xref: /freebsd/sys/netinet6/mld6.c (revision 86c9d9918f1db7cdd968b60f8902466887bcd9e9)
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_ifsoftc *
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 mbufq *, int);
110 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
111 static void	mld_fasttimo_vnet(void);
112 static int	mld_handle_state_change(struct in6_multi *,
113 		    struct mld_ifsoftc *);
114 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
115 		    const int);
116 #ifdef KTR
117 static char *	mld_rec_type_to_str(const int);
118 #endif
119 static void	mld_set_version(struct mld_ifsoftc *, const int);
120 static void	mld_slowtimo_vnet(void);
121 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
122 		    /*const*/ struct mld_hdr *);
123 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static void	mld_v1_process_group_timer(struct mld_ifsoftc *,
126 		    struct in6_multi *);
127 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
128 static int	mld_v1_transmit_report(struct in6_multi *, const int);
129 static void	mld_v1_update_group(struct in6_multi *, const int);
130 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
131 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
132 static struct mbuf *
133 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
134 static int	mld_v2_enqueue_filter_change(struct mbufq *,
135 		    struct in6_multi *);
136 static int	mld_v2_enqueue_group_record(struct mbufq *,
137 		    struct in6_multi *, const int, const int, const int,
138 		    const int);
139 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
140 		    struct mbuf *, const int, const int);
141 static int	mld_v2_merge_state_changes(struct in6_multi *,
142 		    struct mbufq *);
143 static void	mld_v2_process_group_timers(struct mld_ifsoftc *,
144 		    struct mbufq *, struct mbufq *,
145 		    struct in6_multi *, const int);
146 static int	mld_v2_process_group_query(struct in6_multi *,
147 		    struct mld_ifsoftc *mli, int, struct mbuf *, 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_ifsoftc), 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: cuvnet %p m ptr %p",
304 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
305 #endif
306 	return (m->m_pkthdr.flowid);
307 }
308 
309 /*
310  * Retrieve or set threshold between group-source queries in seconds.
311  *
312  * VIMAGE: Assume curvnet set by caller.
313  * SMPng: NOTE: Serialized by MLD lock.
314  */
315 static int
316 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
317 {
318 	int error;
319 	int i;
320 
321 	error = sysctl_wire_old_buffer(req, sizeof(int));
322 	if (error)
323 		return (error);
324 
325 	MLD_LOCK();
326 
327 	i = V_mld_gsrdelay.tv_sec;
328 
329 	error = sysctl_handle_int(oidp, &i, 0, req);
330 	if (error || !req->newptr)
331 		goto out_locked;
332 
333 	if (i < -1 || i >= 60) {
334 		error = EINVAL;
335 		goto out_locked;
336 	}
337 
338 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
339 	     V_mld_gsrdelay.tv_sec, i);
340 	V_mld_gsrdelay.tv_sec = i;
341 
342 out_locked:
343 	MLD_UNLOCK();
344 	return (error);
345 }
346 
347 /*
348  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
349  * For use by ifmcstat(8).
350  *
351  * SMPng: NOTE: Does an unlocked ifindex space read.
352  * VIMAGE: Assume curvnet set by caller. The node handler itself
353  * is not directly virtualized.
354  */
355 static int
356 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
357 {
358 	int			*name;
359 	int			 error;
360 	u_int			 namelen;
361 	struct ifnet		*ifp;
362 	struct mld_ifsoftc	*mli;
363 
364 	name = (int *)arg1;
365 	namelen = arg2;
366 
367 	if (req->newptr != NULL)
368 		return (EPERM);
369 
370 	if (namelen != 1)
371 		return (EINVAL);
372 
373 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
374 	if (error)
375 		return (error);
376 
377 	IN6_MULTI_LOCK();
378 	MLD_LOCK();
379 
380 	if (name[0] <= 0 || name[0] > V_if_index) {
381 		error = ENOENT;
382 		goto out_locked;
383 	}
384 
385 	error = ENOENT;
386 
387 	ifp = ifnet_byindex(name[0]);
388 	if (ifp == NULL)
389 		goto out_locked;
390 
391 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
392 		if (ifp == mli->mli_ifp) {
393 			struct mld_ifinfo info;
394 
395 			info.mli_version = mli->mli_version;
396 			info.mli_v1_timer = mli->mli_v1_timer;
397 			info.mli_v2_timer = mli->mli_v2_timer;
398 			info.mli_flags = mli->mli_flags;
399 			info.mli_rv = mli->mli_rv;
400 			info.mli_qi = mli->mli_qi;
401 			info.mli_qri = mli->mli_qri;
402 			info.mli_uri = mli->mli_uri;
403 			error = SYSCTL_OUT(req, &info, sizeof(info));
404 			break;
405 		}
406 	}
407 
408 out_locked:
409 	MLD_UNLOCK();
410 	IN6_MULTI_UNLOCK();
411 	return (error);
412 }
413 
414 /*
415  * Dispatch an entire queue of pending packet chains.
416  * VIMAGE: Assumes the vnet pointer has been set.
417  */
418 static void
419 mld_dispatch_queue(struct mbufq *mq, int limit)
420 {
421 	struct mbuf *m;
422 
423 	while ((m = mbufq_dequeue(mq)) != NULL) {
424 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
425 		mld_dispatch_packet(m);
426 		if (--limit == 0)
427 			break;
428 	}
429 }
430 
431 /*
432  * Filter outgoing MLD report state by group.
433  *
434  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
435  * and node-local addresses. However, kernel and socket consumers
436  * always embed the KAME scope ID in the address provided, so strip it
437  * when performing comparison.
438  * Note: This is not the same as the *multicast* scope.
439  *
440  * Return zero if the given group is one for which MLD reports
441  * should be suppressed, or non-zero if reports should be issued.
442  */
443 static __inline int
444 mld_is_addr_reported(const struct in6_addr *addr)
445 {
446 
447 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
448 
449 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
450 		return (0);
451 
452 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
453 		struct in6_addr tmp = *addr;
454 		in6_clearscope(&tmp);
455 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
456 			return (0);
457 	}
458 
459 	return (1);
460 }
461 
462 /*
463  * Attach MLD when PF_INET6 is attached to an interface.
464  *
465  * SMPng: Normally called with IF_AFDATA_LOCK held.
466  */
467 struct mld_ifsoftc *
468 mld_domifattach(struct ifnet *ifp)
469 {
470 	struct mld_ifsoftc *mli;
471 
472 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
473 	    __func__, ifp, if_name(ifp));
474 
475 	MLD_LOCK();
476 
477 	mli = mli_alloc_locked(ifp);
478 	if (!(ifp->if_flags & IFF_MULTICAST))
479 		mli->mli_flags |= MLIF_SILENT;
480 	if (mld_use_allow)
481 		mli->mli_flags |= MLIF_USEALLOW;
482 
483 	MLD_UNLOCK();
484 
485 	return (mli);
486 }
487 
488 /*
489  * VIMAGE: assume curvnet set by caller.
490  */
491 static struct mld_ifsoftc *
492 mli_alloc_locked(/*const*/ struct ifnet *ifp)
493 {
494 	struct mld_ifsoftc *mli;
495 
496 	MLD_LOCK_ASSERT();
497 
498 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
499 	if (mli == NULL)
500 		goto out;
501 
502 	mli->mli_ifp = ifp;
503 	mli->mli_version = MLD_VERSION_2;
504 	mli->mli_flags = 0;
505 	mli->mli_rv = MLD_RV_INIT;
506 	mli->mli_qi = MLD_QI_INIT;
507 	mli->mli_qri = MLD_QRI_INIT;
508 	mli->mli_uri = MLD_URI_INIT;
509 	SLIST_INIT(&mli->mli_relinmhead);
510 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
511 
512 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
513 
514 	CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
515 	     ifp, if_name(ifp));
516 
517 out:
518 	return (mli);
519 }
520 
521 /*
522  * Hook for ifdetach.
523  *
524  * NOTE: Some finalization tasks need to run before the protocol domain
525  * is detached, but also before the link layer does its cleanup.
526  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
527  *
528  * SMPng: Caller must hold IN6_MULTI_LOCK().
529  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
530  * XXX This routine is also bitten by unlocked ifma_protospec access.
531  */
532 void
533 mld_ifdetach(struct ifnet *ifp)
534 {
535 	struct mld_ifsoftc	*mli;
536 	struct ifmultiaddr	*ifma;
537 	struct in6_multi	*inm, *tinm;
538 
539 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
540 	    if_name(ifp));
541 
542 	IN6_MULTI_LOCK_ASSERT();
543 	MLD_LOCK();
544 
545 	mli = MLD_IFINFO(ifp);
546 	if (mli->mli_version == MLD_VERSION_2) {
547 		IF_ADDR_RLOCK(ifp);
548 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
549 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
550 			    ifma->ifma_protospec == NULL)
551 				continue;
552 			inm = (struct in6_multi *)ifma->ifma_protospec;
553 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
554 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
555 				    inm, in6m_nrele);
556 			}
557 			in6m_clear_recorded(inm);
558 		}
559 		IF_ADDR_RUNLOCK(ifp);
560 		SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
561 		    tinm) {
562 			SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
563 			in6m_release_locked(inm);
564 		}
565 	}
566 
567 	MLD_UNLOCK();
568 }
569 
570 /*
571  * Hook for domifdetach.
572  * Runs after link-layer cleanup; free MLD state.
573  *
574  * SMPng: Normally called with IF_AFDATA_LOCK held.
575  */
576 void
577 mld_domifdetach(struct ifnet *ifp)
578 {
579 
580 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
581 	    __func__, ifp, if_name(ifp));
582 
583 	MLD_LOCK();
584 	mli_delete_locked(ifp);
585 	MLD_UNLOCK();
586 }
587 
588 static void
589 mli_delete_locked(const struct ifnet *ifp)
590 {
591 	struct mld_ifsoftc *mli, *tmli;
592 
593 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
594 	    __func__, ifp, if_name(ifp));
595 
596 	MLD_LOCK_ASSERT();
597 
598 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
599 		if (mli->mli_ifp == ifp) {
600 			/*
601 			 * Free deferred General Query responses.
602 			 */
603 			mbufq_drain(&mli->mli_gq);
604 
605 			LIST_REMOVE(mli, mli_link);
606 
607 			KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
608 			    ("%s: there are dangling in_multi references",
609 			    __func__));
610 
611 			free(mli, M_MLD);
612 			return;
613 		}
614 	}
615 #ifdef INVARIANTS
616 	panic("%s: mld_ifsoftc not found for ifp %p\n", __func__,  ifp);
617 #endif
618 }
619 
620 /*
621  * Process a received MLDv1 general or address-specific query.
622  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
623  *
624  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
625  * mld_addr. This is OK as we own the mbuf chain.
626  */
627 static int
628 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
629     /*const*/ struct mld_hdr *mld)
630 {
631 	struct ifmultiaddr	*ifma;
632 	struct mld_ifsoftc	*mli;
633 	struct in6_multi	*inm;
634 	int			 is_general_query;
635 	uint16_t		 timer;
636 #ifdef KTR
637 	char			 ip6tbuf[INET6_ADDRSTRLEN];
638 #endif
639 
640 	is_general_query = 0;
641 
642 	if (!mld_v1enable) {
643 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
644 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
645 		    ifp, if_name(ifp));
646 		return (0);
647 	}
648 
649 	/*
650 	 * RFC3810 Section 6.2: MLD queries must originate from
651 	 * a router's link-local address.
652 	 */
653 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
654 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
655 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
656 		    ifp, if_name(ifp));
657 		return (0);
658 	}
659 
660 	/*
661 	 * Do address field validation upfront before we accept
662 	 * the query.
663 	 */
664 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
665 		/*
666 		 * MLDv1 General Query.
667 		 * If this was not sent to the all-nodes group, ignore it.
668 		 */
669 		struct in6_addr		 dst;
670 
671 		dst = ip6->ip6_dst;
672 		in6_clearscope(&dst);
673 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
674 			return (EINVAL);
675 		is_general_query = 1;
676 	} else {
677 		/*
678 		 * Embed scope ID of receiving interface in MLD query for
679 		 * lookup whilst we don't hold other locks.
680 		 */
681 		in6_setscope(&mld->mld_addr, ifp, NULL);
682 	}
683 
684 	IN6_MULTI_LOCK();
685 	MLD_LOCK();
686 
687 	/*
688 	 * Switch to MLDv1 host compatibility mode.
689 	 */
690 	mli = MLD_IFINFO(ifp);
691 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
692 	mld_set_version(mli, MLD_VERSION_1);
693 
694 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
695 	if (timer == 0)
696 		timer = 1;
697 
698 	IF_ADDR_RLOCK(ifp);
699 	if (is_general_query) {
700 		/*
701 		 * For each reporting group joined on this
702 		 * interface, kick the report timer.
703 		 */
704 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
705 		    ifp, if_name(ifp));
706 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
707 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
708 			    ifma->ifma_protospec == NULL)
709 				continue;
710 			inm = (struct in6_multi *)ifma->ifma_protospec;
711 			mld_v1_update_group(inm, timer);
712 		}
713 	} else {
714 		/*
715 		 * MLDv1 Group-Specific Query.
716 		 * If this is a group-specific MLDv1 query, we need only
717 		 * look up the single group to process it.
718 		 */
719 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
720 		if (inm != NULL) {
721 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
722 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
723 			    ifp, if_name(ifp));
724 			mld_v1_update_group(inm, timer);
725 		}
726 		/* XXX Clear embedded scope ID as userland won't expect it. */
727 		in6_clearscope(&mld->mld_addr);
728 	}
729 
730 	IF_ADDR_RUNLOCK(ifp);
731 	MLD_UNLOCK();
732 	IN6_MULTI_UNLOCK();
733 
734 	return (0);
735 }
736 
737 /*
738  * Update the report timer on a group in response to an MLDv1 query.
739  *
740  * If we are becoming the reporting member for this group, start the timer.
741  * If we already are the reporting member for this group, and timer is
742  * below the threshold, reset it.
743  *
744  * We may be updating the group for the first time since we switched
745  * to MLDv2. If we are, then we must clear any recorded source lists,
746  * and transition to REPORTING state; the group timer is overloaded
747  * for group and group-source query responses.
748  *
749  * Unlike MLDv2, the delay per group should be jittered
750  * to avoid bursts of MLDv1 reports.
751  */
752 static void
753 mld_v1_update_group(struct in6_multi *inm, const int timer)
754 {
755 #ifdef KTR
756 	char			 ip6tbuf[INET6_ADDRSTRLEN];
757 #endif
758 
759 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
760 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
761 	    if_name(inm->in6m_ifp), timer);
762 
763 	IN6_MULTI_LOCK_ASSERT();
764 
765 	switch (inm->in6m_state) {
766 	case MLD_NOT_MEMBER:
767 	case MLD_SILENT_MEMBER:
768 		break;
769 	case MLD_REPORTING_MEMBER:
770 		if (inm->in6m_timer != 0 &&
771 		    inm->in6m_timer <= timer) {
772 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
773 			    "skipping.", __func__);
774 			break;
775 		}
776 		/* FALLTHROUGH */
777 	case MLD_SG_QUERY_PENDING_MEMBER:
778 	case MLD_G_QUERY_PENDING_MEMBER:
779 	case MLD_IDLE_MEMBER:
780 	case MLD_LAZY_MEMBER:
781 	case MLD_AWAKENING_MEMBER:
782 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
783 		inm->in6m_state = MLD_REPORTING_MEMBER;
784 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
785 		V_current_state_timers_running6 = 1;
786 		break;
787 	case MLD_SLEEPING_MEMBER:
788 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
789 		inm->in6m_state = MLD_AWAKENING_MEMBER;
790 		break;
791 	case MLD_LEAVING_MEMBER:
792 		break;
793 	}
794 }
795 
796 /*
797  * Process a received MLDv2 general, group-specific or
798  * group-and-source-specific query.
799  *
800  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
801  *
802  * Return 0 if successful, otherwise an appropriate error code is returned.
803  */
804 static int
805 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
806     struct mbuf *m, const int off, const int icmp6len)
807 {
808 	struct mld_ifsoftc	*mli;
809 	struct mldv2_query	*mld;
810 	struct in6_multi	*inm;
811 	uint32_t		 maxdelay, nsrc, qqi;
812 	int			 is_general_query;
813 	uint16_t		 timer;
814 	uint8_t			 qrv;
815 #ifdef KTR
816 	char			 ip6tbuf[INET6_ADDRSTRLEN];
817 #endif
818 
819 	is_general_query = 0;
820 
821 	/*
822 	 * RFC3810 Section 6.2: MLD queries must originate from
823 	 * a router's link-local address.
824 	 */
825 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
826 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
827 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
828 		    ifp, if_name(ifp));
829 		return (0);
830 	}
831 
832 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
833 
834 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
835 
836 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
837 	if (maxdelay >= 32768) {
838 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
839 			   (MLD_MRC_EXP(maxdelay) + 3);
840 	}
841 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
842 	if (timer == 0)
843 		timer = 1;
844 
845 	qrv = MLD_QRV(mld->mld_misc);
846 	if (qrv < 2) {
847 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
848 		    qrv, MLD_RV_INIT);
849 		qrv = MLD_RV_INIT;
850 	}
851 
852 	qqi = mld->mld_qqi;
853 	if (qqi >= 128) {
854 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
855 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
856 	}
857 
858 	nsrc = ntohs(mld->mld_numsrc);
859 	if (nsrc > MLD_MAX_GS_SOURCES)
860 		return (EMSGSIZE);
861 	if (icmp6len < sizeof(struct mldv2_query) +
862 	    (nsrc * sizeof(struct in6_addr)))
863 		return (EMSGSIZE);
864 
865 	/*
866 	 * Do further input validation upfront to avoid resetting timers
867 	 * should we need to discard this query.
868 	 */
869 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
870 		/*
871 		 * A general query with a source list has undefined
872 		 * behaviour; discard it.
873 		 */
874 		if (nsrc > 0)
875 			return (EINVAL);
876 		is_general_query = 1;
877 	} else {
878 		/*
879 		 * Embed scope ID of receiving interface in MLD query for
880 		 * lookup whilst we don't hold other locks (due to KAME
881 		 * locking lameness). We own this mbuf chain just now.
882 		 */
883 		in6_setscope(&mld->mld_addr, ifp, NULL);
884 	}
885 
886 	IN6_MULTI_LOCK();
887 	MLD_LOCK();
888 
889 	mli = MLD_IFINFO(ifp);
890 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
891 
892 	/*
893 	 * Discard the v2 query if we're in Compatibility Mode.
894 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
895 	 * until the Old Version Querier Present timer expires.
896 	 */
897 	if (mli->mli_version != MLD_VERSION_2)
898 		goto out_locked;
899 
900 	mld_set_version(mli, MLD_VERSION_2);
901 	mli->mli_rv = qrv;
902 	mli->mli_qi = qqi;
903 	mli->mli_qri = maxdelay;
904 
905 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
906 	    maxdelay);
907 
908 	if (is_general_query) {
909 		/*
910 		 * MLDv2 General Query.
911 		 *
912 		 * Schedule a current-state report on this ifp for
913 		 * all groups, possibly containing source lists.
914 		 *
915 		 * If there is a pending General Query response
916 		 * scheduled earlier than the selected delay, do
917 		 * not schedule any other reports.
918 		 * Otherwise, reset the interface timer.
919 		 */
920 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
921 		    ifp, if_name(ifp));
922 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
923 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
924 			V_interface_timers_running6 = 1;
925 		}
926 	} else {
927 		/*
928 		 * MLDv2 Group-specific or Group-and-source-specific Query.
929 		 *
930 		 * Group-source-specific queries are throttled on
931 		 * a per-group basis to defeat denial-of-service attempts.
932 		 * Queries for groups we are not a member of on this
933 		 * link are simply ignored.
934 		 */
935 		IF_ADDR_RLOCK(ifp);
936 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
937 		if (inm == NULL) {
938 			IF_ADDR_RUNLOCK(ifp);
939 			goto out_locked;
940 		}
941 		if (nsrc > 0) {
942 			if (!ratecheck(&inm->in6m_lastgsrtv,
943 			    &V_mld_gsrdelay)) {
944 				CTR1(KTR_MLD, "%s: GS query throttled.",
945 				    __func__);
946 				IF_ADDR_RUNLOCK(ifp);
947 				goto out_locked;
948 			}
949 		}
950 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
951 		     ifp, if_name(ifp));
952 		/*
953 		 * If there is a pending General Query response
954 		 * scheduled sooner than the selected delay, no
955 		 * further report need be scheduled.
956 		 * Otherwise, prepare to respond to the
957 		 * group-specific or group-and-source query.
958 		 */
959 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
960 			mld_v2_process_group_query(inm, mli, timer, m, off);
961 
962 		/* XXX Clear embedded scope ID as userland won't expect it. */
963 		in6_clearscope(&mld->mld_addr);
964 		IF_ADDR_RUNLOCK(ifp);
965 	}
966 
967 out_locked:
968 	MLD_UNLOCK();
969 	IN6_MULTI_UNLOCK();
970 
971 	return (0);
972 }
973 
974 /*
975  * Process a received MLDv2 group-specific or group-and-source-specific
976  * query.
977  * Return <0 if any error occurred. Currently this is ignored.
978  */
979 static int
980 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
981     int timer, struct mbuf *m0, const int off)
982 {
983 	struct mldv2_query	*mld;
984 	int			 retval;
985 	uint16_t		 nsrc;
986 
987 	IN6_MULTI_LOCK_ASSERT();
988 	MLD_LOCK_ASSERT();
989 
990 	retval = 0;
991 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
992 
993 	switch (inm->in6m_state) {
994 	case MLD_NOT_MEMBER:
995 	case MLD_SILENT_MEMBER:
996 	case MLD_SLEEPING_MEMBER:
997 	case MLD_LAZY_MEMBER:
998 	case MLD_AWAKENING_MEMBER:
999 	case MLD_IDLE_MEMBER:
1000 	case MLD_LEAVING_MEMBER:
1001 		return (retval);
1002 		break;
1003 	case MLD_REPORTING_MEMBER:
1004 	case MLD_G_QUERY_PENDING_MEMBER:
1005 	case MLD_SG_QUERY_PENDING_MEMBER:
1006 		break;
1007 	}
1008 
1009 	nsrc = ntohs(mld->mld_numsrc);
1010 
1011 	/*
1012 	 * Deal with group-specific queries upfront.
1013 	 * If any group query is already pending, purge any recorded
1014 	 * source-list state if it exists, and schedule a query response
1015 	 * for this group-specific query.
1016 	 */
1017 	if (nsrc == 0) {
1018 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1019 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1020 			in6m_clear_recorded(inm);
1021 			timer = min(inm->in6m_timer, timer);
1022 		}
1023 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1024 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1025 		V_current_state_timers_running6 = 1;
1026 		return (retval);
1027 	}
1028 
1029 	/*
1030 	 * Deal with the case where a group-and-source-specific query has
1031 	 * been received but a group-specific query is already pending.
1032 	 */
1033 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1034 		timer = min(inm->in6m_timer, timer);
1035 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1036 		V_current_state_timers_running6 = 1;
1037 		return (retval);
1038 	}
1039 
1040 	/*
1041 	 * Finally, deal with the case where a group-and-source-specific
1042 	 * query has been received, where a response to a previous g-s-r
1043 	 * query exists, or none exists.
1044 	 * In this case, we need to parse the source-list which the Querier
1045 	 * has provided us with and check if we have any source list filter
1046 	 * entries at T1 for these sources. If we do not, there is no need
1047 	 * schedule a report and the query may be dropped.
1048 	 * If we do, we must record them and schedule a current-state
1049 	 * report for those sources.
1050 	 */
1051 	if (inm->in6m_nsrc > 0) {
1052 		struct mbuf		*m;
1053 		uint8_t			*sp;
1054 		int			 i, nrecorded;
1055 		int			 soff;
1056 
1057 		m = m0;
1058 		soff = off + sizeof(struct mldv2_query);
1059 		nrecorded = 0;
1060 		for (i = 0; i < nsrc; i++) {
1061 			sp = mtod(m, uint8_t *) + soff;
1062 			retval = in6m_record_source(inm,
1063 			    (const struct in6_addr *)sp);
1064 			if (retval < 0)
1065 				break;
1066 			nrecorded += retval;
1067 			soff += sizeof(struct in6_addr);
1068 			if (soff >= m->m_len) {
1069 				soff = soff - m->m_len;
1070 				m = m->m_next;
1071 				if (m == NULL)
1072 					break;
1073 			}
1074 		}
1075 		if (nrecorded > 0) {
1076 			CTR1(KTR_MLD,
1077 			    "%s: schedule response to SG query", __func__);
1078 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1079 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1080 			V_current_state_timers_running6 = 1;
1081 		}
1082 	}
1083 
1084 	return (retval);
1085 }
1086 
1087 /*
1088  * Process a received MLDv1 host membership report.
1089  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1090  *
1091  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1092  * mld_addr. This is OK as we own the mbuf chain.
1093  */
1094 static int
1095 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1096     /*const*/ struct mld_hdr *mld)
1097 {
1098 	struct in6_addr		 src, dst;
1099 	struct in6_ifaddr	*ia;
1100 	struct in6_multi	*inm;
1101 #ifdef KTR
1102 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1103 #endif
1104 
1105 	if (!mld_v1enable) {
1106 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1107 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1108 		    ifp, if_name(ifp));
1109 		return (0);
1110 	}
1111 
1112 	if (ifp->if_flags & IFF_LOOPBACK)
1113 		return (0);
1114 
1115 	/*
1116 	 * MLDv1 reports must originate from a host's link-local address,
1117 	 * or the unspecified address (when booting).
1118 	 */
1119 	src = ip6->ip6_src;
1120 	in6_clearscope(&src);
1121 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1122 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1123 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1124 		    ifp, if_name(ifp));
1125 		return (EINVAL);
1126 	}
1127 
1128 	/*
1129 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1130 	 * group, and must be directed to the group itself.
1131 	 */
1132 	dst = ip6->ip6_dst;
1133 	in6_clearscope(&dst);
1134 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1135 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1136 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1137 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1138 		    ifp, if_name(ifp));
1139 		return (EINVAL);
1140 	}
1141 
1142 	/*
1143 	 * Make sure we don't hear our own membership report, as fast
1144 	 * leave requires knowing that we are the only member of a
1145 	 * group. Assume we used the link-local address if available,
1146 	 * otherwise look for ::.
1147 	 *
1148 	 * XXX Note that scope ID comparison is needed for the address
1149 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1150 	 * performed for the on-wire address.
1151 	 */
1152 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1153 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1154 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1155 		if (ia != NULL)
1156 			ifa_free(&ia->ia_ifa);
1157 		return (0);
1158 	}
1159 	if (ia != NULL)
1160 		ifa_free(&ia->ia_ifa);
1161 
1162 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1163 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1164 
1165 	/*
1166 	 * Embed scope ID of receiving interface in MLD query for lookup
1167 	 * whilst we don't hold other locks (due to KAME locking lameness).
1168 	 */
1169 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1170 		in6_setscope(&mld->mld_addr, ifp, NULL);
1171 
1172 	IN6_MULTI_LOCK();
1173 	MLD_LOCK();
1174 	IF_ADDR_RLOCK(ifp);
1175 
1176 	/*
1177 	 * MLDv1 report suppression.
1178 	 * If we are a member of this group, and our membership should be
1179 	 * reported, and our group timer is pending or about to be reset,
1180 	 * stop our group timer by transitioning to the 'lazy' state.
1181 	 */
1182 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1183 	if (inm != NULL) {
1184 		struct mld_ifsoftc *mli;
1185 
1186 		mli = inm->in6m_mli;
1187 		KASSERT(mli != NULL,
1188 		    ("%s: no mli for ifp %p", __func__, ifp));
1189 
1190 		/*
1191 		 * If we are in MLDv2 host mode, do not allow the
1192 		 * other host's MLDv1 report to suppress our reports.
1193 		 */
1194 		if (mli->mli_version == MLD_VERSION_2)
1195 			goto out_locked;
1196 
1197 		inm->in6m_timer = 0;
1198 
1199 		switch (inm->in6m_state) {
1200 		case MLD_NOT_MEMBER:
1201 		case MLD_SILENT_MEMBER:
1202 		case MLD_SLEEPING_MEMBER:
1203 			break;
1204 		case MLD_REPORTING_MEMBER:
1205 		case MLD_IDLE_MEMBER:
1206 		case MLD_AWAKENING_MEMBER:
1207 			CTR3(KTR_MLD,
1208 			    "report suppressed for %s on ifp %p(%s)",
1209 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1210 			    ifp, if_name(ifp));
1211 		case MLD_LAZY_MEMBER:
1212 			inm->in6m_state = MLD_LAZY_MEMBER;
1213 			break;
1214 		case MLD_G_QUERY_PENDING_MEMBER:
1215 		case MLD_SG_QUERY_PENDING_MEMBER:
1216 		case MLD_LEAVING_MEMBER:
1217 			break;
1218 		}
1219 	}
1220 
1221 out_locked:
1222 	IF_ADDR_RUNLOCK(ifp);
1223 	MLD_UNLOCK();
1224 	IN6_MULTI_UNLOCK();
1225 
1226 	/* XXX Clear embedded scope ID as userland won't expect it. */
1227 	in6_clearscope(&mld->mld_addr);
1228 
1229 	return (0);
1230 }
1231 
1232 /*
1233  * MLD input path.
1234  *
1235  * Assume query messages which fit in a single ICMPv6 message header
1236  * have been pulled up.
1237  * Assume that userland will want to see the message, even if it
1238  * otherwise fails kernel input validation; do not free it.
1239  * Pullup may however free the mbuf chain m if it fails.
1240  *
1241  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1242  */
1243 int
1244 mld_input(struct mbuf *m, int off, int icmp6len)
1245 {
1246 	struct ifnet	*ifp;
1247 	struct ip6_hdr	*ip6;
1248 	struct mld_hdr	*mld;
1249 	int		 mldlen;
1250 
1251 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1252 
1253 	ifp = m->m_pkthdr.rcvif;
1254 
1255 	ip6 = mtod(m, struct ip6_hdr *);
1256 
1257 	/* Pullup to appropriate size. */
1258 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1259 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1260 	    icmp6len >= sizeof(struct mldv2_query)) {
1261 		mldlen = sizeof(struct mldv2_query);
1262 	} else {
1263 		mldlen = sizeof(struct mld_hdr);
1264 	}
1265 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1266 	if (mld == NULL) {
1267 		ICMP6STAT_INC(icp6s_badlen);
1268 		return (IPPROTO_DONE);
1269 	}
1270 
1271 	/*
1272 	 * Userland needs to see all of this traffic for implementing
1273 	 * the endpoint discovery portion of multicast routing.
1274 	 */
1275 	switch (mld->mld_type) {
1276 	case MLD_LISTENER_QUERY:
1277 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1278 		if (icmp6len == sizeof(struct mld_hdr)) {
1279 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1280 				return (0);
1281 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1282 			if (mld_v2_input_query(ifp, ip6, m, off,
1283 			    icmp6len) != 0)
1284 				return (0);
1285 		}
1286 		break;
1287 	case MLD_LISTENER_REPORT:
1288 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1289 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1290 			return (0);
1291 		break;
1292 	case MLDV2_LISTENER_REPORT:
1293 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1294 		break;
1295 	case MLD_LISTENER_DONE:
1296 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1297 		break;
1298 	default:
1299 		break;
1300 	}
1301 
1302 	return (0);
1303 }
1304 
1305 /*
1306  * Fast timeout handler (global).
1307  * VIMAGE: Timeout handlers are expected to service all vimages.
1308  */
1309 void
1310 mld_fasttimo(void)
1311 {
1312 	VNET_ITERATOR_DECL(vnet_iter);
1313 
1314 	VNET_LIST_RLOCK_NOSLEEP();
1315 	VNET_FOREACH(vnet_iter) {
1316 		CURVNET_SET(vnet_iter);
1317 		mld_fasttimo_vnet();
1318 		CURVNET_RESTORE();
1319 	}
1320 	VNET_LIST_RUNLOCK_NOSLEEP();
1321 }
1322 
1323 /*
1324  * Fast timeout handler (per-vnet).
1325  *
1326  * VIMAGE: Assume caller has set up our curvnet.
1327  */
1328 static void
1329 mld_fasttimo_vnet(void)
1330 {
1331 	struct mbufq		 scq;	/* State-change packets */
1332 	struct mbufq		 qrq;	/* Query response packets */
1333 	struct ifnet		*ifp;
1334 	struct mld_ifsoftc	*mli;
1335 	struct ifmultiaddr	*ifma;
1336 	struct in6_multi	*inm, *tinm;
1337 	int			 uri_fasthz;
1338 
1339 	uri_fasthz = 0;
1340 
1341 	/*
1342 	 * Quick check to see if any work needs to be done, in order to
1343 	 * minimize the overhead of fasttimo processing.
1344 	 * SMPng: XXX Unlocked reads.
1345 	 */
1346 	if (!V_current_state_timers_running6 &&
1347 	    !V_interface_timers_running6 &&
1348 	    !V_state_change_timers_running6)
1349 		return;
1350 
1351 	IN6_MULTI_LOCK();
1352 	MLD_LOCK();
1353 
1354 	/*
1355 	 * MLDv2 General Query response timer processing.
1356 	 */
1357 	if (V_interface_timers_running6) {
1358 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1359 
1360 		V_interface_timers_running6 = 0;
1361 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1362 			if (mli->mli_v2_timer == 0) {
1363 				/* Do nothing. */
1364 			} else if (--mli->mli_v2_timer == 0) {
1365 				mld_v2_dispatch_general_query(mli);
1366 			} else {
1367 				V_interface_timers_running6 = 1;
1368 			}
1369 		}
1370 	}
1371 
1372 	if (!V_current_state_timers_running6 &&
1373 	    !V_state_change_timers_running6)
1374 		goto out_locked;
1375 
1376 	V_current_state_timers_running6 = 0;
1377 	V_state_change_timers_running6 = 0;
1378 
1379 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1380 
1381 	/*
1382 	 * MLD host report and state-change timer processing.
1383 	 * Note: Processing a v2 group timer may remove a node.
1384 	 */
1385 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1386 		ifp = mli->mli_ifp;
1387 
1388 		if (mli->mli_version == MLD_VERSION_2) {
1389 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1390 			    PR_FASTHZ);
1391 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1392 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1393 		}
1394 
1395 		IF_ADDR_RLOCK(ifp);
1396 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1397 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1398 			    ifma->ifma_protospec == NULL)
1399 				continue;
1400 			inm = (struct in6_multi *)ifma->ifma_protospec;
1401 			switch (mli->mli_version) {
1402 			case MLD_VERSION_1:
1403 				mld_v1_process_group_timer(mli, inm);
1404 				break;
1405 			case MLD_VERSION_2:
1406 				mld_v2_process_group_timers(mli, &qrq,
1407 				    &scq, inm, uri_fasthz);
1408 				break;
1409 			}
1410 		}
1411 		IF_ADDR_RUNLOCK(ifp);
1412 
1413 		switch (mli->mli_version) {
1414 		case MLD_VERSION_1:
1415 			/*
1416 			 * Transmit reports for this lifecycle.  This
1417 			 * is done while not holding IF_ADDR_LOCK
1418 			 * since this can call
1419 			 * in6ifa_ifpforlinklocal() which locks
1420 			 * IF_ADDR_LOCK internally as well as
1421 			 * ip6_output() to transmit a packet.
1422 			 */
1423 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1424 			    in6m_nrele, tinm) {
1425 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1426 				    in6m_nrele);
1427 				(void)mld_v1_transmit_report(inm,
1428 				    MLD_LISTENER_REPORT);
1429 			}
1430 			break;
1431 		case MLD_VERSION_2:
1432 			mld_dispatch_queue(&qrq, 0);
1433 			mld_dispatch_queue(&scq, 0);
1434 
1435 			/*
1436 			 * Free the in_multi reference(s) for
1437 			 * this lifecycle.
1438 			 */
1439 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1440 			    in6m_nrele, tinm) {
1441 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1442 				    in6m_nrele);
1443 				in6m_release_locked(inm);
1444 			}
1445 			break;
1446 		}
1447 	}
1448 
1449 out_locked:
1450 	MLD_UNLOCK();
1451 	IN6_MULTI_UNLOCK();
1452 }
1453 
1454 /*
1455  * Update host report group timer.
1456  * Will update the global pending timer flags.
1457  */
1458 static void
1459 mld_v1_process_group_timer(struct mld_ifsoftc *mli, struct in6_multi *inm)
1460 {
1461 	int report_timer_expired;
1462 
1463 	IN6_MULTI_LOCK_ASSERT();
1464 	MLD_LOCK_ASSERT();
1465 
1466 	if (inm->in6m_timer == 0) {
1467 		report_timer_expired = 0;
1468 	} else if (--inm->in6m_timer == 0) {
1469 		report_timer_expired = 1;
1470 	} else {
1471 		V_current_state_timers_running6 = 1;
1472 		return;
1473 	}
1474 
1475 	switch (inm->in6m_state) {
1476 	case MLD_NOT_MEMBER:
1477 	case MLD_SILENT_MEMBER:
1478 	case MLD_IDLE_MEMBER:
1479 	case MLD_LAZY_MEMBER:
1480 	case MLD_SLEEPING_MEMBER:
1481 	case MLD_AWAKENING_MEMBER:
1482 		break;
1483 	case MLD_REPORTING_MEMBER:
1484 		if (report_timer_expired) {
1485 			inm->in6m_state = MLD_IDLE_MEMBER;
1486 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
1487 			    in6m_nrele);
1488 		}
1489 		break;
1490 	case MLD_G_QUERY_PENDING_MEMBER:
1491 	case MLD_SG_QUERY_PENDING_MEMBER:
1492 	case MLD_LEAVING_MEMBER:
1493 		break;
1494 	}
1495 }
1496 
1497 /*
1498  * Update a group's timers for MLDv2.
1499  * Will update the global pending timer flags.
1500  * Note: Unlocked read from mli.
1501  */
1502 static void
1503 mld_v2_process_group_timers(struct mld_ifsoftc *mli,
1504     struct mbufq *qrq, struct mbufq *scq,
1505     struct in6_multi *inm, const int uri_fasthz)
1506 {
1507 	int query_response_timer_expired;
1508 	int state_change_retransmit_timer_expired;
1509 #ifdef KTR
1510 	char ip6tbuf[INET6_ADDRSTRLEN];
1511 #endif
1512 
1513 	IN6_MULTI_LOCK_ASSERT();
1514 	MLD_LOCK_ASSERT();
1515 
1516 	query_response_timer_expired = 0;
1517 	state_change_retransmit_timer_expired = 0;
1518 
1519 	/*
1520 	 * During a transition from compatibility mode back to MLDv2,
1521 	 * a group record in REPORTING state may still have its group
1522 	 * timer active. This is a no-op in this function; it is easier
1523 	 * to deal with it here than to complicate the slow-timeout path.
1524 	 */
1525 	if (inm->in6m_timer == 0) {
1526 		query_response_timer_expired = 0;
1527 	} else if (--inm->in6m_timer == 0) {
1528 		query_response_timer_expired = 1;
1529 	} else {
1530 		V_current_state_timers_running6 = 1;
1531 	}
1532 
1533 	if (inm->in6m_sctimer == 0) {
1534 		state_change_retransmit_timer_expired = 0;
1535 	} else if (--inm->in6m_sctimer == 0) {
1536 		state_change_retransmit_timer_expired = 1;
1537 	} else {
1538 		V_state_change_timers_running6 = 1;
1539 	}
1540 
1541 	/* We are in fasttimo, so be quick about it. */
1542 	if (!state_change_retransmit_timer_expired &&
1543 	    !query_response_timer_expired)
1544 		return;
1545 
1546 	switch (inm->in6m_state) {
1547 	case MLD_NOT_MEMBER:
1548 	case MLD_SILENT_MEMBER:
1549 	case MLD_SLEEPING_MEMBER:
1550 	case MLD_LAZY_MEMBER:
1551 	case MLD_AWAKENING_MEMBER:
1552 	case MLD_IDLE_MEMBER:
1553 		break;
1554 	case MLD_G_QUERY_PENDING_MEMBER:
1555 	case MLD_SG_QUERY_PENDING_MEMBER:
1556 		/*
1557 		 * Respond to a previously pending Group-Specific
1558 		 * or Group-and-Source-Specific query by enqueueing
1559 		 * the appropriate Current-State report for
1560 		 * immediate transmission.
1561 		 */
1562 		if (query_response_timer_expired) {
1563 			int retval;
1564 
1565 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1566 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1567 			    0);
1568 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1569 			    __func__, retval);
1570 			inm->in6m_state = MLD_REPORTING_MEMBER;
1571 			in6m_clear_recorded(inm);
1572 		}
1573 		/* FALLTHROUGH */
1574 	case MLD_REPORTING_MEMBER:
1575 	case MLD_LEAVING_MEMBER:
1576 		if (state_change_retransmit_timer_expired) {
1577 			/*
1578 			 * State-change retransmission timer fired.
1579 			 * If there are any further pending retransmissions,
1580 			 * set the global pending state-change flag, and
1581 			 * reset the timer.
1582 			 */
1583 			if (--inm->in6m_scrv > 0) {
1584 				inm->in6m_sctimer = uri_fasthz;
1585 				V_state_change_timers_running6 = 1;
1586 			}
1587 			/*
1588 			 * Retransmit the previously computed state-change
1589 			 * report. If there are no further pending
1590 			 * retransmissions, the mbuf queue will be consumed.
1591 			 * Update T0 state to T1 as we have now sent
1592 			 * a state-change.
1593 			 */
1594 			(void)mld_v2_merge_state_changes(inm, scq);
1595 
1596 			in6m_commit(inm);
1597 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1598 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1599 			    if_name(inm->in6m_ifp));
1600 
1601 			/*
1602 			 * If we are leaving the group for good, make sure
1603 			 * we release MLD's reference to it.
1604 			 * This release must be deferred using a SLIST,
1605 			 * as we are called from a loop which traverses
1606 			 * the in_ifmultiaddr TAILQ.
1607 			 */
1608 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1609 			    inm->in6m_scrv == 0) {
1610 				inm->in6m_state = MLD_NOT_MEMBER;
1611 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1612 				    inm, in6m_nrele);
1613 			}
1614 		}
1615 		break;
1616 	}
1617 }
1618 
1619 /*
1620  * Switch to a different version on the given interface,
1621  * as per Section 9.12.
1622  */
1623 static void
1624 mld_set_version(struct mld_ifsoftc *mli, const int version)
1625 {
1626 	int old_version_timer;
1627 
1628 	MLD_LOCK_ASSERT();
1629 
1630 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1631 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1632 
1633 	if (version == MLD_VERSION_1) {
1634 		/*
1635 		 * Compute the "Older Version Querier Present" timer as per
1636 		 * Section 9.12.
1637 		 */
1638 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1639 		old_version_timer *= PR_SLOWHZ;
1640 		mli->mli_v1_timer = old_version_timer;
1641 	}
1642 
1643 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1644 		mli->mli_version = MLD_VERSION_1;
1645 		mld_v2_cancel_link_timers(mli);
1646 	}
1647 }
1648 
1649 /*
1650  * Cancel pending MLDv2 timers for the given link and all groups
1651  * joined on it; state-change, general-query, and group-query timers.
1652  */
1653 static void
1654 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1655 {
1656 	struct ifmultiaddr	*ifma;
1657 	struct ifnet		*ifp;
1658 	struct in6_multi	*inm, *tinm;
1659 
1660 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1661 	    mli->mli_ifp, if_name(mli->mli_ifp));
1662 
1663 	IN6_MULTI_LOCK_ASSERT();
1664 	MLD_LOCK_ASSERT();
1665 
1666 	/*
1667 	 * Fast-track this potentially expensive operation
1668 	 * by checking all the global 'timer pending' flags.
1669 	 */
1670 	if (!V_interface_timers_running6 &&
1671 	    !V_state_change_timers_running6 &&
1672 	    !V_current_state_timers_running6)
1673 		return;
1674 
1675 	mli->mli_v2_timer = 0;
1676 
1677 	ifp = mli->mli_ifp;
1678 
1679 	IF_ADDR_RLOCK(ifp);
1680 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1681 		if (ifma->ifma_addr->sa_family != AF_INET6)
1682 			continue;
1683 		inm = (struct in6_multi *)ifma->ifma_protospec;
1684 		switch (inm->in6m_state) {
1685 		case MLD_NOT_MEMBER:
1686 		case MLD_SILENT_MEMBER:
1687 		case MLD_IDLE_MEMBER:
1688 		case MLD_LAZY_MEMBER:
1689 		case MLD_SLEEPING_MEMBER:
1690 		case MLD_AWAKENING_MEMBER:
1691 			break;
1692 		case MLD_LEAVING_MEMBER:
1693 			/*
1694 			 * If we are leaving the group and switching
1695 			 * version, we need to release the final
1696 			 * reference held for issuing the INCLUDE {}.
1697 			 */
1698 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
1699 			    in6m_nrele);
1700 			/* FALLTHROUGH */
1701 		case MLD_G_QUERY_PENDING_MEMBER:
1702 		case MLD_SG_QUERY_PENDING_MEMBER:
1703 			in6m_clear_recorded(inm);
1704 			/* FALLTHROUGH */
1705 		case MLD_REPORTING_MEMBER:
1706 			inm->in6m_sctimer = 0;
1707 			inm->in6m_timer = 0;
1708 			inm->in6m_state = MLD_REPORTING_MEMBER;
1709 			/*
1710 			 * Free any pending MLDv2 state-change records.
1711 			 */
1712 			mbufq_drain(&inm->in6m_scq);
1713 			break;
1714 		}
1715 	}
1716 	IF_ADDR_RUNLOCK(ifp);
1717 	SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) {
1718 		SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
1719 		in6m_release_locked(inm);
1720 	}
1721 }
1722 
1723 /*
1724  * Global slowtimo handler.
1725  * VIMAGE: Timeout handlers are expected to service all vimages.
1726  */
1727 void
1728 mld_slowtimo(void)
1729 {
1730 	VNET_ITERATOR_DECL(vnet_iter);
1731 
1732 	VNET_LIST_RLOCK_NOSLEEP();
1733 	VNET_FOREACH(vnet_iter) {
1734 		CURVNET_SET(vnet_iter);
1735 		mld_slowtimo_vnet();
1736 		CURVNET_RESTORE();
1737 	}
1738 	VNET_LIST_RUNLOCK_NOSLEEP();
1739 }
1740 
1741 /*
1742  * Per-vnet slowtimo handler.
1743  */
1744 static void
1745 mld_slowtimo_vnet(void)
1746 {
1747 	struct mld_ifsoftc *mli;
1748 
1749 	MLD_LOCK();
1750 
1751 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1752 		mld_v1_process_querier_timers(mli);
1753 	}
1754 
1755 	MLD_UNLOCK();
1756 }
1757 
1758 /*
1759  * Update the Older Version Querier Present timers for a link.
1760  * See Section 9.12 of RFC 3810.
1761  */
1762 static void
1763 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1764 {
1765 
1766 	MLD_LOCK_ASSERT();
1767 
1768 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1769 		/*
1770 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1771 		 */
1772 		CTR5(KTR_MLD,
1773 		    "%s: transition from v%d -> v%d on %p(%s)",
1774 		    __func__, mli->mli_version, MLD_VERSION_2,
1775 		    mli->mli_ifp, if_name(mli->mli_ifp));
1776 		mli->mli_version = MLD_VERSION_2;
1777 	}
1778 }
1779 
1780 /*
1781  * Transmit an MLDv1 report immediately.
1782  */
1783 static int
1784 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1785 {
1786 	struct ifnet		*ifp;
1787 	struct in6_ifaddr	*ia;
1788 	struct ip6_hdr		*ip6;
1789 	struct mbuf		*mh, *md;
1790 	struct mld_hdr		*mld;
1791 
1792 	IN6_MULTI_LOCK_ASSERT();
1793 	MLD_LOCK_ASSERT();
1794 
1795 	ifp = in6m->in6m_ifp;
1796 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1797 	/* ia may be NULL if link-local address is tentative. */
1798 
1799 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1800 	if (mh == NULL) {
1801 		if (ia != NULL)
1802 			ifa_free(&ia->ia_ifa);
1803 		return (ENOMEM);
1804 	}
1805 	md = m_get(M_NOWAIT, MT_DATA);
1806 	if (md == NULL) {
1807 		m_free(mh);
1808 		if (ia != NULL)
1809 			ifa_free(&ia->ia_ifa);
1810 		return (ENOMEM);
1811 	}
1812 	mh->m_next = md;
1813 
1814 	/*
1815 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1816 	 * that ether_output() does not need to allocate another mbuf
1817 	 * for the header in the most common case.
1818 	 */
1819 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1820 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1821 	mh->m_len = sizeof(struct ip6_hdr);
1822 
1823 	ip6 = mtod(mh, struct ip6_hdr *);
1824 	ip6->ip6_flow = 0;
1825 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1826 	ip6->ip6_vfc |= IPV6_VERSION;
1827 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1828 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1829 	ip6->ip6_dst = in6m->in6m_addr;
1830 
1831 	md->m_len = sizeof(struct mld_hdr);
1832 	mld = mtod(md, struct mld_hdr *);
1833 	mld->mld_type = type;
1834 	mld->mld_code = 0;
1835 	mld->mld_cksum = 0;
1836 	mld->mld_maxdelay = 0;
1837 	mld->mld_reserved = 0;
1838 	mld->mld_addr = in6m->in6m_addr;
1839 	in6_clearscope(&mld->mld_addr);
1840 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1841 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1842 
1843 	mld_save_context(mh, ifp);
1844 	mh->m_flags |= M_MLDV1;
1845 
1846 	mld_dispatch_packet(mh);
1847 
1848 	if (ia != NULL)
1849 		ifa_free(&ia->ia_ifa);
1850 	return (0);
1851 }
1852 
1853 /*
1854  * Process a state change from the upper layer for the given IPv6 group.
1855  *
1856  * Each socket holds a reference on the in_multi in its own ip_moptions.
1857  * The socket layer will have made the necessary updates to.the group
1858  * state, it is now up to MLD to issue a state change report if there
1859  * has been any change between T0 (when the last state-change was issued)
1860  * and T1 (now).
1861  *
1862  * We use the MLDv2 state machine at group level. The MLd module
1863  * however makes the decision as to which MLD protocol version to speak.
1864  * A state change *from* INCLUDE {} always means an initial join.
1865  * A state change *to* INCLUDE {} always means a final leave.
1866  *
1867  * If delay is non-zero, and the state change is an initial multicast
1868  * join, the state change report will be delayed by 'delay' ticks
1869  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1870  * the initial MLDv2 state change report will be delayed by whichever
1871  * is sooner, a pending state-change timer or delay itself.
1872  *
1873  * VIMAGE: curvnet should have been set by caller, as this routine
1874  * is called from the socket option handlers.
1875  */
1876 int
1877 mld_change_state(struct in6_multi *inm, const int delay)
1878 {
1879 	struct mld_ifsoftc *mli;
1880 	struct ifnet *ifp;
1881 	int error;
1882 
1883 	IN6_MULTI_LOCK_ASSERT();
1884 
1885 	error = 0;
1886 
1887 	/*
1888 	 * Try to detect if the upper layer just asked us to change state
1889 	 * for an interface which has now gone away.
1890 	 */
1891 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1892 	ifp = inm->in6m_ifma->ifma_ifp;
1893 	if (ifp != NULL) {
1894 		/*
1895 		 * Sanity check that netinet6's notion of ifp is the
1896 		 * same as net's.
1897 		 */
1898 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1899 	}
1900 
1901 	MLD_LOCK();
1902 
1903 	mli = MLD_IFINFO(ifp);
1904 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1905 
1906 	/*
1907 	 * If we detect a state transition to or from MCAST_UNDEFINED
1908 	 * for this group, then we are starting or finishing an MLD
1909 	 * life cycle for this group.
1910 	 */
1911 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1912 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1913 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1914 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1915 			CTR1(KTR_MLD, "%s: initial join", __func__);
1916 			error = mld_initial_join(inm, mli, delay);
1917 			goto out_locked;
1918 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1919 			CTR1(KTR_MLD, "%s: final leave", __func__);
1920 			mld_final_leave(inm, mli);
1921 			goto out_locked;
1922 		}
1923 	} else {
1924 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1925 	}
1926 
1927 	error = mld_handle_state_change(inm, mli);
1928 
1929 out_locked:
1930 	MLD_UNLOCK();
1931 	return (error);
1932 }
1933 
1934 /*
1935  * Perform the initial join for an MLD group.
1936  *
1937  * When joining a group:
1938  *  If the group should have its MLD traffic suppressed, do nothing.
1939  *  MLDv1 starts sending MLDv1 host membership reports.
1940  *  MLDv2 will schedule an MLDv2 state-change report containing the
1941  *  initial state of the membership.
1942  *
1943  * If the delay argument is non-zero, then we must delay sending the
1944  * initial state change for delay ticks (in units of PR_FASTHZ).
1945  */
1946 static int
1947 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1948     const int delay)
1949 {
1950 	struct ifnet		*ifp;
1951 	struct mbufq		*mq;
1952 	int			 error, retval, syncstates;
1953 	int			 odelay;
1954 #ifdef KTR
1955 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1956 #endif
1957 
1958 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1959 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1960 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1961 
1962 	error = 0;
1963 	syncstates = 1;
1964 
1965 	ifp = inm->in6m_ifp;
1966 
1967 	IN6_MULTI_LOCK_ASSERT();
1968 	MLD_LOCK_ASSERT();
1969 
1970 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1971 
1972 	/*
1973 	 * Groups joined on loopback or marked as 'not reported',
1974 	 * enter the MLD_SILENT_MEMBER state and
1975 	 * are never reported in any protocol exchanges.
1976 	 * All other groups enter the appropriate state machine
1977 	 * for the version in use on this link.
1978 	 * A link marked as MLIF_SILENT causes MLD to be completely
1979 	 * disabled for the link.
1980 	 */
1981 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1982 	    (mli->mli_flags & MLIF_SILENT) ||
1983 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1984 		CTR1(KTR_MLD,
1985 "%s: not kicking state machine for silent group", __func__);
1986 		inm->in6m_state = MLD_SILENT_MEMBER;
1987 		inm->in6m_timer = 0;
1988 	} else {
1989 		/*
1990 		 * Deal with overlapping in_multi lifecycle.
1991 		 * If this group was LEAVING, then make sure
1992 		 * we drop the reference we picked up to keep the
1993 		 * group around for the final INCLUDE {} enqueue.
1994 		 */
1995 		if (mli->mli_version == MLD_VERSION_2 &&
1996 		    inm->in6m_state == MLD_LEAVING_MEMBER)
1997 			in6m_release_locked(inm);
1998 
1999 		inm->in6m_state = MLD_REPORTING_MEMBER;
2000 
2001 		switch (mli->mli_version) {
2002 		case MLD_VERSION_1:
2003 			/*
2004 			 * If a delay was provided, only use it if
2005 			 * it is greater than the delay normally
2006 			 * used for an MLDv1 state change report,
2007 			 * and delay sending the initial MLDv1 report
2008 			 * by not transitioning to the IDLE state.
2009 			 */
2010 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2011 			if (delay) {
2012 				inm->in6m_timer = max(delay, odelay);
2013 				V_current_state_timers_running6 = 1;
2014 			} else {
2015 				inm->in6m_state = MLD_IDLE_MEMBER;
2016 				error = mld_v1_transmit_report(inm,
2017 				     MLD_LISTENER_REPORT);
2018 				if (error == 0) {
2019 					inm->in6m_timer = odelay;
2020 					V_current_state_timers_running6 = 1;
2021 				}
2022 			}
2023 			break;
2024 
2025 		case MLD_VERSION_2:
2026 			/*
2027 			 * Defer update of T0 to T1, until the first copy
2028 			 * of the state change has been transmitted.
2029 			 */
2030 			syncstates = 0;
2031 
2032 			/*
2033 			 * Immediately enqueue a State-Change Report for
2034 			 * this interface, freeing any previous reports.
2035 			 * Don't kick the timers if there is nothing to do,
2036 			 * or if an error occurred.
2037 			 */
2038 			mq = &inm->in6m_scq;
2039 			mbufq_drain(mq);
2040 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2041 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2042 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2043 			    __func__, retval);
2044 			if (retval <= 0) {
2045 				error = retval * -1;
2046 				break;
2047 			}
2048 
2049 			/*
2050 			 * Schedule transmission of pending state-change
2051 			 * report up to RV times for this link. The timer
2052 			 * will fire at the next mld_fasttimo (~200ms),
2053 			 * giving us an opportunity to merge the reports.
2054 			 *
2055 			 * If a delay was provided to this function, only
2056 			 * use this delay if sooner than the existing one.
2057 			 */
2058 			KASSERT(mli->mli_rv > 1,
2059 			   ("%s: invalid robustness %d", __func__,
2060 			    mli->mli_rv));
2061 			inm->in6m_scrv = mli->mli_rv;
2062 			if (delay) {
2063 				if (inm->in6m_sctimer > 1) {
2064 					inm->in6m_sctimer =
2065 					    min(inm->in6m_sctimer, delay);
2066 				} else
2067 					inm->in6m_sctimer = delay;
2068 			} else
2069 				inm->in6m_sctimer = 1;
2070 			V_state_change_timers_running6 = 1;
2071 
2072 			error = 0;
2073 			break;
2074 		}
2075 	}
2076 
2077 	/*
2078 	 * Only update the T0 state if state change is atomic,
2079 	 * i.e. we don't need to wait for a timer to fire before we
2080 	 * can consider the state change to have been communicated.
2081 	 */
2082 	if (syncstates) {
2083 		in6m_commit(inm);
2084 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2085 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2086 		    if_name(inm->in6m_ifp));
2087 	}
2088 
2089 	return (error);
2090 }
2091 
2092 /*
2093  * Issue an intermediate state change during the life-cycle.
2094  */
2095 static int
2096 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2097 {
2098 	struct ifnet		*ifp;
2099 	int			 retval;
2100 #ifdef KTR
2101 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2102 #endif
2103 
2104 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2105 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2106 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2107 
2108 	ifp = inm->in6m_ifp;
2109 
2110 	IN6_MULTI_LOCK_ASSERT();
2111 	MLD_LOCK_ASSERT();
2112 
2113 	KASSERT(mli && mli->mli_ifp == ifp,
2114 	    ("%s: inconsistent ifp", __func__));
2115 
2116 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2117 	    (mli->mli_flags & MLIF_SILENT) ||
2118 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2119 	    (mli->mli_version != MLD_VERSION_2)) {
2120 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2121 			CTR1(KTR_MLD,
2122 "%s: not kicking state machine for silent group", __func__);
2123 		}
2124 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2125 		in6m_commit(inm);
2126 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2127 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2128 		    if_name(inm->in6m_ifp));
2129 		return (0);
2130 	}
2131 
2132 	mbufq_drain(&inm->in6m_scq);
2133 
2134 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2135 	    (mli->mli_flags & MLIF_USEALLOW));
2136 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2137 	if (retval <= 0)
2138 		return (-retval);
2139 
2140 	/*
2141 	 * If record(s) were enqueued, start the state-change
2142 	 * report timer for this group.
2143 	 */
2144 	inm->in6m_scrv = mli->mli_rv;
2145 	inm->in6m_sctimer = 1;
2146 	V_state_change_timers_running6 = 1;
2147 
2148 	return (0);
2149 }
2150 
2151 /*
2152  * Perform the final leave for a multicast address.
2153  *
2154  * When leaving a group:
2155  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2156  *  MLDv2 enqueues a state-change report containing a transition
2157  *  to INCLUDE {} for immediate transmission.
2158  */
2159 static void
2160 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2161 {
2162 	int syncstates;
2163 #ifdef KTR
2164 	char ip6tbuf[INET6_ADDRSTRLEN];
2165 #endif
2166 
2167 	syncstates = 1;
2168 
2169 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2170 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2171 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2172 
2173 	IN6_MULTI_LOCK_ASSERT();
2174 	MLD_LOCK_ASSERT();
2175 
2176 	switch (inm->in6m_state) {
2177 	case MLD_NOT_MEMBER:
2178 	case MLD_SILENT_MEMBER:
2179 	case MLD_LEAVING_MEMBER:
2180 		/* Already leaving or left; do nothing. */
2181 		CTR1(KTR_MLD,
2182 "%s: not kicking state machine for silent group", __func__);
2183 		break;
2184 	case MLD_REPORTING_MEMBER:
2185 	case MLD_IDLE_MEMBER:
2186 	case MLD_G_QUERY_PENDING_MEMBER:
2187 	case MLD_SG_QUERY_PENDING_MEMBER:
2188 		if (mli->mli_version == MLD_VERSION_1) {
2189 #ifdef INVARIANTS
2190 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2191 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2192 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2193 			     __func__);
2194 #endif
2195 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2196 			inm->in6m_state = MLD_NOT_MEMBER;
2197 			V_current_state_timers_running6 = 1;
2198 		} else if (mli->mli_version == MLD_VERSION_2) {
2199 			/*
2200 			 * Stop group timer and all pending reports.
2201 			 * Immediately enqueue a state-change report
2202 			 * TO_IN {} to be sent on the next fast timeout,
2203 			 * giving us an opportunity to merge reports.
2204 			 */
2205 			mbufq_drain(&inm->in6m_scq);
2206 			inm->in6m_timer = 0;
2207 			inm->in6m_scrv = mli->mli_rv;
2208 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2209 			    "pending retransmissions.", __func__,
2210 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2211 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2212 			if (inm->in6m_scrv == 0) {
2213 				inm->in6m_state = MLD_NOT_MEMBER;
2214 				inm->in6m_sctimer = 0;
2215 			} else {
2216 				int retval;
2217 
2218 				in6m_acquire_locked(inm);
2219 
2220 				retval = mld_v2_enqueue_group_record(
2221 				    &inm->in6m_scq, inm, 1, 0, 0,
2222 				    (mli->mli_flags & MLIF_USEALLOW));
2223 				KASSERT(retval != 0,
2224 				    ("%s: enqueue record = %d", __func__,
2225 				     retval));
2226 
2227 				inm->in6m_state = MLD_LEAVING_MEMBER;
2228 				inm->in6m_sctimer = 1;
2229 				V_state_change_timers_running6 = 1;
2230 				syncstates = 0;
2231 			}
2232 			break;
2233 		}
2234 		break;
2235 	case MLD_LAZY_MEMBER:
2236 	case MLD_SLEEPING_MEMBER:
2237 	case MLD_AWAKENING_MEMBER:
2238 		/* Our reports are suppressed; do nothing. */
2239 		break;
2240 	}
2241 
2242 	if (syncstates) {
2243 		in6m_commit(inm);
2244 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2245 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2246 		    if_name(inm->in6m_ifp));
2247 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2248 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2249 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2250 	}
2251 }
2252 
2253 /*
2254  * Enqueue an MLDv2 group record to the given output queue.
2255  *
2256  * If is_state_change is zero, a current-state record is appended.
2257  * If is_state_change is non-zero, a state-change report is appended.
2258  *
2259  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2260  * If is_group_query is zero, and if there is a packet with free space
2261  * at the tail of the queue, it will be appended to providing there
2262  * is enough free space.
2263  * Otherwise a new mbuf packet chain is allocated.
2264  *
2265  * If is_source_query is non-zero, each source is checked to see if
2266  * it was recorded for a Group-Source query, and will be omitted if
2267  * it is not both in-mode and recorded.
2268  *
2269  * If use_block_allow is non-zero, state change reports for initial join
2270  * and final leave, on an inclusive mode group with a source list, will be
2271  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2272  *
2273  * The function will attempt to allocate leading space in the packet
2274  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2275  *
2276  * If successful the size of all data appended to the queue is returned,
2277  * otherwise an error code less than zero is returned, or zero if
2278  * no record(s) were appended.
2279  */
2280 static int
2281 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2282     const int is_state_change, const int is_group_query,
2283     const int is_source_query, const int use_block_allow)
2284 {
2285 	struct mldv2_record	 mr;
2286 	struct mldv2_record	*pmr;
2287 	struct ifnet		*ifp;
2288 	struct ip6_msource	*ims, *nims;
2289 	struct mbuf		*m0, *m, *md;
2290 	int			 error, is_filter_list_change;
2291 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2292 	int			 record_has_sources;
2293 	int			 now;
2294 	int			 type;
2295 	uint8_t			 mode;
2296 #ifdef KTR
2297 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2298 #endif
2299 
2300 	IN6_MULTI_LOCK_ASSERT();
2301 
2302 	error = 0;
2303 	ifp = inm->in6m_ifp;
2304 	is_filter_list_change = 0;
2305 	m = NULL;
2306 	m0 = NULL;
2307 	m0srcs = 0;
2308 	msrcs = 0;
2309 	nbytes = 0;
2310 	nims = NULL;
2311 	record_has_sources = 1;
2312 	pmr = NULL;
2313 	type = MLD_DO_NOTHING;
2314 	mode = inm->in6m_st[1].iss_fmode;
2315 
2316 	/*
2317 	 * If we did not transition out of ASM mode during t0->t1,
2318 	 * and there are no source nodes to process, we can skip
2319 	 * the generation of source records.
2320 	 */
2321 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2322 	    inm->in6m_nsrc == 0)
2323 		record_has_sources = 0;
2324 
2325 	if (is_state_change) {
2326 		/*
2327 		 * Queue a state change record.
2328 		 * If the mode did not change, and there are non-ASM
2329 		 * listeners or source filters present,
2330 		 * we potentially need to issue two records for the group.
2331 		 * If there are ASM listeners, and there was no filter
2332 		 * mode transition of any kind, do nothing.
2333 		 *
2334 		 * If we are transitioning to MCAST_UNDEFINED, we need
2335 		 * not send any sources. A transition to/from this state is
2336 		 * considered inclusive with some special treatment.
2337 		 *
2338 		 * If we are rewriting initial joins/leaves to use
2339 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2340 		 * we need to send sources in all cases.
2341 		 */
2342 		if (mode != inm->in6m_st[0].iss_fmode) {
2343 			if (mode == MCAST_EXCLUDE) {
2344 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2345 				    __func__);
2346 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2347 			} else {
2348 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2349 				    __func__);
2350 				if (use_block_allow) {
2351 					/*
2352 					 * XXX
2353 					 * Here we're interested in state
2354 					 * edges either direction between
2355 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2356 					 * Perhaps we should just check
2357 					 * the group state, rather than
2358 					 * the filter mode.
2359 					 */
2360 					if (mode == MCAST_UNDEFINED) {
2361 						type = MLD_BLOCK_OLD_SOURCES;
2362 					} else {
2363 						type = MLD_ALLOW_NEW_SOURCES;
2364 					}
2365 				} else {
2366 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2367 					if (mode == MCAST_UNDEFINED)
2368 						record_has_sources = 0;
2369 				}
2370 			}
2371 		} else {
2372 			if (record_has_sources) {
2373 				is_filter_list_change = 1;
2374 			} else {
2375 				type = MLD_DO_NOTHING;
2376 			}
2377 		}
2378 	} else {
2379 		/*
2380 		 * Queue a current state record.
2381 		 */
2382 		if (mode == MCAST_EXCLUDE) {
2383 			type = MLD_MODE_IS_EXCLUDE;
2384 		} else if (mode == MCAST_INCLUDE) {
2385 			type = MLD_MODE_IS_INCLUDE;
2386 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2387 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2388 			     __func__, inm, inm->in6m_st[1].iss_asm));
2389 		}
2390 	}
2391 
2392 	/*
2393 	 * Generate the filter list changes using a separate function.
2394 	 */
2395 	if (is_filter_list_change)
2396 		return (mld_v2_enqueue_filter_change(mq, inm));
2397 
2398 	if (type == MLD_DO_NOTHING) {
2399 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2400 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2401 		    if_name(inm->in6m_ifp));
2402 		return (0);
2403 	}
2404 
2405 	/*
2406 	 * If any sources are present, we must be able to fit at least
2407 	 * one in the trailing space of the tail packet's mbuf,
2408 	 * ideally more.
2409 	 */
2410 	minrec0len = sizeof(struct mldv2_record);
2411 	if (record_has_sources)
2412 		minrec0len += sizeof(struct in6_addr);
2413 
2414 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2415 	    mld_rec_type_to_str(type),
2416 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2417 	    if_name(inm->in6m_ifp));
2418 
2419 	/*
2420 	 * Check if we have a packet in the tail of the queue for this
2421 	 * group into which the first group record for this group will fit.
2422 	 * Otherwise allocate a new packet.
2423 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2424 	 * Note: Group records for G/GSR query responses MUST be sent
2425 	 * in their own packet.
2426 	 */
2427 	m0 = mbufq_last(mq);
2428 	if (!is_group_query &&
2429 	    m0 != NULL &&
2430 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2431 	    (m0->m_pkthdr.len + minrec0len) <
2432 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2433 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2434 			    sizeof(struct mldv2_record)) /
2435 			    sizeof(struct in6_addr);
2436 		m = m0;
2437 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2438 	} else {
2439 		if (mbufq_full(mq)) {
2440 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2441 			return (-ENOMEM);
2442 		}
2443 		m = NULL;
2444 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2445 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2446 		if (!is_state_change && !is_group_query)
2447 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2448 		if (m == NULL)
2449 			m = m_gethdr(M_NOWAIT, MT_DATA);
2450 		if (m == NULL)
2451 			return (-ENOMEM);
2452 
2453 		mld_save_context(m, ifp);
2454 
2455 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2456 	}
2457 
2458 	/*
2459 	 * Append group record.
2460 	 * If we have sources, we don't know how many yet.
2461 	 */
2462 	mr.mr_type = type;
2463 	mr.mr_datalen = 0;
2464 	mr.mr_numsrc = 0;
2465 	mr.mr_addr = inm->in6m_addr;
2466 	in6_clearscope(&mr.mr_addr);
2467 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2468 		if (m != m0)
2469 			m_freem(m);
2470 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2471 		return (-ENOMEM);
2472 	}
2473 	nbytes += sizeof(struct mldv2_record);
2474 
2475 	/*
2476 	 * Append as many sources as will fit in the first packet.
2477 	 * If we are appending to a new packet, the chain allocation
2478 	 * may potentially use clusters; use m_getptr() in this case.
2479 	 * If we are appending to an existing packet, we need to obtain
2480 	 * a pointer to the group record after m_append(), in case a new
2481 	 * mbuf was allocated.
2482 	 *
2483 	 * Only append sources which are in-mode at t1. If we are
2484 	 * transitioning to MCAST_UNDEFINED state on the group, and
2485 	 * use_block_allow is zero, do not include source entries.
2486 	 * Otherwise, we need to include this source in the report.
2487 	 *
2488 	 * Only report recorded sources in our filter set when responding
2489 	 * to a group-source query.
2490 	 */
2491 	if (record_has_sources) {
2492 		if (m == m0) {
2493 			md = m_last(m);
2494 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2495 			    md->m_len - nbytes);
2496 		} else {
2497 			md = m_getptr(m, 0, &off);
2498 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2499 			    off);
2500 		}
2501 		msrcs = 0;
2502 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2503 		    nims) {
2504 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2505 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2506 			now = im6s_get_mode(inm, ims, 1);
2507 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2508 			if ((now != mode) ||
2509 			    (now == mode &&
2510 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2511 				CTR1(KTR_MLD, "%s: skip node", __func__);
2512 				continue;
2513 			}
2514 			if (is_source_query && ims->im6s_stp == 0) {
2515 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2516 				    __func__);
2517 				continue;
2518 			}
2519 			CTR1(KTR_MLD, "%s: append node", __func__);
2520 			if (!m_append(m, sizeof(struct in6_addr),
2521 			    (void *)&ims->im6s_addr)) {
2522 				if (m != m0)
2523 					m_freem(m);
2524 				CTR1(KTR_MLD, "%s: m_append() failed.",
2525 				    __func__);
2526 				return (-ENOMEM);
2527 			}
2528 			nbytes += sizeof(struct in6_addr);
2529 			++msrcs;
2530 			if (msrcs == m0srcs)
2531 				break;
2532 		}
2533 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2534 		    msrcs);
2535 		pmr->mr_numsrc = htons(msrcs);
2536 		nbytes += (msrcs * sizeof(struct in6_addr));
2537 	}
2538 
2539 	if (is_source_query && msrcs == 0) {
2540 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2541 		if (m != m0)
2542 			m_freem(m);
2543 		return (0);
2544 	}
2545 
2546 	/*
2547 	 * We are good to go with first packet.
2548 	 */
2549 	if (m != m0) {
2550 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2551 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2552 		mbufq_enqueue(mq, m);
2553 	} else
2554 		m->m_pkthdr.PH_vt.vt_nrecs++;
2555 
2556 	/*
2557 	 * No further work needed if no source list in packet(s).
2558 	 */
2559 	if (!record_has_sources)
2560 		return (nbytes);
2561 
2562 	/*
2563 	 * Whilst sources remain to be announced, we need to allocate
2564 	 * a new packet and fill out as many sources as will fit.
2565 	 * Always try for a cluster first.
2566 	 */
2567 	while (nims != NULL) {
2568 		if (mbufq_full(mq)) {
2569 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2570 			return (-ENOMEM);
2571 		}
2572 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2573 		if (m == NULL)
2574 			m = m_gethdr(M_NOWAIT, MT_DATA);
2575 		if (m == NULL)
2576 			return (-ENOMEM);
2577 		mld_save_context(m, ifp);
2578 		md = m_getptr(m, 0, &off);
2579 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2580 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2581 
2582 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2583 			if (m != m0)
2584 				m_freem(m);
2585 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2586 			return (-ENOMEM);
2587 		}
2588 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2589 		nbytes += sizeof(struct mldv2_record);
2590 
2591 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2592 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2593 
2594 		msrcs = 0;
2595 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2596 			CTR2(KTR_MLD, "%s: visit node %s",
2597 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2598 			now = im6s_get_mode(inm, ims, 1);
2599 			if ((now != mode) ||
2600 			    (now == mode &&
2601 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2602 				CTR1(KTR_MLD, "%s: skip node", __func__);
2603 				continue;
2604 			}
2605 			if (is_source_query && ims->im6s_stp == 0) {
2606 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2607 				    __func__);
2608 				continue;
2609 			}
2610 			CTR1(KTR_MLD, "%s: append node", __func__);
2611 			if (!m_append(m, sizeof(struct in6_addr),
2612 			    (void *)&ims->im6s_addr)) {
2613 				if (m != m0)
2614 					m_freem(m);
2615 				CTR1(KTR_MLD, "%s: m_append() failed.",
2616 				    __func__);
2617 				return (-ENOMEM);
2618 			}
2619 			++msrcs;
2620 			if (msrcs == m0srcs)
2621 				break;
2622 		}
2623 		pmr->mr_numsrc = htons(msrcs);
2624 		nbytes += (msrcs * sizeof(struct in6_addr));
2625 
2626 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2627 		mbufq_enqueue(mq, m);
2628 	}
2629 
2630 	return (nbytes);
2631 }
2632 
2633 /*
2634  * Type used to mark record pass completion.
2635  * We exploit the fact we can cast to this easily from the
2636  * current filter modes on each ip_msource node.
2637  */
2638 typedef enum {
2639 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2640 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2641 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2642 	REC_FULL = REC_ALLOW | REC_BLOCK
2643 } rectype_t;
2644 
2645 /*
2646  * Enqueue an MLDv2 filter list change to the given output queue.
2647  *
2648  * Source list filter state is held in an RB-tree. When the filter list
2649  * for a group is changed without changing its mode, we need to compute
2650  * the deltas between T0 and T1 for each source in the filter set,
2651  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2652  *
2653  * As we may potentially queue two record types, and the entire R-B tree
2654  * needs to be walked at once, we break this out into its own function
2655  * so we can generate a tightly packed queue of packets.
2656  *
2657  * XXX This could be written to only use one tree walk, although that makes
2658  * serializing into the mbuf chains a bit harder. For now we do two walks
2659  * which makes things easier on us, and it may or may not be harder on
2660  * the L2 cache.
2661  *
2662  * If successful the size of all data appended to the queue is returned,
2663  * otherwise an error code less than zero is returned, or zero if
2664  * no record(s) were appended.
2665  */
2666 static int
2667 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2668 {
2669 	static const int MINRECLEN =
2670 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2671 	struct ifnet		*ifp;
2672 	struct mldv2_record	 mr;
2673 	struct mldv2_record	*pmr;
2674 	struct ip6_msource	*ims, *nims;
2675 	struct mbuf		*m, *m0, *md;
2676 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2677 	int			 nallow, nblock;
2678 	uint8_t			 mode, now, then;
2679 	rectype_t		 crt, drt, nrt;
2680 #ifdef KTR
2681 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2682 #endif
2683 
2684 	IN6_MULTI_LOCK_ASSERT();
2685 
2686 	if (inm->in6m_nsrc == 0 ||
2687 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2688 		return (0);
2689 
2690 	ifp = inm->in6m_ifp;			/* interface */
2691 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2692 	crt = REC_NONE;	/* current group record type */
2693 	drt = REC_NONE;	/* mask of completed group record types */
2694 	nrt = REC_NONE;	/* record type for current node */
2695 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2696 	npbytes = 0;	/* # of bytes appended this packet */
2697 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2698 	rsrcs = 0;	/* # sources encoded in current record */
2699 	schanged = 0;	/* # nodes encoded in overall filter change */
2700 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2701 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2702 	nims = NULL;	/* next tree node pointer */
2703 
2704 	/*
2705 	 * For each possible filter record mode.
2706 	 * The first kind of source we encounter tells us which
2707 	 * is the first kind of record we start appending.
2708 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2709 	 * as the inverse of the group's filter mode.
2710 	 */
2711 	while (drt != REC_FULL) {
2712 		do {
2713 			m0 = mbufq_last(mq);
2714 			if (m0 != NULL &&
2715 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2716 			     MLD_V2_REPORT_MAXRECS) &&
2717 			    (m0->m_pkthdr.len + MINRECLEN) <
2718 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2719 				m = m0;
2720 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2721 					    sizeof(struct mldv2_record)) /
2722 					    sizeof(struct in6_addr);
2723 				CTR1(KTR_MLD,
2724 				    "%s: use previous packet", __func__);
2725 			} else {
2726 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2727 				if (m == NULL)
2728 					m = m_gethdr(M_NOWAIT, MT_DATA);
2729 				if (m == NULL) {
2730 					CTR1(KTR_MLD,
2731 					    "%s: m_get*() failed", __func__);
2732 					return (-ENOMEM);
2733 				}
2734 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2735 				mld_save_context(m, ifp);
2736 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2737 				    sizeof(struct mldv2_record)) /
2738 				    sizeof(struct in6_addr);
2739 				npbytes = 0;
2740 				CTR1(KTR_MLD,
2741 				    "%s: allocated new packet", __func__);
2742 			}
2743 			/*
2744 			 * Append the MLD group record header to the
2745 			 * current packet's data area.
2746 			 * Recalculate pointer to free space for next
2747 			 * group record, in case m_append() allocated
2748 			 * a new mbuf or cluster.
2749 			 */
2750 			memset(&mr, 0, sizeof(mr));
2751 			mr.mr_addr = inm->in6m_addr;
2752 			in6_clearscope(&mr.mr_addr);
2753 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2754 				if (m != m0)
2755 					m_freem(m);
2756 				CTR1(KTR_MLD,
2757 				    "%s: m_append() failed", __func__);
2758 				return (-ENOMEM);
2759 			}
2760 			npbytes += sizeof(struct mldv2_record);
2761 			if (m != m0) {
2762 				/* new packet; offset in chain */
2763 				md = m_getptr(m, npbytes -
2764 				    sizeof(struct mldv2_record), &off);
2765 				pmr = (struct mldv2_record *)(mtod(md,
2766 				    uint8_t *) + off);
2767 			} else {
2768 				/* current packet; offset from last append */
2769 				md = m_last(m);
2770 				pmr = (struct mldv2_record *)(mtod(md,
2771 				    uint8_t *) + md->m_len -
2772 				    sizeof(struct mldv2_record));
2773 			}
2774 			/*
2775 			 * Begin walking the tree for this record type
2776 			 * pass, or continue from where we left off
2777 			 * previously if we had to allocate a new packet.
2778 			 * Only report deltas in-mode at t1.
2779 			 * We need not report included sources as allowed
2780 			 * if we are in inclusive mode on the group,
2781 			 * however the converse is not true.
2782 			 */
2783 			rsrcs = 0;
2784 			if (nims == NULL) {
2785 				nims = RB_MIN(ip6_msource_tree,
2786 				    &inm->in6m_srcs);
2787 			}
2788 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2789 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2790 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2791 				now = im6s_get_mode(inm, ims, 1);
2792 				then = im6s_get_mode(inm, ims, 0);
2793 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2794 				    __func__, then, now);
2795 				if (now == then) {
2796 					CTR1(KTR_MLD,
2797 					    "%s: skip unchanged", __func__);
2798 					continue;
2799 				}
2800 				if (mode == MCAST_EXCLUDE &&
2801 				    now == MCAST_INCLUDE) {
2802 					CTR1(KTR_MLD,
2803 					    "%s: skip IN src on EX group",
2804 					    __func__);
2805 					continue;
2806 				}
2807 				nrt = (rectype_t)now;
2808 				if (nrt == REC_NONE)
2809 					nrt = (rectype_t)(~mode & REC_FULL);
2810 				if (schanged++ == 0) {
2811 					crt = nrt;
2812 				} else if (crt != nrt)
2813 					continue;
2814 				if (!m_append(m, sizeof(struct in6_addr),
2815 				    (void *)&ims->im6s_addr)) {
2816 					if (m != m0)
2817 						m_freem(m);
2818 					CTR1(KTR_MLD,
2819 					    "%s: m_append() failed", __func__);
2820 					return (-ENOMEM);
2821 				}
2822 				nallow += !!(crt == REC_ALLOW);
2823 				nblock += !!(crt == REC_BLOCK);
2824 				if (++rsrcs == m0srcs)
2825 					break;
2826 			}
2827 			/*
2828 			 * If we did not append any tree nodes on this
2829 			 * pass, back out of allocations.
2830 			 */
2831 			if (rsrcs == 0) {
2832 				npbytes -= sizeof(struct mldv2_record);
2833 				if (m != m0) {
2834 					CTR1(KTR_MLD,
2835 					    "%s: m_free(m)", __func__);
2836 					m_freem(m);
2837 				} else {
2838 					CTR1(KTR_MLD,
2839 					    "%s: m_adj(m, -mr)", __func__);
2840 					m_adj(m, -((int)sizeof(
2841 					    struct mldv2_record)));
2842 				}
2843 				continue;
2844 			}
2845 			npbytes += (rsrcs * sizeof(struct in6_addr));
2846 			if (crt == REC_ALLOW)
2847 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2848 			else if (crt == REC_BLOCK)
2849 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2850 			pmr->mr_numsrc = htons(rsrcs);
2851 			/*
2852 			 * Count the new group record, and enqueue this
2853 			 * packet if it wasn't already queued.
2854 			 */
2855 			m->m_pkthdr.PH_vt.vt_nrecs++;
2856 			if (m != m0)
2857 				mbufq_enqueue(mq, m);
2858 			nbytes += npbytes;
2859 		} while (nims != NULL);
2860 		drt |= crt;
2861 		crt = (~crt & REC_FULL);
2862 	}
2863 
2864 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2865 	    nallow, nblock);
2866 
2867 	return (nbytes);
2868 }
2869 
2870 static int
2871 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2872 {
2873 	struct mbufq	*gq;
2874 	struct mbuf	*m;		/* pending state-change */
2875 	struct mbuf	*m0;		/* copy of pending state-change */
2876 	struct mbuf	*mt;		/* last state-change in packet */
2877 	int		 docopy, domerge;
2878 	u_int		 recslen;
2879 
2880 	docopy = 0;
2881 	domerge = 0;
2882 	recslen = 0;
2883 
2884 	IN6_MULTI_LOCK_ASSERT();
2885 	MLD_LOCK_ASSERT();
2886 
2887 	/*
2888 	 * If there are further pending retransmissions, make a writable
2889 	 * copy of each queued state-change message before merging.
2890 	 */
2891 	if (inm->in6m_scrv > 0)
2892 		docopy = 1;
2893 
2894 	gq = &inm->in6m_scq;
2895 #ifdef KTR
2896 	if (mbufq_first(gq) == NULL) {
2897 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2898 		    __func__, inm);
2899 	}
2900 #endif
2901 
2902 	m = mbufq_first(gq);
2903 	while (m != NULL) {
2904 		/*
2905 		 * Only merge the report into the current packet if
2906 		 * there is sufficient space to do so; an MLDv2 report
2907 		 * packet may only contain 65,535 group records.
2908 		 * Always use a simple mbuf chain concatentation to do this,
2909 		 * as large state changes for single groups may have
2910 		 * allocated clusters.
2911 		 */
2912 		domerge = 0;
2913 		mt = mbufq_last(scq);
2914 		if (mt != NULL) {
2915 			recslen = m_length(m, NULL);
2916 
2917 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2918 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2919 			    MLD_V2_REPORT_MAXRECS) &&
2920 			    (mt->m_pkthdr.len + recslen <=
2921 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2922 				domerge = 1;
2923 		}
2924 
2925 		if (!domerge && mbufq_full(gq)) {
2926 			CTR2(KTR_MLD,
2927 			    "%s: outbound queue full, skipping whole packet %p",
2928 			    __func__, m);
2929 			mt = m->m_nextpkt;
2930 			if (!docopy)
2931 				m_freem(m);
2932 			m = mt;
2933 			continue;
2934 		}
2935 
2936 		if (!docopy) {
2937 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2938 			m0 = mbufq_dequeue(gq);
2939 			m = m0->m_nextpkt;
2940 		} else {
2941 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2942 			m0 = m_dup(m, M_NOWAIT);
2943 			if (m0 == NULL)
2944 				return (ENOMEM);
2945 			m0->m_nextpkt = NULL;
2946 			m = m->m_nextpkt;
2947 		}
2948 
2949 		if (!domerge) {
2950 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2951 			    __func__, m0, scq);
2952 			mbufq_enqueue(scq, m0);
2953 		} else {
2954 			struct mbuf *mtl;	/* last mbuf of packet mt */
2955 
2956 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2957 			    __func__, m0, mt);
2958 
2959 			mtl = m_last(mt);
2960 			m0->m_flags &= ~M_PKTHDR;
2961 			mt->m_pkthdr.len += recslen;
2962 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2963 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2964 
2965 			mtl->m_next = m0;
2966 		}
2967 	}
2968 
2969 	return (0);
2970 }
2971 
2972 /*
2973  * Respond to a pending MLDv2 General Query.
2974  */
2975 static void
2976 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2977 {
2978 	struct ifmultiaddr	*ifma;
2979 	struct ifnet		*ifp;
2980 	struct in6_multi	*inm;
2981 	int			 retval;
2982 
2983 	IN6_MULTI_LOCK_ASSERT();
2984 	MLD_LOCK_ASSERT();
2985 
2986 	KASSERT(mli->mli_version == MLD_VERSION_2,
2987 	    ("%s: called when version %d", __func__, mli->mli_version));
2988 
2989 	/*
2990 	 * Check that there are some packets queued. If so, send them first.
2991 	 * For large number of groups the reply to general query can take
2992 	 * many packets, we should finish sending them before starting of
2993 	 * queuing the new reply.
2994 	 */
2995 	if (mbufq_len(&mli->mli_gq) != 0)
2996 		goto send;
2997 
2998 	ifp = mli->mli_ifp;
2999 
3000 	IF_ADDR_RLOCK(ifp);
3001 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3002 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
3003 		    ifma->ifma_protospec == NULL)
3004 			continue;
3005 
3006 		inm = (struct in6_multi *)ifma->ifma_protospec;
3007 		KASSERT(ifp == inm->in6m_ifp,
3008 		    ("%s: inconsistent ifp", __func__));
3009 
3010 		switch (inm->in6m_state) {
3011 		case MLD_NOT_MEMBER:
3012 		case MLD_SILENT_MEMBER:
3013 			break;
3014 		case MLD_REPORTING_MEMBER:
3015 		case MLD_IDLE_MEMBER:
3016 		case MLD_LAZY_MEMBER:
3017 		case MLD_SLEEPING_MEMBER:
3018 		case MLD_AWAKENING_MEMBER:
3019 			inm->in6m_state = MLD_REPORTING_MEMBER;
3020 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3021 			    inm, 0, 0, 0, 0);
3022 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3023 			    __func__, retval);
3024 			break;
3025 		case MLD_G_QUERY_PENDING_MEMBER:
3026 		case MLD_SG_QUERY_PENDING_MEMBER:
3027 		case MLD_LEAVING_MEMBER:
3028 			break;
3029 		}
3030 	}
3031 	IF_ADDR_RUNLOCK(ifp);
3032 
3033 send:
3034 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3035 
3036 	/*
3037 	 * Slew transmission of bursts over 500ms intervals.
3038 	 */
3039 	if (mbufq_first(&mli->mli_gq) != NULL) {
3040 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3041 		    MLD_RESPONSE_BURST_INTERVAL);
3042 		V_interface_timers_running6 = 1;
3043 	}
3044 }
3045 
3046 /*
3047  * Transmit the next pending message in the output queue.
3048  *
3049  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3050  * MRT: Nothing needs to be done, as MLD traffic is always local to
3051  * a link and uses a link-scope multicast address.
3052  */
3053 static void
3054 mld_dispatch_packet(struct mbuf *m)
3055 {
3056 	struct ip6_moptions	 im6o;
3057 	struct ifnet		*ifp;
3058 	struct ifnet		*oifp;
3059 	struct mbuf		*m0;
3060 	struct mbuf		*md;
3061 	struct ip6_hdr		*ip6;
3062 	struct mld_hdr		*mld;
3063 	int			 error;
3064 	int			 off;
3065 	int			 type;
3066 	uint32_t		 ifindex;
3067 
3068 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3069 
3070 	/*
3071 	 * Set VNET image pointer from enqueued mbuf chain
3072 	 * before doing anything else. Whilst we use interface
3073 	 * indexes to guard against interface detach, they are
3074 	 * unique to each VIMAGE and must be retrieved.
3075 	 */
3076 	ifindex = mld_restore_context(m);
3077 
3078 	/*
3079 	 * Check if the ifnet still exists. This limits the scope of
3080 	 * any race in the absence of a global ifp lock for low cost
3081 	 * (an array lookup).
3082 	 */
3083 	ifp = ifnet_byindex(ifindex);
3084 	if (ifp == NULL) {
3085 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3086 		    __func__, m, ifindex);
3087 		m_freem(m);
3088 		IP6STAT_INC(ip6s_noroute);
3089 		goto out;
3090 	}
3091 
3092 	im6o.im6o_multicast_hlim  = 1;
3093 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3094 	im6o.im6o_multicast_ifp = ifp;
3095 
3096 	if (m->m_flags & M_MLDV1) {
3097 		m0 = m;
3098 	} else {
3099 		m0 = mld_v2_encap_report(ifp, m);
3100 		if (m0 == NULL) {
3101 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3102 			IP6STAT_INC(ip6s_odropped);
3103 			goto out;
3104 		}
3105 	}
3106 
3107 	mld_scrub_context(m0);
3108 	m_clrprotoflags(m);
3109 	m0->m_pkthdr.rcvif = V_loif;
3110 
3111 	ip6 = mtod(m0, struct ip6_hdr *);
3112 #if 0
3113 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3114 #else
3115 	/*
3116 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3117 	 * occur if we called in6_setscope() at transmission.
3118 	 * See comments at top of file.
3119 	 */
3120 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3121 #endif
3122 
3123 	/*
3124 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3125 	 * so we can bump the stats.
3126 	 */
3127 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3128 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3129 	type = mld->mld_type;
3130 
3131 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3132 	    &oifp, NULL);
3133 	if (error) {
3134 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3135 		goto out;
3136 	}
3137 	ICMP6STAT_INC(icp6s_outhist[type]);
3138 	if (oifp != NULL) {
3139 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3140 		switch (type) {
3141 		case MLD_LISTENER_REPORT:
3142 		case MLDV2_LISTENER_REPORT:
3143 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3144 			break;
3145 		case MLD_LISTENER_DONE:
3146 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3147 			break;
3148 		}
3149 	}
3150 out:
3151 	return;
3152 }
3153 
3154 /*
3155  * Encapsulate an MLDv2 report.
3156  *
3157  * KAME IPv6 requires that hop-by-hop options be passed separately,
3158  * and that the IPv6 header be prepended in a separate mbuf.
3159  *
3160  * Returns a pointer to the new mbuf chain head, or NULL if the
3161  * allocation failed.
3162  */
3163 static struct mbuf *
3164 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3165 {
3166 	struct mbuf		*mh;
3167 	struct mldv2_report	*mld;
3168 	struct ip6_hdr		*ip6;
3169 	struct in6_ifaddr	*ia;
3170 	int			 mldreclen;
3171 
3172 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3173 	KASSERT((m->m_flags & M_PKTHDR),
3174 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3175 
3176 	/*
3177 	 * RFC3590: OK to send as :: or tentative during DAD.
3178 	 */
3179 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3180 	if (ia == NULL)
3181 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3182 
3183 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3184 	if (mh == NULL) {
3185 		if (ia != NULL)
3186 			ifa_free(&ia->ia_ifa);
3187 		m_freem(m);
3188 		return (NULL);
3189 	}
3190 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3191 
3192 	mldreclen = m_length(m, NULL);
3193 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3194 
3195 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3196 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3197 	    sizeof(struct mldv2_report) + mldreclen;
3198 
3199 	ip6 = mtod(mh, struct ip6_hdr *);
3200 	ip6->ip6_flow = 0;
3201 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3202 	ip6->ip6_vfc |= IPV6_VERSION;
3203 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3204 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3205 	if (ia != NULL)
3206 		ifa_free(&ia->ia_ifa);
3207 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3208 	/* scope ID will be set in netisr */
3209 
3210 	mld = (struct mldv2_report *)(ip6 + 1);
3211 	mld->mld_type = MLDV2_LISTENER_REPORT;
3212 	mld->mld_code = 0;
3213 	mld->mld_cksum = 0;
3214 	mld->mld_v2_reserved = 0;
3215 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3216 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3217 
3218 	mh->m_next = m;
3219 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3220 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3221 	return (mh);
3222 }
3223 
3224 #ifdef KTR
3225 static char *
3226 mld_rec_type_to_str(const int type)
3227 {
3228 
3229 	switch (type) {
3230 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3231 			return "TO_EX";
3232 			break;
3233 		case MLD_CHANGE_TO_INCLUDE_MODE:
3234 			return "TO_IN";
3235 			break;
3236 		case MLD_MODE_IS_EXCLUDE:
3237 			return "MODE_EX";
3238 			break;
3239 		case MLD_MODE_IS_INCLUDE:
3240 			return "MODE_IN";
3241 			break;
3242 		case MLD_ALLOW_NEW_SOURCES:
3243 			return "ALLOW_NEW";
3244 			break;
3245 		case MLD_BLOCK_OLD_SOURCES:
3246 			return "BLOCK_OLD";
3247 			break;
3248 		default:
3249 			break;
3250 	}
3251 	return "unknown";
3252 }
3253 #endif
3254 
3255 static void
3256 mld_init(void *unused __unused)
3257 {
3258 
3259 	CTR1(KTR_MLD, "%s: initializing", __func__);
3260 	MLD_LOCK_INIT();
3261 
3262 	ip6_initpktopts(&mld_po);
3263 	mld_po.ip6po_hlim = 1;
3264 	mld_po.ip6po_hbh = &mld_ra.hbh;
3265 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3266 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3267 }
3268 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
3269 
3270 static void
3271 mld_uninit(void *unused __unused)
3272 {
3273 
3274 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3275 	MLD_LOCK_DESTROY();
3276 }
3277 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
3278 
3279 static void
3280 vnet_mld_init(const void *unused __unused)
3281 {
3282 
3283 	CTR1(KTR_MLD, "%s: initializing", __func__);
3284 
3285 	LIST_INIT(&V_mli_head);
3286 }
3287 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
3288     NULL);
3289 
3290 static void
3291 vnet_mld_uninit(const void *unused __unused)
3292 {
3293 
3294 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3295 
3296 	KASSERT(LIST_EMPTY(&V_mli_head),
3297 	    ("%s: mli list not empty; ifnets not detached?", __func__));
3298 }
3299 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
3300     NULL);
3301 
3302 static int
3303 mld_modevent(module_t mod, int type, void *unused __unused)
3304 {
3305 
3306     switch (type) {
3307     case MOD_LOAD:
3308     case MOD_UNLOAD:
3309 	break;
3310     default:
3311 	return (EOPNOTSUPP);
3312     }
3313     return (0);
3314 }
3315 
3316 static moduledata_t mld_mod = {
3317     "mld",
3318     mld_modevent,
3319     0
3320 };
3321 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3322