xref: /freebsd/sys/netinet6/mld6.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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 
83 #include <net/if.h>
84 #include <net/route.h>
85 #include <net/vnet.h>
86 
87 #include <netinet/in.h>
88 #include <netinet/in_var.h>
89 #include <netinet6/in6_var.h>
90 #include <netinet/ip6.h>
91 #include <netinet6/ip6_var.h>
92 #include <netinet6/scope6_var.h>
93 #include <netinet/icmp6.h>
94 #include <netinet6/mld6.h>
95 #include <netinet6/mld6_var.h>
96 
97 #include <security/mac/mac_framework.h>
98 
99 #ifndef KTR_MLD
100 #define KTR_MLD KTR_INET6
101 #endif
102 
103 static struct mld_ifinfo *
104 		mli_alloc_locked(struct ifnet *);
105 static void	mli_delete_locked(const struct ifnet *);
106 static void	mld_dispatch_packet(struct mbuf *);
107 static void	mld_dispatch_queue(struct ifqueue *, int);
108 static void	mld_final_leave(struct in6_multi *, struct mld_ifinfo *);
109 static void	mld_fasttimo_vnet(void);
110 static int	mld_handle_state_change(struct in6_multi *,
111 		    struct mld_ifinfo *);
112 static int	mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
113 		    const int);
114 #ifdef KTR
115 static char *	mld_rec_type_to_str(const int);
116 #endif
117 static void	mld_set_version(struct mld_ifinfo *, const int);
118 static void	mld_slowtimo_vnet(void);
119 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
120 		    /*const*/ struct mld_hdr *);
121 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
122 		    /*const*/ struct mld_hdr *);
123 static void	mld_v1_process_group_timer(struct in6_multi *, const int);
124 static void	mld_v1_process_querier_timers(struct mld_ifinfo *);
125 static int	mld_v1_transmit_report(struct in6_multi *, const int);
126 static void	mld_v1_update_group(struct in6_multi *, const int);
127 static void	mld_v2_cancel_link_timers(struct mld_ifinfo *);
128 static void	mld_v2_dispatch_general_query(struct mld_ifinfo *);
129 static struct mbuf *
130 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
131 static int	mld_v2_enqueue_filter_change(struct ifqueue *,
132 		    struct in6_multi *);
133 static int	mld_v2_enqueue_group_record(struct ifqueue *,
134 		    struct in6_multi *, const int, const int, const int);
135 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
136 		    struct mbuf *, const int, const int);
137 static int	mld_v2_merge_state_changes(struct in6_multi *,
138 		    struct ifqueue *);
139 static void	mld_v2_process_group_timers(struct mld_ifinfo *,
140 		    struct ifqueue *, struct ifqueue *,
141 		    struct in6_multi *, const int);
142 static int	mld_v2_process_group_query(struct in6_multi *,
143 		    struct mld_ifinfo *mli, int, struct mbuf *, const int);
144 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
145 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
146 
147 /*
148  * Normative references: RFC 2710, RFC 3590, RFC 3810.
149  *
150  * Locking:
151  *  * The MLD subsystem lock ends up being system-wide for the moment,
152  *    but could be per-VIMAGE later on.
153  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
154  *    Any may be taken independently; if any are held at the same
155  *    time, the above lock order must be followed.
156  *  * IN6_MULTI_LOCK covers in_multi.
157  *  * MLD_LOCK covers per-link state and any global variables in this file.
158  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
159  *    per-link state iterators.
160  *
161  *  XXX LOR PREVENTION
162  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
163  *  will not accept an ifp; it wants an embedded scope ID, unlike
164  *  ip_output(), which happily takes the ifp given to it. The embedded
165  *  scope ID is only used by MLD to select the outgoing interface.
166  *
167  *  During interface attach and detach, MLD will take MLD_LOCK *after*
168  *  the IF_AFDATA_LOCK.
169  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
170  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
171  *  dispatch could work around this, but we'd rather not do that, as it
172  *  can introduce other races.
173  *
174  *  As such, we exploit the fact that the scope ID is just the interface
175  *  index, and embed it in the IPv6 destination address accordingly.
176  *  This is potentially NOT VALID for MLDv1 reports, as they
177  *  are always sent to the multicast group itself; as MLDv2
178  *  reports are always sent to ff02::16, this is not an issue
179  *  when MLDv2 is in use.
180  *
181  *  This does not however eliminate the LOR when ip6_output() itself
182  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
183  *  trigger a LOR warning in WITNESS when the ifnet is detached.
184  *
185  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
186  *  how it's used across the network stack. Here we're simply exploiting
187  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
188  *
189  * VIMAGE:
190  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
191  *    to a vnet in ifp->if_vnet.
192  */
193 static struct mtx		 mld_mtx;
194 MALLOC_DEFINE(M_MLD, "mld", "mld state");
195 
196 #define	MLD_EMBEDSCOPE(pin6, zoneid) \
197 	(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)
198 
199 /*
200  * VIMAGE-wide globals.
201  */
202 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0};
203 static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head);
204 static VNET_DEFINE(int, interface_timers_running6);
205 static VNET_DEFINE(int, state_change_timers_running6);
206 static VNET_DEFINE(int, current_state_timers_running6);
207 
208 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
209 #define	V_mli_head			VNET(mli_head)
210 #define	V_interface_timers_running6	VNET(interface_timers_running6)
211 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
212 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
213 
214 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
215 
216 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
217     "IPv6 Multicast Listener Discovery");
218 
219 /*
220  * Virtualized sysctls.
221  */
222 SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
223     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
224     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
225     "Rate limit for MLDv2 Group-and-Source queries in seconds");
226 
227 /*
228  * Non-virtualized sysctls.
229  */
230 SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
231     sysctl_mld_ifinfo, "Per-interface MLDv2 state");
232 
233 static int	mld_v1enable = 1;
234 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW,
235     &mld_v1enable, 0, "Enable fallback to MLDv1");
236 TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable);
237 
238 /*
239  * Packed Router Alert option structure declaration.
240  */
241 struct mld_raopt {
242 	struct ip6_hbh		hbh;
243 	struct ip6_opt		pad;
244 	struct ip6_opt_router	ra;
245 } __packed;
246 
247 /*
248  * Router Alert hop-by-hop option header.
249  */
250 static struct mld_raopt mld_ra = {
251 	.hbh = { 0, 0 },
252 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
253 	.ra = {
254 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
255 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
256 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
257 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
258 	}
259 };
260 static struct ip6_pktopts mld_po;
261 
262 static __inline void
263 mld_save_context(struct mbuf *m, struct ifnet *ifp)
264 {
265 
266 #ifdef VIMAGE
267 	m->m_pkthdr.header = ifp->if_vnet;
268 #endif /* VIMAGE */
269 	m->m_pkthdr.flowid = ifp->if_index;
270 }
271 
272 static __inline void
273 mld_scrub_context(struct mbuf *m)
274 {
275 
276 	m->m_pkthdr.header = NULL;
277 	m->m_pkthdr.flowid = 0;
278 }
279 
280 /*
281  * Restore context from a queued output chain.
282  * Return saved ifindex.
283  *
284  * VIMAGE: The assertion is there to make sure that we
285  * actually called CURVNET_SET() with what's in the mbuf chain.
286  */
287 static __inline uint32_t
288 mld_restore_context(struct mbuf *m)
289 {
290 
291 #if defined(VIMAGE) && defined(INVARIANTS)
292 	KASSERT(curvnet == m->m_pkthdr.header,
293 	    ("%s: called when curvnet was not restored", __func__));
294 #endif
295 	return (m->m_pkthdr.flowid);
296 }
297 
298 /*
299  * Retrieve or set threshold between group-source queries in seconds.
300  *
301  * VIMAGE: Assume curvnet set by caller.
302  * SMPng: NOTE: Serialized by MLD lock.
303  */
304 static int
305 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
306 {
307 	int error;
308 	int i;
309 
310 	error = sysctl_wire_old_buffer(req, sizeof(int));
311 	if (error)
312 		return (error);
313 
314 	MLD_LOCK();
315 
316 	i = V_mld_gsrdelay.tv_sec;
317 
318 	error = sysctl_handle_int(oidp, &i, 0, req);
319 	if (error || !req->newptr)
320 		goto out_locked;
321 
322 	if (i < -1 || i >= 60) {
323 		error = EINVAL;
324 		goto out_locked;
325 	}
326 
327 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
328 	     V_mld_gsrdelay.tv_sec, i);
329 	V_mld_gsrdelay.tv_sec = i;
330 
331 out_locked:
332 	MLD_UNLOCK();
333 	return (error);
334 }
335 
336 /*
337  * Expose struct mld_ifinfo to userland, keyed by ifindex.
338  * For use by ifmcstat(8).
339  *
340  * SMPng: NOTE: Does an unlocked ifindex space read.
341  * VIMAGE: Assume curvnet set by caller. The node handler itself
342  * is not directly virtualized.
343  */
344 static int
345 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
346 {
347 	int			*name;
348 	int			 error;
349 	u_int			 namelen;
350 	struct ifnet		*ifp;
351 	struct mld_ifinfo	*mli;
352 
353 	name = (int *)arg1;
354 	namelen = arg2;
355 
356 	if (req->newptr != NULL)
357 		return (EPERM);
358 
359 	if (namelen != 1)
360 		return (EINVAL);
361 
362 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
363 	if (error)
364 		return (error);
365 
366 	IN6_MULTI_LOCK();
367 	MLD_LOCK();
368 
369 	if (name[0] <= 0 || name[0] > V_if_index) {
370 		error = ENOENT;
371 		goto out_locked;
372 	}
373 
374 	error = ENOENT;
375 
376 	ifp = ifnet_byindex(name[0]);
377 	if (ifp == NULL)
378 		goto out_locked;
379 
380 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
381 		if (ifp == mli->mli_ifp) {
382 			error = SYSCTL_OUT(req, mli,
383 			    sizeof(struct mld_ifinfo));
384 			break;
385 		}
386 	}
387 
388 out_locked:
389 	MLD_UNLOCK();
390 	IN6_MULTI_UNLOCK();
391 	return (error);
392 }
393 
394 /*
395  * Dispatch an entire queue of pending packet chains.
396  * VIMAGE: Assumes the vnet pointer has been set.
397  */
398 static void
399 mld_dispatch_queue(struct ifqueue *ifq, int limit)
400 {
401 	struct mbuf *m;
402 
403 	for (;;) {
404 		_IF_DEQUEUE(ifq, m);
405 		if (m == NULL)
406 			break;
407 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m);
408 		mld_dispatch_packet(m);
409 		if (--limit == 0)
410 			break;
411 	}
412 }
413 
414 /*
415  * Filter outgoing MLD report state by group.
416  *
417  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
418  * and node-local addresses. However, kernel and socket consumers
419  * always embed the KAME scope ID in the address provided, so strip it
420  * when performing comparison.
421  * Note: This is not the same as the *multicast* scope.
422  *
423  * Return zero if the given group is one for which MLD reports
424  * should be suppressed, or non-zero if reports should be issued.
425  */
426 static __inline int
427 mld_is_addr_reported(const struct in6_addr *addr)
428 {
429 
430 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
431 
432 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
433 		return (0);
434 
435 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
436 		struct in6_addr tmp = *addr;
437 		in6_clearscope(&tmp);
438 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
439 			return (0);
440 	}
441 
442 	return (1);
443 }
444 
445 /*
446  * Attach MLD when PF_INET6 is attached to an interface.
447  *
448  * SMPng: Normally called with IF_AFDATA_LOCK held.
449  */
450 struct mld_ifinfo *
451 mld_domifattach(struct ifnet *ifp)
452 {
453 	struct mld_ifinfo *mli;
454 
455 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
456 	    __func__, ifp, ifp->if_xname);
457 
458 	MLD_LOCK();
459 
460 	mli = mli_alloc_locked(ifp);
461 	if (!(ifp->if_flags & IFF_MULTICAST))
462 		mli->mli_flags |= MLIF_SILENT;
463 
464 	MLD_UNLOCK();
465 
466 	return (mli);
467 }
468 
469 /*
470  * VIMAGE: assume curvnet set by caller.
471  */
472 static struct mld_ifinfo *
473 mli_alloc_locked(/*const*/ struct ifnet *ifp)
474 {
475 	struct mld_ifinfo *mli;
476 
477 	MLD_LOCK_ASSERT();
478 
479 	mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO);
480 	if (mli == NULL)
481 		goto out;
482 
483 	mli->mli_ifp = ifp;
484 	mli->mli_version = MLD_VERSION_2;
485 	mli->mli_flags = 0;
486 	mli->mli_rv = MLD_RV_INIT;
487 	mli->mli_qi = MLD_QI_INIT;
488 	mli->mli_qri = MLD_QRI_INIT;
489 	mli->mli_uri = MLD_URI_INIT;
490 
491 	SLIST_INIT(&mli->mli_relinmhead);
492 
493 	/*
494 	 * Responses to general queries are subject to bounds.
495 	 */
496 	IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
497 
498 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
499 
500 	CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)",
501 	     ifp, ifp->if_xname);
502 
503 out:
504 	return (mli);
505 }
506 
507 /*
508  * Hook for ifdetach.
509  *
510  * NOTE: Some finalization tasks need to run before the protocol domain
511  * is detached, but also before the link layer does its cleanup.
512  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
513  *
514  * SMPng: Caller must hold IN6_MULTI_LOCK().
515  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
516  * XXX This routine is also bitten by unlocked ifma_protospec access.
517  */
518 void
519 mld_ifdetach(struct ifnet *ifp)
520 {
521 	struct mld_ifinfo	*mli;
522 	struct ifmultiaddr	*ifma;
523 	struct in6_multi	*inm, *tinm;
524 
525 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
526 	    ifp->if_xname);
527 
528 	IN6_MULTI_LOCK_ASSERT();
529 	MLD_LOCK();
530 
531 	mli = MLD_IFINFO(ifp);
532 	if (mli->mli_version == MLD_VERSION_2) {
533 		IF_ADDR_LOCK(ifp);
534 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
535 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
536 			    ifma->ifma_protospec == NULL)
537 				continue;
538 			inm = (struct in6_multi *)ifma->ifma_protospec;
539 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
540 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
541 				    inm, in6m_nrele);
542 			}
543 			in6m_clear_recorded(inm);
544 		}
545 		IF_ADDR_UNLOCK(ifp);
546 		SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
547 		    tinm) {
548 			SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
549 			in6m_release_locked(inm);
550 		}
551 	}
552 
553 	MLD_UNLOCK();
554 }
555 
556 /*
557  * Hook for domifdetach.
558  * Runs after link-layer cleanup; free MLD state.
559  *
560  * SMPng: Normally called with IF_AFDATA_LOCK held.
561  */
562 void
563 mld_domifdetach(struct ifnet *ifp)
564 {
565 
566 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
567 	    __func__, ifp, ifp->if_xname);
568 
569 	MLD_LOCK();
570 	mli_delete_locked(ifp);
571 	MLD_UNLOCK();
572 }
573 
574 static void
575 mli_delete_locked(const struct ifnet *ifp)
576 {
577 	struct mld_ifinfo *mli, *tmli;
578 
579 	CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)",
580 	    __func__, ifp, ifp->if_xname);
581 
582 	MLD_LOCK_ASSERT();
583 
584 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
585 		if (mli->mli_ifp == ifp) {
586 			/*
587 			 * Free deferred General Query responses.
588 			 */
589 			_IF_DRAIN(&mli->mli_gq);
590 
591 			LIST_REMOVE(mli, mli_link);
592 
593 			KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
594 			    ("%s: there are dangling in_multi references",
595 			    __func__));
596 
597 			free(mli, M_MLD);
598 			return;
599 		}
600 	}
601 #ifdef INVARIANTS
602 	panic("%s: mld_ifinfo not found for ifp %p\n", __func__,  ifp);
603 #endif
604 }
605 
606 /*
607  * Process a received MLDv1 general or address-specific query.
608  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
609  *
610  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
611  * mld_addr. This is OK as we own the mbuf chain.
612  */
613 static int
614 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
615     /*const*/ struct mld_hdr *mld)
616 {
617 	struct ifmultiaddr	*ifma;
618 	struct mld_ifinfo	*mli;
619 	struct in6_multi	*inm;
620 	int			 is_general_query;
621 	uint16_t		 timer;
622 #ifdef KTR
623 	char			 ip6tbuf[INET6_ADDRSTRLEN];
624 #endif
625 
626 	is_general_query = 0;
627 
628 	if (!mld_v1enable) {
629 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
630 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
631 		    ifp, ifp->if_xname);
632 		return (0);
633 	}
634 
635 	/*
636 	 * RFC3810 Section 6.2: MLD queries must originate from
637 	 * a router's link-local address.
638 	 */
639 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
640 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
641 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
642 		    ifp, ifp->if_xname);
643 		return (0);
644 	}
645 
646 	/*
647 	 * Do address field validation upfront before we accept
648 	 * the query.
649 	 */
650 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
651 		/*
652 		 * MLDv1 General Query.
653 		 * If this was not sent to the all-nodes group, ignore it.
654 		 */
655 		struct in6_addr		 dst;
656 
657 		dst = ip6->ip6_dst;
658 		in6_clearscope(&dst);
659 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
660 			return (EINVAL);
661 		is_general_query = 1;
662 	} else {
663 		/*
664 		 * Embed scope ID of receiving interface in MLD query for
665 		 * lookup whilst we don't hold other locks.
666 		 */
667 		in6_setscope(&mld->mld_addr, ifp, NULL);
668 	}
669 
670 	IN6_MULTI_LOCK();
671 	MLD_LOCK();
672 	IF_ADDR_LOCK(ifp);
673 
674 	/*
675 	 * Switch to MLDv1 host compatibility mode.
676 	 */
677 	mli = MLD_IFINFO(ifp);
678 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
679 	mld_set_version(mli, MLD_VERSION_1);
680 
681 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
682 	if (timer == 0)
683 		timer = 1;
684 
685 	if (is_general_query) {
686 		/*
687 		 * For each reporting group joined on this
688 		 * interface, kick the report timer.
689 		 */
690 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
691 		    ifp, ifp->if_xname);
692 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
693 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
694 			    ifma->ifma_protospec == NULL)
695 				continue;
696 			inm = (struct in6_multi *)ifma->ifma_protospec;
697 			mld_v1_update_group(inm, timer);
698 		}
699 	} else {
700 		/*
701 		 * MLDv1 Group-Specific Query.
702 		 * If this is a group-specific MLDv1 query, we need only
703 		 * look up the single group to process it.
704 		 */
705 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
706 		if (inm != NULL) {
707 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
708 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
709 			    ifp, ifp->if_xname);
710 			mld_v1_update_group(inm, timer);
711 		}
712 		/* XXX Clear embedded scope ID as userland won't expect it. */
713 		in6_clearscope(&mld->mld_addr);
714 	}
715 
716 	IF_ADDR_UNLOCK(ifp);
717 	MLD_UNLOCK();
718 	IN6_MULTI_UNLOCK();
719 
720 	return (0);
721 }
722 
723 /*
724  * Update the report timer on a group in response to an MLDv1 query.
725  *
726  * If we are becoming the reporting member for this group, start the timer.
727  * If we already are the reporting member for this group, and timer is
728  * below the threshold, reset it.
729  *
730  * We may be updating the group for the first time since we switched
731  * to MLDv2. If we are, then we must clear any recorded source lists,
732  * and transition to REPORTING state; the group timer is overloaded
733  * for group and group-source query responses.
734  *
735  * Unlike MLDv2, the delay per group should be jittered
736  * to avoid bursts of MLDv1 reports.
737  */
738 static void
739 mld_v1_update_group(struct in6_multi *inm, const int timer)
740 {
741 #ifdef KTR
742 	char			 ip6tbuf[INET6_ADDRSTRLEN];
743 #endif
744 
745 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
746 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
747 	    inm->in6m_ifp->if_xname, timer);
748 
749 	IN6_MULTI_LOCK_ASSERT();
750 
751 	switch (inm->in6m_state) {
752 	case MLD_NOT_MEMBER:
753 	case MLD_SILENT_MEMBER:
754 		break;
755 	case MLD_REPORTING_MEMBER:
756 		if (inm->in6m_timer != 0 &&
757 		    inm->in6m_timer <= timer) {
758 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
759 			    "skipping.", __func__);
760 			break;
761 		}
762 		/* FALLTHROUGH */
763 	case MLD_SG_QUERY_PENDING_MEMBER:
764 	case MLD_G_QUERY_PENDING_MEMBER:
765 	case MLD_IDLE_MEMBER:
766 	case MLD_LAZY_MEMBER:
767 	case MLD_AWAKENING_MEMBER:
768 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
769 		inm->in6m_state = MLD_REPORTING_MEMBER;
770 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
771 		V_current_state_timers_running6 = 1;
772 		break;
773 	case MLD_SLEEPING_MEMBER:
774 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
775 		inm->in6m_state = MLD_AWAKENING_MEMBER;
776 		break;
777 	case MLD_LEAVING_MEMBER:
778 		break;
779 	}
780 }
781 
782 /*
783  * Process a received MLDv2 general, group-specific or
784  * group-and-source-specific query.
785  *
786  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
787  *
788  * Return 0 if successful, otherwise an appropriate error code is returned.
789  */
790 static int
791 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
792     struct mbuf *m, const int off, const int icmp6len)
793 {
794 	struct mld_ifinfo	*mli;
795 	struct mldv2_query	*mld;
796 	struct in6_multi	*inm;
797 	uint32_t		 maxdelay, nsrc, qqi;
798 	int			 is_general_query;
799 	uint16_t		 timer;
800 	uint8_t			 qrv;
801 #ifdef KTR
802 	char			 ip6tbuf[INET6_ADDRSTRLEN];
803 #endif
804 
805 	is_general_query = 0;
806 
807 	/*
808 	 * RFC3810 Section 6.2: MLD queries must originate from
809 	 * a router's link-local address.
810 	 */
811 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
812 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
813 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
814 		    ifp, ifp->if_xname);
815 		return (0);
816 	}
817 
818 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname);
819 
820 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
821 
822 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
823 	if (maxdelay >= 32678) {
824 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
825 			   (MLD_MRC_EXP(maxdelay) + 3);
826 	}
827 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
828 	if (timer == 0)
829 		timer = 1;
830 
831 	qrv = MLD_QRV(mld->mld_misc);
832 	if (qrv < 2) {
833 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
834 		    qrv, MLD_RV_INIT);
835 		qrv = MLD_RV_INIT;
836 	}
837 
838 	qqi = mld->mld_qqi;
839 	if (qqi >= 128) {
840 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
841 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
842 	}
843 
844 	nsrc = ntohs(mld->mld_numsrc);
845 	if (nsrc > MLD_MAX_GS_SOURCES)
846 		return (EMSGSIZE);
847 	if (icmp6len < sizeof(struct mldv2_query) +
848 	    (nsrc * sizeof(struct in6_addr)))
849 		return (EMSGSIZE);
850 
851 	/*
852 	 * Do further input validation upfront to avoid resetting timers
853 	 * should we need to discard this query.
854 	 */
855 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
856 		/*
857 		 * General Queries SHOULD be directed to ff02::1.
858 		 * A general query with a source list has undefined
859 		 * behaviour; discard it.
860 		 */
861 		struct in6_addr		 dst;
862 
863 		dst = ip6->ip6_dst;
864 		in6_clearscope(&dst);
865 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) ||
866 		    nsrc > 0)
867 			return (EINVAL);
868 		is_general_query = 1;
869 	} else {
870 		/*
871 		 * Embed scope ID of receiving interface in MLD query for
872 		 * lookup whilst we don't hold other locks (due to KAME
873 		 * locking lameness). We own this mbuf chain just now.
874 		 */
875 		in6_setscope(&mld->mld_addr, ifp, NULL);
876 	}
877 
878 	IN6_MULTI_LOCK();
879 	MLD_LOCK();
880 	IF_ADDR_LOCK(ifp);
881 
882 	mli = MLD_IFINFO(ifp);
883 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
884 
885 	/*
886 	 * Discard the v2 query if we're in Compatibility Mode.
887 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
888 	 * until the Old Version Querier Present timer expires.
889 	 */
890 	if (mli->mli_version != MLD_VERSION_2)
891 		goto out_locked;
892 
893 	mld_set_version(mli, MLD_VERSION_2);
894 	mli->mli_rv = qrv;
895 	mli->mli_qi = qqi;
896 	mli->mli_qri = maxdelay;
897 
898 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
899 	    maxdelay);
900 
901 	if (is_general_query) {
902 		/*
903 		 * MLDv2 General Query.
904 		 *
905 		 * Schedule a current-state report on this ifp for
906 		 * all groups, possibly containing source lists.
907 		 *
908 		 * If there is a pending General Query response
909 		 * scheduled earlier than the selected delay, do
910 		 * not schedule any other reports.
911 		 * Otherwise, reset the interface timer.
912 		 */
913 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
914 		    ifp, ifp->if_xname);
915 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
916 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
917 			V_interface_timers_running6 = 1;
918 		}
919 	} else {
920 		/*
921 		 * MLDv2 Group-specific or Group-and-source-specific Query.
922 		 *
923 		 * Group-source-specific queries are throttled on
924 		 * a per-group basis to defeat denial-of-service attempts.
925 		 * Queries for groups we are not a member of on this
926 		 * link are simply ignored.
927 		 */
928 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
929 		if (inm == NULL)
930 			goto out_locked;
931 		if (nsrc > 0) {
932 			if (!ratecheck(&inm->in6m_lastgsrtv,
933 			    &V_mld_gsrdelay)) {
934 				CTR1(KTR_MLD, "%s: GS query throttled.",
935 				    __func__);
936 				goto out_locked;
937 			}
938 		}
939 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
940 		     ifp, ifp->if_xname);
941 		/*
942 		 * If there is a pending General Query response
943 		 * scheduled sooner than the selected delay, no
944 		 * further report need be scheduled.
945 		 * Otherwise, prepare to respond to the
946 		 * group-specific or group-and-source query.
947 		 */
948 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
949 			mld_v2_process_group_query(inm, mli, timer, m, off);
950 
951 		/* XXX Clear embedded scope ID as userland won't expect it. */
952 		in6_clearscope(&mld->mld_addr);
953 	}
954 
955 out_locked:
956 	IF_ADDR_UNLOCK(ifp);
957 	MLD_UNLOCK();
958 	IN6_MULTI_UNLOCK();
959 
960 	return (0);
961 }
962 
963 /*
964  * Process a recieved MLDv2 group-specific or group-and-source-specific
965  * query.
966  * Return <0 if any error occured. Currently this is ignored.
967  */
968 static int
969 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli,
970     int timer, struct mbuf *m0, const int off)
971 {
972 	struct mldv2_query	*mld;
973 	int			 retval;
974 	uint16_t		 nsrc;
975 
976 	IN6_MULTI_LOCK_ASSERT();
977 	MLD_LOCK_ASSERT();
978 
979 	retval = 0;
980 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
981 
982 	switch (inm->in6m_state) {
983 	case MLD_NOT_MEMBER:
984 	case MLD_SILENT_MEMBER:
985 	case MLD_SLEEPING_MEMBER:
986 	case MLD_LAZY_MEMBER:
987 	case MLD_AWAKENING_MEMBER:
988 	case MLD_IDLE_MEMBER:
989 	case MLD_LEAVING_MEMBER:
990 		return (retval);
991 		break;
992 	case MLD_REPORTING_MEMBER:
993 	case MLD_G_QUERY_PENDING_MEMBER:
994 	case MLD_SG_QUERY_PENDING_MEMBER:
995 		break;
996 	}
997 
998 	nsrc = ntohs(mld->mld_numsrc);
999 
1000 	/*
1001 	 * Deal with group-specific queries upfront.
1002 	 * If any group query is already pending, purge any recorded
1003 	 * source-list state if it exists, and schedule a query response
1004 	 * for this group-specific query.
1005 	 */
1006 	if (nsrc == 0) {
1007 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1008 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1009 			in6m_clear_recorded(inm);
1010 			timer = min(inm->in6m_timer, timer);
1011 		}
1012 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1013 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1014 		V_current_state_timers_running6 = 1;
1015 		return (retval);
1016 	}
1017 
1018 	/*
1019 	 * Deal with the case where a group-and-source-specific query has
1020 	 * been received but a group-specific query is already pending.
1021 	 */
1022 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1023 		timer = min(inm->in6m_timer, timer);
1024 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1025 		V_current_state_timers_running6 = 1;
1026 		return (retval);
1027 	}
1028 
1029 	/*
1030 	 * Finally, deal with the case where a group-and-source-specific
1031 	 * query has been received, where a response to a previous g-s-r
1032 	 * query exists, or none exists.
1033 	 * In this case, we need to parse the source-list which the Querier
1034 	 * has provided us with and check if we have any source list filter
1035 	 * entries at T1 for these sources. If we do not, there is no need
1036 	 * schedule a report and the query may be dropped.
1037 	 * If we do, we must record them and schedule a current-state
1038 	 * report for those sources.
1039 	 */
1040 	if (inm->in6m_nsrc > 0) {
1041 		struct mbuf		*m;
1042 		uint8_t			*sp;
1043 		int			 i, nrecorded;
1044 		int			 soff;
1045 
1046 		m = m0;
1047 		soff = off + sizeof(struct mldv2_query);
1048 		nrecorded = 0;
1049 		for (i = 0; i < nsrc; i++) {
1050 			sp = mtod(m, uint8_t *) + soff;
1051 			retval = in6m_record_source(inm,
1052 			    (const struct in6_addr *)sp);
1053 			if (retval < 0)
1054 				break;
1055 			nrecorded += retval;
1056 			soff += sizeof(struct in6_addr);
1057 			if (soff >= m->m_len) {
1058 				soff = soff - m->m_len;
1059 				m = m->m_next;
1060 				if (m == NULL)
1061 					break;
1062 			}
1063 		}
1064 		if (nrecorded > 0) {
1065 			CTR1(KTR_MLD,
1066 			    "%s: schedule response to SG query", __func__);
1067 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1068 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1069 			V_current_state_timers_running6 = 1;
1070 		}
1071 	}
1072 
1073 	return (retval);
1074 }
1075 
1076 /*
1077  * Process a received MLDv1 host membership report.
1078  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1079  *
1080  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1081  * mld_addr. This is OK as we own the mbuf chain.
1082  */
1083 static int
1084 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1085     /*const*/ struct mld_hdr *mld)
1086 {
1087 	struct in6_addr		 src, dst;
1088 	struct in6_ifaddr	*ia;
1089 	struct in6_multi	*inm;
1090 #ifdef KTR
1091 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1092 #endif
1093 
1094 	if (!mld_v1enable) {
1095 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1096 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1097 		    ifp, ifp->if_xname);
1098 		return (0);
1099 	}
1100 
1101 	if (ifp->if_flags & IFF_LOOPBACK)
1102 		return (0);
1103 
1104 	/*
1105 	 * MLDv1 reports must originate from a host's link-local address,
1106 	 * or the unspecified address (when booting).
1107 	 */
1108 	src = ip6->ip6_src;
1109 	in6_clearscope(&src);
1110 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1111 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1112 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1113 		    ifp, ifp->if_xname);
1114 		return (EINVAL);
1115 	}
1116 
1117 	/*
1118 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1119 	 * group, and must be directed to the group itself.
1120 	 */
1121 	dst = ip6->ip6_dst;
1122 	in6_clearscope(&dst);
1123 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1124 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1125 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1126 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1127 		    ifp, ifp->if_xname);
1128 		return (EINVAL);
1129 	}
1130 
1131 	/*
1132 	 * Make sure we don't hear our own membership report, as fast
1133 	 * leave requires knowing that we are the only member of a
1134 	 * group. Assume we used the link-local address if available,
1135 	 * otherwise look for ::.
1136 	 *
1137 	 * XXX Note that scope ID comparison is needed for the address
1138 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1139 	 * performed for the on-wire address.
1140 	 */
1141 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1142 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1143 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1144 		if (ia != NULL)
1145 			ifa_free(&ia->ia_ifa);
1146 		return (0);
1147 	}
1148 	if (ia != NULL)
1149 		ifa_free(&ia->ia_ifa);
1150 
1151 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1152 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname);
1153 
1154 	/*
1155 	 * Embed scope ID of receiving interface in MLD query for lookup
1156 	 * whilst we don't hold other locks (due to KAME locking lameness).
1157 	 */
1158 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1159 		in6_setscope(&mld->mld_addr, ifp, NULL);
1160 
1161 	IN6_MULTI_LOCK();
1162 	MLD_LOCK();
1163 	IF_ADDR_LOCK(ifp);
1164 
1165 	/*
1166 	 * MLDv1 report suppression.
1167 	 * If we are a member of this group, and our membership should be
1168 	 * reported, and our group timer is pending or about to be reset,
1169 	 * stop our group timer by transitioning to the 'lazy' state.
1170 	 */
1171 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1172 	if (inm != NULL) {
1173 		struct mld_ifinfo *mli;
1174 
1175 		mli = inm->in6m_mli;
1176 		KASSERT(mli != NULL,
1177 		    ("%s: no mli for ifp %p", __func__, ifp));
1178 
1179 		/*
1180 		 * If we are in MLDv2 host mode, do not allow the
1181 		 * other host's MLDv1 report to suppress our reports.
1182 		 */
1183 		if (mli->mli_version == MLD_VERSION_2)
1184 			goto out_locked;
1185 
1186 		inm->in6m_timer = 0;
1187 
1188 		switch (inm->in6m_state) {
1189 		case MLD_NOT_MEMBER:
1190 		case MLD_SILENT_MEMBER:
1191 		case MLD_SLEEPING_MEMBER:
1192 			break;
1193 		case MLD_REPORTING_MEMBER:
1194 		case MLD_IDLE_MEMBER:
1195 		case MLD_AWAKENING_MEMBER:
1196 			CTR3(KTR_MLD,
1197 			    "report suppressed for %s on ifp %p(%s)",
1198 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1199 			    ifp, ifp->if_xname);
1200 		case MLD_LAZY_MEMBER:
1201 			inm->in6m_state = MLD_LAZY_MEMBER;
1202 			break;
1203 		case MLD_G_QUERY_PENDING_MEMBER:
1204 		case MLD_SG_QUERY_PENDING_MEMBER:
1205 		case MLD_LEAVING_MEMBER:
1206 			break;
1207 		}
1208 	}
1209 
1210 out_locked:
1211 	MLD_UNLOCK();
1212 	IF_ADDR_UNLOCK(ifp);
1213 	IN6_MULTI_UNLOCK();
1214 
1215 	/* XXX Clear embedded scope ID as userland won't expect it. */
1216 	in6_clearscope(&mld->mld_addr);
1217 
1218 	return (0);
1219 }
1220 
1221 /*
1222  * MLD input path.
1223  *
1224  * Assume query messages which fit in a single ICMPv6 message header
1225  * have been pulled up.
1226  * Assume that userland will want to see the message, even if it
1227  * otherwise fails kernel input validation; do not free it.
1228  * Pullup may however free the mbuf chain m if it fails.
1229  *
1230  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1231  */
1232 int
1233 mld_input(struct mbuf *m, int off, int icmp6len)
1234 {
1235 	struct ifnet	*ifp;
1236 	struct ip6_hdr	*ip6;
1237 	struct mld_hdr	*mld;
1238 	int		 mldlen;
1239 
1240 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1241 
1242 	ifp = m->m_pkthdr.rcvif;
1243 
1244 	ip6 = mtod(m, struct ip6_hdr *);
1245 
1246 	/* Pullup to appropriate size. */
1247 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1248 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1249 	    icmp6len >= sizeof(struct mldv2_query)) {
1250 		mldlen = sizeof(struct mldv2_query);
1251 	} else {
1252 		mldlen = sizeof(struct mld_hdr);
1253 	}
1254 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1255 	if (mld == NULL) {
1256 		ICMP6STAT_INC(icp6s_badlen);
1257 		return (IPPROTO_DONE);
1258 	}
1259 
1260 	/*
1261 	 * Userland needs to see all of this traffic for implementing
1262 	 * the endpoint discovery portion of multicast routing.
1263 	 */
1264 	switch (mld->mld_type) {
1265 	case MLD_LISTENER_QUERY:
1266 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1267 		if (icmp6len == sizeof(struct mld_hdr)) {
1268 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1269 				return (0);
1270 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1271 			if (mld_v2_input_query(ifp, ip6, m, off,
1272 			    icmp6len) != 0)
1273 				return (0);
1274 		}
1275 		break;
1276 	case MLD_LISTENER_REPORT:
1277 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1278 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1279 			return (0);
1280 		break;
1281 	case MLDV2_LISTENER_REPORT:
1282 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1283 		break;
1284 	case MLD_LISTENER_DONE:
1285 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1286 		break;
1287 	default:
1288 		break;
1289 	}
1290 
1291 	return (0);
1292 }
1293 
1294 /*
1295  * Fast timeout handler (global).
1296  * VIMAGE: Timeout handlers are expected to service all vimages.
1297  */
1298 void
1299 mld_fasttimo(void)
1300 {
1301 	VNET_ITERATOR_DECL(vnet_iter);
1302 
1303 	VNET_LIST_RLOCK_NOSLEEP();
1304 	VNET_FOREACH(vnet_iter) {
1305 		CURVNET_SET(vnet_iter);
1306 		mld_fasttimo_vnet();
1307 		CURVNET_RESTORE();
1308 	}
1309 	VNET_LIST_RUNLOCK_NOSLEEP();
1310 }
1311 
1312 /*
1313  * Fast timeout handler (per-vnet).
1314  *
1315  * VIMAGE: Assume caller has set up our curvnet.
1316  */
1317 static void
1318 mld_fasttimo_vnet(void)
1319 {
1320 	struct ifqueue		 scq;	/* State-change packets */
1321 	struct ifqueue		 qrq;	/* Query response packets */
1322 	struct ifnet		*ifp;
1323 	struct mld_ifinfo	*mli;
1324 	struct ifmultiaddr	*ifma, *tifma;
1325 	struct in6_multi	*inm;
1326 	int			 uri_fasthz;
1327 
1328 	uri_fasthz = 0;
1329 
1330 	/*
1331 	 * Quick check to see if any work needs to be done, in order to
1332 	 * minimize the overhead of fasttimo processing.
1333 	 * SMPng: XXX Unlocked reads.
1334 	 */
1335 	if (!V_current_state_timers_running6 &&
1336 	    !V_interface_timers_running6 &&
1337 	    !V_state_change_timers_running6)
1338 		return;
1339 
1340 	IN6_MULTI_LOCK();
1341 	MLD_LOCK();
1342 
1343 	/*
1344 	 * MLDv2 General Query response timer processing.
1345 	 */
1346 	if (V_interface_timers_running6) {
1347 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1348 
1349 		V_interface_timers_running6 = 0;
1350 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1351 			if (mli->mli_v2_timer == 0) {
1352 				/* Do nothing. */
1353 			} else if (--mli->mli_v2_timer == 0) {
1354 				mld_v2_dispatch_general_query(mli);
1355 			} else {
1356 				V_interface_timers_running6 = 1;
1357 			}
1358 		}
1359 	}
1360 
1361 	if (!V_current_state_timers_running6 &&
1362 	    !V_state_change_timers_running6)
1363 		goto out_locked;
1364 
1365 	V_current_state_timers_running6 = 0;
1366 	V_state_change_timers_running6 = 0;
1367 
1368 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1369 
1370 	/*
1371 	 * MLD host report and state-change timer processing.
1372 	 * Note: Processing a v2 group timer may remove a node.
1373 	 */
1374 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1375 		ifp = mli->mli_ifp;
1376 
1377 		if (mli->mli_version == MLD_VERSION_2) {
1378 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1379 			    PR_FASTHZ);
1380 
1381 			memset(&qrq, 0, sizeof(struct ifqueue));
1382 			IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS);
1383 
1384 			memset(&scq, 0, sizeof(struct ifqueue));
1385 			IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1386 		}
1387 
1388 		IF_ADDR_LOCK(ifp);
1389 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1390 		    tifma) {
1391 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1392 			    ifma->ifma_protospec == NULL)
1393 				continue;
1394 			inm = (struct in6_multi *)ifma->ifma_protospec;
1395 			switch (mli->mli_version) {
1396 			case MLD_VERSION_1:
1397 				/*
1398 				 * XXX Drop IF_ADDR lock temporarily to
1399 				 * avoid recursion caused by a potential
1400 				 * call by in6ifa_ifpforlinklocal().
1401 				 * rwlock candidate?
1402 				 */
1403 				IF_ADDR_UNLOCK(ifp);
1404 				mld_v1_process_group_timer(inm,
1405 				    mli->mli_version);
1406 				IF_ADDR_LOCK(ifp);
1407 				break;
1408 			case MLD_VERSION_2:
1409 				mld_v2_process_group_timers(mli, &qrq,
1410 				    &scq, inm, uri_fasthz);
1411 				break;
1412 			}
1413 		}
1414 		IF_ADDR_UNLOCK(ifp);
1415 
1416 		if (mli->mli_version == MLD_VERSION_2) {
1417 			struct in6_multi		*tinm;
1418 
1419 			mld_dispatch_queue(&qrq, 0);
1420 			mld_dispatch_queue(&scq, 0);
1421 
1422 			/*
1423 			 * Free the in_multi reference(s) for
1424 			 * this lifecycle.
1425 			 */
1426 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1427 			    in6m_nrele, tinm) {
1428 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1429 				    in6m_nrele);
1430 				in6m_release_locked(inm);
1431 			}
1432 		}
1433 	}
1434 
1435 out_locked:
1436 	MLD_UNLOCK();
1437 	IN6_MULTI_UNLOCK();
1438 }
1439 
1440 /*
1441  * Update host report group timer.
1442  * Will update the global pending timer flags.
1443  */
1444 static void
1445 mld_v1_process_group_timer(struct in6_multi *inm, const int version)
1446 {
1447 	int report_timer_expired;
1448 
1449 	IN6_MULTI_LOCK_ASSERT();
1450 	MLD_LOCK_ASSERT();
1451 
1452 	if (inm->in6m_timer == 0) {
1453 		report_timer_expired = 0;
1454 	} else if (--inm->in6m_timer == 0) {
1455 		report_timer_expired = 1;
1456 	} else {
1457 		V_current_state_timers_running6 = 1;
1458 		return;
1459 	}
1460 
1461 	switch (inm->in6m_state) {
1462 	case MLD_NOT_MEMBER:
1463 	case MLD_SILENT_MEMBER:
1464 	case MLD_IDLE_MEMBER:
1465 	case MLD_LAZY_MEMBER:
1466 	case MLD_SLEEPING_MEMBER:
1467 	case MLD_AWAKENING_MEMBER:
1468 		break;
1469 	case MLD_REPORTING_MEMBER:
1470 		if (report_timer_expired) {
1471 			inm->in6m_state = MLD_IDLE_MEMBER;
1472 			(void)mld_v1_transmit_report(inm,
1473 			     MLD_LISTENER_REPORT);
1474 		}
1475 		break;
1476 	case MLD_G_QUERY_PENDING_MEMBER:
1477 	case MLD_SG_QUERY_PENDING_MEMBER:
1478 	case MLD_LEAVING_MEMBER:
1479 		break;
1480 	}
1481 }
1482 
1483 /*
1484  * Update a group's timers for MLDv2.
1485  * Will update the global pending timer flags.
1486  * Note: Unlocked read from mli.
1487  */
1488 static void
1489 mld_v2_process_group_timers(struct mld_ifinfo *mli,
1490     struct ifqueue *qrq, struct ifqueue *scq,
1491     struct in6_multi *inm, const int uri_fasthz)
1492 {
1493 	int query_response_timer_expired;
1494 	int state_change_retransmit_timer_expired;
1495 #ifdef KTR
1496 	char ip6tbuf[INET6_ADDRSTRLEN];
1497 #endif
1498 
1499 	IN6_MULTI_LOCK_ASSERT();
1500 	MLD_LOCK_ASSERT();
1501 
1502 	query_response_timer_expired = 0;
1503 	state_change_retransmit_timer_expired = 0;
1504 
1505 	/*
1506 	 * During a transition from compatibility mode back to MLDv2,
1507 	 * a group record in REPORTING state may still have its group
1508 	 * timer active. This is a no-op in this function; it is easier
1509 	 * to deal with it here than to complicate the slow-timeout path.
1510 	 */
1511 	if (inm->in6m_timer == 0) {
1512 		query_response_timer_expired = 0;
1513 	} else if (--inm->in6m_timer == 0) {
1514 		query_response_timer_expired = 1;
1515 	} else {
1516 		V_current_state_timers_running6 = 1;
1517 	}
1518 
1519 	if (inm->in6m_sctimer == 0) {
1520 		state_change_retransmit_timer_expired = 0;
1521 	} else if (--inm->in6m_sctimer == 0) {
1522 		state_change_retransmit_timer_expired = 1;
1523 	} else {
1524 		V_state_change_timers_running6 = 1;
1525 	}
1526 
1527 	/* We are in fasttimo, so be quick about it. */
1528 	if (!state_change_retransmit_timer_expired &&
1529 	    !query_response_timer_expired)
1530 		return;
1531 
1532 	switch (inm->in6m_state) {
1533 	case MLD_NOT_MEMBER:
1534 	case MLD_SILENT_MEMBER:
1535 	case MLD_SLEEPING_MEMBER:
1536 	case MLD_LAZY_MEMBER:
1537 	case MLD_AWAKENING_MEMBER:
1538 	case MLD_IDLE_MEMBER:
1539 		break;
1540 	case MLD_G_QUERY_PENDING_MEMBER:
1541 	case MLD_SG_QUERY_PENDING_MEMBER:
1542 		/*
1543 		 * Respond to a previously pending Group-Specific
1544 		 * or Group-and-Source-Specific query by enqueueing
1545 		 * the appropriate Current-State report for
1546 		 * immediate transmission.
1547 		 */
1548 		if (query_response_timer_expired) {
1549 			int retval;
1550 
1551 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1552 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER));
1553 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1554 			    __func__, retval);
1555 			inm->in6m_state = MLD_REPORTING_MEMBER;
1556 			in6m_clear_recorded(inm);
1557 		}
1558 		/* FALLTHROUGH */
1559 	case MLD_REPORTING_MEMBER:
1560 	case MLD_LEAVING_MEMBER:
1561 		if (state_change_retransmit_timer_expired) {
1562 			/*
1563 			 * State-change retransmission timer fired.
1564 			 * If there are any further pending retransmissions,
1565 			 * set the global pending state-change flag, and
1566 			 * reset the timer.
1567 			 */
1568 			if (--inm->in6m_scrv > 0) {
1569 				inm->in6m_sctimer = uri_fasthz;
1570 				V_state_change_timers_running6 = 1;
1571 			}
1572 			/*
1573 			 * Retransmit the previously computed state-change
1574 			 * report. If there are no further pending
1575 			 * retransmissions, the mbuf queue will be consumed.
1576 			 * Update T0 state to T1 as we have now sent
1577 			 * a state-change.
1578 			 */
1579 			(void)mld_v2_merge_state_changes(inm, scq);
1580 
1581 			in6m_commit(inm);
1582 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1583 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1584 			    inm->in6m_ifp->if_xname);
1585 
1586 			/*
1587 			 * If we are leaving the group for good, make sure
1588 			 * we release MLD's reference to it.
1589 			 * This release must be deferred using a SLIST,
1590 			 * as we are called from a loop which traverses
1591 			 * the in_ifmultiaddr TAILQ.
1592 			 */
1593 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1594 			    inm->in6m_scrv == 0) {
1595 				inm->in6m_state = MLD_NOT_MEMBER;
1596 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1597 				    inm, in6m_nrele);
1598 			}
1599 		}
1600 		break;
1601 	}
1602 }
1603 
1604 /*
1605  * Switch to a different version on the given interface,
1606  * as per Section 9.12.
1607  */
1608 static void
1609 mld_set_version(struct mld_ifinfo *mli, const int version)
1610 {
1611 	int old_version_timer;
1612 
1613 	MLD_LOCK_ASSERT();
1614 
1615 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1616 	    version, mli->mli_ifp, mli->mli_ifp->if_xname);
1617 
1618 	if (version == MLD_VERSION_1) {
1619 		/*
1620 		 * Compute the "Older Version Querier Present" timer as per
1621 		 * Section 9.12.
1622 		 */
1623 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1624 		old_version_timer *= PR_SLOWHZ;
1625 		mli->mli_v1_timer = old_version_timer;
1626 	}
1627 
1628 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1629 		mli->mli_version = MLD_VERSION_1;
1630 		mld_v2_cancel_link_timers(mli);
1631 	}
1632 }
1633 
1634 /*
1635  * Cancel pending MLDv2 timers for the given link and all groups
1636  * joined on it; state-change, general-query, and group-query timers.
1637  */
1638 static void
1639 mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
1640 {
1641 	struct ifmultiaddr	*ifma;
1642 	struct ifnet		*ifp;
1643 	struct in6_multi		*inm;
1644 
1645 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1646 	    mli->mli_ifp, mli->mli_ifp->if_xname);
1647 
1648 	IN6_MULTI_LOCK_ASSERT();
1649 	MLD_LOCK_ASSERT();
1650 
1651 	/*
1652 	 * Fast-track this potentially expensive operation
1653 	 * by checking all the global 'timer pending' flags.
1654 	 */
1655 	if (!V_interface_timers_running6 &&
1656 	    !V_state_change_timers_running6 &&
1657 	    !V_current_state_timers_running6)
1658 		return;
1659 
1660 	mli->mli_v2_timer = 0;
1661 
1662 	ifp = mli->mli_ifp;
1663 
1664 	IF_ADDR_LOCK(ifp);
1665 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1666 		if (ifma->ifma_addr->sa_family != AF_INET6)
1667 			continue;
1668 		inm = (struct in6_multi *)ifma->ifma_protospec;
1669 		switch (inm->in6m_state) {
1670 		case MLD_NOT_MEMBER:
1671 		case MLD_SILENT_MEMBER:
1672 		case MLD_IDLE_MEMBER:
1673 		case MLD_LAZY_MEMBER:
1674 		case MLD_SLEEPING_MEMBER:
1675 		case MLD_AWAKENING_MEMBER:
1676 			break;
1677 		case MLD_LEAVING_MEMBER:
1678 			/*
1679 			 * If we are leaving the group and switching
1680 			 * version, we need to release the final
1681 			 * reference held for issuing the INCLUDE {}.
1682 			 *
1683 			 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
1684 			 * around in6m_release_locked(), as it is not
1685 			 * a recursive mutex.
1686 			 */
1687 			IF_ADDR_UNLOCK(ifp);
1688 			in6m_release_locked(inm);
1689 			IF_ADDR_LOCK(ifp);
1690 			/* FALLTHROUGH */
1691 		case MLD_G_QUERY_PENDING_MEMBER:
1692 		case MLD_SG_QUERY_PENDING_MEMBER:
1693 			in6m_clear_recorded(inm);
1694 			/* FALLTHROUGH */
1695 		case MLD_REPORTING_MEMBER:
1696 			inm->in6m_sctimer = 0;
1697 			inm->in6m_timer = 0;
1698 			inm->in6m_state = MLD_REPORTING_MEMBER;
1699 			/*
1700 			 * Free any pending MLDv2 state-change records.
1701 			 */
1702 			_IF_DRAIN(&inm->in6m_scq);
1703 			break;
1704 		}
1705 	}
1706 	IF_ADDR_UNLOCK(ifp);
1707 }
1708 
1709 /*
1710  * Global slowtimo handler.
1711  * VIMAGE: Timeout handlers are expected to service all vimages.
1712  */
1713 void
1714 mld_slowtimo(void)
1715 {
1716 	VNET_ITERATOR_DECL(vnet_iter);
1717 
1718 	VNET_LIST_RLOCK_NOSLEEP();
1719 	VNET_FOREACH(vnet_iter) {
1720 		CURVNET_SET(vnet_iter);
1721 		mld_slowtimo_vnet();
1722 		CURVNET_RESTORE();
1723 	}
1724 	VNET_LIST_RUNLOCK_NOSLEEP();
1725 }
1726 
1727 /*
1728  * Per-vnet slowtimo handler.
1729  */
1730 static void
1731 mld_slowtimo_vnet(void)
1732 {
1733 	struct mld_ifinfo *mli;
1734 
1735 	MLD_LOCK();
1736 
1737 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1738 		mld_v1_process_querier_timers(mli);
1739 	}
1740 
1741 	MLD_UNLOCK();
1742 }
1743 
1744 /*
1745  * Update the Older Version Querier Present timers for a link.
1746  * See Section 9.12 of RFC 3810.
1747  */
1748 static void
1749 mld_v1_process_querier_timers(struct mld_ifinfo *mli)
1750 {
1751 
1752 	MLD_LOCK_ASSERT();
1753 
1754 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1755 		/*
1756 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1757 		 */
1758 		CTR5(KTR_MLD,
1759 		    "%s: transition from v%d -> v%d on %p(%s)",
1760 		    __func__, mli->mli_version, MLD_VERSION_2,
1761 		    mli->mli_ifp, mli->mli_ifp->if_xname);
1762 		mli->mli_version = MLD_VERSION_2;
1763 	}
1764 }
1765 
1766 /*
1767  * Transmit an MLDv1 report immediately.
1768  */
1769 static int
1770 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1771 {
1772 	struct ifnet		*ifp;
1773 	struct in6_ifaddr	*ia;
1774 	struct ip6_hdr		*ip6;
1775 	struct mbuf		*mh, *md;
1776 	struct mld_hdr		*mld;
1777 
1778 	IN6_MULTI_LOCK_ASSERT();
1779 	MLD_LOCK_ASSERT();
1780 
1781 	ifp = in6m->in6m_ifp;
1782 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1783 	/* ia may be NULL if link-local address is tentative. */
1784 
1785 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
1786 	if (mh == NULL) {
1787 		if (ia != NULL)
1788 			ifa_free(&ia->ia_ifa);
1789 		return (ENOMEM);
1790 	}
1791 	MGET(md, M_DONTWAIT, MT_DATA);
1792 	if (md == NULL) {
1793 		m_free(mh);
1794 		if (ia != NULL)
1795 			ifa_free(&ia->ia_ifa);
1796 		return (ENOMEM);
1797 	}
1798 	mh->m_next = md;
1799 
1800 	/*
1801 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1802 	 * that ether_output() does not need to allocate another mbuf
1803 	 * for the header in the most common case.
1804 	 */
1805 	MH_ALIGN(mh, sizeof(struct ip6_hdr));
1806 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1807 	mh->m_len = sizeof(struct ip6_hdr);
1808 
1809 	ip6 = mtod(mh, struct ip6_hdr *);
1810 	ip6->ip6_flow = 0;
1811 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1812 	ip6->ip6_vfc |= IPV6_VERSION;
1813 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1814 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1815 	ip6->ip6_dst = in6m->in6m_addr;
1816 
1817 	md->m_len = sizeof(struct mld_hdr);
1818 	mld = mtod(md, struct mld_hdr *);
1819 	mld->mld_type = type;
1820 	mld->mld_code = 0;
1821 	mld->mld_cksum = 0;
1822 	mld->mld_maxdelay = 0;
1823 	mld->mld_reserved = 0;
1824 	mld->mld_addr = in6m->in6m_addr;
1825 	in6_clearscope(&mld->mld_addr);
1826 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1827 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1828 
1829 	mld_save_context(mh, ifp);
1830 	mh->m_flags |= M_MLDV1;
1831 
1832 	mld_dispatch_packet(mh);
1833 
1834 	if (ia != NULL)
1835 		ifa_free(&ia->ia_ifa);
1836 	return (0);
1837 }
1838 
1839 /*
1840  * Process a state change from the upper layer for the given IPv6 group.
1841  *
1842  * Each socket holds a reference on the in_multi in its own ip_moptions.
1843  * The socket layer will have made the necessary updates to.the group
1844  * state, it is now up to MLD to issue a state change report if there
1845  * has been any change between T0 (when the last state-change was issued)
1846  * and T1 (now).
1847  *
1848  * We use the MLDv2 state machine at group level. The MLd module
1849  * however makes the decision as to which MLD protocol version to speak.
1850  * A state change *from* INCLUDE {} always means an initial join.
1851  * A state change *to* INCLUDE {} always means a final leave.
1852  *
1853  * If delay is non-zero, and the state change is an initial multicast
1854  * join, the state change report will be delayed by 'delay' ticks
1855  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1856  * the initial MLDv2 state change report will be delayed by whichever
1857  * is sooner, a pending state-change timer or delay itself.
1858  *
1859  * VIMAGE: curvnet should have been set by caller, as this routine
1860  * is called from the socket option handlers.
1861  */
1862 int
1863 mld_change_state(struct in6_multi *inm, const int delay)
1864 {
1865 	struct mld_ifinfo *mli;
1866 	struct ifnet *ifp;
1867 	int error;
1868 
1869 	IN6_MULTI_LOCK_ASSERT();
1870 
1871 	error = 0;
1872 
1873 	/*
1874 	 * Try to detect if the upper layer just asked us to change state
1875 	 * for an interface which has now gone away.
1876 	 */
1877 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1878 	ifp = inm->in6m_ifma->ifma_ifp;
1879 	if (ifp != NULL) {
1880 		/*
1881 		 * Sanity check that netinet6's notion of ifp is the
1882 		 * same as net's.
1883 		 */
1884 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1885 	}
1886 
1887 	MLD_LOCK();
1888 
1889 	mli = MLD_IFINFO(ifp);
1890 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
1891 
1892 	/*
1893 	 * If we detect a state transition to or from MCAST_UNDEFINED
1894 	 * for this group, then we are starting or finishing an MLD
1895 	 * life cycle for this group.
1896 	 */
1897 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1898 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1899 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1900 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1901 			CTR1(KTR_MLD, "%s: initial join", __func__);
1902 			error = mld_initial_join(inm, mli, delay);
1903 			goto out_locked;
1904 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1905 			CTR1(KTR_MLD, "%s: final leave", __func__);
1906 			mld_final_leave(inm, mli);
1907 			goto out_locked;
1908 		}
1909 	} else {
1910 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1911 	}
1912 
1913 	error = mld_handle_state_change(inm, mli);
1914 
1915 out_locked:
1916 	MLD_UNLOCK();
1917 	return (error);
1918 }
1919 
1920 /*
1921  * Perform the initial join for an MLD group.
1922  *
1923  * When joining a group:
1924  *  If the group should have its MLD traffic suppressed, do nothing.
1925  *  MLDv1 starts sending MLDv1 host membership reports.
1926  *  MLDv2 will schedule an MLDv2 state-change report containing the
1927  *  initial state of the membership.
1928  *
1929  * If the delay argument is non-zero, then we must delay sending the
1930  * initial state change for delay ticks (in units of PR_FASTHZ).
1931  */
1932 static int
1933 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
1934     const int delay)
1935 {
1936 	struct ifnet		*ifp;
1937 	struct ifqueue		*ifq;
1938 	int			 error, retval, syncstates;
1939 	int			 odelay;
1940 #ifdef KTR
1941 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1942 #endif
1943 
1944 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1945 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1946 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
1947 
1948 	error = 0;
1949 	syncstates = 1;
1950 
1951 	ifp = inm->in6m_ifp;
1952 
1953 	IN6_MULTI_LOCK_ASSERT();
1954 	MLD_LOCK_ASSERT();
1955 
1956 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1957 
1958 	/*
1959 	 * Groups joined on loopback or marked as 'not reported',
1960 	 * enter the MLD_SILENT_MEMBER state and
1961 	 * are never reported in any protocol exchanges.
1962 	 * All other groups enter the appropriate state machine
1963 	 * for the version in use on this link.
1964 	 * A link marked as MLIF_SILENT causes MLD to be completely
1965 	 * disabled for the link.
1966 	 */
1967 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1968 	    (mli->mli_flags & MLIF_SILENT) ||
1969 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1970 		CTR1(KTR_MLD,
1971 "%s: not kicking state machine for silent group", __func__);
1972 		inm->in6m_state = MLD_SILENT_MEMBER;
1973 		inm->in6m_timer = 0;
1974 	} else {
1975 		/*
1976 		 * Deal with overlapping in_multi lifecycle.
1977 		 * If this group was LEAVING, then make sure
1978 		 * we drop the reference we picked up to keep the
1979 		 * group around for the final INCLUDE {} enqueue.
1980 		 */
1981 		if (mli->mli_version == MLD_VERSION_2 &&
1982 		    inm->in6m_state == MLD_LEAVING_MEMBER)
1983 			in6m_release_locked(inm);
1984 
1985 		inm->in6m_state = MLD_REPORTING_MEMBER;
1986 
1987 		switch (mli->mli_version) {
1988 		case MLD_VERSION_1:
1989 			/*
1990 			 * If a delay was provided, only use it if
1991 			 * it is greater than the delay normally
1992 			 * used for an MLDv1 state change report,
1993 			 * and delay sending the initial MLDv1 report
1994 			 * by not transitioning to the IDLE state.
1995 			 */
1996 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
1997 			if (delay) {
1998 				inm->in6m_timer = max(delay, odelay);
1999 				V_current_state_timers_running6 = 1;
2000 			} else {
2001 				inm->in6m_state = MLD_IDLE_MEMBER;
2002 				error = mld_v1_transmit_report(inm,
2003 				     MLD_LISTENER_REPORT);
2004 				if (error == 0) {
2005 					inm->in6m_timer = odelay;
2006 					V_current_state_timers_running6 = 1;
2007 				}
2008 			}
2009 			break;
2010 
2011 		case MLD_VERSION_2:
2012 			/*
2013 			 * Defer update of T0 to T1, until the first copy
2014 			 * of the state change has been transmitted.
2015 			 */
2016 			syncstates = 0;
2017 
2018 			/*
2019 			 * Immediately enqueue a State-Change Report for
2020 			 * this interface, freeing any previous reports.
2021 			 * Don't kick the timers if there is nothing to do,
2022 			 * or if an error occurred.
2023 			 */
2024 			ifq = &inm->in6m_scq;
2025 			_IF_DRAIN(ifq);
2026 			retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2027 			    0, 0);
2028 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2029 			    __func__, retval);
2030 			if (retval <= 0) {
2031 				error = retval * -1;
2032 				break;
2033 			}
2034 
2035 			/*
2036 			 * Schedule transmission of pending state-change
2037 			 * report up to RV times for this link. The timer
2038 			 * will fire at the next mld_fasttimo (~200ms),
2039 			 * giving us an opportunity to merge the reports.
2040 			 *
2041 			 * If a delay was provided to this function, only
2042 			 * use this delay if sooner than the existing one.
2043 			 */
2044 			KASSERT(mli->mli_rv > 1,
2045 			   ("%s: invalid robustness %d", __func__,
2046 			    mli->mli_rv));
2047 			inm->in6m_scrv = mli->mli_rv;
2048 			if (delay) {
2049 				if (inm->in6m_sctimer > 1) {
2050 					inm->in6m_sctimer =
2051 					    min(inm->in6m_sctimer, delay);
2052 				} else
2053 					inm->in6m_sctimer = delay;
2054 			} else
2055 				inm->in6m_sctimer = 1;
2056 			V_state_change_timers_running6 = 1;
2057 
2058 			error = 0;
2059 			break;
2060 		}
2061 	}
2062 
2063 	/*
2064 	 * Only update the T0 state if state change is atomic,
2065 	 * i.e. we don't need to wait for a timer to fire before we
2066 	 * can consider the state change to have been communicated.
2067 	 */
2068 	if (syncstates) {
2069 		in6m_commit(inm);
2070 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2071 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2072 		    inm->in6m_ifp->if_xname);
2073 	}
2074 
2075 	return (error);
2076 }
2077 
2078 /*
2079  * Issue an intermediate state change during the life-cycle.
2080  */
2081 static int
2082 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli)
2083 {
2084 	struct ifnet		*ifp;
2085 	int			 retval;
2086 #ifdef KTR
2087 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2088 #endif
2089 
2090 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2091 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2092 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
2093 
2094 	ifp = inm->in6m_ifp;
2095 
2096 	IN6_MULTI_LOCK_ASSERT();
2097 	MLD_LOCK_ASSERT();
2098 
2099 	KASSERT(mli && mli->mli_ifp == ifp,
2100 	    ("%s: inconsistent ifp", __func__));
2101 
2102 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2103 	    (mli->mli_flags & MLIF_SILENT) ||
2104 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2105 	    (mli->mli_version != MLD_VERSION_2)) {
2106 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2107 			CTR1(KTR_MLD,
2108 "%s: not kicking state machine for silent group", __func__);
2109 		}
2110 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2111 		in6m_commit(inm);
2112 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2113 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2114 		    inm->in6m_ifp->if_xname);
2115 		return (0);
2116 	}
2117 
2118 	_IF_DRAIN(&inm->in6m_scq);
2119 
2120 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0);
2121 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2122 	if (retval <= 0)
2123 		return (-retval);
2124 
2125 	/*
2126 	 * If record(s) were enqueued, start the state-change
2127 	 * report timer for this group.
2128 	 */
2129 	inm->in6m_scrv = mli->mli_rv;
2130 	inm->in6m_sctimer = 1;
2131 	V_state_change_timers_running6 = 1;
2132 
2133 	return (0);
2134 }
2135 
2136 /*
2137  * Perform the final leave for a multicast address.
2138  *
2139  * When leaving a group:
2140  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2141  *  MLDv2 enqueues a state-change report containing a transition
2142  *  to INCLUDE {} for immediate transmission.
2143  */
2144 static void
2145 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli)
2146 {
2147 	int syncstates;
2148 #ifdef KTR
2149 	char ip6tbuf[INET6_ADDRSTRLEN];
2150 #endif
2151 
2152 	syncstates = 1;
2153 
2154 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2155 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2156 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
2157 
2158 	IN6_MULTI_LOCK_ASSERT();
2159 	MLD_LOCK_ASSERT();
2160 
2161 	switch (inm->in6m_state) {
2162 	case MLD_NOT_MEMBER:
2163 	case MLD_SILENT_MEMBER:
2164 	case MLD_LEAVING_MEMBER:
2165 		/* Already leaving or left; do nothing. */
2166 		CTR1(KTR_MLD,
2167 "%s: not kicking state machine for silent group", __func__);
2168 		break;
2169 	case MLD_REPORTING_MEMBER:
2170 	case MLD_IDLE_MEMBER:
2171 	case MLD_G_QUERY_PENDING_MEMBER:
2172 	case MLD_SG_QUERY_PENDING_MEMBER:
2173 		if (mli->mli_version == MLD_VERSION_1) {
2174 #ifdef INVARIANTS
2175 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2176 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2177 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2178 			     __func__);
2179 #endif
2180 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2181 			inm->in6m_state = MLD_NOT_MEMBER;
2182 		} else if (mli->mli_version == MLD_VERSION_2) {
2183 			/*
2184 			 * Stop group timer and all pending reports.
2185 			 * Immediately enqueue a state-change report
2186 			 * TO_IN {} to be sent on the next fast timeout,
2187 			 * giving us an opportunity to merge reports.
2188 			 */
2189 			_IF_DRAIN(&inm->in6m_scq);
2190 			inm->in6m_timer = 0;
2191 			inm->in6m_scrv = mli->mli_rv;
2192 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2193 			    "pending retransmissions.", __func__,
2194 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2195 			    inm->in6m_ifp->if_xname, inm->in6m_scrv);
2196 			if (inm->in6m_scrv == 0) {
2197 				inm->in6m_state = MLD_NOT_MEMBER;
2198 				inm->in6m_sctimer = 0;
2199 			} else {
2200 				int retval;
2201 
2202 				in6m_acquire_locked(inm);
2203 
2204 				retval = mld_v2_enqueue_group_record(
2205 				    &inm->in6m_scq, inm, 1, 0, 0);
2206 				KASSERT(retval != 0,
2207 				    ("%s: enqueue record = %d", __func__,
2208 				     retval));
2209 
2210 				inm->in6m_state = MLD_LEAVING_MEMBER;
2211 				inm->in6m_sctimer = 1;
2212 				V_state_change_timers_running6 = 1;
2213 				syncstates = 0;
2214 			}
2215 			break;
2216 		}
2217 		break;
2218 	case MLD_LAZY_MEMBER:
2219 	case MLD_SLEEPING_MEMBER:
2220 	case MLD_AWAKENING_MEMBER:
2221 		/* Our reports are suppressed; do nothing. */
2222 		break;
2223 	}
2224 
2225 	if (syncstates) {
2226 		in6m_commit(inm);
2227 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2228 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2229 		    inm->in6m_ifp->if_xname);
2230 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2231 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2232 		    __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname);
2233 	}
2234 }
2235 
2236 /*
2237  * Enqueue an MLDv2 group record to the given output queue.
2238  *
2239  * If is_state_change is zero, a current-state record is appended.
2240  * If is_state_change is non-zero, a state-change report is appended.
2241  *
2242  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2243  * If is_group_query is zero, and if there is a packet with free space
2244  * at the tail of the queue, it will be appended to providing there
2245  * is enough free space.
2246  * Otherwise a new mbuf packet chain is allocated.
2247  *
2248  * If is_source_query is non-zero, each source is checked to see if
2249  * it was recorded for a Group-Source query, and will be omitted if
2250  * it is not both in-mode and recorded.
2251  *
2252  * The function will attempt to allocate leading space in the packet
2253  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2254  *
2255  * If successful the size of all data appended to the queue is returned,
2256  * otherwise an error code less than zero is returned, or zero if
2257  * no record(s) were appended.
2258  */
2259 static int
2260 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2261     const int is_state_change, const int is_group_query,
2262     const int is_source_query)
2263 {
2264 	struct mldv2_record	 mr;
2265 	struct mldv2_record	*pmr;
2266 	struct ifnet		*ifp;
2267 	struct ip6_msource	*ims, *nims;
2268 	struct mbuf		*m0, *m, *md;
2269 	int			 error, is_filter_list_change;
2270 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2271 	int			 record_has_sources;
2272 	int			 now;
2273 	int			 type;
2274 	uint8_t			 mode;
2275 #ifdef KTR
2276 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2277 #endif
2278 
2279 	IN6_MULTI_LOCK_ASSERT();
2280 
2281 	error = 0;
2282 	ifp = inm->in6m_ifp;
2283 	is_filter_list_change = 0;
2284 	m = NULL;
2285 	m0 = NULL;
2286 	m0srcs = 0;
2287 	msrcs = 0;
2288 	nbytes = 0;
2289 	nims = NULL;
2290 	record_has_sources = 1;
2291 	pmr = NULL;
2292 	type = MLD_DO_NOTHING;
2293 	mode = inm->in6m_st[1].iss_fmode;
2294 
2295 	/*
2296 	 * If we did not transition out of ASM mode during t0->t1,
2297 	 * and there are no source nodes to process, we can skip
2298 	 * the generation of source records.
2299 	 */
2300 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2301 	    inm->in6m_nsrc == 0)
2302 		record_has_sources = 0;
2303 
2304 	if (is_state_change) {
2305 		/*
2306 		 * Queue a state change record.
2307 		 * If the mode did not change, and there are non-ASM
2308 		 * listeners or source filters present,
2309 		 * we potentially need to issue two records for the group.
2310 		 * If we are transitioning to MCAST_UNDEFINED, we need
2311 		 * not send any sources.
2312 		 * If there are ASM listeners, and there was no filter
2313 		 * mode transition of any kind, do nothing.
2314 		 */
2315 		if (mode != inm->in6m_st[0].iss_fmode) {
2316 			if (mode == MCAST_EXCLUDE) {
2317 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2318 				    __func__);
2319 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2320 			} else {
2321 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2322 				    __func__);
2323 				type = MLD_CHANGE_TO_INCLUDE_MODE;
2324 				if (mode == MCAST_UNDEFINED)
2325 					record_has_sources = 0;
2326 			}
2327 		} else {
2328 			if (record_has_sources) {
2329 				is_filter_list_change = 1;
2330 			} else {
2331 				type = MLD_DO_NOTHING;
2332 			}
2333 		}
2334 	} else {
2335 		/*
2336 		 * Queue a current state record.
2337 		 */
2338 		if (mode == MCAST_EXCLUDE) {
2339 			type = MLD_MODE_IS_EXCLUDE;
2340 		} else if (mode == MCAST_INCLUDE) {
2341 			type = MLD_MODE_IS_INCLUDE;
2342 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2343 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2344 			     __func__, inm, inm->in6m_st[1].iss_asm));
2345 		}
2346 	}
2347 
2348 	/*
2349 	 * Generate the filter list changes using a separate function.
2350 	 */
2351 	if (is_filter_list_change)
2352 		return (mld_v2_enqueue_filter_change(ifq, inm));
2353 
2354 	if (type == MLD_DO_NOTHING) {
2355 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2356 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2357 		    inm->in6m_ifp->if_xname);
2358 		return (0);
2359 	}
2360 
2361 	/*
2362 	 * If any sources are present, we must be able to fit at least
2363 	 * one in the trailing space of the tail packet's mbuf,
2364 	 * ideally more.
2365 	 */
2366 	minrec0len = sizeof(struct mldv2_record);
2367 	if (record_has_sources)
2368 		minrec0len += sizeof(struct in6_addr);
2369 
2370 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2371 	    mld_rec_type_to_str(type),
2372 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2373 	    inm->in6m_ifp->if_xname);
2374 
2375 	/*
2376 	 * Check if we have a packet in the tail of the queue for this
2377 	 * group into which the first group record for this group will fit.
2378 	 * Otherwise allocate a new packet.
2379 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2380 	 * Note: Group records for G/GSR query responses MUST be sent
2381 	 * in their own packet.
2382 	 */
2383 	m0 = ifq->ifq_tail;
2384 	if (!is_group_query &&
2385 	    m0 != NULL &&
2386 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2387 	    (m0->m_pkthdr.len + minrec0len) <
2388 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2389 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2390 			    sizeof(struct mldv2_record)) /
2391 			    sizeof(struct in6_addr);
2392 		m = m0;
2393 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2394 	} else {
2395 		if (_IF_QFULL(ifq)) {
2396 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2397 			return (-ENOMEM);
2398 		}
2399 		m = NULL;
2400 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2401 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2402 		if (!is_state_change && !is_group_query)
2403 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2404 		if (m == NULL)
2405 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2406 		if (m == NULL)
2407 			return (-ENOMEM);
2408 
2409 		mld_save_context(m, ifp);
2410 
2411 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2412 	}
2413 
2414 	/*
2415 	 * Append group record.
2416 	 * If we have sources, we don't know how many yet.
2417 	 */
2418 	mr.mr_type = type;
2419 	mr.mr_datalen = 0;
2420 	mr.mr_numsrc = 0;
2421 	mr.mr_addr = inm->in6m_addr;
2422 	in6_clearscope(&mr.mr_addr);
2423 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2424 		if (m != m0)
2425 			m_freem(m);
2426 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2427 		return (-ENOMEM);
2428 	}
2429 	nbytes += sizeof(struct mldv2_record);
2430 
2431 	/*
2432 	 * Append as many sources as will fit in the first packet.
2433 	 * If we are appending to a new packet, the chain allocation
2434 	 * may potentially use clusters; use m_getptr() in this case.
2435 	 * If we are appending to an existing packet, we need to obtain
2436 	 * a pointer to the group record after m_append(), in case a new
2437 	 * mbuf was allocated.
2438 	 * Only append sources which are in-mode at t1. If we are
2439 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2440 	 * include source entries.
2441 	 * Only report recorded sources in our filter set when responding
2442 	 * to a group-source query.
2443 	 */
2444 	if (record_has_sources) {
2445 		if (m == m0) {
2446 			md = m_last(m);
2447 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2448 			    md->m_len - nbytes);
2449 		} else {
2450 			md = m_getptr(m, 0, &off);
2451 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2452 			    off);
2453 		}
2454 		msrcs = 0;
2455 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2456 		    nims) {
2457 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2458 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2459 			now = im6s_get_mode(inm, ims, 1);
2460 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2461 			if ((now != mode) ||
2462 			    (now == mode && mode == MCAST_UNDEFINED)) {
2463 				CTR1(KTR_MLD, "%s: skip node", __func__);
2464 				continue;
2465 			}
2466 			if (is_source_query && ims->im6s_stp == 0) {
2467 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2468 				    __func__);
2469 				continue;
2470 			}
2471 			CTR1(KTR_MLD, "%s: append node", __func__);
2472 			if (!m_append(m, sizeof(struct in6_addr),
2473 			    (void *)&ims->im6s_addr)) {
2474 				if (m != m0)
2475 					m_freem(m);
2476 				CTR1(KTR_MLD, "%s: m_append() failed.",
2477 				    __func__);
2478 				return (-ENOMEM);
2479 			}
2480 			nbytes += sizeof(struct in6_addr);
2481 			++msrcs;
2482 			if (msrcs == m0srcs)
2483 				break;
2484 		}
2485 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2486 		    msrcs);
2487 		pmr->mr_numsrc = htons(msrcs);
2488 		nbytes += (msrcs * sizeof(struct in6_addr));
2489 	}
2490 
2491 	if (is_source_query && msrcs == 0) {
2492 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2493 		if (m != m0)
2494 			m_freem(m);
2495 		return (0);
2496 	}
2497 
2498 	/*
2499 	 * We are good to go with first packet.
2500 	 */
2501 	if (m != m0) {
2502 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2503 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2504 		_IF_ENQUEUE(ifq, m);
2505 	} else
2506 		m->m_pkthdr.PH_vt.vt_nrecs++;
2507 
2508 	/*
2509 	 * No further work needed if no source list in packet(s).
2510 	 */
2511 	if (!record_has_sources)
2512 		return (nbytes);
2513 
2514 	/*
2515 	 * Whilst sources remain to be announced, we need to allocate
2516 	 * a new packet and fill out as many sources as will fit.
2517 	 * Always try for a cluster first.
2518 	 */
2519 	while (nims != NULL) {
2520 		if (_IF_QFULL(ifq)) {
2521 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2522 			return (-ENOMEM);
2523 		}
2524 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2525 		if (m == NULL)
2526 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2527 		if (m == NULL)
2528 			return (-ENOMEM);
2529 		mld_save_context(m, ifp);
2530 		md = m_getptr(m, 0, &off);
2531 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2532 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2533 
2534 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2535 			if (m != m0)
2536 				m_freem(m);
2537 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2538 			return (-ENOMEM);
2539 		}
2540 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2541 		nbytes += sizeof(struct mldv2_record);
2542 
2543 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2544 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2545 
2546 		msrcs = 0;
2547 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2548 			CTR2(KTR_MLD, "%s: visit node %s",
2549 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2550 			now = im6s_get_mode(inm, ims, 1);
2551 			if ((now != mode) ||
2552 			    (now == mode && mode == MCAST_UNDEFINED)) {
2553 				CTR1(KTR_MLD, "%s: skip node", __func__);
2554 				continue;
2555 			}
2556 			if (is_source_query && ims->im6s_stp == 0) {
2557 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2558 				    __func__);
2559 				continue;
2560 			}
2561 			CTR1(KTR_MLD, "%s: append node", __func__);
2562 			if (!m_append(m, sizeof(struct in6_addr),
2563 			    (void *)&ims->im6s_addr)) {
2564 				if (m != m0)
2565 					m_freem(m);
2566 				CTR1(KTR_MLD, "%s: m_append() failed.",
2567 				    __func__);
2568 				return (-ENOMEM);
2569 			}
2570 			++msrcs;
2571 			if (msrcs == m0srcs)
2572 				break;
2573 		}
2574 		pmr->mr_numsrc = htons(msrcs);
2575 		nbytes += (msrcs * sizeof(struct in6_addr));
2576 
2577 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2578 		_IF_ENQUEUE(ifq, m);
2579 	}
2580 
2581 	return (nbytes);
2582 }
2583 
2584 /*
2585  * Type used to mark record pass completion.
2586  * We exploit the fact we can cast to this easily from the
2587  * current filter modes on each ip_msource node.
2588  */
2589 typedef enum {
2590 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2591 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2592 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2593 	REC_FULL = REC_ALLOW | REC_BLOCK
2594 } rectype_t;
2595 
2596 /*
2597  * Enqueue an MLDv2 filter list change to the given output queue.
2598  *
2599  * Source list filter state is held in an RB-tree. When the filter list
2600  * for a group is changed without changing its mode, we need to compute
2601  * the deltas between T0 and T1 for each source in the filter set,
2602  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2603  *
2604  * As we may potentially queue two record types, and the entire R-B tree
2605  * needs to be walked at once, we break this out into its own function
2606  * so we can generate a tightly packed queue of packets.
2607  *
2608  * XXX This could be written to only use one tree walk, although that makes
2609  * serializing into the mbuf chains a bit harder. For now we do two walks
2610  * which makes things easier on us, and it may or may not be harder on
2611  * the L2 cache.
2612  *
2613  * If successful the size of all data appended to the queue is returned,
2614  * otherwise an error code less than zero is returned, or zero if
2615  * no record(s) were appended.
2616  */
2617 static int
2618 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
2619 {
2620 	static const int MINRECLEN =
2621 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2622 	struct ifnet		*ifp;
2623 	struct mldv2_record	 mr;
2624 	struct mldv2_record	*pmr;
2625 	struct ip6_msource	*ims, *nims;
2626 	struct mbuf		*m, *m0, *md;
2627 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2628 	int			 nallow, nblock;
2629 	uint8_t			 mode, now, then;
2630 	rectype_t		 crt, drt, nrt;
2631 #ifdef KTR
2632 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2633 #endif
2634 
2635 	IN6_MULTI_LOCK_ASSERT();
2636 
2637 	if (inm->in6m_nsrc == 0 ||
2638 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2639 		return (0);
2640 
2641 	ifp = inm->in6m_ifp;			/* interface */
2642 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2643 	crt = REC_NONE;	/* current group record type */
2644 	drt = REC_NONE;	/* mask of completed group record types */
2645 	nrt = REC_NONE;	/* record type for current node */
2646 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2647 	npbytes = 0;	/* # of bytes appended this packet */
2648 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2649 	rsrcs = 0;	/* # sources encoded in current record */
2650 	schanged = 0;	/* # nodes encoded in overall filter change */
2651 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2652 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2653 	nims = NULL;	/* next tree node pointer */
2654 
2655 	/*
2656 	 * For each possible filter record mode.
2657 	 * The first kind of source we encounter tells us which
2658 	 * is the first kind of record we start appending.
2659 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2660 	 * as the inverse of the group's filter mode.
2661 	 */
2662 	while (drt != REC_FULL) {
2663 		do {
2664 			m0 = ifq->ifq_tail;
2665 			if (m0 != NULL &&
2666 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2667 			     MLD_V2_REPORT_MAXRECS) &&
2668 			    (m0->m_pkthdr.len + MINRECLEN) <
2669 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2670 				m = m0;
2671 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2672 					    sizeof(struct mldv2_record)) /
2673 					    sizeof(struct in6_addr);
2674 				CTR1(KTR_MLD,
2675 				    "%s: use previous packet", __func__);
2676 			} else {
2677 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2678 				if (m == NULL)
2679 					m = m_gethdr(M_DONTWAIT, MT_DATA);
2680 				if (m == NULL) {
2681 					CTR1(KTR_MLD,
2682 					    "%s: m_get*() failed", __func__);
2683 					return (-ENOMEM);
2684 				}
2685 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2686 				mld_save_context(m, ifp);
2687 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2688 				    sizeof(struct mldv2_record)) /
2689 				    sizeof(struct in6_addr);
2690 				npbytes = 0;
2691 				CTR1(KTR_MLD,
2692 				    "%s: allocated new packet", __func__);
2693 			}
2694 			/*
2695 			 * Append the MLD group record header to the
2696 			 * current packet's data area.
2697 			 * Recalculate pointer to free space for next
2698 			 * group record, in case m_append() allocated
2699 			 * a new mbuf or cluster.
2700 			 */
2701 			memset(&mr, 0, sizeof(mr));
2702 			mr.mr_addr = inm->in6m_addr;
2703 			in6_clearscope(&mr.mr_addr);
2704 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2705 				if (m != m0)
2706 					m_freem(m);
2707 				CTR1(KTR_MLD,
2708 				    "%s: m_append() failed", __func__);
2709 				return (-ENOMEM);
2710 			}
2711 			npbytes += sizeof(struct mldv2_record);
2712 			if (m != m0) {
2713 				/* new packet; offset in chain */
2714 				md = m_getptr(m, npbytes -
2715 				    sizeof(struct mldv2_record), &off);
2716 				pmr = (struct mldv2_record *)(mtod(md,
2717 				    uint8_t *) + off);
2718 			} else {
2719 				/* current packet; offset from last append */
2720 				md = m_last(m);
2721 				pmr = (struct mldv2_record *)(mtod(md,
2722 				    uint8_t *) + md->m_len -
2723 				    sizeof(struct mldv2_record));
2724 			}
2725 			/*
2726 			 * Begin walking the tree for this record type
2727 			 * pass, or continue from where we left off
2728 			 * previously if we had to allocate a new packet.
2729 			 * Only report deltas in-mode at t1.
2730 			 * We need not report included sources as allowed
2731 			 * if we are in inclusive mode on the group,
2732 			 * however the converse is not true.
2733 			 */
2734 			rsrcs = 0;
2735 			if (nims == NULL) {
2736 				nims = RB_MIN(ip6_msource_tree,
2737 				    &inm->in6m_srcs);
2738 			}
2739 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2740 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2741 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2742 				now = im6s_get_mode(inm, ims, 1);
2743 				then = im6s_get_mode(inm, ims, 0);
2744 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2745 				    __func__, then, now);
2746 				if (now == then) {
2747 					CTR1(KTR_MLD,
2748 					    "%s: skip unchanged", __func__);
2749 					continue;
2750 				}
2751 				if (mode == MCAST_EXCLUDE &&
2752 				    now == MCAST_INCLUDE) {
2753 					CTR1(KTR_MLD,
2754 					    "%s: skip IN src on EX group",
2755 					    __func__);
2756 					continue;
2757 				}
2758 				nrt = (rectype_t)now;
2759 				if (nrt == REC_NONE)
2760 					nrt = (rectype_t)(~mode & REC_FULL);
2761 				if (schanged++ == 0) {
2762 					crt = nrt;
2763 				} else if (crt != nrt)
2764 					continue;
2765 				if (!m_append(m, sizeof(struct in6_addr),
2766 				    (void *)&ims->im6s_addr)) {
2767 					if (m != m0)
2768 						m_freem(m);
2769 					CTR1(KTR_MLD,
2770 					    "%s: m_append() failed", __func__);
2771 					return (-ENOMEM);
2772 				}
2773 				nallow += !!(crt == REC_ALLOW);
2774 				nblock += !!(crt == REC_BLOCK);
2775 				if (++rsrcs == m0srcs)
2776 					break;
2777 			}
2778 			/*
2779 			 * If we did not append any tree nodes on this
2780 			 * pass, back out of allocations.
2781 			 */
2782 			if (rsrcs == 0) {
2783 				npbytes -= sizeof(struct mldv2_record);
2784 				if (m != m0) {
2785 					CTR1(KTR_MLD,
2786 					    "%s: m_free(m)", __func__);
2787 					m_freem(m);
2788 				} else {
2789 					CTR1(KTR_MLD,
2790 					    "%s: m_adj(m, -mr)", __func__);
2791 					m_adj(m, -((int)sizeof(
2792 					    struct mldv2_record)));
2793 				}
2794 				continue;
2795 			}
2796 			npbytes += (rsrcs * sizeof(struct in6_addr));
2797 			if (crt == REC_ALLOW)
2798 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2799 			else if (crt == REC_BLOCK)
2800 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2801 			pmr->mr_numsrc = htons(rsrcs);
2802 			/*
2803 			 * Count the new group record, and enqueue this
2804 			 * packet if it wasn't already queued.
2805 			 */
2806 			m->m_pkthdr.PH_vt.vt_nrecs++;
2807 			if (m != m0)
2808 				_IF_ENQUEUE(ifq, m);
2809 			nbytes += npbytes;
2810 		} while (nims != NULL);
2811 		drt |= crt;
2812 		crt = (~crt & REC_FULL);
2813 	}
2814 
2815 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2816 	    nallow, nblock);
2817 
2818 	return (nbytes);
2819 }
2820 
2821 static int
2822 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
2823 {
2824 	struct ifqueue	*gq;
2825 	struct mbuf	*m;		/* pending state-change */
2826 	struct mbuf	*m0;		/* copy of pending state-change */
2827 	struct mbuf	*mt;		/* last state-change in packet */
2828 	int		 docopy, domerge;
2829 	u_int		 recslen;
2830 
2831 	docopy = 0;
2832 	domerge = 0;
2833 	recslen = 0;
2834 
2835 	IN6_MULTI_LOCK_ASSERT();
2836 	MLD_LOCK_ASSERT();
2837 
2838 	/*
2839 	 * If there are further pending retransmissions, make a writable
2840 	 * copy of each queued state-change message before merging.
2841 	 */
2842 	if (inm->in6m_scrv > 0)
2843 		docopy = 1;
2844 
2845 	gq = &inm->in6m_scq;
2846 #ifdef KTR
2847 	if (gq->ifq_head == NULL) {
2848 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2849 		    __func__, inm);
2850 	}
2851 #endif
2852 
2853 	m = gq->ifq_head;
2854 	while (m != NULL) {
2855 		/*
2856 		 * Only merge the report into the current packet if
2857 		 * there is sufficient space to do so; an MLDv2 report
2858 		 * packet may only contain 65,535 group records.
2859 		 * Always use a simple mbuf chain concatentation to do this,
2860 		 * as large state changes for single groups may have
2861 		 * allocated clusters.
2862 		 */
2863 		domerge = 0;
2864 		mt = ifscq->ifq_tail;
2865 		if (mt != NULL) {
2866 			recslen = m_length(m, NULL);
2867 
2868 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2869 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2870 			    MLD_V2_REPORT_MAXRECS) &&
2871 			    (mt->m_pkthdr.len + recslen <=
2872 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2873 				domerge = 1;
2874 		}
2875 
2876 		if (!domerge && _IF_QFULL(gq)) {
2877 			CTR2(KTR_MLD,
2878 			    "%s: outbound queue full, skipping whole packet %p",
2879 			    __func__, m);
2880 			mt = m->m_nextpkt;
2881 			if (!docopy)
2882 				m_freem(m);
2883 			m = mt;
2884 			continue;
2885 		}
2886 
2887 		if (!docopy) {
2888 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2889 			_IF_DEQUEUE(gq, m0);
2890 			m = m0->m_nextpkt;
2891 		} else {
2892 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2893 			m0 = m_dup(m, M_NOWAIT);
2894 			if (m0 == NULL)
2895 				return (ENOMEM);
2896 			m0->m_nextpkt = NULL;
2897 			m = m->m_nextpkt;
2898 		}
2899 
2900 		if (!domerge) {
2901 			CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)",
2902 			    __func__, m0, ifscq);
2903 			_IF_ENQUEUE(ifscq, m0);
2904 		} else {
2905 			struct mbuf *mtl;	/* last mbuf of packet mt */
2906 
2907 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2908 			    __func__, m0, mt);
2909 
2910 			mtl = m_last(mt);
2911 			m0->m_flags &= ~M_PKTHDR;
2912 			mt->m_pkthdr.len += recslen;
2913 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2914 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2915 
2916 			mtl->m_next = m0;
2917 		}
2918 	}
2919 
2920 	return (0);
2921 }
2922 
2923 /*
2924  * Respond to a pending MLDv2 General Query.
2925  */
2926 static void
2927 mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
2928 {
2929 	struct ifmultiaddr	*ifma, *tifma;
2930 	struct ifnet		*ifp;
2931 	struct in6_multi	*inm;
2932 	int			 retval;
2933 
2934 	IN6_MULTI_LOCK_ASSERT();
2935 	MLD_LOCK_ASSERT();
2936 
2937 	KASSERT(mli->mli_version == MLD_VERSION_2,
2938 	    ("%s: called when version %d", __func__, mli->mli_version));
2939 
2940 	ifp = mli->mli_ifp;
2941 
2942 	IF_ADDR_LOCK(ifp);
2943 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
2944 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2945 		    ifma->ifma_protospec == NULL)
2946 			continue;
2947 
2948 		inm = (struct in6_multi *)ifma->ifma_protospec;
2949 		KASSERT(ifp == inm->in6m_ifp,
2950 		    ("%s: inconsistent ifp", __func__));
2951 
2952 		switch (inm->in6m_state) {
2953 		case MLD_NOT_MEMBER:
2954 		case MLD_SILENT_MEMBER:
2955 			break;
2956 		case MLD_REPORTING_MEMBER:
2957 		case MLD_IDLE_MEMBER:
2958 		case MLD_LAZY_MEMBER:
2959 		case MLD_SLEEPING_MEMBER:
2960 		case MLD_AWAKENING_MEMBER:
2961 			inm->in6m_state = MLD_REPORTING_MEMBER;
2962 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
2963 			    inm, 0, 0, 0);
2964 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2965 			    __func__, retval);
2966 			break;
2967 		case MLD_G_QUERY_PENDING_MEMBER:
2968 		case MLD_SG_QUERY_PENDING_MEMBER:
2969 		case MLD_LEAVING_MEMBER:
2970 			break;
2971 		}
2972 	}
2973 	IF_ADDR_UNLOCK(ifp);
2974 
2975 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
2976 
2977 	/*
2978 	 * Slew transmission of bursts over 500ms intervals.
2979 	 */
2980 	if (mli->mli_gq.ifq_head != NULL) {
2981 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
2982 		    MLD_RESPONSE_BURST_INTERVAL);
2983 		V_interface_timers_running6 = 1;
2984 	}
2985 }
2986 
2987 /*
2988  * Transmit the next pending message in the output queue.
2989  *
2990  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
2991  * MRT: Nothing needs to be done, as MLD traffic is always local to
2992  * a link and uses a link-scope multicast address.
2993  */
2994 static void
2995 mld_dispatch_packet(struct mbuf *m)
2996 {
2997 	struct ip6_moptions	 im6o;
2998 	struct ifnet		*ifp;
2999 	struct ifnet		*oifp;
3000 	struct mbuf		*m0;
3001 	struct mbuf		*md;
3002 	struct ip6_hdr		*ip6;
3003 	struct mld_hdr		*mld;
3004 	int			 error;
3005 	int			 off;
3006 	int			 type;
3007 	uint32_t		 ifindex;
3008 
3009 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3010 
3011 	/*
3012 	 * Set VNET image pointer from enqueued mbuf chain
3013 	 * before doing anything else. Whilst we use interface
3014 	 * indexes to guard against interface detach, they are
3015 	 * unique to each VIMAGE and must be retrieved.
3016 	 */
3017 	ifindex = mld_restore_context(m);
3018 
3019 	/*
3020 	 * Check if the ifnet still exists. This limits the scope of
3021 	 * any race in the absence of a global ifp lock for low cost
3022 	 * (an array lookup).
3023 	 */
3024 	ifp = ifnet_byindex(ifindex);
3025 	if (ifp == NULL) {
3026 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3027 		    __func__, m, ifindex);
3028 		m_freem(m);
3029 		IP6STAT_INC(ip6s_noroute);
3030 		goto out;
3031 	}
3032 
3033 	im6o.im6o_multicast_hlim  = 1;
3034 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3035 	im6o.im6o_multicast_ifp = ifp;
3036 
3037 	if (m->m_flags & M_MLDV1) {
3038 		m0 = m;
3039 	} else {
3040 		m0 = mld_v2_encap_report(ifp, m);
3041 		if (m0 == NULL) {
3042 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3043 			m_freem(m);
3044 			IP6STAT_INC(ip6s_odropped);
3045 			goto out;
3046 		}
3047 	}
3048 
3049 	mld_scrub_context(m0);
3050 	m->m_flags &= ~(M_PROTOFLAGS);
3051 	m0->m_pkthdr.rcvif = V_loif;
3052 
3053 	ip6 = mtod(m0, struct ip6_hdr *);
3054 #if 0
3055 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3056 #else
3057 	/*
3058 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3059 	 * occur if we called in6_setscope() at transmission.
3060 	 * See comments at top of file.
3061 	 */
3062 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3063 #endif
3064 
3065 	/*
3066 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3067 	 * so we can bump the stats.
3068 	 */
3069 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3070 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3071 	type = mld->mld_type;
3072 
3073 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3074 	    &oifp, NULL);
3075 	if (error) {
3076 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3077 		goto out;
3078 	}
3079 	ICMP6STAT_INC(icp6s_outhist[type]);
3080 	if (oifp != NULL) {
3081 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3082 		switch (type) {
3083 		case MLD_LISTENER_REPORT:
3084 		case MLDV2_LISTENER_REPORT:
3085 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3086 			break;
3087 		case MLD_LISTENER_DONE:
3088 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3089 			break;
3090 		}
3091 	}
3092 out:
3093 	return;
3094 }
3095 
3096 /*
3097  * Encapsulate an MLDv2 report.
3098  *
3099  * KAME IPv6 requires that hop-by-hop options be passed separately,
3100  * and that the IPv6 header be prepended in a separate mbuf.
3101  *
3102  * Returns a pointer to the new mbuf chain head, or NULL if the
3103  * allocation failed.
3104  */
3105 static struct mbuf *
3106 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3107 {
3108 	struct mbuf		*mh;
3109 	struct mldv2_report	*mld;
3110 	struct ip6_hdr		*ip6;
3111 	struct in6_ifaddr	*ia;
3112 	int			 mldreclen;
3113 
3114 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3115 	KASSERT((m->m_flags & M_PKTHDR),
3116 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3117 
3118 	/*
3119 	 * RFC3590: OK to send as :: or tentative during DAD.
3120 	 */
3121 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3122 	if (ia == NULL)
3123 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3124 
3125 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3126 	if (mh == NULL) {
3127 		if (ia != NULL)
3128 			ifa_free(&ia->ia_ifa);
3129 		m_freem(m);
3130 		return (NULL);
3131 	}
3132 	MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3133 
3134 	mldreclen = m_length(m, NULL);
3135 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3136 
3137 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3138 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3139 	    sizeof(struct mldv2_report) + mldreclen;
3140 
3141 	ip6 = mtod(mh, struct ip6_hdr *);
3142 	ip6->ip6_flow = 0;
3143 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3144 	ip6->ip6_vfc |= IPV6_VERSION;
3145 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3146 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3147 	if (ia != NULL)
3148 		ifa_free(&ia->ia_ifa);
3149 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3150 	/* scope ID will be set in netisr */
3151 
3152 	mld = (struct mldv2_report *)(ip6 + 1);
3153 	mld->mld_type = MLDV2_LISTENER_REPORT;
3154 	mld->mld_code = 0;
3155 	mld->mld_cksum = 0;
3156 	mld->mld_v2_reserved = 0;
3157 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3158 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3159 
3160 	mh->m_next = m;
3161 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3162 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3163 	return (mh);
3164 }
3165 
3166 #ifdef KTR
3167 static char *
3168 mld_rec_type_to_str(const int type)
3169 {
3170 
3171 	switch (type) {
3172 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3173 			return "TO_EX";
3174 			break;
3175 		case MLD_CHANGE_TO_INCLUDE_MODE:
3176 			return "TO_IN";
3177 			break;
3178 		case MLD_MODE_IS_EXCLUDE:
3179 			return "MODE_EX";
3180 			break;
3181 		case MLD_MODE_IS_INCLUDE:
3182 			return "MODE_IN";
3183 			break;
3184 		case MLD_ALLOW_NEW_SOURCES:
3185 			return "ALLOW_NEW";
3186 			break;
3187 		case MLD_BLOCK_OLD_SOURCES:
3188 			return "BLOCK_OLD";
3189 			break;
3190 		default:
3191 			break;
3192 	}
3193 	return "unknown";
3194 }
3195 #endif
3196 
3197 static void
3198 mld_init(void *unused __unused)
3199 {
3200 
3201 	CTR1(KTR_MLD, "%s: initializing", __func__);
3202 	MLD_LOCK_INIT();
3203 
3204 	ip6_initpktopts(&mld_po);
3205 	mld_po.ip6po_hlim = 1;
3206 	mld_po.ip6po_hbh = &mld_ra.hbh;
3207 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3208 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3209 }
3210 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
3211 
3212 static void
3213 mld_uninit(void *unused __unused)
3214 {
3215 
3216 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3217 	MLD_LOCK_DESTROY();
3218 }
3219 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
3220 
3221 static void
3222 vnet_mld_init(const void *unused __unused)
3223 {
3224 
3225 	CTR1(KTR_MLD, "%s: initializing", __func__);
3226 
3227 	LIST_INIT(&V_mli_head);
3228 }
3229 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
3230     NULL);
3231 
3232 static void
3233 vnet_mld_uninit(const void *unused __unused)
3234 {
3235 
3236 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3237 
3238 	KASSERT(LIST_EMPTY(&V_mli_head),
3239 	    ("%s: mli list not empty; ifnets not detached?", __func__));
3240 }
3241 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
3242     NULL);
3243 
3244 static int
3245 mld_modevent(module_t mod, int type, void *unused __unused)
3246 {
3247 
3248     switch (type) {
3249     case MOD_LOAD:
3250     case MOD_UNLOAD:
3251 	break;
3252     default:
3253 	return (EOPNOTSUPP);
3254     }
3255     return (0);
3256 }
3257 
3258 static moduledata_t mld_mod = {
3259     "mld",
3260     mld_modevent,
3261     0
3262 };
3263 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3264