xref: /freebsd/sys/netinet6/mld6.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
1 /*-
2  * Copyright (c) 2009 Bruce Simpson.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. The name of the author may not be used to endorse or promote
13  *    products derived from this software without specific prior written
14  *    permission.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
29  */
30 
31 /*-
32  * Copyright (c) 1988 Stephen Deering.
33  * Copyright (c) 1992, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * Stephen Deering of Stanford University.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 4. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
64  */
65 
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
68 
69 #include "opt_inet.h"
70 #include "opt_inet6.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/mbuf.h>
75 #include <sys/socket.h>
76 #include <sys/protosw.h>
77 #include <sys/sysctl.h>
78 #include <sys/kernel.h>
79 #include <sys/callout.h>
80 #include <sys/malloc.h>
81 #include <sys/module.h>
82 #include <sys/ktr.h>
83 
84 #include <net/if.h>
85 #include <net/if_var.h>
86 #include <net/route.h>
87 #include <net/vnet.h>
88 
89 #include <netinet/in.h>
90 #include <netinet/in_var.h>
91 #include <netinet6/in6_var.h>
92 #include <netinet/ip6.h>
93 #include <netinet6/ip6_var.h>
94 #include <netinet6/scope6_var.h>
95 #include <netinet/icmp6.h>
96 #include <netinet6/mld6.h>
97 #include <netinet6/mld6_var.h>
98 
99 #include <security/mac/mac_framework.h>
100 
101 #ifndef KTR_MLD
102 #define KTR_MLD KTR_INET6
103 #endif
104 
105 static struct mld_ifinfo *
106 		mli_alloc_locked(struct ifnet *);
107 static void	mli_delete_locked(const struct ifnet *);
108 static void	mld_dispatch_packet(struct mbuf *);
109 static void	mld_dispatch_queue(struct ifqueue *, int);
110 static void	mld_final_leave(struct in6_multi *, struct mld_ifinfo *);
111 static void	mld_fasttimo_vnet(void);
112 static int	mld_handle_state_change(struct in6_multi *,
113 		    struct mld_ifinfo *);
114 static int	mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
115 		    const int);
116 #ifdef KTR
117 static char *	mld_rec_type_to_str(const int);
118 #endif
119 static void	mld_set_version(struct mld_ifinfo *, const int);
120 static void	mld_slowtimo_vnet(void);
121 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
122 		    /*const*/ struct mld_hdr *);
123 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static void	mld_v1_process_group_timer(struct mld_ifinfo *,
126 		    struct in6_multi *);
127 static void	mld_v1_process_querier_timers(struct mld_ifinfo *);
128 static int	mld_v1_transmit_report(struct in6_multi *, const int);
129 static void	mld_v1_update_group(struct in6_multi *, const int);
130 static void	mld_v2_cancel_link_timers(struct mld_ifinfo *);
131 static void	mld_v2_dispatch_general_query(struct mld_ifinfo *);
132 static struct mbuf *
133 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
134 static int	mld_v2_enqueue_filter_change(struct ifqueue *,
135 		    struct in6_multi *);
136 static int	mld_v2_enqueue_group_record(struct ifqueue *,
137 		    struct in6_multi *, const int, const int, const int,
138 		    const int);
139 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
140 		    struct mbuf *, const int, const int);
141 static int	mld_v2_merge_state_changes(struct in6_multi *,
142 		    struct ifqueue *);
143 static void	mld_v2_process_group_timers(struct mld_ifinfo *,
144 		    struct ifqueue *, struct ifqueue *,
145 		    struct in6_multi *, const int);
146 static int	mld_v2_process_group_query(struct in6_multi *,
147 		    struct mld_ifinfo *mli, int, struct mbuf *, const int);
148 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
149 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
150 
151 /*
152  * Normative references: RFC 2710, RFC 3590, RFC 3810.
153  *
154  * Locking:
155  *  * The MLD subsystem lock ends up being system-wide for the moment,
156  *    but could be per-VIMAGE later on.
157  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
158  *    Any may be taken independently; if any are held at the same
159  *    time, the above lock order must be followed.
160  *  * IN6_MULTI_LOCK covers in_multi.
161  *  * MLD_LOCK covers per-link state and any global variables in this file.
162  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
163  *    per-link state iterators.
164  *
165  *  XXX LOR PREVENTION
166  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
167  *  will not accept an ifp; it wants an embedded scope ID, unlike
168  *  ip_output(), which happily takes the ifp given to it. The embedded
169  *  scope ID is only used by MLD to select the outgoing interface.
170  *
171  *  During interface attach and detach, MLD will take MLD_LOCK *after*
172  *  the IF_AFDATA_LOCK.
173  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
174  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
175  *  dispatch could work around this, but we'd rather not do that, as it
176  *  can introduce other races.
177  *
178  *  As such, we exploit the fact that the scope ID is just the interface
179  *  index, and embed it in the IPv6 destination address accordingly.
180  *  This is potentially NOT VALID for MLDv1 reports, as they
181  *  are always sent to the multicast group itself; as MLDv2
182  *  reports are always sent to ff02::16, this is not an issue
183  *  when MLDv2 is in use.
184  *
185  *  This does not however eliminate the LOR when ip6_output() itself
186  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
187  *  trigger a LOR warning in WITNESS when the ifnet is detached.
188  *
189  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
190  *  how it's used across the network stack. Here we're simply exploiting
191  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
192  *
193  * VIMAGE:
194  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
195  *    to a vnet in ifp->if_vnet.
196  */
197 static struct mtx		 mld_mtx;
198 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
199 
200 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
201 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
202 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
203 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
204 
205 /*
206  * VIMAGE-wide globals.
207  */
208 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0};
209 static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head);
210 static VNET_DEFINE(int, interface_timers_running6);
211 static VNET_DEFINE(int, state_change_timers_running6);
212 static VNET_DEFINE(int, current_state_timers_running6);
213 
214 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
215 #define	V_mli_head			VNET(mli_head)
216 #define	V_interface_timers_running6	VNET(interface_timers_running6)
217 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
218 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
219 
220 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
221 
222 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
223     "IPv6 Multicast Listener Discovery");
224 
225 /*
226  * Virtualized sysctls.
227  */
228 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
229     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
230     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
231     "Rate limit for MLDv2 Group-and-Source queries in seconds");
232 
233 /*
234  * Non-virtualized sysctls.
235  */
236 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
237     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
238     "Per-interface MLDv2 state");
239 
240 static int	mld_v1enable = 1;
241 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
242     &mld_v1enable, 0, "Enable fallback to MLDv1");
243 
244 static int	mld_use_allow = 1;
245 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
246     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
247 
248 /*
249  * Packed Router Alert option structure declaration.
250  */
251 struct mld_raopt {
252 	struct ip6_hbh		hbh;
253 	struct ip6_opt		pad;
254 	struct ip6_opt_router	ra;
255 } __packed;
256 
257 /*
258  * Router Alert hop-by-hop option header.
259  */
260 static struct mld_raopt mld_ra = {
261 	.hbh = { 0, 0 },
262 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
263 	.ra = {
264 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
265 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
266 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
267 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
268 	}
269 };
270 static struct ip6_pktopts mld_po;
271 
272 static __inline void
273 mld_save_context(struct mbuf *m, struct ifnet *ifp)
274 {
275 
276 #ifdef VIMAGE
277 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
278 #endif /* VIMAGE */
279 	m->m_pkthdr.flowid = ifp->if_index;
280 }
281 
282 static __inline void
283 mld_scrub_context(struct mbuf *m)
284 {
285 
286 	m->m_pkthdr.PH_loc.ptr = NULL;
287 	m->m_pkthdr.flowid = 0;
288 }
289 
290 /*
291  * Restore context from a queued output chain.
292  * Return saved ifindex.
293  *
294  * VIMAGE: The assertion is there to make sure that we
295  * actually called CURVNET_SET() with what's in the mbuf chain.
296  */
297 static __inline uint32_t
298 mld_restore_context(struct mbuf *m)
299 {
300 
301 #if defined(VIMAGE) && defined(INVARIANTS)
302 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
303 	    ("%s: called when curvnet was not restored", __func__));
304 #endif
305 	return (m->m_pkthdr.flowid);
306 }
307 
308 /*
309  * Retrieve or set threshold between group-source queries in seconds.
310  *
311  * VIMAGE: Assume curvnet set by caller.
312  * SMPng: NOTE: Serialized by MLD lock.
313  */
314 static int
315 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
316 {
317 	int error;
318 	int i;
319 
320 	error = sysctl_wire_old_buffer(req, sizeof(int));
321 	if (error)
322 		return (error);
323 
324 	MLD_LOCK();
325 
326 	i = V_mld_gsrdelay.tv_sec;
327 
328 	error = sysctl_handle_int(oidp, &i, 0, req);
329 	if (error || !req->newptr)
330 		goto out_locked;
331 
332 	if (i < -1 || i >= 60) {
333 		error = EINVAL;
334 		goto out_locked;
335 	}
336 
337 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
338 	     V_mld_gsrdelay.tv_sec, i);
339 	V_mld_gsrdelay.tv_sec = i;
340 
341 out_locked:
342 	MLD_UNLOCK();
343 	return (error);
344 }
345 
346 /*
347  * Expose struct mld_ifinfo to userland, keyed by ifindex.
348  * For use by ifmcstat(8).
349  *
350  * SMPng: NOTE: Does an unlocked ifindex space read.
351  * VIMAGE: Assume curvnet set by caller. The node handler itself
352  * is not directly virtualized.
353  */
354 static int
355 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
356 {
357 	int			*name;
358 	int			 error;
359 	u_int			 namelen;
360 	struct ifnet		*ifp;
361 	struct mld_ifinfo	*mli;
362 
363 	name = (int *)arg1;
364 	namelen = arg2;
365 
366 	if (req->newptr != NULL)
367 		return (EPERM);
368 
369 	if (namelen != 1)
370 		return (EINVAL);
371 
372 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
373 	if (error)
374 		return (error);
375 
376 	IN6_MULTI_LOCK();
377 	MLD_LOCK();
378 
379 	if (name[0] <= 0 || name[0] > V_if_index) {
380 		error = ENOENT;
381 		goto out_locked;
382 	}
383 
384 	error = ENOENT;
385 
386 	ifp = ifnet_byindex(name[0]);
387 	if (ifp == NULL)
388 		goto out_locked;
389 
390 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
391 		if (ifp == mli->mli_ifp) {
392 			error = SYSCTL_OUT(req, mli,
393 			    sizeof(struct mld_ifinfo));
394 			break;
395 		}
396 	}
397 
398 out_locked:
399 	MLD_UNLOCK();
400 	IN6_MULTI_UNLOCK();
401 	return (error);
402 }
403 
404 /*
405  * Dispatch an entire queue of pending packet chains.
406  * VIMAGE: Assumes the vnet pointer has been set.
407  */
408 static void
409 mld_dispatch_queue(struct ifqueue *ifq, int limit)
410 {
411 	struct mbuf *m;
412 
413 	for (;;) {
414 		_IF_DEQUEUE(ifq, m);
415 		if (m == NULL)
416 			break;
417 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m);
418 		mld_dispatch_packet(m);
419 		if (--limit == 0)
420 			break;
421 	}
422 }
423 
424 /*
425  * Filter outgoing MLD report state by group.
426  *
427  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
428  * and node-local addresses. However, kernel and socket consumers
429  * always embed the KAME scope ID in the address provided, so strip it
430  * when performing comparison.
431  * Note: This is not the same as the *multicast* scope.
432  *
433  * Return zero if the given group is one for which MLD reports
434  * should be suppressed, or non-zero if reports should be issued.
435  */
436 static __inline int
437 mld_is_addr_reported(const struct in6_addr *addr)
438 {
439 
440 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
441 
442 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
443 		return (0);
444 
445 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
446 		struct in6_addr tmp = *addr;
447 		in6_clearscope(&tmp);
448 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
449 			return (0);
450 	}
451 
452 	return (1);
453 }
454 
455 /*
456  * Attach MLD when PF_INET6 is attached to an interface.
457  *
458  * SMPng: Normally called with IF_AFDATA_LOCK held.
459  */
460 struct mld_ifinfo *
461 mld_domifattach(struct ifnet *ifp)
462 {
463 	struct mld_ifinfo *mli;
464 
465 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
466 	    __func__, ifp, if_name(ifp));
467 
468 	MLD_LOCK();
469 
470 	mli = mli_alloc_locked(ifp);
471 	if (!(ifp->if_flags & IFF_MULTICAST))
472 		mli->mli_flags |= MLIF_SILENT;
473 	if (mld_use_allow)
474 		mli->mli_flags |= MLIF_USEALLOW;
475 
476 	MLD_UNLOCK();
477 
478 	return (mli);
479 }
480 
481 /*
482  * VIMAGE: assume curvnet set by caller.
483  */
484 static struct mld_ifinfo *
485 mli_alloc_locked(/*const*/ struct ifnet *ifp)
486 {
487 	struct mld_ifinfo *mli;
488 
489 	MLD_LOCK_ASSERT();
490 
491 	mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO);
492 	if (mli == NULL)
493 		goto out;
494 
495 	mli->mli_ifp = ifp;
496 	mli->mli_version = MLD_VERSION_2;
497 	mli->mli_flags = 0;
498 	mli->mli_rv = MLD_RV_INIT;
499 	mli->mli_qi = MLD_QI_INIT;
500 	mli->mli_qri = MLD_QRI_INIT;
501 	mli->mli_uri = MLD_URI_INIT;
502 
503 	SLIST_INIT(&mli->mli_relinmhead);
504 
505 	/*
506 	 * Responses to general queries are subject to bounds.
507 	 */
508 	IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
509 
510 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
511 
512 	CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)",
513 	     ifp, if_name(ifp));
514 
515 out:
516 	return (mli);
517 }
518 
519 /*
520  * Hook for ifdetach.
521  *
522  * NOTE: Some finalization tasks need to run before the protocol domain
523  * is detached, but also before the link layer does its cleanup.
524  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
525  *
526  * SMPng: Caller must hold IN6_MULTI_LOCK().
527  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
528  * XXX This routine is also bitten by unlocked ifma_protospec access.
529  */
530 void
531 mld_ifdetach(struct ifnet *ifp)
532 {
533 	struct mld_ifinfo	*mli;
534 	struct ifmultiaddr	*ifma;
535 	struct in6_multi	*inm, *tinm;
536 
537 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
538 	    if_name(ifp));
539 
540 	IN6_MULTI_LOCK_ASSERT();
541 	MLD_LOCK();
542 
543 	mli = MLD_IFINFO(ifp);
544 	if (mli->mli_version == MLD_VERSION_2) {
545 		IF_ADDR_RLOCK(ifp);
546 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
547 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
548 			    ifma->ifma_protospec == NULL)
549 				continue;
550 			inm = (struct in6_multi *)ifma->ifma_protospec;
551 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
552 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
553 				    inm, in6m_nrele);
554 			}
555 			in6m_clear_recorded(inm);
556 		}
557 		IF_ADDR_RUNLOCK(ifp);
558 		SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
559 		    tinm) {
560 			SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
561 			in6m_release_locked(inm);
562 		}
563 	}
564 
565 	MLD_UNLOCK();
566 }
567 
568 /*
569  * Hook for domifdetach.
570  * Runs after link-layer cleanup; free MLD state.
571  *
572  * SMPng: Normally called with IF_AFDATA_LOCK held.
573  */
574 void
575 mld_domifdetach(struct ifnet *ifp)
576 {
577 
578 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
579 	    __func__, ifp, if_name(ifp));
580 
581 	MLD_LOCK();
582 	mli_delete_locked(ifp);
583 	MLD_UNLOCK();
584 }
585 
586 static void
587 mli_delete_locked(const struct ifnet *ifp)
588 {
589 	struct mld_ifinfo *mli, *tmli;
590 
591 	CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)",
592 	    __func__, ifp, if_name(ifp));
593 
594 	MLD_LOCK_ASSERT();
595 
596 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
597 		if (mli->mli_ifp == ifp) {
598 			/*
599 			 * Free deferred General Query responses.
600 			 */
601 			_IF_DRAIN(&mli->mli_gq);
602 
603 			LIST_REMOVE(mli, mli_link);
604 
605 			KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
606 			    ("%s: there are dangling in_multi references",
607 			    __func__));
608 
609 			free(mli, M_MLD);
610 			return;
611 		}
612 	}
613 #ifdef INVARIANTS
614 	panic("%s: mld_ifinfo not found for ifp %p\n", __func__,  ifp);
615 #endif
616 }
617 
618 /*
619  * Process a received MLDv1 general or address-specific query.
620  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
621  *
622  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
623  * mld_addr. This is OK as we own the mbuf chain.
624  */
625 static int
626 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
627     /*const*/ struct mld_hdr *mld)
628 {
629 	struct ifmultiaddr	*ifma;
630 	struct mld_ifinfo	*mli;
631 	struct in6_multi	*inm;
632 	int			 is_general_query;
633 	uint16_t		 timer;
634 #ifdef KTR
635 	char			 ip6tbuf[INET6_ADDRSTRLEN];
636 #endif
637 
638 	is_general_query = 0;
639 
640 	if (!mld_v1enable) {
641 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
642 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
643 		    ifp, if_name(ifp));
644 		return (0);
645 	}
646 
647 	/*
648 	 * RFC3810 Section 6.2: MLD queries must originate from
649 	 * a router's link-local address.
650 	 */
651 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
652 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
653 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
654 		    ifp, if_name(ifp));
655 		return (0);
656 	}
657 
658 	/*
659 	 * Do address field validation upfront before we accept
660 	 * the query.
661 	 */
662 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
663 		/*
664 		 * MLDv1 General Query.
665 		 * If this was not sent to the all-nodes group, ignore it.
666 		 */
667 		struct in6_addr		 dst;
668 
669 		dst = ip6->ip6_dst;
670 		in6_clearscope(&dst);
671 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
672 			return (EINVAL);
673 		is_general_query = 1;
674 	} else {
675 		/*
676 		 * Embed scope ID of receiving interface in MLD query for
677 		 * lookup whilst we don't hold other locks.
678 		 */
679 		in6_setscope(&mld->mld_addr, ifp, NULL);
680 	}
681 
682 	IN6_MULTI_LOCK();
683 	MLD_LOCK();
684 
685 	/*
686 	 * Switch to MLDv1 host compatibility mode.
687 	 */
688 	mli = MLD_IFINFO(ifp);
689 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
690 	mld_set_version(mli, MLD_VERSION_1);
691 
692 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
693 	if (timer == 0)
694 		timer = 1;
695 
696 	IF_ADDR_RLOCK(ifp);
697 	if (is_general_query) {
698 		/*
699 		 * For each reporting group joined on this
700 		 * interface, kick the report timer.
701 		 */
702 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
703 		    ifp, if_name(ifp));
704 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
705 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
706 			    ifma->ifma_protospec == NULL)
707 				continue;
708 			inm = (struct in6_multi *)ifma->ifma_protospec;
709 			mld_v1_update_group(inm, timer);
710 		}
711 	} else {
712 		/*
713 		 * MLDv1 Group-Specific Query.
714 		 * If this is a group-specific MLDv1 query, we need only
715 		 * look up the single group to process it.
716 		 */
717 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
718 		if (inm != NULL) {
719 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
720 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
721 			    ifp, if_name(ifp));
722 			mld_v1_update_group(inm, timer);
723 		}
724 		/* XXX Clear embedded scope ID as userland won't expect it. */
725 		in6_clearscope(&mld->mld_addr);
726 	}
727 
728 	IF_ADDR_RUNLOCK(ifp);
729 	MLD_UNLOCK();
730 	IN6_MULTI_UNLOCK();
731 
732 	return (0);
733 }
734 
735 /*
736  * Update the report timer on a group in response to an MLDv1 query.
737  *
738  * If we are becoming the reporting member for this group, start the timer.
739  * If we already are the reporting member for this group, and timer is
740  * below the threshold, reset it.
741  *
742  * We may be updating the group for the first time since we switched
743  * to MLDv2. If we are, then we must clear any recorded source lists,
744  * and transition to REPORTING state; the group timer is overloaded
745  * for group and group-source query responses.
746  *
747  * Unlike MLDv2, the delay per group should be jittered
748  * to avoid bursts of MLDv1 reports.
749  */
750 static void
751 mld_v1_update_group(struct in6_multi *inm, const int timer)
752 {
753 #ifdef KTR
754 	char			 ip6tbuf[INET6_ADDRSTRLEN];
755 #endif
756 
757 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
758 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
759 	    if_name(inm->in6m_ifp), timer);
760 
761 	IN6_MULTI_LOCK_ASSERT();
762 
763 	switch (inm->in6m_state) {
764 	case MLD_NOT_MEMBER:
765 	case MLD_SILENT_MEMBER:
766 		break;
767 	case MLD_REPORTING_MEMBER:
768 		if (inm->in6m_timer != 0 &&
769 		    inm->in6m_timer <= timer) {
770 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
771 			    "skipping.", __func__);
772 			break;
773 		}
774 		/* FALLTHROUGH */
775 	case MLD_SG_QUERY_PENDING_MEMBER:
776 	case MLD_G_QUERY_PENDING_MEMBER:
777 	case MLD_IDLE_MEMBER:
778 	case MLD_LAZY_MEMBER:
779 	case MLD_AWAKENING_MEMBER:
780 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
781 		inm->in6m_state = MLD_REPORTING_MEMBER;
782 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
783 		V_current_state_timers_running6 = 1;
784 		break;
785 	case MLD_SLEEPING_MEMBER:
786 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
787 		inm->in6m_state = MLD_AWAKENING_MEMBER;
788 		break;
789 	case MLD_LEAVING_MEMBER:
790 		break;
791 	}
792 }
793 
794 /*
795  * Process a received MLDv2 general, group-specific or
796  * group-and-source-specific query.
797  *
798  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
799  *
800  * Return 0 if successful, otherwise an appropriate error code is returned.
801  */
802 static int
803 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
804     struct mbuf *m, const int off, const int icmp6len)
805 {
806 	struct mld_ifinfo	*mli;
807 	struct mldv2_query	*mld;
808 	struct in6_multi	*inm;
809 	uint32_t		 maxdelay, nsrc, qqi;
810 	int			 is_general_query;
811 	uint16_t		 timer;
812 	uint8_t			 qrv;
813 #ifdef KTR
814 	char			 ip6tbuf[INET6_ADDRSTRLEN];
815 #endif
816 
817 	is_general_query = 0;
818 
819 	/*
820 	 * RFC3810 Section 6.2: MLD queries must originate from
821 	 * a router's link-local address.
822 	 */
823 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
824 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
825 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
826 		    ifp, if_name(ifp));
827 		return (0);
828 	}
829 
830 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
831 
832 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
833 
834 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
835 	if (maxdelay >= 32768) {
836 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
837 			   (MLD_MRC_EXP(maxdelay) + 3);
838 	}
839 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
840 	if (timer == 0)
841 		timer = 1;
842 
843 	qrv = MLD_QRV(mld->mld_misc);
844 	if (qrv < 2) {
845 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
846 		    qrv, MLD_RV_INIT);
847 		qrv = MLD_RV_INIT;
848 	}
849 
850 	qqi = mld->mld_qqi;
851 	if (qqi >= 128) {
852 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
853 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
854 	}
855 
856 	nsrc = ntohs(mld->mld_numsrc);
857 	if (nsrc > MLD_MAX_GS_SOURCES)
858 		return (EMSGSIZE);
859 	if (icmp6len < sizeof(struct mldv2_query) +
860 	    (nsrc * sizeof(struct in6_addr)))
861 		return (EMSGSIZE);
862 
863 	/*
864 	 * Do further input validation upfront to avoid resetting timers
865 	 * should we need to discard this query.
866 	 */
867 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
868 		/*
869 		 * A general query with a source list has undefined
870 		 * behaviour; discard it.
871 		 */
872 		if (nsrc > 0)
873 			return (EINVAL);
874 		is_general_query = 1;
875 	} else {
876 		/*
877 		 * Embed scope ID of receiving interface in MLD query for
878 		 * lookup whilst we don't hold other locks (due to KAME
879 		 * locking lameness). We own this mbuf chain just now.
880 		 */
881 		in6_setscope(&mld->mld_addr, ifp, NULL);
882 	}
883 
884 	IN6_MULTI_LOCK();
885 	MLD_LOCK();
886 
887 	mli = MLD_IFINFO(ifp);
888 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
889 
890 	/*
891 	 * Discard the v2 query if we're in Compatibility Mode.
892 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
893 	 * until the Old Version Querier Present timer expires.
894 	 */
895 	if (mli->mli_version != MLD_VERSION_2)
896 		goto out_locked;
897 
898 	mld_set_version(mli, MLD_VERSION_2);
899 	mli->mli_rv = qrv;
900 	mli->mli_qi = qqi;
901 	mli->mli_qri = maxdelay;
902 
903 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
904 	    maxdelay);
905 
906 	if (is_general_query) {
907 		/*
908 		 * MLDv2 General Query.
909 		 *
910 		 * Schedule a current-state report on this ifp for
911 		 * all groups, possibly containing source lists.
912 		 *
913 		 * If there is a pending General Query response
914 		 * scheduled earlier than the selected delay, do
915 		 * not schedule any other reports.
916 		 * Otherwise, reset the interface timer.
917 		 */
918 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
919 		    ifp, if_name(ifp));
920 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
921 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
922 			V_interface_timers_running6 = 1;
923 		}
924 	} else {
925 		/*
926 		 * MLDv2 Group-specific or Group-and-source-specific Query.
927 		 *
928 		 * Group-source-specific queries are throttled on
929 		 * a per-group basis to defeat denial-of-service attempts.
930 		 * Queries for groups we are not a member of on this
931 		 * link are simply ignored.
932 		 */
933 		IF_ADDR_RLOCK(ifp);
934 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
935 		if (inm == NULL) {
936 			IF_ADDR_RUNLOCK(ifp);
937 			goto out_locked;
938 		}
939 		if (nsrc > 0) {
940 			if (!ratecheck(&inm->in6m_lastgsrtv,
941 			    &V_mld_gsrdelay)) {
942 				CTR1(KTR_MLD, "%s: GS query throttled.",
943 				    __func__);
944 				IF_ADDR_RUNLOCK(ifp);
945 				goto out_locked;
946 			}
947 		}
948 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
949 		     ifp, if_name(ifp));
950 		/*
951 		 * If there is a pending General Query response
952 		 * scheduled sooner than the selected delay, no
953 		 * further report need be scheduled.
954 		 * Otherwise, prepare to respond to the
955 		 * group-specific or group-and-source query.
956 		 */
957 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
958 			mld_v2_process_group_query(inm, mli, timer, m, off);
959 
960 		/* XXX Clear embedded scope ID as userland won't expect it. */
961 		in6_clearscope(&mld->mld_addr);
962 		IF_ADDR_RUNLOCK(ifp);
963 	}
964 
965 out_locked:
966 	MLD_UNLOCK();
967 	IN6_MULTI_UNLOCK();
968 
969 	return (0);
970 }
971 
972 /*
973  * Process a recieved MLDv2 group-specific or group-and-source-specific
974  * query.
975  * Return <0 if any error occured. Currently this is ignored.
976  */
977 static int
978 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli,
979     int timer, struct mbuf *m0, const int off)
980 {
981 	struct mldv2_query	*mld;
982 	int			 retval;
983 	uint16_t		 nsrc;
984 
985 	IN6_MULTI_LOCK_ASSERT();
986 	MLD_LOCK_ASSERT();
987 
988 	retval = 0;
989 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
990 
991 	switch (inm->in6m_state) {
992 	case MLD_NOT_MEMBER:
993 	case MLD_SILENT_MEMBER:
994 	case MLD_SLEEPING_MEMBER:
995 	case MLD_LAZY_MEMBER:
996 	case MLD_AWAKENING_MEMBER:
997 	case MLD_IDLE_MEMBER:
998 	case MLD_LEAVING_MEMBER:
999 		return (retval);
1000 		break;
1001 	case MLD_REPORTING_MEMBER:
1002 	case MLD_G_QUERY_PENDING_MEMBER:
1003 	case MLD_SG_QUERY_PENDING_MEMBER:
1004 		break;
1005 	}
1006 
1007 	nsrc = ntohs(mld->mld_numsrc);
1008 
1009 	/*
1010 	 * Deal with group-specific queries upfront.
1011 	 * If any group query is already pending, purge any recorded
1012 	 * source-list state if it exists, and schedule a query response
1013 	 * for this group-specific query.
1014 	 */
1015 	if (nsrc == 0) {
1016 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1017 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1018 			in6m_clear_recorded(inm);
1019 			timer = min(inm->in6m_timer, timer);
1020 		}
1021 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1022 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1023 		V_current_state_timers_running6 = 1;
1024 		return (retval);
1025 	}
1026 
1027 	/*
1028 	 * Deal with the case where a group-and-source-specific query has
1029 	 * been received but a group-specific query is already pending.
1030 	 */
1031 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1032 		timer = min(inm->in6m_timer, timer);
1033 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1034 		V_current_state_timers_running6 = 1;
1035 		return (retval);
1036 	}
1037 
1038 	/*
1039 	 * Finally, deal with the case where a group-and-source-specific
1040 	 * query has been received, where a response to a previous g-s-r
1041 	 * query exists, or none exists.
1042 	 * In this case, we need to parse the source-list which the Querier
1043 	 * has provided us with and check if we have any source list filter
1044 	 * entries at T1 for these sources. If we do not, there is no need
1045 	 * schedule a report and the query may be dropped.
1046 	 * If we do, we must record them and schedule a current-state
1047 	 * report for those sources.
1048 	 */
1049 	if (inm->in6m_nsrc > 0) {
1050 		struct mbuf		*m;
1051 		uint8_t			*sp;
1052 		int			 i, nrecorded;
1053 		int			 soff;
1054 
1055 		m = m0;
1056 		soff = off + sizeof(struct mldv2_query);
1057 		nrecorded = 0;
1058 		for (i = 0; i < nsrc; i++) {
1059 			sp = mtod(m, uint8_t *) + soff;
1060 			retval = in6m_record_source(inm,
1061 			    (const struct in6_addr *)sp);
1062 			if (retval < 0)
1063 				break;
1064 			nrecorded += retval;
1065 			soff += sizeof(struct in6_addr);
1066 			if (soff >= m->m_len) {
1067 				soff = soff - m->m_len;
1068 				m = m->m_next;
1069 				if (m == NULL)
1070 					break;
1071 			}
1072 		}
1073 		if (nrecorded > 0) {
1074 			CTR1(KTR_MLD,
1075 			    "%s: schedule response to SG query", __func__);
1076 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1077 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1078 			V_current_state_timers_running6 = 1;
1079 		}
1080 	}
1081 
1082 	return (retval);
1083 }
1084 
1085 /*
1086  * Process a received MLDv1 host membership report.
1087  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1088  *
1089  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1090  * mld_addr. This is OK as we own the mbuf chain.
1091  */
1092 static int
1093 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1094     /*const*/ struct mld_hdr *mld)
1095 {
1096 	struct in6_addr		 src, dst;
1097 	struct in6_ifaddr	*ia;
1098 	struct in6_multi	*inm;
1099 #ifdef KTR
1100 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1101 #endif
1102 
1103 	if (!mld_v1enable) {
1104 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1105 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1106 		    ifp, if_name(ifp));
1107 		return (0);
1108 	}
1109 
1110 	if (ifp->if_flags & IFF_LOOPBACK)
1111 		return (0);
1112 
1113 	/*
1114 	 * MLDv1 reports must originate from a host's link-local address,
1115 	 * or the unspecified address (when booting).
1116 	 */
1117 	src = ip6->ip6_src;
1118 	in6_clearscope(&src);
1119 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1120 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1121 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1122 		    ifp, if_name(ifp));
1123 		return (EINVAL);
1124 	}
1125 
1126 	/*
1127 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1128 	 * group, and must be directed to the group itself.
1129 	 */
1130 	dst = ip6->ip6_dst;
1131 	in6_clearscope(&dst);
1132 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1133 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1134 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1135 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1136 		    ifp, if_name(ifp));
1137 		return (EINVAL);
1138 	}
1139 
1140 	/*
1141 	 * Make sure we don't hear our own membership report, as fast
1142 	 * leave requires knowing that we are the only member of a
1143 	 * group. Assume we used the link-local address if available,
1144 	 * otherwise look for ::.
1145 	 *
1146 	 * XXX Note that scope ID comparison is needed for the address
1147 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1148 	 * performed for the on-wire address.
1149 	 */
1150 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1151 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1152 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1153 		if (ia != NULL)
1154 			ifa_free(&ia->ia_ifa);
1155 		return (0);
1156 	}
1157 	if (ia != NULL)
1158 		ifa_free(&ia->ia_ifa);
1159 
1160 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1161 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1162 
1163 	/*
1164 	 * Embed scope ID of receiving interface in MLD query for lookup
1165 	 * whilst we don't hold other locks (due to KAME locking lameness).
1166 	 */
1167 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1168 		in6_setscope(&mld->mld_addr, ifp, NULL);
1169 
1170 	IN6_MULTI_LOCK();
1171 	MLD_LOCK();
1172 	IF_ADDR_RLOCK(ifp);
1173 
1174 	/*
1175 	 * MLDv1 report suppression.
1176 	 * If we are a member of this group, and our membership should be
1177 	 * reported, and our group timer is pending or about to be reset,
1178 	 * stop our group timer by transitioning to the 'lazy' state.
1179 	 */
1180 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1181 	if (inm != NULL) {
1182 		struct mld_ifinfo *mli;
1183 
1184 		mli = inm->in6m_mli;
1185 		KASSERT(mli != NULL,
1186 		    ("%s: no mli for ifp %p", __func__, ifp));
1187 
1188 		/*
1189 		 * If we are in MLDv2 host mode, do not allow the
1190 		 * other host's MLDv1 report to suppress our reports.
1191 		 */
1192 		if (mli->mli_version == MLD_VERSION_2)
1193 			goto out_locked;
1194 
1195 		inm->in6m_timer = 0;
1196 
1197 		switch (inm->in6m_state) {
1198 		case MLD_NOT_MEMBER:
1199 		case MLD_SILENT_MEMBER:
1200 		case MLD_SLEEPING_MEMBER:
1201 			break;
1202 		case MLD_REPORTING_MEMBER:
1203 		case MLD_IDLE_MEMBER:
1204 		case MLD_AWAKENING_MEMBER:
1205 			CTR3(KTR_MLD,
1206 			    "report suppressed for %s on ifp %p(%s)",
1207 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1208 			    ifp, if_name(ifp));
1209 		case MLD_LAZY_MEMBER:
1210 			inm->in6m_state = MLD_LAZY_MEMBER;
1211 			break;
1212 		case MLD_G_QUERY_PENDING_MEMBER:
1213 		case MLD_SG_QUERY_PENDING_MEMBER:
1214 		case MLD_LEAVING_MEMBER:
1215 			break;
1216 		}
1217 	}
1218 
1219 out_locked:
1220 	IF_ADDR_RUNLOCK(ifp);
1221 	MLD_UNLOCK();
1222 	IN6_MULTI_UNLOCK();
1223 
1224 	/* XXX Clear embedded scope ID as userland won't expect it. */
1225 	in6_clearscope(&mld->mld_addr);
1226 
1227 	return (0);
1228 }
1229 
1230 /*
1231  * MLD input path.
1232  *
1233  * Assume query messages which fit in a single ICMPv6 message header
1234  * have been pulled up.
1235  * Assume that userland will want to see the message, even if it
1236  * otherwise fails kernel input validation; do not free it.
1237  * Pullup may however free the mbuf chain m if it fails.
1238  *
1239  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1240  */
1241 int
1242 mld_input(struct mbuf *m, int off, int icmp6len)
1243 {
1244 	struct ifnet	*ifp;
1245 	struct ip6_hdr	*ip6;
1246 	struct mld_hdr	*mld;
1247 	int		 mldlen;
1248 
1249 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1250 
1251 	ifp = m->m_pkthdr.rcvif;
1252 
1253 	ip6 = mtod(m, struct ip6_hdr *);
1254 
1255 	/* Pullup to appropriate size. */
1256 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1257 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1258 	    icmp6len >= sizeof(struct mldv2_query)) {
1259 		mldlen = sizeof(struct mldv2_query);
1260 	} else {
1261 		mldlen = sizeof(struct mld_hdr);
1262 	}
1263 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1264 	if (mld == NULL) {
1265 		ICMP6STAT_INC(icp6s_badlen);
1266 		return (IPPROTO_DONE);
1267 	}
1268 
1269 	/*
1270 	 * Userland needs to see all of this traffic for implementing
1271 	 * the endpoint discovery portion of multicast routing.
1272 	 */
1273 	switch (mld->mld_type) {
1274 	case MLD_LISTENER_QUERY:
1275 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1276 		if (icmp6len == sizeof(struct mld_hdr)) {
1277 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1278 				return (0);
1279 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1280 			if (mld_v2_input_query(ifp, ip6, m, off,
1281 			    icmp6len) != 0)
1282 				return (0);
1283 		}
1284 		break;
1285 	case MLD_LISTENER_REPORT:
1286 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1287 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1288 			return (0);
1289 		break;
1290 	case MLDV2_LISTENER_REPORT:
1291 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1292 		break;
1293 	case MLD_LISTENER_DONE:
1294 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1295 		break;
1296 	default:
1297 		break;
1298 	}
1299 
1300 	return (0);
1301 }
1302 
1303 /*
1304  * Fast timeout handler (global).
1305  * VIMAGE: Timeout handlers are expected to service all vimages.
1306  */
1307 void
1308 mld_fasttimo(void)
1309 {
1310 	VNET_ITERATOR_DECL(vnet_iter);
1311 
1312 	VNET_LIST_RLOCK_NOSLEEP();
1313 	VNET_FOREACH(vnet_iter) {
1314 		CURVNET_SET(vnet_iter);
1315 		mld_fasttimo_vnet();
1316 		CURVNET_RESTORE();
1317 	}
1318 	VNET_LIST_RUNLOCK_NOSLEEP();
1319 }
1320 
1321 /*
1322  * Fast timeout handler (per-vnet).
1323  *
1324  * VIMAGE: Assume caller has set up our curvnet.
1325  */
1326 static void
1327 mld_fasttimo_vnet(void)
1328 {
1329 	struct ifqueue		 scq;	/* State-change packets */
1330 	struct ifqueue		 qrq;	/* Query response packets */
1331 	struct ifnet		*ifp;
1332 	struct mld_ifinfo	*mli;
1333 	struct ifmultiaddr	*ifma;
1334 	struct in6_multi	*inm, *tinm;
1335 	int			 uri_fasthz;
1336 
1337 	uri_fasthz = 0;
1338 
1339 	/*
1340 	 * Quick check to see if any work needs to be done, in order to
1341 	 * minimize the overhead of fasttimo processing.
1342 	 * SMPng: XXX Unlocked reads.
1343 	 */
1344 	if (!V_current_state_timers_running6 &&
1345 	    !V_interface_timers_running6 &&
1346 	    !V_state_change_timers_running6)
1347 		return;
1348 
1349 	IN6_MULTI_LOCK();
1350 	MLD_LOCK();
1351 
1352 	/*
1353 	 * MLDv2 General Query response timer processing.
1354 	 */
1355 	if (V_interface_timers_running6) {
1356 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1357 
1358 		V_interface_timers_running6 = 0;
1359 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1360 			if (mli->mli_v2_timer == 0) {
1361 				/* Do nothing. */
1362 			} else if (--mli->mli_v2_timer == 0) {
1363 				mld_v2_dispatch_general_query(mli);
1364 			} else {
1365 				V_interface_timers_running6 = 1;
1366 			}
1367 		}
1368 	}
1369 
1370 	if (!V_current_state_timers_running6 &&
1371 	    !V_state_change_timers_running6)
1372 		goto out_locked;
1373 
1374 	V_current_state_timers_running6 = 0;
1375 	V_state_change_timers_running6 = 0;
1376 
1377 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1378 
1379 	/*
1380 	 * MLD host report and state-change timer processing.
1381 	 * Note: Processing a v2 group timer may remove a node.
1382 	 */
1383 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1384 		ifp = mli->mli_ifp;
1385 
1386 		if (mli->mli_version == MLD_VERSION_2) {
1387 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1388 			    PR_FASTHZ);
1389 
1390 			memset(&qrq, 0, sizeof(struct ifqueue));
1391 			IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS);
1392 
1393 			memset(&scq, 0, sizeof(struct ifqueue));
1394 			IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1395 		}
1396 
1397 		IF_ADDR_RLOCK(ifp);
1398 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1399 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1400 			    ifma->ifma_protospec == NULL)
1401 				continue;
1402 			inm = (struct in6_multi *)ifma->ifma_protospec;
1403 			switch (mli->mli_version) {
1404 			case MLD_VERSION_1:
1405 				mld_v1_process_group_timer(mli, inm);
1406 				break;
1407 			case MLD_VERSION_2:
1408 				mld_v2_process_group_timers(mli, &qrq,
1409 				    &scq, inm, uri_fasthz);
1410 				break;
1411 			}
1412 		}
1413 		IF_ADDR_RUNLOCK(ifp);
1414 
1415 		switch (mli->mli_version) {
1416 		case MLD_VERSION_1:
1417 			/*
1418 			 * Transmit reports for this lifecycle.  This
1419 			 * is done while not holding IF_ADDR_LOCK
1420 			 * since this can call
1421 			 * in6ifa_ifpforlinklocal() which locks
1422 			 * IF_ADDR_LOCK internally as well as
1423 			 * ip6_output() to transmit a packet.
1424 			 */
1425 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1426 			    in6m_nrele, tinm) {
1427 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1428 				    in6m_nrele);
1429 				(void)mld_v1_transmit_report(inm,
1430 				    MLD_LISTENER_REPORT);
1431 			}
1432 			break;
1433 		case MLD_VERSION_2:
1434 			mld_dispatch_queue(&qrq, 0);
1435 			mld_dispatch_queue(&scq, 0);
1436 
1437 			/*
1438 			 * Free the in_multi reference(s) for
1439 			 * this lifecycle.
1440 			 */
1441 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1442 			    in6m_nrele, tinm) {
1443 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1444 				    in6m_nrele);
1445 				in6m_release_locked(inm);
1446 			}
1447 			break;
1448 		}
1449 	}
1450 
1451 out_locked:
1452 	MLD_UNLOCK();
1453 	IN6_MULTI_UNLOCK();
1454 }
1455 
1456 /*
1457  * Update host report group timer.
1458  * Will update the global pending timer flags.
1459  */
1460 static void
1461 mld_v1_process_group_timer(struct mld_ifinfo *mli, struct in6_multi *inm)
1462 {
1463 	int report_timer_expired;
1464 
1465 	IN6_MULTI_LOCK_ASSERT();
1466 	MLD_LOCK_ASSERT();
1467 
1468 	if (inm->in6m_timer == 0) {
1469 		report_timer_expired = 0;
1470 	} else if (--inm->in6m_timer == 0) {
1471 		report_timer_expired = 1;
1472 	} else {
1473 		V_current_state_timers_running6 = 1;
1474 		return;
1475 	}
1476 
1477 	switch (inm->in6m_state) {
1478 	case MLD_NOT_MEMBER:
1479 	case MLD_SILENT_MEMBER:
1480 	case MLD_IDLE_MEMBER:
1481 	case MLD_LAZY_MEMBER:
1482 	case MLD_SLEEPING_MEMBER:
1483 	case MLD_AWAKENING_MEMBER:
1484 		break;
1485 	case MLD_REPORTING_MEMBER:
1486 		if (report_timer_expired) {
1487 			inm->in6m_state = MLD_IDLE_MEMBER;
1488 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
1489 			    in6m_nrele);
1490 		}
1491 		break;
1492 	case MLD_G_QUERY_PENDING_MEMBER:
1493 	case MLD_SG_QUERY_PENDING_MEMBER:
1494 	case MLD_LEAVING_MEMBER:
1495 		break;
1496 	}
1497 }
1498 
1499 /*
1500  * Update a group's timers for MLDv2.
1501  * Will update the global pending timer flags.
1502  * Note: Unlocked read from mli.
1503  */
1504 static void
1505 mld_v2_process_group_timers(struct mld_ifinfo *mli,
1506     struct ifqueue *qrq, struct ifqueue *scq,
1507     struct in6_multi *inm, const int uri_fasthz)
1508 {
1509 	int query_response_timer_expired;
1510 	int state_change_retransmit_timer_expired;
1511 #ifdef KTR
1512 	char ip6tbuf[INET6_ADDRSTRLEN];
1513 #endif
1514 
1515 	IN6_MULTI_LOCK_ASSERT();
1516 	MLD_LOCK_ASSERT();
1517 
1518 	query_response_timer_expired = 0;
1519 	state_change_retransmit_timer_expired = 0;
1520 
1521 	/*
1522 	 * During a transition from compatibility mode back to MLDv2,
1523 	 * a group record in REPORTING state may still have its group
1524 	 * timer active. This is a no-op in this function; it is easier
1525 	 * to deal with it here than to complicate the slow-timeout path.
1526 	 */
1527 	if (inm->in6m_timer == 0) {
1528 		query_response_timer_expired = 0;
1529 	} else if (--inm->in6m_timer == 0) {
1530 		query_response_timer_expired = 1;
1531 	} else {
1532 		V_current_state_timers_running6 = 1;
1533 	}
1534 
1535 	if (inm->in6m_sctimer == 0) {
1536 		state_change_retransmit_timer_expired = 0;
1537 	} else if (--inm->in6m_sctimer == 0) {
1538 		state_change_retransmit_timer_expired = 1;
1539 	} else {
1540 		V_state_change_timers_running6 = 1;
1541 	}
1542 
1543 	/* We are in fasttimo, so be quick about it. */
1544 	if (!state_change_retransmit_timer_expired &&
1545 	    !query_response_timer_expired)
1546 		return;
1547 
1548 	switch (inm->in6m_state) {
1549 	case MLD_NOT_MEMBER:
1550 	case MLD_SILENT_MEMBER:
1551 	case MLD_SLEEPING_MEMBER:
1552 	case MLD_LAZY_MEMBER:
1553 	case MLD_AWAKENING_MEMBER:
1554 	case MLD_IDLE_MEMBER:
1555 		break;
1556 	case MLD_G_QUERY_PENDING_MEMBER:
1557 	case MLD_SG_QUERY_PENDING_MEMBER:
1558 		/*
1559 		 * Respond to a previously pending Group-Specific
1560 		 * or Group-and-Source-Specific query by enqueueing
1561 		 * the appropriate Current-State report for
1562 		 * immediate transmission.
1563 		 */
1564 		if (query_response_timer_expired) {
1565 			int retval;
1566 
1567 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1568 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1569 			    0);
1570 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1571 			    __func__, retval);
1572 			inm->in6m_state = MLD_REPORTING_MEMBER;
1573 			in6m_clear_recorded(inm);
1574 		}
1575 		/* FALLTHROUGH */
1576 	case MLD_REPORTING_MEMBER:
1577 	case MLD_LEAVING_MEMBER:
1578 		if (state_change_retransmit_timer_expired) {
1579 			/*
1580 			 * State-change retransmission timer fired.
1581 			 * If there are any further pending retransmissions,
1582 			 * set the global pending state-change flag, and
1583 			 * reset the timer.
1584 			 */
1585 			if (--inm->in6m_scrv > 0) {
1586 				inm->in6m_sctimer = uri_fasthz;
1587 				V_state_change_timers_running6 = 1;
1588 			}
1589 			/*
1590 			 * Retransmit the previously computed state-change
1591 			 * report. If there are no further pending
1592 			 * retransmissions, the mbuf queue will be consumed.
1593 			 * Update T0 state to T1 as we have now sent
1594 			 * a state-change.
1595 			 */
1596 			(void)mld_v2_merge_state_changes(inm, scq);
1597 
1598 			in6m_commit(inm);
1599 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1600 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1601 			    if_name(inm->in6m_ifp));
1602 
1603 			/*
1604 			 * If we are leaving the group for good, make sure
1605 			 * we release MLD's reference to it.
1606 			 * This release must be deferred using a SLIST,
1607 			 * as we are called from a loop which traverses
1608 			 * the in_ifmultiaddr TAILQ.
1609 			 */
1610 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1611 			    inm->in6m_scrv == 0) {
1612 				inm->in6m_state = MLD_NOT_MEMBER;
1613 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1614 				    inm, in6m_nrele);
1615 			}
1616 		}
1617 		break;
1618 	}
1619 }
1620 
1621 /*
1622  * Switch to a different version on the given interface,
1623  * as per Section 9.12.
1624  */
1625 static void
1626 mld_set_version(struct mld_ifinfo *mli, const int version)
1627 {
1628 	int old_version_timer;
1629 
1630 	MLD_LOCK_ASSERT();
1631 
1632 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1633 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1634 
1635 	if (version == MLD_VERSION_1) {
1636 		/*
1637 		 * Compute the "Older Version Querier Present" timer as per
1638 		 * Section 9.12.
1639 		 */
1640 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1641 		old_version_timer *= PR_SLOWHZ;
1642 		mli->mli_v1_timer = old_version_timer;
1643 	}
1644 
1645 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1646 		mli->mli_version = MLD_VERSION_1;
1647 		mld_v2_cancel_link_timers(mli);
1648 	}
1649 }
1650 
1651 /*
1652  * Cancel pending MLDv2 timers for the given link and all groups
1653  * joined on it; state-change, general-query, and group-query timers.
1654  */
1655 static void
1656 mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
1657 {
1658 	struct ifmultiaddr	*ifma;
1659 	struct ifnet		*ifp;
1660 	struct in6_multi	*inm, *tinm;
1661 
1662 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1663 	    mli->mli_ifp, if_name(mli->mli_ifp));
1664 
1665 	IN6_MULTI_LOCK_ASSERT();
1666 	MLD_LOCK_ASSERT();
1667 
1668 	/*
1669 	 * Fast-track this potentially expensive operation
1670 	 * by checking all the global 'timer pending' flags.
1671 	 */
1672 	if (!V_interface_timers_running6 &&
1673 	    !V_state_change_timers_running6 &&
1674 	    !V_current_state_timers_running6)
1675 		return;
1676 
1677 	mli->mli_v2_timer = 0;
1678 
1679 	ifp = mli->mli_ifp;
1680 
1681 	IF_ADDR_RLOCK(ifp);
1682 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1683 		if (ifma->ifma_addr->sa_family != AF_INET6)
1684 			continue;
1685 		inm = (struct in6_multi *)ifma->ifma_protospec;
1686 		switch (inm->in6m_state) {
1687 		case MLD_NOT_MEMBER:
1688 		case MLD_SILENT_MEMBER:
1689 		case MLD_IDLE_MEMBER:
1690 		case MLD_LAZY_MEMBER:
1691 		case MLD_SLEEPING_MEMBER:
1692 		case MLD_AWAKENING_MEMBER:
1693 			break;
1694 		case MLD_LEAVING_MEMBER:
1695 			/*
1696 			 * If we are leaving the group and switching
1697 			 * version, we need to release the final
1698 			 * reference held for issuing the INCLUDE {}.
1699 			 */
1700 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
1701 			    in6m_nrele);
1702 			/* FALLTHROUGH */
1703 		case MLD_G_QUERY_PENDING_MEMBER:
1704 		case MLD_SG_QUERY_PENDING_MEMBER:
1705 			in6m_clear_recorded(inm);
1706 			/* FALLTHROUGH */
1707 		case MLD_REPORTING_MEMBER:
1708 			inm->in6m_sctimer = 0;
1709 			inm->in6m_timer = 0;
1710 			inm->in6m_state = MLD_REPORTING_MEMBER;
1711 			/*
1712 			 * Free any pending MLDv2 state-change records.
1713 			 */
1714 			_IF_DRAIN(&inm->in6m_scq);
1715 			break;
1716 		}
1717 	}
1718 	IF_ADDR_RUNLOCK(ifp);
1719 	SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) {
1720 		SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
1721 		in6m_release_locked(inm);
1722 	}
1723 }
1724 
1725 /*
1726  * Global slowtimo handler.
1727  * VIMAGE: Timeout handlers are expected to service all vimages.
1728  */
1729 void
1730 mld_slowtimo(void)
1731 {
1732 	VNET_ITERATOR_DECL(vnet_iter);
1733 
1734 	VNET_LIST_RLOCK_NOSLEEP();
1735 	VNET_FOREACH(vnet_iter) {
1736 		CURVNET_SET(vnet_iter);
1737 		mld_slowtimo_vnet();
1738 		CURVNET_RESTORE();
1739 	}
1740 	VNET_LIST_RUNLOCK_NOSLEEP();
1741 }
1742 
1743 /*
1744  * Per-vnet slowtimo handler.
1745  */
1746 static void
1747 mld_slowtimo_vnet(void)
1748 {
1749 	struct mld_ifinfo *mli;
1750 
1751 	MLD_LOCK();
1752 
1753 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1754 		mld_v1_process_querier_timers(mli);
1755 	}
1756 
1757 	MLD_UNLOCK();
1758 }
1759 
1760 /*
1761  * Update the Older Version Querier Present timers for a link.
1762  * See Section 9.12 of RFC 3810.
1763  */
1764 static void
1765 mld_v1_process_querier_timers(struct mld_ifinfo *mli)
1766 {
1767 
1768 	MLD_LOCK_ASSERT();
1769 
1770 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1771 		/*
1772 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1773 		 */
1774 		CTR5(KTR_MLD,
1775 		    "%s: transition from v%d -> v%d on %p(%s)",
1776 		    __func__, mli->mli_version, MLD_VERSION_2,
1777 		    mli->mli_ifp, if_name(mli->mli_ifp));
1778 		mli->mli_version = MLD_VERSION_2;
1779 	}
1780 }
1781 
1782 /*
1783  * Transmit an MLDv1 report immediately.
1784  */
1785 static int
1786 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1787 {
1788 	struct ifnet		*ifp;
1789 	struct in6_ifaddr	*ia;
1790 	struct ip6_hdr		*ip6;
1791 	struct mbuf		*mh, *md;
1792 	struct mld_hdr		*mld;
1793 
1794 	IN6_MULTI_LOCK_ASSERT();
1795 	MLD_LOCK_ASSERT();
1796 
1797 	ifp = in6m->in6m_ifp;
1798 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1799 	/* ia may be NULL if link-local address is tentative. */
1800 
1801 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1802 	if (mh == NULL) {
1803 		if (ia != NULL)
1804 			ifa_free(&ia->ia_ifa);
1805 		return (ENOMEM);
1806 	}
1807 	md = m_get(M_NOWAIT, MT_DATA);
1808 	if (md == NULL) {
1809 		m_free(mh);
1810 		if (ia != NULL)
1811 			ifa_free(&ia->ia_ifa);
1812 		return (ENOMEM);
1813 	}
1814 	mh->m_next = md;
1815 
1816 	/*
1817 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1818 	 * that ether_output() does not need to allocate another mbuf
1819 	 * for the header in the most common case.
1820 	 */
1821 	MH_ALIGN(mh, sizeof(struct ip6_hdr));
1822 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1823 	mh->m_len = sizeof(struct ip6_hdr);
1824 
1825 	ip6 = mtod(mh, struct ip6_hdr *);
1826 	ip6->ip6_flow = 0;
1827 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1828 	ip6->ip6_vfc |= IPV6_VERSION;
1829 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1830 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1831 	ip6->ip6_dst = in6m->in6m_addr;
1832 
1833 	md->m_len = sizeof(struct mld_hdr);
1834 	mld = mtod(md, struct mld_hdr *);
1835 	mld->mld_type = type;
1836 	mld->mld_code = 0;
1837 	mld->mld_cksum = 0;
1838 	mld->mld_maxdelay = 0;
1839 	mld->mld_reserved = 0;
1840 	mld->mld_addr = in6m->in6m_addr;
1841 	in6_clearscope(&mld->mld_addr);
1842 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1843 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1844 
1845 	mld_save_context(mh, ifp);
1846 	mh->m_flags |= M_MLDV1;
1847 
1848 	mld_dispatch_packet(mh);
1849 
1850 	if (ia != NULL)
1851 		ifa_free(&ia->ia_ifa);
1852 	return (0);
1853 }
1854 
1855 /*
1856  * Process a state change from the upper layer for the given IPv6 group.
1857  *
1858  * Each socket holds a reference on the in_multi in its own ip_moptions.
1859  * The socket layer will have made the necessary updates to.the group
1860  * state, it is now up to MLD to issue a state change report if there
1861  * has been any change between T0 (when the last state-change was issued)
1862  * and T1 (now).
1863  *
1864  * We use the MLDv2 state machine at group level. The MLd module
1865  * however makes the decision as to which MLD protocol version to speak.
1866  * A state change *from* INCLUDE {} always means an initial join.
1867  * A state change *to* INCLUDE {} always means a final leave.
1868  *
1869  * If delay is non-zero, and the state change is an initial multicast
1870  * join, the state change report will be delayed by 'delay' ticks
1871  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1872  * the initial MLDv2 state change report will be delayed by whichever
1873  * is sooner, a pending state-change timer or delay itself.
1874  *
1875  * VIMAGE: curvnet should have been set by caller, as this routine
1876  * is called from the socket option handlers.
1877  */
1878 int
1879 mld_change_state(struct in6_multi *inm, const int delay)
1880 {
1881 	struct mld_ifinfo *mli;
1882 	struct ifnet *ifp;
1883 	int error;
1884 
1885 	IN6_MULTI_LOCK_ASSERT();
1886 
1887 	error = 0;
1888 
1889 	/*
1890 	 * Try to detect if the upper layer just asked us to change state
1891 	 * for an interface which has now gone away.
1892 	 */
1893 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1894 	ifp = inm->in6m_ifma->ifma_ifp;
1895 	if (ifp != NULL) {
1896 		/*
1897 		 * Sanity check that netinet6's notion of ifp is the
1898 		 * same as net's.
1899 		 */
1900 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1901 	}
1902 
1903 	MLD_LOCK();
1904 
1905 	mli = MLD_IFINFO(ifp);
1906 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
1907 
1908 	/*
1909 	 * If we detect a state transition to or from MCAST_UNDEFINED
1910 	 * for this group, then we are starting or finishing an MLD
1911 	 * life cycle for this group.
1912 	 */
1913 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1914 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1915 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1916 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1917 			CTR1(KTR_MLD, "%s: initial join", __func__);
1918 			error = mld_initial_join(inm, mli, delay);
1919 			goto out_locked;
1920 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1921 			CTR1(KTR_MLD, "%s: final leave", __func__);
1922 			mld_final_leave(inm, mli);
1923 			goto out_locked;
1924 		}
1925 	} else {
1926 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1927 	}
1928 
1929 	error = mld_handle_state_change(inm, mli);
1930 
1931 out_locked:
1932 	MLD_UNLOCK();
1933 	return (error);
1934 }
1935 
1936 /*
1937  * Perform the initial join for an MLD group.
1938  *
1939  * When joining a group:
1940  *  If the group should have its MLD traffic suppressed, do nothing.
1941  *  MLDv1 starts sending MLDv1 host membership reports.
1942  *  MLDv2 will schedule an MLDv2 state-change report containing the
1943  *  initial state of the membership.
1944  *
1945  * If the delay argument is non-zero, then we must delay sending the
1946  * initial state change for delay ticks (in units of PR_FASTHZ).
1947  */
1948 static int
1949 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
1950     const int delay)
1951 {
1952 	struct ifnet		*ifp;
1953 	struct ifqueue		*ifq;
1954 	int			 error, retval, syncstates;
1955 	int			 odelay;
1956 #ifdef KTR
1957 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1958 #endif
1959 
1960 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1961 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1962 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1963 
1964 	error = 0;
1965 	syncstates = 1;
1966 
1967 	ifp = inm->in6m_ifp;
1968 
1969 	IN6_MULTI_LOCK_ASSERT();
1970 	MLD_LOCK_ASSERT();
1971 
1972 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1973 
1974 	/*
1975 	 * Groups joined on loopback or marked as 'not reported',
1976 	 * enter the MLD_SILENT_MEMBER state and
1977 	 * are never reported in any protocol exchanges.
1978 	 * All other groups enter the appropriate state machine
1979 	 * for the version in use on this link.
1980 	 * A link marked as MLIF_SILENT causes MLD to be completely
1981 	 * disabled for the link.
1982 	 */
1983 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1984 	    (mli->mli_flags & MLIF_SILENT) ||
1985 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1986 		CTR1(KTR_MLD,
1987 "%s: not kicking state machine for silent group", __func__);
1988 		inm->in6m_state = MLD_SILENT_MEMBER;
1989 		inm->in6m_timer = 0;
1990 	} else {
1991 		/*
1992 		 * Deal with overlapping in_multi lifecycle.
1993 		 * If this group was LEAVING, then make sure
1994 		 * we drop the reference we picked up to keep the
1995 		 * group around for the final INCLUDE {} enqueue.
1996 		 */
1997 		if (mli->mli_version == MLD_VERSION_2 &&
1998 		    inm->in6m_state == MLD_LEAVING_MEMBER)
1999 			in6m_release_locked(inm);
2000 
2001 		inm->in6m_state = MLD_REPORTING_MEMBER;
2002 
2003 		switch (mli->mli_version) {
2004 		case MLD_VERSION_1:
2005 			/*
2006 			 * If a delay was provided, only use it if
2007 			 * it is greater than the delay normally
2008 			 * used for an MLDv1 state change report,
2009 			 * and delay sending the initial MLDv1 report
2010 			 * by not transitioning to the IDLE state.
2011 			 */
2012 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2013 			if (delay) {
2014 				inm->in6m_timer = max(delay, odelay);
2015 				V_current_state_timers_running6 = 1;
2016 			} else {
2017 				inm->in6m_state = MLD_IDLE_MEMBER;
2018 				error = mld_v1_transmit_report(inm,
2019 				     MLD_LISTENER_REPORT);
2020 				if (error == 0) {
2021 					inm->in6m_timer = odelay;
2022 					V_current_state_timers_running6 = 1;
2023 				}
2024 			}
2025 			break;
2026 
2027 		case MLD_VERSION_2:
2028 			/*
2029 			 * Defer update of T0 to T1, until the first copy
2030 			 * of the state change has been transmitted.
2031 			 */
2032 			syncstates = 0;
2033 
2034 			/*
2035 			 * Immediately enqueue a State-Change Report for
2036 			 * this interface, freeing any previous reports.
2037 			 * Don't kick the timers if there is nothing to do,
2038 			 * or if an error occurred.
2039 			 */
2040 			ifq = &inm->in6m_scq;
2041 			_IF_DRAIN(ifq);
2042 			retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2043 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2044 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2045 			    __func__, retval);
2046 			if (retval <= 0) {
2047 				error = retval * -1;
2048 				break;
2049 			}
2050 
2051 			/*
2052 			 * Schedule transmission of pending state-change
2053 			 * report up to RV times for this link. The timer
2054 			 * will fire at the next mld_fasttimo (~200ms),
2055 			 * giving us an opportunity to merge the reports.
2056 			 *
2057 			 * If a delay was provided to this function, only
2058 			 * use this delay if sooner than the existing one.
2059 			 */
2060 			KASSERT(mli->mli_rv > 1,
2061 			   ("%s: invalid robustness %d", __func__,
2062 			    mli->mli_rv));
2063 			inm->in6m_scrv = mli->mli_rv;
2064 			if (delay) {
2065 				if (inm->in6m_sctimer > 1) {
2066 					inm->in6m_sctimer =
2067 					    min(inm->in6m_sctimer, delay);
2068 				} else
2069 					inm->in6m_sctimer = delay;
2070 			} else
2071 				inm->in6m_sctimer = 1;
2072 			V_state_change_timers_running6 = 1;
2073 
2074 			error = 0;
2075 			break;
2076 		}
2077 	}
2078 
2079 	/*
2080 	 * Only update the T0 state if state change is atomic,
2081 	 * i.e. we don't need to wait for a timer to fire before we
2082 	 * can consider the state change to have been communicated.
2083 	 */
2084 	if (syncstates) {
2085 		in6m_commit(inm);
2086 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2087 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2088 		    if_name(inm->in6m_ifp));
2089 	}
2090 
2091 	return (error);
2092 }
2093 
2094 /*
2095  * Issue an intermediate state change during the life-cycle.
2096  */
2097 static int
2098 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli)
2099 {
2100 	struct ifnet		*ifp;
2101 	int			 retval;
2102 #ifdef KTR
2103 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2104 #endif
2105 
2106 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2107 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2108 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2109 
2110 	ifp = inm->in6m_ifp;
2111 
2112 	IN6_MULTI_LOCK_ASSERT();
2113 	MLD_LOCK_ASSERT();
2114 
2115 	KASSERT(mli && mli->mli_ifp == ifp,
2116 	    ("%s: inconsistent ifp", __func__));
2117 
2118 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2119 	    (mli->mli_flags & MLIF_SILENT) ||
2120 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2121 	    (mli->mli_version != MLD_VERSION_2)) {
2122 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2123 			CTR1(KTR_MLD,
2124 "%s: not kicking state machine for silent group", __func__);
2125 		}
2126 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2127 		in6m_commit(inm);
2128 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2129 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2130 		    if_name(inm->in6m_ifp));
2131 		return (0);
2132 	}
2133 
2134 	_IF_DRAIN(&inm->in6m_scq);
2135 
2136 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2137 	    (mli->mli_flags & MLIF_USEALLOW));
2138 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2139 	if (retval <= 0)
2140 		return (-retval);
2141 
2142 	/*
2143 	 * If record(s) were enqueued, start the state-change
2144 	 * report timer for this group.
2145 	 */
2146 	inm->in6m_scrv = mli->mli_rv;
2147 	inm->in6m_sctimer = 1;
2148 	V_state_change_timers_running6 = 1;
2149 
2150 	return (0);
2151 }
2152 
2153 /*
2154  * Perform the final leave for a multicast address.
2155  *
2156  * When leaving a group:
2157  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2158  *  MLDv2 enqueues a state-change report containing a transition
2159  *  to INCLUDE {} for immediate transmission.
2160  */
2161 static void
2162 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli)
2163 {
2164 	int syncstates;
2165 #ifdef KTR
2166 	char ip6tbuf[INET6_ADDRSTRLEN];
2167 #endif
2168 
2169 	syncstates = 1;
2170 
2171 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2172 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2173 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2174 
2175 	IN6_MULTI_LOCK_ASSERT();
2176 	MLD_LOCK_ASSERT();
2177 
2178 	switch (inm->in6m_state) {
2179 	case MLD_NOT_MEMBER:
2180 	case MLD_SILENT_MEMBER:
2181 	case MLD_LEAVING_MEMBER:
2182 		/* Already leaving or left; do nothing. */
2183 		CTR1(KTR_MLD,
2184 "%s: not kicking state machine for silent group", __func__);
2185 		break;
2186 	case MLD_REPORTING_MEMBER:
2187 	case MLD_IDLE_MEMBER:
2188 	case MLD_G_QUERY_PENDING_MEMBER:
2189 	case MLD_SG_QUERY_PENDING_MEMBER:
2190 		if (mli->mli_version == MLD_VERSION_1) {
2191 #ifdef INVARIANTS
2192 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2193 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2194 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2195 			     __func__);
2196 #endif
2197 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2198 			inm->in6m_state = MLD_NOT_MEMBER;
2199 			V_current_state_timers_running6 = 1;
2200 		} else if (mli->mli_version == MLD_VERSION_2) {
2201 			/*
2202 			 * Stop group timer and all pending reports.
2203 			 * Immediately enqueue a state-change report
2204 			 * TO_IN {} to be sent on the next fast timeout,
2205 			 * giving us an opportunity to merge reports.
2206 			 */
2207 			_IF_DRAIN(&inm->in6m_scq);
2208 			inm->in6m_timer = 0;
2209 			inm->in6m_scrv = mli->mli_rv;
2210 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2211 			    "pending retransmissions.", __func__,
2212 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2213 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2214 			if (inm->in6m_scrv == 0) {
2215 				inm->in6m_state = MLD_NOT_MEMBER;
2216 				inm->in6m_sctimer = 0;
2217 			} else {
2218 				int retval;
2219 
2220 				in6m_acquire_locked(inm);
2221 
2222 				retval = mld_v2_enqueue_group_record(
2223 				    &inm->in6m_scq, inm, 1, 0, 0,
2224 				    (mli->mli_flags & MLIF_USEALLOW));
2225 				KASSERT(retval != 0,
2226 				    ("%s: enqueue record = %d", __func__,
2227 				     retval));
2228 
2229 				inm->in6m_state = MLD_LEAVING_MEMBER;
2230 				inm->in6m_sctimer = 1;
2231 				V_state_change_timers_running6 = 1;
2232 				syncstates = 0;
2233 			}
2234 			break;
2235 		}
2236 		break;
2237 	case MLD_LAZY_MEMBER:
2238 	case MLD_SLEEPING_MEMBER:
2239 	case MLD_AWAKENING_MEMBER:
2240 		/* Our reports are suppressed; do nothing. */
2241 		break;
2242 	}
2243 
2244 	if (syncstates) {
2245 		in6m_commit(inm);
2246 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2247 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2248 		    if_name(inm->in6m_ifp));
2249 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2250 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2251 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2252 	}
2253 }
2254 
2255 /*
2256  * Enqueue an MLDv2 group record to the given output queue.
2257  *
2258  * If is_state_change is zero, a current-state record is appended.
2259  * If is_state_change is non-zero, a state-change report is appended.
2260  *
2261  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2262  * If is_group_query is zero, and if there is a packet with free space
2263  * at the tail of the queue, it will be appended to providing there
2264  * is enough free space.
2265  * Otherwise a new mbuf packet chain is allocated.
2266  *
2267  * If is_source_query is non-zero, each source is checked to see if
2268  * it was recorded for a Group-Source query, and will be omitted if
2269  * it is not both in-mode and recorded.
2270  *
2271  * If use_block_allow is non-zero, state change reports for initial join
2272  * and final leave, on an inclusive mode group with a source list, will be
2273  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2274  *
2275  * The function will attempt to allocate leading space in the packet
2276  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2277  *
2278  * If successful the size of all data appended to the queue is returned,
2279  * otherwise an error code less than zero is returned, or zero if
2280  * no record(s) were appended.
2281  */
2282 static int
2283 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2284     const int is_state_change, const int is_group_query,
2285     const int is_source_query, const int use_block_allow)
2286 {
2287 	struct mldv2_record	 mr;
2288 	struct mldv2_record	*pmr;
2289 	struct ifnet		*ifp;
2290 	struct ip6_msource	*ims, *nims;
2291 	struct mbuf		*m0, *m, *md;
2292 	int			 error, is_filter_list_change;
2293 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2294 	int			 record_has_sources;
2295 	int			 now;
2296 	int			 type;
2297 	uint8_t			 mode;
2298 #ifdef KTR
2299 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2300 #endif
2301 
2302 	IN6_MULTI_LOCK_ASSERT();
2303 
2304 	error = 0;
2305 	ifp = inm->in6m_ifp;
2306 	is_filter_list_change = 0;
2307 	m = NULL;
2308 	m0 = NULL;
2309 	m0srcs = 0;
2310 	msrcs = 0;
2311 	nbytes = 0;
2312 	nims = NULL;
2313 	record_has_sources = 1;
2314 	pmr = NULL;
2315 	type = MLD_DO_NOTHING;
2316 	mode = inm->in6m_st[1].iss_fmode;
2317 
2318 	/*
2319 	 * If we did not transition out of ASM mode during t0->t1,
2320 	 * and there are no source nodes to process, we can skip
2321 	 * the generation of source records.
2322 	 */
2323 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2324 	    inm->in6m_nsrc == 0)
2325 		record_has_sources = 0;
2326 
2327 	if (is_state_change) {
2328 		/*
2329 		 * Queue a state change record.
2330 		 * If the mode did not change, and there are non-ASM
2331 		 * listeners or source filters present,
2332 		 * we potentially need to issue two records for the group.
2333 		 * If there are ASM listeners, and there was no filter
2334 		 * mode transition of any kind, do nothing.
2335 		 *
2336 		 * If we are transitioning to MCAST_UNDEFINED, we need
2337 		 * not send any sources. A transition to/from this state is
2338 		 * considered inclusive with some special treatment.
2339 		 *
2340 		 * If we are rewriting initial joins/leaves to use
2341 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2342 		 * we need to send sources in all cases.
2343 		 */
2344 		if (mode != inm->in6m_st[0].iss_fmode) {
2345 			if (mode == MCAST_EXCLUDE) {
2346 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2347 				    __func__);
2348 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2349 			} else {
2350 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2351 				    __func__);
2352 				if (use_block_allow) {
2353 					/*
2354 					 * XXX
2355 					 * Here we're interested in state
2356 					 * edges either direction between
2357 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2358 					 * Perhaps we should just check
2359 					 * the group state, rather than
2360 					 * the filter mode.
2361 					 */
2362 					if (mode == MCAST_UNDEFINED) {
2363 						type = MLD_BLOCK_OLD_SOURCES;
2364 					} else {
2365 						type = MLD_ALLOW_NEW_SOURCES;
2366 					}
2367 				} else {
2368 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2369 					if (mode == MCAST_UNDEFINED)
2370 						record_has_sources = 0;
2371 				}
2372 			}
2373 		} else {
2374 			if (record_has_sources) {
2375 				is_filter_list_change = 1;
2376 			} else {
2377 				type = MLD_DO_NOTHING;
2378 			}
2379 		}
2380 	} else {
2381 		/*
2382 		 * Queue a current state record.
2383 		 */
2384 		if (mode == MCAST_EXCLUDE) {
2385 			type = MLD_MODE_IS_EXCLUDE;
2386 		} else if (mode == MCAST_INCLUDE) {
2387 			type = MLD_MODE_IS_INCLUDE;
2388 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2389 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2390 			     __func__, inm, inm->in6m_st[1].iss_asm));
2391 		}
2392 	}
2393 
2394 	/*
2395 	 * Generate the filter list changes using a separate function.
2396 	 */
2397 	if (is_filter_list_change)
2398 		return (mld_v2_enqueue_filter_change(ifq, inm));
2399 
2400 	if (type == MLD_DO_NOTHING) {
2401 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2402 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2403 		    if_name(inm->in6m_ifp));
2404 		return (0);
2405 	}
2406 
2407 	/*
2408 	 * If any sources are present, we must be able to fit at least
2409 	 * one in the trailing space of the tail packet's mbuf,
2410 	 * ideally more.
2411 	 */
2412 	minrec0len = sizeof(struct mldv2_record);
2413 	if (record_has_sources)
2414 		minrec0len += sizeof(struct in6_addr);
2415 
2416 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2417 	    mld_rec_type_to_str(type),
2418 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2419 	    if_name(inm->in6m_ifp));
2420 
2421 	/*
2422 	 * Check if we have a packet in the tail of the queue for this
2423 	 * group into which the first group record for this group will fit.
2424 	 * Otherwise allocate a new packet.
2425 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2426 	 * Note: Group records for G/GSR query responses MUST be sent
2427 	 * in their own packet.
2428 	 */
2429 	m0 = ifq->ifq_tail;
2430 	if (!is_group_query &&
2431 	    m0 != NULL &&
2432 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2433 	    (m0->m_pkthdr.len + minrec0len) <
2434 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2435 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2436 			    sizeof(struct mldv2_record)) /
2437 			    sizeof(struct in6_addr);
2438 		m = m0;
2439 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2440 	} else {
2441 		if (_IF_QFULL(ifq)) {
2442 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2443 			return (-ENOMEM);
2444 		}
2445 		m = NULL;
2446 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2447 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2448 		if (!is_state_change && !is_group_query)
2449 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2450 		if (m == NULL)
2451 			m = m_gethdr(M_NOWAIT, MT_DATA);
2452 		if (m == NULL)
2453 			return (-ENOMEM);
2454 
2455 		mld_save_context(m, ifp);
2456 
2457 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2458 	}
2459 
2460 	/*
2461 	 * Append group record.
2462 	 * If we have sources, we don't know how many yet.
2463 	 */
2464 	mr.mr_type = type;
2465 	mr.mr_datalen = 0;
2466 	mr.mr_numsrc = 0;
2467 	mr.mr_addr = inm->in6m_addr;
2468 	in6_clearscope(&mr.mr_addr);
2469 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2470 		if (m != m0)
2471 			m_freem(m);
2472 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2473 		return (-ENOMEM);
2474 	}
2475 	nbytes += sizeof(struct mldv2_record);
2476 
2477 	/*
2478 	 * Append as many sources as will fit in the first packet.
2479 	 * If we are appending to a new packet, the chain allocation
2480 	 * may potentially use clusters; use m_getptr() in this case.
2481 	 * If we are appending to an existing packet, we need to obtain
2482 	 * a pointer to the group record after m_append(), in case a new
2483 	 * mbuf was allocated.
2484 	 *
2485 	 * Only append sources which are in-mode at t1. If we are
2486 	 * transitioning to MCAST_UNDEFINED state on the group, and
2487 	 * use_block_allow is zero, do not include source entries.
2488 	 * Otherwise, we need to include this source in the report.
2489 	 *
2490 	 * Only report recorded sources in our filter set when responding
2491 	 * to a group-source query.
2492 	 */
2493 	if (record_has_sources) {
2494 		if (m == m0) {
2495 			md = m_last(m);
2496 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2497 			    md->m_len - nbytes);
2498 		} else {
2499 			md = m_getptr(m, 0, &off);
2500 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2501 			    off);
2502 		}
2503 		msrcs = 0;
2504 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2505 		    nims) {
2506 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2507 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2508 			now = im6s_get_mode(inm, ims, 1);
2509 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2510 			if ((now != mode) ||
2511 			    (now == mode &&
2512 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2513 				CTR1(KTR_MLD, "%s: skip node", __func__);
2514 				continue;
2515 			}
2516 			if (is_source_query && ims->im6s_stp == 0) {
2517 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2518 				    __func__);
2519 				continue;
2520 			}
2521 			CTR1(KTR_MLD, "%s: append node", __func__);
2522 			if (!m_append(m, sizeof(struct in6_addr),
2523 			    (void *)&ims->im6s_addr)) {
2524 				if (m != m0)
2525 					m_freem(m);
2526 				CTR1(KTR_MLD, "%s: m_append() failed.",
2527 				    __func__);
2528 				return (-ENOMEM);
2529 			}
2530 			nbytes += sizeof(struct in6_addr);
2531 			++msrcs;
2532 			if (msrcs == m0srcs)
2533 				break;
2534 		}
2535 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2536 		    msrcs);
2537 		pmr->mr_numsrc = htons(msrcs);
2538 		nbytes += (msrcs * sizeof(struct in6_addr));
2539 	}
2540 
2541 	if (is_source_query && msrcs == 0) {
2542 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2543 		if (m != m0)
2544 			m_freem(m);
2545 		return (0);
2546 	}
2547 
2548 	/*
2549 	 * We are good to go with first packet.
2550 	 */
2551 	if (m != m0) {
2552 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2553 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2554 		_IF_ENQUEUE(ifq, m);
2555 	} else
2556 		m->m_pkthdr.PH_vt.vt_nrecs++;
2557 
2558 	/*
2559 	 * No further work needed if no source list in packet(s).
2560 	 */
2561 	if (!record_has_sources)
2562 		return (nbytes);
2563 
2564 	/*
2565 	 * Whilst sources remain to be announced, we need to allocate
2566 	 * a new packet and fill out as many sources as will fit.
2567 	 * Always try for a cluster first.
2568 	 */
2569 	while (nims != NULL) {
2570 		if (_IF_QFULL(ifq)) {
2571 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2572 			return (-ENOMEM);
2573 		}
2574 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2575 		if (m == NULL)
2576 			m = m_gethdr(M_NOWAIT, MT_DATA);
2577 		if (m == NULL)
2578 			return (-ENOMEM);
2579 		mld_save_context(m, ifp);
2580 		md = m_getptr(m, 0, &off);
2581 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2582 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2583 
2584 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2585 			if (m != m0)
2586 				m_freem(m);
2587 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2588 			return (-ENOMEM);
2589 		}
2590 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2591 		nbytes += sizeof(struct mldv2_record);
2592 
2593 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2594 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2595 
2596 		msrcs = 0;
2597 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2598 			CTR2(KTR_MLD, "%s: visit node %s",
2599 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2600 			now = im6s_get_mode(inm, ims, 1);
2601 			if ((now != mode) ||
2602 			    (now == mode &&
2603 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2604 				CTR1(KTR_MLD, "%s: skip node", __func__);
2605 				continue;
2606 			}
2607 			if (is_source_query && ims->im6s_stp == 0) {
2608 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2609 				    __func__);
2610 				continue;
2611 			}
2612 			CTR1(KTR_MLD, "%s: append node", __func__);
2613 			if (!m_append(m, sizeof(struct in6_addr),
2614 			    (void *)&ims->im6s_addr)) {
2615 				if (m != m0)
2616 					m_freem(m);
2617 				CTR1(KTR_MLD, "%s: m_append() failed.",
2618 				    __func__);
2619 				return (-ENOMEM);
2620 			}
2621 			++msrcs;
2622 			if (msrcs == m0srcs)
2623 				break;
2624 		}
2625 		pmr->mr_numsrc = htons(msrcs);
2626 		nbytes += (msrcs * sizeof(struct in6_addr));
2627 
2628 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2629 		_IF_ENQUEUE(ifq, m);
2630 	}
2631 
2632 	return (nbytes);
2633 }
2634 
2635 /*
2636  * Type used to mark record pass completion.
2637  * We exploit the fact we can cast to this easily from the
2638  * current filter modes on each ip_msource node.
2639  */
2640 typedef enum {
2641 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2642 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2643 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2644 	REC_FULL = REC_ALLOW | REC_BLOCK
2645 } rectype_t;
2646 
2647 /*
2648  * Enqueue an MLDv2 filter list change to the given output queue.
2649  *
2650  * Source list filter state is held in an RB-tree. When the filter list
2651  * for a group is changed without changing its mode, we need to compute
2652  * the deltas between T0 and T1 for each source in the filter set,
2653  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2654  *
2655  * As we may potentially queue two record types, and the entire R-B tree
2656  * needs to be walked at once, we break this out into its own function
2657  * so we can generate a tightly packed queue of packets.
2658  *
2659  * XXX This could be written to only use one tree walk, although that makes
2660  * serializing into the mbuf chains a bit harder. For now we do two walks
2661  * which makes things easier on us, and it may or may not be harder on
2662  * the L2 cache.
2663  *
2664  * If successful the size of all data appended to the queue is returned,
2665  * otherwise an error code less than zero is returned, or zero if
2666  * no record(s) were appended.
2667  */
2668 static int
2669 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
2670 {
2671 	static const int MINRECLEN =
2672 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2673 	struct ifnet		*ifp;
2674 	struct mldv2_record	 mr;
2675 	struct mldv2_record	*pmr;
2676 	struct ip6_msource	*ims, *nims;
2677 	struct mbuf		*m, *m0, *md;
2678 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2679 	int			 nallow, nblock;
2680 	uint8_t			 mode, now, then;
2681 	rectype_t		 crt, drt, nrt;
2682 #ifdef KTR
2683 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2684 #endif
2685 
2686 	IN6_MULTI_LOCK_ASSERT();
2687 
2688 	if (inm->in6m_nsrc == 0 ||
2689 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2690 		return (0);
2691 
2692 	ifp = inm->in6m_ifp;			/* interface */
2693 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2694 	crt = REC_NONE;	/* current group record type */
2695 	drt = REC_NONE;	/* mask of completed group record types */
2696 	nrt = REC_NONE;	/* record type for current node */
2697 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2698 	npbytes = 0;	/* # of bytes appended this packet */
2699 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2700 	rsrcs = 0;	/* # sources encoded in current record */
2701 	schanged = 0;	/* # nodes encoded in overall filter change */
2702 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2703 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2704 	nims = NULL;	/* next tree node pointer */
2705 
2706 	/*
2707 	 * For each possible filter record mode.
2708 	 * The first kind of source we encounter tells us which
2709 	 * is the first kind of record we start appending.
2710 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2711 	 * as the inverse of the group's filter mode.
2712 	 */
2713 	while (drt != REC_FULL) {
2714 		do {
2715 			m0 = ifq->ifq_tail;
2716 			if (m0 != NULL &&
2717 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2718 			     MLD_V2_REPORT_MAXRECS) &&
2719 			    (m0->m_pkthdr.len + MINRECLEN) <
2720 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2721 				m = m0;
2722 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2723 					    sizeof(struct mldv2_record)) /
2724 					    sizeof(struct in6_addr);
2725 				CTR1(KTR_MLD,
2726 				    "%s: use previous packet", __func__);
2727 			} else {
2728 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2729 				if (m == NULL)
2730 					m = m_gethdr(M_NOWAIT, MT_DATA);
2731 				if (m == NULL) {
2732 					CTR1(KTR_MLD,
2733 					    "%s: m_get*() failed", __func__);
2734 					return (-ENOMEM);
2735 				}
2736 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2737 				mld_save_context(m, ifp);
2738 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2739 				    sizeof(struct mldv2_record)) /
2740 				    sizeof(struct in6_addr);
2741 				npbytes = 0;
2742 				CTR1(KTR_MLD,
2743 				    "%s: allocated new packet", __func__);
2744 			}
2745 			/*
2746 			 * Append the MLD group record header to the
2747 			 * current packet's data area.
2748 			 * Recalculate pointer to free space for next
2749 			 * group record, in case m_append() allocated
2750 			 * a new mbuf or cluster.
2751 			 */
2752 			memset(&mr, 0, sizeof(mr));
2753 			mr.mr_addr = inm->in6m_addr;
2754 			in6_clearscope(&mr.mr_addr);
2755 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2756 				if (m != m0)
2757 					m_freem(m);
2758 				CTR1(KTR_MLD,
2759 				    "%s: m_append() failed", __func__);
2760 				return (-ENOMEM);
2761 			}
2762 			npbytes += sizeof(struct mldv2_record);
2763 			if (m != m0) {
2764 				/* new packet; offset in chain */
2765 				md = m_getptr(m, npbytes -
2766 				    sizeof(struct mldv2_record), &off);
2767 				pmr = (struct mldv2_record *)(mtod(md,
2768 				    uint8_t *) + off);
2769 			} else {
2770 				/* current packet; offset from last append */
2771 				md = m_last(m);
2772 				pmr = (struct mldv2_record *)(mtod(md,
2773 				    uint8_t *) + md->m_len -
2774 				    sizeof(struct mldv2_record));
2775 			}
2776 			/*
2777 			 * Begin walking the tree for this record type
2778 			 * pass, or continue from where we left off
2779 			 * previously if we had to allocate a new packet.
2780 			 * Only report deltas in-mode at t1.
2781 			 * We need not report included sources as allowed
2782 			 * if we are in inclusive mode on the group,
2783 			 * however the converse is not true.
2784 			 */
2785 			rsrcs = 0;
2786 			if (nims == NULL) {
2787 				nims = RB_MIN(ip6_msource_tree,
2788 				    &inm->in6m_srcs);
2789 			}
2790 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2791 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2792 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2793 				now = im6s_get_mode(inm, ims, 1);
2794 				then = im6s_get_mode(inm, ims, 0);
2795 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2796 				    __func__, then, now);
2797 				if (now == then) {
2798 					CTR1(KTR_MLD,
2799 					    "%s: skip unchanged", __func__);
2800 					continue;
2801 				}
2802 				if (mode == MCAST_EXCLUDE &&
2803 				    now == MCAST_INCLUDE) {
2804 					CTR1(KTR_MLD,
2805 					    "%s: skip IN src on EX group",
2806 					    __func__);
2807 					continue;
2808 				}
2809 				nrt = (rectype_t)now;
2810 				if (nrt == REC_NONE)
2811 					nrt = (rectype_t)(~mode & REC_FULL);
2812 				if (schanged++ == 0) {
2813 					crt = nrt;
2814 				} else if (crt != nrt)
2815 					continue;
2816 				if (!m_append(m, sizeof(struct in6_addr),
2817 				    (void *)&ims->im6s_addr)) {
2818 					if (m != m0)
2819 						m_freem(m);
2820 					CTR1(KTR_MLD,
2821 					    "%s: m_append() failed", __func__);
2822 					return (-ENOMEM);
2823 				}
2824 				nallow += !!(crt == REC_ALLOW);
2825 				nblock += !!(crt == REC_BLOCK);
2826 				if (++rsrcs == m0srcs)
2827 					break;
2828 			}
2829 			/*
2830 			 * If we did not append any tree nodes on this
2831 			 * pass, back out of allocations.
2832 			 */
2833 			if (rsrcs == 0) {
2834 				npbytes -= sizeof(struct mldv2_record);
2835 				if (m != m0) {
2836 					CTR1(KTR_MLD,
2837 					    "%s: m_free(m)", __func__);
2838 					m_freem(m);
2839 				} else {
2840 					CTR1(KTR_MLD,
2841 					    "%s: m_adj(m, -mr)", __func__);
2842 					m_adj(m, -((int)sizeof(
2843 					    struct mldv2_record)));
2844 				}
2845 				continue;
2846 			}
2847 			npbytes += (rsrcs * sizeof(struct in6_addr));
2848 			if (crt == REC_ALLOW)
2849 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2850 			else if (crt == REC_BLOCK)
2851 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2852 			pmr->mr_numsrc = htons(rsrcs);
2853 			/*
2854 			 * Count the new group record, and enqueue this
2855 			 * packet if it wasn't already queued.
2856 			 */
2857 			m->m_pkthdr.PH_vt.vt_nrecs++;
2858 			if (m != m0)
2859 				_IF_ENQUEUE(ifq, m);
2860 			nbytes += npbytes;
2861 		} while (nims != NULL);
2862 		drt |= crt;
2863 		crt = (~crt & REC_FULL);
2864 	}
2865 
2866 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2867 	    nallow, nblock);
2868 
2869 	return (nbytes);
2870 }
2871 
2872 static int
2873 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
2874 {
2875 	struct ifqueue	*gq;
2876 	struct mbuf	*m;		/* pending state-change */
2877 	struct mbuf	*m0;		/* copy of pending state-change */
2878 	struct mbuf	*mt;		/* last state-change in packet */
2879 	int		 docopy, domerge;
2880 	u_int		 recslen;
2881 
2882 	docopy = 0;
2883 	domerge = 0;
2884 	recslen = 0;
2885 
2886 	IN6_MULTI_LOCK_ASSERT();
2887 	MLD_LOCK_ASSERT();
2888 
2889 	/*
2890 	 * If there are further pending retransmissions, make a writable
2891 	 * copy of each queued state-change message before merging.
2892 	 */
2893 	if (inm->in6m_scrv > 0)
2894 		docopy = 1;
2895 
2896 	gq = &inm->in6m_scq;
2897 #ifdef KTR
2898 	if (gq->ifq_head == NULL) {
2899 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2900 		    __func__, inm);
2901 	}
2902 #endif
2903 
2904 	m = gq->ifq_head;
2905 	while (m != NULL) {
2906 		/*
2907 		 * Only merge the report into the current packet if
2908 		 * there is sufficient space to do so; an MLDv2 report
2909 		 * packet may only contain 65,535 group records.
2910 		 * Always use a simple mbuf chain concatentation to do this,
2911 		 * as large state changes for single groups may have
2912 		 * allocated clusters.
2913 		 */
2914 		domerge = 0;
2915 		mt = ifscq->ifq_tail;
2916 		if (mt != NULL) {
2917 			recslen = m_length(m, NULL);
2918 
2919 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2920 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2921 			    MLD_V2_REPORT_MAXRECS) &&
2922 			    (mt->m_pkthdr.len + recslen <=
2923 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2924 				domerge = 1;
2925 		}
2926 
2927 		if (!domerge && _IF_QFULL(gq)) {
2928 			CTR2(KTR_MLD,
2929 			    "%s: outbound queue full, skipping whole packet %p",
2930 			    __func__, m);
2931 			mt = m->m_nextpkt;
2932 			if (!docopy)
2933 				m_freem(m);
2934 			m = mt;
2935 			continue;
2936 		}
2937 
2938 		if (!docopy) {
2939 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2940 			_IF_DEQUEUE(gq, m0);
2941 			m = m0->m_nextpkt;
2942 		} else {
2943 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2944 			m0 = m_dup(m, M_NOWAIT);
2945 			if (m0 == NULL)
2946 				return (ENOMEM);
2947 			m0->m_nextpkt = NULL;
2948 			m = m->m_nextpkt;
2949 		}
2950 
2951 		if (!domerge) {
2952 			CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)",
2953 			    __func__, m0, ifscq);
2954 			_IF_ENQUEUE(ifscq, m0);
2955 		} else {
2956 			struct mbuf *mtl;	/* last mbuf of packet mt */
2957 
2958 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2959 			    __func__, m0, mt);
2960 
2961 			mtl = m_last(mt);
2962 			m0->m_flags &= ~M_PKTHDR;
2963 			mt->m_pkthdr.len += recslen;
2964 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2965 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2966 
2967 			mtl->m_next = m0;
2968 		}
2969 	}
2970 
2971 	return (0);
2972 }
2973 
2974 /*
2975  * Respond to a pending MLDv2 General Query.
2976  */
2977 static void
2978 mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
2979 {
2980 	struct ifmultiaddr	*ifma;
2981 	struct ifnet		*ifp;
2982 	struct in6_multi	*inm;
2983 	int			 retval;
2984 
2985 	IN6_MULTI_LOCK_ASSERT();
2986 	MLD_LOCK_ASSERT();
2987 
2988 	KASSERT(mli->mli_version == MLD_VERSION_2,
2989 	    ("%s: called when version %d", __func__, mli->mli_version));
2990 
2991 	ifp = mli->mli_ifp;
2992 
2993 	IF_ADDR_RLOCK(ifp);
2994 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2995 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2996 		    ifma->ifma_protospec == NULL)
2997 			continue;
2998 
2999 		inm = (struct in6_multi *)ifma->ifma_protospec;
3000 		KASSERT(ifp == inm->in6m_ifp,
3001 		    ("%s: inconsistent ifp", __func__));
3002 
3003 		switch (inm->in6m_state) {
3004 		case MLD_NOT_MEMBER:
3005 		case MLD_SILENT_MEMBER:
3006 			break;
3007 		case MLD_REPORTING_MEMBER:
3008 		case MLD_IDLE_MEMBER:
3009 		case MLD_LAZY_MEMBER:
3010 		case MLD_SLEEPING_MEMBER:
3011 		case MLD_AWAKENING_MEMBER:
3012 			inm->in6m_state = MLD_REPORTING_MEMBER;
3013 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3014 			    inm, 0, 0, 0, 0);
3015 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3016 			    __func__, retval);
3017 			break;
3018 		case MLD_G_QUERY_PENDING_MEMBER:
3019 		case MLD_SG_QUERY_PENDING_MEMBER:
3020 		case MLD_LEAVING_MEMBER:
3021 			break;
3022 		}
3023 	}
3024 	IF_ADDR_RUNLOCK(ifp);
3025 
3026 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3027 
3028 	/*
3029 	 * Slew transmission of bursts over 500ms intervals.
3030 	 */
3031 	if (mli->mli_gq.ifq_head != NULL) {
3032 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3033 		    MLD_RESPONSE_BURST_INTERVAL);
3034 		V_interface_timers_running6 = 1;
3035 	}
3036 }
3037 
3038 /*
3039  * Transmit the next pending message in the output queue.
3040  *
3041  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3042  * MRT: Nothing needs to be done, as MLD traffic is always local to
3043  * a link and uses a link-scope multicast address.
3044  */
3045 static void
3046 mld_dispatch_packet(struct mbuf *m)
3047 {
3048 	struct ip6_moptions	 im6o;
3049 	struct ifnet		*ifp;
3050 	struct ifnet		*oifp;
3051 	struct mbuf		*m0;
3052 	struct mbuf		*md;
3053 	struct ip6_hdr		*ip6;
3054 	struct mld_hdr		*mld;
3055 	int			 error;
3056 	int			 off;
3057 	int			 type;
3058 	uint32_t		 ifindex;
3059 
3060 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3061 
3062 	/*
3063 	 * Set VNET image pointer from enqueued mbuf chain
3064 	 * before doing anything else. Whilst we use interface
3065 	 * indexes to guard against interface detach, they are
3066 	 * unique to each VIMAGE and must be retrieved.
3067 	 */
3068 	ifindex = mld_restore_context(m);
3069 
3070 	/*
3071 	 * Check if the ifnet still exists. This limits the scope of
3072 	 * any race in the absence of a global ifp lock for low cost
3073 	 * (an array lookup).
3074 	 */
3075 	ifp = ifnet_byindex(ifindex);
3076 	if (ifp == NULL) {
3077 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3078 		    __func__, m, ifindex);
3079 		m_freem(m);
3080 		IP6STAT_INC(ip6s_noroute);
3081 		goto out;
3082 	}
3083 
3084 	im6o.im6o_multicast_hlim  = 1;
3085 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3086 	im6o.im6o_multicast_ifp = ifp;
3087 
3088 	if (m->m_flags & M_MLDV1) {
3089 		m0 = m;
3090 	} else {
3091 		m0 = mld_v2_encap_report(ifp, m);
3092 		if (m0 == NULL) {
3093 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3094 			IP6STAT_INC(ip6s_odropped);
3095 			goto out;
3096 		}
3097 	}
3098 
3099 	mld_scrub_context(m0);
3100 	m_clrprotoflags(m);
3101 	m0->m_pkthdr.rcvif = V_loif;
3102 
3103 	ip6 = mtod(m0, struct ip6_hdr *);
3104 #if 0
3105 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3106 #else
3107 	/*
3108 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3109 	 * occur if we called in6_setscope() at transmission.
3110 	 * See comments at top of file.
3111 	 */
3112 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3113 #endif
3114 
3115 	/*
3116 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3117 	 * so we can bump the stats.
3118 	 */
3119 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3120 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3121 	type = mld->mld_type;
3122 
3123 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3124 	    &oifp, NULL);
3125 	if (error) {
3126 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3127 		goto out;
3128 	}
3129 	ICMP6STAT_INC(icp6s_outhist[type]);
3130 	if (oifp != NULL) {
3131 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3132 		switch (type) {
3133 		case MLD_LISTENER_REPORT:
3134 		case MLDV2_LISTENER_REPORT:
3135 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3136 			break;
3137 		case MLD_LISTENER_DONE:
3138 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3139 			break;
3140 		}
3141 	}
3142 out:
3143 	return;
3144 }
3145 
3146 /*
3147  * Encapsulate an MLDv2 report.
3148  *
3149  * KAME IPv6 requires that hop-by-hop options be passed separately,
3150  * and that the IPv6 header be prepended in a separate mbuf.
3151  *
3152  * Returns a pointer to the new mbuf chain head, or NULL if the
3153  * allocation failed.
3154  */
3155 static struct mbuf *
3156 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3157 {
3158 	struct mbuf		*mh;
3159 	struct mldv2_report	*mld;
3160 	struct ip6_hdr		*ip6;
3161 	struct in6_ifaddr	*ia;
3162 	int			 mldreclen;
3163 
3164 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3165 	KASSERT((m->m_flags & M_PKTHDR),
3166 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3167 
3168 	/*
3169 	 * RFC3590: OK to send as :: or tentative during DAD.
3170 	 */
3171 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3172 	if (ia == NULL)
3173 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3174 
3175 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3176 	if (mh == NULL) {
3177 		if (ia != NULL)
3178 			ifa_free(&ia->ia_ifa);
3179 		m_freem(m);
3180 		return (NULL);
3181 	}
3182 	MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3183 
3184 	mldreclen = m_length(m, NULL);
3185 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3186 
3187 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3188 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3189 	    sizeof(struct mldv2_report) + mldreclen;
3190 
3191 	ip6 = mtod(mh, struct ip6_hdr *);
3192 	ip6->ip6_flow = 0;
3193 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3194 	ip6->ip6_vfc |= IPV6_VERSION;
3195 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3196 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3197 	if (ia != NULL)
3198 		ifa_free(&ia->ia_ifa);
3199 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3200 	/* scope ID will be set in netisr */
3201 
3202 	mld = (struct mldv2_report *)(ip6 + 1);
3203 	mld->mld_type = MLDV2_LISTENER_REPORT;
3204 	mld->mld_code = 0;
3205 	mld->mld_cksum = 0;
3206 	mld->mld_v2_reserved = 0;
3207 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3208 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3209 
3210 	mh->m_next = m;
3211 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3212 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3213 	return (mh);
3214 }
3215 
3216 #ifdef KTR
3217 static char *
3218 mld_rec_type_to_str(const int type)
3219 {
3220 
3221 	switch (type) {
3222 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3223 			return "TO_EX";
3224 			break;
3225 		case MLD_CHANGE_TO_INCLUDE_MODE:
3226 			return "TO_IN";
3227 			break;
3228 		case MLD_MODE_IS_EXCLUDE:
3229 			return "MODE_EX";
3230 			break;
3231 		case MLD_MODE_IS_INCLUDE:
3232 			return "MODE_IN";
3233 			break;
3234 		case MLD_ALLOW_NEW_SOURCES:
3235 			return "ALLOW_NEW";
3236 			break;
3237 		case MLD_BLOCK_OLD_SOURCES:
3238 			return "BLOCK_OLD";
3239 			break;
3240 		default:
3241 			break;
3242 	}
3243 	return "unknown";
3244 }
3245 #endif
3246 
3247 static void
3248 mld_init(void *unused __unused)
3249 {
3250 
3251 	CTR1(KTR_MLD, "%s: initializing", __func__);
3252 	MLD_LOCK_INIT();
3253 
3254 	ip6_initpktopts(&mld_po);
3255 	mld_po.ip6po_hlim = 1;
3256 	mld_po.ip6po_hbh = &mld_ra.hbh;
3257 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3258 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3259 }
3260 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
3261 
3262 static void
3263 mld_uninit(void *unused __unused)
3264 {
3265 
3266 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3267 	MLD_LOCK_DESTROY();
3268 }
3269 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
3270 
3271 static void
3272 vnet_mld_init(const void *unused __unused)
3273 {
3274 
3275 	CTR1(KTR_MLD, "%s: initializing", __func__);
3276 
3277 	LIST_INIT(&V_mli_head);
3278 }
3279 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
3280     NULL);
3281 
3282 static void
3283 vnet_mld_uninit(const void *unused __unused)
3284 {
3285 
3286 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3287 
3288 	KASSERT(LIST_EMPTY(&V_mli_head),
3289 	    ("%s: mli list not empty; ifnets not detached?", __func__));
3290 }
3291 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
3292     NULL);
3293 
3294 static int
3295 mld_modevent(module_t mod, int type, void *unused __unused)
3296 {
3297 
3298     switch (type) {
3299     case MOD_LOAD:
3300     case MOD_UNLOAD:
3301 	break;
3302     default:
3303 	return (EOPNOTSUPP);
3304     }
3305     return (0);
3306 }
3307 
3308 static moduledata_t mld_mod = {
3309     "mld",
3310     mld_modevent,
3311     0
3312 };
3313 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3314