xref: /freebsd/sys/netinet6/mld6.c (revision 6bfca4dcab07dad45a805879d954876b353c0810)
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 
66 #include <sys/cdefs.h>
67 #include "opt_inet.h"
68 #include "opt_inet6.h"
69 
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/mbuf.h>
73 #include <sys/socket.h>
74 #include <sys/sysctl.h>
75 #include <sys/kernel.h>
76 #include <sys/callout.h>
77 #include <sys/malloc.h>
78 #include <sys/module.h>
79 #include <sys/ktr.h>
80 
81 #include <net/if.h>
82 #include <net/if_var.h>
83 #include <net/if_private.h>
84 #include <net/route.h>
85 #include <net/vnet.h>
86 
87 #include <netinet/in.h>
88 #include <netinet/in_var.h>
89 #include <netinet6/in6_var.h>
90 #include <netinet/ip6.h>
91 #include <netinet6/ip6_var.h>
92 #include <netinet6/scope6_var.h>
93 #include <netinet/icmp6.h>
94 #include <netinet6/mld6.h>
95 #include <netinet6/mld6_var.h>
96 
97 #include <security/mac/mac_framework.h>
98 
99 #ifndef KTR_MLD
100 #define KTR_MLD KTR_INET6
101 #endif
102 
103 static void	mli_delete_locked(struct ifnet *);
104 static void	mld_dispatch_packet(struct mbuf *);
105 static void	mld_dispatch_queue(struct mbufq *, int);
106 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
107 static void	mld_fasttimo_vnet(struct in6_multi_head *inmh);
108 static int	mld_handle_state_change(struct in6_multi *,
109 		    struct mld_ifsoftc *);
110 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
111 		    const int);
112 #ifdef KTR
113 static char *	mld_rec_type_to_str(const int);
114 #endif
115 static void	mld_set_version(struct mld_ifsoftc *, const int);
116 static void	mld_slowtimo_vnet(void);
117 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
118 		    /*const*/ struct mld_hdr *);
119 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
120 		    /*const*/ struct mld_hdr *);
121 static void	mld_v1_process_group_timer(struct in6_multi_head *,
122 		    struct in6_multi *);
123 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
124 static int	mld_v1_transmit_report(struct in6_multi *, const int);
125 static void	mld_v1_update_group(struct in6_multi *, const int);
126 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
127 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
128 static struct mbuf *
129 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
130 static int	mld_v2_enqueue_filter_change(struct mbufq *,
131 		    struct in6_multi *);
132 static int	mld_v2_enqueue_group_record(struct mbufq *,
133 		    struct in6_multi *, const int, const int, const int,
134 		    const int);
135 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
136 		    struct mbuf *, struct mldv2_query *, const int, const int);
137 static int	mld_v2_merge_state_changes(struct in6_multi *,
138 		    struct mbufq *);
139 static void	mld_v2_process_group_timers(struct in6_multi_head *,
140 		    struct mbufq *, struct mbufq *,
141 		    struct in6_multi *, const int);
142 static int	mld_v2_process_group_query(struct in6_multi *,
143 		    struct mld_ifsoftc *mli, int, struct mbuf *,
144 		    struct mldv2_query *, const int);
145 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
146 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
147 
148 /*
149  * Normative references: RFC 2710, RFC 3590, RFC 3810.
150  *
151  * Locking:
152  *  * The MLD subsystem lock ends up being system-wide for the moment,
153  *    but could be per-VIMAGE later on.
154  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
155  *    Any may be taken independently; if any are held at the same
156  *    time, the above lock order must be followed.
157  *  * IN6_MULTI_LOCK covers in_multi.
158  *  * MLD_LOCK covers per-link state and any global variables in this file.
159  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
160  *    per-link state iterators.
161  *
162  *  XXX LOR PREVENTION
163  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
164  *  will not accept an ifp; it wants an embedded scope ID, unlike
165  *  ip_output(), which happily takes the ifp given to it. The embedded
166  *  scope ID is only used by MLD to select the outgoing interface.
167  *
168  *  During interface attach and detach, MLD will take MLD_LOCK *after*
169  *  the IF_AFDATA_LOCK.
170  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
171  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
172  *  dispatch could work around this, but we'd rather not do that, as it
173  *  can introduce other races.
174  *
175  *  As such, we exploit the fact that the scope ID is just the interface
176  *  index, and embed it in the IPv6 destination address accordingly.
177  *  This is potentially NOT VALID for MLDv1 reports, as they
178  *  are always sent to the multicast group itself; as MLDv2
179  *  reports are always sent to ff02::16, this is not an issue
180  *  when MLDv2 is in use.
181  *
182  *  This does not however eliminate the LOR when ip6_output() itself
183  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
184  *  trigger a LOR warning in WITNESS when the ifnet is detached.
185  *
186  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
187  *  how it's used across the network stack. Here we're simply exploiting
188  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
189  *
190  * VIMAGE:
191  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
192  *    to a vnet in ifp->if_vnet.
193  */
194 static struct mtx		 mld_mtx;
195 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
196 
197 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
198 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
199 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
200 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
201 
202 /*
203  * VIMAGE-wide globals.
204  */
205 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
206 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
207 VNET_DEFINE_STATIC(int, interface_timers_running6);
208 VNET_DEFINE_STATIC(int, state_change_timers_running6);
209 VNET_DEFINE_STATIC(int, current_state_timers_running6);
210 
211 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
212 #define	V_mli_head			VNET(mli_head)
213 #define	V_interface_timers_running6	VNET(interface_timers_running6)
214 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
215 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
216 
217 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
218 
219 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
220     "IPv6 Multicast Listener Discovery");
221 
222 /*
223  * Virtualized sysctls.
224  */
225 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
226     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
227     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
228     "Rate limit for MLDv2 Group-and-Source queries in seconds");
229 
230 /*
231  * Non-virtualized sysctls.
232  */
233 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
234     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
235     "Per-interface MLDv2 state");
236 
237 static int	mld_v1enable = 1;
238 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
239     &mld_v1enable, 0, "Enable fallback to MLDv1");
240 
241 static int	mld_v2enable = 1;
242 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
243     &mld_v2enable, 0, "Enable MLDv2");
244 
245 static int	mld_use_allow = 1;
246 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
247     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
248 
249 /*
250  * Packed Router Alert option structure declaration.
251  */
252 struct mld_raopt {
253 	struct ip6_hbh		hbh;
254 	struct ip6_opt		pad;
255 	struct ip6_opt_router	ra;
256 } __packed;
257 
258 /*
259  * Router Alert hop-by-hop option header.
260  */
261 static struct mld_raopt mld_ra = {
262 	.hbh = { 0, 0 },
263 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
264 	.ra = {
265 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
266 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
267 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
268 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
269 	}
270 };
271 static struct ip6_pktopts mld_po;
272 
273 static __inline void
274 mld_save_context(struct mbuf *m, struct ifnet *ifp)
275 {
276 
277 #ifdef VIMAGE
278 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
279 #endif /* VIMAGE */
280 	m->m_pkthdr.rcvif = ifp;
281 	m->m_pkthdr.flowid = ifp->if_index;
282 }
283 
284 static __inline void
285 mld_scrub_context(struct mbuf *m)
286 {
287 
288 	m->m_pkthdr.PH_loc.ptr = NULL;
289 	m->m_pkthdr.flowid = 0;
290 }
291 
292 /*
293  * Restore context from a queued output chain.
294  * Return saved ifindex.
295  *
296  * VIMAGE: The assertion is there to make sure that we
297  * actually called CURVNET_SET() with what's in the mbuf chain.
298  */
299 static __inline uint32_t
300 mld_restore_context(struct mbuf *m)
301 {
302 
303 #if defined(VIMAGE) && defined(INVARIANTS)
304 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
305 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
306 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
307 #endif
308 	return (m->m_pkthdr.flowid);
309 }
310 
311 /*
312  * Retrieve or set threshold between group-source queries in seconds.
313  *
314  * VIMAGE: Assume curvnet set by caller.
315  * SMPng: NOTE: Serialized by MLD lock.
316  */
317 static int
318 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
319 {
320 	int error;
321 	int i;
322 
323 	error = sysctl_wire_old_buffer(req, sizeof(int));
324 	if (error)
325 		return (error);
326 
327 	MLD_LOCK();
328 
329 	i = V_mld_gsrdelay.tv_sec;
330 
331 	error = sysctl_handle_int(oidp, &i, 0, req);
332 	if (error || !req->newptr)
333 		goto out_locked;
334 
335 	if (i < -1 || i >= 60) {
336 		error = EINVAL;
337 		goto out_locked;
338 	}
339 
340 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
341 	     V_mld_gsrdelay.tv_sec, i);
342 	V_mld_gsrdelay.tv_sec = i;
343 
344 out_locked:
345 	MLD_UNLOCK();
346 	return (error);
347 }
348 
349 /*
350  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
351  * For use by ifmcstat(8).
352  *
353  * VIMAGE: Assume curvnet set by caller. The node handler itself
354  * is not directly virtualized.
355  */
356 static int
357 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
358 {
359 	struct epoch_tracker	 et;
360 	int			*name;
361 	int			 error;
362 	u_int			 namelen;
363 	struct ifnet		*ifp;
364 	struct mld_ifsoftc	*mli;
365 
366 	name = (int *)arg1;
367 	namelen = arg2;
368 
369 	if (req->newptr != NULL)
370 		return (EPERM);
371 
372 	if (namelen != 1)
373 		return (EINVAL);
374 
375 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
376 	if (error)
377 		return (error);
378 
379 	IN6_MULTI_LOCK();
380 	IN6_MULTI_LIST_LOCK();
381 	MLD_LOCK();
382 	NET_EPOCH_ENTER(et);
383 
384 	error = ENOENT;
385 	ifp = ifnet_byindex(name[0]);
386 	if (ifp == NULL)
387 		goto out_locked;
388 
389 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
390 		if (ifp == mli->mli_ifp) {
391 			struct mld_ifinfo info;
392 
393 			info.mli_version = mli->mli_version;
394 			info.mli_v1_timer = mli->mli_v1_timer;
395 			info.mli_v2_timer = mli->mli_v2_timer;
396 			info.mli_flags = mli->mli_flags;
397 			info.mli_rv = mli->mli_rv;
398 			info.mli_qi = mli->mli_qi;
399 			info.mli_qri = mli->mli_qri;
400 			info.mli_uri = mli->mli_uri;
401 			error = SYSCTL_OUT(req, &info, sizeof(info));
402 			break;
403 		}
404 	}
405 
406 out_locked:
407 	NET_EPOCH_EXIT(et);
408 	MLD_UNLOCK();
409 	IN6_MULTI_LIST_UNLOCK();
410 	IN6_MULTI_UNLOCK();
411 	return (error);
412 }
413 
414 /*
415  * Dispatch an entire queue of pending packet chains.
416  * VIMAGE: Assumes the vnet pointer has been set.
417  */
418 static void
419 mld_dispatch_queue(struct mbufq *mq, int limit)
420 {
421 	struct mbuf *m;
422 
423 	while ((m = mbufq_dequeue(mq)) != NULL) {
424 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
425 		mld_dispatch_packet(m);
426 		if (--limit == 0)
427 			break;
428 	}
429 }
430 
431 /*
432  * Filter outgoing MLD report state by group.
433  *
434  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
435  * and node-local addresses. However, kernel and socket consumers
436  * always embed the KAME scope ID in the address provided, so strip it
437  * when performing comparison.
438  * Note: This is not the same as the *multicast* scope.
439  *
440  * Return zero if the given group is one for which MLD reports
441  * should be suppressed, or non-zero if reports should be issued.
442  */
443 static __inline int
444 mld_is_addr_reported(const struct in6_addr *addr)
445 {
446 
447 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
448 
449 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
450 		return (0);
451 
452 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
453 		struct in6_addr tmp = *addr;
454 		in6_clearscope(&tmp);
455 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
456 			return (0);
457 	}
458 
459 	return (1);
460 }
461 
462 /*
463  * Attach MLD when PF_INET6 is attached to an interface.  Assumes that the
464  * current VNET is set by the caller.
465  */
466 struct mld_ifsoftc *
467 mld_domifattach(struct ifnet *ifp)
468 {
469 	struct mld_ifsoftc *mli;
470 
471 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp));
472 
473 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_WAITOK | M_ZERO);
474 	mli->mli_ifp = ifp;
475 	mli->mli_version = MLD_VERSION_2;
476 	mli->mli_flags = 0;
477 	mli->mli_rv = MLD_RV_INIT;
478 	mli->mli_qi = MLD_QI_INIT;
479 	mli->mli_qri = MLD_QRI_INIT;
480 	mli->mli_uri = MLD_URI_INIT;
481 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
482 	if ((ifp->if_flags & IFF_MULTICAST) == 0)
483 		mli->mli_flags |= MLIF_SILENT;
484 	if (mld_use_allow)
485 		mli->mli_flags |= MLIF_USEALLOW;
486 
487 	MLD_LOCK();
488 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
489 	MLD_UNLOCK();
490 
491 	return (mli);
492 }
493 
494 /*
495  * Hook for ifdetach.
496  *
497  * NOTE: Some finalization tasks need to run before the protocol domain
498  * is detached, but also before the link layer does its cleanup.
499  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
500  *
501  * SMPng: Caller must hold IN6_MULTI_LOCK().
502  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
503  * XXX This routine is also bitten by unlocked ifma_protospec access.
504  */
505 void
506 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
507 {
508 	struct epoch_tracker     et;
509 	struct mld_ifsoftc	*mli;
510 	struct ifmultiaddr	*ifma;
511 	struct in6_multi	*inm;
512 
513 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
514 	    if_name(ifp));
515 
516 	IN6_MULTI_LIST_LOCK_ASSERT();
517 	MLD_LOCK();
518 
519 	mli = MLD_IFINFO(ifp);
520 	IF_ADDR_WLOCK(ifp);
521 	/*
522 	 * Extract list of in6_multi associated with the detaching ifp
523 	 * which the PF_INET6 layer is about to release.
524 	 */
525 	NET_EPOCH_ENTER(et);
526 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
527 		inm = in6m_ifmultiaddr_get_inm(ifma);
528 		if (inm == NULL)
529 			continue;
530 		in6m_disconnect_locked(inmh, inm);
531 
532 		if (mli->mli_version == MLD_VERSION_2) {
533 			in6m_clear_recorded(inm);
534 
535 			/*
536 			 * We need to release the final reference held
537 			 * for issuing the INCLUDE {}.
538 			 */
539 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
540 				inm->in6m_state = MLD_NOT_MEMBER;
541 				in6m_rele_locked(inmh, inm);
542 			}
543 		}
544 	}
545 	NET_EPOCH_EXIT(et);
546 	IF_ADDR_WUNLOCK(ifp);
547 	MLD_UNLOCK();
548 }
549 
550 /*
551  * Hook for domifdetach.
552  * Runs after link-layer cleanup; free MLD state.
553  *
554  * SMPng: Normally called with IF_AFDATA_LOCK held.
555  */
556 void
557 mld_domifdetach(struct ifnet *ifp)
558 {
559 
560 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
561 	    __func__, ifp, if_name(ifp));
562 
563 	MLD_LOCK();
564 	mli_delete_locked(ifp);
565 	MLD_UNLOCK();
566 }
567 
568 static void
569 mli_delete_locked(struct ifnet *ifp)
570 {
571 	struct mld_ifsoftc *mli, *tmli;
572 
573 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
574 	    __func__, ifp, if_name(ifp));
575 
576 	MLD_LOCK_ASSERT();
577 
578 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
579 		if (mli->mli_ifp == ifp) {
580 			/*
581 			 * Free deferred General Query responses.
582 			 */
583 			mbufq_drain(&mli->mli_gq);
584 
585 			LIST_REMOVE(mli, mli_link);
586 
587 			free(mli, M_MLD);
588 			return;
589 		}
590 	}
591 }
592 
593 /*
594  * Process a received MLDv1 general or address-specific query.
595  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
596  *
597  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
598  * mld_addr. This is OK as we own the mbuf chain.
599  */
600 static int
601 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
602     /*const*/ struct mld_hdr *mld)
603 {
604 	struct ifmultiaddr	*ifma;
605 	struct mld_ifsoftc	*mli;
606 	struct in6_multi	*inm;
607 	int			 is_general_query;
608 	uint16_t		 timer;
609 #ifdef KTR
610 	char			 ip6tbuf[INET6_ADDRSTRLEN];
611 #endif
612 
613 	NET_EPOCH_ASSERT();
614 
615 	is_general_query = 0;
616 
617 	if (!mld_v1enable) {
618 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
619 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
620 		    ifp, if_name(ifp));
621 		return (0);
622 	}
623 
624 	/*
625 	 * RFC3810 Section 6.2: MLD queries must originate from
626 	 * a router's link-local address.
627 	 */
628 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
629 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
630 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
631 		    ifp, if_name(ifp));
632 		return (0);
633 	}
634 
635 	/*
636 	 * Do address field validation upfront before we accept
637 	 * the query.
638 	 */
639 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
640 		/*
641 		 * MLDv1 General Query.
642 		 * If this was not sent to the all-nodes group, ignore it.
643 		 */
644 		struct in6_addr		 dst;
645 
646 		dst = ip6->ip6_dst;
647 		in6_clearscope(&dst);
648 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
649 			return (EINVAL);
650 		is_general_query = 1;
651 	} else {
652 		/*
653 		 * Embed scope ID of receiving interface in MLD query for
654 		 * lookup whilst we don't hold other locks.
655 		 */
656 		in6_setscope(&mld->mld_addr, ifp, NULL);
657 	}
658 
659 	IN6_MULTI_LIST_LOCK();
660 	MLD_LOCK();
661 
662 	/*
663 	 * Switch to MLDv1 host compatibility mode.
664 	 */
665 	mli = MLD_IFINFO(ifp);
666 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
667 	mld_set_version(mli, MLD_VERSION_1);
668 
669 	timer = (ntohs(mld->mld_maxdelay) * MLD_FASTHZ) / MLD_TIMER_SCALE;
670 	if (timer == 0)
671 		timer = 1;
672 
673 	if (is_general_query) {
674 		/*
675 		 * For each reporting group joined on this
676 		 * interface, kick the report timer.
677 		 */
678 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
679 			 ifp, if_name(ifp));
680 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
681 			inm = in6m_ifmultiaddr_get_inm(ifma);
682 			if (inm == NULL)
683 				continue;
684 			mld_v1_update_group(inm, timer);
685 		}
686 	} else {
687 		/*
688 		 * MLDv1 Group-Specific Query.
689 		 * If this is a group-specific MLDv1 query, we need only
690 		 * look up the single group to process it.
691 		 */
692 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
693 		if (inm != NULL) {
694 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
695 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
696 			    ifp, if_name(ifp));
697 			mld_v1_update_group(inm, timer);
698 		}
699 		/* XXX Clear embedded scope ID as userland won't expect it. */
700 		in6_clearscope(&mld->mld_addr);
701 	}
702 
703 	MLD_UNLOCK();
704 	IN6_MULTI_LIST_UNLOCK();
705 
706 	return (0);
707 }
708 
709 /*
710  * Update the report timer on a group in response to an MLDv1 query.
711  *
712  * If we are becoming the reporting member for this group, start the timer.
713  * If we already are the reporting member for this group, and timer is
714  * below the threshold, reset it.
715  *
716  * We may be updating the group for the first time since we switched
717  * to MLDv2. If we are, then we must clear any recorded source lists,
718  * and transition to REPORTING state; the group timer is overloaded
719  * for group and group-source query responses.
720  *
721  * Unlike MLDv2, the delay per group should be jittered
722  * to avoid bursts of MLDv1 reports.
723  */
724 static void
725 mld_v1_update_group(struct in6_multi *inm, const int timer)
726 {
727 #ifdef KTR
728 	char			 ip6tbuf[INET6_ADDRSTRLEN];
729 #endif
730 
731 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
732 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
733 	    if_name(inm->in6m_ifp), timer);
734 
735 	IN6_MULTI_LIST_LOCK_ASSERT();
736 
737 	switch (inm->in6m_state) {
738 	case MLD_NOT_MEMBER:
739 	case MLD_SILENT_MEMBER:
740 		break;
741 	case MLD_REPORTING_MEMBER:
742 		if (inm->in6m_timer != 0 &&
743 		    inm->in6m_timer <= timer) {
744 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
745 			    "skipping.", __func__);
746 			break;
747 		}
748 		/* FALLTHROUGH */
749 	case MLD_SG_QUERY_PENDING_MEMBER:
750 	case MLD_G_QUERY_PENDING_MEMBER:
751 	case MLD_IDLE_MEMBER:
752 	case MLD_LAZY_MEMBER:
753 	case MLD_AWAKENING_MEMBER:
754 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
755 		inm->in6m_state = MLD_REPORTING_MEMBER;
756 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
757 		V_current_state_timers_running6 = 1;
758 		break;
759 	case MLD_SLEEPING_MEMBER:
760 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
761 		inm->in6m_state = MLD_AWAKENING_MEMBER;
762 		break;
763 	case MLD_LEAVING_MEMBER:
764 		break;
765 	}
766 }
767 
768 /*
769  * Process a received MLDv2 general, group-specific or
770  * group-and-source-specific query.
771  *
772  * Assumes that mld points to a struct mldv2_query which is stored in
773  * contiguous memory.
774  *
775  * Return 0 if successful, otherwise an appropriate error code is returned.
776  */
777 static int
778 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
779     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
780 {
781 	struct mld_ifsoftc	*mli;
782 	struct in6_multi	*inm;
783 	uint32_t		 maxdelay, nsrc, qqi;
784 	int			 is_general_query;
785 	uint16_t		 timer;
786 	uint8_t			 qrv;
787 #ifdef KTR
788 	char			 ip6tbuf[INET6_ADDRSTRLEN];
789 #endif
790 
791 	NET_EPOCH_ASSERT();
792 
793 	if (!mld_v2enable) {
794 		CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
795 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
796 		    ifp, if_name(ifp));
797 		return (0);
798 	}
799 
800 	/*
801 	 * RFC3810 Section 6.2: MLD queries must originate from
802 	 * a router's link-local address.
803 	 */
804 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
805 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
806 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
807 		    ifp, if_name(ifp));
808 		return (0);
809 	}
810 
811 	is_general_query = 0;
812 
813 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
814 
815 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
816 	if (maxdelay >= 32768) {
817 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
818 			   (MLD_MRC_EXP(maxdelay) + 3);
819 	}
820 	timer = (maxdelay * MLD_FASTHZ) / MLD_TIMER_SCALE;
821 	if (timer == 0)
822 		timer = 1;
823 
824 	qrv = MLD_QRV(mld->mld_misc);
825 	if (qrv < 2) {
826 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
827 		    qrv, MLD_RV_INIT);
828 		qrv = MLD_RV_INIT;
829 	}
830 
831 	qqi = mld->mld_qqi;
832 	if (qqi >= 128) {
833 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
834 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
835 	}
836 
837 	nsrc = ntohs(mld->mld_numsrc);
838 	if (nsrc > MLD_MAX_GS_SOURCES)
839 		return (EMSGSIZE);
840 	if (icmp6len < sizeof(struct mldv2_query) +
841 	    (nsrc * sizeof(struct in6_addr)))
842 		return (EMSGSIZE);
843 
844 	/*
845 	 * Do further input validation upfront to avoid resetting timers
846 	 * should we need to discard this query.
847 	 */
848 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
849 		/*
850 		 * A general query with a source list has undefined
851 		 * behaviour; discard it.
852 		 */
853 		if (nsrc > 0)
854 			return (EINVAL);
855 		is_general_query = 1;
856 	} else {
857 		/*
858 		 * Embed scope ID of receiving interface in MLD query for
859 		 * lookup whilst we don't hold other locks (due to KAME
860 		 * locking lameness). We own this mbuf chain just now.
861 		 */
862 		in6_setscope(&mld->mld_addr, ifp, NULL);
863 	}
864 
865 	IN6_MULTI_LIST_LOCK();
866 	MLD_LOCK();
867 
868 	mli = MLD_IFINFO(ifp);
869 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
870 
871 	/*
872 	 * Discard the v2 query if we're in Compatibility Mode.
873 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
874 	 * until the Old Version Querier Present timer expires.
875 	 */
876 	if (mli->mli_version != MLD_VERSION_2)
877 		goto out_locked;
878 
879 	mld_set_version(mli, MLD_VERSION_2);
880 	mli->mli_rv = qrv;
881 	mli->mli_qi = qqi;
882 	mli->mli_qri = maxdelay;
883 
884 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
885 	    maxdelay);
886 
887 	if (is_general_query) {
888 		/*
889 		 * MLDv2 General Query.
890 		 *
891 		 * Schedule a current-state report on this ifp for
892 		 * all groups, possibly containing source lists.
893 		 *
894 		 * If there is a pending General Query response
895 		 * scheduled earlier than the selected delay, do
896 		 * not schedule any other reports.
897 		 * Otherwise, reset the interface timer.
898 		 */
899 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
900 		    ifp, if_name(ifp));
901 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
902 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
903 			V_interface_timers_running6 = 1;
904 		}
905 	} else {
906 		/*
907 		 * MLDv2 Group-specific or Group-and-source-specific Query.
908 		 *
909 		 * Group-source-specific queries are throttled on
910 		 * a per-group basis to defeat denial-of-service attempts.
911 		 * Queries for groups we are not a member of on this
912 		 * link are simply ignored.
913 		 */
914 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
915 		if (inm == NULL)
916 			goto out_locked;
917 		if (nsrc > 0) {
918 			if (!ratecheck(&inm->in6m_lastgsrtv,
919 			    &V_mld_gsrdelay)) {
920 				CTR1(KTR_MLD, "%s: GS query throttled.",
921 				    __func__);
922 				goto out_locked;
923 			}
924 		}
925 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
926 		     ifp, if_name(ifp));
927 		/*
928 		 * If there is a pending General Query response
929 		 * scheduled sooner than the selected delay, no
930 		 * further report need be scheduled.
931 		 * Otherwise, prepare to respond to the
932 		 * group-specific or group-and-source query.
933 		 */
934 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
935 			mld_v2_process_group_query(inm, mli, timer, m, mld, off);
936 
937 		/* XXX Clear embedded scope ID as userland won't expect it. */
938 		in6_clearscope(&mld->mld_addr);
939 	}
940 
941 out_locked:
942 	MLD_UNLOCK();
943 	IN6_MULTI_LIST_UNLOCK();
944 
945 	return (0);
946 }
947 
948 /*
949  * Process a received MLDv2 group-specific or group-and-source-specific
950  * query.
951  * Return <0 if any error occurred. Currently this is ignored.
952  */
953 static int
954 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
955     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
956 {
957 	int			 retval;
958 	uint16_t		 nsrc;
959 
960 	IN6_MULTI_LIST_LOCK_ASSERT();
961 	MLD_LOCK_ASSERT();
962 
963 	retval = 0;
964 
965 	switch (inm->in6m_state) {
966 	case MLD_NOT_MEMBER:
967 	case MLD_SILENT_MEMBER:
968 	case MLD_SLEEPING_MEMBER:
969 	case MLD_LAZY_MEMBER:
970 	case MLD_AWAKENING_MEMBER:
971 	case MLD_IDLE_MEMBER:
972 	case MLD_LEAVING_MEMBER:
973 		return (retval);
974 		break;
975 	case MLD_REPORTING_MEMBER:
976 	case MLD_G_QUERY_PENDING_MEMBER:
977 	case MLD_SG_QUERY_PENDING_MEMBER:
978 		break;
979 	}
980 
981 	nsrc = ntohs(mld->mld_numsrc);
982 
983 	/* Length should be checked by calling function. */
984 	KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
985 	    m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
986 	    nsrc * sizeof(struct in6_addr),
987 	    ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
988 	    m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
989 	    nsrc * sizeof(struct in6_addr), m0));
990 
991 	/*
992 	 * Deal with group-specific queries upfront.
993 	 * If any group query is already pending, purge any recorded
994 	 * source-list state if it exists, and schedule a query response
995 	 * for this group-specific query.
996 	 */
997 	if (nsrc == 0) {
998 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
999 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1000 			in6m_clear_recorded(inm);
1001 			timer = min(inm->in6m_timer, timer);
1002 		}
1003 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1004 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1005 		V_current_state_timers_running6 = 1;
1006 		return (retval);
1007 	}
1008 
1009 	/*
1010 	 * Deal with the case where a group-and-source-specific query has
1011 	 * been received but a group-specific query is already pending.
1012 	 */
1013 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1014 		timer = min(inm->in6m_timer, timer);
1015 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1016 		V_current_state_timers_running6 = 1;
1017 		return (retval);
1018 	}
1019 
1020 	/*
1021 	 * Finally, deal with the case where a group-and-source-specific
1022 	 * query has been received, where a response to a previous g-s-r
1023 	 * query exists, or none exists.
1024 	 * In this case, we need to parse the source-list which the Querier
1025 	 * has provided us with and check if we have any source list filter
1026 	 * entries at T1 for these sources. If we do not, there is no need
1027 	 * schedule a report and the query may be dropped.
1028 	 * If we do, we must record them and schedule a current-state
1029 	 * report for those sources.
1030 	 */
1031 	if (inm->in6m_nsrc > 0) {
1032 		struct in6_addr		 srcaddr;
1033 		int			 i, nrecorded;
1034 		int			 soff;
1035 
1036 		soff = off + sizeof(struct mldv2_query);
1037 		nrecorded = 0;
1038 		for (i = 0; i < nsrc; i++) {
1039 			m_copydata(m0, soff, sizeof(struct in6_addr),
1040 			    (caddr_t)&srcaddr);
1041 			retval = in6m_record_source(inm, &srcaddr);
1042 			if (retval < 0)
1043 				break;
1044 			nrecorded += retval;
1045 			soff += sizeof(struct in6_addr);
1046 		}
1047 		if (nrecorded > 0) {
1048 			CTR1(KTR_MLD,
1049 			    "%s: schedule response to SG query", __func__);
1050 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1051 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1052 			V_current_state_timers_running6 = 1;
1053 		}
1054 	}
1055 
1056 	return (retval);
1057 }
1058 
1059 /*
1060  * Process a received MLDv1 host membership report.
1061  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1062  *
1063  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1064  * mld_addr. This is OK as we own the mbuf chain.
1065  */
1066 static int
1067 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1068     /*const*/ struct mld_hdr *mld)
1069 {
1070 	struct in6_addr		 src, dst;
1071 	struct in6_ifaddr	*ia;
1072 	struct in6_multi	*inm;
1073 #ifdef KTR
1074 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1075 #endif
1076 
1077 	NET_EPOCH_ASSERT();
1078 
1079 	if (!mld_v1enable) {
1080 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1081 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1082 		    ifp, if_name(ifp));
1083 		return (0);
1084 	}
1085 
1086 	if (ifp->if_flags & IFF_LOOPBACK)
1087 		return (0);
1088 
1089 	/*
1090 	 * MLDv1 reports must originate from a host's link-local address,
1091 	 * or the unspecified address (when booting).
1092 	 */
1093 	src = ip6->ip6_src;
1094 	in6_clearscope(&src);
1095 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1096 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1097 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1098 		    ifp, if_name(ifp));
1099 		return (EINVAL);
1100 	}
1101 
1102 	/*
1103 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1104 	 * group, and must be directed to the group itself.
1105 	 */
1106 	dst = ip6->ip6_dst;
1107 	in6_clearscope(&dst);
1108 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1109 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1110 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1111 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1112 		    ifp, if_name(ifp));
1113 		return (EINVAL);
1114 	}
1115 
1116 	/*
1117 	 * Make sure we don't hear our own membership report, as fast
1118 	 * leave requires knowing that we are the only member of a
1119 	 * group. Assume we used the link-local address if available,
1120 	 * otherwise look for ::.
1121 	 *
1122 	 * XXX Note that scope ID comparison is needed for the address
1123 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1124 	 * performed for the on-wire address.
1125 	 */
1126 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1127 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1128 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1129 		if (ia != NULL)
1130 			ifa_free(&ia->ia_ifa);
1131 		return (0);
1132 	}
1133 	if (ia != NULL)
1134 		ifa_free(&ia->ia_ifa);
1135 
1136 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1137 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1138 
1139 	/*
1140 	 * Embed scope ID of receiving interface in MLD query for lookup
1141 	 * whilst we don't hold other locks (due to KAME locking lameness).
1142 	 */
1143 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1144 		in6_setscope(&mld->mld_addr, ifp, NULL);
1145 
1146 	IN6_MULTI_LIST_LOCK();
1147 	MLD_LOCK();
1148 
1149 	/*
1150 	 * MLDv1 report suppression.
1151 	 * If we are a member of this group, and our membership should be
1152 	 * reported, and our group timer is pending or about to be reset,
1153 	 * stop our group timer by transitioning to the 'lazy' state.
1154 	 */
1155 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1156 	if (inm != NULL) {
1157 		struct mld_ifsoftc *mli;
1158 
1159 		mli = inm->in6m_mli;
1160 		KASSERT(mli != NULL,
1161 		    ("%s: no mli for ifp %p", __func__, ifp));
1162 
1163 		/*
1164 		 * If we are in MLDv2 host mode, do not allow the
1165 		 * other host's MLDv1 report to suppress our reports.
1166 		 */
1167 		if (mli->mli_version == MLD_VERSION_2)
1168 			goto out_locked;
1169 
1170 		inm->in6m_timer = 0;
1171 
1172 		switch (inm->in6m_state) {
1173 		case MLD_NOT_MEMBER:
1174 		case MLD_SILENT_MEMBER:
1175 		case MLD_SLEEPING_MEMBER:
1176 			break;
1177 		case MLD_REPORTING_MEMBER:
1178 		case MLD_IDLE_MEMBER:
1179 		case MLD_AWAKENING_MEMBER:
1180 			CTR3(KTR_MLD,
1181 			    "report suppressed for %s on ifp %p(%s)",
1182 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1183 			    ifp, if_name(ifp));
1184 		case MLD_LAZY_MEMBER:
1185 			inm->in6m_state = MLD_LAZY_MEMBER;
1186 			break;
1187 		case MLD_G_QUERY_PENDING_MEMBER:
1188 		case MLD_SG_QUERY_PENDING_MEMBER:
1189 		case MLD_LEAVING_MEMBER:
1190 			break;
1191 		}
1192 	}
1193 
1194 out_locked:
1195 	MLD_UNLOCK();
1196 	IN6_MULTI_LIST_UNLOCK();
1197 
1198 	/* XXX Clear embedded scope ID as userland won't expect it. */
1199 	in6_clearscope(&mld->mld_addr);
1200 
1201 	return (0);
1202 }
1203 
1204 /*
1205  * MLD input path.
1206  *
1207  * Assume query messages which fit in a single ICMPv6 message header
1208  * have been pulled up.
1209  * Assume that userland will want to see the message, even if it
1210  * otherwise fails kernel input validation; do not free it.
1211  * Pullup may however free the mbuf chain m if it fails.
1212  *
1213  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1214  */
1215 int
1216 mld_input(struct mbuf **mp, int off, int icmp6len)
1217 {
1218 	struct ifnet	*ifp;
1219 	struct ip6_hdr	*ip6;
1220 	struct mbuf	*m;
1221 	struct mld_hdr	*mld;
1222 	int		 mldlen;
1223 
1224 	m = *mp;
1225 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1226 
1227 	ifp = m->m_pkthdr.rcvif;
1228 
1229 	/* Pullup to appropriate size. */
1230 	if (m->m_len < off + sizeof(*mld)) {
1231 		m = m_pullup(m, off + sizeof(*mld));
1232 		if (m == NULL) {
1233 			ICMP6STAT_INC(icp6s_badlen);
1234 			return (IPPROTO_DONE);
1235 		}
1236 	}
1237 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1238 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1239 	    icmp6len >= sizeof(struct mldv2_query)) {
1240 		mldlen = sizeof(struct mldv2_query);
1241 	} else {
1242 		mldlen = sizeof(struct mld_hdr);
1243 	}
1244 	if (m->m_len < off + mldlen) {
1245 		m = m_pullup(m, off + mldlen);
1246 		if (m == NULL) {
1247 			ICMP6STAT_INC(icp6s_badlen);
1248 			return (IPPROTO_DONE);
1249 		}
1250 	}
1251 	*mp = m;
1252 	ip6 = mtod(m, struct ip6_hdr *);
1253 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1254 
1255 	/*
1256 	 * Userland needs to see all of this traffic for implementing
1257 	 * the endpoint discovery portion of multicast routing.
1258 	 */
1259 	switch (mld->mld_type) {
1260 	case MLD_LISTENER_QUERY:
1261 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1262 		if (icmp6len == sizeof(struct mld_hdr)) {
1263 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1264 				return (0);
1265 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1266 			if (mld_v2_input_query(ifp, ip6, m,
1267 			    (struct mldv2_query *)mld, off, icmp6len) != 0)
1268 				return (0);
1269 		}
1270 		break;
1271 	case MLD_LISTENER_REPORT:
1272 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1273 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1274 			return (0);
1275 		break;
1276 	case MLDV2_LISTENER_REPORT:
1277 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1278 		break;
1279 	case MLD_LISTENER_DONE:
1280 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1281 		break;
1282 	default:
1283 		break;
1284 	}
1285 
1286 	return (0);
1287 }
1288 
1289 /*
1290  * Fast timeout handler (global).
1291  * VIMAGE: Timeout handlers are expected to service all vimages.
1292  */
1293 static struct callout mldfast_callout;
1294 static void
1295 mld_fasttimo(void *arg __unused)
1296 {
1297 	struct epoch_tracker et;
1298 	struct in6_multi_head inmh;
1299 	VNET_ITERATOR_DECL(vnet_iter);
1300 
1301 	SLIST_INIT(&inmh);
1302 
1303 	NET_EPOCH_ENTER(et);
1304 	VNET_LIST_RLOCK_NOSLEEP();
1305 	VNET_FOREACH(vnet_iter) {
1306 		CURVNET_SET(vnet_iter);
1307 		mld_fasttimo_vnet(&inmh);
1308 		CURVNET_RESTORE();
1309 	}
1310 	VNET_LIST_RUNLOCK_NOSLEEP();
1311 	NET_EPOCH_EXIT(et);
1312 	in6m_release_list_deferred(&inmh);
1313 
1314 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
1315 }
1316 
1317 /*
1318  * Fast timeout handler (per-vnet).
1319  *
1320  * VIMAGE: Assume caller has set up our curvnet.
1321  */
1322 static void
1323 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1324 {
1325 	struct mbufq		 scq;	/* State-change packets */
1326 	struct mbufq		 qrq;	/* Query response packets */
1327 	struct ifnet		*ifp;
1328 	struct mld_ifsoftc	*mli;
1329 	struct ifmultiaddr	*ifma;
1330 	struct in6_multi	*inm;
1331 	int			 uri_fasthz;
1332 
1333 	uri_fasthz = 0;
1334 
1335 	/*
1336 	 * Quick check to see if any work needs to be done, in order to
1337 	 * minimize the overhead of fasttimo processing.
1338 	 * SMPng: XXX Unlocked reads.
1339 	 */
1340 	if (!V_current_state_timers_running6 &&
1341 	    !V_interface_timers_running6 &&
1342 	    !V_state_change_timers_running6)
1343 		return;
1344 
1345 	IN6_MULTI_LIST_LOCK();
1346 	MLD_LOCK();
1347 
1348 	/*
1349 	 * MLDv2 General Query response timer processing.
1350 	 */
1351 	if (V_interface_timers_running6) {
1352 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1353 
1354 		V_interface_timers_running6 = 0;
1355 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1356 			if (mli->mli_v2_timer == 0) {
1357 				/* Do nothing. */
1358 			} else if (--mli->mli_v2_timer == 0) {
1359 				mld_v2_dispatch_general_query(mli);
1360 			} else {
1361 				V_interface_timers_running6 = 1;
1362 			}
1363 		}
1364 	}
1365 
1366 	if (!V_current_state_timers_running6 &&
1367 	    !V_state_change_timers_running6)
1368 		goto out_locked;
1369 
1370 	V_current_state_timers_running6 = 0;
1371 	V_state_change_timers_running6 = 0;
1372 
1373 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1374 
1375 	/*
1376 	 * MLD host report and state-change timer processing.
1377 	 * Note: Processing a v2 group timer may remove a node.
1378 	 */
1379 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1380 		ifp = mli->mli_ifp;
1381 
1382 		if (mli->mli_version == MLD_VERSION_2) {
1383 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1384 			    MLD_FASTHZ);
1385 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1386 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1387 		}
1388 
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 	}
1428 
1429 out_locked:
1430 	MLD_UNLOCK();
1431 	IN6_MULTI_LIST_UNLOCK();
1432 }
1433 
1434 /*
1435  * Update host report group timer.
1436  * Will update the global pending timer flags.
1437  */
1438 static void
1439 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1440 {
1441 	int report_timer_expired;
1442 
1443 	IN6_MULTI_LIST_LOCK_ASSERT();
1444 	MLD_LOCK_ASSERT();
1445 
1446 	if (inm->in6m_timer == 0) {
1447 		report_timer_expired = 0;
1448 	} else if (--inm->in6m_timer == 0) {
1449 		report_timer_expired = 1;
1450 	} else {
1451 		V_current_state_timers_running6 = 1;
1452 		return;
1453 	}
1454 
1455 	switch (inm->in6m_state) {
1456 	case MLD_NOT_MEMBER:
1457 	case MLD_SILENT_MEMBER:
1458 	case MLD_IDLE_MEMBER:
1459 	case MLD_LAZY_MEMBER:
1460 	case MLD_SLEEPING_MEMBER:
1461 	case MLD_AWAKENING_MEMBER:
1462 		break;
1463 	case MLD_REPORTING_MEMBER:
1464 		if (report_timer_expired) {
1465 			inm->in6m_state = MLD_IDLE_MEMBER;
1466 			SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1467 		}
1468 		break;
1469 	case MLD_G_QUERY_PENDING_MEMBER:
1470 	case MLD_SG_QUERY_PENDING_MEMBER:
1471 	case MLD_LEAVING_MEMBER:
1472 		break;
1473 	}
1474 }
1475 
1476 /*
1477  * Update a group's timers for MLDv2.
1478  * Will update the global pending timer flags.
1479  * Note: Unlocked read from mli.
1480  */
1481 static void
1482 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1483     struct mbufq *qrq, struct mbufq *scq,
1484     struct in6_multi *inm, const int uri_fasthz)
1485 {
1486 	int query_response_timer_expired;
1487 	int state_change_retransmit_timer_expired;
1488 #ifdef KTR
1489 	char ip6tbuf[INET6_ADDRSTRLEN];
1490 #endif
1491 
1492 	IN6_MULTI_LIST_LOCK_ASSERT();
1493 	MLD_LOCK_ASSERT();
1494 
1495 	query_response_timer_expired = 0;
1496 	state_change_retransmit_timer_expired = 0;
1497 
1498 	/*
1499 	 * During a transition from compatibility mode back to MLDv2,
1500 	 * a group record in REPORTING state may still have its group
1501 	 * timer active. This is a no-op in this function; it is easier
1502 	 * to deal with it here than to complicate the slow-timeout path.
1503 	 */
1504 	if (inm->in6m_timer == 0) {
1505 		query_response_timer_expired = 0;
1506 	} else if (--inm->in6m_timer == 0) {
1507 		query_response_timer_expired = 1;
1508 	} else {
1509 		V_current_state_timers_running6 = 1;
1510 	}
1511 
1512 	if (inm->in6m_sctimer == 0) {
1513 		state_change_retransmit_timer_expired = 0;
1514 	} else if (--inm->in6m_sctimer == 0) {
1515 		state_change_retransmit_timer_expired = 1;
1516 	} else {
1517 		V_state_change_timers_running6 = 1;
1518 	}
1519 
1520 	/* We are in fasttimo, so be quick about it. */
1521 	if (!state_change_retransmit_timer_expired &&
1522 	    !query_response_timer_expired)
1523 		return;
1524 
1525 	switch (inm->in6m_state) {
1526 	case MLD_NOT_MEMBER:
1527 	case MLD_SILENT_MEMBER:
1528 	case MLD_SLEEPING_MEMBER:
1529 	case MLD_LAZY_MEMBER:
1530 	case MLD_AWAKENING_MEMBER:
1531 	case MLD_IDLE_MEMBER:
1532 		break;
1533 	case MLD_G_QUERY_PENDING_MEMBER:
1534 	case MLD_SG_QUERY_PENDING_MEMBER:
1535 		/*
1536 		 * Respond to a previously pending Group-Specific
1537 		 * or Group-and-Source-Specific query by enqueueing
1538 		 * the appropriate Current-State report for
1539 		 * immediate transmission.
1540 		 */
1541 		if (query_response_timer_expired) {
1542 			int retval __unused;
1543 
1544 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1545 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1546 			    0);
1547 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1548 			    __func__, retval);
1549 			inm->in6m_state = MLD_REPORTING_MEMBER;
1550 			in6m_clear_recorded(inm);
1551 		}
1552 		/* FALLTHROUGH */
1553 	case MLD_REPORTING_MEMBER:
1554 	case MLD_LEAVING_MEMBER:
1555 		if (state_change_retransmit_timer_expired) {
1556 			/*
1557 			 * State-change retransmission timer fired.
1558 			 * If there are any further pending retransmissions,
1559 			 * set the global pending state-change flag, and
1560 			 * reset the timer.
1561 			 */
1562 			if (--inm->in6m_scrv > 0) {
1563 				inm->in6m_sctimer = uri_fasthz;
1564 				V_state_change_timers_running6 = 1;
1565 			}
1566 			/*
1567 			 * Retransmit the previously computed state-change
1568 			 * report. If there are no further pending
1569 			 * retransmissions, the mbuf queue will be consumed.
1570 			 * Update T0 state to T1 as we have now sent
1571 			 * a state-change.
1572 			 */
1573 			(void)mld_v2_merge_state_changes(inm, scq);
1574 
1575 			in6m_commit(inm);
1576 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1577 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1578 			    if_name(inm->in6m_ifp));
1579 
1580 			/*
1581 			 * If we are leaving the group for good, make sure
1582 			 * we release MLD's reference to it.
1583 			 * This release must be deferred using a SLIST,
1584 			 * as we are called from a loop which traverses
1585 			 * the in_ifmultiaddr TAILQ.
1586 			 */
1587 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1588 			    inm->in6m_scrv == 0) {
1589 				inm->in6m_state = MLD_NOT_MEMBER;
1590 				in6m_disconnect_locked(inmh, inm);
1591 				in6m_rele_locked(inmh, inm);
1592 			}
1593 		}
1594 		break;
1595 	}
1596 }
1597 
1598 /*
1599  * Switch to a different version on the given interface,
1600  * as per Section 9.12.
1601  */
1602 static void
1603 mld_set_version(struct mld_ifsoftc *mli, const int version)
1604 {
1605 	int old_version_timer;
1606 
1607 	MLD_LOCK_ASSERT();
1608 
1609 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1610 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1611 
1612 	if (version == MLD_VERSION_1) {
1613 		/*
1614 		 * Compute the "Older Version Querier Present" timer as per
1615 		 * Section 9.12.
1616 		 */
1617 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1618 		old_version_timer *= MLD_SLOWHZ;
1619 		mli->mli_v1_timer = old_version_timer;
1620 	}
1621 
1622 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1623 		mli->mli_version = MLD_VERSION_1;
1624 		mld_v2_cancel_link_timers(mli);
1625 	}
1626 }
1627 
1628 /*
1629  * Cancel pending MLDv2 timers for the given link and all groups
1630  * joined on it; state-change, general-query, and group-query timers.
1631  */
1632 static void
1633 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1634 {
1635 	struct epoch_tracker	 et;
1636 	struct in6_multi_head	 inmh;
1637 	struct ifmultiaddr	*ifma;
1638 	struct ifnet		*ifp;
1639 	struct in6_multi	*inm;
1640 
1641 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1642 	    mli->mli_ifp, if_name(mli->mli_ifp));
1643 
1644 	SLIST_INIT(&inmh);
1645 	IN6_MULTI_LIST_LOCK_ASSERT();
1646 	MLD_LOCK_ASSERT();
1647 
1648 	/*
1649 	 * Fast-track this potentially expensive operation
1650 	 * by checking all the global 'timer pending' flags.
1651 	 */
1652 	if (!V_interface_timers_running6 &&
1653 	    !V_state_change_timers_running6 &&
1654 	    !V_current_state_timers_running6)
1655 		return;
1656 
1657 	mli->mli_v2_timer = 0;
1658 
1659 	ifp = mli->mli_ifp;
1660 
1661 	IF_ADDR_WLOCK(ifp);
1662 	NET_EPOCH_ENTER(et);
1663 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1664 		inm = in6m_ifmultiaddr_get_inm(ifma);
1665 		if (inm == NULL)
1666 			continue;
1667 		switch (inm->in6m_state) {
1668 		case MLD_NOT_MEMBER:
1669 		case MLD_SILENT_MEMBER:
1670 		case MLD_IDLE_MEMBER:
1671 		case MLD_LAZY_MEMBER:
1672 		case MLD_SLEEPING_MEMBER:
1673 		case MLD_AWAKENING_MEMBER:
1674 			break;
1675 		case MLD_LEAVING_MEMBER:
1676 			/*
1677 			 * If we are leaving the group and switching
1678 			 * version, we need to release the final
1679 			 * reference held for issuing the INCLUDE {}.
1680 			 */
1681 			if (inm->in6m_refcount == 1)
1682 				in6m_disconnect_locked(&inmh, inm);
1683 			in6m_rele_locked(&inmh, inm);
1684 			/* FALLTHROUGH */
1685 		case MLD_G_QUERY_PENDING_MEMBER:
1686 		case MLD_SG_QUERY_PENDING_MEMBER:
1687 			in6m_clear_recorded(inm);
1688 			/* FALLTHROUGH */
1689 		case MLD_REPORTING_MEMBER:
1690 			inm->in6m_sctimer = 0;
1691 			inm->in6m_timer = 0;
1692 			inm->in6m_state = MLD_REPORTING_MEMBER;
1693 			/*
1694 			 * Free any pending MLDv2 state-change records.
1695 			 */
1696 			mbufq_drain(&inm->in6m_scq);
1697 			break;
1698 		}
1699 	}
1700 	NET_EPOCH_EXIT(et);
1701 	IF_ADDR_WUNLOCK(ifp);
1702 	in6m_release_list_deferred(&inmh);
1703 }
1704 
1705 /*
1706  * Global slowtimo handler.
1707  * VIMAGE: Timeout handlers are expected to service all vimages.
1708  */
1709 static struct callout mldslow_callout;
1710 static void
1711 mld_slowtimo(void *arg __unused)
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 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
1724 }
1725 
1726 /*
1727  * Per-vnet slowtimo handler.
1728  */
1729 static void
1730 mld_slowtimo_vnet(void)
1731 {
1732 	struct mld_ifsoftc *mli;
1733 
1734 	MLD_LOCK();
1735 
1736 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1737 		mld_v1_process_querier_timers(mli);
1738 	}
1739 
1740 	MLD_UNLOCK();
1741 }
1742 
1743 /*
1744  * Update the Older Version Querier Present timers for a link.
1745  * See Section 9.12 of RFC 3810.
1746  */
1747 static void
1748 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1749 {
1750 
1751 	MLD_LOCK_ASSERT();
1752 
1753 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1754 		/*
1755 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1756 		 */
1757 		CTR5(KTR_MLD,
1758 		    "%s: transition from v%d -> v%d on %p(%s)",
1759 		    __func__, mli->mli_version, MLD_VERSION_2,
1760 		    mli->mli_ifp, if_name(mli->mli_ifp));
1761 		mli->mli_version = MLD_VERSION_2;
1762 	}
1763 }
1764 
1765 /*
1766  * Transmit an MLDv1 report immediately.
1767  */
1768 static int
1769 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1770 {
1771 	struct ifnet		*ifp;
1772 	struct in6_ifaddr	*ia;
1773 	struct ip6_hdr		*ip6;
1774 	struct mbuf		*mh, *md;
1775 	struct mld_hdr		*mld;
1776 
1777 	NET_EPOCH_ASSERT();
1778 	IN6_MULTI_LIST_LOCK_ASSERT();
1779 	MLD_LOCK_ASSERT();
1780 
1781 	ifp = in6m->in6m_ifp;
1782 	/* in process of being freed */
1783 	if (ifp == NULL)
1784 		return (0);
1785 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1786 	/* ia may be NULL if link-local address is tentative. */
1787 
1788 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1789 	if (mh == NULL) {
1790 		if (ia != NULL)
1791 			ifa_free(&ia->ia_ifa);
1792 		return (ENOMEM);
1793 	}
1794 	md = m_get(M_NOWAIT, MT_DATA);
1795 	if (md == NULL) {
1796 		m_free(mh);
1797 		if (ia != NULL)
1798 			ifa_free(&ia->ia_ifa);
1799 		return (ENOMEM);
1800 	}
1801 	mh->m_next = md;
1802 
1803 	/*
1804 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1805 	 * that ether_output() does not need to allocate another mbuf
1806 	 * for the header in the most common case.
1807 	 */
1808 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1809 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1810 	mh->m_len = sizeof(struct ip6_hdr);
1811 
1812 	ip6 = mtod(mh, struct ip6_hdr *);
1813 	ip6->ip6_flow = 0;
1814 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1815 	ip6->ip6_vfc |= IPV6_VERSION;
1816 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1817 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1818 	ip6->ip6_dst = in6m->in6m_addr;
1819 
1820 	md->m_len = sizeof(struct mld_hdr);
1821 	mld = mtod(md, struct mld_hdr *);
1822 	mld->mld_type = type;
1823 	mld->mld_code = 0;
1824 	mld->mld_cksum = 0;
1825 	mld->mld_maxdelay = 0;
1826 	mld->mld_reserved = 0;
1827 	mld->mld_addr = in6m->in6m_addr;
1828 	in6_clearscope(&mld->mld_addr);
1829 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1830 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1831 
1832 	mld_save_context(mh, ifp);
1833 	mh->m_flags |= M_MLDV1;
1834 
1835 	mld_dispatch_packet(mh);
1836 
1837 	if (ia != NULL)
1838 		ifa_free(&ia->ia_ifa);
1839 	return (0);
1840 }
1841 
1842 /*
1843  * Process a state change from the upper layer for the given IPv6 group.
1844  *
1845  * Each socket holds a reference on the in_multi in its own ip_moptions.
1846  * The socket layer will have made the necessary updates to.the group
1847  * state, it is now up to MLD to issue a state change report if there
1848  * has been any change between T0 (when the last state-change was issued)
1849  * and T1 (now).
1850  *
1851  * We use the MLDv2 state machine at group level. The MLd module
1852  * however makes the decision as to which MLD protocol version to speak.
1853  * A state change *from* INCLUDE {} always means an initial join.
1854  * A state change *to* INCLUDE {} always means a final leave.
1855  *
1856  * If delay is non-zero, and the state change is an initial multicast
1857  * join, the state change report will be delayed by 'delay' ticks
1858  * in units of MLD_FASTHZ if MLDv1 is active on the link; otherwise
1859  * the initial MLDv2 state change report will be delayed by whichever
1860  * is sooner, a pending state-change timer or delay itself.
1861  *
1862  * VIMAGE: curvnet should have been set by caller, as this routine
1863  * is called from the socket option handlers.
1864  */
1865 int
1866 mld_change_state(struct in6_multi *inm, const int delay)
1867 {
1868 	struct mld_ifsoftc *mli;
1869 	struct ifnet *ifp;
1870 	int error;
1871 
1872 	IN6_MULTI_LIST_LOCK_ASSERT();
1873 
1874 	error = 0;
1875 
1876 	/*
1877 	 * Check if the in6_multi has already been disconnected.
1878 	 */
1879 	if (inm->in6m_ifp == NULL) {
1880 		CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1881 		return (0);
1882 	}
1883 
1884 	/*
1885 	 * Try to detect if the upper layer just asked us to change state
1886 	 * for an interface which has now gone away.
1887 	 */
1888 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1889 	ifp = inm->in6m_ifma->ifma_ifp;
1890 	if (ifp == NULL)
1891 		return (0);
1892 	/*
1893 	 * Sanity check that netinet6's notion of ifp is the
1894 	 * same as net's.
1895 	 */
1896 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1897 
1898 	MLD_LOCK();
1899 	mli = MLD_IFINFO(ifp);
1900 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1901 
1902 	/*
1903 	 * If we detect a state transition to or from MCAST_UNDEFINED
1904 	 * for this group, then we are starting or finishing an MLD
1905 	 * life cycle for this group.
1906 	 */
1907 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1908 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1909 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1910 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1911 			CTR1(KTR_MLD, "%s: initial join", __func__);
1912 			error = mld_initial_join(inm, mli, delay);
1913 			goto out_locked;
1914 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1915 			CTR1(KTR_MLD, "%s: final leave", __func__);
1916 			mld_final_leave(inm, mli);
1917 			goto out_locked;
1918 		}
1919 	} else {
1920 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1921 	}
1922 
1923 	error = mld_handle_state_change(inm, mli);
1924 
1925 out_locked:
1926 	MLD_UNLOCK();
1927 	return (error);
1928 }
1929 
1930 /*
1931  * Perform the initial join for an MLD group.
1932  *
1933  * When joining a group:
1934  *  If the group should have its MLD traffic suppressed, do nothing.
1935  *  MLDv1 starts sending MLDv1 host membership reports.
1936  *  MLDv2 will schedule an MLDv2 state-change report containing the
1937  *  initial state of the membership.
1938  *
1939  * If the delay argument is non-zero, then we must delay sending the
1940  * initial state change for delay ticks (in units of MLD_FASTHZ).
1941  */
1942 static int
1943 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1944     const int delay)
1945 {
1946 	struct epoch_tracker     et;
1947 	struct ifnet		*ifp;
1948 	struct mbufq		*mq;
1949 	int			 error, retval, syncstates;
1950 	int			 odelay;
1951 #ifdef KTR
1952 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1953 #endif
1954 
1955 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1956 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1957 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1958 
1959 	error = 0;
1960 	syncstates = 1;
1961 
1962 	ifp = inm->in6m_ifp;
1963 
1964 	IN6_MULTI_LIST_LOCK_ASSERT();
1965 	MLD_LOCK_ASSERT();
1966 
1967 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1968 
1969 	/*
1970 	 * Groups joined on loopback or marked as 'not reported',
1971 	 * enter the MLD_SILENT_MEMBER state and
1972 	 * are never reported in any protocol exchanges.
1973 	 * All other groups enter the appropriate state machine
1974 	 * for the version in use on this link.
1975 	 * A link marked as MLIF_SILENT causes MLD to be completely
1976 	 * disabled for the link.
1977 	 */
1978 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1979 	    (mli->mli_flags & MLIF_SILENT) ||
1980 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1981 		CTR1(KTR_MLD,
1982 "%s: not kicking state machine for silent group", __func__);
1983 		inm->in6m_state = MLD_SILENT_MEMBER;
1984 		inm->in6m_timer = 0;
1985 	} else {
1986 		/*
1987 		 * Deal with overlapping in_multi lifecycle.
1988 		 * If this group was LEAVING, then make sure
1989 		 * we drop the reference we picked up to keep the
1990 		 * group around for the final INCLUDE {} enqueue.
1991 		 */
1992 		if (mli->mli_version == MLD_VERSION_2 &&
1993 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
1994 			inm->in6m_refcount--;
1995 			MPASS(inm->in6m_refcount > 0);
1996 		}
1997 		inm->in6m_state = MLD_REPORTING_MEMBER;
1998 
1999 		switch (mli->mli_version) {
2000 		case MLD_VERSION_1:
2001 			/*
2002 			 * If a delay was provided, only use it if
2003 			 * it is greater than the delay normally
2004 			 * used for an MLDv1 state change report,
2005 			 * and delay sending the initial MLDv1 report
2006 			 * by not transitioning to the IDLE state.
2007 			 */
2008 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * MLD_FASTHZ);
2009 			if (delay) {
2010 				inm->in6m_timer = max(delay, odelay);
2011 				V_current_state_timers_running6 = 1;
2012 			} else {
2013 				inm->in6m_state = MLD_IDLE_MEMBER;
2014 				NET_EPOCH_ENTER(et);
2015 				error = mld_v1_transmit_report(inm,
2016 				     MLD_LISTENER_REPORT);
2017 				NET_EPOCH_EXIT(et);
2018 				if (error == 0) {
2019 					inm->in6m_timer = odelay;
2020 					V_current_state_timers_running6 = 1;
2021 				}
2022 			}
2023 			break;
2024 
2025 		case MLD_VERSION_2:
2026 			/*
2027 			 * Defer update of T0 to T1, until the first copy
2028 			 * of the state change has been transmitted.
2029 			 */
2030 			syncstates = 0;
2031 
2032 			/*
2033 			 * Immediately enqueue a State-Change Report for
2034 			 * this interface, freeing any previous reports.
2035 			 * Don't kick the timers if there is nothing to do,
2036 			 * or if an error occurred.
2037 			 */
2038 			mq = &inm->in6m_scq;
2039 			mbufq_drain(mq);
2040 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2041 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2042 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2043 			    __func__, retval);
2044 			if (retval <= 0) {
2045 				error = retval * -1;
2046 				break;
2047 			}
2048 
2049 			/*
2050 			 * Schedule transmission of pending state-change
2051 			 * report up to RV times for this link. The timer
2052 			 * will fire at the next mld_fasttimo (~200ms),
2053 			 * giving us an opportunity to merge the reports.
2054 			 *
2055 			 * If a delay was provided to this function, only
2056 			 * use this delay if sooner than the existing one.
2057 			 */
2058 			KASSERT(mli->mli_rv > 1,
2059 			   ("%s: invalid robustness %d", __func__,
2060 			    mli->mli_rv));
2061 			inm->in6m_scrv = mli->mli_rv;
2062 			if (delay) {
2063 				if (inm->in6m_sctimer > 1) {
2064 					inm->in6m_sctimer =
2065 					    min(inm->in6m_sctimer, delay);
2066 				} else
2067 					inm->in6m_sctimer = delay;
2068 			} else
2069 				inm->in6m_sctimer = 1;
2070 			V_state_change_timers_running6 = 1;
2071 
2072 			error = 0;
2073 			break;
2074 		}
2075 	}
2076 
2077 	/*
2078 	 * Only update the T0 state if state change is atomic,
2079 	 * i.e. we don't need to wait for a timer to fire before we
2080 	 * can consider the state change to have been communicated.
2081 	 */
2082 	if (syncstates) {
2083 		in6m_commit(inm);
2084 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2085 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2086 		    if_name(inm->in6m_ifp));
2087 	}
2088 
2089 	return (error);
2090 }
2091 
2092 /*
2093  * Issue an intermediate state change during the life-cycle.
2094  */
2095 static int
2096 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2097 {
2098 	struct ifnet		*ifp;
2099 	int			 retval;
2100 #ifdef KTR
2101 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2102 #endif
2103 
2104 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2105 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2106 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2107 
2108 	ifp = inm->in6m_ifp;
2109 
2110 	IN6_MULTI_LIST_LOCK_ASSERT();
2111 	MLD_LOCK_ASSERT();
2112 
2113 	KASSERT(mli && mli->mli_ifp == ifp,
2114 	    ("%s: inconsistent ifp", __func__));
2115 
2116 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2117 	    (mli->mli_flags & MLIF_SILENT) ||
2118 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2119 	    (mli->mli_version != MLD_VERSION_2)) {
2120 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2121 			CTR1(KTR_MLD,
2122 "%s: not kicking state machine for silent group", __func__);
2123 		}
2124 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2125 		in6m_commit(inm);
2126 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2127 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2128 		    if_name(inm->in6m_ifp));
2129 		return (0);
2130 	}
2131 
2132 	mbufq_drain(&inm->in6m_scq);
2133 
2134 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2135 	    (mli->mli_flags & MLIF_USEALLOW));
2136 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2137 	if (retval <= 0)
2138 		return (-retval);
2139 
2140 	/*
2141 	 * If record(s) were enqueued, start the state-change
2142 	 * report timer for this group.
2143 	 */
2144 	inm->in6m_scrv = mli->mli_rv;
2145 	inm->in6m_sctimer = 1;
2146 	V_state_change_timers_running6 = 1;
2147 
2148 	return (0);
2149 }
2150 
2151 /*
2152  * Perform the final leave for a multicast address.
2153  *
2154  * When leaving a group:
2155  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2156  *  MLDv2 enqueues a state-change report containing a transition
2157  *  to INCLUDE {} for immediate transmission.
2158  */
2159 static void
2160 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2161 {
2162 	struct epoch_tracker     et;
2163 #ifdef KTR
2164 	char ip6tbuf[INET6_ADDRSTRLEN];
2165 #endif
2166 
2167 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2168 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2169 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2170 
2171 	IN6_MULTI_LIST_LOCK_ASSERT();
2172 	MLD_LOCK_ASSERT();
2173 
2174 	switch (inm->in6m_state) {
2175 	case MLD_NOT_MEMBER:
2176 	case MLD_SILENT_MEMBER:
2177 	case MLD_LEAVING_MEMBER:
2178 		/* Already leaving or left; do nothing. */
2179 		CTR1(KTR_MLD,
2180 "%s: not kicking state machine for silent group", __func__);
2181 		break;
2182 	case MLD_REPORTING_MEMBER:
2183 	case MLD_IDLE_MEMBER:
2184 	case MLD_G_QUERY_PENDING_MEMBER:
2185 	case MLD_SG_QUERY_PENDING_MEMBER:
2186 		if (mli->mli_version == MLD_VERSION_1) {
2187 #ifdef INVARIANTS
2188 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2189 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2190 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2191 			     __func__);
2192 #endif
2193 			NET_EPOCH_ENTER(et);
2194 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2195 			NET_EPOCH_EXIT(et);
2196 			inm->in6m_state = MLD_NOT_MEMBER;
2197 			V_current_state_timers_running6 = 1;
2198 		} else if (mli->mli_version == MLD_VERSION_2) {
2199 			/*
2200 			 * Stop group timer and all pending reports.
2201 			 * Immediately enqueue a state-change report
2202 			 * TO_IN {} to be sent on the next fast timeout,
2203 			 * giving us an opportunity to merge reports.
2204 			 */
2205 			mbufq_drain(&inm->in6m_scq);
2206 			inm->in6m_timer = 0;
2207 			inm->in6m_scrv = mli->mli_rv;
2208 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2209 			    "pending retransmissions.", __func__,
2210 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2211 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2212 			if (inm->in6m_scrv == 0) {
2213 				inm->in6m_state = MLD_NOT_MEMBER;
2214 				inm->in6m_sctimer = 0;
2215 			} else {
2216 				int retval __diagused;
2217 
2218 				in6m_acquire_locked(inm);
2219 
2220 				retval = mld_v2_enqueue_group_record(
2221 				    &inm->in6m_scq, inm, 1, 0, 0,
2222 				    (mli->mli_flags & MLIF_USEALLOW));
2223 				KASSERT(retval != 0,
2224 				    ("%s: enqueue record = %d", __func__,
2225 				     retval));
2226 
2227 				inm->in6m_state = MLD_LEAVING_MEMBER;
2228 				inm->in6m_sctimer = 1;
2229 				V_state_change_timers_running6 = 1;
2230 			}
2231 			break;
2232 		}
2233 		break;
2234 	case MLD_LAZY_MEMBER:
2235 	case MLD_SLEEPING_MEMBER:
2236 	case MLD_AWAKENING_MEMBER:
2237 		/* Our reports are suppressed; do nothing. */
2238 		break;
2239 	}
2240 
2241 	in6m_commit(inm);
2242 	CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2243 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2244 	    if_name(inm->in6m_ifp));
2245 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2246 	CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2247 	    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2248 }
2249 
2250 /*
2251  * Enqueue an MLDv2 group record to the given output queue.
2252  *
2253  * If is_state_change is zero, a current-state record is appended.
2254  * If is_state_change is non-zero, a state-change report is appended.
2255  *
2256  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2257  * If is_group_query is zero, and if there is a packet with free space
2258  * at the tail of the queue, it will be appended to providing there
2259  * is enough free space.
2260  * Otherwise a new mbuf packet chain is allocated.
2261  *
2262  * If is_source_query is non-zero, each source is checked to see if
2263  * it was recorded for a Group-Source query, and will be omitted if
2264  * it is not both in-mode and recorded.
2265  *
2266  * If use_block_allow is non-zero, state change reports for initial join
2267  * and final leave, on an inclusive mode group with a source list, will be
2268  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2269  *
2270  * The function will attempt to allocate leading space in the packet
2271  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2272  *
2273  * If successful the size of all data appended to the queue is returned,
2274  * otherwise an error code less than zero is returned, or zero if
2275  * no record(s) were appended.
2276  */
2277 static int
2278 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2279     const int is_state_change, const int is_group_query,
2280     const int is_source_query, const int use_block_allow)
2281 {
2282 	struct mldv2_record	 mr;
2283 	struct mldv2_record	*pmr;
2284 	struct ifnet		*ifp;
2285 	struct ip6_msource	*ims, *nims;
2286 	struct mbuf		*m0, *m, *md;
2287 	int			 is_filter_list_change;
2288 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2289 	int			 record_has_sources;
2290 	int			 now;
2291 	int			 type;
2292 	uint8_t			 mode;
2293 #ifdef KTR
2294 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2295 #endif
2296 
2297 	IN6_MULTI_LIST_LOCK_ASSERT();
2298 
2299 	ifp = inm->in6m_ifp;
2300 	is_filter_list_change = 0;
2301 	m = NULL;
2302 	m0 = NULL;
2303 	m0srcs = 0;
2304 	msrcs = 0;
2305 	nbytes = 0;
2306 	nims = NULL;
2307 	record_has_sources = 1;
2308 	pmr = NULL;
2309 	type = MLD_DO_NOTHING;
2310 	mode = inm->in6m_st[1].iss_fmode;
2311 
2312 	/*
2313 	 * If we did not transition out of ASM mode during t0->t1,
2314 	 * and there are no source nodes to process, we can skip
2315 	 * the generation of source records.
2316 	 */
2317 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2318 	    inm->in6m_nsrc == 0)
2319 		record_has_sources = 0;
2320 
2321 	if (is_state_change) {
2322 		/*
2323 		 * Queue a state change record.
2324 		 * If the mode did not change, and there are non-ASM
2325 		 * listeners or source filters present,
2326 		 * we potentially need to issue two records for the group.
2327 		 * If there are ASM listeners, and there was no filter
2328 		 * mode transition of any kind, do nothing.
2329 		 *
2330 		 * If we are transitioning to MCAST_UNDEFINED, we need
2331 		 * not send any sources. A transition to/from this state is
2332 		 * considered inclusive with some special treatment.
2333 		 *
2334 		 * If we are rewriting initial joins/leaves to use
2335 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2336 		 * we need to send sources in all cases.
2337 		 */
2338 		if (mode != inm->in6m_st[0].iss_fmode) {
2339 			if (mode == MCAST_EXCLUDE) {
2340 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2341 				    __func__);
2342 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2343 			} else {
2344 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2345 				    __func__);
2346 				if (use_block_allow) {
2347 					/*
2348 					 * XXX
2349 					 * Here we're interested in state
2350 					 * edges either direction between
2351 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2352 					 * Perhaps we should just check
2353 					 * the group state, rather than
2354 					 * the filter mode.
2355 					 */
2356 					if (mode == MCAST_UNDEFINED) {
2357 						type = MLD_BLOCK_OLD_SOURCES;
2358 					} else {
2359 						type = MLD_ALLOW_NEW_SOURCES;
2360 					}
2361 				} else {
2362 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2363 					if (mode == MCAST_UNDEFINED)
2364 						record_has_sources = 0;
2365 				}
2366 			}
2367 		} else {
2368 			if (record_has_sources) {
2369 				is_filter_list_change = 1;
2370 			} else {
2371 				type = MLD_DO_NOTHING;
2372 			}
2373 		}
2374 	} else {
2375 		/*
2376 		 * Queue a current state record.
2377 		 */
2378 		if (mode == MCAST_EXCLUDE) {
2379 			type = MLD_MODE_IS_EXCLUDE;
2380 		} else if (mode == MCAST_INCLUDE) {
2381 			type = MLD_MODE_IS_INCLUDE;
2382 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2383 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2384 			     __func__, inm, inm->in6m_st[1].iss_asm));
2385 		}
2386 	}
2387 
2388 	/*
2389 	 * Generate the filter list changes using a separate function.
2390 	 */
2391 	if (is_filter_list_change)
2392 		return (mld_v2_enqueue_filter_change(mq, inm));
2393 
2394 	if (type == MLD_DO_NOTHING) {
2395 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2396 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2397 		    if_name(inm->in6m_ifp));
2398 		return (0);
2399 	}
2400 
2401 	/*
2402 	 * If any sources are present, we must be able to fit at least
2403 	 * one in the trailing space of the tail packet's mbuf,
2404 	 * ideally more.
2405 	 */
2406 	minrec0len = sizeof(struct mldv2_record);
2407 	if (record_has_sources)
2408 		minrec0len += sizeof(struct in6_addr);
2409 
2410 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2411 	    mld_rec_type_to_str(type),
2412 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2413 	    if_name(inm->in6m_ifp));
2414 
2415 	/*
2416 	 * Check if we have a packet in the tail of the queue for this
2417 	 * group into which the first group record for this group will fit.
2418 	 * Otherwise allocate a new packet.
2419 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2420 	 * Note: Group records for G/GSR query responses MUST be sent
2421 	 * in their own packet.
2422 	 */
2423 	m0 = mbufq_last(mq);
2424 	if (!is_group_query &&
2425 	    m0 != NULL &&
2426 	    (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2427 	    (m0->m_pkthdr.len + minrec0len) <
2428 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2429 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2430 			    sizeof(struct mldv2_record)) /
2431 			    sizeof(struct in6_addr);
2432 		m = m0;
2433 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2434 	} else {
2435 		if (mbufq_full(mq)) {
2436 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2437 			return (-ENOMEM);
2438 		}
2439 		m = NULL;
2440 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2441 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2442 		if (!is_state_change && !is_group_query)
2443 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2444 		if (m == NULL)
2445 			m = m_gethdr(M_NOWAIT, MT_DATA);
2446 		if (m == NULL)
2447 			return (-ENOMEM);
2448 
2449 		mld_save_context(m, ifp);
2450 
2451 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2452 	}
2453 
2454 	/*
2455 	 * Append group record.
2456 	 * If we have sources, we don't know how many yet.
2457 	 */
2458 	mr.mr_type = type;
2459 	mr.mr_datalen = 0;
2460 	mr.mr_numsrc = 0;
2461 	mr.mr_addr = inm->in6m_addr;
2462 	in6_clearscope(&mr.mr_addr);
2463 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2464 		if (m != m0)
2465 			m_freem(m);
2466 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2467 		return (-ENOMEM);
2468 	}
2469 	nbytes += sizeof(struct mldv2_record);
2470 
2471 	/*
2472 	 * Append as many sources as will fit in the first packet.
2473 	 * If we are appending to a new packet, the chain allocation
2474 	 * may potentially use clusters; use m_getptr() in this case.
2475 	 * If we are appending to an existing packet, we need to obtain
2476 	 * a pointer to the group record after m_append(), in case a new
2477 	 * mbuf was allocated.
2478 	 *
2479 	 * Only append sources which are in-mode at t1. If we are
2480 	 * transitioning to MCAST_UNDEFINED state on the group, and
2481 	 * use_block_allow is zero, do not include source entries.
2482 	 * Otherwise, we need to include this source in the report.
2483 	 *
2484 	 * Only report recorded sources in our filter set when responding
2485 	 * to a group-source query.
2486 	 */
2487 	if (record_has_sources) {
2488 		if (m == m0) {
2489 			md = m_last(m);
2490 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2491 			    md->m_len - nbytes);
2492 		} else {
2493 			md = m_getptr(m, 0, &off);
2494 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2495 			    off);
2496 		}
2497 		msrcs = 0;
2498 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2499 		    nims) {
2500 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2501 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2502 			now = im6s_get_mode(inm, ims, 1);
2503 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2504 			if ((now != mode) ||
2505 			    (now == mode &&
2506 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2507 				CTR1(KTR_MLD, "%s: skip node", __func__);
2508 				continue;
2509 			}
2510 			if (is_source_query && ims->im6s_stp == 0) {
2511 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2512 				    __func__);
2513 				continue;
2514 			}
2515 			CTR1(KTR_MLD, "%s: append node", __func__);
2516 			if (!m_append(m, sizeof(struct in6_addr),
2517 			    (void *)&ims->im6s_addr)) {
2518 				if (m != m0)
2519 					m_freem(m);
2520 				CTR1(KTR_MLD, "%s: m_append() failed.",
2521 				    __func__);
2522 				return (-ENOMEM);
2523 			}
2524 			nbytes += sizeof(struct in6_addr);
2525 			++msrcs;
2526 			if (msrcs == m0srcs)
2527 				break;
2528 		}
2529 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2530 		    msrcs);
2531 		pmr->mr_numsrc = htons(msrcs);
2532 		nbytes += (msrcs * sizeof(struct in6_addr));
2533 	}
2534 
2535 	if (is_source_query && msrcs == 0) {
2536 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2537 		if (m != m0)
2538 			m_freem(m);
2539 		return (0);
2540 	}
2541 
2542 	/*
2543 	 * We are good to go with first packet.
2544 	 */
2545 	if (m != m0) {
2546 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2547 		m->m_pkthdr.vt_nrecs = 1;
2548 		mbufq_enqueue(mq, m);
2549 	} else
2550 		m->m_pkthdr.vt_nrecs++;
2551 
2552 	/*
2553 	 * No further work needed if no source list in packet(s).
2554 	 */
2555 	if (!record_has_sources)
2556 		return (nbytes);
2557 
2558 	/*
2559 	 * Whilst sources remain to be announced, we need to allocate
2560 	 * a new packet and fill out as many sources as will fit.
2561 	 * Always try for a cluster first.
2562 	 */
2563 	while (nims != NULL) {
2564 		if (mbufq_full(mq)) {
2565 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2566 			return (-ENOMEM);
2567 		}
2568 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2569 		if (m == NULL)
2570 			m = m_gethdr(M_NOWAIT, MT_DATA);
2571 		if (m == NULL)
2572 			return (-ENOMEM);
2573 		mld_save_context(m, ifp);
2574 		md = m_getptr(m, 0, &off);
2575 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2576 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2577 
2578 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2579 			if (m != m0)
2580 				m_freem(m);
2581 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2582 			return (-ENOMEM);
2583 		}
2584 		m->m_pkthdr.vt_nrecs = 1;
2585 		nbytes += sizeof(struct mldv2_record);
2586 
2587 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2588 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2589 
2590 		msrcs = 0;
2591 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2592 			CTR2(KTR_MLD, "%s: visit node %s",
2593 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2594 			now = im6s_get_mode(inm, ims, 1);
2595 			if ((now != mode) ||
2596 			    (now == mode &&
2597 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2598 				CTR1(KTR_MLD, "%s: skip node", __func__);
2599 				continue;
2600 			}
2601 			if (is_source_query && ims->im6s_stp == 0) {
2602 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2603 				    __func__);
2604 				continue;
2605 			}
2606 			CTR1(KTR_MLD, "%s: append node", __func__);
2607 			if (!m_append(m, sizeof(struct in6_addr),
2608 			    (void *)&ims->im6s_addr)) {
2609 				if (m != m0)
2610 					m_freem(m);
2611 				CTR1(KTR_MLD, "%s: m_append() failed.",
2612 				    __func__);
2613 				return (-ENOMEM);
2614 			}
2615 			++msrcs;
2616 			if (msrcs == m0srcs)
2617 				break;
2618 		}
2619 		pmr->mr_numsrc = htons(msrcs);
2620 		nbytes += (msrcs * sizeof(struct in6_addr));
2621 
2622 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2623 		mbufq_enqueue(mq, m);
2624 	}
2625 
2626 	return (nbytes);
2627 }
2628 
2629 /*
2630  * Type used to mark record pass completion.
2631  * We exploit the fact we can cast to this easily from the
2632  * current filter modes on each ip_msource node.
2633  */
2634 typedef enum {
2635 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2636 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2637 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2638 	REC_FULL = REC_ALLOW | REC_BLOCK
2639 } rectype_t;
2640 
2641 /*
2642  * Enqueue an MLDv2 filter list change to the given output queue.
2643  *
2644  * Source list filter state is held in an RB-tree. When the filter list
2645  * for a group is changed without changing its mode, we need to compute
2646  * the deltas between T0 and T1 for each source in the filter set,
2647  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2648  *
2649  * As we may potentially queue two record types, and the entire R-B tree
2650  * needs to be walked at once, we break this out into its own function
2651  * so we can generate a tightly packed queue of packets.
2652  *
2653  * XXX This could be written to only use one tree walk, although that makes
2654  * serializing into the mbuf chains a bit harder. For now we do two walks
2655  * which makes things easier on us, and it may or may not be harder on
2656  * the L2 cache.
2657  *
2658  * If successful the size of all data appended to the queue is returned,
2659  * otherwise an error code less than zero is returned, or zero if
2660  * no record(s) were appended.
2661  */
2662 static int
2663 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2664 {
2665 	static const int MINRECLEN =
2666 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2667 	struct ifnet		*ifp;
2668 	struct mldv2_record	 mr;
2669 	struct mldv2_record	*pmr;
2670 	struct ip6_msource	*ims, *nims;
2671 	struct mbuf		*m, *m0, *md;
2672 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2673 	uint8_t			 mode, now, then;
2674 	rectype_t		 crt, drt, nrt;
2675 #ifdef KTR
2676 	int			 nallow, nblock;
2677 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2678 #endif
2679 
2680 	IN6_MULTI_LIST_LOCK_ASSERT();
2681 
2682 	if (inm->in6m_nsrc == 0 ||
2683 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2684 		return (0);
2685 
2686 	ifp = inm->in6m_ifp;			/* interface */
2687 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2688 	crt = REC_NONE;	/* current group record type */
2689 	drt = REC_NONE;	/* mask of completed group record types */
2690 	nrt = REC_NONE;	/* record type for current node */
2691 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2692 	npbytes = 0;	/* # of bytes appended this packet */
2693 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2694 	rsrcs = 0;	/* # sources encoded in current record */
2695 	schanged = 0;	/* # nodes encoded in overall filter change */
2696 #ifdef KTR
2697 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2698 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2699 #endif
2700 	nims = NULL;	/* next tree node pointer */
2701 
2702 	/*
2703 	 * For each possible filter record mode.
2704 	 * The first kind of source we encounter tells us which
2705 	 * is the first kind of record we start appending.
2706 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2707 	 * as the inverse of the group's filter mode.
2708 	 */
2709 	while (drt != REC_FULL) {
2710 		do {
2711 			m0 = mbufq_last(mq);
2712 			if (m0 != NULL &&
2713 			    (m0->m_pkthdr.vt_nrecs + 1 <=
2714 			     MLD_V2_REPORT_MAXRECS) &&
2715 			    (m0->m_pkthdr.len + MINRECLEN) <
2716 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2717 				m = m0;
2718 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2719 					    sizeof(struct mldv2_record)) /
2720 					    sizeof(struct in6_addr);
2721 				CTR1(KTR_MLD,
2722 				    "%s: use previous packet", __func__);
2723 			} else {
2724 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2725 				if (m == NULL)
2726 					m = m_gethdr(M_NOWAIT, MT_DATA);
2727 				if (m == NULL) {
2728 					CTR1(KTR_MLD,
2729 					    "%s: m_get*() failed", __func__);
2730 					return (-ENOMEM);
2731 				}
2732 				m->m_pkthdr.vt_nrecs = 0;
2733 				mld_save_context(m, ifp);
2734 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2735 				    sizeof(struct mldv2_record)) /
2736 				    sizeof(struct in6_addr);
2737 				npbytes = 0;
2738 				CTR1(KTR_MLD,
2739 				    "%s: allocated new packet", __func__);
2740 			}
2741 			/*
2742 			 * Append the MLD group record header to the
2743 			 * current packet's data area.
2744 			 * Recalculate pointer to free space for next
2745 			 * group record, in case m_append() allocated
2746 			 * a new mbuf or cluster.
2747 			 */
2748 			memset(&mr, 0, sizeof(mr));
2749 			mr.mr_addr = inm->in6m_addr;
2750 			in6_clearscope(&mr.mr_addr);
2751 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2752 				if (m != m0)
2753 					m_freem(m);
2754 				CTR1(KTR_MLD,
2755 				    "%s: m_append() failed", __func__);
2756 				return (-ENOMEM);
2757 			}
2758 			npbytes += sizeof(struct mldv2_record);
2759 			if (m != m0) {
2760 				/* new packet; offset in chain */
2761 				md = m_getptr(m, npbytes -
2762 				    sizeof(struct mldv2_record), &off);
2763 				pmr = (struct mldv2_record *)(mtod(md,
2764 				    uint8_t *) + off);
2765 			} else {
2766 				/* current packet; offset from last append */
2767 				md = m_last(m);
2768 				pmr = (struct mldv2_record *)(mtod(md,
2769 				    uint8_t *) + md->m_len -
2770 				    sizeof(struct mldv2_record));
2771 			}
2772 			/*
2773 			 * Begin walking the tree for this record type
2774 			 * pass, or continue from where we left off
2775 			 * previously if we had to allocate a new packet.
2776 			 * Only report deltas in-mode at t1.
2777 			 * We need not report included sources as allowed
2778 			 * if we are in inclusive mode on the group,
2779 			 * however the converse is not true.
2780 			 */
2781 			rsrcs = 0;
2782 			if (nims == NULL) {
2783 				nims = RB_MIN(ip6_msource_tree,
2784 				    &inm->in6m_srcs);
2785 			}
2786 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2787 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2788 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2789 				now = im6s_get_mode(inm, ims, 1);
2790 				then = im6s_get_mode(inm, ims, 0);
2791 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2792 				    __func__, then, now);
2793 				if (now == then) {
2794 					CTR1(KTR_MLD,
2795 					    "%s: skip unchanged", __func__);
2796 					continue;
2797 				}
2798 				if (mode == MCAST_EXCLUDE &&
2799 				    now == MCAST_INCLUDE) {
2800 					CTR1(KTR_MLD,
2801 					    "%s: skip IN src on EX group",
2802 					    __func__);
2803 					continue;
2804 				}
2805 				nrt = (rectype_t)now;
2806 				if (nrt == REC_NONE)
2807 					nrt = (rectype_t)(~mode & REC_FULL);
2808 				if (schanged++ == 0) {
2809 					crt = nrt;
2810 				} else if (crt != nrt)
2811 					continue;
2812 				if (!m_append(m, sizeof(struct in6_addr),
2813 				    (void *)&ims->im6s_addr)) {
2814 					if (m != m0)
2815 						m_freem(m);
2816 					CTR1(KTR_MLD,
2817 					    "%s: m_append() failed", __func__);
2818 					return (-ENOMEM);
2819 				}
2820 #ifdef KTR
2821 				nallow += !!(crt == REC_ALLOW);
2822 				nblock += !!(crt == REC_BLOCK);
2823 #endif
2824 				if (++rsrcs == m0srcs)
2825 					break;
2826 			}
2827 			/*
2828 			 * If we did not append any tree nodes on this
2829 			 * pass, back out of allocations.
2830 			 */
2831 			if (rsrcs == 0) {
2832 				npbytes -= sizeof(struct mldv2_record);
2833 				if (m != m0) {
2834 					CTR1(KTR_MLD,
2835 					    "%s: m_free(m)", __func__);
2836 					m_freem(m);
2837 				} else {
2838 					CTR1(KTR_MLD,
2839 					    "%s: m_adj(m, -mr)", __func__);
2840 					m_adj(m, -((int)sizeof(
2841 					    struct mldv2_record)));
2842 				}
2843 				continue;
2844 			}
2845 			npbytes += (rsrcs * sizeof(struct in6_addr));
2846 			if (crt == REC_ALLOW)
2847 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2848 			else if (crt == REC_BLOCK)
2849 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2850 			pmr->mr_numsrc = htons(rsrcs);
2851 			/*
2852 			 * Count the new group record, and enqueue this
2853 			 * packet if it wasn't already queued.
2854 			 */
2855 			m->m_pkthdr.vt_nrecs++;
2856 			if (m != m0)
2857 				mbufq_enqueue(mq, m);
2858 			nbytes += npbytes;
2859 		} while (nims != NULL);
2860 		drt |= crt;
2861 		crt = (~crt & REC_FULL);
2862 	}
2863 
2864 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2865 	    nallow, nblock);
2866 
2867 	return (nbytes);
2868 }
2869 
2870 static int
2871 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2872 {
2873 	struct mbufq	*gq;
2874 	struct mbuf	*m;		/* pending state-change */
2875 	struct mbuf	*m0;		/* copy of pending state-change */
2876 	struct mbuf	*mt;		/* last state-change in packet */
2877 	int		 docopy, domerge;
2878 	u_int		 recslen;
2879 
2880 	docopy = 0;
2881 	domerge = 0;
2882 	recslen = 0;
2883 
2884 	IN6_MULTI_LIST_LOCK_ASSERT();
2885 	MLD_LOCK_ASSERT();
2886 
2887 	/*
2888 	 * If there are further pending retransmissions, make a writable
2889 	 * copy of each queued state-change message before merging.
2890 	 */
2891 	if (inm->in6m_scrv > 0)
2892 		docopy = 1;
2893 
2894 	gq = &inm->in6m_scq;
2895 #ifdef KTR
2896 	if (mbufq_first(gq) == NULL) {
2897 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2898 		    __func__, inm);
2899 	}
2900 #endif
2901 
2902 	m = mbufq_first(gq);
2903 	while (m != NULL) {
2904 		/*
2905 		 * Only merge the report into the current packet if
2906 		 * there is sufficient space to do so; an MLDv2 report
2907 		 * packet may only contain 65,535 group records.
2908 		 * Always use a simple mbuf chain concatentation to do this,
2909 		 * as large state changes for single groups may have
2910 		 * allocated clusters.
2911 		 */
2912 		domerge = 0;
2913 		mt = mbufq_last(scq);
2914 		if (mt != NULL) {
2915 			recslen = m_length(m, NULL);
2916 
2917 			if ((mt->m_pkthdr.vt_nrecs +
2918 			    m->m_pkthdr.vt_nrecs <=
2919 			    MLD_V2_REPORT_MAXRECS) &&
2920 			    (mt->m_pkthdr.len + recslen <=
2921 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2922 				domerge = 1;
2923 		}
2924 
2925 		if (!domerge && mbufq_full(gq)) {
2926 			CTR2(KTR_MLD,
2927 			    "%s: outbound queue full, skipping whole packet %p",
2928 			    __func__, m);
2929 			mt = m->m_nextpkt;
2930 			if (!docopy)
2931 				m_freem(m);
2932 			m = mt;
2933 			continue;
2934 		}
2935 
2936 		if (!docopy) {
2937 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2938 			m0 = mbufq_dequeue(gq);
2939 			m = m0->m_nextpkt;
2940 		} else {
2941 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2942 			m0 = m_dup(m, M_NOWAIT);
2943 			if (m0 == NULL)
2944 				return (ENOMEM);
2945 			m0->m_nextpkt = NULL;
2946 			m = m->m_nextpkt;
2947 		}
2948 
2949 		if (!domerge) {
2950 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2951 			    __func__, m0, scq);
2952 			mbufq_enqueue(scq, m0);
2953 		} else {
2954 			struct mbuf *mtl;	/* last mbuf of packet mt */
2955 
2956 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2957 			    __func__, m0, mt);
2958 
2959 			mtl = m_last(mt);
2960 			m0->m_flags &= ~M_PKTHDR;
2961 			mt->m_pkthdr.len += recslen;
2962 			mt->m_pkthdr.vt_nrecs +=
2963 			    m0->m_pkthdr.vt_nrecs;
2964 
2965 			mtl->m_next = m0;
2966 		}
2967 	}
2968 
2969 	return (0);
2970 }
2971 
2972 /*
2973  * Respond to a pending MLDv2 General Query.
2974  */
2975 static void
2976 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2977 {
2978 	struct ifmultiaddr	*ifma;
2979 	struct ifnet		*ifp;
2980 	struct in6_multi	*inm;
2981 	int			 retval __unused;
2982 
2983 	NET_EPOCH_ASSERT();
2984 	IN6_MULTI_LIST_LOCK_ASSERT();
2985 	MLD_LOCK_ASSERT();
2986 
2987 	KASSERT(mli->mli_version == MLD_VERSION_2,
2988 	    ("%s: called when version %d", __func__, mli->mli_version));
2989 
2990 	/*
2991 	 * Check that there are some packets queued. If so, send them first.
2992 	 * For large number of groups the reply to general query can take
2993 	 * many packets, we should finish sending them before starting of
2994 	 * queuing the new reply.
2995 	 */
2996 	if (mbufq_len(&mli->mli_gq) != 0)
2997 		goto send;
2998 
2999 	ifp = mli->mli_ifp;
3000 
3001 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3002 		inm = in6m_ifmultiaddr_get_inm(ifma);
3003 		if (inm == NULL)
3004 			continue;
3005 		KASSERT(ifp == inm->in6m_ifp,
3006 		    ("%s: inconsistent ifp", __func__));
3007 
3008 		switch (inm->in6m_state) {
3009 		case MLD_NOT_MEMBER:
3010 		case MLD_SILENT_MEMBER:
3011 			break;
3012 		case MLD_REPORTING_MEMBER:
3013 		case MLD_IDLE_MEMBER:
3014 		case MLD_LAZY_MEMBER:
3015 		case MLD_SLEEPING_MEMBER:
3016 		case MLD_AWAKENING_MEMBER:
3017 			inm->in6m_state = MLD_REPORTING_MEMBER;
3018 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3019 			    inm, 0, 0, 0, 0);
3020 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3021 			    __func__, retval);
3022 			break;
3023 		case MLD_G_QUERY_PENDING_MEMBER:
3024 		case MLD_SG_QUERY_PENDING_MEMBER:
3025 		case MLD_LEAVING_MEMBER:
3026 			break;
3027 		}
3028 	}
3029 
3030 send:
3031 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3032 
3033 	/*
3034 	 * Slew transmission of bursts over 500ms intervals.
3035 	 */
3036 	if (mbufq_first(&mli->mli_gq) != NULL) {
3037 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3038 		    MLD_RESPONSE_BURST_INTERVAL);
3039 		V_interface_timers_running6 = 1;
3040 	}
3041 }
3042 
3043 /*
3044  * Transmit the next pending message in the output queue.
3045  *
3046  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3047  * MRT: Nothing needs to be done, as MLD traffic is always local to
3048  * a link and uses a link-scope multicast address.
3049  */
3050 static void
3051 mld_dispatch_packet(struct mbuf *m)
3052 {
3053 	struct ip6_moptions	 im6o;
3054 	struct ifnet		*ifp;
3055 	struct ifnet		*oifp;
3056 	struct mbuf		*m0;
3057 	struct mbuf		*md;
3058 	struct ip6_hdr		*ip6;
3059 	struct mld_hdr		*mld;
3060 	int			 error;
3061 	int			 off;
3062 	int			 type;
3063 	uint32_t		 ifindex;
3064 
3065 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3066 	NET_EPOCH_ASSERT();
3067 
3068 	/*
3069 	 * Set VNET image pointer from enqueued mbuf chain
3070 	 * before doing anything else. Whilst we use interface
3071 	 * indexes to guard against interface detach, they are
3072 	 * unique to each VIMAGE and must be retrieved.
3073 	 */
3074 	ifindex = mld_restore_context(m);
3075 
3076 	/*
3077 	 * Check if the ifnet still exists. This limits the scope of
3078 	 * any race in the absence of a global ifp lock for low cost
3079 	 * (an array lookup).
3080 	 */
3081 	ifp = ifnet_byindex(ifindex);
3082 	if (ifp == NULL) {
3083 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3084 		    __func__, m, ifindex);
3085 		m_freem(m);
3086 		IP6STAT_INC(ip6s_noroute);
3087 		goto out;
3088 	}
3089 
3090 	im6o.im6o_multicast_hlim  = 1;
3091 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3092 	im6o.im6o_multicast_ifp = ifp;
3093 
3094 	if (m->m_flags & M_MLDV1) {
3095 		m0 = m;
3096 	} else {
3097 		m0 = mld_v2_encap_report(ifp, m);
3098 		if (m0 == NULL) {
3099 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3100 			IP6STAT_INC(ip6s_odropped);
3101 			goto out;
3102 		}
3103 	}
3104 
3105 	mld_scrub_context(m0);
3106 	m_clrprotoflags(m);
3107 	m0->m_pkthdr.rcvif = V_loif;
3108 
3109 	ip6 = mtod(m0, struct ip6_hdr *);
3110 #if 0
3111 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3112 #else
3113 	/*
3114 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3115 	 * occur if we called in6_setscope() at transmission.
3116 	 * See comments at top of file.
3117 	 */
3118 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3119 #endif
3120 
3121 	/*
3122 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3123 	 * so we can bump the stats.
3124 	 */
3125 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3126 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3127 	type = mld->mld_type;
3128 
3129 	oifp = NULL;
3130 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3131 	    &oifp, NULL);
3132 	if (error) {
3133 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3134 		goto out;
3135 	}
3136 	ICMP6STAT_INC(icp6s_outhist[type]);
3137 	if (oifp != NULL) {
3138 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3139 		switch (type) {
3140 		case MLD_LISTENER_REPORT:
3141 		case MLDV2_LISTENER_REPORT:
3142 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3143 			break;
3144 		case MLD_LISTENER_DONE:
3145 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3146 			break;
3147 		}
3148 	}
3149 out:
3150 	return;
3151 }
3152 
3153 /*
3154  * Encapsulate an MLDv2 report.
3155  *
3156  * KAME IPv6 requires that hop-by-hop options be passed separately,
3157  * and that the IPv6 header be prepended in a separate mbuf.
3158  *
3159  * Returns a pointer to the new mbuf chain head, or NULL if the
3160  * allocation failed.
3161  */
3162 static struct mbuf *
3163 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3164 {
3165 	struct mbuf		*mh;
3166 	struct mldv2_report	*mld;
3167 	struct ip6_hdr		*ip6;
3168 	struct in6_ifaddr	*ia;
3169 	int			 mldreclen;
3170 
3171 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3172 	KASSERT((m->m_flags & M_PKTHDR),
3173 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3174 
3175 	/*
3176 	 * RFC3590: OK to send as :: or tentative during DAD.
3177 	 */
3178 	NET_EPOCH_ASSERT();
3179 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3180 	if (ia == NULL)
3181 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3182 
3183 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3184 	if (mh == NULL) {
3185 		if (ia != NULL)
3186 			ifa_free(&ia->ia_ifa);
3187 		m_freem(m);
3188 		return (NULL);
3189 	}
3190 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3191 
3192 	mldreclen = m_length(m, NULL);
3193 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3194 
3195 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3196 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3197 	    sizeof(struct mldv2_report) + mldreclen;
3198 
3199 	ip6 = mtod(mh, struct ip6_hdr *);
3200 	ip6->ip6_flow = 0;
3201 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3202 	ip6->ip6_vfc |= IPV6_VERSION;
3203 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3204 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3205 	if (ia != NULL)
3206 		ifa_free(&ia->ia_ifa);
3207 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3208 	/* scope ID will be set in netisr */
3209 
3210 	mld = (struct mldv2_report *)(ip6 + 1);
3211 	mld->mld_type = MLDV2_LISTENER_REPORT;
3212 	mld->mld_code = 0;
3213 	mld->mld_cksum = 0;
3214 	mld->mld_v2_reserved = 0;
3215 	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3216 	m->m_pkthdr.vt_nrecs = 0;
3217 
3218 	mh->m_next = m;
3219 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3220 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3221 	return (mh);
3222 }
3223 
3224 #ifdef KTR
3225 static char *
3226 mld_rec_type_to_str(const int type)
3227 {
3228 
3229 	switch (type) {
3230 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3231 			return "TO_EX";
3232 			break;
3233 		case MLD_CHANGE_TO_INCLUDE_MODE:
3234 			return "TO_IN";
3235 			break;
3236 		case MLD_MODE_IS_EXCLUDE:
3237 			return "MODE_EX";
3238 			break;
3239 		case MLD_MODE_IS_INCLUDE:
3240 			return "MODE_IN";
3241 			break;
3242 		case MLD_ALLOW_NEW_SOURCES:
3243 			return "ALLOW_NEW";
3244 			break;
3245 		case MLD_BLOCK_OLD_SOURCES:
3246 			return "BLOCK_OLD";
3247 			break;
3248 		default:
3249 			break;
3250 	}
3251 	return "unknown";
3252 }
3253 #endif
3254 
3255 static void
3256 mld_init(void *unused __unused)
3257 {
3258 
3259 	CTR1(KTR_MLD, "%s: initializing", __func__);
3260 	MLD_LOCK_INIT();
3261 
3262 	ip6_initpktopts(&mld_po);
3263 	mld_po.ip6po_hlim = 1;
3264 	mld_po.ip6po_hbh = &mld_ra.hbh;
3265 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3266 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3267 
3268 	callout_init(&mldslow_callout, 1);
3269 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
3270 	callout_init(&mldfast_callout, 1);
3271 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
3272 }
3273 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3274 
3275 static void
3276 mld_uninit(void *unused __unused)
3277 {
3278 
3279 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3280 	callout_drain(&mldslow_callout);
3281 	callout_drain(&mldfast_callout);
3282 	MLD_LOCK_DESTROY();
3283 }
3284 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3285 
3286 static void
3287 vnet_mld_init(const void *unused __unused)
3288 {
3289 
3290 	CTR1(KTR_MLD, "%s: initializing", __func__);
3291 
3292 	LIST_INIT(&V_mli_head);
3293 }
3294 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3295     NULL);
3296 
3297 static void
3298 vnet_mld_uninit(const void *unused __unused)
3299 {
3300 
3301 	/* This can happen if we shutdown the network stack. */
3302 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3303 }
3304 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3305     NULL);
3306 
3307 static int
3308 mld_modevent(module_t mod, int type, void *unused __unused)
3309 {
3310 
3311     switch (type) {
3312     case MOD_LOAD:
3313     case MOD_UNLOAD:
3314 	break;
3315     default:
3316 	return (EOPNOTSUPP);
3317     }
3318     return (0);
3319 }
3320 
3321 static moduledata_t mld_mod = {
3322     "mld",
3323     mld_modevent,
3324     0
3325 };
3326 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3327