xref: /freebsd/sys/netinet6/mld6.c (revision 8eb2bee6c0f4957c6c1cea826e59cda4d18a2a64)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote
15  *    products derived from this software without specific prior written
16  *    permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
31  */
32 
33 /*-
34  * Copyright (c) 1988 Stephen Deering.
35  * Copyright (c) 1992, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * This code is derived from software contributed to Berkeley by
39  * Stephen Deering of Stanford University.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
66  */
67 
68 #include <sys/cdefs.h>
69 __FBSDID("$FreeBSD$");
70 
71 #include "opt_inet.h"
72 #include "opt_inet6.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/protosw.h>
79 #include <sys/sysctl.h>
80 #include <sys/kernel.h>
81 #include <sys/callout.h>
82 #include <sys/malloc.h>
83 #include <sys/module.h>
84 #include <sys/ktr.h>
85 
86 #include <net/if.h>
87 #include <net/if_var.h>
88 #include <net/route.h>
89 #include <net/vnet.h>
90 
91 #include <netinet/in.h>
92 #include <netinet/in_var.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/mld6.h>
99 #include <netinet6/mld6_var.h>
100 
101 #include <security/mac/mac_framework.h>
102 
103 #ifndef KTR_MLD
104 #define KTR_MLD KTR_INET6
105 #endif
106 
107 static void	mli_delete_locked(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(struct in6_multi_head *inmh);
112 static int	mld_handle_state_change(struct in6_multi *,
113 		    struct mld_ifsoftc *);
114 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
115 		    const int);
116 #ifdef KTR
117 static char *	mld_rec_type_to_str(const int);
118 #endif
119 static void	mld_set_version(struct mld_ifsoftc *, const int);
120 static void	mld_slowtimo_vnet(void);
121 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
122 		    /*const*/ struct mld_hdr *);
123 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static void	mld_v1_process_group_timer(struct in6_multi_head *,
126 		    struct in6_multi *);
127 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
128 static int	mld_v1_transmit_report(struct in6_multi *, const int);
129 static void	mld_v1_update_group(struct in6_multi *, const int);
130 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
131 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
132 static struct mbuf *
133 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
134 static int	mld_v2_enqueue_filter_change(struct mbufq *,
135 		    struct in6_multi *);
136 static int	mld_v2_enqueue_group_record(struct mbufq *,
137 		    struct in6_multi *, const int, const int, const int,
138 		    const int);
139 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
140 		    struct mbuf *, struct mldv2_query *, const int, const int);
141 static int	mld_v2_merge_state_changes(struct in6_multi *,
142 		    struct mbufq *);
143 static void	mld_v2_process_group_timers(struct in6_multi_head *,
144 		    struct mbufq *, struct mbufq *,
145 		    struct in6_multi *, const int);
146 static int	mld_v2_process_group_query(struct in6_multi *,
147 		    struct mld_ifsoftc *mli, int, struct mbuf *,
148 		    struct mldv2_query *, const int);
149 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
150 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
151 
152 /*
153  * Normative references: RFC 2710, RFC 3590, RFC 3810.
154  *
155  * Locking:
156  *  * The MLD subsystem lock ends up being system-wide for the moment,
157  *    but could be per-VIMAGE later on.
158  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
159  *    Any may be taken independently; if any are held at the same
160  *    time, the above lock order must be followed.
161  *  * IN6_MULTI_LOCK covers in_multi.
162  *  * MLD_LOCK covers per-link state and any global variables in this file.
163  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
164  *    per-link state iterators.
165  *
166  *  XXX LOR PREVENTION
167  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
168  *  will not accept an ifp; it wants an embedded scope ID, unlike
169  *  ip_output(), which happily takes the ifp given to it. The embedded
170  *  scope ID is only used by MLD to select the outgoing interface.
171  *
172  *  During interface attach and detach, MLD will take MLD_LOCK *after*
173  *  the IF_AFDATA_LOCK.
174  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
175  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
176  *  dispatch could work around this, but we'd rather not do that, as it
177  *  can introduce other races.
178  *
179  *  As such, we exploit the fact that the scope ID is just the interface
180  *  index, and embed it in the IPv6 destination address accordingly.
181  *  This is potentially NOT VALID for MLDv1 reports, as they
182  *  are always sent to the multicast group itself; as MLDv2
183  *  reports are always sent to ff02::16, this is not an issue
184  *  when MLDv2 is in use.
185  *
186  *  This does not however eliminate the LOR when ip6_output() itself
187  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
188  *  trigger a LOR warning in WITNESS when the ifnet is detached.
189  *
190  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
191  *  how it's used across the network stack. Here we're simply exploiting
192  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
193  *
194  * VIMAGE:
195  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
196  *    to a vnet in ifp->if_vnet.
197  */
198 static struct mtx		 mld_mtx;
199 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
200 
201 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
202 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
203 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
204 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
205 
206 /*
207  * VIMAGE-wide globals.
208  */
209 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
210 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
211 VNET_DEFINE_STATIC(int, interface_timers_running6);
212 VNET_DEFINE_STATIC(int, state_change_timers_running6);
213 VNET_DEFINE_STATIC(int, current_state_timers_running6);
214 
215 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
216 #define	V_mli_head			VNET(mli_head)
217 #define	V_interface_timers_running6	VNET(interface_timers_running6)
218 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
219 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
220 
221 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
222 
223 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
224     "IPv6 Multicast Listener Discovery");
225 
226 /*
227  * Virtualized sysctls.
228  */
229 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
230     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
231     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
232     "Rate limit for MLDv2 Group-and-Source queries in seconds");
233 
234 /*
235  * Non-virtualized sysctls.
236  */
237 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
238     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
239     "Per-interface MLDv2 state");
240 
241 static int	mld_v1enable = 1;
242 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
243     &mld_v1enable, 0, "Enable fallback to MLDv1");
244 
245 static int	mld_v2enable = 1;
246 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
247     &mld_v2enable, 0, "Enable MLDv2");
248 
249 static int	mld_use_allow = 1;
250 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
251     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
252 
253 /*
254  * Packed Router Alert option structure declaration.
255  */
256 struct mld_raopt {
257 	struct ip6_hbh		hbh;
258 	struct ip6_opt		pad;
259 	struct ip6_opt_router	ra;
260 } __packed;
261 
262 /*
263  * Router Alert hop-by-hop option header.
264  */
265 static struct mld_raopt mld_ra = {
266 	.hbh = { 0, 0 },
267 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
268 	.ra = {
269 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
270 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
271 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
272 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
273 	}
274 };
275 static struct ip6_pktopts mld_po;
276 
277 static __inline void
278 mld_save_context(struct mbuf *m, struct ifnet *ifp)
279 {
280 
281 #ifdef VIMAGE
282 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
283 #endif /* VIMAGE */
284 	m->m_pkthdr.rcvif = ifp;
285 	m->m_pkthdr.flowid = ifp->if_index;
286 }
287 
288 static __inline void
289 mld_scrub_context(struct mbuf *m)
290 {
291 
292 	m->m_pkthdr.PH_loc.ptr = NULL;
293 	m->m_pkthdr.flowid = 0;
294 }
295 
296 /*
297  * Restore context from a queued output chain.
298  * Return saved ifindex.
299  *
300  * VIMAGE: The assertion is there to make sure that we
301  * actually called CURVNET_SET() with what's in the mbuf chain.
302  */
303 static __inline uint32_t
304 mld_restore_context(struct mbuf *m)
305 {
306 
307 #if defined(VIMAGE) && defined(INVARIANTS)
308 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
309 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
310 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
311 #endif
312 	return (m->m_pkthdr.flowid);
313 }
314 
315 /*
316  * Retrieve or set threshold between group-source queries in seconds.
317  *
318  * VIMAGE: Assume curvnet set by caller.
319  * SMPng: NOTE: Serialized by MLD lock.
320  */
321 static int
322 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
323 {
324 	int error;
325 	int i;
326 
327 	error = sysctl_wire_old_buffer(req, sizeof(int));
328 	if (error)
329 		return (error);
330 
331 	MLD_LOCK();
332 
333 	i = V_mld_gsrdelay.tv_sec;
334 
335 	error = sysctl_handle_int(oidp, &i, 0, req);
336 	if (error || !req->newptr)
337 		goto out_locked;
338 
339 	if (i < -1 || i >= 60) {
340 		error = EINVAL;
341 		goto out_locked;
342 	}
343 
344 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
345 	     V_mld_gsrdelay.tv_sec, i);
346 	V_mld_gsrdelay.tv_sec = i;
347 
348 out_locked:
349 	MLD_UNLOCK();
350 	return (error);
351 }
352 
353 /*
354  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
355  * For use by ifmcstat(8).
356  *
357  * VIMAGE: Assume curvnet set by caller. The node handler itself
358  * is not directly virtualized.
359  */
360 static int
361 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
362 {
363 	struct epoch_tracker	 et;
364 	int			*name;
365 	int			 error;
366 	u_int			 namelen;
367 	struct ifnet		*ifp;
368 	struct mld_ifsoftc	*mli;
369 
370 	name = (int *)arg1;
371 	namelen = arg2;
372 
373 	if (req->newptr != NULL)
374 		return (EPERM);
375 
376 	if (namelen != 1)
377 		return (EINVAL);
378 
379 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
380 	if (error)
381 		return (error);
382 
383 	IN6_MULTI_LOCK();
384 	IN6_MULTI_LIST_LOCK();
385 	MLD_LOCK();
386 	NET_EPOCH_ENTER(et);
387 
388 	error = ENOENT;
389 	ifp = ifnet_byindex(name[0]);
390 	if (ifp == NULL)
391 		goto out_locked;
392 
393 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
394 		if (ifp == mli->mli_ifp) {
395 			struct mld_ifinfo info;
396 
397 			info.mli_version = mli->mli_version;
398 			info.mli_v1_timer = mli->mli_v1_timer;
399 			info.mli_v2_timer = mli->mli_v2_timer;
400 			info.mli_flags = mli->mli_flags;
401 			info.mli_rv = mli->mli_rv;
402 			info.mli_qi = mli->mli_qi;
403 			info.mli_qri = mli->mli_qri;
404 			info.mli_uri = mli->mli_uri;
405 			error = SYSCTL_OUT(req, &info, sizeof(info));
406 			break;
407 		}
408 	}
409 
410 out_locked:
411 	NET_EPOCH_EXIT(et);
412 	MLD_UNLOCK();
413 	IN6_MULTI_LIST_UNLOCK();
414 	IN6_MULTI_UNLOCK();
415 	return (error);
416 }
417 
418 /*
419  * Dispatch an entire queue of pending packet chains.
420  * VIMAGE: Assumes the vnet pointer has been set.
421  */
422 static void
423 mld_dispatch_queue(struct mbufq *mq, int limit)
424 {
425 	struct mbuf *m;
426 
427 	while ((m = mbufq_dequeue(mq)) != NULL) {
428 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
429 		mld_dispatch_packet(m);
430 		if (--limit == 0)
431 			break;
432 	}
433 }
434 
435 /*
436  * Filter outgoing MLD report state by group.
437  *
438  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
439  * and node-local addresses. However, kernel and socket consumers
440  * always embed the KAME scope ID in the address provided, so strip it
441  * when performing comparison.
442  * Note: This is not the same as the *multicast* scope.
443  *
444  * Return zero if the given group is one for which MLD reports
445  * should be suppressed, or non-zero if reports should be issued.
446  */
447 static __inline int
448 mld_is_addr_reported(const struct in6_addr *addr)
449 {
450 
451 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
452 
453 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
454 		return (0);
455 
456 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
457 		struct in6_addr tmp = *addr;
458 		in6_clearscope(&tmp);
459 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
460 			return (0);
461 	}
462 
463 	return (1);
464 }
465 
466 /*
467  * Attach MLD when PF_INET6 is attached to an interface.  Assumes that the
468  * current VNET is set by the caller.
469  */
470 struct mld_ifsoftc *
471 mld_domifattach(struct ifnet *ifp)
472 {
473 	struct mld_ifsoftc *mli;
474 
475 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp));
476 
477 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_WAITOK | M_ZERO);
478 	mli->mli_ifp = ifp;
479 	mli->mli_version = MLD_VERSION_2;
480 	mli->mli_flags = 0;
481 	mli->mli_rv = MLD_RV_INIT;
482 	mli->mli_qi = MLD_QI_INIT;
483 	mli->mli_qri = MLD_QRI_INIT;
484 	mli->mli_uri = MLD_URI_INIT;
485 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
486 	if ((ifp->if_flags & IFF_MULTICAST) == 0)
487 		mli->mli_flags |= MLIF_SILENT;
488 	if (mld_use_allow)
489 		mli->mli_flags |= MLIF_USEALLOW;
490 
491 	MLD_LOCK();
492 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
493 	MLD_UNLOCK();
494 
495 	return (mli);
496 }
497 
498 /*
499  * Hook for ifdetach.
500  *
501  * NOTE: Some finalization tasks need to run before the protocol domain
502  * is detached, but also before the link layer does its cleanup.
503  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
504  *
505  * SMPng: Caller must hold IN6_MULTI_LOCK().
506  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
507  * XXX This routine is also bitten by unlocked ifma_protospec access.
508  */
509 void
510 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
511 {
512 	struct epoch_tracker     et;
513 	struct mld_ifsoftc	*mli;
514 	struct ifmultiaddr	*ifma;
515 	struct in6_multi	*inm;
516 
517 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
518 	    if_name(ifp));
519 
520 	IN6_MULTI_LIST_LOCK_ASSERT();
521 	MLD_LOCK();
522 
523 	mli = MLD_IFINFO(ifp);
524 	IF_ADDR_WLOCK(ifp);
525 	/*
526 	 * Extract list of in6_multi associated with the detaching ifp
527 	 * which the PF_INET6 layer is about to release.
528 	 */
529 	NET_EPOCH_ENTER(et);
530 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
531 		inm = in6m_ifmultiaddr_get_inm(ifma);
532 		if (inm == NULL)
533 			continue;
534 		in6m_disconnect_locked(inmh, inm);
535 
536 		if (mli->mli_version == MLD_VERSION_2) {
537 			in6m_clear_recorded(inm);
538 
539 			/*
540 			 * We need to release the final reference held
541 			 * for issuing the INCLUDE {}.
542 			 */
543 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
544 				inm->in6m_state = MLD_NOT_MEMBER;
545 				in6m_rele_locked(inmh, inm);
546 			}
547 		}
548 	}
549 	NET_EPOCH_EXIT(et);
550 	IF_ADDR_WUNLOCK(ifp);
551 	MLD_UNLOCK();
552 }
553 
554 /*
555  * Hook for domifdetach.
556  * Runs after link-layer cleanup; free MLD state.
557  *
558  * SMPng: Normally called with IF_AFDATA_LOCK held.
559  */
560 void
561 mld_domifdetach(struct ifnet *ifp)
562 {
563 
564 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
565 	    __func__, ifp, if_name(ifp));
566 
567 	MLD_LOCK();
568 	mli_delete_locked(ifp);
569 	MLD_UNLOCK();
570 }
571 
572 static void
573 mli_delete_locked(const struct ifnet *ifp)
574 {
575 	struct mld_ifsoftc *mli, *tmli;
576 
577 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
578 	    __func__, ifp, if_name(ifp));
579 
580 	MLD_LOCK_ASSERT();
581 
582 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
583 		if (mli->mli_ifp == ifp) {
584 			/*
585 			 * Free deferred General Query responses.
586 			 */
587 			mbufq_drain(&mli->mli_gq);
588 
589 			LIST_REMOVE(mli, mli_link);
590 
591 			free(mli, M_MLD);
592 			return;
593 		}
594 	}
595 }
596 
597 /*
598  * Process a received MLDv1 general or address-specific query.
599  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
600  *
601  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
602  * mld_addr. This is OK as we own the mbuf chain.
603  */
604 static int
605 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
606     /*const*/ struct mld_hdr *mld)
607 {
608 	struct ifmultiaddr	*ifma;
609 	struct mld_ifsoftc	*mli;
610 	struct in6_multi	*inm;
611 	int			 is_general_query;
612 	uint16_t		 timer;
613 #ifdef KTR
614 	char			 ip6tbuf[INET6_ADDRSTRLEN];
615 #endif
616 
617 	NET_EPOCH_ASSERT();
618 
619 	is_general_query = 0;
620 
621 	if (!mld_v1enable) {
622 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
623 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
624 		    ifp, if_name(ifp));
625 		return (0);
626 	}
627 
628 	/*
629 	 * RFC3810 Section 6.2: MLD queries must originate from
630 	 * a router's link-local address.
631 	 */
632 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
633 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
634 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
635 		    ifp, if_name(ifp));
636 		return (0);
637 	}
638 
639 	/*
640 	 * Do address field validation upfront before we accept
641 	 * the query.
642 	 */
643 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
644 		/*
645 		 * MLDv1 General Query.
646 		 * If this was not sent to the all-nodes group, ignore it.
647 		 */
648 		struct in6_addr		 dst;
649 
650 		dst = ip6->ip6_dst;
651 		in6_clearscope(&dst);
652 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
653 			return (EINVAL);
654 		is_general_query = 1;
655 	} else {
656 		/*
657 		 * Embed scope ID of receiving interface in MLD query for
658 		 * lookup whilst we don't hold other locks.
659 		 */
660 		in6_setscope(&mld->mld_addr, ifp, NULL);
661 	}
662 
663 	IN6_MULTI_LIST_LOCK();
664 	MLD_LOCK();
665 
666 	/*
667 	 * Switch to MLDv1 host compatibility mode.
668 	 */
669 	mli = MLD_IFINFO(ifp);
670 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
671 	mld_set_version(mli, MLD_VERSION_1);
672 
673 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
674 	if (timer == 0)
675 		timer = 1;
676 
677 	if (is_general_query) {
678 		/*
679 		 * For each reporting group joined on this
680 		 * interface, kick the report timer.
681 		 */
682 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
683 			 ifp, if_name(ifp));
684 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
685 			inm = in6m_ifmultiaddr_get_inm(ifma);
686 			if (inm == NULL)
687 				continue;
688 			mld_v1_update_group(inm, timer);
689 		}
690 	} else {
691 		/*
692 		 * MLDv1 Group-Specific Query.
693 		 * If this is a group-specific MLDv1 query, we need only
694 		 * look up the single group to process it.
695 		 */
696 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
697 		if (inm != NULL) {
698 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
699 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
700 			    ifp, if_name(ifp));
701 			mld_v1_update_group(inm, timer);
702 		}
703 		/* XXX Clear embedded scope ID as userland won't expect it. */
704 		in6_clearscope(&mld->mld_addr);
705 	}
706 
707 	MLD_UNLOCK();
708 	IN6_MULTI_LIST_UNLOCK();
709 
710 	return (0);
711 }
712 
713 /*
714  * Update the report timer on a group in response to an MLDv1 query.
715  *
716  * If we are becoming the reporting member for this group, start the timer.
717  * If we already are the reporting member for this group, and timer is
718  * below the threshold, reset it.
719  *
720  * We may be updating the group for the first time since we switched
721  * to MLDv2. If we are, then we must clear any recorded source lists,
722  * and transition to REPORTING state; the group timer is overloaded
723  * for group and group-source query responses.
724  *
725  * Unlike MLDv2, the delay per group should be jittered
726  * to avoid bursts of MLDv1 reports.
727  */
728 static void
729 mld_v1_update_group(struct in6_multi *inm, const int timer)
730 {
731 #ifdef KTR
732 	char			 ip6tbuf[INET6_ADDRSTRLEN];
733 #endif
734 
735 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
736 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
737 	    if_name(inm->in6m_ifp), timer);
738 
739 	IN6_MULTI_LIST_LOCK_ASSERT();
740 
741 	switch (inm->in6m_state) {
742 	case MLD_NOT_MEMBER:
743 	case MLD_SILENT_MEMBER:
744 		break;
745 	case MLD_REPORTING_MEMBER:
746 		if (inm->in6m_timer != 0 &&
747 		    inm->in6m_timer <= timer) {
748 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
749 			    "skipping.", __func__);
750 			break;
751 		}
752 		/* FALLTHROUGH */
753 	case MLD_SG_QUERY_PENDING_MEMBER:
754 	case MLD_G_QUERY_PENDING_MEMBER:
755 	case MLD_IDLE_MEMBER:
756 	case MLD_LAZY_MEMBER:
757 	case MLD_AWAKENING_MEMBER:
758 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
759 		inm->in6m_state = MLD_REPORTING_MEMBER;
760 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
761 		V_current_state_timers_running6 = 1;
762 		break;
763 	case MLD_SLEEPING_MEMBER:
764 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
765 		inm->in6m_state = MLD_AWAKENING_MEMBER;
766 		break;
767 	case MLD_LEAVING_MEMBER:
768 		break;
769 	}
770 }
771 
772 /*
773  * Process a received MLDv2 general, group-specific or
774  * group-and-source-specific query.
775  *
776  * Assumes that mld points to a struct mldv2_query which is stored in
777  * contiguous memory.
778  *
779  * Return 0 if successful, otherwise an appropriate error code is returned.
780  */
781 static int
782 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
783     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
784 {
785 	struct mld_ifsoftc	*mli;
786 	struct in6_multi	*inm;
787 	uint32_t		 maxdelay, nsrc, qqi;
788 	int			 is_general_query;
789 	uint16_t		 timer;
790 	uint8_t			 qrv;
791 #ifdef KTR
792 	char			 ip6tbuf[INET6_ADDRSTRLEN];
793 #endif
794 
795 	NET_EPOCH_ASSERT();
796 
797 	if (!mld_v2enable) {
798 		CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
799 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
800 		    ifp, if_name(ifp));
801 		return (0);
802 	}
803 
804 	/*
805 	 * RFC3810 Section 6.2: MLD queries must originate from
806 	 * a router's link-local address.
807 	 */
808 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
809 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
810 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
811 		    ifp, if_name(ifp));
812 		return (0);
813 	}
814 
815 	is_general_query = 0;
816 
817 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
818 
819 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
820 	if (maxdelay >= 32768) {
821 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
822 			   (MLD_MRC_EXP(maxdelay) + 3);
823 	}
824 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
825 	if (timer == 0)
826 		timer = 1;
827 
828 	qrv = MLD_QRV(mld->mld_misc);
829 	if (qrv < 2) {
830 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
831 		    qrv, MLD_RV_INIT);
832 		qrv = MLD_RV_INIT;
833 	}
834 
835 	qqi = mld->mld_qqi;
836 	if (qqi >= 128) {
837 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
838 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
839 	}
840 
841 	nsrc = ntohs(mld->mld_numsrc);
842 	if (nsrc > MLD_MAX_GS_SOURCES)
843 		return (EMSGSIZE);
844 	if (icmp6len < sizeof(struct mldv2_query) +
845 	    (nsrc * sizeof(struct in6_addr)))
846 		return (EMSGSIZE);
847 
848 	/*
849 	 * Do further input validation upfront to avoid resetting timers
850 	 * should we need to discard this query.
851 	 */
852 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
853 		/*
854 		 * A general query with a source list has undefined
855 		 * behaviour; discard it.
856 		 */
857 		if (nsrc > 0)
858 			return (EINVAL);
859 		is_general_query = 1;
860 	} else {
861 		/*
862 		 * Embed scope ID of receiving interface in MLD query for
863 		 * lookup whilst we don't hold other locks (due to KAME
864 		 * locking lameness). We own this mbuf chain just now.
865 		 */
866 		in6_setscope(&mld->mld_addr, ifp, NULL);
867 	}
868 
869 	IN6_MULTI_LIST_LOCK();
870 	MLD_LOCK();
871 
872 	mli = MLD_IFINFO(ifp);
873 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
874 
875 	/*
876 	 * Discard the v2 query if we're in Compatibility Mode.
877 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
878 	 * until the Old Version Querier Present timer expires.
879 	 */
880 	if (mli->mli_version != MLD_VERSION_2)
881 		goto out_locked;
882 
883 	mld_set_version(mli, MLD_VERSION_2);
884 	mli->mli_rv = qrv;
885 	mli->mli_qi = qqi;
886 	mli->mli_qri = maxdelay;
887 
888 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
889 	    maxdelay);
890 
891 	if (is_general_query) {
892 		/*
893 		 * MLDv2 General Query.
894 		 *
895 		 * Schedule a current-state report on this ifp for
896 		 * all groups, possibly containing source lists.
897 		 *
898 		 * If there is a pending General Query response
899 		 * scheduled earlier than the selected delay, do
900 		 * not schedule any other reports.
901 		 * Otherwise, reset the interface timer.
902 		 */
903 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
904 		    ifp, if_name(ifp));
905 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
906 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
907 			V_interface_timers_running6 = 1;
908 		}
909 	} else {
910 		/*
911 		 * MLDv2 Group-specific or Group-and-source-specific Query.
912 		 *
913 		 * Group-source-specific queries are throttled on
914 		 * a per-group basis to defeat denial-of-service attempts.
915 		 * Queries for groups we are not a member of on this
916 		 * link are simply ignored.
917 		 */
918 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
919 		if (inm == NULL)
920 			goto out_locked;
921 		if (nsrc > 0) {
922 			if (!ratecheck(&inm->in6m_lastgsrtv,
923 			    &V_mld_gsrdelay)) {
924 				CTR1(KTR_MLD, "%s: GS query throttled.",
925 				    __func__);
926 				goto out_locked;
927 			}
928 		}
929 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
930 		     ifp, if_name(ifp));
931 		/*
932 		 * If there is a pending General Query response
933 		 * scheduled sooner than the selected delay, no
934 		 * further report need be scheduled.
935 		 * Otherwise, prepare to respond to the
936 		 * group-specific or group-and-source query.
937 		 */
938 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
939 			mld_v2_process_group_query(inm, mli, timer, m, mld, off);
940 
941 		/* XXX Clear embedded scope ID as userland won't expect it. */
942 		in6_clearscope(&mld->mld_addr);
943 	}
944 
945 out_locked:
946 	MLD_UNLOCK();
947 	IN6_MULTI_LIST_UNLOCK();
948 
949 	return (0);
950 }
951 
952 /*
953  * Process a received MLDv2 group-specific or group-and-source-specific
954  * query.
955  * Return <0 if any error occurred. Currently this is ignored.
956  */
957 static int
958 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
959     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
960 {
961 	int			 retval;
962 	uint16_t		 nsrc;
963 
964 	IN6_MULTI_LIST_LOCK_ASSERT();
965 	MLD_LOCK_ASSERT();
966 
967 	retval = 0;
968 
969 	switch (inm->in6m_state) {
970 	case MLD_NOT_MEMBER:
971 	case MLD_SILENT_MEMBER:
972 	case MLD_SLEEPING_MEMBER:
973 	case MLD_LAZY_MEMBER:
974 	case MLD_AWAKENING_MEMBER:
975 	case MLD_IDLE_MEMBER:
976 	case MLD_LEAVING_MEMBER:
977 		return (retval);
978 		break;
979 	case MLD_REPORTING_MEMBER:
980 	case MLD_G_QUERY_PENDING_MEMBER:
981 	case MLD_SG_QUERY_PENDING_MEMBER:
982 		break;
983 	}
984 
985 	nsrc = ntohs(mld->mld_numsrc);
986 
987 	/* Length should be checked by calling function. */
988 	KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
989 	    m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
990 	    nsrc * sizeof(struct in6_addr),
991 	    ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
992 	    m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
993 	    nsrc * sizeof(struct in6_addr), m0));
994 
995 	/*
996 	 * Deal with group-specific queries upfront.
997 	 * If any group query is already pending, purge any recorded
998 	 * source-list state if it exists, and schedule a query response
999 	 * for this group-specific query.
1000 	 */
1001 	if (nsrc == 0) {
1002 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1003 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1004 			in6m_clear_recorded(inm);
1005 			timer = min(inm->in6m_timer, timer);
1006 		}
1007 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1008 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1009 		V_current_state_timers_running6 = 1;
1010 		return (retval);
1011 	}
1012 
1013 	/*
1014 	 * Deal with the case where a group-and-source-specific query has
1015 	 * been received but a group-specific query is already pending.
1016 	 */
1017 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1018 		timer = min(inm->in6m_timer, timer);
1019 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1020 		V_current_state_timers_running6 = 1;
1021 		return (retval);
1022 	}
1023 
1024 	/*
1025 	 * Finally, deal with the case where a group-and-source-specific
1026 	 * query has been received, where a response to a previous g-s-r
1027 	 * query exists, or none exists.
1028 	 * In this case, we need to parse the source-list which the Querier
1029 	 * has provided us with and check if we have any source list filter
1030 	 * entries at T1 for these sources. If we do not, there is no need
1031 	 * schedule a report and the query may be dropped.
1032 	 * If we do, we must record them and schedule a current-state
1033 	 * report for those sources.
1034 	 */
1035 	if (inm->in6m_nsrc > 0) {
1036 		struct in6_addr		 srcaddr;
1037 		int			 i, nrecorded;
1038 		int			 soff;
1039 
1040 		soff = off + sizeof(struct mldv2_query);
1041 		nrecorded = 0;
1042 		for (i = 0; i < nsrc; i++) {
1043 			m_copydata(m0, soff, sizeof(struct in6_addr),
1044 			    (caddr_t)&srcaddr);
1045 			retval = in6m_record_source(inm, &srcaddr);
1046 			if (retval < 0)
1047 				break;
1048 			nrecorded += retval;
1049 			soff += sizeof(struct in6_addr);
1050 		}
1051 		if (nrecorded > 0) {
1052 			CTR1(KTR_MLD,
1053 			    "%s: schedule response to SG query", __func__);
1054 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1055 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1056 			V_current_state_timers_running6 = 1;
1057 		}
1058 	}
1059 
1060 	return (retval);
1061 }
1062 
1063 /*
1064  * Process a received MLDv1 host membership report.
1065  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1066  *
1067  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1068  * mld_addr. This is OK as we own the mbuf chain.
1069  */
1070 static int
1071 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1072     /*const*/ struct mld_hdr *mld)
1073 {
1074 	struct in6_addr		 src, dst;
1075 	struct in6_ifaddr	*ia;
1076 	struct in6_multi	*inm;
1077 #ifdef KTR
1078 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1079 #endif
1080 
1081 	NET_EPOCH_ASSERT();
1082 
1083 	if (!mld_v1enable) {
1084 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1085 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1086 		    ifp, if_name(ifp));
1087 		return (0);
1088 	}
1089 
1090 	if (ifp->if_flags & IFF_LOOPBACK)
1091 		return (0);
1092 
1093 	/*
1094 	 * MLDv1 reports must originate from a host's link-local address,
1095 	 * or the unspecified address (when booting).
1096 	 */
1097 	src = ip6->ip6_src;
1098 	in6_clearscope(&src);
1099 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1100 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1101 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1102 		    ifp, if_name(ifp));
1103 		return (EINVAL);
1104 	}
1105 
1106 	/*
1107 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1108 	 * group, and must be directed to the group itself.
1109 	 */
1110 	dst = ip6->ip6_dst;
1111 	in6_clearscope(&dst);
1112 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1113 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1114 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1115 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1116 		    ifp, if_name(ifp));
1117 		return (EINVAL);
1118 	}
1119 
1120 	/*
1121 	 * Make sure we don't hear our own membership report, as fast
1122 	 * leave requires knowing that we are the only member of a
1123 	 * group. Assume we used the link-local address if available,
1124 	 * otherwise look for ::.
1125 	 *
1126 	 * XXX Note that scope ID comparison is needed for the address
1127 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1128 	 * performed for the on-wire address.
1129 	 */
1130 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1131 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1132 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1133 		if (ia != NULL)
1134 			ifa_free(&ia->ia_ifa);
1135 		return (0);
1136 	}
1137 	if (ia != NULL)
1138 		ifa_free(&ia->ia_ifa);
1139 
1140 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1141 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1142 
1143 	/*
1144 	 * Embed scope ID of receiving interface in MLD query for lookup
1145 	 * whilst we don't hold other locks (due to KAME locking lameness).
1146 	 */
1147 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1148 		in6_setscope(&mld->mld_addr, ifp, NULL);
1149 
1150 	IN6_MULTI_LIST_LOCK();
1151 	MLD_LOCK();
1152 
1153 	/*
1154 	 * MLDv1 report suppression.
1155 	 * If we are a member of this group, and our membership should be
1156 	 * reported, and our group timer is pending or about to be reset,
1157 	 * stop our group timer by transitioning to the 'lazy' state.
1158 	 */
1159 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1160 	if (inm != NULL) {
1161 		struct mld_ifsoftc *mli;
1162 
1163 		mli = inm->in6m_mli;
1164 		KASSERT(mli != NULL,
1165 		    ("%s: no mli for ifp %p", __func__, ifp));
1166 
1167 		/*
1168 		 * If we are in MLDv2 host mode, do not allow the
1169 		 * other host's MLDv1 report to suppress our reports.
1170 		 */
1171 		if (mli->mli_version == MLD_VERSION_2)
1172 			goto out_locked;
1173 
1174 		inm->in6m_timer = 0;
1175 
1176 		switch (inm->in6m_state) {
1177 		case MLD_NOT_MEMBER:
1178 		case MLD_SILENT_MEMBER:
1179 		case MLD_SLEEPING_MEMBER:
1180 			break;
1181 		case MLD_REPORTING_MEMBER:
1182 		case MLD_IDLE_MEMBER:
1183 		case MLD_AWAKENING_MEMBER:
1184 			CTR3(KTR_MLD,
1185 			    "report suppressed for %s on ifp %p(%s)",
1186 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1187 			    ifp, if_name(ifp));
1188 		case MLD_LAZY_MEMBER:
1189 			inm->in6m_state = MLD_LAZY_MEMBER;
1190 			break;
1191 		case MLD_G_QUERY_PENDING_MEMBER:
1192 		case MLD_SG_QUERY_PENDING_MEMBER:
1193 		case MLD_LEAVING_MEMBER:
1194 			break;
1195 		}
1196 	}
1197 
1198 out_locked:
1199 	MLD_UNLOCK();
1200 	IN6_MULTI_LIST_UNLOCK();
1201 
1202 	/* XXX Clear embedded scope ID as userland won't expect it. */
1203 	in6_clearscope(&mld->mld_addr);
1204 
1205 	return (0);
1206 }
1207 
1208 /*
1209  * MLD input path.
1210  *
1211  * Assume query messages which fit in a single ICMPv6 message header
1212  * have been pulled up.
1213  * Assume that userland will want to see the message, even if it
1214  * otherwise fails kernel input validation; do not free it.
1215  * Pullup may however free the mbuf chain m if it fails.
1216  *
1217  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1218  */
1219 int
1220 mld_input(struct mbuf **mp, int off, int icmp6len)
1221 {
1222 	struct ifnet	*ifp;
1223 	struct ip6_hdr	*ip6;
1224 	struct mbuf	*m;
1225 	struct mld_hdr	*mld;
1226 	int		 mldlen;
1227 
1228 	m = *mp;
1229 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1230 
1231 	ifp = m->m_pkthdr.rcvif;
1232 
1233 	/* Pullup to appropriate size. */
1234 	if (m->m_len < off + sizeof(*mld)) {
1235 		m = m_pullup(m, off + sizeof(*mld));
1236 		if (m == NULL) {
1237 			ICMP6STAT_INC(icp6s_badlen);
1238 			return (IPPROTO_DONE);
1239 		}
1240 	}
1241 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1242 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1243 	    icmp6len >= sizeof(struct mldv2_query)) {
1244 		mldlen = sizeof(struct mldv2_query);
1245 	} else {
1246 		mldlen = sizeof(struct mld_hdr);
1247 	}
1248 	if (m->m_len < off + mldlen) {
1249 		m = m_pullup(m, off + mldlen);
1250 		if (m == NULL) {
1251 			ICMP6STAT_INC(icp6s_badlen);
1252 			return (IPPROTO_DONE);
1253 		}
1254 	}
1255 	*mp = m;
1256 	ip6 = mtod(m, struct ip6_hdr *);
1257 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1258 
1259 	/*
1260 	 * Userland needs to see all of this traffic for implementing
1261 	 * the endpoint discovery portion of multicast routing.
1262 	 */
1263 	switch (mld->mld_type) {
1264 	case MLD_LISTENER_QUERY:
1265 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1266 		if (icmp6len == sizeof(struct mld_hdr)) {
1267 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1268 				return (0);
1269 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1270 			if (mld_v2_input_query(ifp, ip6, m,
1271 			    (struct mldv2_query *)mld, off, icmp6len) != 0)
1272 				return (0);
1273 		}
1274 		break;
1275 	case MLD_LISTENER_REPORT:
1276 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1277 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1278 			return (0);
1279 		break;
1280 	case MLDV2_LISTENER_REPORT:
1281 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1282 		break;
1283 	case MLD_LISTENER_DONE:
1284 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1285 		break;
1286 	default:
1287 		break;
1288 	}
1289 
1290 	return (0);
1291 }
1292 
1293 /*
1294  * Fast timeout handler (global).
1295  * VIMAGE: Timeout handlers are expected to service all vimages.
1296  */
1297 void
1298 mld_fasttimo(void)
1299 {
1300 	struct in6_multi_head inmh;
1301 	VNET_ITERATOR_DECL(vnet_iter);
1302 
1303 	SLIST_INIT(&inmh);
1304 
1305 	VNET_LIST_RLOCK_NOSLEEP();
1306 	VNET_FOREACH(vnet_iter) {
1307 		CURVNET_SET(vnet_iter);
1308 		mld_fasttimo_vnet(&inmh);
1309 		CURVNET_RESTORE();
1310 	}
1311 	VNET_LIST_RUNLOCK_NOSLEEP();
1312 	in6m_release_list_deferred(&inmh);
1313 }
1314 
1315 /*
1316  * Fast timeout handler (per-vnet).
1317  *
1318  * VIMAGE: Assume caller has set up our curvnet.
1319  */
1320 static void
1321 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1322 {
1323 	struct epoch_tracker     et;
1324 	struct mbufq		 scq;	/* State-change packets */
1325 	struct mbufq		 qrq;	/* Query response packets */
1326 	struct ifnet		*ifp;
1327 	struct mld_ifsoftc	*mli;
1328 	struct ifmultiaddr	*ifma;
1329 	struct in6_multi	*inm;
1330 	int			 uri_fasthz;
1331 
1332 	uri_fasthz = 0;
1333 
1334 	/*
1335 	 * Quick check to see if any work needs to be done, in order to
1336 	 * minimize the overhead of fasttimo processing.
1337 	 * SMPng: XXX Unlocked reads.
1338 	 */
1339 	if (!V_current_state_timers_running6 &&
1340 	    !V_interface_timers_running6 &&
1341 	    !V_state_change_timers_running6)
1342 		return;
1343 
1344 	IN6_MULTI_LIST_LOCK();
1345 	MLD_LOCK();
1346 
1347 	/*
1348 	 * MLDv2 General Query response timer processing.
1349 	 */
1350 	if (V_interface_timers_running6) {
1351 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1352 
1353 		V_interface_timers_running6 = 0;
1354 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1355 			if (mli->mli_v2_timer == 0) {
1356 				/* Do nothing. */
1357 			} else if (--mli->mli_v2_timer == 0) {
1358 				mld_v2_dispatch_general_query(mli);
1359 			} else {
1360 				V_interface_timers_running6 = 1;
1361 			}
1362 		}
1363 	}
1364 
1365 	if (!V_current_state_timers_running6 &&
1366 	    !V_state_change_timers_running6)
1367 		goto out_locked;
1368 
1369 	V_current_state_timers_running6 = 0;
1370 	V_state_change_timers_running6 = 0;
1371 
1372 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1373 
1374 	/*
1375 	 * MLD host report and state-change timer processing.
1376 	 * Note: Processing a v2 group timer may remove a node.
1377 	 */
1378 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1379 		ifp = mli->mli_ifp;
1380 
1381 		if (mli->mli_version == MLD_VERSION_2) {
1382 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1383 			    PR_FASTHZ);
1384 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1385 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1386 		}
1387 
1388 		NET_EPOCH_ENTER(et);
1389 		IF_ADDR_WLOCK(ifp);
1390 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1391 			inm = in6m_ifmultiaddr_get_inm(ifma);
1392 			if (inm == NULL)
1393 				continue;
1394 			switch (mli->mli_version) {
1395 			case MLD_VERSION_1:
1396 				mld_v1_process_group_timer(inmh, inm);
1397 				break;
1398 			case MLD_VERSION_2:
1399 				mld_v2_process_group_timers(inmh, &qrq,
1400 				    &scq, inm, uri_fasthz);
1401 				break;
1402 			}
1403 		}
1404 		IF_ADDR_WUNLOCK(ifp);
1405 
1406 		switch (mli->mli_version) {
1407 		case MLD_VERSION_1:
1408 			/*
1409 			 * Transmit reports for this lifecycle.  This
1410 			 * is done while not holding IF_ADDR_LOCK
1411 			 * since this can call
1412 			 * in6ifa_ifpforlinklocal() which locks
1413 			 * IF_ADDR_LOCK internally as well as
1414 			 * ip6_output() to transmit a packet.
1415 			 */
1416 			while ((inm = SLIST_FIRST(inmh)) != NULL) {
1417 				SLIST_REMOVE_HEAD(inmh, in6m_defer);
1418 				(void)mld_v1_transmit_report(inm,
1419 				    MLD_LISTENER_REPORT);
1420 			}
1421 			break;
1422 		case MLD_VERSION_2:
1423 			mld_dispatch_queue(&qrq, 0);
1424 			mld_dispatch_queue(&scq, 0);
1425 			break;
1426 		}
1427 		NET_EPOCH_EXIT(et);
1428 	}
1429 
1430 out_locked:
1431 	MLD_UNLOCK();
1432 	IN6_MULTI_LIST_UNLOCK();
1433 }
1434 
1435 /*
1436  * Update host report group timer.
1437  * Will update the global pending timer flags.
1438  */
1439 static void
1440 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1441 {
1442 	int report_timer_expired;
1443 
1444 	IN6_MULTI_LIST_LOCK_ASSERT();
1445 	MLD_LOCK_ASSERT();
1446 
1447 	if (inm->in6m_timer == 0) {
1448 		report_timer_expired = 0;
1449 	} else if (--inm->in6m_timer == 0) {
1450 		report_timer_expired = 1;
1451 	} else {
1452 		V_current_state_timers_running6 = 1;
1453 		return;
1454 	}
1455 
1456 	switch (inm->in6m_state) {
1457 	case MLD_NOT_MEMBER:
1458 	case MLD_SILENT_MEMBER:
1459 	case MLD_IDLE_MEMBER:
1460 	case MLD_LAZY_MEMBER:
1461 	case MLD_SLEEPING_MEMBER:
1462 	case MLD_AWAKENING_MEMBER:
1463 		break;
1464 	case MLD_REPORTING_MEMBER:
1465 		if (report_timer_expired) {
1466 			inm->in6m_state = MLD_IDLE_MEMBER;
1467 			SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1468 		}
1469 		break;
1470 	case MLD_G_QUERY_PENDING_MEMBER:
1471 	case MLD_SG_QUERY_PENDING_MEMBER:
1472 	case MLD_LEAVING_MEMBER:
1473 		break;
1474 	}
1475 }
1476 
1477 /*
1478  * Update a group's timers for MLDv2.
1479  * Will update the global pending timer flags.
1480  * Note: Unlocked read from mli.
1481  */
1482 static void
1483 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1484     struct mbufq *qrq, struct mbufq *scq,
1485     struct in6_multi *inm, const int uri_fasthz)
1486 {
1487 	int query_response_timer_expired;
1488 	int state_change_retransmit_timer_expired;
1489 #ifdef KTR
1490 	char ip6tbuf[INET6_ADDRSTRLEN];
1491 #endif
1492 
1493 	IN6_MULTI_LIST_LOCK_ASSERT();
1494 	MLD_LOCK_ASSERT();
1495 
1496 	query_response_timer_expired = 0;
1497 	state_change_retransmit_timer_expired = 0;
1498 
1499 	/*
1500 	 * During a transition from compatibility mode back to MLDv2,
1501 	 * a group record in REPORTING state may still have its group
1502 	 * timer active. This is a no-op in this function; it is easier
1503 	 * to deal with it here than to complicate the slow-timeout path.
1504 	 */
1505 	if (inm->in6m_timer == 0) {
1506 		query_response_timer_expired = 0;
1507 	} else if (--inm->in6m_timer == 0) {
1508 		query_response_timer_expired = 1;
1509 	} else {
1510 		V_current_state_timers_running6 = 1;
1511 	}
1512 
1513 	if (inm->in6m_sctimer == 0) {
1514 		state_change_retransmit_timer_expired = 0;
1515 	} else if (--inm->in6m_sctimer == 0) {
1516 		state_change_retransmit_timer_expired = 1;
1517 	} else {
1518 		V_state_change_timers_running6 = 1;
1519 	}
1520 
1521 	/* We are in fasttimo, so be quick about it. */
1522 	if (!state_change_retransmit_timer_expired &&
1523 	    !query_response_timer_expired)
1524 		return;
1525 
1526 	switch (inm->in6m_state) {
1527 	case MLD_NOT_MEMBER:
1528 	case MLD_SILENT_MEMBER:
1529 	case MLD_SLEEPING_MEMBER:
1530 	case MLD_LAZY_MEMBER:
1531 	case MLD_AWAKENING_MEMBER:
1532 	case MLD_IDLE_MEMBER:
1533 		break;
1534 	case MLD_G_QUERY_PENDING_MEMBER:
1535 	case MLD_SG_QUERY_PENDING_MEMBER:
1536 		/*
1537 		 * Respond to a previously pending Group-Specific
1538 		 * or Group-and-Source-Specific query by enqueueing
1539 		 * the appropriate Current-State report for
1540 		 * immediate transmission.
1541 		 */
1542 		if (query_response_timer_expired) {
1543 			int retval __unused;
1544 
1545 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1546 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1547 			    0);
1548 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1549 			    __func__, retval);
1550 			inm->in6m_state = MLD_REPORTING_MEMBER;
1551 			in6m_clear_recorded(inm);
1552 		}
1553 		/* FALLTHROUGH */
1554 	case MLD_REPORTING_MEMBER:
1555 	case MLD_LEAVING_MEMBER:
1556 		if (state_change_retransmit_timer_expired) {
1557 			/*
1558 			 * State-change retransmission timer fired.
1559 			 * If there are any further pending retransmissions,
1560 			 * set the global pending state-change flag, and
1561 			 * reset the timer.
1562 			 */
1563 			if (--inm->in6m_scrv > 0) {
1564 				inm->in6m_sctimer = uri_fasthz;
1565 				V_state_change_timers_running6 = 1;
1566 			}
1567 			/*
1568 			 * Retransmit the previously computed state-change
1569 			 * report. If there are no further pending
1570 			 * retransmissions, the mbuf queue will be consumed.
1571 			 * Update T0 state to T1 as we have now sent
1572 			 * a state-change.
1573 			 */
1574 			(void)mld_v2_merge_state_changes(inm, scq);
1575 
1576 			in6m_commit(inm);
1577 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1578 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1579 			    if_name(inm->in6m_ifp));
1580 
1581 			/*
1582 			 * If we are leaving the group for good, make sure
1583 			 * we release MLD's reference to it.
1584 			 * This release must be deferred using a SLIST,
1585 			 * as we are called from a loop which traverses
1586 			 * the in_ifmultiaddr TAILQ.
1587 			 */
1588 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1589 			    inm->in6m_scrv == 0) {
1590 				inm->in6m_state = MLD_NOT_MEMBER;
1591 				in6m_disconnect_locked(inmh, inm);
1592 				in6m_rele_locked(inmh, inm);
1593 			}
1594 		}
1595 		break;
1596 	}
1597 }
1598 
1599 /*
1600  * Switch to a different version on the given interface,
1601  * as per Section 9.12.
1602  */
1603 static void
1604 mld_set_version(struct mld_ifsoftc *mli, const int version)
1605 {
1606 	int old_version_timer;
1607 
1608 	MLD_LOCK_ASSERT();
1609 
1610 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1611 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1612 
1613 	if (version == MLD_VERSION_1) {
1614 		/*
1615 		 * Compute the "Older Version Querier Present" timer as per
1616 		 * Section 9.12.
1617 		 */
1618 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1619 		old_version_timer *= PR_SLOWHZ;
1620 		mli->mli_v1_timer = old_version_timer;
1621 	}
1622 
1623 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1624 		mli->mli_version = MLD_VERSION_1;
1625 		mld_v2_cancel_link_timers(mli);
1626 	}
1627 }
1628 
1629 /*
1630  * Cancel pending MLDv2 timers for the given link and all groups
1631  * joined on it; state-change, general-query, and group-query timers.
1632  */
1633 static void
1634 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1635 {
1636 	struct epoch_tracker	 et;
1637 	struct in6_multi_head	 inmh;
1638 	struct ifmultiaddr	*ifma;
1639 	struct ifnet		*ifp;
1640 	struct in6_multi	*inm;
1641 
1642 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1643 	    mli->mli_ifp, if_name(mli->mli_ifp));
1644 
1645 	SLIST_INIT(&inmh);
1646 	IN6_MULTI_LIST_LOCK_ASSERT();
1647 	MLD_LOCK_ASSERT();
1648 
1649 	/*
1650 	 * Fast-track this potentially expensive operation
1651 	 * by checking all the global 'timer pending' flags.
1652 	 */
1653 	if (!V_interface_timers_running6 &&
1654 	    !V_state_change_timers_running6 &&
1655 	    !V_current_state_timers_running6)
1656 		return;
1657 
1658 	mli->mli_v2_timer = 0;
1659 
1660 	ifp = mli->mli_ifp;
1661 
1662 	IF_ADDR_WLOCK(ifp);
1663 	NET_EPOCH_ENTER(et);
1664 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1665 		inm = in6m_ifmultiaddr_get_inm(ifma);
1666 		if (inm == NULL)
1667 			continue;
1668 		switch (inm->in6m_state) {
1669 		case MLD_NOT_MEMBER:
1670 		case MLD_SILENT_MEMBER:
1671 		case MLD_IDLE_MEMBER:
1672 		case MLD_LAZY_MEMBER:
1673 		case MLD_SLEEPING_MEMBER:
1674 		case MLD_AWAKENING_MEMBER:
1675 			break;
1676 		case MLD_LEAVING_MEMBER:
1677 			/*
1678 			 * If we are leaving the group and switching
1679 			 * version, we need to release the final
1680 			 * reference held for issuing the INCLUDE {}.
1681 			 */
1682 			if (inm->in6m_refcount == 1)
1683 				in6m_disconnect_locked(&inmh, inm);
1684 			in6m_rele_locked(&inmh, inm);
1685 			/* FALLTHROUGH */
1686 		case MLD_G_QUERY_PENDING_MEMBER:
1687 		case MLD_SG_QUERY_PENDING_MEMBER:
1688 			in6m_clear_recorded(inm);
1689 			/* FALLTHROUGH */
1690 		case MLD_REPORTING_MEMBER:
1691 			inm->in6m_sctimer = 0;
1692 			inm->in6m_timer = 0;
1693 			inm->in6m_state = MLD_REPORTING_MEMBER;
1694 			/*
1695 			 * Free any pending MLDv2 state-change records.
1696 			 */
1697 			mbufq_drain(&inm->in6m_scq);
1698 			break;
1699 		}
1700 	}
1701 	NET_EPOCH_EXIT(et);
1702 	IF_ADDR_WUNLOCK(ifp);
1703 	in6m_release_list_deferred(&inmh);
1704 }
1705 
1706 /*
1707  * Global slowtimo handler.
1708  * VIMAGE: Timeout handlers are expected to service all vimages.
1709  */
1710 void
1711 mld_slowtimo(void)
1712 {
1713 	VNET_ITERATOR_DECL(vnet_iter);
1714 
1715 	VNET_LIST_RLOCK_NOSLEEP();
1716 	VNET_FOREACH(vnet_iter) {
1717 		CURVNET_SET(vnet_iter);
1718 		mld_slowtimo_vnet();
1719 		CURVNET_RESTORE();
1720 	}
1721 	VNET_LIST_RUNLOCK_NOSLEEP();
1722 }
1723 
1724 /*
1725  * Per-vnet slowtimo handler.
1726  */
1727 static void
1728 mld_slowtimo_vnet(void)
1729 {
1730 	struct mld_ifsoftc *mli;
1731 
1732 	MLD_LOCK();
1733 
1734 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1735 		mld_v1_process_querier_timers(mli);
1736 	}
1737 
1738 	MLD_UNLOCK();
1739 }
1740 
1741 /*
1742  * Update the Older Version Querier Present timers for a link.
1743  * See Section 9.12 of RFC 3810.
1744  */
1745 static void
1746 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1747 {
1748 
1749 	MLD_LOCK_ASSERT();
1750 
1751 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1752 		/*
1753 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1754 		 */
1755 		CTR5(KTR_MLD,
1756 		    "%s: transition from v%d -> v%d on %p(%s)",
1757 		    __func__, mli->mli_version, MLD_VERSION_2,
1758 		    mli->mli_ifp, if_name(mli->mli_ifp));
1759 		mli->mli_version = MLD_VERSION_2;
1760 	}
1761 }
1762 
1763 /*
1764  * Transmit an MLDv1 report immediately.
1765  */
1766 static int
1767 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1768 {
1769 	struct ifnet		*ifp;
1770 	struct in6_ifaddr	*ia;
1771 	struct ip6_hdr		*ip6;
1772 	struct mbuf		*mh, *md;
1773 	struct mld_hdr		*mld;
1774 
1775 	NET_EPOCH_ASSERT();
1776 	IN6_MULTI_LIST_LOCK_ASSERT();
1777 	MLD_LOCK_ASSERT();
1778 
1779 	ifp = in6m->in6m_ifp;
1780 	/* in process of being freed */
1781 	if (ifp == NULL)
1782 		return (0);
1783 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1784 	/* ia may be NULL if link-local address is tentative. */
1785 
1786 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1787 	if (mh == NULL) {
1788 		if (ia != NULL)
1789 			ifa_free(&ia->ia_ifa);
1790 		return (ENOMEM);
1791 	}
1792 	md = m_get(M_NOWAIT, MT_DATA);
1793 	if (md == NULL) {
1794 		m_free(mh);
1795 		if (ia != NULL)
1796 			ifa_free(&ia->ia_ifa);
1797 		return (ENOMEM);
1798 	}
1799 	mh->m_next = md;
1800 
1801 	/*
1802 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1803 	 * that ether_output() does not need to allocate another mbuf
1804 	 * for the header in the most common case.
1805 	 */
1806 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1807 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1808 	mh->m_len = sizeof(struct ip6_hdr);
1809 
1810 	ip6 = mtod(mh, struct ip6_hdr *);
1811 	ip6->ip6_flow = 0;
1812 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1813 	ip6->ip6_vfc |= IPV6_VERSION;
1814 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1815 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1816 	ip6->ip6_dst = in6m->in6m_addr;
1817 
1818 	md->m_len = sizeof(struct mld_hdr);
1819 	mld = mtod(md, struct mld_hdr *);
1820 	mld->mld_type = type;
1821 	mld->mld_code = 0;
1822 	mld->mld_cksum = 0;
1823 	mld->mld_maxdelay = 0;
1824 	mld->mld_reserved = 0;
1825 	mld->mld_addr = in6m->in6m_addr;
1826 	in6_clearscope(&mld->mld_addr);
1827 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1828 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1829 
1830 	mld_save_context(mh, ifp);
1831 	mh->m_flags |= M_MLDV1;
1832 
1833 	mld_dispatch_packet(mh);
1834 
1835 	if (ia != NULL)
1836 		ifa_free(&ia->ia_ifa);
1837 	return (0);
1838 }
1839 
1840 /*
1841  * Process a state change from the upper layer for the given IPv6 group.
1842  *
1843  * Each socket holds a reference on the in_multi in its own ip_moptions.
1844  * The socket layer will have made the necessary updates to.the group
1845  * state, it is now up to MLD to issue a state change report if there
1846  * has been any change between T0 (when the last state-change was issued)
1847  * and T1 (now).
1848  *
1849  * We use the MLDv2 state machine at group level. The MLd module
1850  * however makes the decision as to which MLD protocol version to speak.
1851  * A state change *from* INCLUDE {} always means an initial join.
1852  * A state change *to* INCLUDE {} always means a final leave.
1853  *
1854  * If delay is non-zero, and the state change is an initial multicast
1855  * join, the state change report will be delayed by 'delay' ticks
1856  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1857  * the initial MLDv2 state change report will be delayed by whichever
1858  * is sooner, a pending state-change timer or delay itself.
1859  *
1860  * VIMAGE: curvnet should have been set by caller, as this routine
1861  * is called from the socket option handlers.
1862  */
1863 int
1864 mld_change_state(struct in6_multi *inm, const int delay)
1865 {
1866 	struct mld_ifsoftc *mli;
1867 	struct ifnet *ifp;
1868 	int error;
1869 
1870 	IN6_MULTI_LIST_LOCK_ASSERT();
1871 
1872 	error = 0;
1873 
1874 	/*
1875 	 * Check if the in6_multi has already been disconnected.
1876 	 */
1877 	if (inm->in6m_ifp == NULL) {
1878 		CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1879 		return (0);
1880 	}
1881 
1882 	/*
1883 	 * Try to detect if the upper layer just asked us to change state
1884 	 * for an interface which has now gone away.
1885 	 */
1886 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1887 	ifp = inm->in6m_ifma->ifma_ifp;
1888 	if (ifp == NULL)
1889 		return (0);
1890 	/*
1891 	 * Sanity check that netinet6's notion of ifp is the
1892 	 * same as net's.
1893 	 */
1894 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1895 
1896 	MLD_LOCK();
1897 	mli = MLD_IFINFO(ifp);
1898 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1899 
1900 	/*
1901 	 * If we detect a state transition to or from MCAST_UNDEFINED
1902 	 * for this group, then we are starting or finishing an MLD
1903 	 * life cycle for this group.
1904 	 */
1905 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1906 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1907 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1908 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1909 			CTR1(KTR_MLD, "%s: initial join", __func__);
1910 			error = mld_initial_join(inm, mli, delay);
1911 			goto out_locked;
1912 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1913 			CTR1(KTR_MLD, "%s: final leave", __func__);
1914 			mld_final_leave(inm, mli);
1915 			goto out_locked;
1916 		}
1917 	} else {
1918 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1919 	}
1920 
1921 	error = mld_handle_state_change(inm, mli);
1922 
1923 out_locked:
1924 	MLD_UNLOCK();
1925 	return (error);
1926 }
1927 
1928 /*
1929  * Perform the initial join for an MLD group.
1930  *
1931  * When joining a group:
1932  *  If the group should have its MLD traffic suppressed, do nothing.
1933  *  MLDv1 starts sending MLDv1 host membership reports.
1934  *  MLDv2 will schedule an MLDv2 state-change report containing the
1935  *  initial state of the membership.
1936  *
1937  * If the delay argument is non-zero, then we must delay sending the
1938  * initial state change for delay ticks (in units of PR_FASTHZ).
1939  */
1940 static int
1941 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1942     const int delay)
1943 {
1944 	struct epoch_tracker     et;
1945 	struct ifnet		*ifp;
1946 	struct mbufq		*mq;
1947 	int			 error, retval, syncstates;
1948 	int			 odelay;
1949 #ifdef KTR
1950 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1951 #endif
1952 
1953 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1954 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1955 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1956 
1957 	error = 0;
1958 	syncstates = 1;
1959 
1960 	ifp = inm->in6m_ifp;
1961 
1962 	IN6_MULTI_LIST_LOCK_ASSERT();
1963 	MLD_LOCK_ASSERT();
1964 
1965 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1966 
1967 	/*
1968 	 * Groups joined on loopback or marked as 'not reported',
1969 	 * enter the MLD_SILENT_MEMBER state and
1970 	 * are never reported in any protocol exchanges.
1971 	 * All other groups enter the appropriate state machine
1972 	 * for the version in use on this link.
1973 	 * A link marked as MLIF_SILENT causes MLD to be completely
1974 	 * disabled for the link.
1975 	 */
1976 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1977 	    (mli->mli_flags & MLIF_SILENT) ||
1978 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1979 		CTR1(KTR_MLD,
1980 "%s: not kicking state machine for silent group", __func__);
1981 		inm->in6m_state = MLD_SILENT_MEMBER;
1982 		inm->in6m_timer = 0;
1983 	} else {
1984 		/*
1985 		 * Deal with overlapping in_multi lifecycle.
1986 		 * If this group was LEAVING, then make sure
1987 		 * we drop the reference we picked up to keep the
1988 		 * group around for the final INCLUDE {} enqueue.
1989 		 */
1990 		if (mli->mli_version == MLD_VERSION_2 &&
1991 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
1992 			inm->in6m_refcount--;
1993 			MPASS(inm->in6m_refcount > 0);
1994 		}
1995 		inm->in6m_state = MLD_REPORTING_MEMBER;
1996 
1997 		switch (mli->mli_version) {
1998 		case MLD_VERSION_1:
1999 			/*
2000 			 * If a delay was provided, only use it if
2001 			 * it is greater than the delay normally
2002 			 * used for an MLDv1 state change report,
2003 			 * and delay sending the initial MLDv1 report
2004 			 * by not transitioning to the IDLE state.
2005 			 */
2006 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2007 			if (delay) {
2008 				inm->in6m_timer = max(delay, odelay);
2009 				V_current_state_timers_running6 = 1;
2010 			} else {
2011 				inm->in6m_state = MLD_IDLE_MEMBER;
2012 				NET_EPOCH_ENTER(et);
2013 				error = mld_v1_transmit_report(inm,
2014 				     MLD_LISTENER_REPORT);
2015 				NET_EPOCH_EXIT(et);
2016 				if (error == 0) {
2017 					inm->in6m_timer = odelay;
2018 					V_current_state_timers_running6 = 1;
2019 				}
2020 			}
2021 			break;
2022 
2023 		case MLD_VERSION_2:
2024 			/*
2025 			 * Defer update of T0 to T1, until the first copy
2026 			 * of the state change has been transmitted.
2027 			 */
2028 			syncstates = 0;
2029 
2030 			/*
2031 			 * Immediately enqueue a State-Change Report for
2032 			 * this interface, freeing any previous reports.
2033 			 * Don't kick the timers if there is nothing to do,
2034 			 * or if an error occurred.
2035 			 */
2036 			mq = &inm->in6m_scq;
2037 			mbufq_drain(mq);
2038 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2039 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2040 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2041 			    __func__, retval);
2042 			if (retval <= 0) {
2043 				error = retval * -1;
2044 				break;
2045 			}
2046 
2047 			/*
2048 			 * Schedule transmission of pending state-change
2049 			 * report up to RV times for this link. The timer
2050 			 * will fire at the next mld_fasttimo (~200ms),
2051 			 * giving us an opportunity to merge the reports.
2052 			 *
2053 			 * If a delay was provided to this function, only
2054 			 * use this delay if sooner than the existing one.
2055 			 */
2056 			KASSERT(mli->mli_rv > 1,
2057 			   ("%s: invalid robustness %d", __func__,
2058 			    mli->mli_rv));
2059 			inm->in6m_scrv = mli->mli_rv;
2060 			if (delay) {
2061 				if (inm->in6m_sctimer > 1) {
2062 					inm->in6m_sctimer =
2063 					    min(inm->in6m_sctimer, delay);
2064 				} else
2065 					inm->in6m_sctimer = delay;
2066 			} else
2067 				inm->in6m_sctimer = 1;
2068 			V_state_change_timers_running6 = 1;
2069 
2070 			error = 0;
2071 			break;
2072 		}
2073 	}
2074 
2075 	/*
2076 	 * Only update the T0 state if state change is atomic,
2077 	 * i.e. we don't need to wait for a timer to fire before we
2078 	 * can consider the state change to have been communicated.
2079 	 */
2080 	if (syncstates) {
2081 		in6m_commit(inm);
2082 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2083 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2084 		    if_name(inm->in6m_ifp));
2085 	}
2086 
2087 	return (error);
2088 }
2089 
2090 /*
2091  * Issue an intermediate state change during the life-cycle.
2092  */
2093 static int
2094 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2095 {
2096 	struct ifnet		*ifp;
2097 	int			 retval;
2098 #ifdef KTR
2099 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2100 #endif
2101 
2102 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2103 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2104 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2105 
2106 	ifp = inm->in6m_ifp;
2107 
2108 	IN6_MULTI_LIST_LOCK_ASSERT();
2109 	MLD_LOCK_ASSERT();
2110 
2111 	KASSERT(mli && mli->mli_ifp == ifp,
2112 	    ("%s: inconsistent ifp", __func__));
2113 
2114 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2115 	    (mli->mli_flags & MLIF_SILENT) ||
2116 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2117 	    (mli->mli_version != MLD_VERSION_2)) {
2118 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2119 			CTR1(KTR_MLD,
2120 "%s: not kicking state machine for silent group", __func__);
2121 		}
2122 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2123 		in6m_commit(inm);
2124 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2125 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2126 		    if_name(inm->in6m_ifp));
2127 		return (0);
2128 	}
2129 
2130 	mbufq_drain(&inm->in6m_scq);
2131 
2132 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2133 	    (mli->mli_flags & MLIF_USEALLOW));
2134 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2135 	if (retval <= 0)
2136 		return (-retval);
2137 
2138 	/*
2139 	 * If record(s) were enqueued, start the state-change
2140 	 * report timer for this group.
2141 	 */
2142 	inm->in6m_scrv = mli->mli_rv;
2143 	inm->in6m_sctimer = 1;
2144 	V_state_change_timers_running6 = 1;
2145 
2146 	return (0);
2147 }
2148 
2149 /*
2150  * Perform the final leave for a multicast address.
2151  *
2152  * When leaving a group:
2153  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2154  *  MLDv2 enqueues a state-change report containing a transition
2155  *  to INCLUDE {} for immediate transmission.
2156  */
2157 static void
2158 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2159 {
2160 	struct epoch_tracker     et;
2161 	int syncstates;
2162 #ifdef KTR
2163 	char ip6tbuf[INET6_ADDRSTRLEN];
2164 #endif
2165 
2166 	syncstates = 1;
2167 
2168 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2169 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2170 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2171 
2172 	IN6_MULTI_LIST_LOCK_ASSERT();
2173 	MLD_LOCK_ASSERT();
2174 
2175 	switch (inm->in6m_state) {
2176 	case MLD_NOT_MEMBER:
2177 	case MLD_SILENT_MEMBER:
2178 	case MLD_LEAVING_MEMBER:
2179 		/* Already leaving or left; do nothing. */
2180 		CTR1(KTR_MLD,
2181 "%s: not kicking state machine for silent group", __func__);
2182 		break;
2183 	case MLD_REPORTING_MEMBER:
2184 	case MLD_IDLE_MEMBER:
2185 	case MLD_G_QUERY_PENDING_MEMBER:
2186 	case MLD_SG_QUERY_PENDING_MEMBER:
2187 		if (mli->mli_version == MLD_VERSION_1) {
2188 #ifdef INVARIANTS
2189 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2190 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2191 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2192 			     __func__);
2193 #endif
2194 			NET_EPOCH_ENTER(et);
2195 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2196 			NET_EPOCH_EXIT(et);
2197 			inm->in6m_state = MLD_NOT_MEMBER;
2198 			V_current_state_timers_running6 = 1;
2199 		} else if (mli->mli_version == MLD_VERSION_2) {
2200 			/*
2201 			 * Stop group timer and all pending reports.
2202 			 * Immediately enqueue a state-change report
2203 			 * TO_IN {} to be sent on the next fast timeout,
2204 			 * giving us an opportunity to merge reports.
2205 			 */
2206 			mbufq_drain(&inm->in6m_scq);
2207 			inm->in6m_timer = 0;
2208 			inm->in6m_scrv = mli->mli_rv;
2209 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2210 			    "pending retransmissions.", __func__,
2211 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2212 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2213 			if (inm->in6m_scrv == 0) {
2214 				inm->in6m_state = MLD_NOT_MEMBER;
2215 				inm->in6m_sctimer = 0;
2216 			} else {
2217 				int retval __diagused;
2218 
2219 				in6m_acquire_locked(inm);
2220 
2221 				retval = mld_v2_enqueue_group_record(
2222 				    &inm->in6m_scq, inm, 1, 0, 0,
2223 				    (mli->mli_flags & MLIF_USEALLOW));
2224 				KASSERT(retval != 0,
2225 				    ("%s: enqueue record = %d", __func__,
2226 				     retval));
2227 
2228 				inm->in6m_state = MLD_LEAVING_MEMBER;
2229 				inm->in6m_sctimer = 1;
2230 				V_state_change_timers_running6 = 1;
2231 				syncstates = 0;
2232 			}
2233 			break;
2234 		}
2235 		break;
2236 	case MLD_LAZY_MEMBER:
2237 	case MLD_SLEEPING_MEMBER:
2238 	case MLD_AWAKENING_MEMBER:
2239 		/* Our reports are suppressed; do nothing. */
2240 		break;
2241 	}
2242 
2243 	if (syncstates) {
2244 		in6m_commit(inm);
2245 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2246 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2247 		    if_name(inm->in6m_ifp));
2248 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2249 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2250 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2251 	}
2252 }
2253 
2254 /*
2255  * Enqueue an MLDv2 group record to the given output queue.
2256  *
2257  * If is_state_change is zero, a current-state record is appended.
2258  * If is_state_change is non-zero, a state-change report is appended.
2259  *
2260  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2261  * If is_group_query is zero, and if there is a packet with free space
2262  * at the tail of the queue, it will be appended to providing there
2263  * is enough free space.
2264  * Otherwise a new mbuf packet chain is allocated.
2265  *
2266  * If is_source_query is non-zero, each source is checked to see if
2267  * it was recorded for a Group-Source query, and will be omitted if
2268  * it is not both in-mode and recorded.
2269  *
2270  * If use_block_allow is non-zero, state change reports for initial join
2271  * and final leave, on an inclusive mode group with a source list, will be
2272  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2273  *
2274  * The function will attempt to allocate leading space in the packet
2275  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2276  *
2277  * If successful the size of all data appended to the queue is returned,
2278  * otherwise an error code less than zero is returned, or zero if
2279  * no record(s) were appended.
2280  */
2281 static int
2282 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2283     const int is_state_change, const int is_group_query,
2284     const int is_source_query, const int use_block_allow)
2285 {
2286 	struct mldv2_record	 mr;
2287 	struct mldv2_record	*pmr;
2288 	struct ifnet		*ifp;
2289 	struct ip6_msource	*ims, *nims;
2290 	struct mbuf		*m0, *m, *md;
2291 	int			 is_filter_list_change;
2292 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2293 	int			 record_has_sources;
2294 	int			 now;
2295 	int			 type;
2296 	uint8_t			 mode;
2297 #ifdef KTR
2298 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2299 #endif
2300 
2301 	IN6_MULTI_LIST_LOCK_ASSERT();
2302 
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.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.vt_nrecs = 1;
2552 		mbufq_enqueue(mq, m);
2553 	} else
2554 		m->m_pkthdr.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.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 	uint8_t			 mode, now, then;
2678 	rectype_t		 crt, drt, nrt;
2679 #ifdef KTR
2680 	int			 nallow, nblock;
2681 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2682 #endif
2683 
2684 	IN6_MULTI_LIST_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 #ifdef KTR
2701 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2702 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2703 #endif
2704 	nims = NULL;	/* next tree node pointer */
2705 
2706 	/*
2707 	 * For each possible filter record mode.
2708 	 * The first kind of source we encounter tells us which
2709 	 * is the first kind of record we start appending.
2710 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2711 	 * as the inverse of the group's filter mode.
2712 	 */
2713 	while (drt != REC_FULL) {
2714 		do {
2715 			m0 = mbufq_last(mq);
2716 			if (m0 != NULL &&
2717 			    (m0->m_pkthdr.vt_nrecs + 1 <=
2718 			     MLD_V2_REPORT_MAXRECS) &&
2719 			    (m0->m_pkthdr.len + MINRECLEN) <
2720 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2721 				m = m0;
2722 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2723 					    sizeof(struct mldv2_record)) /
2724 					    sizeof(struct in6_addr);
2725 				CTR1(KTR_MLD,
2726 				    "%s: use previous packet", __func__);
2727 			} else {
2728 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2729 				if (m == NULL)
2730 					m = m_gethdr(M_NOWAIT, MT_DATA);
2731 				if (m == NULL) {
2732 					CTR1(KTR_MLD,
2733 					    "%s: m_get*() failed", __func__);
2734 					return (-ENOMEM);
2735 				}
2736 				m->m_pkthdr.vt_nrecs = 0;
2737 				mld_save_context(m, ifp);
2738 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2739 				    sizeof(struct mldv2_record)) /
2740 				    sizeof(struct in6_addr);
2741 				npbytes = 0;
2742 				CTR1(KTR_MLD,
2743 				    "%s: allocated new packet", __func__);
2744 			}
2745 			/*
2746 			 * Append the MLD group record header to the
2747 			 * current packet's data area.
2748 			 * Recalculate pointer to free space for next
2749 			 * group record, in case m_append() allocated
2750 			 * a new mbuf or cluster.
2751 			 */
2752 			memset(&mr, 0, sizeof(mr));
2753 			mr.mr_addr = inm->in6m_addr;
2754 			in6_clearscope(&mr.mr_addr);
2755 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2756 				if (m != m0)
2757 					m_freem(m);
2758 				CTR1(KTR_MLD,
2759 				    "%s: m_append() failed", __func__);
2760 				return (-ENOMEM);
2761 			}
2762 			npbytes += sizeof(struct mldv2_record);
2763 			if (m != m0) {
2764 				/* new packet; offset in chain */
2765 				md = m_getptr(m, npbytes -
2766 				    sizeof(struct mldv2_record), &off);
2767 				pmr = (struct mldv2_record *)(mtod(md,
2768 				    uint8_t *) + off);
2769 			} else {
2770 				/* current packet; offset from last append */
2771 				md = m_last(m);
2772 				pmr = (struct mldv2_record *)(mtod(md,
2773 				    uint8_t *) + md->m_len -
2774 				    sizeof(struct mldv2_record));
2775 			}
2776 			/*
2777 			 * Begin walking the tree for this record type
2778 			 * pass, or continue from where we left off
2779 			 * previously if we had to allocate a new packet.
2780 			 * Only report deltas in-mode at t1.
2781 			 * We need not report included sources as allowed
2782 			 * if we are in inclusive mode on the group,
2783 			 * however the converse is not true.
2784 			 */
2785 			rsrcs = 0;
2786 			if (nims == NULL) {
2787 				nims = RB_MIN(ip6_msource_tree,
2788 				    &inm->in6m_srcs);
2789 			}
2790 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2791 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2792 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2793 				now = im6s_get_mode(inm, ims, 1);
2794 				then = im6s_get_mode(inm, ims, 0);
2795 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2796 				    __func__, then, now);
2797 				if (now == then) {
2798 					CTR1(KTR_MLD,
2799 					    "%s: skip unchanged", __func__);
2800 					continue;
2801 				}
2802 				if (mode == MCAST_EXCLUDE &&
2803 				    now == MCAST_INCLUDE) {
2804 					CTR1(KTR_MLD,
2805 					    "%s: skip IN src on EX group",
2806 					    __func__);
2807 					continue;
2808 				}
2809 				nrt = (rectype_t)now;
2810 				if (nrt == REC_NONE)
2811 					nrt = (rectype_t)(~mode & REC_FULL);
2812 				if (schanged++ == 0) {
2813 					crt = nrt;
2814 				} else if (crt != nrt)
2815 					continue;
2816 				if (!m_append(m, sizeof(struct in6_addr),
2817 				    (void *)&ims->im6s_addr)) {
2818 					if (m != m0)
2819 						m_freem(m);
2820 					CTR1(KTR_MLD,
2821 					    "%s: m_append() failed", __func__);
2822 					return (-ENOMEM);
2823 				}
2824 #ifdef KTR
2825 				nallow += !!(crt == REC_ALLOW);
2826 				nblock += !!(crt == REC_BLOCK);
2827 #endif
2828 				if (++rsrcs == m0srcs)
2829 					break;
2830 			}
2831 			/*
2832 			 * If we did not append any tree nodes on this
2833 			 * pass, back out of allocations.
2834 			 */
2835 			if (rsrcs == 0) {
2836 				npbytes -= sizeof(struct mldv2_record);
2837 				if (m != m0) {
2838 					CTR1(KTR_MLD,
2839 					    "%s: m_free(m)", __func__);
2840 					m_freem(m);
2841 				} else {
2842 					CTR1(KTR_MLD,
2843 					    "%s: m_adj(m, -mr)", __func__);
2844 					m_adj(m, -((int)sizeof(
2845 					    struct mldv2_record)));
2846 				}
2847 				continue;
2848 			}
2849 			npbytes += (rsrcs * sizeof(struct in6_addr));
2850 			if (crt == REC_ALLOW)
2851 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2852 			else if (crt == REC_BLOCK)
2853 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2854 			pmr->mr_numsrc = htons(rsrcs);
2855 			/*
2856 			 * Count the new group record, and enqueue this
2857 			 * packet if it wasn't already queued.
2858 			 */
2859 			m->m_pkthdr.vt_nrecs++;
2860 			if (m != m0)
2861 				mbufq_enqueue(mq, m);
2862 			nbytes += npbytes;
2863 		} while (nims != NULL);
2864 		drt |= crt;
2865 		crt = (~crt & REC_FULL);
2866 	}
2867 
2868 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2869 	    nallow, nblock);
2870 
2871 	return (nbytes);
2872 }
2873 
2874 static int
2875 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2876 {
2877 	struct mbufq	*gq;
2878 	struct mbuf	*m;		/* pending state-change */
2879 	struct mbuf	*m0;		/* copy of pending state-change */
2880 	struct mbuf	*mt;		/* last state-change in packet */
2881 	int		 docopy, domerge;
2882 	u_int		 recslen;
2883 
2884 	docopy = 0;
2885 	domerge = 0;
2886 	recslen = 0;
2887 
2888 	IN6_MULTI_LIST_LOCK_ASSERT();
2889 	MLD_LOCK_ASSERT();
2890 
2891 	/*
2892 	 * If there are further pending retransmissions, make a writable
2893 	 * copy of each queued state-change message before merging.
2894 	 */
2895 	if (inm->in6m_scrv > 0)
2896 		docopy = 1;
2897 
2898 	gq = &inm->in6m_scq;
2899 #ifdef KTR
2900 	if (mbufq_first(gq) == NULL) {
2901 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2902 		    __func__, inm);
2903 	}
2904 #endif
2905 
2906 	m = mbufq_first(gq);
2907 	while (m != NULL) {
2908 		/*
2909 		 * Only merge the report into the current packet if
2910 		 * there is sufficient space to do so; an MLDv2 report
2911 		 * packet may only contain 65,535 group records.
2912 		 * Always use a simple mbuf chain concatentation to do this,
2913 		 * as large state changes for single groups may have
2914 		 * allocated clusters.
2915 		 */
2916 		domerge = 0;
2917 		mt = mbufq_last(scq);
2918 		if (mt != NULL) {
2919 			recslen = m_length(m, NULL);
2920 
2921 			if ((mt->m_pkthdr.vt_nrecs +
2922 			    m->m_pkthdr.vt_nrecs <=
2923 			    MLD_V2_REPORT_MAXRECS) &&
2924 			    (mt->m_pkthdr.len + recslen <=
2925 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2926 				domerge = 1;
2927 		}
2928 
2929 		if (!domerge && mbufq_full(gq)) {
2930 			CTR2(KTR_MLD,
2931 			    "%s: outbound queue full, skipping whole packet %p",
2932 			    __func__, m);
2933 			mt = m->m_nextpkt;
2934 			if (!docopy)
2935 				m_freem(m);
2936 			m = mt;
2937 			continue;
2938 		}
2939 
2940 		if (!docopy) {
2941 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2942 			m0 = mbufq_dequeue(gq);
2943 			m = m0->m_nextpkt;
2944 		} else {
2945 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2946 			m0 = m_dup(m, M_NOWAIT);
2947 			if (m0 == NULL)
2948 				return (ENOMEM);
2949 			m0->m_nextpkt = NULL;
2950 			m = m->m_nextpkt;
2951 		}
2952 
2953 		if (!domerge) {
2954 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2955 			    __func__, m0, scq);
2956 			mbufq_enqueue(scq, m0);
2957 		} else {
2958 			struct mbuf *mtl;	/* last mbuf of packet mt */
2959 
2960 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2961 			    __func__, m0, mt);
2962 
2963 			mtl = m_last(mt);
2964 			m0->m_flags &= ~M_PKTHDR;
2965 			mt->m_pkthdr.len += recslen;
2966 			mt->m_pkthdr.vt_nrecs +=
2967 			    m0->m_pkthdr.vt_nrecs;
2968 
2969 			mtl->m_next = m0;
2970 		}
2971 	}
2972 
2973 	return (0);
2974 }
2975 
2976 /*
2977  * Respond to a pending MLDv2 General Query.
2978  */
2979 static void
2980 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2981 {
2982 	struct ifmultiaddr	*ifma;
2983 	struct ifnet		*ifp;
2984 	struct in6_multi	*inm;
2985 	int			 retval __unused;
2986 
2987 	NET_EPOCH_ASSERT();
2988 	IN6_MULTI_LIST_LOCK_ASSERT();
2989 	MLD_LOCK_ASSERT();
2990 
2991 	KASSERT(mli->mli_version == MLD_VERSION_2,
2992 	    ("%s: called when version %d", __func__, mli->mli_version));
2993 
2994 	/*
2995 	 * Check that there are some packets queued. If so, send them first.
2996 	 * For large number of groups the reply to general query can take
2997 	 * many packets, we should finish sending them before starting of
2998 	 * queuing the new reply.
2999 	 */
3000 	if (mbufq_len(&mli->mli_gq) != 0)
3001 		goto send;
3002 
3003 	ifp = mli->mli_ifp;
3004 
3005 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3006 		inm = in6m_ifmultiaddr_get_inm(ifma);
3007 		if (inm == NULL)
3008 			continue;
3009 		KASSERT(ifp == inm->in6m_ifp,
3010 		    ("%s: inconsistent ifp", __func__));
3011 
3012 		switch (inm->in6m_state) {
3013 		case MLD_NOT_MEMBER:
3014 		case MLD_SILENT_MEMBER:
3015 			break;
3016 		case MLD_REPORTING_MEMBER:
3017 		case MLD_IDLE_MEMBER:
3018 		case MLD_LAZY_MEMBER:
3019 		case MLD_SLEEPING_MEMBER:
3020 		case MLD_AWAKENING_MEMBER:
3021 			inm->in6m_state = MLD_REPORTING_MEMBER;
3022 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3023 			    inm, 0, 0, 0, 0);
3024 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3025 			    __func__, retval);
3026 			break;
3027 		case MLD_G_QUERY_PENDING_MEMBER:
3028 		case MLD_SG_QUERY_PENDING_MEMBER:
3029 		case MLD_LEAVING_MEMBER:
3030 			break;
3031 		}
3032 	}
3033 
3034 send:
3035 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3036 
3037 	/*
3038 	 * Slew transmission of bursts over 500ms intervals.
3039 	 */
3040 	if (mbufq_first(&mli->mli_gq) != NULL) {
3041 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3042 		    MLD_RESPONSE_BURST_INTERVAL);
3043 		V_interface_timers_running6 = 1;
3044 	}
3045 }
3046 
3047 /*
3048  * Transmit the next pending message in the output queue.
3049  *
3050  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3051  * MRT: Nothing needs to be done, as MLD traffic is always local to
3052  * a link and uses a link-scope multicast address.
3053  */
3054 static void
3055 mld_dispatch_packet(struct mbuf *m)
3056 {
3057 	struct ip6_moptions	 im6o;
3058 	struct ifnet		*ifp;
3059 	struct ifnet		*oifp;
3060 	struct mbuf		*m0;
3061 	struct mbuf		*md;
3062 	struct ip6_hdr		*ip6;
3063 	struct mld_hdr		*mld;
3064 	int			 error;
3065 	int			 off;
3066 	int			 type;
3067 	uint32_t		 ifindex;
3068 
3069 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3070 	NET_EPOCH_ASSERT();
3071 
3072 	/*
3073 	 * Set VNET image pointer from enqueued mbuf chain
3074 	 * before doing anything else. Whilst we use interface
3075 	 * indexes to guard against interface detach, they are
3076 	 * unique to each VIMAGE and must be retrieved.
3077 	 */
3078 	ifindex = mld_restore_context(m);
3079 
3080 	/*
3081 	 * Check if the ifnet still exists. This limits the scope of
3082 	 * any race in the absence of a global ifp lock for low cost
3083 	 * (an array lookup).
3084 	 */
3085 	ifp = ifnet_byindex(ifindex);
3086 	if (ifp == NULL) {
3087 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3088 		    __func__, m, ifindex);
3089 		m_freem(m);
3090 		IP6STAT_INC(ip6s_noroute);
3091 		goto out;
3092 	}
3093 
3094 	im6o.im6o_multicast_hlim  = 1;
3095 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3096 	im6o.im6o_multicast_ifp = ifp;
3097 
3098 	if (m->m_flags & M_MLDV1) {
3099 		m0 = m;
3100 	} else {
3101 		m0 = mld_v2_encap_report(ifp, m);
3102 		if (m0 == NULL) {
3103 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3104 			IP6STAT_INC(ip6s_odropped);
3105 			goto out;
3106 		}
3107 	}
3108 
3109 	mld_scrub_context(m0);
3110 	m_clrprotoflags(m);
3111 	m0->m_pkthdr.rcvif = V_loif;
3112 
3113 	ip6 = mtod(m0, struct ip6_hdr *);
3114 #if 0
3115 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3116 #else
3117 	/*
3118 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3119 	 * occur if we called in6_setscope() at transmission.
3120 	 * See comments at top of file.
3121 	 */
3122 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3123 #endif
3124 
3125 	/*
3126 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3127 	 * so we can bump the stats.
3128 	 */
3129 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3130 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3131 	type = mld->mld_type;
3132 
3133 	oifp = NULL;
3134 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3135 	    &oifp, NULL);
3136 	if (error) {
3137 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3138 		goto out;
3139 	}
3140 	ICMP6STAT_INC(icp6s_outhist[type]);
3141 	if (oifp != NULL) {
3142 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3143 		switch (type) {
3144 		case MLD_LISTENER_REPORT:
3145 		case MLDV2_LISTENER_REPORT:
3146 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3147 			break;
3148 		case MLD_LISTENER_DONE:
3149 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3150 			break;
3151 		}
3152 	}
3153 out:
3154 	return;
3155 }
3156 
3157 /*
3158  * Encapsulate an MLDv2 report.
3159  *
3160  * KAME IPv6 requires that hop-by-hop options be passed separately,
3161  * and that the IPv6 header be prepended in a separate mbuf.
3162  *
3163  * Returns a pointer to the new mbuf chain head, or NULL if the
3164  * allocation failed.
3165  */
3166 static struct mbuf *
3167 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3168 {
3169 	struct mbuf		*mh;
3170 	struct mldv2_report	*mld;
3171 	struct ip6_hdr		*ip6;
3172 	struct in6_ifaddr	*ia;
3173 	int			 mldreclen;
3174 
3175 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3176 	KASSERT((m->m_flags & M_PKTHDR),
3177 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3178 
3179 	/*
3180 	 * RFC3590: OK to send as :: or tentative during DAD.
3181 	 */
3182 	NET_EPOCH_ASSERT();
3183 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3184 	if (ia == NULL)
3185 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3186 
3187 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3188 	if (mh == NULL) {
3189 		if (ia != NULL)
3190 			ifa_free(&ia->ia_ifa);
3191 		m_freem(m);
3192 		return (NULL);
3193 	}
3194 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3195 
3196 	mldreclen = m_length(m, NULL);
3197 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3198 
3199 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3200 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3201 	    sizeof(struct mldv2_report) + mldreclen;
3202 
3203 	ip6 = mtod(mh, struct ip6_hdr *);
3204 	ip6->ip6_flow = 0;
3205 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3206 	ip6->ip6_vfc |= IPV6_VERSION;
3207 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3208 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3209 	if (ia != NULL)
3210 		ifa_free(&ia->ia_ifa);
3211 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3212 	/* scope ID will be set in netisr */
3213 
3214 	mld = (struct mldv2_report *)(ip6 + 1);
3215 	mld->mld_type = MLDV2_LISTENER_REPORT;
3216 	mld->mld_code = 0;
3217 	mld->mld_cksum = 0;
3218 	mld->mld_v2_reserved = 0;
3219 	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3220 	m->m_pkthdr.vt_nrecs = 0;
3221 
3222 	mh->m_next = m;
3223 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3224 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3225 	return (mh);
3226 }
3227 
3228 #ifdef KTR
3229 static char *
3230 mld_rec_type_to_str(const int type)
3231 {
3232 
3233 	switch (type) {
3234 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3235 			return "TO_EX";
3236 			break;
3237 		case MLD_CHANGE_TO_INCLUDE_MODE:
3238 			return "TO_IN";
3239 			break;
3240 		case MLD_MODE_IS_EXCLUDE:
3241 			return "MODE_EX";
3242 			break;
3243 		case MLD_MODE_IS_INCLUDE:
3244 			return "MODE_IN";
3245 			break;
3246 		case MLD_ALLOW_NEW_SOURCES:
3247 			return "ALLOW_NEW";
3248 			break;
3249 		case MLD_BLOCK_OLD_SOURCES:
3250 			return "BLOCK_OLD";
3251 			break;
3252 		default:
3253 			break;
3254 	}
3255 	return "unknown";
3256 }
3257 #endif
3258 
3259 static void
3260 mld_init(void *unused __unused)
3261 {
3262 
3263 	CTR1(KTR_MLD, "%s: initializing", __func__);
3264 	MLD_LOCK_INIT();
3265 
3266 	ip6_initpktopts(&mld_po);
3267 	mld_po.ip6po_hlim = 1;
3268 	mld_po.ip6po_hbh = &mld_ra.hbh;
3269 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3270 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3271 }
3272 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3273 
3274 static void
3275 mld_uninit(void *unused __unused)
3276 {
3277 
3278 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3279 	MLD_LOCK_DESTROY();
3280 }
3281 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3282 
3283 static void
3284 vnet_mld_init(const void *unused __unused)
3285 {
3286 
3287 	CTR1(KTR_MLD, "%s: initializing", __func__);
3288 
3289 	LIST_INIT(&V_mli_head);
3290 }
3291 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3292     NULL);
3293 
3294 static void
3295 vnet_mld_uninit(const void *unused __unused)
3296 {
3297 
3298 	/* This can happen if we shutdown the network stack. */
3299 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3300 }
3301 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3302     NULL);
3303 
3304 static int
3305 mld_modevent(module_t mod, int type, void *unused __unused)
3306 {
3307 
3308     switch (type) {
3309     case MOD_LOAD:
3310     case MOD_UNLOAD:
3311 	break;
3312     default:
3313 	return (EOPNOTSUPP);
3314     }
3315     return (0);
3316 }
3317 
3318 static moduledata_t mld_mod = {
3319     "mld",
3320     mld_modevent,
3321     0
3322 };
3323 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3324