xref: /freebsd/sys/netinet6/mld6.c (revision eb6d21b4ca6d668cf89afd99eef7baeafa712197)
1 /*-
2  * Copyright (c) 2009 Bruce Simpson.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. The name of the author may not be used to endorse or promote
13  *    products derived from this software without specific prior written
14  *    permission.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
29  */
30 
31 /*-
32  * Copyright (c) 1988 Stephen Deering.
33  * Copyright (c) 1992, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * Stephen Deering of Stanford University.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 4. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
64  */
65 
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
68 
69 #include "opt_inet.h"
70 #include "opt_inet6.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/mbuf.h>
75 #include <sys/socket.h>
76 #include <sys/protosw.h>
77 #include <sys/sysctl.h>
78 #include <sys/kernel.h>
79 #include <sys/callout.h>
80 #include <sys/malloc.h>
81 #include <sys/module.h>
82 #include <sys/ktr.h>
83 
84 #include <net/if.h>
85 #include <net/route.h>
86 #include <net/vnet.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 #include <netinet6/in6_var.h>
91 #include <netinet/ip6.h>
92 #include <netinet6/ip6_var.h>
93 #include <netinet6/scope6_var.h>
94 #include <netinet/icmp6.h>
95 #include <netinet6/mld6.h>
96 #include <netinet6/mld6_var.h>
97 
98 #include <security/mac/mac_framework.h>
99 
100 #ifndef KTR_MLD
101 #define KTR_MLD KTR_INET6
102 #endif
103 
104 static struct mld_ifinfo *
105 		mli_alloc_locked(struct ifnet *);
106 static void	mli_delete_locked(const struct ifnet *);
107 static void	mld_dispatch_packet(struct mbuf *);
108 static void	mld_dispatch_queue(struct ifqueue *, int);
109 static void	mld_final_leave(struct in6_multi *, struct mld_ifinfo *);
110 static void	mld_fasttimo_vnet(void);
111 static int	mld_handle_state_change(struct in6_multi *,
112 		    struct mld_ifinfo *);
113 static int	mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
114 		    const int);
115 #ifdef KTR
116 static char *	mld_rec_type_to_str(const int);
117 #endif
118 static void	mld_set_version(struct mld_ifinfo *, const int);
119 static void	mld_slowtimo_vnet(void);
120 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
121 		    /*const*/ struct mld_hdr *);
122 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
123 		    /*const*/ struct mld_hdr *);
124 static void	mld_v1_process_group_timer(struct in6_multi *, const int);
125 static void	mld_v1_process_querier_timers(struct mld_ifinfo *);
126 static int	mld_v1_transmit_report(struct in6_multi *, const int);
127 static void	mld_v1_update_group(struct in6_multi *, const int);
128 static void	mld_v2_cancel_link_timers(struct mld_ifinfo *);
129 static void	mld_v2_dispatch_general_query(struct mld_ifinfo *);
130 static struct mbuf *
131 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
132 static int	mld_v2_enqueue_filter_change(struct ifqueue *,
133 		    struct in6_multi *);
134 static int	mld_v2_enqueue_group_record(struct ifqueue *,
135 		    struct in6_multi *, const int, const int, const int);
136 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
137 		    struct mbuf *, const int, const int);
138 static int	mld_v2_merge_state_changes(struct in6_multi *,
139 		    struct ifqueue *);
140 static void	mld_v2_process_group_timers(struct mld_ifinfo *,
141 		    struct ifqueue *, struct ifqueue *,
142 		    struct in6_multi *, const int);
143 static int	mld_v2_process_group_query(struct in6_multi *,
144 		    struct mld_ifinfo *mli, int, struct mbuf *, const int);
145 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
146 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
147 
148 /*
149  * Normative references: RFC 2710, RFC 3590, RFC 3810.
150  *
151  * Locking:
152  *  * The MLD subsystem lock ends up being system-wide for the moment,
153  *    but could be per-VIMAGE later on.
154  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
155  *    Any may be taken independently; if any are held at the same
156  *    time, the above lock order must be followed.
157  *  * IN6_MULTI_LOCK covers in_multi.
158  *  * MLD_LOCK covers per-link state and any global variables in this file.
159  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
160  *    per-link state iterators.
161  *
162  *  XXX LOR PREVENTION
163  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
164  *  will not accept an ifp; it wants an embedded scope ID, unlike
165  *  ip_output(), which happily takes the ifp given to it. The embedded
166  *  scope ID is only used by MLD to select the outgoing interface.
167  *
168  *  During interface attach and detach, MLD will take MLD_LOCK *after*
169  *  the IF_AFDATA_LOCK.
170  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
171  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
172  *  dispatch could work around this, but we'd rather not do that, as it
173  *  can introduce other races.
174  *
175  *  As such, we exploit the fact that the scope ID is just the interface
176  *  index, and embed it in the IPv6 destination address accordingly.
177  *  This is potentially NOT VALID for MLDv1 reports, as they
178  *  are always sent to the multicast group itself; as MLDv2
179  *  reports are always sent to ff02::16, this is not an issue
180  *  when MLDv2 is in use.
181  *
182  *  This does not however eliminate the LOR when ip6_output() itself
183  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
184  *  trigger a LOR warning in WITNESS when the ifnet is detached.
185  *
186  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
187  *  how it's used across the network stack. Here we're simply exploiting
188  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
189  *
190  * VIMAGE:
191  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
192  *    to a vnet in ifp->if_vnet.
193  */
194 static struct mtx		 mld_mtx;
195 MALLOC_DEFINE(M_MLD, "mld", "mld state");
196 
197 #define	MLD_EMBEDSCOPE(pin6, zoneid) \
198 	(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)
199 
200 /*
201  * VIMAGE-wide globals.
202  */
203 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0};
204 static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head);
205 static VNET_DEFINE(int, interface_timers_running6);
206 static VNET_DEFINE(int, state_change_timers_running6);
207 static VNET_DEFINE(int, current_state_timers_running6);
208 
209 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
210 #define	V_mli_head			VNET(mli_head)
211 #define	V_interface_timers_running6	VNET(interface_timers_running6)
212 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
213 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
214 
215 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
216 
217 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
218     "IPv6 Multicast Listener Discovery");
219 
220 /*
221  * Virtualized sysctls.
222  */
223 SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
224     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
225     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
226     "Rate limit for MLDv2 Group-and-Source queries in seconds");
227 
228 /*
229  * Non-virtualized sysctls.
230  */
231 SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
232     sysctl_mld_ifinfo, "Per-interface MLDv2 state");
233 
234 static int	mld_v1enable = 1;
235 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW,
236     &mld_v1enable, 0, "Enable fallback to MLDv1");
237 TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable);
238 
239 /*
240  * Packed Router Alert option structure declaration.
241  */
242 struct mld_raopt {
243 	struct ip6_hbh		hbh;
244 	struct ip6_opt		pad;
245 	struct ip6_opt_router	ra;
246 } __packed;
247 
248 /*
249  * Router Alert hop-by-hop option header.
250  */
251 static struct mld_raopt mld_ra = {
252 	.hbh = { 0, 0 },
253 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
254 	.ra = {
255 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
256 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
257 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
258 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
259 	}
260 };
261 static struct ip6_pktopts mld_po;
262 
263 static __inline void
264 mld_save_context(struct mbuf *m, struct ifnet *ifp)
265 {
266 
267 #ifdef VIMAGE
268 	m->m_pkthdr.header = ifp->if_vnet;
269 #endif /* VIMAGE */
270 	m->m_pkthdr.flowid = ifp->if_index;
271 }
272 
273 static __inline void
274 mld_scrub_context(struct mbuf *m)
275 {
276 
277 	m->m_pkthdr.header = NULL;
278 	m->m_pkthdr.flowid = 0;
279 }
280 
281 /*
282  * Restore context from a queued output chain.
283  * Return saved ifindex.
284  *
285  * VIMAGE: The assertion is there to make sure that we
286  * actually called CURVNET_SET() with what's in the mbuf chain.
287  */
288 static __inline uint32_t
289 mld_restore_context(struct mbuf *m)
290 {
291 
292 #if defined(VIMAGE) && defined(INVARIANTS)
293 	KASSERT(curvnet == m->m_pkthdr.header,
294 	    ("%s: called when curvnet was not restored", __func__));
295 #endif
296 	return (m->m_pkthdr.flowid);
297 }
298 
299 /*
300  * Retrieve or set threshold between group-source queries in seconds.
301  *
302  * VIMAGE: Assume curvnet set by caller.
303  * SMPng: NOTE: Serialized by MLD lock.
304  */
305 static int
306 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
307 {
308 	int error;
309 	int i;
310 
311 	error = sysctl_wire_old_buffer(req, sizeof(int));
312 	if (error)
313 		return (error);
314 
315 	MLD_LOCK();
316 
317 	i = V_mld_gsrdelay.tv_sec;
318 
319 	error = sysctl_handle_int(oidp, &i, 0, req);
320 	if (error || !req->newptr)
321 		goto out_locked;
322 
323 	if (i < -1 || i >= 60) {
324 		error = EINVAL;
325 		goto out_locked;
326 	}
327 
328 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
329 	     V_mld_gsrdelay.tv_sec, i);
330 	V_mld_gsrdelay.tv_sec = i;
331 
332 out_locked:
333 	MLD_UNLOCK();
334 	return (error);
335 }
336 
337 /*
338  * Expose struct mld_ifinfo to userland, keyed by ifindex.
339  * For use by ifmcstat(8).
340  *
341  * SMPng: NOTE: Does an unlocked ifindex space read.
342  * VIMAGE: Assume curvnet set by caller. The node handler itself
343  * is not directly virtualized.
344  */
345 static int
346 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
347 {
348 	int			*name;
349 	int			 error;
350 	u_int			 namelen;
351 	struct ifnet		*ifp;
352 	struct mld_ifinfo	*mli;
353 
354 	name = (int *)arg1;
355 	namelen = arg2;
356 
357 	if (req->newptr != NULL)
358 		return (EPERM);
359 
360 	if (namelen != 1)
361 		return (EINVAL);
362 
363 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
364 	if (error)
365 		return (error);
366 
367 	IN6_MULTI_LOCK();
368 	MLD_LOCK();
369 
370 	if (name[0] <= 0 || name[0] > V_if_index) {
371 		error = ENOENT;
372 		goto out_locked;
373 	}
374 
375 	error = ENOENT;
376 
377 	ifp = ifnet_byindex(name[0]);
378 	if (ifp == NULL)
379 		goto out_locked;
380 
381 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
382 		if (ifp == mli->mli_ifp) {
383 			error = SYSCTL_OUT(req, mli,
384 			    sizeof(struct mld_ifinfo));
385 			break;
386 		}
387 	}
388 
389 out_locked:
390 	MLD_UNLOCK();
391 	IN6_MULTI_UNLOCK();
392 	return (error);
393 }
394 
395 /*
396  * Dispatch an entire queue of pending packet chains.
397  * VIMAGE: Assumes the vnet pointer has been set.
398  */
399 static void
400 mld_dispatch_queue(struct ifqueue *ifq, int limit)
401 {
402 	struct mbuf *m;
403 
404 	for (;;) {
405 		_IF_DEQUEUE(ifq, m);
406 		if (m == NULL)
407 			break;
408 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m);
409 		mld_dispatch_packet(m);
410 		if (--limit == 0)
411 			break;
412 	}
413 }
414 
415 /*
416  * Filter outgoing MLD report state by group.
417  *
418  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
419  * and node-local addresses. However, kernel and socket consumers
420  * always embed the KAME scope ID in the address provided, so strip it
421  * when performing comparison.
422  * Note: This is not the same as the *multicast* scope.
423  *
424  * Return zero if the given group is one for which MLD reports
425  * should be suppressed, or non-zero if reports should be issued.
426  */
427 static __inline int
428 mld_is_addr_reported(const struct in6_addr *addr)
429 {
430 
431 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
432 
433 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
434 		return (0);
435 
436 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
437 		struct in6_addr tmp = *addr;
438 		in6_clearscope(&tmp);
439 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
440 			return (0);
441 	}
442 
443 	return (1);
444 }
445 
446 /*
447  * Attach MLD when PF_INET6 is attached to an interface.
448  *
449  * SMPng: Normally called with IF_AFDATA_LOCK held.
450  */
451 struct mld_ifinfo *
452 mld_domifattach(struct ifnet *ifp)
453 {
454 	struct mld_ifinfo *mli;
455 
456 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
457 	    __func__, ifp, ifp->if_xname);
458 
459 	MLD_LOCK();
460 
461 	mli = mli_alloc_locked(ifp);
462 	if (!(ifp->if_flags & IFF_MULTICAST))
463 		mli->mli_flags |= MLIF_SILENT;
464 
465 	MLD_UNLOCK();
466 
467 	return (mli);
468 }
469 
470 /*
471  * VIMAGE: assume curvnet set by caller.
472  */
473 static struct mld_ifinfo *
474 mli_alloc_locked(/*const*/ struct ifnet *ifp)
475 {
476 	struct mld_ifinfo *mli;
477 
478 	MLD_LOCK_ASSERT();
479 
480 	mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO);
481 	if (mli == NULL)
482 		goto out;
483 
484 	mli->mli_ifp = ifp;
485 	mli->mli_version = MLD_VERSION_2;
486 	mli->mli_flags = 0;
487 	mli->mli_rv = MLD_RV_INIT;
488 	mli->mli_qi = MLD_QI_INIT;
489 	mli->mli_qri = MLD_QRI_INIT;
490 	mli->mli_uri = MLD_URI_INIT;
491 
492 	SLIST_INIT(&mli->mli_relinmhead);
493 
494 	/*
495 	 * Responses to general queries are subject to bounds.
496 	 */
497 	IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
498 
499 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
500 
501 	CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)",
502 	     ifp, ifp->if_xname);
503 
504 out:
505 	return (mli);
506 }
507 
508 /*
509  * Hook for ifdetach.
510  *
511  * NOTE: Some finalization tasks need to run before the protocol domain
512  * is detached, but also before the link layer does its cleanup.
513  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
514  *
515  * SMPng: Caller must hold IN6_MULTI_LOCK().
516  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
517  * XXX This routine is also bitten by unlocked ifma_protospec access.
518  */
519 void
520 mld_ifdetach(struct ifnet *ifp)
521 {
522 	struct mld_ifinfo	*mli;
523 	struct ifmultiaddr	*ifma;
524 	struct in6_multi	*inm, *tinm;
525 
526 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
527 	    ifp->if_xname);
528 
529 	IN6_MULTI_LOCK_ASSERT();
530 	MLD_LOCK();
531 
532 	mli = MLD_IFINFO(ifp);
533 	if (mli->mli_version == MLD_VERSION_2) {
534 		IF_ADDR_LOCK(ifp);
535 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
536 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
537 			    ifma->ifma_protospec == NULL)
538 				continue;
539 			inm = (struct in6_multi *)ifma->ifma_protospec;
540 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
541 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
542 				    inm, in6m_nrele);
543 			}
544 			in6m_clear_recorded(inm);
545 		}
546 		IF_ADDR_UNLOCK(ifp);
547 		SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
548 		    tinm) {
549 			SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
550 			in6m_release_locked(inm);
551 		}
552 	}
553 
554 	MLD_UNLOCK();
555 }
556 
557 /*
558  * Hook for domifdetach.
559  * Runs after link-layer cleanup; free MLD state.
560  *
561  * SMPng: Normally called with IF_AFDATA_LOCK held.
562  */
563 void
564 mld_domifdetach(struct ifnet *ifp)
565 {
566 
567 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
568 	    __func__, ifp, ifp->if_xname);
569 
570 	MLD_LOCK();
571 	mli_delete_locked(ifp);
572 	MLD_UNLOCK();
573 }
574 
575 static void
576 mli_delete_locked(const struct ifnet *ifp)
577 {
578 	struct mld_ifinfo *mli, *tmli;
579 
580 	CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)",
581 	    __func__, ifp, ifp->if_xname);
582 
583 	MLD_LOCK_ASSERT();
584 
585 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
586 		if (mli->mli_ifp == ifp) {
587 			/*
588 			 * Free deferred General Query responses.
589 			 */
590 			_IF_DRAIN(&mli->mli_gq);
591 
592 			LIST_REMOVE(mli, mli_link);
593 
594 			KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
595 			    ("%s: there are dangling in_multi references",
596 			    __func__));
597 
598 			free(mli, M_MLD);
599 			return;
600 		}
601 	}
602 #ifdef INVARIANTS
603 	panic("%s: mld_ifinfo not found for ifp %p\n", __func__,  ifp);
604 #endif
605 }
606 
607 /*
608  * Process a received MLDv1 general or address-specific query.
609  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
610  *
611  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
612  * mld_addr. This is OK as we own the mbuf chain.
613  */
614 static int
615 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
616     /*const*/ struct mld_hdr *mld)
617 {
618 	struct ifmultiaddr	*ifma;
619 	struct mld_ifinfo	*mli;
620 	struct in6_multi	*inm;
621 	int			 is_general_query;
622 	uint16_t		 timer;
623 #ifdef KTR
624 	char			 ip6tbuf[INET6_ADDRSTRLEN];
625 #endif
626 
627 	is_general_query = 0;
628 
629 	if (!mld_v1enable) {
630 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
631 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
632 		    ifp, ifp->if_xname);
633 		return (0);
634 	}
635 
636 	/*
637 	 * RFC3810 Section 6.2: MLD queries must originate from
638 	 * a router's link-local address.
639 	 */
640 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
641 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
642 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
643 		    ifp, ifp->if_xname);
644 		return (0);
645 	}
646 
647 	/*
648 	 * Do address field validation upfront before we accept
649 	 * the query.
650 	 */
651 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
652 		/*
653 		 * MLDv1 General Query.
654 		 * If this was not sent to the all-nodes group, ignore it.
655 		 */
656 		struct in6_addr		 dst;
657 
658 		dst = ip6->ip6_dst;
659 		in6_clearscope(&dst);
660 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
661 			return (EINVAL);
662 		is_general_query = 1;
663 	} else {
664 		/*
665 		 * Embed scope ID of receiving interface in MLD query for
666 		 * lookup whilst we don't hold other locks.
667 		 */
668 		in6_setscope(&mld->mld_addr, ifp, NULL);
669 	}
670 
671 	IN6_MULTI_LOCK();
672 	MLD_LOCK();
673 	IF_ADDR_LOCK(ifp);
674 
675 	/*
676 	 * Switch to MLDv1 host compatibility mode.
677 	 */
678 	mli = MLD_IFINFO(ifp);
679 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
680 	mld_set_version(mli, MLD_VERSION_1);
681 
682 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
683 	if (timer == 0)
684 		timer = 1;
685 
686 	if (is_general_query) {
687 		/*
688 		 * For each reporting group joined on this
689 		 * interface, kick the report timer.
690 		 */
691 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
692 		    ifp, ifp->if_xname);
693 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
694 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
695 			    ifma->ifma_protospec == NULL)
696 				continue;
697 			inm = (struct in6_multi *)ifma->ifma_protospec;
698 			mld_v1_update_group(inm, timer);
699 		}
700 	} else {
701 		/*
702 		 * MLDv1 Group-Specific Query.
703 		 * If this is a group-specific MLDv1 query, we need only
704 		 * look up the single group to process it.
705 		 */
706 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
707 		if (inm != NULL) {
708 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
709 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
710 			    ifp, ifp->if_xname);
711 			mld_v1_update_group(inm, timer);
712 		}
713 		/* XXX Clear embedded scope ID as userland won't expect it. */
714 		in6_clearscope(&mld->mld_addr);
715 	}
716 
717 	IF_ADDR_UNLOCK(ifp);
718 	MLD_UNLOCK();
719 	IN6_MULTI_UNLOCK();
720 
721 	return (0);
722 }
723 
724 /*
725  * Update the report timer on a group in response to an MLDv1 query.
726  *
727  * If we are becoming the reporting member for this group, start the timer.
728  * If we already are the reporting member for this group, and timer is
729  * below the threshold, reset it.
730  *
731  * We may be updating the group for the first time since we switched
732  * to MLDv2. If we are, then we must clear any recorded source lists,
733  * and transition to REPORTING state; the group timer is overloaded
734  * for group and group-source query responses.
735  *
736  * Unlike MLDv2, the delay per group should be jittered
737  * to avoid bursts of MLDv1 reports.
738  */
739 static void
740 mld_v1_update_group(struct in6_multi *inm, const int timer)
741 {
742 #ifdef KTR
743 	char			 ip6tbuf[INET6_ADDRSTRLEN];
744 #endif
745 
746 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
747 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
748 	    inm->in6m_ifp->if_xname, timer);
749 
750 	IN6_MULTI_LOCK_ASSERT();
751 
752 	switch (inm->in6m_state) {
753 	case MLD_NOT_MEMBER:
754 	case MLD_SILENT_MEMBER:
755 		break;
756 	case MLD_REPORTING_MEMBER:
757 		if (inm->in6m_timer != 0 &&
758 		    inm->in6m_timer <= timer) {
759 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
760 			    "skipping.", __func__);
761 			break;
762 		}
763 		/* FALLTHROUGH */
764 	case MLD_SG_QUERY_PENDING_MEMBER:
765 	case MLD_G_QUERY_PENDING_MEMBER:
766 	case MLD_IDLE_MEMBER:
767 	case MLD_LAZY_MEMBER:
768 	case MLD_AWAKENING_MEMBER:
769 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
770 		inm->in6m_state = MLD_REPORTING_MEMBER;
771 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
772 		V_current_state_timers_running6 = 1;
773 		break;
774 	case MLD_SLEEPING_MEMBER:
775 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
776 		inm->in6m_state = MLD_AWAKENING_MEMBER;
777 		break;
778 	case MLD_LEAVING_MEMBER:
779 		break;
780 	}
781 }
782 
783 /*
784  * Process a received MLDv2 general, group-specific or
785  * group-and-source-specific query.
786  *
787  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
788  *
789  * Return 0 if successful, otherwise an appropriate error code is returned.
790  */
791 static int
792 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
793     struct mbuf *m, const int off, const int icmp6len)
794 {
795 	struct mld_ifinfo	*mli;
796 	struct mldv2_query	*mld;
797 	struct in6_multi	*inm;
798 	uint32_t		 maxdelay, nsrc, qqi;
799 	int			 is_general_query;
800 	uint16_t		 timer;
801 	uint8_t			 qrv;
802 #ifdef KTR
803 	char			 ip6tbuf[INET6_ADDRSTRLEN];
804 #endif
805 
806 	is_general_query = 0;
807 
808 	/*
809 	 * RFC3810 Section 6.2: MLD queries must originate from
810 	 * a router's link-local address.
811 	 */
812 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
813 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
814 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
815 		    ifp, ifp->if_xname);
816 		return (0);
817 	}
818 
819 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname);
820 
821 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
822 
823 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
824 	if (maxdelay >= 32678) {
825 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
826 			   (MLD_MRC_EXP(maxdelay) + 3);
827 	}
828 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
829 	if (timer == 0)
830 		timer = 1;
831 
832 	qrv = MLD_QRV(mld->mld_misc);
833 	if (qrv < 2) {
834 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
835 		    qrv, MLD_RV_INIT);
836 		qrv = MLD_RV_INIT;
837 	}
838 
839 	qqi = mld->mld_qqi;
840 	if (qqi >= 128) {
841 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
842 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
843 	}
844 
845 	nsrc = ntohs(mld->mld_numsrc);
846 	if (nsrc > MLD_MAX_GS_SOURCES)
847 		return (EMSGSIZE);
848 	if (icmp6len < sizeof(struct mldv2_query) +
849 	    (nsrc * sizeof(struct in6_addr)))
850 		return (EMSGSIZE);
851 
852 	/*
853 	 * Do further input validation upfront to avoid resetting timers
854 	 * should we need to discard this query.
855 	 */
856 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
857 		/*
858 		 * General Queries SHOULD be directed to ff02::1.
859 		 * A general query with a source list has undefined
860 		 * behaviour; discard it.
861 		 */
862 		struct in6_addr		 dst;
863 
864 		dst = ip6->ip6_dst;
865 		in6_clearscope(&dst);
866 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) ||
867 		    nsrc > 0)
868 			return (EINVAL);
869 		is_general_query = 1;
870 	} else {
871 		/*
872 		 * Embed scope ID of receiving interface in MLD query for
873 		 * lookup whilst we don't hold other locks (due to KAME
874 		 * locking lameness). We own this mbuf chain just now.
875 		 */
876 		in6_setscope(&mld->mld_addr, ifp, NULL);
877 	}
878 
879 	IN6_MULTI_LOCK();
880 	MLD_LOCK();
881 	IF_ADDR_LOCK(ifp);
882 
883 	mli = MLD_IFINFO(ifp);
884 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
885 
886 	/*
887 	 * Discard the v2 query if we're in Compatibility Mode.
888 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
889 	 * until the Old Version Querier Present timer expires.
890 	 */
891 	if (mli->mli_version != MLD_VERSION_2)
892 		goto out_locked;
893 
894 	mld_set_version(mli, MLD_VERSION_2);
895 	mli->mli_rv = qrv;
896 	mli->mli_qi = qqi;
897 	mli->mli_qri = maxdelay;
898 
899 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
900 	    maxdelay);
901 
902 	if (is_general_query) {
903 		/*
904 		 * MLDv2 General Query.
905 		 *
906 		 * Schedule a current-state report on this ifp for
907 		 * all groups, possibly containing source lists.
908 		 *
909 		 * If there is a pending General Query response
910 		 * scheduled earlier than the selected delay, do
911 		 * not schedule any other reports.
912 		 * Otherwise, reset the interface timer.
913 		 */
914 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
915 		    ifp, ifp->if_xname);
916 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
917 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
918 			V_interface_timers_running6 = 1;
919 		}
920 	} else {
921 		/*
922 		 * MLDv2 Group-specific or Group-and-source-specific Query.
923 		 *
924 		 * Group-source-specific queries are throttled on
925 		 * a per-group basis to defeat denial-of-service attempts.
926 		 * Queries for groups we are not a member of on this
927 		 * link are simply ignored.
928 		 */
929 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
930 		if (inm == NULL)
931 			goto out_locked;
932 		if (nsrc > 0) {
933 			if (!ratecheck(&inm->in6m_lastgsrtv,
934 			    &V_mld_gsrdelay)) {
935 				CTR1(KTR_MLD, "%s: GS query throttled.",
936 				    __func__);
937 				goto out_locked;
938 			}
939 		}
940 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
941 		     ifp, ifp->if_xname);
942 		/*
943 		 * If there is a pending General Query response
944 		 * scheduled sooner than the selected delay, no
945 		 * further report need be scheduled.
946 		 * Otherwise, prepare to respond to the
947 		 * group-specific or group-and-source query.
948 		 */
949 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
950 			mld_v2_process_group_query(inm, mli, timer, m, off);
951 
952 		/* XXX Clear embedded scope ID as userland won't expect it. */
953 		in6_clearscope(&mld->mld_addr);
954 	}
955 
956 out_locked:
957 	IF_ADDR_UNLOCK(ifp);
958 	MLD_UNLOCK();
959 	IN6_MULTI_UNLOCK();
960 
961 	return (0);
962 }
963 
964 /*
965  * Process a recieved MLDv2 group-specific or group-and-source-specific
966  * query.
967  * Return <0 if any error occured. Currently this is ignored.
968  */
969 static int
970 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli,
971     int timer, struct mbuf *m0, const int off)
972 {
973 	struct mldv2_query	*mld;
974 	int			 retval;
975 	uint16_t		 nsrc;
976 
977 	IN6_MULTI_LOCK_ASSERT();
978 	MLD_LOCK_ASSERT();
979 
980 	retval = 0;
981 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
982 
983 	switch (inm->in6m_state) {
984 	case MLD_NOT_MEMBER:
985 	case MLD_SILENT_MEMBER:
986 	case MLD_SLEEPING_MEMBER:
987 	case MLD_LAZY_MEMBER:
988 	case MLD_AWAKENING_MEMBER:
989 	case MLD_IDLE_MEMBER:
990 	case MLD_LEAVING_MEMBER:
991 		return (retval);
992 		break;
993 	case MLD_REPORTING_MEMBER:
994 	case MLD_G_QUERY_PENDING_MEMBER:
995 	case MLD_SG_QUERY_PENDING_MEMBER:
996 		break;
997 	}
998 
999 	nsrc = ntohs(mld->mld_numsrc);
1000 
1001 	/*
1002 	 * Deal with group-specific queries upfront.
1003 	 * If any group query is already pending, purge any recorded
1004 	 * source-list state if it exists, and schedule a query response
1005 	 * for this group-specific query.
1006 	 */
1007 	if (nsrc == 0) {
1008 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1009 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1010 			in6m_clear_recorded(inm);
1011 			timer = min(inm->in6m_timer, timer);
1012 		}
1013 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1014 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1015 		V_current_state_timers_running6 = 1;
1016 		return (retval);
1017 	}
1018 
1019 	/*
1020 	 * Deal with the case where a group-and-source-specific query has
1021 	 * been received but a group-specific query is already pending.
1022 	 */
1023 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1024 		timer = min(inm->in6m_timer, timer);
1025 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1026 		V_current_state_timers_running6 = 1;
1027 		return (retval);
1028 	}
1029 
1030 	/*
1031 	 * Finally, deal with the case where a group-and-source-specific
1032 	 * query has been received, where a response to a previous g-s-r
1033 	 * query exists, or none exists.
1034 	 * In this case, we need to parse the source-list which the Querier
1035 	 * has provided us with and check if we have any source list filter
1036 	 * entries at T1 for these sources. If we do not, there is no need
1037 	 * schedule a report and the query may be dropped.
1038 	 * If we do, we must record them and schedule a current-state
1039 	 * report for those sources.
1040 	 */
1041 	if (inm->in6m_nsrc > 0) {
1042 		struct mbuf		*m;
1043 		uint8_t			*sp;
1044 		int			 i, nrecorded;
1045 		int			 soff;
1046 
1047 		m = m0;
1048 		soff = off + sizeof(struct mldv2_query);
1049 		nrecorded = 0;
1050 		for (i = 0; i < nsrc; i++) {
1051 			sp = mtod(m, uint8_t *) + soff;
1052 			retval = in6m_record_source(inm,
1053 			    (const struct in6_addr *)sp);
1054 			if (retval < 0)
1055 				break;
1056 			nrecorded += retval;
1057 			soff += sizeof(struct in6_addr);
1058 			if (soff >= m->m_len) {
1059 				soff = soff - m->m_len;
1060 				m = m->m_next;
1061 				if (m == NULL)
1062 					break;
1063 			}
1064 		}
1065 		if (nrecorded > 0) {
1066 			CTR1(KTR_MLD,
1067 			    "%s: schedule response to SG query", __func__);
1068 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1069 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1070 			V_current_state_timers_running6 = 1;
1071 		}
1072 	}
1073 
1074 	return (retval);
1075 }
1076 
1077 /*
1078  * Process a received MLDv1 host membership report.
1079  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1080  *
1081  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1082  * mld_addr. This is OK as we own the mbuf chain.
1083  */
1084 static int
1085 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1086     /*const*/ struct mld_hdr *mld)
1087 {
1088 	struct in6_addr		 src, dst;
1089 	struct in6_ifaddr	*ia;
1090 	struct in6_multi	*inm;
1091 #ifdef KTR
1092 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1093 #endif
1094 
1095 	if (!mld_v1enable) {
1096 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1097 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1098 		    ifp, ifp->if_xname);
1099 		return (0);
1100 	}
1101 
1102 	if (ifp->if_flags & IFF_LOOPBACK)
1103 		return (0);
1104 
1105 	/*
1106 	 * MLDv1 reports must originate from a host's link-local address,
1107 	 * or the unspecified address (when booting).
1108 	 */
1109 	src = ip6->ip6_src;
1110 	in6_clearscope(&src);
1111 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1112 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1113 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1114 		    ifp, ifp->if_xname);
1115 		return (EINVAL);
1116 	}
1117 
1118 	/*
1119 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1120 	 * group, and must be directed to the group itself.
1121 	 */
1122 	dst = ip6->ip6_dst;
1123 	in6_clearscope(&dst);
1124 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1125 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1126 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1127 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1128 		    ifp, ifp->if_xname);
1129 		return (EINVAL);
1130 	}
1131 
1132 	/*
1133 	 * Make sure we don't hear our own membership report, as fast
1134 	 * leave requires knowing that we are the only member of a
1135 	 * group. Assume we used the link-local address if available,
1136 	 * otherwise look for ::.
1137 	 *
1138 	 * XXX Note that scope ID comparison is needed for the address
1139 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1140 	 * performed for the on-wire address.
1141 	 */
1142 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1143 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1144 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1145 		if (ia != NULL)
1146 			ifa_free(&ia->ia_ifa);
1147 		return (0);
1148 	}
1149 	if (ia != NULL)
1150 		ifa_free(&ia->ia_ifa);
1151 
1152 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1153 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname);
1154 
1155 	/*
1156 	 * Embed scope ID of receiving interface in MLD query for lookup
1157 	 * whilst we don't hold other locks (due to KAME locking lameness).
1158 	 */
1159 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1160 		in6_setscope(&mld->mld_addr, ifp, NULL);
1161 
1162 	IN6_MULTI_LOCK();
1163 	MLD_LOCK();
1164 	IF_ADDR_LOCK(ifp);
1165 
1166 	/*
1167 	 * MLDv1 report suppression.
1168 	 * If we are a member of this group, and our membership should be
1169 	 * reported, and our group timer is pending or about to be reset,
1170 	 * stop our group timer by transitioning to the 'lazy' state.
1171 	 */
1172 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1173 	if (inm != NULL) {
1174 		struct mld_ifinfo *mli;
1175 
1176 		mli = inm->in6m_mli;
1177 		KASSERT(mli != NULL,
1178 		    ("%s: no mli for ifp %p", __func__, ifp));
1179 
1180 		/*
1181 		 * If we are in MLDv2 host mode, do not allow the
1182 		 * other host's MLDv1 report to suppress our reports.
1183 		 */
1184 		if (mli->mli_version == MLD_VERSION_2)
1185 			goto out_locked;
1186 
1187 		inm->in6m_timer = 0;
1188 
1189 		switch (inm->in6m_state) {
1190 		case MLD_NOT_MEMBER:
1191 		case MLD_SILENT_MEMBER:
1192 		case MLD_SLEEPING_MEMBER:
1193 			break;
1194 		case MLD_REPORTING_MEMBER:
1195 		case MLD_IDLE_MEMBER:
1196 		case MLD_AWAKENING_MEMBER:
1197 			CTR3(KTR_MLD,
1198 			    "report suppressed for %s on ifp %p(%s)",
1199 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1200 			    ifp, ifp->if_xname);
1201 		case MLD_LAZY_MEMBER:
1202 			inm->in6m_state = MLD_LAZY_MEMBER;
1203 			break;
1204 		case MLD_G_QUERY_PENDING_MEMBER:
1205 		case MLD_SG_QUERY_PENDING_MEMBER:
1206 		case MLD_LEAVING_MEMBER:
1207 			break;
1208 		}
1209 	}
1210 
1211 out_locked:
1212 	MLD_UNLOCK();
1213 	IF_ADDR_UNLOCK(ifp);
1214 	IN6_MULTI_UNLOCK();
1215 
1216 	/* XXX Clear embedded scope ID as userland won't expect it. */
1217 	in6_clearscope(&mld->mld_addr);
1218 
1219 	return (0);
1220 }
1221 
1222 /*
1223  * MLD input path.
1224  *
1225  * Assume query messages which fit in a single ICMPv6 message header
1226  * have been pulled up.
1227  * Assume that userland will want to see the message, even if it
1228  * otherwise fails kernel input validation; do not free it.
1229  * Pullup may however free the mbuf chain m if it fails.
1230  *
1231  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1232  */
1233 int
1234 mld_input(struct mbuf *m, int off, int icmp6len)
1235 {
1236 	struct ifnet	*ifp;
1237 	struct ip6_hdr	*ip6;
1238 	struct mld_hdr	*mld;
1239 	int		 mldlen;
1240 
1241 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1242 
1243 	ifp = m->m_pkthdr.rcvif;
1244 
1245 	ip6 = mtod(m, struct ip6_hdr *);
1246 
1247 	/* Pullup to appropriate size. */
1248 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1249 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1250 	    icmp6len >= sizeof(struct mldv2_query)) {
1251 		mldlen = sizeof(struct mldv2_query);
1252 	} else {
1253 		mldlen = sizeof(struct mld_hdr);
1254 	}
1255 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1256 	if (mld == NULL) {
1257 		ICMP6STAT_INC(icp6s_badlen);
1258 		return (IPPROTO_DONE);
1259 	}
1260 
1261 	/*
1262 	 * Userland needs to see all of this traffic for implementing
1263 	 * the endpoint discovery portion of multicast routing.
1264 	 */
1265 	switch (mld->mld_type) {
1266 	case MLD_LISTENER_QUERY:
1267 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1268 		if (icmp6len == sizeof(struct mld_hdr)) {
1269 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1270 				return (0);
1271 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1272 			if (mld_v2_input_query(ifp, ip6, m, off,
1273 			    icmp6len) != 0)
1274 				return (0);
1275 		}
1276 		break;
1277 	case MLD_LISTENER_REPORT:
1278 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1279 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1280 			return (0);
1281 		break;
1282 	case MLDV2_LISTENER_REPORT:
1283 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1284 		break;
1285 	case MLD_LISTENER_DONE:
1286 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1287 		break;
1288 	default:
1289 		break;
1290 	}
1291 
1292 	return (0);
1293 }
1294 
1295 /*
1296  * Fast timeout handler (global).
1297  * VIMAGE: Timeout handlers are expected to service all vimages.
1298  */
1299 void
1300 mld_fasttimo(void)
1301 {
1302 	VNET_ITERATOR_DECL(vnet_iter);
1303 
1304 	VNET_LIST_RLOCK_NOSLEEP();
1305 	VNET_FOREACH(vnet_iter) {
1306 		CURVNET_SET(vnet_iter);
1307 		mld_fasttimo_vnet();
1308 		CURVNET_RESTORE();
1309 	}
1310 	VNET_LIST_RUNLOCK_NOSLEEP();
1311 }
1312 
1313 /*
1314  * Fast timeout handler (per-vnet).
1315  *
1316  * VIMAGE: Assume caller has set up our curvnet.
1317  */
1318 static void
1319 mld_fasttimo_vnet(void)
1320 {
1321 	struct ifqueue		 scq;	/* State-change packets */
1322 	struct ifqueue		 qrq;	/* Query response packets */
1323 	struct ifnet		*ifp;
1324 	struct mld_ifinfo	*mli;
1325 	struct ifmultiaddr	*ifma, *tifma;
1326 	struct in6_multi	*inm;
1327 	int			 uri_fasthz;
1328 
1329 	uri_fasthz = 0;
1330 
1331 	/*
1332 	 * Quick check to see if any work needs to be done, in order to
1333 	 * minimize the overhead of fasttimo processing.
1334 	 * SMPng: XXX Unlocked reads.
1335 	 */
1336 	if (!V_current_state_timers_running6 &&
1337 	    !V_interface_timers_running6 &&
1338 	    !V_state_change_timers_running6)
1339 		return;
1340 
1341 	IN6_MULTI_LOCK();
1342 	MLD_LOCK();
1343 
1344 	/*
1345 	 * MLDv2 General Query response timer processing.
1346 	 */
1347 	if (V_interface_timers_running6) {
1348 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1349 
1350 		V_interface_timers_running6 = 0;
1351 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1352 			if (mli->mli_v2_timer == 0) {
1353 				/* Do nothing. */
1354 			} else if (--mli->mli_v2_timer == 0) {
1355 				mld_v2_dispatch_general_query(mli);
1356 			} else {
1357 				V_interface_timers_running6 = 1;
1358 			}
1359 		}
1360 	}
1361 
1362 	if (!V_current_state_timers_running6 &&
1363 	    !V_state_change_timers_running6)
1364 		goto out_locked;
1365 
1366 	V_current_state_timers_running6 = 0;
1367 	V_state_change_timers_running6 = 0;
1368 
1369 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1370 
1371 	/*
1372 	 * MLD host report and state-change timer processing.
1373 	 * Note: Processing a v2 group timer may remove a node.
1374 	 */
1375 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1376 		ifp = mli->mli_ifp;
1377 
1378 		if (mli->mli_version == MLD_VERSION_2) {
1379 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1380 			    PR_FASTHZ);
1381 
1382 			memset(&qrq, 0, sizeof(struct ifqueue));
1383 			IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS);
1384 
1385 			memset(&scq, 0, sizeof(struct ifqueue));
1386 			IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1387 		}
1388 
1389 		IF_ADDR_LOCK(ifp);
1390 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1391 		    tifma) {
1392 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1393 			    ifma->ifma_protospec == NULL)
1394 				continue;
1395 			inm = (struct in6_multi *)ifma->ifma_protospec;
1396 			switch (mli->mli_version) {
1397 			case MLD_VERSION_1:
1398 				/*
1399 				 * XXX Drop IF_ADDR lock temporarily to
1400 				 * avoid recursion caused by a potential
1401 				 * call by in6ifa_ifpforlinklocal().
1402 				 * rwlock candidate?
1403 				 */
1404 				IF_ADDR_UNLOCK(ifp);
1405 				mld_v1_process_group_timer(inm,
1406 				    mli->mli_version);
1407 				IF_ADDR_LOCK(ifp);
1408 				break;
1409 			case MLD_VERSION_2:
1410 				mld_v2_process_group_timers(mli, &qrq,
1411 				    &scq, inm, uri_fasthz);
1412 				break;
1413 			}
1414 		}
1415 		IF_ADDR_UNLOCK(ifp);
1416 
1417 		if (mli->mli_version == MLD_VERSION_2) {
1418 			struct in6_multi		*tinm;
1419 
1420 			mld_dispatch_queue(&qrq, 0);
1421 			mld_dispatch_queue(&scq, 0);
1422 
1423 			/*
1424 			 * Free the in_multi reference(s) for
1425 			 * this lifecycle.
1426 			 */
1427 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1428 			    in6m_nrele, tinm) {
1429 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1430 				    in6m_nrele);
1431 				in6m_release_locked(inm);
1432 			}
1433 		}
1434 	}
1435 
1436 out_locked:
1437 	MLD_UNLOCK();
1438 	IN6_MULTI_UNLOCK();
1439 }
1440 
1441 /*
1442  * Update host report group timer.
1443  * Will update the global pending timer flags.
1444  */
1445 static void
1446 mld_v1_process_group_timer(struct in6_multi *inm, const int version)
1447 {
1448 	int report_timer_expired;
1449 
1450 	IN6_MULTI_LOCK_ASSERT();
1451 	MLD_LOCK_ASSERT();
1452 
1453 	if (inm->in6m_timer == 0) {
1454 		report_timer_expired = 0;
1455 	} else if (--inm->in6m_timer == 0) {
1456 		report_timer_expired = 1;
1457 	} else {
1458 		V_current_state_timers_running6 = 1;
1459 		return;
1460 	}
1461 
1462 	switch (inm->in6m_state) {
1463 	case MLD_NOT_MEMBER:
1464 	case MLD_SILENT_MEMBER:
1465 	case MLD_IDLE_MEMBER:
1466 	case MLD_LAZY_MEMBER:
1467 	case MLD_SLEEPING_MEMBER:
1468 	case MLD_AWAKENING_MEMBER:
1469 		break;
1470 	case MLD_REPORTING_MEMBER:
1471 		if (report_timer_expired) {
1472 			inm->in6m_state = MLD_IDLE_MEMBER;
1473 			(void)mld_v1_transmit_report(inm,
1474 			     MLD_LISTENER_REPORT);
1475 		}
1476 		break;
1477 	case MLD_G_QUERY_PENDING_MEMBER:
1478 	case MLD_SG_QUERY_PENDING_MEMBER:
1479 	case MLD_LEAVING_MEMBER:
1480 		break;
1481 	}
1482 }
1483 
1484 /*
1485  * Update a group's timers for MLDv2.
1486  * Will update the global pending timer flags.
1487  * Note: Unlocked read from mli.
1488  */
1489 static void
1490 mld_v2_process_group_timers(struct mld_ifinfo *mli,
1491     struct ifqueue *qrq, struct ifqueue *scq,
1492     struct in6_multi *inm, const int uri_fasthz)
1493 {
1494 	int query_response_timer_expired;
1495 	int state_change_retransmit_timer_expired;
1496 #ifdef KTR
1497 	char ip6tbuf[INET6_ADDRSTRLEN];
1498 #endif
1499 
1500 	IN6_MULTI_LOCK_ASSERT();
1501 	MLD_LOCK_ASSERT();
1502 
1503 	query_response_timer_expired = 0;
1504 	state_change_retransmit_timer_expired = 0;
1505 
1506 	/*
1507 	 * During a transition from compatibility mode back to MLDv2,
1508 	 * a group record in REPORTING state may still have its group
1509 	 * timer active. This is a no-op in this function; it is easier
1510 	 * to deal with it here than to complicate the slow-timeout path.
1511 	 */
1512 	if (inm->in6m_timer == 0) {
1513 		query_response_timer_expired = 0;
1514 	} else if (--inm->in6m_timer == 0) {
1515 		query_response_timer_expired = 1;
1516 	} else {
1517 		V_current_state_timers_running6 = 1;
1518 	}
1519 
1520 	if (inm->in6m_sctimer == 0) {
1521 		state_change_retransmit_timer_expired = 0;
1522 	} else if (--inm->in6m_sctimer == 0) {
1523 		state_change_retransmit_timer_expired = 1;
1524 	} else {
1525 		V_state_change_timers_running6 = 1;
1526 	}
1527 
1528 	/* We are in fasttimo, so be quick about it. */
1529 	if (!state_change_retransmit_timer_expired &&
1530 	    !query_response_timer_expired)
1531 		return;
1532 
1533 	switch (inm->in6m_state) {
1534 	case MLD_NOT_MEMBER:
1535 	case MLD_SILENT_MEMBER:
1536 	case MLD_SLEEPING_MEMBER:
1537 	case MLD_LAZY_MEMBER:
1538 	case MLD_AWAKENING_MEMBER:
1539 	case MLD_IDLE_MEMBER:
1540 		break;
1541 	case MLD_G_QUERY_PENDING_MEMBER:
1542 	case MLD_SG_QUERY_PENDING_MEMBER:
1543 		/*
1544 		 * Respond to a previously pending Group-Specific
1545 		 * or Group-and-Source-Specific query by enqueueing
1546 		 * the appropriate Current-State report for
1547 		 * immediate transmission.
1548 		 */
1549 		if (query_response_timer_expired) {
1550 			int retval;
1551 
1552 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1553 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER));
1554 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1555 			    __func__, retval);
1556 			inm->in6m_state = MLD_REPORTING_MEMBER;
1557 			in6m_clear_recorded(inm);
1558 		}
1559 		/* FALLTHROUGH */
1560 	case MLD_REPORTING_MEMBER:
1561 	case MLD_LEAVING_MEMBER:
1562 		if (state_change_retransmit_timer_expired) {
1563 			/*
1564 			 * State-change retransmission timer fired.
1565 			 * If there are any further pending retransmissions,
1566 			 * set the global pending state-change flag, and
1567 			 * reset the timer.
1568 			 */
1569 			if (--inm->in6m_scrv > 0) {
1570 				inm->in6m_sctimer = uri_fasthz;
1571 				V_state_change_timers_running6 = 1;
1572 			}
1573 			/*
1574 			 * Retransmit the previously computed state-change
1575 			 * report. If there are no further pending
1576 			 * retransmissions, the mbuf queue will be consumed.
1577 			 * Update T0 state to T1 as we have now sent
1578 			 * a state-change.
1579 			 */
1580 			(void)mld_v2_merge_state_changes(inm, scq);
1581 
1582 			in6m_commit(inm);
1583 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1584 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1585 			    inm->in6m_ifp->if_xname);
1586 
1587 			/*
1588 			 * If we are leaving the group for good, make sure
1589 			 * we release MLD's reference to it.
1590 			 * This release must be deferred using a SLIST,
1591 			 * as we are called from a loop which traverses
1592 			 * the in_ifmultiaddr TAILQ.
1593 			 */
1594 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1595 			    inm->in6m_scrv == 0) {
1596 				inm->in6m_state = MLD_NOT_MEMBER;
1597 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1598 				    inm, in6m_nrele);
1599 			}
1600 		}
1601 		break;
1602 	}
1603 }
1604 
1605 /*
1606  * Switch to a different version on the given interface,
1607  * as per Section 9.12.
1608  */
1609 static void
1610 mld_set_version(struct mld_ifinfo *mli, const int version)
1611 {
1612 	int old_version_timer;
1613 
1614 	MLD_LOCK_ASSERT();
1615 
1616 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1617 	    version, mli->mli_ifp, mli->mli_ifp->if_xname);
1618 
1619 	if (version == MLD_VERSION_1) {
1620 		/*
1621 		 * Compute the "Older Version Querier Present" timer as per
1622 		 * Section 9.12.
1623 		 */
1624 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1625 		old_version_timer *= PR_SLOWHZ;
1626 		mli->mli_v1_timer = old_version_timer;
1627 	}
1628 
1629 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1630 		mli->mli_version = MLD_VERSION_1;
1631 		mld_v2_cancel_link_timers(mli);
1632 	}
1633 }
1634 
1635 /*
1636  * Cancel pending MLDv2 timers for the given link and all groups
1637  * joined on it; state-change, general-query, and group-query timers.
1638  */
1639 static void
1640 mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
1641 {
1642 	struct ifmultiaddr	*ifma;
1643 	struct ifnet		*ifp;
1644 	struct in6_multi		*inm;
1645 
1646 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1647 	    mli->mli_ifp, mli->mli_ifp->if_xname);
1648 
1649 	IN6_MULTI_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_LOCK(ifp);
1666 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1667 		if (ifma->ifma_addr->sa_family != AF_INET6)
1668 			continue;
1669 		inm = (struct in6_multi *)ifma->ifma_protospec;
1670 		switch (inm->in6m_state) {
1671 		case MLD_NOT_MEMBER:
1672 		case MLD_SILENT_MEMBER:
1673 		case MLD_IDLE_MEMBER:
1674 		case MLD_LAZY_MEMBER:
1675 		case MLD_SLEEPING_MEMBER:
1676 		case MLD_AWAKENING_MEMBER:
1677 			break;
1678 		case MLD_LEAVING_MEMBER:
1679 			/*
1680 			 * If we are leaving the group and switching
1681 			 * version, we need to release the final
1682 			 * reference held for issuing the INCLUDE {}.
1683 			 *
1684 			 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
1685 			 * around in6m_release_locked(), as it is not
1686 			 * a recursive mutex.
1687 			 */
1688 			IF_ADDR_UNLOCK(ifp);
1689 			in6m_release_locked(inm);
1690 			IF_ADDR_LOCK(ifp);
1691 			/* FALLTHROUGH */
1692 		case MLD_G_QUERY_PENDING_MEMBER:
1693 		case MLD_SG_QUERY_PENDING_MEMBER:
1694 			in6m_clear_recorded(inm);
1695 			/* FALLTHROUGH */
1696 		case MLD_REPORTING_MEMBER:
1697 			inm->in6m_sctimer = 0;
1698 			inm->in6m_timer = 0;
1699 			inm->in6m_state = MLD_REPORTING_MEMBER;
1700 			/*
1701 			 * Free any pending MLDv2 state-change records.
1702 			 */
1703 			_IF_DRAIN(&inm->in6m_scq);
1704 			break;
1705 		}
1706 	}
1707 	IF_ADDR_UNLOCK(ifp);
1708 }
1709 
1710 /*
1711  * Global slowtimo handler.
1712  * VIMAGE: Timeout handlers are expected to service all vimages.
1713  */
1714 void
1715 mld_slowtimo(void)
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 
1728 /*
1729  * Per-vnet slowtimo handler.
1730  */
1731 static void
1732 mld_slowtimo_vnet(void)
1733 {
1734 	struct mld_ifinfo *mli;
1735 
1736 	MLD_LOCK();
1737 
1738 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1739 		mld_v1_process_querier_timers(mli);
1740 	}
1741 
1742 	MLD_UNLOCK();
1743 }
1744 
1745 /*
1746  * Update the Older Version Querier Present timers for a link.
1747  * See Section 9.12 of RFC 3810.
1748  */
1749 static void
1750 mld_v1_process_querier_timers(struct mld_ifinfo *mli)
1751 {
1752 
1753 	MLD_LOCK_ASSERT();
1754 
1755 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1756 		/*
1757 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1758 		 */
1759 		CTR5(KTR_MLD,
1760 		    "%s: transition from v%d -> v%d on %p(%s)",
1761 		    __func__, mli->mli_version, MLD_VERSION_2,
1762 		    mli->mli_ifp, mli->mli_ifp->if_xname);
1763 		mli->mli_version = MLD_VERSION_2;
1764 	}
1765 }
1766 
1767 /*
1768  * Transmit an MLDv1 report immediately.
1769  */
1770 static int
1771 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1772 {
1773 	struct ifnet		*ifp;
1774 	struct in6_ifaddr	*ia;
1775 	struct ip6_hdr		*ip6;
1776 	struct mbuf		*mh, *md;
1777 	struct mld_hdr		*mld;
1778 
1779 	IN6_MULTI_LOCK_ASSERT();
1780 	MLD_LOCK_ASSERT();
1781 
1782 	ifp = in6m->in6m_ifp;
1783 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1784 	/* ia may be NULL if link-local address is tentative. */
1785 
1786 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
1787 	if (mh == NULL) {
1788 		if (ia != NULL)
1789 			ifa_free(&ia->ia_ifa);
1790 		return (ENOMEM);
1791 	}
1792 	MGET(md, M_DONTWAIT, MT_DATA);
1793 	if (md == NULL) {
1794 		m_free(mh);
1795 		if (ia != NULL)
1796 			ifa_free(&ia->ia_ifa);
1797 		return (ENOMEM);
1798 	}
1799 	mh->m_next = md;
1800 
1801 	/*
1802 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1803 	 * that ether_output() does not need to allocate another mbuf
1804 	 * for the header in the most common case.
1805 	 */
1806 	MH_ALIGN(mh, sizeof(struct ip6_hdr));
1807 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1808 	mh->m_len = sizeof(struct ip6_hdr);
1809 
1810 	ip6 = mtod(mh, struct ip6_hdr *);
1811 	ip6->ip6_flow = 0;
1812 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1813 	ip6->ip6_vfc |= IPV6_VERSION;
1814 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1815 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1816 	ip6->ip6_dst = in6m->in6m_addr;
1817 
1818 	md->m_len = sizeof(struct mld_hdr);
1819 	mld = mtod(md, struct mld_hdr *);
1820 	mld->mld_type = type;
1821 	mld->mld_code = 0;
1822 	mld->mld_cksum = 0;
1823 	mld->mld_maxdelay = 0;
1824 	mld->mld_reserved = 0;
1825 	mld->mld_addr = in6m->in6m_addr;
1826 	in6_clearscope(&mld->mld_addr);
1827 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1828 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1829 
1830 	mld_save_context(mh, ifp);
1831 	mh->m_flags |= M_MLDV1;
1832 
1833 	mld_dispatch_packet(mh);
1834 
1835 	if (ia != NULL)
1836 		ifa_free(&ia->ia_ifa);
1837 	return (0);
1838 }
1839 
1840 /*
1841  * Process a state change from the upper layer for the given IPv6 group.
1842  *
1843  * Each socket holds a reference on the in_multi in its own ip_moptions.
1844  * The socket layer will have made the necessary updates to.the group
1845  * state, it is now up to MLD to issue a state change report if there
1846  * has been any change between T0 (when the last state-change was issued)
1847  * and T1 (now).
1848  *
1849  * We use the MLDv2 state machine at group level. The MLd module
1850  * however makes the decision as to which MLD protocol version to speak.
1851  * A state change *from* INCLUDE {} always means an initial join.
1852  * A state change *to* INCLUDE {} always means a final leave.
1853  *
1854  * If delay is non-zero, and the state change is an initial multicast
1855  * join, the state change report will be delayed by 'delay' ticks
1856  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1857  * the initial MLDv2 state change report will be delayed by whichever
1858  * is sooner, a pending state-change timer or delay itself.
1859  *
1860  * VIMAGE: curvnet should have been set by caller, as this routine
1861  * is called from the socket option handlers.
1862  */
1863 int
1864 mld_change_state(struct in6_multi *inm, const int delay)
1865 {
1866 	struct mld_ifinfo *mli;
1867 	struct ifnet *ifp;
1868 	int error;
1869 
1870 	IN6_MULTI_LOCK_ASSERT();
1871 
1872 	error = 0;
1873 
1874 	/*
1875 	 * Try to detect if the upper layer just asked us to change state
1876 	 * for an interface which has now gone away.
1877 	 */
1878 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1879 	ifp = inm->in6m_ifma->ifma_ifp;
1880 	if (ifp != NULL) {
1881 		/*
1882 		 * Sanity check that netinet6's notion of ifp is the
1883 		 * same as net's.
1884 		 */
1885 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1886 	}
1887 
1888 	MLD_LOCK();
1889 
1890 	mli = MLD_IFINFO(ifp);
1891 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
1892 
1893 	/*
1894 	 * If we detect a state transition to or from MCAST_UNDEFINED
1895 	 * for this group, then we are starting or finishing an MLD
1896 	 * life cycle for this group.
1897 	 */
1898 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1899 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1900 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1901 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1902 			CTR1(KTR_MLD, "%s: initial join", __func__);
1903 			error = mld_initial_join(inm, mli, delay);
1904 			goto out_locked;
1905 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1906 			CTR1(KTR_MLD, "%s: final leave", __func__);
1907 			mld_final_leave(inm, mli);
1908 			goto out_locked;
1909 		}
1910 	} else {
1911 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1912 	}
1913 
1914 	error = mld_handle_state_change(inm, mli);
1915 
1916 out_locked:
1917 	MLD_UNLOCK();
1918 	return (error);
1919 }
1920 
1921 /*
1922  * Perform the initial join for an MLD group.
1923  *
1924  * When joining a group:
1925  *  If the group should have its MLD traffic suppressed, do nothing.
1926  *  MLDv1 starts sending MLDv1 host membership reports.
1927  *  MLDv2 will schedule an MLDv2 state-change report containing the
1928  *  initial state of the membership.
1929  *
1930  * If the delay argument is non-zero, then we must delay sending the
1931  * initial state change for delay ticks (in units of PR_FASTHZ).
1932  */
1933 static int
1934 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
1935     const int delay)
1936 {
1937 	struct ifnet		*ifp;
1938 	struct ifqueue		*ifq;
1939 	int			 error, retval, syncstates;
1940 	int			 odelay;
1941 #ifdef KTR
1942 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1943 #endif
1944 
1945 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1946 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1947 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
1948 
1949 	error = 0;
1950 	syncstates = 1;
1951 
1952 	ifp = inm->in6m_ifp;
1953 
1954 	IN6_MULTI_LOCK_ASSERT();
1955 	MLD_LOCK_ASSERT();
1956 
1957 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1958 
1959 	/*
1960 	 * Groups joined on loopback or marked as 'not reported',
1961 	 * enter the MLD_SILENT_MEMBER state and
1962 	 * are never reported in any protocol exchanges.
1963 	 * All other groups enter the appropriate state machine
1964 	 * for the version in use on this link.
1965 	 * A link marked as MLIF_SILENT causes MLD to be completely
1966 	 * disabled for the link.
1967 	 */
1968 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1969 	    (mli->mli_flags & MLIF_SILENT) ||
1970 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1971 		CTR1(KTR_MLD,
1972 "%s: not kicking state machine for silent group", __func__);
1973 		inm->in6m_state = MLD_SILENT_MEMBER;
1974 		inm->in6m_timer = 0;
1975 	} else {
1976 		/*
1977 		 * Deal with overlapping in_multi lifecycle.
1978 		 * If this group was LEAVING, then make sure
1979 		 * we drop the reference we picked up to keep the
1980 		 * group around for the final INCLUDE {} enqueue.
1981 		 */
1982 		if (mli->mli_version == MLD_VERSION_2 &&
1983 		    inm->in6m_state == MLD_LEAVING_MEMBER)
1984 			in6m_release_locked(inm);
1985 
1986 		inm->in6m_state = MLD_REPORTING_MEMBER;
1987 
1988 		switch (mli->mli_version) {
1989 		case MLD_VERSION_1:
1990 			/*
1991 			 * If a delay was provided, only use it if
1992 			 * it is greater than the delay normally
1993 			 * used for an MLDv1 state change report,
1994 			 * and delay sending the initial MLDv1 report
1995 			 * by not transitioning to the IDLE state.
1996 			 */
1997 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
1998 			if (delay) {
1999 				inm->in6m_timer = max(delay, odelay);
2000 				V_current_state_timers_running6 = 1;
2001 			} else {
2002 				inm->in6m_state = MLD_IDLE_MEMBER;
2003 				error = mld_v1_transmit_report(inm,
2004 				     MLD_LISTENER_REPORT);
2005 				if (error == 0) {
2006 					inm->in6m_timer = odelay;
2007 					V_current_state_timers_running6 = 1;
2008 				}
2009 			}
2010 			break;
2011 
2012 		case MLD_VERSION_2:
2013 			/*
2014 			 * Defer update of T0 to T1, until the first copy
2015 			 * of the state change has been transmitted.
2016 			 */
2017 			syncstates = 0;
2018 
2019 			/*
2020 			 * Immediately enqueue a State-Change Report for
2021 			 * this interface, freeing any previous reports.
2022 			 * Don't kick the timers if there is nothing to do,
2023 			 * or if an error occurred.
2024 			 */
2025 			ifq = &inm->in6m_scq;
2026 			_IF_DRAIN(ifq);
2027 			retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2028 			    0, 0);
2029 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2030 			    __func__, retval);
2031 			if (retval <= 0) {
2032 				error = retval * -1;
2033 				break;
2034 			}
2035 
2036 			/*
2037 			 * Schedule transmission of pending state-change
2038 			 * report up to RV times for this link. The timer
2039 			 * will fire at the next mld_fasttimo (~200ms),
2040 			 * giving us an opportunity to merge the reports.
2041 			 *
2042 			 * If a delay was provided to this function, only
2043 			 * use this delay if sooner than the existing one.
2044 			 */
2045 			KASSERT(mli->mli_rv > 1,
2046 			   ("%s: invalid robustness %d", __func__,
2047 			    mli->mli_rv));
2048 			inm->in6m_scrv = mli->mli_rv;
2049 			if (delay) {
2050 				if (inm->in6m_sctimer > 1) {
2051 					inm->in6m_sctimer =
2052 					    min(inm->in6m_sctimer, delay);
2053 				} else
2054 					inm->in6m_sctimer = delay;
2055 			} else
2056 				inm->in6m_sctimer = 1;
2057 			V_state_change_timers_running6 = 1;
2058 
2059 			error = 0;
2060 			break;
2061 		}
2062 	}
2063 
2064 	/*
2065 	 * Only update the T0 state if state change is atomic,
2066 	 * i.e. we don't need to wait for a timer to fire before we
2067 	 * can consider the state change to have been communicated.
2068 	 */
2069 	if (syncstates) {
2070 		in6m_commit(inm);
2071 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2072 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2073 		    inm->in6m_ifp->if_xname);
2074 	}
2075 
2076 	return (error);
2077 }
2078 
2079 /*
2080  * Issue an intermediate state change during the life-cycle.
2081  */
2082 static int
2083 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli)
2084 {
2085 	struct ifnet		*ifp;
2086 	int			 retval;
2087 #ifdef KTR
2088 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2089 #endif
2090 
2091 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2092 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2093 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
2094 
2095 	ifp = inm->in6m_ifp;
2096 
2097 	IN6_MULTI_LOCK_ASSERT();
2098 	MLD_LOCK_ASSERT();
2099 
2100 	KASSERT(mli && mli->mli_ifp == ifp,
2101 	    ("%s: inconsistent ifp", __func__));
2102 
2103 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2104 	    (mli->mli_flags & MLIF_SILENT) ||
2105 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2106 	    (mli->mli_version != MLD_VERSION_2)) {
2107 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2108 			CTR1(KTR_MLD,
2109 "%s: not kicking state machine for silent group", __func__);
2110 		}
2111 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2112 		in6m_commit(inm);
2113 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2114 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2115 		    inm->in6m_ifp->if_xname);
2116 		return (0);
2117 	}
2118 
2119 	_IF_DRAIN(&inm->in6m_scq);
2120 
2121 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0);
2122 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2123 	if (retval <= 0)
2124 		return (-retval);
2125 
2126 	/*
2127 	 * If record(s) were enqueued, start the state-change
2128 	 * report timer for this group.
2129 	 */
2130 	inm->in6m_scrv = mli->mli_rv;
2131 	inm->in6m_sctimer = 1;
2132 	V_state_change_timers_running6 = 1;
2133 
2134 	return (0);
2135 }
2136 
2137 /*
2138  * Perform the final leave for a multicast address.
2139  *
2140  * When leaving a group:
2141  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2142  *  MLDv2 enqueues a state-change report containing a transition
2143  *  to INCLUDE {} for immediate transmission.
2144  */
2145 static void
2146 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli)
2147 {
2148 	int syncstates;
2149 #ifdef KTR
2150 	char ip6tbuf[INET6_ADDRSTRLEN];
2151 #endif
2152 
2153 	syncstates = 1;
2154 
2155 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2156 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2157 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
2158 
2159 	IN6_MULTI_LOCK_ASSERT();
2160 	MLD_LOCK_ASSERT();
2161 
2162 	switch (inm->in6m_state) {
2163 	case MLD_NOT_MEMBER:
2164 	case MLD_SILENT_MEMBER:
2165 	case MLD_LEAVING_MEMBER:
2166 		/* Already leaving or left; do nothing. */
2167 		CTR1(KTR_MLD,
2168 "%s: not kicking state machine for silent group", __func__);
2169 		break;
2170 	case MLD_REPORTING_MEMBER:
2171 	case MLD_IDLE_MEMBER:
2172 	case MLD_G_QUERY_PENDING_MEMBER:
2173 	case MLD_SG_QUERY_PENDING_MEMBER:
2174 		if (mli->mli_version == MLD_VERSION_1) {
2175 #ifdef INVARIANTS
2176 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2177 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2178 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2179 			     __func__);
2180 #endif
2181 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2182 			inm->in6m_state = MLD_NOT_MEMBER;
2183 		} else if (mli->mli_version == MLD_VERSION_2) {
2184 			/*
2185 			 * Stop group timer and all pending reports.
2186 			 * Immediately enqueue a state-change report
2187 			 * TO_IN {} to be sent on the next fast timeout,
2188 			 * giving us an opportunity to merge reports.
2189 			 */
2190 			_IF_DRAIN(&inm->in6m_scq);
2191 			inm->in6m_timer = 0;
2192 			inm->in6m_scrv = mli->mli_rv;
2193 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2194 			    "pending retransmissions.", __func__,
2195 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2196 			    inm->in6m_ifp->if_xname, inm->in6m_scrv);
2197 			if (inm->in6m_scrv == 0) {
2198 				inm->in6m_state = MLD_NOT_MEMBER;
2199 				inm->in6m_sctimer = 0;
2200 			} else {
2201 				int retval;
2202 
2203 				in6m_acquire_locked(inm);
2204 
2205 				retval = mld_v2_enqueue_group_record(
2206 				    &inm->in6m_scq, inm, 1, 0, 0);
2207 				KASSERT(retval != 0,
2208 				    ("%s: enqueue record = %d", __func__,
2209 				     retval));
2210 
2211 				inm->in6m_state = MLD_LEAVING_MEMBER;
2212 				inm->in6m_sctimer = 1;
2213 				V_state_change_timers_running6 = 1;
2214 				syncstates = 0;
2215 			}
2216 			break;
2217 		}
2218 		break;
2219 	case MLD_LAZY_MEMBER:
2220 	case MLD_SLEEPING_MEMBER:
2221 	case MLD_AWAKENING_MEMBER:
2222 		/* Our reports are suppressed; do nothing. */
2223 		break;
2224 	}
2225 
2226 	if (syncstates) {
2227 		in6m_commit(inm);
2228 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2229 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2230 		    inm->in6m_ifp->if_xname);
2231 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2232 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2233 		    __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname);
2234 	}
2235 }
2236 
2237 /*
2238  * Enqueue an MLDv2 group record to the given output queue.
2239  *
2240  * If is_state_change is zero, a current-state record is appended.
2241  * If is_state_change is non-zero, a state-change report is appended.
2242  *
2243  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2244  * If is_group_query is zero, and if there is a packet with free space
2245  * at the tail of the queue, it will be appended to providing there
2246  * is enough free space.
2247  * Otherwise a new mbuf packet chain is allocated.
2248  *
2249  * If is_source_query is non-zero, each source is checked to see if
2250  * it was recorded for a Group-Source query, and will be omitted if
2251  * it is not both in-mode and recorded.
2252  *
2253  * The function will attempt to allocate leading space in the packet
2254  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2255  *
2256  * If successful the size of all data appended to the queue is returned,
2257  * otherwise an error code less than zero is returned, or zero if
2258  * no record(s) were appended.
2259  */
2260 static int
2261 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2262     const int is_state_change, const int is_group_query,
2263     const int is_source_query)
2264 {
2265 	struct mldv2_record	 mr;
2266 	struct mldv2_record	*pmr;
2267 	struct ifnet		*ifp;
2268 	struct ip6_msource	*ims, *nims;
2269 	struct mbuf		*m0, *m, *md;
2270 	int			 error, is_filter_list_change;
2271 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2272 	int			 record_has_sources;
2273 	int			 now;
2274 	int			 type;
2275 	uint8_t			 mode;
2276 #ifdef KTR
2277 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2278 #endif
2279 
2280 	IN6_MULTI_LOCK_ASSERT();
2281 
2282 	error = 0;
2283 	ifp = inm->in6m_ifp;
2284 	is_filter_list_change = 0;
2285 	m = NULL;
2286 	m0 = NULL;
2287 	m0srcs = 0;
2288 	msrcs = 0;
2289 	nbytes = 0;
2290 	nims = NULL;
2291 	record_has_sources = 1;
2292 	pmr = NULL;
2293 	type = MLD_DO_NOTHING;
2294 	mode = inm->in6m_st[1].iss_fmode;
2295 
2296 	/*
2297 	 * If we did not transition out of ASM mode during t0->t1,
2298 	 * and there are no source nodes to process, we can skip
2299 	 * the generation of source records.
2300 	 */
2301 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2302 	    inm->in6m_nsrc == 0)
2303 		record_has_sources = 0;
2304 
2305 	if (is_state_change) {
2306 		/*
2307 		 * Queue a state change record.
2308 		 * If the mode did not change, and there are non-ASM
2309 		 * listeners or source filters present,
2310 		 * we potentially need to issue two records for the group.
2311 		 * If we are transitioning to MCAST_UNDEFINED, we need
2312 		 * not send any sources.
2313 		 * If there are ASM listeners, and there was no filter
2314 		 * mode transition of any kind, do nothing.
2315 		 */
2316 		if (mode != inm->in6m_st[0].iss_fmode) {
2317 			if (mode == MCAST_EXCLUDE) {
2318 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2319 				    __func__);
2320 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2321 			} else {
2322 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2323 				    __func__);
2324 				type = MLD_CHANGE_TO_INCLUDE_MODE;
2325 				if (mode == MCAST_UNDEFINED)
2326 					record_has_sources = 0;
2327 			}
2328 		} else {
2329 			if (record_has_sources) {
2330 				is_filter_list_change = 1;
2331 			} else {
2332 				type = MLD_DO_NOTHING;
2333 			}
2334 		}
2335 	} else {
2336 		/*
2337 		 * Queue a current state record.
2338 		 */
2339 		if (mode == MCAST_EXCLUDE) {
2340 			type = MLD_MODE_IS_EXCLUDE;
2341 		} else if (mode == MCAST_INCLUDE) {
2342 			type = MLD_MODE_IS_INCLUDE;
2343 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2344 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2345 			     __func__, inm, inm->in6m_st[1].iss_asm));
2346 		}
2347 	}
2348 
2349 	/*
2350 	 * Generate the filter list changes using a separate function.
2351 	 */
2352 	if (is_filter_list_change)
2353 		return (mld_v2_enqueue_filter_change(ifq, inm));
2354 
2355 	if (type == MLD_DO_NOTHING) {
2356 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2357 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2358 		    inm->in6m_ifp->if_xname);
2359 		return (0);
2360 	}
2361 
2362 	/*
2363 	 * If any sources are present, we must be able to fit at least
2364 	 * one in the trailing space of the tail packet's mbuf,
2365 	 * ideally more.
2366 	 */
2367 	minrec0len = sizeof(struct mldv2_record);
2368 	if (record_has_sources)
2369 		minrec0len += sizeof(struct in6_addr);
2370 
2371 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2372 	    mld_rec_type_to_str(type),
2373 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2374 	    inm->in6m_ifp->if_xname);
2375 
2376 	/*
2377 	 * Check if we have a packet in the tail of the queue for this
2378 	 * group into which the first group record for this group will fit.
2379 	 * Otherwise allocate a new packet.
2380 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2381 	 * Note: Group records for G/GSR query responses MUST be sent
2382 	 * in their own packet.
2383 	 */
2384 	m0 = ifq->ifq_tail;
2385 	if (!is_group_query &&
2386 	    m0 != NULL &&
2387 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2388 	    (m0->m_pkthdr.len + minrec0len) <
2389 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2390 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2391 			    sizeof(struct mldv2_record)) /
2392 			    sizeof(struct in6_addr);
2393 		m = m0;
2394 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2395 	} else {
2396 		if (_IF_QFULL(ifq)) {
2397 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2398 			return (-ENOMEM);
2399 		}
2400 		m = NULL;
2401 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2402 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2403 		if (!is_state_change && !is_group_query)
2404 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2405 		if (m == NULL)
2406 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2407 		if (m == NULL)
2408 			return (-ENOMEM);
2409 
2410 		mld_save_context(m, ifp);
2411 
2412 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2413 	}
2414 
2415 	/*
2416 	 * Append group record.
2417 	 * If we have sources, we don't know how many yet.
2418 	 */
2419 	mr.mr_type = type;
2420 	mr.mr_datalen = 0;
2421 	mr.mr_numsrc = 0;
2422 	mr.mr_addr = inm->in6m_addr;
2423 	in6_clearscope(&mr.mr_addr);
2424 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2425 		if (m != m0)
2426 			m_freem(m);
2427 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2428 		return (-ENOMEM);
2429 	}
2430 	nbytes += sizeof(struct mldv2_record);
2431 
2432 	/*
2433 	 * Append as many sources as will fit in the first packet.
2434 	 * If we are appending to a new packet, the chain allocation
2435 	 * may potentially use clusters; use m_getptr() in this case.
2436 	 * If we are appending to an existing packet, we need to obtain
2437 	 * a pointer to the group record after m_append(), in case a new
2438 	 * mbuf was allocated.
2439 	 * Only append sources which are in-mode at t1. If we are
2440 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2441 	 * include source entries.
2442 	 * Only report recorded sources in our filter set when responding
2443 	 * to a group-source query.
2444 	 */
2445 	if (record_has_sources) {
2446 		if (m == m0) {
2447 			md = m_last(m);
2448 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2449 			    md->m_len - nbytes);
2450 		} else {
2451 			md = m_getptr(m, 0, &off);
2452 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2453 			    off);
2454 		}
2455 		msrcs = 0;
2456 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2457 		    nims) {
2458 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2459 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2460 			now = im6s_get_mode(inm, ims, 1);
2461 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2462 			if ((now != mode) ||
2463 			    (now == mode && mode == MCAST_UNDEFINED)) {
2464 				CTR1(KTR_MLD, "%s: skip node", __func__);
2465 				continue;
2466 			}
2467 			if (is_source_query && ims->im6s_stp == 0) {
2468 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2469 				    __func__);
2470 				continue;
2471 			}
2472 			CTR1(KTR_MLD, "%s: append node", __func__);
2473 			if (!m_append(m, sizeof(struct in6_addr),
2474 			    (void *)&ims->im6s_addr)) {
2475 				if (m != m0)
2476 					m_freem(m);
2477 				CTR1(KTR_MLD, "%s: m_append() failed.",
2478 				    __func__);
2479 				return (-ENOMEM);
2480 			}
2481 			nbytes += sizeof(struct in6_addr);
2482 			++msrcs;
2483 			if (msrcs == m0srcs)
2484 				break;
2485 		}
2486 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2487 		    msrcs);
2488 		pmr->mr_numsrc = htons(msrcs);
2489 		nbytes += (msrcs * sizeof(struct in6_addr));
2490 	}
2491 
2492 	if (is_source_query && msrcs == 0) {
2493 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2494 		if (m != m0)
2495 			m_freem(m);
2496 		return (0);
2497 	}
2498 
2499 	/*
2500 	 * We are good to go with first packet.
2501 	 */
2502 	if (m != m0) {
2503 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2504 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2505 		_IF_ENQUEUE(ifq, m);
2506 	} else
2507 		m->m_pkthdr.PH_vt.vt_nrecs++;
2508 
2509 	/*
2510 	 * No further work needed if no source list in packet(s).
2511 	 */
2512 	if (!record_has_sources)
2513 		return (nbytes);
2514 
2515 	/*
2516 	 * Whilst sources remain to be announced, we need to allocate
2517 	 * a new packet and fill out as many sources as will fit.
2518 	 * Always try for a cluster first.
2519 	 */
2520 	while (nims != NULL) {
2521 		if (_IF_QFULL(ifq)) {
2522 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2523 			return (-ENOMEM);
2524 		}
2525 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2526 		if (m == NULL)
2527 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2528 		if (m == NULL)
2529 			return (-ENOMEM);
2530 		mld_save_context(m, ifp);
2531 		md = m_getptr(m, 0, &off);
2532 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2533 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2534 
2535 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2536 			if (m != m0)
2537 				m_freem(m);
2538 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2539 			return (-ENOMEM);
2540 		}
2541 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2542 		nbytes += sizeof(struct mldv2_record);
2543 
2544 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2545 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2546 
2547 		msrcs = 0;
2548 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2549 			CTR2(KTR_MLD, "%s: visit node %s",
2550 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2551 			now = im6s_get_mode(inm, ims, 1);
2552 			if ((now != mode) ||
2553 			    (now == mode && mode == MCAST_UNDEFINED)) {
2554 				CTR1(KTR_MLD, "%s: skip node", __func__);
2555 				continue;
2556 			}
2557 			if (is_source_query && ims->im6s_stp == 0) {
2558 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2559 				    __func__);
2560 				continue;
2561 			}
2562 			CTR1(KTR_MLD, "%s: append node", __func__);
2563 			if (!m_append(m, sizeof(struct in6_addr),
2564 			    (void *)&ims->im6s_addr)) {
2565 				if (m != m0)
2566 					m_freem(m);
2567 				CTR1(KTR_MLD, "%s: m_append() failed.",
2568 				    __func__);
2569 				return (-ENOMEM);
2570 			}
2571 			++msrcs;
2572 			if (msrcs == m0srcs)
2573 				break;
2574 		}
2575 		pmr->mr_numsrc = htons(msrcs);
2576 		nbytes += (msrcs * sizeof(struct in6_addr));
2577 
2578 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2579 		_IF_ENQUEUE(ifq, m);
2580 	}
2581 
2582 	return (nbytes);
2583 }
2584 
2585 /*
2586  * Type used to mark record pass completion.
2587  * We exploit the fact we can cast to this easily from the
2588  * current filter modes on each ip_msource node.
2589  */
2590 typedef enum {
2591 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2592 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2593 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2594 	REC_FULL = REC_ALLOW | REC_BLOCK
2595 } rectype_t;
2596 
2597 /*
2598  * Enqueue an MLDv2 filter list change to the given output queue.
2599  *
2600  * Source list filter state is held in an RB-tree. When the filter list
2601  * for a group is changed without changing its mode, we need to compute
2602  * the deltas between T0 and T1 for each source in the filter set,
2603  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2604  *
2605  * As we may potentially queue two record types, and the entire R-B tree
2606  * needs to be walked at once, we break this out into its own function
2607  * so we can generate a tightly packed queue of packets.
2608  *
2609  * XXX This could be written to only use one tree walk, although that makes
2610  * serializing into the mbuf chains a bit harder. For now we do two walks
2611  * which makes things easier on us, and it may or may not be harder on
2612  * the L2 cache.
2613  *
2614  * If successful the size of all data appended to the queue is returned,
2615  * otherwise an error code less than zero is returned, or zero if
2616  * no record(s) were appended.
2617  */
2618 static int
2619 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
2620 {
2621 	static const int MINRECLEN =
2622 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2623 	struct ifnet		*ifp;
2624 	struct mldv2_record	 mr;
2625 	struct mldv2_record	*pmr;
2626 	struct ip6_msource	*ims, *nims;
2627 	struct mbuf		*m, *m0, *md;
2628 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2629 	int			 nallow, nblock;
2630 	uint8_t			 mode, now, then;
2631 	rectype_t		 crt, drt, nrt;
2632 #ifdef KTR
2633 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2634 #endif
2635 
2636 	IN6_MULTI_LOCK_ASSERT();
2637 
2638 	if (inm->in6m_nsrc == 0 ||
2639 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2640 		return (0);
2641 
2642 	ifp = inm->in6m_ifp;			/* interface */
2643 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2644 	crt = REC_NONE;	/* current group record type */
2645 	drt = REC_NONE;	/* mask of completed group record types */
2646 	nrt = REC_NONE;	/* record type for current node */
2647 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2648 	npbytes = 0;	/* # of bytes appended this packet */
2649 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2650 	rsrcs = 0;	/* # sources encoded in current record */
2651 	schanged = 0;	/* # nodes encoded in overall filter change */
2652 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2653 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2654 	nims = NULL;	/* next tree node pointer */
2655 
2656 	/*
2657 	 * For each possible filter record mode.
2658 	 * The first kind of source we encounter tells us which
2659 	 * is the first kind of record we start appending.
2660 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2661 	 * as the inverse of the group's filter mode.
2662 	 */
2663 	while (drt != REC_FULL) {
2664 		do {
2665 			m0 = ifq->ifq_tail;
2666 			if (m0 != NULL &&
2667 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2668 			     MLD_V2_REPORT_MAXRECS) &&
2669 			    (m0->m_pkthdr.len + MINRECLEN) <
2670 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2671 				m = m0;
2672 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2673 					    sizeof(struct mldv2_record)) /
2674 					    sizeof(struct in6_addr);
2675 				CTR1(KTR_MLD,
2676 				    "%s: use previous packet", __func__);
2677 			} else {
2678 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2679 				if (m == NULL)
2680 					m = m_gethdr(M_DONTWAIT, MT_DATA);
2681 				if (m == NULL) {
2682 					CTR1(KTR_MLD,
2683 					    "%s: m_get*() failed", __func__);
2684 					return (-ENOMEM);
2685 				}
2686 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2687 				mld_save_context(m, ifp);
2688 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2689 				    sizeof(struct mldv2_record)) /
2690 				    sizeof(struct in6_addr);
2691 				npbytes = 0;
2692 				CTR1(KTR_MLD,
2693 				    "%s: allocated new packet", __func__);
2694 			}
2695 			/*
2696 			 * Append the MLD group record header to the
2697 			 * current packet's data area.
2698 			 * Recalculate pointer to free space for next
2699 			 * group record, in case m_append() allocated
2700 			 * a new mbuf or cluster.
2701 			 */
2702 			memset(&mr, 0, sizeof(mr));
2703 			mr.mr_addr = inm->in6m_addr;
2704 			in6_clearscope(&mr.mr_addr);
2705 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2706 				if (m != m0)
2707 					m_freem(m);
2708 				CTR1(KTR_MLD,
2709 				    "%s: m_append() failed", __func__);
2710 				return (-ENOMEM);
2711 			}
2712 			npbytes += sizeof(struct mldv2_record);
2713 			if (m != m0) {
2714 				/* new packet; offset in chain */
2715 				md = m_getptr(m, npbytes -
2716 				    sizeof(struct mldv2_record), &off);
2717 				pmr = (struct mldv2_record *)(mtod(md,
2718 				    uint8_t *) + off);
2719 			} else {
2720 				/* current packet; offset from last append */
2721 				md = m_last(m);
2722 				pmr = (struct mldv2_record *)(mtod(md,
2723 				    uint8_t *) + md->m_len -
2724 				    sizeof(struct mldv2_record));
2725 			}
2726 			/*
2727 			 * Begin walking the tree for this record type
2728 			 * pass, or continue from where we left off
2729 			 * previously if we had to allocate a new packet.
2730 			 * Only report deltas in-mode at t1.
2731 			 * We need not report included sources as allowed
2732 			 * if we are in inclusive mode on the group,
2733 			 * however the converse is not true.
2734 			 */
2735 			rsrcs = 0;
2736 			if (nims == NULL) {
2737 				nims = RB_MIN(ip6_msource_tree,
2738 				    &inm->in6m_srcs);
2739 			}
2740 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2741 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2742 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2743 				now = im6s_get_mode(inm, ims, 1);
2744 				then = im6s_get_mode(inm, ims, 0);
2745 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2746 				    __func__, then, now);
2747 				if (now == then) {
2748 					CTR1(KTR_MLD,
2749 					    "%s: skip unchanged", __func__);
2750 					continue;
2751 				}
2752 				if (mode == MCAST_EXCLUDE &&
2753 				    now == MCAST_INCLUDE) {
2754 					CTR1(KTR_MLD,
2755 					    "%s: skip IN src on EX group",
2756 					    __func__);
2757 					continue;
2758 				}
2759 				nrt = (rectype_t)now;
2760 				if (nrt == REC_NONE)
2761 					nrt = (rectype_t)(~mode & REC_FULL);
2762 				if (schanged++ == 0) {
2763 					crt = nrt;
2764 				} else if (crt != nrt)
2765 					continue;
2766 				if (!m_append(m, sizeof(struct in6_addr),
2767 				    (void *)&ims->im6s_addr)) {
2768 					if (m != m0)
2769 						m_freem(m);
2770 					CTR1(KTR_MLD,
2771 					    "%s: m_append() failed", __func__);
2772 					return (-ENOMEM);
2773 				}
2774 				nallow += !!(crt == REC_ALLOW);
2775 				nblock += !!(crt == REC_BLOCK);
2776 				if (++rsrcs == m0srcs)
2777 					break;
2778 			}
2779 			/*
2780 			 * If we did not append any tree nodes on this
2781 			 * pass, back out of allocations.
2782 			 */
2783 			if (rsrcs == 0) {
2784 				npbytes -= sizeof(struct mldv2_record);
2785 				if (m != m0) {
2786 					CTR1(KTR_MLD,
2787 					    "%s: m_free(m)", __func__);
2788 					m_freem(m);
2789 				} else {
2790 					CTR1(KTR_MLD,
2791 					    "%s: m_adj(m, -mr)", __func__);
2792 					m_adj(m, -((int)sizeof(
2793 					    struct mldv2_record)));
2794 				}
2795 				continue;
2796 			}
2797 			npbytes += (rsrcs * sizeof(struct in6_addr));
2798 			if (crt == REC_ALLOW)
2799 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2800 			else if (crt == REC_BLOCK)
2801 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2802 			pmr->mr_numsrc = htons(rsrcs);
2803 			/*
2804 			 * Count the new group record, and enqueue this
2805 			 * packet if it wasn't already queued.
2806 			 */
2807 			m->m_pkthdr.PH_vt.vt_nrecs++;
2808 			if (m != m0)
2809 				_IF_ENQUEUE(ifq, m);
2810 			nbytes += npbytes;
2811 		} while (nims != NULL);
2812 		drt |= crt;
2813 		crt = (~crt & REC_FULL);
2814 	}
2815 
2816 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2817 	    nallow, nblock);
2818 
2819 	return (nbytes);
2820 }
2821 
2822 static int
2823 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
2824 {
2825 	struct ifqueue	*gq;
2826 	struct mbuf	*m;		/* pending state-change */
2827 	struct mbuf	*m0;		/* copy of pending state-change */
2828 	struct mbuf	*mt;		/* last state-change in packet */
2829 	int		 docopy, domerge;
2830 	u_int		 recslen;
2831 
2832 	docopy = 0;
2833 	domerge = 0;
2834 	recslen = 0;
2835 
2836 	IN6_MULTI_LOCK_ASSERT();
2837 	MLD_LOCK_ASSERT();
2838 
2839 	/*
2840 	 * If there are further pending retransmissions, make a writable
2841 	 * copy of each queued state-change message before merging.
2842 	 */
2843 	if (inm->in6m_scrv > 0)
2844 		docopy = 1;
2845 
2846 	gq = &inm->in6m_scq;
2847 #ifdef KTR
2848 	if (gq->ifq_head == NULL) {
2849 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2850 		    __func__, inm);
2851 	}
2852 #endif
2853 
2854 	m = gq->ifq_head;
2855 	while (m != NULL) {
2856 		/*
2857 		 * Only merge the report into the current packet if
2858 		 * there is sufficient space to do so; an MLDv2 report
2859 		 * packet may only contain 65,535 group records.
2860 		 * Always use a simple mbuf chain concatentation to do this,
2861 		 * as large state changes for single groups may have
2862 		 * allocated clusters.
2863 		 */
2864 		domerge = 0;
2865 		mt = ifscq->ifq_tail;
2866 		if (mt != NULL) {
2867 			recslen = m_length(m, NULL);
2868 
2869 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2870 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2871 			    MLD_V2_REPORT_MAXRECS) &&
2872 			    (mt->m_pkthdr.len + recslen <=
2873 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2874 				domerge = 1;
2875 		}
2876 
2877 		if (!domerge && _IF_QFULL(gq)) {
2878 			CTR2(KTR_MLD,
2879 			    "%s: outbound queue full, skipping whole packet %p",
2880 			    __func__, m);
2881 			mt = m->m_nextpkt;
2882 			if (!docopy)
2883 				m_freem(m);
2884 			m = mt;
2885 			continue;
2886 		}
2887 
2888 		if (!docopy) {
2889 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2890 			_IF_DEQUEUE(gq, m0);
2891 			m = m0->m_nextpkt;
2892 		} else {
2893 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2894 			m0 = m_dup(m, M_NOWAIT);
2895 			if (m0 == NULL)
2896 				return (ENOMEM);
2897 			m0->m_nextpkt = NULL;
2898 			m = m->m_nextpkt;
2899 		}
2900 
2901 		if (!domerge) {
2902 			CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)",
2903 			    __func__, m0, ifscq);
2904 			_IF_ENQUEUE(ifscq, m0);
2905 		} else {
2906 			struct mbuf *mtl;	/* last mbuf of packet mt */
2907 
2908 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2909 			    __func__, m0, mt);
2910 
2911 			mtl = m_last(mt);
2912 			m0->m_flags &= ~M_PKTHDR;
2913 			mt->m_pkthdr.len += recslen;
2914 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2915 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2916 
2917 			mtl->m_next = m0;
2918 		}
2919 	}
2920 
2921 	return (0);
2922 }
2923 
2924 /*
2925  * Respond to a pending MLDv2 General Query.
2926  */
2927 static void
2928 mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
2929 {
2930 	struct ifmultiaddr	*ifma, *tifma;
2931 	struct ifnet		*ifp;
2932 	struct in6_multi	*inm;
2933 	int			 retval;
2934 
2935 	IN6_MULTI_LOCK_ASSERT();
2936 	MLD_LOCK_ASSERT();
2937 
2938 	KASSERT(mli->mli_version == MLD_VERSION_2,
2939 	    ("%s: called when version %d", __func__, mli->mli_version));
2940 
2941 	ifp = mli->mli_ifp;
2942 
2943 	IF_ADDR_LOCK(ifp);
2944 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
2945 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2946 		    ifma->ifma_protospec == NULL)
2947 			continue;
2948 
2949 		inm = (struct in6_multi *)ifma->ifma_protospec;
2950 		KASSERT(ifp == inm->in6m_ifp,
2951 		    ("%s: inconsistent ifp", __func__));
2952 
2953 		switch (inm->in6m_state) {
2954 		case MLD_NOT_MEMBER:
2955 		case MLD_SILENT_MEMBER:
2956 			break;
2957 		case MLD_REPORTING_MEMBER:
2958 		case MLD_IDLE_MEMBER:
2959 		case MLD_LAZY_MEMBER:
2960 		case MLD_SLEEPING_MEMBER:
2961 		case MLD_AWAKENING_MEMBER:
2962 			inm->in6m_state = MLD_REPORTING_MEMBER;
2963 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
2964 			    inm, 0, 0, 0);
2965 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2966 			    __func__, retval);
2967 			break;
2968 		case MLD_G_QUERY_PENDING_MEMBER:
2969 		case MLD_SG_QUERY_PENDING_MEMBER:
2970 		case MLD_LEAVING_MEMBER:
2971 			break;
2972 		}
2973 	}
2974 	IF_ADDR_UNLOCK(ifp);
2975 
2976 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
2977 
2978 	/*
2979 	 * Slew transmission of bursts over 500ms intervals.
2980 	 */
2981 	if (mli->mli_gq.ifq_head != NULL) {
2982 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
2983 		    MLD_RESPONSE_BURST_INTERVAL);
2984 		V_interface_timers_running6 = 1;
2985 	}
2986 }
2987 
2988 /*
2989  * Transmit the next pending message in the output queue.
2990  *
2991  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
2992  * MRT: Nothing needs to be done, as MLD traffic is always local to
2993  * a link and uses a link-scope multicast address.
2994  */
2995 static void
2996 mld_dispatch_packet(struct mbuf *m)
2997 {
2998 	struct ip6_moptions	 im6o;
2999 	struct ifnet		*ifp;
3000 	struct ifnet		*oifp;
3001 	struct mbuf		*m0;
3002 	struct mbuf		*md;
3003 	struct ip6_hdr		*ip6;
3004 	struct mld_hdr		*mld;
3005 	int			 error;
3006 	int			 off;
3007 	int			 type;
3008 	uint32_t		 ifindex;
3009 
3010 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3011 
3012 	/*
3013 	 * Set VNET image pointer from enqueued mbuf chain
3014 	 * before doing anything else. Whilst we use interface
3015 	 * indexes to guard against interface detach, they are
3016 	 * unique to each VIMAGE and must be retrieved.
3017 	 */
3018 	ifindex = mld_restore_context(m);
3019 
3020 	/*
3021 	 * Check if the ifnet still exists. This limits the scope of
3022 	 * any race in the absence of a global ifp lock for low cost
3023 	 * (an array lookup).
3024 	 */
3025 	ifp = ifnet_byindex(ifindex);
3026 	if (ifp == NULL) {
3027 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3028 		    __func__, m, ifindex);
3029 		m_freem(m);
3030 		IP6STAT_INC(ip6s_noroute);
3031 		goto out;
3032 	}
3033 
3034 	im6o.im6o_multicast_hlim  = 1;
3035 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3036 	im6o.im6o_multicast_ifp = ifp;
3037 
3038 	if (m->m_flags & M_MLDV1) {
3039 		m0 = m;
3040 	} else {
3041 		m0 = mld_v2_encap_report(ifp, m);
3042 		if (m0 == NULL) {
3043 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3044 			m_freem(m);
3045 			IP6STAT_INC(ip6s_odropped);
3046 			goto out;
3047 		}
3048 	}
3049 
3050 	mld_scrub_context(m0);
3051 	m->m_flags &= ~(M_PROTOFLAGS);
3052 	m0->m_pkthdr.rcvif = V_loif;
3053 
3054 	ip6 = mtod(m0, struct ip6_hdr *);
3055 #if 0
3056 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3057 #else
3058 	/*
3059 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3060 	 * occur if we called in6_setscope() at transmission.
3061 	 * See comments at top of file.
3062 	 */
3063 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3064 #endif
3065 
3066 	/*
3067 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3068 	 * so we can bump the stats.
3069 	 */
3070 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3071 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3072 	type = mld->mld_type;
3073 
3074 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3075 	    &oifp, NULL);
3076 	if (error) {
3077 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3078 		goto out;
3079 	}
3080 	ICMP6STAT_INC(icp6s_outhist[type]);
3081 	if (oifp != NULL) {
3082 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3083 		switch (type) {
3084 		case MLD_LISTENER_REPORT:
3085 		case MLDV2_LISTENER_REPORT:
3086 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3087 			break;
3088 		case MLD_LISTENER_DONE:
3089 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3090 			break;
3091 		}
3092 	}
3093 out:
3094 	return;
3095 }
3096 
3097 /*
3098  * Encapsulate an MLDv2 report.
3099  *
3100  * KAME IPv6 requires that hop-by-hop options be passed separately,
3101  * and that the IPv6 header be prepended in a separate mbuf.
3102  *
3103  * Returns a pointer to the new mbuf chain head, or NULL if the
3104  * allocation failed.
3105  */
3106 static struct mbuf *
3107 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3108 {
3109 	struct mbuf		*mh;
3110 	struct mldv2_report	*mld;
3111 	struct ip6_hdr		*ip6;
3112 	struct in6_ifaddr	*ia;
3113 	int			 mldreclen;
3114 
3115 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3116 	KASSERT((m->m_flags & M_PKTHDR),
3117 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3118 
3119 	/*
3120 	 * RFC3590: OK to send as :: or tentative during DAD.
3121 	 */
3122 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3123 	if (ia == NULL)
3124 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3125 
3126 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3127 	if (mh == NULL) {
3128 		if (ia != NULL)
3129 			ifa_free(&ia->ia_ifa);
3130 		m_freem(m);
3131 		return (NULL);
3132 	}
3133 	MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3134 
3135 	mldreclen = m_length(m, NULL);
3136 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3137 
3138 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3139 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3140 	    sizeof(struct mldv2_report) + mldreclen;
3141 
3142 	ip6 = mtod(mh, struct ip6_hdr *);
3143 	ip6->ip6_flow = 0;
3144 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3145 	ip6->ip6_vfc |= IPV6_VERSION;
3146 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3147 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3148 	if (ia != NULL)
3149 		ifa_free(&ia->ia_ifa);
3150 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3151 	/* scope ID will be set in netisr */
3152 
3153 	mld = (struct mldv2_report *)(ip6 + 1);
3154 	mld->mld_type = MLDV2_LISTENER_REPORT;
3155 	mld->mld_code = 0;
3156 	mld->mld_cksum = 0;
3157 	mld->mld_v2_reserved = 0;
3158 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3159 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3160 
3161 	mh->m_next = m;
3162 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3163 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3164 	return (mh);
3165 }
3166 
3167 #ifdef KTR
3168 static char *
3169 mld_rec_type_to_str(const int type)
3170 {
3171 
3172 	switch (type) {
3173 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3174 			return "TO_EX";
3175 			break;
3176 		case MLD_CHANGE_TO_INCLUDE_MODE:
3177 			return "TO_IN";
3178 			break;
3179 		case MLD_MODE_IS_EXCLUDE:
3180 			return "MODE_EX";
3181 			break;
3182 		case MLD_MODE_IS_INCLUDE:
3183 			return "MODE_IN";
3184 			break;
3185 		case MLD_ALLOW_NEW_SOURCES:
3186 			return "ALLOW_NEW";
3187 			break;
3188 		case MLD_BLOCK_OLD_SOURCES:
3189 			return "BLOCK_OLD";
3190 			break;
3191 		default:
3192 			break;
3193 	}
3194 	return "unknown";
3195 }
3196 #endif
3197 
3198 static void
3199 mld_init(void *unused __unused)
3200 {
3201 
3202 	CTR1(KTR_MLD, "%s: initializing", __func__);
3203 	MLD_LOCK_INIT();
3204 
3205 	ip6_initpktopts(&mld_po);
3206 	mld_po.ip6po_hlim = 1;
3207 	mld_po.ip6po_hbh = &mld_ra.hbh;
3208 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3209 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3210 }
3211 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
3212 
3213 static void
3214 mld_uninit(void *unused __unused)
3215 {
3216 
3217 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3218 	MLD_LOCK_DESTROY();
3219 }
3220 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
3221 
3222 static void
3223 vnet_mld_init(const void *unused __unused)
3224 {
3225 
3226 	CTR1(KTR_MLD, "%s: initializing", __func__);
3227 
3228 	LIST_INIT(&V_mli_head);
3229 }
3230 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
3231     NULL);
3232 
3233 static void
3234 vnet_mld_uninit(const void *unused __unused)
3235 {
3236 
3237 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3238 
3239 	KASSERT(LIST_EMPTY(&V_mli_head),
3240 	    ("%s: mli list not empty; ifnets not detached?", __func__));
3241 }
3242 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
3243     NULL);
3244 
3245 static int
3246 mld_modevent(module_t mod, int type, void *unused __unused)
3247 {
3248 
3249     switch (type) {
3250     case MOD_LOAD:
3251     case MOD_UNLOAD:
3252 	break;
3253     default:
3254 	return (EOPNOTSUPP);
3255     }
3256     return (0);
3257 }
3258 
3259 static moduledata_t mld_mod = {
3260     "mld",
3261     mld_modevent,
3262     0
3263 };
3264 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3265