 
1876 	IN6_MULTI_LIST_LOCK_ASSERT();
1877 
1878 	error = 0;
1879 
1880 	/*
1881 	 * Check if the in6_multi has already been disconnected.
1882 	 */
1883 	if (inm->in6m_ifp == NULL) {
1884 		CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1885 		return (0);
1886 	}
1887 
1888 	/*
1889 	 * Try to detect if the upper layer just asked us to change state
1890 	 * for an interface which has now gone away.
1891 	 */
1892 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1893 	ifp = inm->in6m_ifma->ifma_ifp;
1894 	if (ifp == NULL)
1895 		return (0);
1896 	/*
1897 	 * Sanity check that netinet6's notion of ifp is the
1898 	 * same as net's.
1899 	 */
1900 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1901 
1902 	MLD_LOCK();
1903 	mli = MLD_IFINFO(ifp);
1904 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1905 
1906 	/*
1907 	 * If we detect a state transition to or from MCAST_UNDEFINED
1908 	 * for this group, then we are starting or finishing an MLD
1909 	 * life cycle for this group.
1910 	 */
1911 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1912 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1913 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1914 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1915 			CTR1(KTR_MLD, "%s: initial join", __func__);
1916 			error = mld_initial_join(inm, mli, delay);
1917 			goto out_locked;
1918 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1919 			CTR1(KTR_MLD, "%s: final leave", __func__);
1920 			mld_final_leave(inm, mli);
1921 			goto out_locked;
1922 		}
1923 	} else {
1924 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1925 	}
1926 
1927 	error = mld_handle_state_change(inm, mli);
1928 
1929 out_locked:
1930 	MLD_UNLOCK();
1931 	return (error);
1932 }
1933 
1934 /*
1935  * Perform the initial join for an MLD group.
1936  *
1937  * When joining a group:
1938  *  If the group should have its MLD traffic suppressed, do nothing.
1939  *  MLDv1 starts sending MLDv1 host membership reports.
1940  *  MLDv2 will schedule an MLDv2 state-change report containing the
1941  *  initial state of the membership.
1942  *
1943  * If the delay argument is non-zero, then we must delay sending the
1944  * initial state change for delay ticks (in units of MLD_FASTHZ).
1945  */
1946 static int
1947 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1948     const int delay)
1949 {
1950 	struct epoch_tracker     et;
1951 	struct ifnet		*ifp;
1952 	struct mbufq		*mq;
1953 	int			 error, retval, syncstates;
1954 	int			 odelay;
1955 #ifdef KTR
1956 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1957 #endif
1958 
1959 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1960 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1961 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1962 
1963 	error = 0;
1964 	syncstates = 1;
1965 
1966 	ifp = inm->in6m_ifp;
1967 
1968 	IN6_MULTI_LIST_LOCK_ASSERT();
1969 	MLD_LOCK_ASSERT();
1970 
1971 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1972 
1973 	/*
1974 	 * Groups joined on loopback or marked as 'not reported',
1975 	 * enter the MLD_SILENT_MEMBER state and
1976 	 * are never reported in any protocol exchanges.
1977 	 * All other groups enter the appropriate state machine
1978 	 * for the version in use on this link.
1979 	 * A link marked as MLIF_SILENT causes MLD to be completely
1980 	 * disabled for the link.
1981 	 */
1982 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1983 	    (mli->mli_flags & MLIF_SILENT) ||
1984 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1985 		CTR1(KTR_MLD,
1986 "%s: not kicking state machine for silent group", __func__);
1987 		inm->in6m_state = MLD_SILENT_MEMBER;
1988 		inm->in6m_timer = 0;
1989 	} else {
1990 		/*
1991 		 * Deal with overlapping in_multi lifecycle.
1992 		 * If this group was LEAVING, then make sure
1993 		 * we drop the reference we picked up to keep the
1994 		 * group around for the final INCLUDE {} enqueue.
1995 		 */
1996 		if (mli->mli_version == MLD_VERSION_2 &&
1997 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
1998 			inm->in6m_refcount--;
1999 			MPASS(inm->in6m_refcount > 0);
2000 		}
2001 		inm->in6m_state = MLD_REPORTING_MEMBER;
2002 
2003 		switch (mli->mli_version) {
2004 		case MLD_VERSION_1:
2005 			/*
2006 			 * If a delay was provided, only use it if
2007 			 * it is greater than the delay normally
2008 			 * used for an MLDv1 state change report,
2009 			 * and delay sending the initial MLDv1 report
2010 			 * by not transitioning to the IDLE state.
2011 			 */
2012 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * MLD_FASTHZ);
2013 			if (delay) {
2014 				inm->in6m_timer = max(delay, odelay);
2015 				V_current_state_timers_running6 = 1;
2016 			} else {
2017 				inm->in6m_state = MLD_IDLE_MEMBER;
2018 				NET_EPOCH_ENTER(et);
2019 				error = mld_v1_transmit_report(inm,
2020 				     MLD_LISTENER_REPORT);
2021 				NET_EPOCH_EXIT(et);
2022 				if (error == 0) {
2023 					inm->in6m_timer = odelay;
2024 					V_current_state_timers_running6 = 1;
2025 				}
2026 			}
2027 			break;
2028 
2029 		case MLD_VERSION_2:
2030 			/*
2031 			 * Defer update of T0 to T1, until the first copy
2032 			 * of the state change has been transmitted.
2033 			 */
2034 			syncstates = 0;
2035 
2036 			/*
2037 			 * Immediately enqueue a State-Change Report for
2038 			 * this interface, freeing any previous reports.
2039 			 * Don't kick the timers if there is nothing to do,
2040 			 * or if an error occurred.
2041 			 */
2042 			mq = &inm->in6m_scq;
2043 			mbufq_drain(mq);
2044 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2045 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2046 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2047 			    __func__, retval);
2048 			if (retval <= 0) {
2049 				error = retval * -1;
2050 				break;
2051 			}
2052 
2053 			/*
2054 			 * Schedule transmission of pending state-change
2055 			 * report up to RV times for this link. The timer
2056 			 * will fire at the next mld_fasttimo (~200ms),
2057 			 * giving us an opportunity to merge the reports.
2058 			 *
2059 			 * If a delay was provided to this function, only
2060 			 * use this delay if sooner than the existing one.
2061 			 */
2062 			KASSERT(mli->mli_rv > 1,
2063 			   ("%s: invalid robustness %d", __func__,
2064 			    mli->mli_rv));
2065 			inm->in6m_scrv = mli->mli_rv;
2066 			if (delay) {
2067 				if (inm->in6m_sctimer > 1) {
2068 					inm->in6m_sctimer =
2069 					    min(inm->in6m_sctimer, delay);
2070 				} else
2071 					inm->in6m_sctimer = delay;
2072 			} else
2073 				inm->in6m_sctimer = 1;
2074 			V_state_change_timers_running6 = 1;
2075 
2076 			error = 0;
2077 			break;
2078 		}
2079 	}
2080 
2081 	/*
2082 	 * Only update the T0 state if state change is atomic,
2083 	 * i.e. we don't need to wait for a timer to fire before we
2084 	 * can consider the state change to have been communicated.
2085 	 */
2086 	if (syncstates) {
2087 		in6m_commit(inm);
2088 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2089 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2090 		    if_name(inm->in6m_ifp));
2091 	}
2092 
2093 	return (error);
2094 }
2095 
2096 /*
2097  * Issue an intermediate state change during the life-cycle.
2098  */
2099 static int
2100 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2101 {
2102 	struct ifnet		*ifp;
2103 	int			 retval;
2104 #ifdef KTR
2105 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2106 #endif
2107 
2108 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2109 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2110 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2111 
2112 	ifp = inm->in6m_ifp;
2113 
2114 	IN6_MULTI_LIST_LOCK_ASSERT();
2115 	MLD_LOCK_ASSERT();
2116 
2117 	KASSERT(mli && mli->mli_ifp == ifp,
2118 	    ("%s: inconsistent ifp", __func__));
2119 
2120 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2121 	    (mli->mli_flags & MLIF_SILENT) ||
2122 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2123 	    (mli->mli_version != MLD_VERSION_2)) {
2124 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2125 			CTR1(KTR_MLD,
2126 "%s: not kicking state machine for silent group", __func__);
2127 		}
2128 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2129 		in6m_commit(inm);
2130 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2131 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2132 		    if_name(inm->in6m_ifp));
2133 		return (0);
2134 	}
2135 
2136 	mbufq_drain(&inm->in6m_scq);
2137 
2138 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2139 	    (mli->mli_flags & MLIF_USEALLOW));
2140 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2141 	if (retval <= 0)
2142 		return (-retval);
2143 
2144 	/*
2145 	 * If record(s) were enqueued, start the state-change
2146 	 * report timer for this group.
2147 	 */
2148 	inm->in6m_scrv = mli->mli_rv;
2149 	inm->in6m_sctimer = 1;
2150 	V_state_change_timers_running6 = 1;
2151 
2152 	return (0);
2153 }
2154 
2155 /*
2156  * Perform the final leave for a multicast address.
2157  *
2158  * When leaving a group:
2159  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2160  *  MLDv2 enqueues a state-change report containing a transition
2161  *  to INCLUDE {} for immediate transmission.
2162  */
2163 static void
2164 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2165 {
2166 	struct epoch_tracker     et;
2167 	int syncstates;
2168 #ifdef KTR
2169 	char ip6tbuf[INET6_ADDRSTRLEN];
2170 #endif
2171 
2172 	syncstates = 1;
2173 
2174 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2175 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2176 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2177 
2178 	IN6_MULTI_LIST_LOCK_ASSERT();
2179 	MLD_LOCK_ASSERT();
2180 
2181 	switch (inm->in6m_state) {
2182 	case MLD_NOT_MEMBER:
2183 	case MLD_SILENT_MEMBER:
2184 	case MLD_LEAVING_MEMBER:
2185 		/* Already leaving or left; do nothing. */
2186 		CTR1(KTR_MLD,
2187 "%s: not kicking state machine for silent group", __func__);
2188 		break;
2189 	case MLD_REPORTING_MEMBER:
2190 	case MLD_IDLE_MEMBER:
2191 	case MLD_G_QUERY_PENDING_MEMBER:
2192 	case MLD_SG_QUERY_PENDING_MEMBER:
2193 		if (mli->mli_version == MLD_VERSION_1) {
2194 #ifdef INVARIANTS
2195 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2196 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2197 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2198 			     __func__);
2199 #endif
2200 			NET_EPOCH_ENTER(et);
2201 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2202 			NET_EPOCH_EXIT(et);
2203 			inm->in6m_state = MLD_NOT_MEMBER;
2204 			V_current_state_timers_running6 = 1;
2205 		} else if (mli->mli_version == MLD_VERSION_2) {
2206 			/*
2207 			 * Stop group timer and all pending reports.
2208 			 * Immediately enqueue a state-change report
2209 			 * TO_IN {} to be sent on the next fast timeout,
2210 			 * giving us an opportunity to merge reports.
2211 			 */
2212 			mbufq_drain(&inm->in6m_scq);
2213 			inm->in6m_timer = 0;
2214 			inm->in6m_scrv = mli->mli_rv;
2215 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2216 			    "pending retransmissions.", __func__,
2217 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2218 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2219 			if (inm->in6m_scrv == 0) {
2220 				inm->in6m_state = MLD_NOT_MEMBER;
2221 				inm->in6m_sctimer = 0;
2222 			} else {
2223 				int retval __diagused;
2224 
2225 				in6m_acquire_locked(inm);
2226 
2227 				retval = mld_v2_enqueue_group_record(
2228 				    &inm->in6m_scq, inm, 1, 0, 0,
2229 				    (mli->mli_flags & MLIF_USEALLOW));
2230 				KASSERT(retval != 0,
2231 				    ("%s: enqueue record = %d", __func__,
2232 				     retval));
2233 
2234 				inm->in6m_state = MLD_LEAVING_MEMBER;
2235 				inm->in6m_sctimer = 1;
2236 				V_state_change_timers_running6 = 1;
2237 				syncstates = 0;
2238 			}
2239 			break;
2240 		}
2241 		break;
2242 	case MLD_LAZY_MEMBER:
2243 	case MLD_SLEEPING_MEMBER:
2244 	case MLD_AWAKENING_MEMBER:
2245 		/* Our reports are suppressed; do nothing. */
2246 		break;
2247 	}
2248 
2249 	if (syncstates) {
2250 		in6m_commit(inm);
2251 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2252 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2253 		    if_name(inm->in6m_ifp));
2254 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2255 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2256 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2257 	}
2258 }
2259 
2260 /*
2261  * Enqueue an MLDv2 group record to the given output queue.
2262  *
2263  * If is_state_change is zero, a current-state record is appended.
2264  * If is_state_change is non-zero, a state-change report is appended.
2265  *
2266  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2267  * If is_group_query is zero, and if there is a packet with free space
2268  * at the tail of the queue, it will be appended to providing there
2269  * is enough free space.
2270  * Otherwise a new mbuf packet chain is allocated.
2271  *
2272  * If is_source_query is non-zero, each source is checked to see if
2273  * it was recorded for a Group-Source query, and will be omitted if
2274  * it is not both in-mode and recorded.
2275  *
2276  * If use_block_allow is non-zero, state change reports for initial join
2277  * and final leave, on an inclusive mode group with a source list, will be
2278  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2279  *
2280  * The function will attempt to allocate leading space in the packet
2281  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2282  *
2283  * If successful the size of all data appended to the queue is returned,
2284  * otherwise an error code less than zero is returned, or zero if
2285  * no record(s) were appended.
2286  */
2287 static int
2288 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2289     const int is_state_change, const int is_group_query,
2290     const int is_source_query, const int use_block_allow)
2291 {
2292 	struct mldv2_record	 mr;
2293 	struct mldv2_record	*pmr;
2294 	struct ifnet		*ifp;
2295 	struct ip6_msource	*ims, *nims;
2296 	struct mbuf		*m0, *m, *md;
2297 	int			 is_filter_list_change;
2298 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2299 	int			 record_has_sources;
2300 	int			 now;
2301 	int			 type;
2302 	uint8_t			 mode;
2303 #ifdef KTR
2304 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2305 #endif
2306 
2307 	IN6_MULTI_LIST_LOCK_ASSERT();
2308 
2309 	ifp = inm->in6m_ifp;
2310 	is_filter_list_change = 0;
2311 	m = NULL;
2312 	m0 = NULL;
2313 	m0srcs = 0;
2314 	msrcs = 0;
2315 	nbytes = 0;
2316 	nims = NULL;
2317 	record_has_sources = 1;
2318 	pmr = NULL;
2319 	type = MLD_DO_NOTHING;
2320 	mode = inm->in6m_st[1].iss_fmode;
2321 
2322 	/*
2323 	 * If we did not transition out of ASM mode during t0->t1,
2324 	 * and there are no source nodes to process, we can skip
2325 	 * the generation of source records.
2326 	 */
2327 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2328 	    inm->in6m_nsrc == 0)
2329 		record_has_sources = 0;
2330 
2331 	if (is_state_change) {
2332 		/*
2333 		 * Queue a state change record.
2334 		 * If the mode did not change, and there are non-ASM
2335 		 * listeners or source filters present,
2336 		 * we potentially need to issue two records for the group.
2337 		 * If there are ASM listeners, and there was no filter
2338 		 * mode transition of any kind, do nothing.
2339 		 *
2340 		 * If we are transitioning to MCAST_UNDEFINED, we need
2341 		 * not send any sources. A transition to/from this state is
2342 		 * considered inclusive with some special treatment.
2343 		 *
2344 		 * If we are rewriting initial joins/leaves to use
2345 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2346 		 * we need to send sources in all cases.
2347 		 */
2348 		if (mode != inm->in6m_st[0].iss_fmode) {
2349 			if (mode == MCAST_EXCLUDE) {
2350 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2351 				    __func__);
2352 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2353 			} else {
2354 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2355 				    __func__);
2356 				if (use_block_allow) {
2357 					/*
2358 					 * XXX
2359 					 * Here we're interested in state
2360 					 * edges either direction between
2361 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2362 					 * Perhaps we should just check
2363 					 * the group state, rather than
2364 					 * the filter mode.
2365 					 */
2366 					if (mode == MCAST_UNDEFINED) {
2367 						type = MLD_BLOCK_OLD_SOURCES;
2368 					} else {
2369 						type = MLD_ALLOW_NEW_SOURCES;
2370 					}
2371 				} else {
2372 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2373 					if (mode == MCAST_UNDEFINED)
2374 						record_has_sources = 0;
2375 				}
2376 			}
2377 		} else {
2378 			if (record_has_sources) {
2379 				is_filter_list_change = 1;
2380 			} else {
2381 				type = MLD_DO_NOTHING;
2382 			}
2383 		}
2384 	} else {
2385 		/*
2386 		 * Queue a current state record.
2387 		 */
2388 		if (mode == MCAST_EXCLUDE) {
2389 			type = MLD_MODE_IS_EXCLUDE;
2390 		} else if (mode == MCAST_INCLUDE) {
2391 			type = MLD_MODE_IS_INCLUDE;
2392 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2393 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2394 			     __func__, inm, inm->in6m_st[1].iss_asm));
2395 		}
2396 	}
2397 
2398 	/*
2399 	 * Generate the filter list changes using a separate function.
2400 	 */
2401 	if (is_filter_list_change)
2402 		return (mld_v2_enqueue_filter_change(mq, inm));
2403 
2404 	if (type == MLD_DO_NOTHING) {
2405 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2406 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2407 		    if_name(inm->in6m_ifp));
2408 		return (0);
2409 	}
2410 
2411 	/*
2412 	 * If any sources are present, we must be able to fit at least
2413 	 * one in the trailing space of the tail packet's mbuf,
2414 	 * ideally more.
2415 	 */
2416 	minrec0len = sizeof(struct mldv2_record);
2417 	if (record_has_sources)
2418 		minrec0len += sizeof(struct in6_addr);
2419 
2420 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2421 	    mld_rec_type_to_str(type),
2422 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2423 	    if_name(inm->in6m_ifp));
2424 
2425 	/*
2426 	 * Check if we have a packet in the tail of the queue for this
2427 	 * group into which the first group record for this group will fit.
2428 	 * Otherwise allocate a new packet.
2429 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2430 	 * Note: Group records for G/GSR query responses MUST be sent
2431 	 * in their own packet.
2432 	 */
2433 	m0 = mbufq_last(mq);
2434 	if (!is_group_query &&
2435 	    m0 != NULL &&
2436 	    (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2437 	    (m0->m_pkthdr.len + minrec0len) <
2438 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2439 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2440 			    sizeof(struct mldv2_record)) /
2441 			    sizeof(struct in6_addr);
2442 		m = m0;
2443 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2444 	} else {
2445 		if (mbufq_full(mq)) {
2446 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2447 			return (-ENOMEM);
2448 		}
2449 		m = NULL;
2450 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2451 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2452 		if (!is_state_change && !is_group_query)
2453 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2454 		if (m == NULL)
2455 			m = m_gethdr(M_NOWAIT, MT_DATA);
2456 		if (m == NULL)
2457 			return (-ENOMEM);
2458 
2459 		mld_save_context(m, ifp);
2460 
2461 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2462 	}
2463 
2464 	/*
2465 	 * Append group record.
2466 	 * If we have sources, we don't know how many yet.
2467 	 */
2468 	mr.mr_type = type;
2469 	mr.mr_datalen = 0;
2470 	mr.mr_numsrc = 0;
2471 	mr.mr_addr = inm->in6m_addr;
2472 	in6_clearscope(&mr.mr_addr);
2473 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2474 		if (m != m0)
2475 			m_freem(m);
2476 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2477 		return (-ENOMEM);
2478 	}
2479 	nbytes += sizeof(struct mldv2_record);
2480 
2481 	/*
2482 	 * Append as many sources as will fit in the first packet.
2483 	 * If we are appending to a new packet, the chain allocation
2484 	 * may potentially use clusters; use m_getptr() in this case.
2485 	 * If we are appending to an existing packet, we need to obtain
2486 	 * a pointer to the group record after m_append(), in case a new
2487 	 * mbuf was allocated.
2488 	 *
2489 	 * Only append sources which are in-mode at t1. If we are
2490 	 * transitioning to MCAST_UNDEFINED state on the group, and
2491 	 * use_block_allow is zero, do not include source entries.
2492 	 * Otherwise, we need to include this source in the report.
2493 	 *
2494 	 * Only report recorded sources in our filter set when responding
2495 	 * to a group-source query.
2496 	 */
2497 	if (record_has_sources) {
2498 		if (m == m0) {
2499 			md = m_last(m);
2500 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2501 			    md->m_len - nbytes);
2502 		} else {
2503 			md = m_getptr(m, 0, &off);
2504 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2505 			    off);
2506 		}
2507 		msrcs = 0;
2508 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2509 		    nims) {
2510 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2511 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2512 			now = im6s_get_mode(inm, ims, 1);
2513 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2514 			if ((now != mode) ||
2515 			    (now == mode &&
2516 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2517 				CTR1(KTR_MLD, "%s: skip node", __func__);
2518 				continue;
2519 			}
2520 			if (is_source_query && ims->im6s_stp == 0) {
2521 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2522 				    __func__);
2523 				continue;
2524 			}
2525 			CTR1(KTR_MLD, "%s: append node", __func__);
2526 			if (!m_append(m, sizeof(struct in6_addr),
2527 			    (void *)&ims->im6s_addr)) {
2528 				if (m != m0)
2529 					m_freem(m);
2530 				CTR1(KTR_MLD, "%s: m_append() failed.",
2531 				    __func__);
2532 				return (-ENOMEM);
2533 			}
2534 			nbytes += sizeof(struct in6_addr);
2535 			++msrcs;
2536 			if (msrcs == m0srcs)
2537 				break;
2538 		}
2539 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2540 		    msrcs);
2541 		pmr->mr_numsrc = htons(msrcs);
2542 		nbytes += (msrcs * sizeof(struct in6_addr));
2543 	}
2544 
2545 	if (is_source_query && msrcs == 0) {
2546 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2547 		if (m != m0)
2548 			m_freem(m);
2549 		return (0);
2550 	}
2551 
2552 	/*
2553 	 * We are good to go with first packet.
2554 	 */
2555 	if (m != m0) {
2556 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2557 		m->m_pkthdr.vt_nrecs = 1;
2558 		mbufq_enqueue(mq, m);
2559 	} else
2560 		m->m_pkthdr.vt_nrecs++;
2561 
2562 	/*
2563 	 * No further work needed if no source list in packet(s).
2564 	 */
2565 	if (!record_has_sources)
2566 		return (nbytes);
2567 
2568 	/*
2569 	 * Whilst sources remain to be announced, we need to allocate
2570 	 * a new packet and fill out as many sources as will fit.
2571 	 * Always try for a cluster first.
2572 	 */
2573 	while (nims != NULL) {
2574 		if (mbufq_full(mq)) {
2575 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2576 			return (-ENOMEM);
2577 		}
2578 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2579 		if (m == NULL)
2580 			m = m_gethdr(M_NOWAIT, MT_DATA);
2581 		if (m == NULL)
2582 			return (-ENOMEM);
2583 		mld_save_context(m, ifp);
2584 		md = m_getptr(m, 0, &off);
2585 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2586 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2587 
2588 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2589 			if (m != m0)
2590 				m_freem(m);
2591 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2592 			return (-ENOMEM);
2593 		}
2594 		m->m_pkthdr.vt_nrecs = 1;
2595 		nbytes += sizeof(struct mldv2_record);
2596 
2597 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2598 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2599 
2600 		msrcs = 0;
2601 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2602 			CTR2(KTR_MLD, "%s: visit node %s",
2603 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2604 			now = im6s_get_mode(inm, ims, 1);
2605 			if ((now != mode) ||
2606 			    (now == mode &&
2607 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2608 				CTR1(KTR_MLD, "%s: skip node", __func__);
2609 				continue;
2610 			}
2611 			if (is_source_query && ims->im6s_stp == 0) {
2612 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2613 				    __func__);
2614 				continue;
2615 			}
2616 			CTR1(KTR_MLD, "%s: append node", __func__);
2617 			if (!m_append(m, sizeof(struct in6_addr),
2618 			    (void *)&ims->im6s_addr)) {
2619 				if (m != m0)
2620 					m_freem(m);
2621 				CTR1(KTR_MLD, "%s: m_append() failed.",
2622 				    __func__);
2623 				return (-ENOMEM);
2624 			}
2625 			++msrcs;
2626 			if (msrcs == m0srcs)
2627 				break;
2628 		}
2629 		pmr->mr_numsrc = htons(msrcs);
2630 		nbytes += (msrcs * sizeof(struct in6_addr));
2631 
2632 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2633 		mbufq_enqueue(mq, m);
2634 	}
2635 
2636 	return (nbytes);
2637 }
2638 
2639 /*
2640  * Type used to mark record pass completion.
2641  * We exploit the fact we can cast to this easily from the
2642  * current filter modes on each ip_msource node.
2643  */
2644 typedef enum {
2645 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2646 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2647 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2648 	REC_FULL = REC_ALLOW | REC_BLOCK
2649 } rectype_t;
2650 
2651 /*
2652  * Enqueue an MLDv2 filter list change to the given output queue.
2653  *
2654  * Source list filter state is held in an RB-tree. When the filter list
2655  * for a group is changed without changing its mode, we need to compute
2656  * the deltas between T0 and T1 for each source in the filter set,
2657  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2658  *
2659  * As we may potentially queue two record types, and the entire R-B tree
2660  * needs to be walked at once, we break this out into its own function
2661  * so we can generate a tightly packed queue of packets.
2662  *
2663  * XXX This could be written to only use one tree walk, although that makes
2664  * serializing into the mbuf chains a bit harder. For now we do two walks
2665  * which makes things easier on us, and it may or may not be harder on
2666  * the L2 cache.
2667  *
2668  * If successful the size of all data appended to the queue is returned,
2669  * otherwise an error code less than zero is returned, or zero if
2670  * no record(s) were appended.
2671  */
2672 static int
2673 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2674 {
2675 	static const int MINRECLEN =
2676 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2677 	struct ifnet		*ifp;
2678 	struct mldv2_record	 mr;
2679 	struct mldv2_record	*pmr;
2680 	struct ip6_msource	*ims, *nims;
2681 	struct mbuf		*m, *m0, *md;
2682 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2683 	uint8_t			 mode, now, then;
2684 	rectype_t		 crt, drt, nrt;
2685 #ifdef KTR
2686 	int			 nallow, nblock;
2687 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2688 #endif
2689 
2690 	IN6_MULTI_LIST_LOCK_ASSERT();
2691 
2692 	if (inm->in6m_nsrc == 0 ||
2693 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2694 		return (0);
2695 
2696 	ifp = inm->in6m_ifp;			/* interface */
2697 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2698 	crt = REC_NONE;	/* current group record type */
2699 	drt = REC_NONE;	/* mask of completed group record types */
2700 	nrt = REC_NONE;	/* record type for current node */
2701 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2702 	npbytes = 0;	/* # of bytes appended this packet */
2703 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2704 	rsrcs = 0;	/* # sources encoded in current record */
2705 	schanged = 0;	/* # nodes encoded in overall filter change */
2706 #ifdef KTR
2707 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2708 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2709 #endif
2710 	nims = NULL;	/* next tree node pointer */
2711 
2712 	/*
2713 	 * For each possible filter record mode.
2714 	 * The first kind of source we encounter tells us which
2715 	 * is the first kind of record we start appending.
2716 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2717 	 * as the inverse of the group's filter mode.
2718 	 */
2719 	while (drt != REC_FULL) {
2720 		do {
2721 			m0 = mbufq_last(mq);
2722 			if (m0 != NULL &&
2723 			    (m0->m_pkthdr.vt_nrecs + 1 <=
2724 			     MLD_V2_REPORT_MAXRECS) &&
2725 			    (m0->m_pkthdr.len + MINRECLEN) <
2726 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2727 				m = m0;
2728 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2729 					    sizeof(struct mldv2_record)) /
2730 					    sizeof(struct in6_addr);
2731 				CTR1(KTR_MLD,
2732 				    "%s: use previous packet", __func__);
2733 			} else {
2734 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2735 				if (m == NULL)
2736 					m = m_gethdr(M_NOWAIT, MT_DATA);
2737 				if (m == NULL) {
2738 					CTR1(KTR_MLD,
2739 					    "%s: m_get*() failed", __func__);
2740 					return (-ENOMEM);
2741 				}
2742 				m->m_pkthdr.vt_nrecs = 0;
2743 				mld_save_context(m, ifp);
2744 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2745 				    sizeof(struct mldv2_record)) /
2746 				    sizeof(struct in6_addr);
2747 				npbytes = 0;
2748 				CTR1(KTR_MLD,
2749 				    "%s: allocated new packet", __func__);
2750 			}
2751 			/*
2752 			 * Append the MLD group record header to the
2753 			 * current packet's data area.
2754 			 * Recalculate pointer to free space for next
2755 			 * group record, in case m_append() allocated
2756 			 * a new mbuf or cluster.
2757 			 */
2758 			memset(&mr, 0, sizeof(mr));
2759 			mr.mr_addr = inm->in6m_addr;
2760 			in6_clearscope(&mr.mr_addr);
2761 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2762 				if (m != m0)
2763 					m_freem(m);
2764 				CTR1(KTR_MLD,
2765 				    "%s: m_append() failed", __func__);
2766 				return (-ENOMEM);
2767 			}
2768 			npbytes += sizeof(struct mldv2_record);
2769 			if (m != m0) {
2770 				/* new packet; offset in chain */
2771 				md = m_getptr(m, npbytes -
2772 				    sizeof(struct mldv2_record), &off);
2773 				pmr = (struct mldv2_record *)(mtod(md,
2774 				    uint8_t *) + off);
2775 			} else {
2776 				/* current packet; offset from last append */
2777 				md = m_last(m);
2778 				pmr = (struct mldv2_record *)(mtod(md,
2779 				    uint8_t *) + md->m_len -
2780 				    sizeof(struct mldv2_record));
2781 			}
2782 			/*
2783 			 * Begin walking the tree for this record type
2784 			 * pass, or continue from where we left off
2785 			 * previously if we had to allocate a new packet.
2786 			 * Only report deltas in-mode at t1.
2787 			 * We need not report included sources as allowed
2788 			 * if we are in inclusive mode on the group,
2789 			 * however the converse is not true.
2790 			 */
2791 			rsrcs = 0;
2792 			if (nims == NULL) {
2793 				nims = RB_MIN(ip6_msource_tree,
2794 				    &inm->in6m_srcs);
2795 			}
2796 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2797 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2798 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2799 				now = im6s_get_mode(inm, ims, 1);
2800 				then = im6s_get_mode(inm, ims, 0);
2801 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2802 				    __func__, then, now);
2803 				if (now == then) {
2804 					CTR1(KTR_MLD,
2805 					    "%s: skip unchanged", __func__);
2806 					continue;
2807 				}
2808 				if (mode == MCAST_EXCLUDE &&
2809 				    now == MCAST_INCLUDE) {
2810 					CTR1(KTR_MLD,
2811 					    "%s: skip IN src on EX group",
2812 					    __func__);
2813 					continue;
2814 				}
2815 				nrt = (rectype_t)now;
2816 				if (nrt == REC_NONE)
2817 					nrt = (rectype_t)(~mode & REC_FULL);
2818 				if (schanged++ == 0) {
2819 					crt = nrt;
2820 				} else if (crt != nrt)
2821 					continue;
2822 				if (!m_append(m, sizeof(struct in6_addr),
2823 				    (void *)&ims->im6s_addr)) {
2824 					if (m != m0)
2825 						m_freem(m);
2826 					CTR1(KTR_MLD,
2827 					    "%s: m_append() failed", __func__);
2828 					return (-ENOMEM);
2829 				}
2830 #ifdef KTR
2831 				nallow += !!(crt == REC_ALLOW);
2832 				nblock += !!(crt == REC_BLOCK);
2833 #endif
2834 				if (++rsrcs == m0srcs)
2835 					break;
2836 			}
2837 			/*
2838 			 * If we did not append any tree nodes on this
2839 			 * pass, back out of allocations.
2840 			 */
2841 			if (rsrcs == 0) {
2842 				npbytes -= sizeof(struct mldv2_record);
2843 				if (m != m0) {
2844 					CTR1(KTR_MLD,
2845 					    "%s: m_free(m)", __func__);
2846 					m_freem(m);
2847 				} else {
2848 					CTR1(KTR_MLD,
2849 					    "%s: m_adj(m, -mr)", __func__);
2850 					m_adj(m, -((int)sizeof(
2851 					    struct mldv2_record)));
2852 				}
2853 				continue;
2854 			}
2855 			npbytes += (rsrcs * sizeof(struct in6_addr));
2856 			if (crt == REC_ALLOW)
2857 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2858 			else if (crt == REC_BLOCK)
2859 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2860 			pmr->mr_numsrc = htons(rsrcs);
2861 			/*
2862 			 * Count the new group record, and enqueue this
2863 			 * packet if it wasn't already queued.
2864 			 */
2865 			m->m_pkthdr.vt_nrecs++;
2866 			if (m != m0)
2867 				mbufq_enqueue(mq, m);
2868 			nbytes += npbytes;
2869 		} while (nims != NULL);
2870 		drt |= crt;
2871 		crt = (~crt & REC_FULL);
2872 	}
2873 
2874 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2875 	    nallow, nblock);
2876 
2877 	return (nbytes);
2878 }
2879 
2880 static int
2881 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2882 {
2883 	struct mbufq	*gq;
2884 	struct mbuf	*m;		/* pending state-change */
2885 	struct mbuf	*m0;		/* copy of pending state-change */
2886 	struct mbuf	*mt;		/* last state-change in packet */
2887 	int		 docopy, domerge;
2888 	u_int		 recslen;
2889 
2890 	docopy = 0;
2891 	domerge = 0;
2892 	recslen = 0;
2893 
2894 	IN6_MULTI_LIST_LOCK_ASSERT();
2895 	MLD_LOCK_ASSERT();
2896 
2897 	/*
2898 	 * If there are further pending retransmissions, make a writable
2899 	 * copy of each queued state-change message before merging.
2900 	 */
2901 	if (inm->in6m_scrv > 0)
2902 		docopy = 1;
2903 
2904 	gq = &inm->in6m_scq;
2905 #ifdef KTR
2906 	if (mbufq_first(gq) == NULL) {
2907 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2908 		    __func__, inm);
2909 	}
2910 #endif
2911 
2912 	m = mbufq_first(gq);
2913 	while (m != NULL) {
2914 		/*
2915 		 * Only merge the report into the current packet if
2916 		 * there is sufficient space to do so; an MLDv2 report
2917 		 * packet may only contain 65,535 group records.
2918 		 * Always use a simple mbuf chain concatentation to do this,
2919 		 * as large state changes for single groups may have
2920 		 * allocated clusters.
2921 		 */
2922 		domerge = 0;
2923 		mt = mbufq_last(scq);
2924 		if (mt != NULL) {
2925 			recslen = m_length(m, NULL);
2926 
2927 			if ((mt->m_pkthdr.vt_nrecs +
2928 			    m->m_pkthdr.vt_nrecs <=
2929 			    MLD_V2_REPORT_MAXRECS) &&
2930 			    (mt->m_pkthdr.len + recslen <=
2931 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2932 				domerge = 1;
2933 		}
2934 
2935 		if (!domerge && mbufq_full(gq)) {
2936 			CTR2(KTR_MLD,
2937 			    "%s: outbound queue full, skipping whole packet %p",
2938 			    __func__, m);
2939 			mt = m->m_nextpkt;
2940 			if (!docopy)
2941 				m_freem(m);
2942 			m = mt;
2943 			continue;
2944 		}
2945 
2946 		if (!docopy) {
2947 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2948 			m0 = mbufq_dequeue(gq);
2949 			m = m0->m_nextpkt;
2950 		} else {
2951 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2952 			m0 = m_dup(m, M_NOWAIT);
2953 			if (m0 == NULL)
2954 				return (ENOMEM);
2955 			m0->m_nextpkt = NULL;
2956 			m = m->m_nextpkt;
2957 		}
2958 
2959 		if (!domerge) {
2960 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2961 			    __func__, m0, scq);
2962 			mbufq_enqueue(scq, m0);
2963 		} else {
2964 			struct mbuf *mtl;	/* last mbuf of packet mt */
2965 
2966 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2967 			    __func__, m0, mt);
2968 
2969 			mtl = m_last(mt);
2970 			m0->m_flags &= ~M_PKTHDR;
2971 			mt->m_pkthdr.len += recslen;
2972 			mt->m_pkthdr.vt_nrecs +=
2973 			    m0->m_pkthdr.vt_nrecs;
2974 
2975 			mtl->m_next = m0;
2976 		}
2977 	}
2978 
2979 	return (0);
2980 }
2981 
2982 /*
2983  * Respond to a pending MLDv2 General Query.
2984  */
2985 static void
2986 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2987 {
2988 	struct ifmultiaddr	*ifma;
2989 	struct ifnet		*ifp;
2990 	struct in6_multi	*inm;
2991 	int			 retval __unused;
2992 
2993 	NET_EPOCH_ASSERT();
2994 	IN6_MULTI_LIST_LOCK_ASSERT();
2995 	MLD_LOCK_ASSERT();
2996 
2997 	KASSERT(mli->mli_version == MLD_VERSION_2,
2998 	    ("%s: called when version %d", __func__, mli->mli_version));
2999 
3000 	/*
3001 	 * Check that there are some packets queued. If so, send them first.
3002 	 * For large number of groups the reply to general query can take
3003 	 * many packets, we should finish sending them before starting of
3004 	 * queuing the new reply.
3005 	 */
3006 	if (mbufq_len(&mli->mli_gq) != 0)
3007 		goto send;
3008 
3009 	ifp = mli->mli_ifp;
3010 
3011 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3012 		inm = in6m_ifmultiaddr_get_inm(ifma);
3013 		if (inm == NULL)
3014 			continue;
3015 		KASSERT(ifp == inm->in6m_ifp,
3016 		    ("%s: inconsistent ifp", __func__));
3017 
3018 		switch (inm->in6m_state) {
3019 		case MLD_NOT_MEMBER:
3020 		case MLD_SILENT_MEMBER:
3021 			break;
3022 		case MLD_REPORTING_MEMBER:
3023 		case MLD_IDLE_MEMBER:
3024 		case MLD_LAZY_MEMBER:
3025 		case MLD_SLEEPING_MEMBER:
3026 		case MLD_AWAKENING_MEMBER:
3027 			inm->in6m_state = MLD_REPORTING_MEMBER;
3028 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3029 			    inm, 0, 0, 0, 0);
3030 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3031 			    __func__, retval);
3032 			break;
3033 		case MLD_G_QUERY_PENDING_MEMBER:
3034 		case MLD_SG_QUERY_PENDING_MEMBER:
3035 		case MLD_LEAVING_MEMBER:
3036 			break;
3037 		}
3038 	}
3039 
3040 send:
3041 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3042 
3043 	/*
3044 	 * Slew transmission of bursts over 500ms intervals.
3045 	 */
3046 	if (mbufq_first(&mli->mli_gq) != NULL) {
3047 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3048 		    MLD_RESPONSE_BURST_INTERVAL);
3049 		V_interface_timers_running6 = 1;
3050 	}
3051 }
3052 
3053 /*
3054  * Transmit the next pending message in the output queue.
3055  *
3056  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3057  * MRT: Nothing needs to be done, as MLD traffic is always local to
3058  * a link and uses a link-scope multicast address.
3059  */
3060 static void
3061 mld_dispatch_packet(struct mbuf *m)
3062 {
3063 	struct ip6_moptions	 im6o;
3064 	struct ifnet		*ifp;
3065 	struct ifnet		*oifp;
3066 	struct mbuf		*m0;
3067 	struct mbuf		*md;
3068 	struct ip6_hdr		*ip6;
3069 	struct mld_hdr		*mld;
3070 	int			 error;
3071 	int			 off;
3072 	int			 type;
3073 	uint32_t		 ifindex;
3074 
3075 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3076 	NET_EPOCH_ASSERT();
3077 
3078 	/*
3079 	 * Set VNET image pointer from enqueued mbuf chain
3080 	 * before doing anything else. Whilst we use interface
3081 	 * indexes to guard against interface detach, they are
3082 	 * unique to each VIMAGE and must be retrieved.
3083 	 */
3084 	ifindex = mld_restore_context(m);
3085 
3086 	/*
3087 	 * Check if the ifnet still exists. This limits the scope of
3088 	 * any race in the absence of a global ifp lock for low cost
3089 	 * (an array lookup).
3090 	 */
3091 	ifp = ifnet_byindex(ifindex);
3092 	if (ifp == NULL) {
3093 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3094 		    __func__, m, ifindex);
3095 		m_freem(m);
3096 		IP6STAT_INC(ip6s_noroute);
3097 		goto out;
3098 	}
3099 
3100 	im6o.im6o_multicast_hlim  = 1;
3101 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3102 	im6o.im6o_multicast_ifp = ifp;
3103 
3104 	if (m->m_flags & M_MLDV1) {
3105 		m0 = m;
3106 	} else {
3107 		m0 = mld_v2_encap_report(ifp, m);
3108 		if (m0 == NULL) {
3109 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3110 			IP6STAT_INC(ip6s_odropped);
3111 			goto out;
3112 		}
3113 	}
3114 
3115 	mld_scrub_context(m0);
3116 	m_clrprotoflags(m);
3117 	m0->m_pkthdr.rcvif = V_loif;
3118 
3119 	ip6 = mtod(m0, struct ip6_hdr *);
3120 #if 0
3121 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3122 #else
3123 	/*
3124 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3125 	 * occur if we called in6_setscope() at transmission.
3126 	 * See comments at top of file.
3127 	 */
3128 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3129 #endif
3130 
3131 	/*
3132 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3133 	 * so we can bump the stats.
3134 	 */
3135 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3136 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3137 	type = mld->mld_type;
3138 
3139 	oifp = NULL;
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 	NET_EPOCH_ASSERT();
3189 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3190 	if (ia == NULL)
3191 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3192 
3193 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3194 	if (mh == NULL) {
3195 		if (ia != NULL)
3196 			ifa_free(&ia->ia_ifa);
3197 		m_freem(m);
3198 		return (NULL);
3199 	}
3200 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3201 
3202 	mldreclen = m_length(m, NULL);
3203 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3204 
3205 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3206 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3207 	    sizeof(struct mldv2_report) + mldreclen;
3208 
3209 	ip6 = mtod(mh, struct ip6_hdr *);
3210 	ip6->ip6_flow = 0;
3211 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3212 	ip6->ip6_vfc |= IPV6_VERSION;
3213 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3214 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3215 	if (ia != NULL)
3216 		ifa_free(&ia->ia_ifa);
3217 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3218 	/* scope ID will be set in netisr */
3219 
3220 	mld = (struct mldv2_report *)(ip6 + 1);
3221 	mld->mld_type = MLDV2_LISTENER_REPORT;
3222 	mld->mld_code = 0;
3223 	mld->mld_cksum = 0;
3224 	mld->mld_v2_reserved = 0;
3225 	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3226 	m->m_pkthdr.vt_nrecs = 0;
3227 
3228 	mh->m_next = m;
3229 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3230 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3231 	return (mh);
3232 }
3233 
3234 #ifdef KTR
3235 static char *
3236 mld_rec_type_to_str(const int type)
3237 {
3238 
3239 	switch (type) {
3240 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3241 			return "TO_EX";
3242 			break;
3243 		case MLD_CHANGE_TO_INCLUDE_MODE:
3244 			return "TO_IN";
3245 			break;
3246 		case MLD_MODE_IS_EXCLUDE:
3247 			return "MODE_EX";
3248 			break;
3249 		case MLD_MODE_IS_INCLUDE:
3250 			return "MODE_IN";
3251 			break;
3252 		case MLD_ALLOW_NEW_SOURCES:
3253 			return "ALLOW_NEW";
3254 			break;
3255 		case MLD_BLOCK_OLD_SOURCES:
3256 			return "BLOCK_OLD";
3257 			break;
3258 		default:
3259 			break;
3260 	}
3261 	return "unknown";
3262 }
3263 #endif
3264 
3265 static void
3266 mld_init(void *unused __unused)
3267 {
3268 
3269 	CTR1(KTR_MLD, "%s: initializing", __func__);
3270 	MLD_LOCK_INIT();
3271 
3272 	ip6_initpktopts(&mld_po);
3273 	mld_po.ip6po_hlim = 1;
3274 	mld_po.ip6po_hbh = &mld_ra.hbh;
3275 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3276 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3277 
3278 	callout_init(&mldslow_callout, 1);
3279 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
3280 	callout_init(&mldfast_callout, 1);
3281 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
3282 }
3283 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3284 
3285 static void
3286 mld_uninit(void *unused __unused)
3287 {
3288 
3289 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3290 	callout_drain(&mldslow_callout);
3291 	callout_drain(&mldfast_callout);
3292 	MLD_LOCK_DESTROY();
3293 }
3294 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3295 
3296 static void
3297 vnet_mld_init(const void *unused __unused)
3298 {
3299 
3300 	CTR1(KTR_MLD, "%s: initializing", __func__);
3301 
3302 	LIST_INIT(&V_mli_head);
3303 }
3304 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3305     NULL);
3306 
3307 static void
3308 vnet_mld_uninit(const void *unused __unused)
3309 {
3310 
3311 	/* This can happen if we shutdown the network stack. */
3312 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3313 }
3314 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3315     NULL);
3316 
3317 static int
3318 mld_modevent(module_t mod, int type, void *unused __unused)
3319 {
3320 
3321     switch (type) {
3322     case MOD_LOAD:
3323     case MOD_UNLOAD:
3324 	break;
3325     default:
3326 	return (EOPNOTSUPP);
3327     }
3328     return (0);
3329 }
3330 
3331 static moduledata_t mld_mod = {
3332     "mld",
3333     mld_modevent,
3334     0
3335 };
3336 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3337