xref: /freebsd/sys/netinet6/mld6.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote
15  *    products derived from this software without specific prior written
16  *    permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
31  */
32 
33 /*-
34  * Copyright (c) 1988 Stephen Deering.
35  * Copyright (c) 1992, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * This code is derived from software contributed to Berkeley by
39  * Stephen Deering of Stanford University.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
66  */
67 
68 #include <sys/cdefs.h>
69 #include "opt_inet.h"
70 #include "opt_inet6.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/mbuf.h>
75 #include <sys/socket.h>
76 #include <sys/sysctl.h>
77 #include <sys/kernel.h>
78 #include <sys/callout.h>
79 #include <sys/malloc.h>
80 #include <sys/module.h>
81 #include <sys/ktr.h>
82 
83 #include <net/if.h>
84 #include <net/if_var.h>
85 #include <net/if_private.h>
86 #include <net/route.h>
87 #include <net/vnet.h>
88 
89 #include <netinet/in.h>
90 #include <netinet/in_var.h>
91 #include <netinet6/in6_var.h>
92 #include <netinet/ip6.h>
93 #include <netinet6/ip6_var.h>
94 #include <netinet6/scope6_var.h>
95 #include <netinet/icmp6.h>
96 #include <netinet6/mld6.h>
97 #include <netinet6/mld6_var.h>
98 
99 #include <security/mac/mac_framework.h>
100 
101 #ifndef KTR_MLD
102 #define KTR_MLD KTR_INET6
103 #endif
104 
105 static void	mli_delete_locked(struct ifnet *);
106 static void	mld_dispatch_packet(struct mbuf *);
107 static void	mld_dispatch_queue(struct mbufq *, int);
108 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
109 static void	mld_fasttimo_vnet(struct in6_multi_head *inmh);
110 static int	mld_handle_state_change(struct in6_multi *,
111 		    struct mld_ifsoftc *);
112 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
113 		    const int);
114 #ifdef KTR
115 static char *	mld_rec_type_to_str(const int);
116 #endif
117 static void	mld_set_version(struct mld_ifsoftc *, const int);
118 static void	mld_slowtimo_vnet(void);
119 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
120 		    /*const*/ struct mld_hdr *);
121 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
122 		    /*const*/ struct mld_hdr *);
123 static void	mld_v1_process_group_timer(struct in6_multi_head *,
124 		    struct in6_multi *);
125 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
126 static int	mld_v1_transmit_report(struct in6_multi *, const int);
127 static void	mld_v1_update_group(struct in6_multi *, const int);
128 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
129 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
130 static struct mbuf *
131 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
132 static int	mld_v2_enqueue_filter_change(struct mbufq *,
133 		    struct in6_multi *);
134 static int	mld_v2_enqueue_group_record(struct mbufq *,
135 		    struct in6_multi *, const int, const int, const int,
136 		    const int);
137 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
138 		    struct mbuf *, struct mldv2_query *, const int, const int);
139 static int	mld_v2_merge_state_changes(struct in6_multi *,
140 		    struct mbufq *);
141 static void	mld_v2_process_group_timers(struct in6_multi_head *,
142 		    struct mbufq *, struct mbufq *,
143 		    struct in6_multi *, const int);
144 static int	mld_v2_process_group_query(struct in6_multi *,
145 		    struct mld_ifsoftc *mli, int, struct mbuf *,
146 		    struct mldv2_query *, const int);
147 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
148 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
149 
150 /*
151  * Normative references: RFC 2710, RFC 3590, RFC 3810.
152  *
153  * Locking:
154  *  * The MLD subsystem lock ends up being system-wide for the moment,
155  *    but could be per-VIMAGE later on.
156  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
157  *    Any may be taken independently; if any are held at the same
158  *    time, the above lock order must be followed.
159  *  * IN6_MULTI_LOCK covers in_multi.
160  *  * MLD_LOCK covers per-link state and any global variables in this file.
161  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
162  *    per-link state iterators.
163  *
164  *  XXX LOR PREVENTION
165  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
166  *  will not accept an ifp; it wants an embedded scope ID, unlike
167  *  ip_output(), which happily takes the ifp given to it. The embedded
168  *  scope ID is only used by MLD to select the outgoing interface.
169  *
170  *  During interface attach and detach, MLD will take MLD_LOCK *after*
171  *  the IF_AFDATA_LOCK.
172  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
173  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
174  *  dispatch could work around this, but we'd rather not do that, as it
175  *  can introduce other races.
176  *
177  *  As such, we exploit the fact that the scope ID is just the interface
178  *  index, and embed it in the IPv6 destination address accordingly.
179  *  This is potentially NOT VALID for MLDv1 reports, as they
180  *  are always sent to the multicast group itself; as MLDv2
181  *  reports are always sent to ff02::16, this is not an issue
182  *  when MLDv2 is in use.
183  *
184  *  This does not however eliminate the LOR when ip6_output() itself
185  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
186  *  trigger a LOR warning in WITNESS when the ifnet is detached.
187  *
188  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
189  *  how it's used across the network stack. Here we're simply exploiting
190  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
191  *
192  * VIMAGE:
193  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
194  *    to a vnet in ifp->if_vnet.
195  */
196 static struct mtx		 mld_mtx;
197 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
198 
199 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
200 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
201 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
202 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
203 
204 /*
205  * VIMAGE-wide globals.
206  */
207 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
208 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
209 VNET_DEFINE_STATIC(int, interface_timers_running6);
210 VNET_DEFINE_STATIC(int, state_change_timers_running6);
211 VNET_DEFINE_STATIC(int, current_state_timers_running6);
212 
213 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
214 #define	V_mli_head			VNET(mli_head)
215 #define	V_interface_timers_running6	VNET(interface_timers_running6)
216 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
217 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
218 
219 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
220 
221 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
222     "IPv6 Multicast Listener Discovery");
223 
224 /*
225  * Virtualized sysctls.
226  */
227 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
228     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
229     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
230     "Rate limit for MLDv2 Group-and-Source queries in seconds");
231 
232 /*
233  * Non-virtualized sysctls.
234  */
235 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
236     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
237     "Per-interface MLDv2 state");
238 
239 static int	mld_v1enable = 1;
240 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
241     &mld_v1enable, 0, "Enable fallback to MLDv1");
242 
243 static int	mld_v2enable = 1;
244 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
245     &mld_v2enable, 0, "Enable MLDv2");
246 
247 static int	mld_use_allow = 1;
248 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
249     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
250 
251 /*
252  * Packed Router Alert option structure declaration.
253  */
254 struct mld_raopt {
255 	struct ip6_hbh		hbh;
256 	struct ip6_opt		pad;
257 	struct ip6_opt_router	ra;
258 } __packed;
259 
260 /*
261  * Router Alert hop-by-hop option header.
262  */
263 static struct mld_raopt mld_ra = {
264 	.hbh = { 0, 0 },
265 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
266 	.ra = {
267 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
268 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
269 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
270 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
271 	}
272 };
273 static struct ip6_pktopts mld_po;
274 
275 static __inline void
276 mld_save_context(struct mbuf *m, struct ifnet *ifp)
277 {
278 
279 #ifdef VIMAGE
280 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
281 #endif /* VIMAGE */
282 	m->m_pkthdr.rcvif = ifp;
283 	m->m_pkthdr.flowid = ifp->if_index;
284 }
285 
286 static __inline void
287 mld_scrub_context(struct mbuf *m)
288 {
289 
290 	m->m_pkthdr.PH_loc.ptr = NULL;
291 	m->m_pkthdr.flowid = 0;
292 }
293 
294 /*
295  * Restore context from a queued output chain.
296  * Return saved ifindex.
297  *
298  * VIMAGE: The assertion is there to make sure that we
299  * actually called CURVNET_SET() with what's in the mbuf chain.
300  */
301 static __inline uint32_t
302 mld_restore_context(struct mbuf *m)
303 {
304 
305 #if defined(VIMAGE) && defined(INVARIANTS)
306 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
307 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
308 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
309 #endif
310 	return (m->m_pkthdr.flowid);
311 }
312 
313 /*
314  * Retrieve or set threshold between group-source queries in seconds.
315  *
316  * VIMAGE: Assume curvnet set by caller.
317  * SMPng: NOTE: Serialized by MLD lock.
318  */
319 static int
320 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
321 {
322 	int error;
323 	int i;
324 
325 	error = sysctl_wire_old_buffer(req, sizeof(int));
326 	if (error)
327 		return (error);
328 
329 	MLD_LOCK();
330 
331 	i = V_mld_gsrdelay.tv_sec;
332 
333 	error = sysctl_handle_int(oidp, &i, 0, req);
334 	if (error || !req->newptr)
335 		goto out_locked;
336 
337 	if (i < -1 || i >= 60) {
338 		error = EINVAL;
339 		goto out_locked;
340 	}
341 
342 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
343 	     V_mld_gsrdelay.tv_sec, i);
344 	V_mld_gsrdelay.tv_sec = i;
345 
346 out_locked:
347 	MLD_UNLOCK();
348 	return (error);
349 }
350 
351 /*
352  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
353  * For use by ifmcstat(8).
354  *
355  * VIMAGE: Assume curvnet set by caller. The node handler itself
356  * is not directly virtualized.
357  */
358 static int
359 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
360 {
361 	struct epoch_tracker	 et;
362 	int			*name;
363 	int			 error;
364 	u_int			 namelen;
365 	struct ifnet		*ifp;
366 	struct mld_ifsoftc	*mli;
367 
368 	name = (int *)arg1;
369 	namelen = arg2;
370 
371 	if (req->newptr != NULL)
372 		return (EPERM);
373 
374 	if (namelen != 1)
375 		return (EINVAL);
376 
377 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
378 	if (error)
379 		return (error);
380 
381 	IN6_MULTI_LOCK();
382 	IN6_MULTI_LIST_LOCK();
383 	MLD_LOCK();
384 	NET_EPOCH_ENTER(et);
385 
386 	error = ENOENT;
387 	ifp = ifnet_byindex(name[0]);
388 	if (ifp == NULL)
389 		goto out_locked;
390 
391 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
392 		if (ifp == mli->mli_ifp) {
393 			struct mld_ifinfo info;
394 
395 			info.mli_version = mli->mli_version;
396 			info.mli_v1_timer = mli->mli_v1_timer;
397 			info.mli_v2_timer = mli->mli_v2_timer;
398 			info.mli_flags = mli->mli_flags;
399 			info.mli_rv = mli->mli_rv;
400 			info.mli_qi = mli->mli_qi;
401 			info.mli_qri = mli->mli_qri;
402 			info.mli_uri = mli->mli_uri;
403 			error = SYSCTL_OUT(req, &info, sizeof(info));
404 			break;
405 		}
406 	}
407 
408 out_locked:
409 	NET_EPOCH_EXIT(et);
410 	MLD_UNLOCK();
411 	IN6_MULTI_LIST_UNLOCK();
412 	IN6_MULTI_UNLOCK();
413 	return (error);
414 }
415 
416 /*
417  * Dispatch an entire queue of pending packet chains.
418  * VIMAGE: Assumes the vnet pointer has been set.
419  */
420 static void
421 mld_dispatch_queue(struct mbufq *mq, int limit)
422 {
423 	struct mbuf *m;
424 
425 	while ((m = mbufq_dequeue(mq)) != NULL) {
426 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
427 		mld_dispatch_packet(m);
428 		if (--limit == 0)
429 			break;
430 	}
431 }
432 
433 /*
434  * Filter outgoing MLD report state by group.
435  *
436  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
437  * and node-local addresses. However, kernel and socket consumers
438  * always embed the KAME scope ID in the address provided, so strip it
439  * when performing comparison.
440  * Note: This is not the same as the *multicast* scope.
441  *
442  * Return zero if the given group is one for which MLD reports
443  * should be suppressed, or non-zero if reports should be issued.
444  */
445 static __inline int
446 mld_is_addr_reported(const struct in6_addr *addr)
447 {
448 
449 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
450 
451 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
452 		return (0);
453 
454 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
455 		struct in6_addr tmp = *addr;
456 		in6_clearscope(&tmp);
457 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
458 			return (0);
459 	}
460 
461 	return (1);
462 }
463 
464 /*
465  * Attach MLD when PF_INET6 is attached to an interface.  Assumes that the
466  * current VNET is set by the caller.
467  */
468 struct mld_ifsoftc *
469 mld_domifattach(struct ifnet *ifp)
470 {
471 	struct mld_ifsoftc *mli;
472 
473 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp));
474 
475 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_WAITOK | M_ZERO);
476 	mli->mli_ifp = ifp;
477 	mli->mli_version = MLD_VERSION_2;
478 	mli->mli_flags = 0;
479 	mli->mli_rv = MLD_RV_INIT;
480 	mli->mli_qi = MLD_QI_INIT;
481 	mli->mli_qri = MLD_QRI_INIT;
482 	mli->mli_uri = MLD_URI_INIT;
483 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
484 	if ((ifp->if_flags & IFF_MULTICAST) == 0)
485 		mli->mli_flags |= MLIF_SILENT;
486 	if (mld_use_allow)
487 		mli->mli_flags |= MLIF_USEALLOW;
488 
489 	MLD_LOCK();
490 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
491 	MLD_UNLOCK();
492 
493 	return (mli);
494 }
495 
496 /*
497  * Hook for ifdetach.
498  *
499  * NOTE: Some finalization tasks need to run before the protocol domain
500  * is detached, but also before the link layer does its cleanup.
501  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
502  *
503  * SMPng: Caller must hold IN6_MULTI_LOCK().
504  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
505  * XXX This routine is also bitten by unlocked ifma_protospec access.
506  */
507 void
508 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
509 {
510 	struct epoch_tracker     et;
511 	struct mld_ifsoftc	*mli;
512 	struct ifmultiaddr	*ifma;
513 	struct in6_multi	*inm;
514 
515 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
516 	    if_name(ifp));
517 
518 	IN6_MULTI_LIST_LOCK_ASSERT();
519 	MLD_LOCK();
520 
521 	mli = MLD_IFINFO(ifp);
522 	IF_ADDR_WLOCK(ifp);
523 	/*
524 	 * Extract list of in6_multi associated with the detaching ifp
525 	 * which the PF_INET6 layer is about to release.
526 	 */
527 	NET_EPOCH_ENTER(et);
528 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
529 		inm = in6m_ifmultiaddr_get_inm(ifma);
530 		if (inm == NULL)
531 			continue;
532 		in6m_disconnect_locked(inmh, inm);
533 
534 		if (mli->mli_version == MLD_VERSION_2) {
535 			in6m_clear_recorded(inm);
536 
537 			/*
538 			 * We need to release the final reference held
539 			 * for issuing the INCLUDE {}.
540 			 */
541 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
542 				inm->in6m_state = MLD_NOT_MEMBER;
543 				in6m_rele_locked(inmh, inm);
544 			}
545 		}
546 	}
547 	NET_EPOCH_EXIT(et);
548 	IF_ADDR_WUNLOCK(ifp);
549 	MLD_UNLOCK();
550 }
551 
552 /*
553  * Hook for domifdetach.
554  * Runs after link-layer cleanup; free MLD state.
555  *
556  * SMPng: Normally called with IF_AFDATA_LOCK held.
557  */
558 void
559 mld_domifdetach(struct ifnet *ifp)
560 {
561 
562 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
563 	    __func__, ifp, if_name(ifp));
564 
565 	MLD_LOCK();
566 	mli_delete_locked(ifp);
567 	MLD_UNLOCK();
568 }
569 
570 static void
571 mli_delete_locked(struct ifnet *ifp)
572 {
573 	struct mld_ifsoftc *mli, *tmli;
574 
575 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
576 	    __func__, ifp, if_name(ifp));
577 
578 	MLD_LOCK_ASSERT();
579 
580 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
581 		if (mli->mli_ifp == ifp) {
582 			/*
583 			 * Free deferred General Query responses.
584 			 */
585 			mbufq_drain(&mli->mli_gq);
586 
587 			LIST_REMOVE(mli, mli_link);
588 
589 			free(mli, M_MLD);
590 			return;
591 		}
592 	}
593 }
594 
595 /*
596  * Process a received MLDv1 general or address-specific query.
597  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
598  *
599  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
600  * mld_addr. This is OK as we own the mbuf chain.
601  */
602 static int
603 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
604     /*const*/ struct mld_hdr *mld)
605 {
606 	struct ifmultiaddr	*ifma;
607 	struct mld_ifsoftc	*mli;
608 	struct in6_multi	*inm;
609 	int			 is_general_query;
610 	uint16_t		 timer;
611 #ifdef KTR
612 	char			 ip6tbuf[INET6_ADDRSTRLEN];
613 #endif
614 
615 	NET_EPOCH_ASSERT();
616 
617 	is_general_query = 0;
618 
619 	if (!mld_v1enable) {
620 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
621 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
622 		    ifp, if_name(ifp));
623 		return (0);
624 	}
625 
626 	/*
627 	 * RFC3810 Section 6.2: MLD queries must originate from
628 	 * a router's link-local address.
629 	 */
630 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
631 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
632 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
633 		    ifp, if_name(ifp));
634 		return (0);
635 	}
636 
637 	/*
638 	 * Do address field validation upfront before we accept
639 	 * the query.
640 	 */
641 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
642 		/*
643 		 * MLDv1 General Query.
644 		 * If this was not sent to the all-nodes group, ignore it.
645 		 */
646 		struct in6_addr		 dst;
647 
648 		dst = ip6->ip6_dst;
649 		in6_clearscope(&dst);
650 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
651 			return (EINVAL);
652 		is_general_query = 1;
653 	} else {
654 		/*
655 		 * Embed scope ID of receiving interface in MLD query for
656 		 * lookup whilst we don't hold other locks.
657 		 */
658 		in6_setscope(&mld->mld_addr, ifp, NULL);
659 	}
660 
661 	IN6_MULTI_LIST_LOCK();
662 	MLD_LOCK();
663 
664 	/*
665 	 * Switch to MLDv1 host compatibility mode.
666 	 */
667 	mli = MLD_IFINFO(ifp);
668 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
669 	mld_set_version(mli, MLD_VERSION_1);
670 
671 	timer = (ntohs(mld->mld_maxdelay) * MLD_FASTHZ) / MLD_TIMER_SCALE;
672 	if (timer == 0)
673 		timer = 1;
674 
675 	if (is_general_query) {
676 		/*
677 		 * For each reporting group joined on this
678 		 * interface, kick the report timer.
679 		 */
680 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
681 			 ifp, if_name(ifp));
682 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
683 			inm = in6m_ifmultiaddr_get_inm(ifma);
684 			if (inm == NULL)
685 				continue;
686 			mld_v1_update_group(inm, timer);
687 		}
688 	} else {
689 		/*
690 		 * MLDv1 Group-Specific Query.
691 		 * If this is a group-specific MLDv1 query, we need only
692 		 * look up the single group to process it.
693 		 */
694 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
695 		if (inm != NULL) {
696 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
697 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
698 			    ifp, if_name(ifp));
699 			mld_v1_update_group(inm, timer);
700 		}
701 		/* XXX Clear embedded scope ID as userland won't expect it. */
702 		in6_clearscope(&mld->mld_addr);
703 	}
704 
705 	MLD_UNLOCK();
706 	IN6_MULTI_LIST_UNLOCK();
707 
708 	return (0);
709 }
710 
711 /*
712  * Update the report timer on a group in response to an MLDv1 query.
713  *
714  * If we are becoming the reporting member for this group, start the timer.
715  * If we already are the reporting member for this group, and timer is
716  * below the threshold, reset it.
717  *
718  * We may be updating the group for the first time since we switched
719  * to MLDv2. If we are, then we must clear any recorded source lists,
720  * and transition to REPORTING state; the group timer is overloaded
721  * for group and group-source query responses.
722  *
723  * Unlike MLDv2, the delay per group should be jittered
724  * to avoid bursts of MLDv1 reports.
725  */
726 static void
727 mld_v1_update_group(struct in6_multi *inm, const int timer)
728 {
729 #ifdef KTR
730 	char			 ip6tbuf[INET6_ADDRSTRLEN];
731 #endif
732 
733 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
734 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
735 	    if_name(inm->in6m_ifp), timer);
736 
737 	IN6_MULTI_LIST_LOCK_ASSERT();
738 
739 	switch (inm->in6m_state) {
740 	case MLD_NOT_MEMBER:
741 	case MLD_SILENT_MEMBER:
742 		break;
743 	case MLD_REPORTING_MEMBER:
744 		if (inm->in6m_timer != 0 &&
745 		    inm->in6m_timer <= timer) {
746 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
747 			    "skipping.", __func__);
748 			break;
749 		}
750 		/* FALLTHROUGH */
751 	case MLD_SG_QUERY_PENDING_MEMBER:
752 	case MLD_G_QUERY_PENDING_MEMBER:
753 	case MLD_IDLE_MEMBER:
754 	case MLD_LAZY_MEMBER:
755 	case MLD_AWAKENING_MEMBER:
756 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
757 		inm->in6m_state = MLD_REPORTING_MEMBER;
758 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
759 		V_current_state_timers_running6 = 1;
760 		break;
761 	case MLD_SLEEPING_MEMBER:
762 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
763 		inm->in6m_state = MLD_AWAKENING_MEMBER;
764 		break;
765 	case MLD_LEAVING_MEMBER:
766 		break;
767 	}
768 }
769 
770 /*
771  * Process a received MLDv2 general, group-specific or
772  * group-and-source-specific query.
773  *
774  * Assumes that mld points to a struct mldv2_query which is stored in
775  * contiguous memory.
776  *
777  * Return 0 if successful, otherwise an appropriate error code is returned.
778  */
779 static int
780 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
781     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
782 {
783 	struct mld_ifsoftc	*mli;
784 	struct in6_multi	*inm;
785 	uint32_t		 maxdelay, nsrc, qqi;
786 	int			 is_general_query;
787 	uint16_t		 timer;
788 	uint8_t			 qrv;
789 #ifdef KTR
790 	char			 ip6tbuf[INET6_ADDRSTRLEN];
791 #endif
792 
793 	NET_EPOCH_ASSERT();
794 
795 	if (!mld_v2enable) {
796 		CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
797 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
798 		    ifp, if_name(ifp));
799 		return (0);
800 	}
801 
802 	/*
803 	 * RFC3810 Section 6.2: MLD queries must originate from
804 	 * a router's link-local address.
805 	 */
806 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
807 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
808 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
809 		    ifp, if_name(ifp));
810 		return (0);
811 	}
812 
813 	is_general_query = 0;
814 
815 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
816 
817 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
818 	if (maxdelay >= 32768) {
819 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
820 			   (MLD_MRC_EXP(maxdelay) + 3);
821 	}
822 	timer = (maxdelay * MLD_FASTHZ) / MLD_TIMER_SCALE;
823 	if (timer == 0)
824 		timer = 1;
825 
826 	qrv = MLD_QRV(mld->mld_misc);
827 	if (qrv < 2) {
828 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
829 		    qrv, MLD_RV_INIT);
830 		qrv = MLD_RV_INIT;
831 	}
832 
833 	qqi = mld->mld_qqi;
834 	if (qqi >= 128) {
835 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
836 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
837 	}
838 
839 	nsrc = ntohs(mld->mld_numsrc);
840 	if (nsrc > MLD_MAX_GS_SOURCES)
841 		return (EMSGSIZE);
842 	if (icmp6len < sizeof(struct mldv2_query) +
843 	    (nsrc * sizeof(struct in6_addr)))
844 		return (EMSGSIZE);
845 
846 	/*
847 	 * Do further input validation upfront to avoid resetting timers
848 	 * should we need to discard this query.
849 	 */
850 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
851 		/*
852 		 * A general query with a source list has undefined
853 		 * behaviour; discard it.
854 		 */
855 		if (nsrc > 0)
856 			return (EINVAL);
857 		is_general_query = 1;
858 	} else {
859 		/*
860 		 * Embed scope ID of receiving interface in MLD query for
861 		 * lookup whilst we don't hold other locks (due to KAME
862 		 * locking lameness). We own this mbuf chain just now.
863 		 */
864 		in6_setscope(&mld->mld_addr, ifp, NULL);
865 	}
866 
867 	IN6_MULTI_LIST_LOCK();
868 	MLD_LOCK();
869 
870 	mli = MLD_IFINFO(ifp);
871 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
872 
873 	/*
874 	 * Discard the v2 query if we're in Compatibility Mode.
875 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
876 	 * until the Old Version Querier Present timer expires.
877 	 */
878 	if (mli->mli_version != MLD_VERSION_2)
879 		goto out_locked;
880 
881 	mld_set_version(mli, MLD_VERSION_2);
882 	mli->mli_rv = qrv;
883 	mli->mli_qi = qqi;
884 	mli->mli_qri = maxdelay;
885 
886 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
887 	    maxdelay);
888 
889 	if (is_general_query) {
890 		/*
891 		 * MLDv2 General Query.
892 		 *
893 		 * Schedule a current-state report on this ifp for
894 		 * all groups, possibly containing source lists.
895 		 *
896 		 * If there is a pending General Query response
897 		 * scheduled earlier than the selected delay, do
898 		 * not schedule any other reports.
899 		 * Otherwise, reset the interface timer.
900 		 */
901 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
902 		    ifp, if_name(ifp));
903 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
904 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
905 			V_interface_timers_running6 = 1;
906 		}
907 	} else {
908 		/*
909 		 * MLDv2 Group-specific or Group-and-source-specific Query.
910 		 *
911 		 * Group-source-specific queries are throttled on
912 		 * a per-group basis to defeat denial-of-service attempts.
913 		 * Queries for groups we are not a member of on this
914 		 * link are simply ignored.
915 		 */
916 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
917 		if (inm == NULL)
918 			goto out_locked;
919 		if (nsrc > 0) {
920 			if (!ratecheck(&inm->in6m_lastgsrtv,
921 			    &V_mld_gsrdelay)) {
922 				CTR1(KTR_MLD, "%s: GS query throttled.",
923 				    __func__);
924 				goto out_locked;
925 			}
926 		}
927 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
928 		     ifp, if_name(ifp));
929 		/*
930 		 * If there is a pending General Query response
931 		 * scheduled sooner than the selected delay, no
932 		 * further report need be scheduled.
933 		 * Otherwise, prepare to respond to the
934 		 * group-specific or group-and-source query.
935 		 */
936 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
937 			mld_v2_process_group_query(inm, mli, timer, m, mld, off);
938 
939 		/* XXX Clear embedded scope ID as userland won't expect it. */
940 		in6_clearscope(&mld->mld_addr);
941 	}
942 
943 out_locked:
944 	MLD_UNLOCK();
945 	IN6_MULTI_LIST_UNLOCK();
946 
947 	return (0);
948 }
949 
950 /*
951  * Process a received MLDv2 group-specific or group-and-source-specific
952  * query.
953  * Return <0 if any error occurred. Currently this is ignored.
954  */
955 static int
956 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
957     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
958 {
959 	int			 retval;
960 	uint16_t		 nsrc;
961 
962 	IN6_MULTI_LIST_LOCK_ASSERT();
963 	MLD_LOCK_ASSERT();
964 
965 	retval = 0;
966 
967 	switch (inm->in6m_state) {
968 	case MLD_NOT_MEMBER:
969 	case MLD_SILENT_MEMBER:
970 	case MLD_SLEEPING_MEMBER:
971 	case MLD_LAZY_MEMBER:
972 	case MLD_AWAKENING_MEMBER:
973 	case MLD_IDLE_MEMBER:
974 	case MLD_LEAVING_MEMBER:
975 		return (retval);
976 		break;
977 	case MLD_REPORTING_MEMBER:
978 	case MLD_G_QUERY_PENDING_MEMBER:
979 	case MLD_SG_QUERY_PENDING_MEMBER:
980 		break;
981 	}
982 
983 	nsrc = ntohs(mld->mld_numsrc);
984 
985 	/* Length should be checked by calling function. */
986 	KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
987 	    m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
988 	    nsrc * sizeof(struct in6_addr),
989 	    ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
990 	    m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
991 	    nsrc * sizeof(struct in6_addr), m0));
992 
993 	/*
994 	 * Deal with group-specific queries upfront.
995 	 * If any group query is already pending, purge any recorded
996 	 * source-list state if it exists, and schedule a query response
997 	 * for this group-specific query.
998 	 */
999 	if (nsrc == 0) {
1000 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1001 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1002 			in6m_clear_recorded(inm);
1003 			timer = min(inm->in6m_timer, timer);
1004 		}
1005 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1006 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1007 		V_current_state_timers_running6 = 1;
1008 		return (retval);
1009 	}
1010 
1011 	/*
1012 	 * Deal with the case where a group-and-source-specific query has
1013 	 * been received but a group-specific query is already pending.
1014 	 */
1015 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1016 		timer = min(inm->in6m_timer, timer);
1017 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1018 		V_current_state_timers_running6 = 1;
1019 		return (retval);
1020 	}
1021 
1022 	/*
1023 	 * Finally, deal with the case where a group-and-source-specific
1024 	 * query has been received, where a response to a previous g-s-r
1025 	 * query exists, or none exists.
1026 	 * In this case, we need to parse the source-list which the Querier
1027 	 * has provided us with and check if we have any source list filter
1028 	 * entries at T1 for these sources. If we do not, there is no need
1029 	 * schedule a report and the query may be dropped.
1030 	 * If we do, we must record them and schedule a current-state
1031 	 * report for those sources.
1032 	 */
1033 	if (inm->in6m_nsrc > 0) {
1034 		struct in6_addr		 srcaddr;
1035 		int			 i, nrecorded;
1036 		int			 soff;
1037 
1038 		soff = off + sizeof(struct mldv2_query);
1039 		nrecorded = 0;
1040 		for (i = 0; i < nsrc; i++) {
1041 			m_copydata(m0, soff, sizeof(struct in6_addr),
1042 			    (caddr_t)&srcaddr);
1043 			retval = in6m_record_source(inm, &srcaddr);
1044 			if (retval < 0)
1045 				break;
1046 			nrecorded += retval;
1047 			soff += sizeof(struct in6_addr);
1048 		}
1049 		if (nrecorded > 0) {
1050 			CTR1(KTR_MLD,
1051 			    "%s: schedule response to SG query", __func__);
1052 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1053 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1054 			V_current_state_timers_running6 = 1;
1055 		}
1056 	}
1057 
1058 	return (retval);
1059 }
1060 
1061 /*
1062  * Process a received MLDv1 host membership report.
1063  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1064  *
1065  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1066  * mld_addr. This is OK as we own the mbuf chain.
1067  */
1068 static int
1069 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1070     /*const*/ struct mld_hdr *mld)
1071 {
1072 	struct in6_addr		 src, dst;
1073 	struct in6_ifaddr	*ia;
1074 	struct in6_multi	*inm;
1075 #ifdef KTR
1076 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1077 #endif
1078 
1079 	NET_EPOCH_ASSERT();
1080 
1081 	if (!mld_v1enable) {
1082 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1083 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1084 		    ifp, if_name(ifp));
1085 		return (0);
1086 	}
1087 
1088 	if (ifp->if_flags & IFF_LOOPBACK)
1089 		return (0);
1090 
1091 	/*
1092 	 * MLDv1 reports must originate from a host's link-local address,
1093 	 * or the unspecified address (when booting).
1094 	 */
1095 	src = ip6->ip6_src;
1096 	in6_clearscope(&src);
1097 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1098 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1099 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1100 		    ifp, if_name(ifp));
1101 		return (EINVAL);
1102 	}
1103 
1104 	/*
1105 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1106 	 * group, and must be directed to the group itself.
1107 	 */
1108 	dst = ip6->ip6_dst;
1109 	in6_clearscope(&dst);
1110 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1111 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1112 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1113 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1114 		    ifp, if_name(ifp));
1115 		return (EINVAL);
1116 	}
1117 
1118 	/*
1119 	 * Make sure we don't hear our own membership report, as fast
1120 	 * leave requires knowing that we are the only member of a
1121 	 * group. Assume we used the link-local address if available,
1122 	 * otherwise look for ::.
1123 	 *
1124 	 * XXX Note that scope ID comparison is needed for the address
1125 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1126 	 * performed for the on-wire address.
1127 	 */
1128 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1129 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1130 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1131 		if (ia != NULL)
1132 			ifa_free(&ia->ia_ifa);
1133 		return (0);
1134 	}
1135 	if (ia != NULL)
1136 		ifa_free(&ia->ia_ifa);
1137 
1138 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1139 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1140 
1141 	/*
1142 	 * Embed scope ID of receiving interface in MLD query for lookup
1143 	 * whilst we don't hold other locks (due to KAME locking lameness).
1144 	 */
1145 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1146 		in6_setscope(&mld->mld_addr, ifp, NULL);
1147 
1148 	IN6_MULTI_LIST_LOCK();
1149 	MLD_LOCK();
1150 
1151 	/*
1152 	 * MLDv1 report suppression.
1153 	 * If we are a member of this group, and our membership should be
1154 	 * reported, and our group timer is pending or about to be reset,
1155 	 * stop our group timer by transitioning to the 'lazy' state.
1156 	 */
1157 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1158 	if (inm != NULL) {
1159 		struct mld_ifsoftc *mli;
1160 
1161 		mli = inm->in6m_mli;
1162 		KASSERT(mli != NULL,
1163 		    ("%s: no mli for ifp %p", __func__, ifp));
1164 
1165 		/*
1166 		 * If we are in MLDv2 host mode, do not allow the
1167 		 * other host's MLDv1 report to suppress our reports.
1168 		 */
1169 		if (mli->mli_version == MLD_VERSION_2)
1170 			goto out_locked;
1171 
1172 		inm->in6m_timer = 0;
1173 
1174 		switch (inm->in6m_state) {
1175 		case MLD_NOT_MEMBER:
1176 		case MLD_SILENT_MEMBER:
1177 		case MLD_SLEEPING_MEMBER:
1178 			break;
1179 		case MLD_REPORTING_MEMBER:
1180 		case MLD_IDLE_MEMBER:
1181 		case MLD_AWAKENING_MEMBER:
1182 			CTR3(KTR_MLD,
1183 			    "report suppressed for %s on ifp %p(%s)",
1184 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1185 			    ifp, if_name(ifp));
1186 		case MLD_LAZY_MEMBER:
1187 			inm->in6m_state = MLD_LAZY_MEMBER;
1188 			break;
1189 		case MLD_G_QUERY_PENDING_MEMBER:
1190 		case MLD_SG_QUERY_PENDING_MEMBER:
1191 		case MLD_LEAVING_MEMBER:
1192 			break;
1193 		}
1194 	}
1195 
1196 out_locked:
1197 	MLD_UNLOCK();
1198 	IN6_MULTI_LIST_UNLOCK();
1199 
1200 	/* XXX Clear embedded scope ID as userland won't expect it. */
1201 	in6_clearscope(&mld->mld_addr);
1202 
1203 	return (0);
1204 }
1205 
1206 /*
1207  * MLD input path.
1208  *
1209  * Assume query messages which fit in a single ICMPv6 message header
1210  * have been pulled up.
1211  * Assume that userland will want to see the message, even if it
1212  * otherwise fails kernel input validation; do not free it.
1213  * Pullup may however free the mbuf chain m if it fails.
1214  *
1215  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1216  */
1217 int
1218 mld_input(struct mbuf **mp, int off, int icmp6len)
1219 {
1220 	struct ifnet	*ifp;
1221 	struct ip6_hdr	*ip6;
1222 	struct mbuf	*m;
1223 	struct mld_hdr	*mld;
1224 	int		 mldlen;
1225 
1226 	m = *mp;
1227 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1228 
1229 	ifp = m->m_pkthdr.rcvif;
1230 
1231 	/* Pullup to appropriate size. */
1232 	if (m->m_len < off + sizeof(*mld)) {
1233 		m = m_pullup(m, off + sizeof(*mld));
1234 		if (m == NULL) {
1235 			ICMP6STAT_INC(icp6s_badlen);
1236 			return (IPPROTO_DONE);
1237 		}
1238 	}
1239 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1240 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1241 	    icmp6len >= sizeof(struct mldv2_query)) {
1242 		mldlen = sizeof(struct mldv2_query);
1243 	} else {
1244 		mldlen = sizeof(struct mld_hdr);
1245 	}
1246 	if (m->m_len < off + mldlen) {
1247 		m = m_pullup(m, off + mldlen);
1248 		if (m == NULL) {
1249 			ICMP6STAT_INC(icp6s_badlen);
1250 			return (IPPROTO_DONE);
1251 		}
1252 	}
1253 	*mp = m;
1254 	ip6 = mtod(m, struct ip6_hdr *);
1255 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1256 
1257 	/*
1258 	 * Userland needs to see all of this traffic for implementing
1259 	 * the endpoint discovery portion of multicast routing.
1260 	 */
1261 	switch (mld->mld_type) {
1262 	case MLD_LISTENER_QUERY:
1263 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1264 		if (icmp6len == sizeof(struct mld_hdr)) {
1265 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1266 				return (0);
1267 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1268 			if (mld_v2_input_query(ifp, ip6, m,
1269 			    (struct mldv2_query *)mld, off, icmp6len) != 0)
1270 				return (0);
1271 		}
1272 		break;
1273 	case MLD_LISTENER_REPORT:
1274 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1275 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1276 			return (0);
1277 		break;
1278 	case MLDV2_LISTENER_REPORT:
1279 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1280 		break;
1281 	case MLD_LISTENER_DONE:
1282 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1283 		break;
1284 	default:
1285 		break;
1286 	}
1287 
1288 	return (0);
1289 }
1290 
1291 /*
1292  * Fast timeout handler (global).
1293  * VIMAGE: Timeout handlers are expected to service all vimages.
1294  */
1295 static struct callout mldfast_callout;
1296 static void
1297 mld_fasttimo(void *arg __unused)
1298 {
1299 	struct epoch_tracker et;
1300 	struct in6_multi_head inmh;
1301 	VNET_ITERATOR_DECL(vnet_iter);
1302 
1303 	SLIST_INIT(&inmh);
1304 
1305 	NET_EPOCH_ENTER(et);
1306 	VNET_LIST_RLOCK_NOSLEEP();
1307 	VNET_FOREACH(vnet_iter) {
1308 		CURVNET_SET(vnet_iter);
1309 		mld_fasttimo_vnet(&inmh);
1310 		CURVNET_RESTORE();
1311 	}
1312 	VNET_LIST_RUNLOCK_NOSLEEP();
1313 	NET_EPOCH_EXIT(et);
1314 	in6m_release_list_deferred(&inmh);
1315 
1316 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
1317 }
1318 
1319 /*
1320  * Fast timeout handler (per-vnet).
1321  *
1322  * VIMAGE: Assume caller has set up our curvnet.
1323  */
1324 static void
1325 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1326 {
1327 	struct mbufq		 scq;	/* State-change packets */
1328 	struct mbufq		 qrq;	/* Query response packets */
1329 	struct ifnet		*ifp;
1330 	struct mld_ifsoftc	*mli;
1331 	struct ifmultiaddr	*ifma;
1332 	struct in6_multi	*inm;
1333 	int			 uri_fasthz;
1334 
1335 	uri_fasthz = 0;
1336 
1337 	/*
1338 	 * Quick check to see if any work needs to be done, in order to
1339 	 * minimize the overhead of fasttimo processing.
1340 	 * SMPng: XXX Unlocked reads.
1341 	 */
1342 	if (!V_current_state_timers_running6 &&
1343 	    !V_interface_timers_running6 &&
1344 	    !V_state_change_timers_running6)
1345 		return;
1346 
1347 	IN6_MULTI_LIST_LOCK();
1348 	MLD_LOCK();
1349 
1350 	/*
1351 	 * MLDv2 General Query response timer processing.
1352 	 */
1353 	if (V_interface_timers_running6) {
1354 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1355 
1356 		V_interface_timers_running6 = 0;
1357 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1358 			if (mli->mli_v2_timer == 0) {
1359 				/* Do nothing. */
1360 			} else if (--mli->mli_v2_timer == 0) {
1361 				mld_v2_dispatch_general_query(mli);
1362 			} else {
1363 				V_interface_timers_running6 = 1;
1364 			}
1365 		}
1366 	}
1367 
1368 	if (!V_current_state_timers_running6 &&
1369 	    !V_state_change_timers_running6)
1370 		goto out_locked;
1371 
1372 	V_current_state_timers_running6 = 0;
1373 	V_state_change_timers_running6 = 0;
1374 
1375 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1376 
1377 	/*
1378 	 * MLD host report and state-change timer processing.
1379 	 * Note: Processing a v2 group timer may remove a node.
1380 	 */
1381 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1382 		ifp = mli->mli_ifp;
1383 
1384 		if (mli->mli_version == MLD_VERSION_2) {
1385 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1386 			    MLD_FASTHZ);
1387 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1388 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1389 		}
1390 
1391 		IF_ADDR_WLOCK(ifp);
1392 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1393 			inm = in6m_ifmultiaddr_get_inm(ifma);
1394 			if (inm == NULL)
1395 				continue;
1396 			switch (mli->mli_version) {
1397 			case MLD_VERSION_1:
1398 				mld_v1_process_group_timer(inmh, inm);
1399 				break;
1400 			case MLD_VERSION_2:
1401 				mld_v2_process_group_timers(inmh, &qrq,
1402 				    &scq, inm, uri_fasthz);
1403 				break;
1404 			}
1405 		}
1406 		IF_ADDR_WUNLOCK(ifp);
1407 
1408 		switch (mli->mli_version) {
1409 		case MLD_VERSION_1:
1410 			/*
1411 			 * Transmit reports for this lifecycle.  This
1412 			 * is done while not holding IF_ADDR_LOCK
1413 			 * since this can call
1414 			 * in6ifa_ifpforlinklocal() which locks
1415 			 * IF_ADDR_LOCK internally as well as
1416 			 * ip6_output() to transmit a packet.
1417 			 */
1418 			while ((inm = SLIST_FIRST(inmh)) != NULL) {
1419 				SLIST_REMOVE_HEAD(inmh, in6m_defer);
1420 				(void)mld_v1_transmit_report(inm,
1421 				    MLD_LISTENER_REPORT);
1422 			}
1423 			break;
1424 		case MLD_VERSION_2:
1425 			mld_dispatch_queue(&qrq, 0);
1426 			mld_dispatch_queue(&scq, 0);
1427 			break;
1428 		}
1429 	}
1430 
1431 out_locked:
1432 	MLD_UNLOCK();
1433 	IN6_MULTI_LIST_UNLOCK();
1434 }
1435 
1436 /*
1437  * Update host report group timer.
1438  * Will update the global pending timer flags.
1439  */
1440 static void
1441 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1442 {
1443 	int report_timer_expired;
1444 
1445 	IN6_MULTI_LIST_LOCK_ASSERT();
1446 	MLD_LOCK_ASSERT();
1447 
1448 	if (inm->in6m_timer == 0) {
1449 		report_timer_expired = 0;
1450 	} else if (--inm->in6m_timer == 0) {
1451 		report_timer_expired = 1;
1452 	} else {
1453 		V_current_state_timers_running6 = 1;
1454 		return;
1455 	}
1456 
1457 	switch (inm->in6m_state) {
1458 	case MLD_NOT_MEMBER:
1459 	case MLD_SILENT_MEMBER:
1460 	case MLD_IDLE_MEMBER:
1461 	case MLD_LAZY_MEMBER:
1462 	case MLD_SLEEPING_MEMBER:
1463 	case MLD_AWAKENING_MEMBER:
1464 		break;
1465 	case MLD_REPORTING_MEMBER:
1466 		if (report_timer_expired) {
1467 			inm->in6m_state = MLD_IDLE_MEMBER;
1468 			SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1469 		}
1470 		break;
1471 	case MLD_G_QUERY_PENDING_MEMBER:
1472 	case MLD_SG_QUERY_PENDING_MEMBER:
1473 	case MLD_LEAVING_MEMBER:
1474 		break;
1475 	}
1476 }
1477 
1478 /*
1479  * Update a group's timers for MLDv2.
1480  * Will update the global pending timer flags.
1481  * Note: Unlocked read from mli.
1482  */
1483 static void
1484 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1485     struct mbufq *qrq, struct mbufq *scq,
1486     struct in6_multi *inm, const int uri_fasthz)
1487 {
1488 	int query_response_timer_expired;
1489 	int state_change_retransmit_timer_expired;
1490 #ifdef KTR
1491 	char ip6tbuf[INET6_ADDRSTRLEN];
1492 #endif
1493 
1494 	IN6_MULTI_LIST_LOCK_ASSERT();
1495 	MLD_LOCK_ASSERT();
1496 
1497 	query_response_timer_expired = 0;
1498 	state_change_retransmit_timer_expired = 0;
1499 
1500 	/*
1501 	 * During a transition from compatibility mode back to MLDv2,
1502 	 * a group record in REPORTING state may still have its group
1503 	 * timer active. This is a no-op in this function; it is easier
1504 	 * to deal with it here than to complicate the slow-timeout path.
1505 	 */
1506 	if (inm->in6m_timer == 0) {
1507 		query_response_timer_expired = 0;
1508 	} else if (--inm->in6m_timer == 0) {
1509 		query_response_timer_expired = 1;
1510 	} else {
1511 		V_current_state_timers_running6 = 1;
1512 	}
1513 
1514 	if (inm->in6m_sctimer == 0) {
1515 		state_change_retransmit_timer_expired = 0;
1516 	} else if (--inm->in6m_sctimer == 0) {
1517 		state_change_retransmit_timer_expired = 1;
1518 	} else {
1519 		V_state_change_timers_running6 = 1;
1520 	}
1521 
1522 	/* We are in fasttimo, so be quick about it. */
1523 	if (!state_change_retransmit_timer_expired &&
1524 	    !query_response_timer_expired)
1525 		return;
1526 
1527 	switch (inm->in6m_state) {
1528 	case MLD_NOT_MEMBER:
1529 	case MLD_SILENT_MEMBER:
1530 	case MLD_SLEEPING_MEMBER:
1531 	case MLD_LAZY_MEMBER:
1532 	case MLD_AWAKENING_MEMBER:
1533 	case MLD_IDLE_MEMBER:
1534 		break;
1535 	case MLD_G_QUERY_PENDING_MEMBER:
1536 	case MLD_SG_QUERY_PENDING_MEMBER:
1537 		/*
1538 		 * Respond to a previously pending Group-Specific
1539 		 * or Group-and-Source-Specific query by enqueueing
1540 		 * the appropriate Current-State report for
1541 		 * immediate transmission.
1542 		 */
1543 		if (query_response_timer_expired) {
1544 			int retval __unused;
1545 
1546 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1547 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1548 			    0);
1549 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1550 			    __func__, retval);
1551 			inm->in6m_state = MLD_REPORTING_MEMBER;
1552 			in6m_clear_recorded(inm);
1553 		}
1554 		/* FALLTHROUGH */
1555 	case MLD_REPORTING_MEMBER:
1556 	case MLD_LEAVING_MEMBER:
1557 		if (state_change_retransmit_timer_expired) {
1558 			/*
1559 			 * State-change retransmission timer fired.
1560 			 * If there are any further pending retransmissions,
1561 			 * set the global pending state-change flag, and
1562 			 * reset the timer.
1563 			 */
1564 			if (--inm->in6m_scrv > 0) {
1565 				inm->in6m_sctimer = uri_fasthz;
1566 				V_state_change_timers_running6 = 1;
1567 			}
1568 			/*
1569 			 * Retransmit the previously computed state-change
1570 			 * report. If there are no further pending
1571 			 * retransmissions, the mbuf queue will be consumed.
1572 			 * Update T0 state to T1 as we have now sent
1573 			 * a state-change.
1574 			 */
1575 			(void)mld_v2_merge_state_changes(inm, scq);
1576 
1577 			in6m_commit(inm);
1578 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1579 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1580 			    if_name(inm->in6m_ifp));
1581 
1582 			/*
1583 			 * If we are leaving the group for good, make sure
1584 			 * we release MLD's reference to it.
1585 			 * This release must be deferred using a SLIST,
1586 			 * as we are called from a loop which traverses
1587 			 * the in_ifmultiaddr TAILQ.
1588 			 */
1589 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1590 			    inm->in6m_scrv == 0) {
1591 				inm->in6m_state = MLD_NOT_MEMBER;
1592 				in6m_disconnect_locked(inmh, inm);
1593 				in6m_rele_locked(inmh, inm);
1594 			}
1595 		}
1596 		break;
1597 	}
1598 }
1599 
1600 /*
1601  * Switch to a different version on the given interface,
1602  * as per Section 9.12.
1603  */
1604 static void
1605 mld_set_version(struct mld_ifsoftc *mli, const int version)
1606 {
1607 	int old_version_timer;
1608 
1609 	MLD_LOCK_ASSERT();
1610 
1611 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1612 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1613 
1614 	if (version == MLD_VERSION_1) {
1615 		/*
1616 		 * Compute the "Older Version Querier Present" timer as per
1617 		 * Section 9.12.
1618 		 */
1619 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1620 		old_version_timer *= MLD_SLOWHZ;
1621 		mli->mli_v1_timer = old_version_timer;
1622 	}
1623 
1624 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1625 		mli->mli_version = MLD_VERSION_1;
1626 		mld_v2_cancel_link_timers(mli);
1627 	}
1628 }
1629 
1630 /*
1631  * Cancel pending MLDv2 timers for the given link and all groups
1632  * joined on it; state-change, general-query, and group-query timers.
1633  */
1634 static void
1635 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1636 {
1637 	struct epoch_tracker	 et;
1638 	struct in6_multi_head	 inmh;
1639 	struct ifmultiaddr	*ifma;
1640 	struct ifnet		*ifp;
1641 	struct in6_multi	*inm;
1642 
1643 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1644 	    mli->mli_ifp, if_name(mli->mli_ifp));
1645 
1646 	SLIST_INIT(&inmh);
1647 	IN6_MULTI_LIST_LOCK_ASSERT();
1648 	MLD_LOCK_ASSERT();
1649 
1650 	/*
1651 	 * Fast-track this potentially expensive operation
1652 	 * by checking all the global 'timer pending' flags.
1653 	 */
1654 	if (!V_interface_timers_running6 &&
1655 	    !V_state_change_timers_running6 &&
1656 	    !V_current_state_timers_running6)
1657 		return;
1658 
1659 	mli->mli_v2_timer = 0;
1660 
1661 	ifp = mli->mli_ifp;
1662 
1663 	IF_ADDR_WLOCK(ifp);
1664 	NET_EPOCH_ENTER(et);
1665 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1666 		inm = in6m_ifmultiaddr_get_inm(ifma);
1667 		if (inm == NULL)
1668 			continue;
1669 		switch (inm->in6m_state) {
1670 		case MLD_NOT_MEMBER:
1671 		case MLD_SILENT_MEMBER:
1672 		case MLD_IDLE_MEMBER:
1673 		case MLD_LAZY_MEMBER:
1674 		case MLD_SLEEPING_MEMBER:
1675 		case MLD_AWAKENING_MEMBER:
1676 			break;
1677 		case MLD_LEAVING_MEMBER:
1678 			/*
1679 			 * If we are leaving the group and switching
1680 			 * version, we need to release the final
1681 			 * reference held for issuing the INCLUDE {}.
1682 			 */
1683 			if (inm->in6m_refcount == 1)
1684 				in6m_disconnect_locked(&inmh, inm);
1685 			in6m_rele_locked(&inmh, inm);
1686 			/* FALLTHROUGH */
1687 		case MLD_G_QUERY_PENDING_MEMBER:
1688 		case MLD_SG_QUERY_PENDING_MEMBER:
1689 			in6m_clear_recorded(inm);
1690 			/* FALLTHROUGH */
1691 		case MLD_REPORTING_MEMBER:
1692 			inm->in6m_sctimer = 0;
1693 			inm->in6m_timer = 0;
1694 			inm->in6m_state = MLD_REPORTING_MEMBER;
1695 			/*
1696 			 * Free any pending MLDv2 state-change records.
1697 			 */
1698 			mbufq_drain(&inm->in6m_scq);
1699 			break;
1700 		}
1701 	}
1702 	NET_EPOCH_EXIT(et);
1703 	IF_ADDR_WUNLOCK(ifp);
1704 	in6m_release_list_deferred(&inmh);
1705 }
1706 
1707 /*
1708  * Global slowtimo handler.
1709  * VIMAGE: Timeout handlers are expected to service all vimages.
1710  */
1711 static struct callout mldslow_callout;
1712 static void
1713 mld_slowtimo(void *arg __unused)
1714 {
1715 	VNET_ITERATOR_DECL(vnet_iter);
1716 
1717 	VNET_LIST_RLOCK_NOSLEEP();
1718 	VNET_FOREACH(vnet_iter) {
1719 		CURVNET_SET(vnet_iter);
1720 		mld_slowtimo_vnet();
1721 		CURVNET_RESTORE();
1722 	}
1723 	VNET_LIST_RUNLOCK_NOSLEEP();
1724 
1725 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
1726 }
1727 
1728 /*
1729  * Per-vnet slowtimo handler.
1730  */
1731 static void
1732 mld_slowtimo_vnet(void)
1733 {
1734 	struct mld_ifsoftc *mli;
1735 
1736 	MLD_LOCK();
1737 
1738 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1739 		mld_v1_process_querier_timers(mli);
1740 	}
1741 
1742 	MLD_UNLOCK();
1743 }
1744 
1745 /*
1746  * Update the Older Version Querier Present timers for a link.
1747  * See Section 9.12 of RFC 3810.
1748  */
1749 static void
1750 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1751 {
1752 
1753 	MLD_LOCK_ASSERT();
1754 
1755 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1756 		/*
1757 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1758 		 */
1759 		CTR5(KTR_MLD,
1760 		    "%s: transition from v%d -> v%d on %p(%s)",
1761 		    __func__, mli->mli_version, MLD_VERSION_2,
1762 		    mli->mli_ifp, if_name(mli->mli_ifp));
1763 		mli->mli_version = MLD_VERSION_2;
1764 	}
1765 }
1766 
1767 /*
1768  * Transmit an MLDv1 report immediately.
1769  */
1770 static int
1771 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1772 {
1773 	struct ifnet		*ifp;
1774 	struct in6_ifaddr	*ia;
1775 	struct ip6_hdr		*ip6;
1776 	struct mbuf		*mh, *md;
1777 	struct mld_hdr		*mld;
1778 
1779 	NET_EPOCH_ASSERT();
1780 	IN6_MULTI_LIST_LOCK_ASSERT();
1781 	MLD_LOCK_ASSERT();
1782 
1783 	ifp = in6m->in6m_ifp;
1784 	/* in process of being freed */
1785 	if (ifp == NULL)
1786 		return (0);
1787 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1788 	/* ia may be NULL if link-local address is tentative. */
1789 
1790 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1791 	if (mh == NULL) {
1792 		if (ia != NULL)
1793 			ifa_free(&ia->ia_ifa);
1794 		return (ENOMEM);
1795 	}
1796 	md = m_get(M_NOWAIT, MT_DATA);
1797 	if (md == NULL) {
1798 		m_free(mh);
1799 		if (ia != NULL)
1800 			ifa_free(&ia->ia_ifa);
1801 		return (ENOMEM);
1802 	}
1803 	mh->m_next = md;
1804 
1805 	/*
1806 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1807 	 * that ether_output() does not need to allocate another mbuf
1808 	 * for the header in the most common case.
1809 	 */
1810 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1811 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1812 	mh->m_len = sizeof(struct ip6_hdr);
1813 
1814 	ip6 = mtod(mh, struct ip6_hdr *);
1815 	ip6->ip6_flow = 0;
1816 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1817 	ip6->ip6_vfc |= IPV6_VERSION;
1818 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1819 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1820 	ip6->ip6_dst = in6m->in6m_addr;
1821 
1822 	md->m_len = sizeof(struct mld_hdr);
1823 	mld = mtod(md, struct mld_hdr *);
1824 	mld->mld_type = type;
1825 	mld->mld_code = 0;
1826 	mld->mld_cksum = 0;
1827 	mld->mld_maxdelay = 0;
1828 	mld->mld_reserved = 0;
1829 	mld->mld_addr = in6m->in6m_addr;
1830 	in6_clearscope(&mld->mld_addr);
1831 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1832 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1833 
1834 	mld_save_context(mh, ifp);
1835 	mh->m_flags |= M_MLDV1;
1836 
1837 	mld_dispatch_packet(mh);
1838 
1839 	if (ia != NULL)
1840 		ifa_free(&ia->ia_ifa);
1841 	return (0);
1842 }
1843 
1844 /*
1845  * Process a state change from the upper layer for the given IPv6 group.
1846  *
1847  * Each socket holds a reference on the in_multi in its own ip_moptions.
1848  * The socket layer will have made the necessary updates to.the group
1849  * state, it is now up to MLD to issue a state change report if there
1850  * has been any change between T0 (when the last state-change was issued)
1851  * and T1 (now).
1852  *
1853  * We use the MLDv2 state machine at group level. The MLd module
1854  * however makes the decision as to which MLD protocol version to speak.
1855  * A state change *from* INCLUDE {} always means an initial join.
1856  * A state change *to* INCLUDE {} always means a final leave.
1857  *
1858  * If delay is non-zero, and the state change is an initial multicast
1859  * join, the state change report will be delayed by 'delay' ticks
1860  * in units of MLD_FASTHZ if MLDv1 is active on the link; otherwise
1861  * the initial MLDv2 state change report will be delayed by whichever
1862  * is sooner, a pending state-change timer or delay itself.
1863  *
1864  * VIMAGE: curvnet should have been set by caller, as this routine
1865  * is called from the socket option handlers.
1866  */
1867 int
1868 mld_change_state(struct in6_multi *inm, const int delay)
1869 {
1870 	struct mld_ifsoftc *mli;
1871 	struct ifnet *ifp;
1872 	int error;
1873 
1874 	IN6_MULTI_LIST_LOCK_ASSERT();
1875 
1876 	error = 0;
1877 
1878 	/*
1879 	 * Check if the in6_multi has already been disconnected.
1880 	 */
1881 	if (inm->in6m_ifp == NULL) {
1882 		CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1883 		return (0);
1884 	}
1885 
1886 	/*
1887 	 * Try to detect if the upper layer just asked us to change state
1888 	 * for an interface which has now gone away.
1889 	 */
1890 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1891 	ifp = inm->in6m_ifma->ifma_ifp;
1892 	if (ifp == NULL)
1893 		return (0);
1894 	/*
1895 	 * Sanity check that netinet6's notion of ifp is the
1896 	 * same as net's.
1897 	 */
1898 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1899 
1900 	MLD_LOCK();
1901 	mli = MLD_IFINFO(ifp);
1902 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1903 
1904 	/*
1905 	 * If we detect a state transition to or from MCAST_UNDEFINED
1906 	 * for this group, then we are starting or finishing an MLD
1907 	 * life cycle for this group.
1908 	 */
1909 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1910 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1911 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1912 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1913 			CTR1(KTR_MLD, "%s: initial join", __func__);
1914 			error = mld_initial_join(inm, mli, delay);
1915 			goto out_locked;
1916 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1917 			CTR1(KTR_MLD, "%s: final leave", __func__);
1918 			mld_final_leave(inm, mli);
1919 			goto out_locked;
1920 		}
1921 	} else {
1922 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1923 	}
1924 
1925 	error = mld_handle_state_change(inm, mli);
1926 
1927 out_locked:
1928 	MLD_UNLOCK();
1929 	return (error);
1930 }
1931 
1932 /*
1933  * Perform the initial join for an MLD group.
1934  *
1935  * When joining a group:
1936  *  If the group should have its MLD traffic suppressed, do nothing.
1937  *  MLDv1 starts sending MLDv1 host membership reports.
1938  *  MLDv2 will schedule an MLDv2 state-change report containing the
1939  *  initial state of the membership.
1940  *
1941  * If the delay argument is non-zero, then we must delay sending the
1942  * initial state change for delay ticks (in units of MLD_FASTHZ).
1943  */
1944 static int
1945 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1946     const int delay)
1947 {
1948 	struct epoch_tracker     et;
1949 	struct ifnet		*ifp;
1950 	struct mbufq		*mq;
1951 	int			 error, retval, syncstates;
1952 	int			 odelay;
1953 #ifdef KTR
1954 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1955 #endif
1956 
1957 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1958 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1959 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1960 
1961 	error = 0;
1962 	syncstates = 1;
1963 
1964 	ifp = inm->in6m_ifp;
1965 
1966 	IN6_MULTI_LIST_LOCK_ASSERT();
1967 	MLD_LOCK_ASSERT();
1968 
1969 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1970 
1971 	/*
1972 	 * Groups joined on loopback or marked as 'not reported',
1973 	 * enter the MLD_SILENT_MEMBER state and
1974 	 * are never reported in any protocol exchanges.
1975 	 * All other groups enter the appropriate state machine
1976 	 * for the version in use on this link.
1977 	 * A link marked as MLIF_SILENT causes MLD to be completely
1978 	 * disabled for the link.
1979 	 */
1980 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1981 	    (mli->mli_flags & MLIF_SILENT) ||
1982 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1983 		CTR1(KTR_MLD,
1984 "%s: not kicking state machine for silent group", __func__);
1985 		inm->in6m_state = MLD_SILENT_MEMBER;
1986 		inm->in6m_timer = 0;
1987 	} else {
1988 		/*
1989 		 * Deal with overlapping in_multi lifecycle.
1990 		 * If this group was LEAVING, then make sure
1991 		 * we drop the reference we picked up to keep the
1992 		 * group around for the final INCLUDE {} enqueue.
1993 		 */
1994 		if (mli->mli_version == MLD_VERSION_2 &&
1995 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
1996 			inm->in6m_refcount--;
1997 			MPASS(inm->in6m_refcount > 0);
1998 		}
1999 		inm->in6m_state = MLD_REPORTING_MEMBER;
2000 
2001 		switch (mli->mli_version) {
2002 		case MLD_VERSION_1:
2003 			/*
2004 			 * If a delay was provided, only use it if
2005 			 * it is greater than the delay normally
2006 			 * used for an MLDv1 state change report,
2007 			 * and delay sending the initial MLDv1 report
2008 			 * by not transitioning to the IDLE state.
2009 			 */
2010 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * MLD_FASTHZ);
2011 			if (delay) {
2012 				inm->in6m_timer = max(delay, odelay);
2013 				V_current_state_timers_running6 = 1;
2014 			} else {
2015 				inm->in6m_state = MLD_IDLE_MEMBER;
2016 				NET_EPOCH_ENTER(et);
2017 				error = mld_v1_transmit_report(inm,
2018 				     MLD_LISTENER_REPORT);
2019 				NET_EPOCH_EXIT(et);
2020 				if (error == 0) {
2021 					inm->in6m_timer = odelay;
2022 					V_current_state_timers_running6 = 1;
2023 				}
2024 			}
2025 			break;
2026 
2027 		case MLD_VERSION_2:
2028 			/*
2029 			 * Defer update of T0 to T1, until the first copy
2030 			 * of the state change has been transmitted.
2031 			 */
2032 			syncstates = 0;
2033 
2034 			/*
2035 			 * Immediately enqueue a State-Change Report for
2036 			 * this interface, freeing any previous reports.
2037 			 * Don't kick the timers if there is nothing to do,
2038 			 * or if an error occurred.
2039 			 */
2040 			mq = &inm->in6m_scq;
2041 			mbufq_drain(mq);
2042 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2043 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2044 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2045 			    __func__, retval);
2046 			if (retval <= 0) {
2047 				error = retval * -1;
2048 				break;
2049 			}
2050 
2051 			/*
2052 			 * Schedule transmission of pending state-change
2053 			 * report up to RV times for this link. The timer
2054 			 * will fire at the next mld_fasttimo (~200ms),
2055 			 * giving us an opportunity to merge the reports.
2056 			 *
2057 			 * If a delay was provided to this function, only
2058 			 * use this delay if sooner than the existing one.
2059 			 */
2060 			KASSERT(mli->mli_rv > 1,
2061 			   ("%s: invalid robustness %d", __func__,
2062 			    mli->mli_rv));
2063 			inm->in6m_scrv = mli->mli_rv;
2064 			if (delay) {
2065 				if (inm->in6m_sctimer > 1) {
2066 					inm->in6m_sctimer =
2067 					    min(inm->in6m_sctimer, delay);
2068 				} else
2069 					inm->in6m_sctimer = delay;
2070 			} else
2071 				inm->in6m_sctimer = 1;
2072 			V_state_change_timers_running6 = 1;
2073 
2074 			error = 0;
2075 			break;
2076 		}
2077 	}
2078 
2079 	/*
2080 	 * Only update the T0 state if state change is atomic,
2081 	 * i.e. we don't need to wait for a timer to fire before we
2082 	 * can consider the state change to have been communicated.
2083 	 */
2084 	if (syncstates) {
2085 		in6m_commit(inm);
2086 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2087 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2088 		    if_name(inm->in6m_ifp));
2089 	}
2090 
2091 	return (error);
2092 }
2093 
2094 /*
2095  * Issue an intermediate state change during the life-cycle.
2096  */
2097 static int
2098 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2099 {
2100 	struct ifnet		*ifp;
2101 	int			 retval;
2102 #ifdef KTR
2103 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2104 #endif
2105 
2106 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2107 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2108 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2109 
2110 	ifp = inm->in6m_ifp;
2111 
2112 	IN6_MULTI_LIST_LOCK_ASSERT();
2113 	MLD_LOCK_ASSERT();
2114 
2115 	KASSERT(mli && mli->mli_ifp == ifp,
2116 	    ("%s: inconsistent ifp", __func__));
2117 
2118 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2119 	    (mli->mli_flags & MLIF_SILENT) ||
2120 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2121 	    (mli->mli_version != MLD_VERSION_2)) {
2122 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2123 			CTR1(KTR_MLD,
2124 "%s: not kicking state machine for silent group", __func__);
2125 		}
2126 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2127 		in6m_commit(inm);
2128 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2129 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2130 		    if_name(inm->in6m_ifp));
2131 		return (0);
2132 	}
2133 
2134 	mbufq_drain(&inm->in6m_scq);
2135 
2136 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2137 	    (mli->mli_flags & MLIF_USEALLOW));
2138 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2139 	if (retval <= 0)
2140 		return (-retval);
2141 
2142 	/*
2143 	 * If record(s) were enqueued, start the state-change
2144 	 * report timer for this group.
2145 	 */
2146 	inm->in6m_scrv = mli->mli_rv;
2147 	inm->in6m_sctimer = 1;
2148 	V_state_change_timers_running6 = 1;
2149 
2150 	return (0);
2151 }
2152 
2153 /*
2154  * Perform the final leave for a multicast address.
2155  *
2156  * When leaving a group:
2157  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2158  *  MLDv2 enqueues a state-change report containing a transition
2159  *  to INCLUDE {} for immediate transmission.
2160  */
2161 static void
2162 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2163 {
2164 	struct epoch_tracker     et;
2165 #ifdef KTR
2166 	char ip6tbuf[INET6_ADDRSTRLEN];
2167 #endif
2168 
2169 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2170 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2171 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2172 
2173 	IN6_MULTI_LIST_LOCK_ASSERT();
2174 	MLD_LOCK_ASSERT();
2175 
2176 	switch (inm->in6m_state) {
2177 	case MLD_NOT_MEMBER:
2178 	case MLD_SILENT_MEMBER:
2179 	case MLD_LEAVING_MEMBER:
2180 		/* Already leaving or left; do nothing. */
2181 		CTR1(KTR_MLD,
2182 "%s: not kicking state machine for silent group", __func__);
2183 		break;
2184 	case MLD_REPORTING_MEMBER:
2185 	case MLD_IDLE_MEMBER:
2186 	case MLD_G_QUERY_PENDING_MEMBER:
2187 	case MLD_SG_QUERY_PENDING_MEMBER:
2188 		if (mli->mli_version == MLD_VERSION_1) {
2189 #ifdef INVARIANTS
2190 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2191 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2192 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2193 			     __func__);
2194 #endif
2195 			NET_EPOCH_ENTER(et);
2196 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2197 			NET_EPOCH_EXIT(et);
2198 			inm->in6m_state = MLD_NOT_MEMBER;
2199 			V_current_state_timers_running6 = 1;
2200 		} else if (mli->mli_version == MLD_VERSION_2) {
2201 			/*
2202 			 * Stop group timer and all pending reports.
2203 			 * Immediately enqueue a state-change report
2204 			 * TO_IN {} to be sent on the next fast timeout,
2205 			 * giving us an opportunity to merge reports.
2206 			 */
2207 			mbufq_drain(&inm->in6m_scq);
2208 			inm->in6m_timer = 0;
2209 			inm->in6m_scrv = mli->mli_rv;
2210 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2211 			    "pending retransmissions.", __func__,
2212 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2213 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2214 			if (inm->in6m_scrv == 0) {
2215 				inm->in6m_state = MLD_NOT_MEMBER;
2216 				inm->in6m_sctimer = 0;
2217 			} else {
2218 				int retval __diagused;
2219 
2220 				in6m_acquire_locked(inm);
2221 
2222 				retval = mld_v2_enqueue_group_record(
2223 				    &inm->in6m_scq, inm, 1, 0, 0,
2224 				    (mli->mli_flags & MLIF_USEALLOW));
2225 				KASSERT(retval != 0,
2226 				    ("%s: enqueue record = %d", __func__,
2227 				     retval));
2228 
2229 				inm->in6m_state = MLD_LEAVING_MEMBER;
2230 				inm->in6m_sctimer = 1;
2231 				V_state_change_timers_running6 = 1;
2232 			}
2233 			break;
2234 		}
2235 		break;
2236 	case MLD_LAZY_MEMBER:
2237 	case MLD_SLEEPING_MEMBER:
2238 	case MLD_AWAKENING_MEMBER:
2239 		/* Our reports are suppressed; do nothing. */
2240 		break;
2241 	}
2242 
2243 	in6m_commit(inm);
2244 	CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2245 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2246 	    if_name(inm->in6m_ifp));
2247 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2248 	CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2249 	    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2250 }
2251 
2252 /*
2253  * Enqueue an MLDv2 group record to the given output queue.
2254  *
2255  * If is_state_change is zero, a current-state record is appended.
2256  * If is_state_change is non-zero, a state-change report is appended.
2257  *
2258  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2259  * If is_group_query is zero, and if there is a packet with free space
2260  * at the tail of the queue, it will be appended to providing there
2261  * is enough free space.
2262  * Otherwise a new mbuf packet chain is allocated.
2263  *
2264  * If is_source_query is non-zero, each source is checked to see if
2265  * it was recorded for a Group-Source query, and will be omitted if
2266  * it is not both in-mode and recorded.
2267  *
2268  * If use_block_allow is non-zero, state change reports for initial join
2269  * and final leave, on an inclusive mode group with a source list, will be
2270  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2271  *
2272  * The function will attempt to allocate leading space in the packet
2273  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2274  *
2275  * If successful the size of all data appended to the queue is returned,
2276  * otherwise an error code less than zero is returned, or zero if
2277  * no record(s) were appended.
2278  */
2279 static int
2280 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2281     const int is_state_change, const int is_group_query,
2282     const int is_source_query, const int use_block_allow)
2283 {
2284 	struct mldv2_record	 mr;
2285 	struct mldv2_record	*pmr;
2286 	struct ifnet		*ifp;
2287 	struct ip6_msource	*ims, *nims;
2288 	struct mbuf		*m0, *m, *md;
2289 	int			 is_filter_list_change;
2290 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2291 	int			 record_has_sources;
2292 	int			 now;
2293 	int			 type;
2294 	uint8_t			 mode;
2295 #ifdef KTR
2296 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2297 #endif
2298 
2299 	IN6_MULTI_LIST_LOCK_ASSERT();
2300 
2301 	ifp = inm->in6m_ifp;
2302 	is_filter_list_change = 0;
2303 	m = NULL;
2304 	m0 = NULL;
2305 	m0srcs = 0;
2306 	msrcs = 0;
2307 	nbytes = 0;
2308 	nims = NULL;
2309 	record_has_sources = 1;
2310 	pmr = NULL;
2311 	type = MLD_DO_NOTHING;
2312 	mode = inm->in6m_st[1].iss_fmode;
2313 
2314 	/*
2315 	 * If we did not transition out of ASM mode during t0->t1,
2316 	 * and there are no source nodes to process, we can skip
2317 	 * the generation of source records.
2318 	 */
2319 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2320 	    inm->in6m_nsrc == 0)
2321 		record_has_sources = 0;
2322 
2323 	if (is_state_change) {
2324 		/*
2325 		 * Queue a state change record.
2326 		 * If the mode did not change, and there are non-ASM
2327 		 * listeners or source filters present,
2328 		 * we potentially need to issue two records for the group.
2329 		 * If there are ASM listeners, and there was no filter
2330 		 * mode transition of any kind, do nothing.
2331 		 *
2332 		 * If we are transitioning to MCAST_UNDEFINED, we need
2333 		 * not send any sources. A transition to/from this state is
2334 		 * considered inclusive with some special treatment.
2335 		 *
2336 		 * If we are rewriting initial joins/leaves to use
2337 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2338 		 * we need to send sources in all cases.
2339 		 */
2340 		if (mode != inm->in6m_st[0].iss_fmode) {
2341 			if (mode == MCAST_EXCLUDE) {
2342 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2343 				    __func__);
2344 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2345 			} else {
2346 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2347 				    __func__);
2348 				if (use_block_allow) {
2349 					/*
2350 					 * XXX
2351 					 * Here we're interested in state
2352 					 * edges either direction between
2353 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2354 					 * Perhaps we should just check
2355 					 * the group state, rather than
2356 					 * the filter mode.
2357 					 */
2358 					if (mode == MCAST_UNDEFINED) {
2359 						type = MLD_BLOCK_OLD_SOURCES;
2360 					} else {
2361 						type = MLD_ALLOW_NEW_SOURCES;
2362 					}
2363 				} else {
2364 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2365 					if (mode == MCAST_UNDEFINED)
2366 						record_has_sources = 0;
2367 				}
2368 			}
2369 		} else {
2370 			if (record_has_sources) {
2371 				is_filter_list_change = 1;
2372 			} else {
2373 				type = MLD_DO_NOTHING;
2374 			}
2375 		}
2376 	} else {
2377 		/*
2378 		 * Queue a current state record.
2379 		 */
2380 		if (mode == MCAST_EXCLUDE) {
2381 			type = MLD_MODE_IS_EXCLUDE;
2382 		} else if (mode == MCAST_INCLUDE) {
2383 			type = MLD_MODE_IS_INCLUDE;
2384 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2385 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2386 			     __func__, inm, inm->in6m_st[1].iss_asm));
2387 		}
2388 	}
2389 
2390 	/*
2391 	 * Generate the filter list changes using a separate function.
2392 	 */
2393 	if (is_filter_list_change)
2394 		return (mld_v2_enqueue_filter_change(mq, inm));
2395 
2396 	if (type == MLD_DO_NOTHING) {
2397 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2398 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2399 		    if_name(inm->in6m_ifp));
2400 		return (0);
2401 	}
2402 
2403 	/*
2404 	 * If any sources are present, we must be able to fit at least
2405 	 * one in the trailing space of the tail packet's mbuf,
2406 	 * ideally more.
2407 	 */
2408 	minrec0len = sizeof(struct mldv2_record);
2409 	if (record_has_sources)
2410 		minrec0len += sizeof(struct in6_addr);
2411 
2412 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2413 	    mld_rec_type_to_str(type),
2414 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2415 	    if_name(inm->in6m_ifp));
2416 
2417 	/*
2418 	 * Check if we have a packet in the tail of the queue for this
2419 	 * group into which the first group record for this group will fit.
2420 	 * Otherwise allocate a new packet.
2421 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2422 	 * Note: Group records for G/GSR query responses MUST be sent
2423 	 * in their own packet.
2424 	 */
2425 	m0 = mbufq_last(mq);
2426 	if (!is_group_query &&
2427 	    m0 != NULL &&
2428 	    (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2429 	    (m0->m_pkthdr.len + minrec0len) <
2430 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2431 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2432 			    sizeof(struct mldv2_record)) /
2433 			    sizeof(struct in6_addr);
2434 		m = m0;
2435 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2436 	} else {
2437 		if (mbufq_full(mq)) {
2438 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2439 			return (-ENOMEM);
2440 		}
2441 		m = NULL;
2442 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2443 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2444 		if (!is_state_change && !is_group_query)
2445 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2446 		if (m == NULL)
2447 			m = m_gethdr(M_NOWAIT, MT_DATA);
2448 		if (m == NULL)
2449 			return (-ENOMEM);
2450 
2451 		mld_save_context(m, ifp);
2452 
2453 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2454 	}
2455 
2456 	/*
2457 	 * Append group record.
2458 	 * If we have sources, we don't know how many yet.
2459 	 */
2460 	mr.mr_type = type;
2461 	mr.mr_datalen = 0;
2462 	mr.mr_numsrc = 0;
2463 	mr.mr_addr = inm->in6m_addr;
2464 	in6_clearscope(&mr.mr_addr);
2465 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2466 		if (m != m0)
2467 			m_freem(m);
2468 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2469 		return (-ENOMEM);
2470 	}
2471 	nbytes += sizeof(struct mldv2_record);
2472 
2473 	/*
2474 	 * Append as many sources as will fit in the first packet.
2475 	 * If we are appending to a new packet, the chain allocation
2476 	 * may potentially use clusters; use m_getptr() in this case.
2477 	 * If we are appending to an existing packet, we need to obtain
2478 	 * a pointer to the group record after m_append(), in case a new
2479 	 * mbuf was allocated.
2480 	 *
2481 	 * Only append sources which are in-mode at t1. If we are
2482 	 * transitioning to MCAST_UNDEFINED state on the group, and
2483 	 * use_block_allow is zero, do not include source entries.
2484 	 * Otherwise, we need to include this source in the report.
2485 	 *
2486 	 * Only report recorded sources in our filter set when responding
2487 	 * to a group-source query.
2488 	 */
2489 	if (record_has_sources) {
2490 		if (m == m0) {
2491 			md = m_last(m);
2492 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2493 			    md->m_len - nbytes);
2494 		} else {
2495 			md = m_getptr(m, 0, &off);
2496 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2497 			    off);
2498 		}
2499 		msrcs = 0;
2500 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2501 		    nims) {
2502 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2503 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2504 			now = im6s_get_mode(inm, ims, 1);
2505 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2506 			if ((now != mode) ||
2507 			    (now == mode &&
2508 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2509 				CTR1(KTR_MLD, "%s: skip node", __func__);
2510 				continue;
2511 			}
2512 			if (is_source_query && ims->im6s_stp == 0) {
2513 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2514 				    __func__);
2515 				continue;
2516 			}
2517 			CTR1(KTR_MLD, "%s: append node", __func__);
2518 			if (!m_append(m, sizeof(struct in6_addr),
2519 			    (void *)&ims->im6s_addr)) {
2520 				if (m != m0)
2521 					m_freem(m);
2522 				CTR1(KTR_MLD, "%s: m_append() failed.",
2523 				    __func__);
2524 				return (-ENOMEM);
2525 			}
2526 			nbytes += sizeof(struct in6_addr);
2527 			++msrcs;
2528 			if (msrcs == m0srcs)
2529 				break;
2530 		}
2531 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2532 		    msrcs);
2533 		pmr->mr_numsrc = htons(msrcs);
2534 		nbytes += (msrcs * sizeof(struct in6_addr));
2535 	}
2536 
2537 	if (is_source_query && msrcs == 0) {
2538 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2539 		if (m != m0)
2540 			m_freem(m);
2541 		return (0);
2542 	}
2543 
2544 	/*
2545 	 * We are good to go with first packet.
2546 	 */
2547 	if (m != m0) {
2548 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2549 		m->m_pkthdr.vt_nrecs = 1;
2550 		mbufq_enqueue(mq, m);
2551 	} else
2552 		m->m_pkthdr.vt_nrecs++;
2553 
2554 	/*
2555 	 * No further work needed if no source list in packet(s).
2556 	 */
2557 	if (!record_has_sources)
2558 		return (nbytes);
2559 
2560 	/*
2561 	 * Whilst sources remain to be announced, we need to allocate
2562 	 * a new packet and fill out as many sources as will fit.
2563 	 * Always try for a cluster first.
2564 	 */
2565 	while (nims != NULL) {
2566 		if (mbufq_full(mq)) {
2567 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2568 			return (-ENOMEM);
2569 		}
2570 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2571 		if (m == NULL)
2572 			m = m_gethdr(M_NOWAIT, MT_DATA);
2573 		if (m == NULL)
2574 			return (-ENOMEM);
2575 		mld_save_context(m, ifp);
2576 		md = m_getptr(m, 0, &off);
2577 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2578 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2579 
2580 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2581 			if (m != m0)
2582 				m_freem(m);
2583 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2584 			return (-ENOMEM);
2585 		}
2586 		m->m_pkthdr.vt_nrecs = 1;
2587 		nbytes += sizeof(struct mldv2_record);
2588 
2589 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2590 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2591 
2592 		msrcs = 0;
2593 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2594 			CTR2(KTR_MLD, "%s: visit node %s",
2595 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2596 			now = im6s_get_mode(inm, ims, 1);
2597 			if ((now != mode) ||
2598 			    (now == mode &&
2599 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2600 				CTR1(KTR_MLD, "%s: skip node", __func__);
2601 				continue;
2602 			}
2603 			if (is_source_query && ims->im6s_stp == 0) {
2604 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2605 				    __func__);
2606 				continue;
2607 			}
2608 			CTR1(KTR_MLD, "%s: append node", __func__);
2609 			if (!m_append(m, sizeof(struct in6_addr),
2610 			    (void *)&ims->im6s_addr)) {
2611 				if (m != m0)
2612 					m_freem(m);
2613 				CTR1(KTR_MLD, "%s: m_append() failed.",
2614 				    __func__);
2615 				return (-ENOMEM);
2616 			}
2617 			++msrcs;
2618 			if (msrcs == m0srcs)
2619 				break;
2620 		}
2621 		pmr->mr_numsrc = htons(msrcs);
2622 		nbytes += (msrcs * sizeof(struct in6_addr));
2623 
2624 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2625 		mbufq_enqueue(mq, m);
2626 	}
2627 
2628 	return (nbytes);
2629 }
2630 
2631 /*
2632  * Type used to mark record pass completion.
2633  * We exploit the fact we can cast to this easily from the
2634  * current filter modes on each ip_msource node.
2635  */
2636 typedef enum {
2637 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2638 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2639 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2640 	REC_FULL = REC_ALLOW | REC_BLOCK
2641 } rectype_t;
2642 
2643 /*
2644  * Enqueue an MLDv2 filter list change to the given output queue.
2645  *
2646  * Source list filter state is held in an RB-tree. When the filter list
2647  * for a group is changed without changing its mode, we need to compute
2648  * the deltas between T0 and T1 for each source in the filter set,
2649  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2650  *
2651  * As we may potentially queue two record types, and the entire R-B tree
2652  * needs to be walked at once, we break this out into its own function
2653  * so we can generate a tightly packed queue of packets.
2654  *
2655  * XXX This could be written to only use one tree walk, although that makes
2656  * serializing into the mbuf chains a bit harder. For now we do two walks
2657  * which makes things easier on us, and it may or may not be harder on
2658  * the L2 cache.
2659  *
2660  * If successful the size of all data appended to the queue is returned,
2661  * otherwise an error code less than zero is returned, or zero if
2662  * no record(s) were appended.
2663  */
2664 static int
2665 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2666 {
2667 	static const int MINRECLEN =
2668 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2669 	struct ifnet		*ifp;
2670 	struct mldv2_record	 mr;
2671 	struct mldv2_record	*pmr;
2672 	struct ip6_msource	*ims, *nims;
2673 	struct mbuf		*m, *m0, *md;
2674 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2675 	uint8_t			 mode, now, then;
2676 	rectype_t		 crt, drt, nrt;
2677 #ifdef KTR
2678 	int			 nallow, nblock;
2679 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2680 #endif
2681 
2682 	IN6_MULTI_LIST_LOCK_ASSERT();
2683 
2684 	if (inm->in6m_nsrc == 0 ||
2685 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2686 		return (0);
2687 
2688 	ifp = inm->in6m_ifp;			/* interface */
2689 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2690 	crt = REC_NONE;	/* current group record type */
2691 	drt = REC_NONE;	/* mask of completed group record types */
2692 	nrt = REC_NONE;	/* record type for current node */
2693 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2694 	npbytes = 0;	/* # of bytes appended this packet */
2695 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2696 	rsrcs = 0;	/* # sources encoded in current record */
2697 	schanged = 0;	/* # nodes encoded in overall filter change */
2698 #ifdef KTR
2699 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2700 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2701 #endif
2702 	nims = NULL;	/* next tree node pointer */
2703 
2704 	/*
2705 	 * For each possible filter record mode.
2706 	 * The first kind of source we encounter tells us which
2707 	 * is the first kind of record we start appending.
2708 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2709 	 * as the inverse of the group's filter mode.
2710 	 */
2711 	while (drt != REC_FULL) {
2712 		do {
2713 			m0 = mbufq_last(mq);
2714 			if (m0 != NULL &&
2715 			    (m0->m_pkthdr.vt_nrecs + 1 <=
2716 			     MLD_V2_REPORT_MAXRECS) &&
2717 			    (m0->m_pkthdr.len + MINRECLEN) <
2718 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2719 				m = m0;
2720 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2721 					    sizeof(struct mldv2_record)) /
2722 					    sizeof(struct in6_addr);
2723 				CTR1(KTR_MLD,
2724 				    "%s: use previous packet", __func__);
2725 			} else {
2726 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2727 				if (m == NULL)
2728 					m = m_gethdr(M_NOWAIT, MT_DATA);
2729 				if (m == NULL) {
2730 					CTR1(KTR_MLD,
2731 					    "%s: m_get*() failed", __func__);
2732 					return (-ENOMEM);
2733 				}
2734 				m->m_pkthdr.vt_nrecs = 0;
2735 				mld_save_context(m, ifp);
2736 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2737 				    sizeof(struct mldv2_record)) /
2738 				    sizeof(struct in6_addr);
2739 				npbytes = 0;
2740 				CTR1(KTR_MLD,
2741 				    "%s: allocated new packet", __func__);
2742 			}
2743 			/*
2744 			 * Append the MLD group record header to the
2745 			 * current packet's data area.
2746 			 * Recalculate pointer to free space for next
2747 			 * group record, in case m_append() allocated
2748 			 * a new mbuf or cluster.
2749 			 */
2750 			memset(&mr, 0, sizeof(mr));
2751 			mr.mr_addr = inm->in6m_addr;
2752 			in6_clearscope(&mr.mr_addr);
2753 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2754 				if (m != m0)
2755 					m_freem(m);
2756 				CTR1(KTR_MLD,
2757 				    "%s: m_append() failed", __func__);
2758 				return (-ENOMEM);
2759 			}
2760 			npbytes += sizeof(struct mldv2_record);
2761 			if (m != m0) {
2762 				/* new packet; offset in chain */
2763 				md = m_getptr(m, npbytes -
2764 				    sizeof(struct mldv2_record), &off);
2765 				pmr = (struct mldv2_record *)(mtod(md,
2766 				    uint8_t *) + off);
2767 			} else {
2768 				/* current packet; offset from last append */
2769 				md = m_last(m);
2770 				pmr = (struct mldv2_record *)(mtod(md,
2771 				    uint8_t *) + md->m_len -
2772 				    sizeof(struct mldv2_record));
2773 			}
2774 			/*
2775 			 * Begin walking the tree for this record type
2776 			 * pass, or continue from where we left off
2777 			 * previously if we had to allocate a new packet.
2778 			 * Only report deltas in-mode at t1.
2779 			 * We need not report included sources as allowed
2780 			 * if we are in inclusive mode on the group,
2781 			 * however the converse is not true.
2782 			 */
2783 			rsrcs = 0;
2784 			if (nims == NULL) {
2785 				nims = RB_MIN(ip6_msource_tree,
2786 				    &inm->in6m_srcs);
2787 			}
2788 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2789 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2790 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2791 				now = im6s_get_mode(inm, ims, 1);
2792 				then = im6s_get_mode(inm, ims, 0);
2793 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2794 				    __func__, then, now);
2795 				if (now == then) {
2796 					CTR1(KTR_MLD,
2797 					    "%s: skip unchanged", __func__);
2798 					continue;
2799 				}
2800 				if (mode == MCAST_EXCLUDE &&
2801 				    now == MCAST_INCLUDE) {
2802 					CTR1(KTR_MLD,
2803 					    "%s: skip IN src on EX group",
2804 					    __func__);
2805 					continue;
2806 				}
2807 				nrt = (rectype_t)now;
2808 				if (nrt == REC_NONE)
2809 					nrt = (rectype_t)(~mode & REC_FULL);
2810 				if (schanged++ == 0) {
2811 					crt = nrt;
2812 				} else if (crt != nrt)
2813 					continue;
2814 				if (!m_append(m, sizeof(struct in6_addr),
2815 				    (void *)&ims->im6s_addr)) {
2816 					if (m != m0)
2817 						m_freem(m);
2818 					CTR1(KTR_MLD,
2819 					    "%s: m_append() failed", __func__);
2820 					return (-ENOMEM);
2821 				}
2822 #ifdef KTR
2823 				nallow += !!(crt == REC_ALLOW);
2824 				nblock += !!(crt == REC_BLOCK);
2825 #endif
2826 				if (++rsrcs == m0srcs)
2827 					break;
2828 			}
2829 			/*
2830 			 * If we did not append any tree nodes on this
2831 			 * pass, back out of allocations.
2832 			 */
2833 			if (rsrcs == 0) {
2834 				npbytes -= sizeof(struct mldv2_record);
2835 				if (m != m0) {
2836 					CTR1(KTR_MLD,
2837 					    "%s: m_free(m)", __func__);
2838 					m_freem(m);
2839 				} else {
2840 					CTR1(KTR_MLD,
2841 					    "%s: m_adj(m, -mr)", __func__);
2842 					m_adj(m, -((int)sizeof(
2843 					    struct mldv2_record)));
2844 				}
2845 				continue;
2846 			}
2847 			npbytes += (rsrcs * sizeof(struct in6_addr));
2848 			if (crt == REC_ALLOW)
2849 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2850 			else if (crt == REC_BLOCK)
2851 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2852 			pmr->mr_numsrc = htons(rsrcs);
2853 			/*
2854 			 * Count the new group record, and enqueue this
2855 			 * packet if it wasn't already queued.
2856 			 */
2857 			m->m_pkthdr.vt_nrecs++;
2858 			if (m != m0)
2859 				mbufq_enqueue(mq, m);
2860 			nbytes += npbytes;
2861 		} while (nims != NULL);
2862 		drt |= crt;
2863 		crt = (~crt & REC_FULL);
2864 	}
2865 
2866 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2867 	    nallow, nblock);
2868 
2869 	return (nbytes);
2870 }
2871 
2872 static int
2873 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2874 {
2875 	struct mbufq	*gq;
2876 	struct mbuf	*m;		/* pending state-change */
2877 	struct mbuf	*m0;		/* copy of pending state-change */
2878 	struct mbuf	*mt;		/* last state-change in packet */
2879 	int		 docopy, domerge;
2880 	u_int		 recslen;
2881 
2882 	docopy = 0;
2883 	domerge = 0;
2884 	recslen = 0;
2885 
2886 	IN6_MULTI_LIST_LOCK_ASSERT();
2887 	MLD_LOCK_ASSERT();
2888 
2889 	/*
2890 	 * If there are further pending retransmissions, make a writable
2891 	 * copy of each queued state-change message before merging.
2892 	 */
2893 	if (inm->in6m_scrv > 0)
2894 		docopy = 1;
2895 
2896 	gq = &inm->in6m_scq;
2897 #ifdef KTR
2898 	if (mbufq_first(gq) == NULL) {
2899 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2900 		    __func__, inm);
2901 	}
2902 #endif
2903 
2904 	m = mbufq_first(gq);
2905 	while (m != NULL) {
2906 		/*
2907 		 * Only merge the report into the current packet if
2908 		 * there is sufficient space to do so; an MLDv2 report
2909 		 * packet may only contain 65,535 group records.
2910 		 * Always use a simple mbuf chain concatentation to do this,
2911 		 * as large state changes for single groups may have
2912 		 * allocated clusters.
2913 		 */
2914 		domerge = 0;
2915 		mt = mbufq_last(scq);
2916 		if (mt != NULL) {
2917 			recslen = m_length(m, NULL);
2918 
2919 			if ((mt->m_pkthdr.vt_nrecs +
2920 			    m->m_pkthdr.vt_nrecs <=
2921 			    MLD_V2_REPORT_MAXRECS) &&
2922 			    (mt->m_pkthdr.len + recslen <=
2923 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2924 				domerge = 1;
2925 		}
2926 
2927 		if (!domerge && mbufq_full(gq)) {
2928 			CTR2(KTR_MLD,
2929 			    "%s: outbound queue full, skipping whole packet %p",
2930 			    __func__, m);
2931 			mt = m->m_nextpkt;
2932 			if (!docopy)
2933 				m_freem(m);
2934 			m = mt;
2935 			continue;
2936 		}
2937 
2938 		if (!docopy) {
2939 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2940 			m0 = mbufq_dequeue(gq);
2941 			m = m0->m_nextpkt;
2942 		} else {
2943 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2944 			m0 = m_dup(m, M_NOWAIT);
2945 			if (m0 == NULL)
2946 				return (ENOMEM);
2947 			m0->m_nextpkt = NULL;
2948 			m = m->m_nextpkt;
2949 		}
2950 
2951 		if (!domerge) {
2952 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2953 			    __func__, m0, scq);
2954 			mbufq_enqueue(scq, m0);
2955 		} else {
2956 			struct mbuf *mtl;	/* last mbuf of packet mt */
2957 
2958 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2959 			    __func__, m0, mt);
2960 
2961 			mtl = m_last(mt);
2962 			m0->m_flags &= ~M_PKTHDR;
2963 			mt->m_pkthdr.len += recslen;
2964 			mt->m_pkthdr.vt_nrecs +=
2965 			    m0->m_pkthdr.vt_nrecs;
2966 
2967 			mtl->m_next = m0;
2968 		}
2969 	}
2970 
2971 	return (0);
2972 }
2973 
2974 /*
2975  * Respond to a pending MLDv2 General Query.
2976  */
2977 static void
2978 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2979 {
2980 	struct ifmultiaddr	*ifma;
2981 	struct ifnet		*ifp;
2982 	struct in6_multi	*inm;
2983 	int			 retval __unused;
2984 
2985 	NET_EPOCH_ASSERT();
2986 	IN6_MULTI_LIST_LOCK_ASSERT();
2987 	MLD_LOCK_ASSERT();
2988 
2989 	KASSERT(mli->mli_version == MLD_VERSION_2,
2990 	    ("%s: called when version %d", __func__, mli->mli_version));
2991 
2992 	/*
2993 	 * Check that there are some packets queued. If so, send them first.
2994 	 * For large number of groups the reply to general query can take
2995 	 * many packets, we should finish sending them before starting of
2996 	 * queuing the new reply.
2997 	 */
2998 	if (mbufq_len(&mli->mli_gq) != 0)
2999 		goto send;
3000 
3001 	ifp = mli->mli_ifp;
3002 
3003 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3004 		inm = in6m_ifmultiaddr_get_inm(ifma);
3005 		if (inm == NULL)
3006 			continue;
3007 		KASSERT(ifp == inm->in6m_ifp,
3008 		    ("%s: inconsistent ifp", __func__));
3009 
3010 		switch (inm->in6m_state) {
3011 		case MLD_NOT_MEMBER:
3012 		case MLD_SILENT_MEMBER:
3013 			break;
3014 		case MLD_REPORTING_MEMBER:
3015 		case MLD_IDLE_MEMBER:
3016 		case MLD_LAZY_MEMBER:
3017 		case MLD_SLEEPING_MEMBER:
3018 		case MLD_AWAKENING_MEMBER:
3019 			inm->in6m_state = MLD_REPORTING_MEMBER;
3020 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3021 			    inm, 0, 0, 0, 0);
3022 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3023 			    __func__, retval);
3024 			break;
3025 		case MLD_G_QUERY_PENDING_MEMBER:
3026 		case MLD_SG_QUERY_PENDING_MEMBER:
3027 		case MLD_LEAVING_MEMBER:
3028 			break;
3029 		}
3030 	}
3031 
3032 send:
3033 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3034 
3035 	/*
3036 	 * Slew transmission of bursts over 500ms intervals.
3037 	 */
3038 	if (mbufq_first(&mli->mli_gq) != NULL) {
3039 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3040 		    MLD_RESPONSE_BURST_INTERVAL);
3041 		V_interface_timers_running6 = 1;
3042 	}
3043 }
3044 
3045 /*
3046  * Transmit the next pending message in the output queue.
3047  *
3048  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3049  * MRT: Nothing needs to be done, as MLD traffic is always local to
3050  * a link and uses a link-scope multicast address.
3051  */
3052 static void
3053 mld_dispatch_packet(struct mbuf *m)
3054 {
3055 	struct ip6_moptions	 im6o;
3056 	struct ifnet		*ifp;
3057 	struct ifnet		*oifp;
3058 	struct mbuf		*m0;
3059 	struct mbuf		*md;
3060 	struct ip6_hdr		*ip6;
3061 	struct mld_hdr		*mld;
3062 	int			 error;
3063 	int			 off;
3064 	int			 type;
3065 	uint32_t		 ifindex;
3066 
3067 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3068 	NET_EPOCH_ASSERT();
3069 
3070 	/*
3071 	 * Set VNET image pointer from enqueued mbuf chain
3072 	 * before doing anything else. Whilst we use interface
3073 	 * indexes to guard against interface detach, they are
3074 	 * unique to each VIMAGE and must be retrieved.
3075 	 */
3076 	ifindex = mld_restore_context(m);
3077 
3078 	/*
3079 	 * Check if the ifnet still exists. This limits the scope of
3080 	 * any race in the absence of a global ifp lock for low cost
3081 	 * (an array lookup).
3082 	 */
3083 	ifp = ifnet_byindex(ifindex);
3084 	if (ifp == NULL) {
3085 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3086 		    __func__, m, ifindex);
3087 		m_freem(m);
3088 		IP6STAT_INC(ip6s_noroute);
3089 		goto out;
3090 	}
3091 
3092 	im6o.im6o_multicast_hlim  = 1;
3093 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3094 	im6o.im6o_multicast_ifp = ifp;
3095 
3096 	if (m->m_flags & M_MLDV1) {
3097 		m0 = m;
3098 	} else {
3099 		m0 = mld_v2_encap_report(ifp, m);
3100 		if (m0 == NULL) {
3101 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3102 			IP6STAT_INC(ip6s_odropped);
3103 			goto out;
3104 		}
3105 	}
3106 
3107 	mld_scrub_context(m0);
3108 	m_clrprotoflags(m);
3109 	m0->m_pkthdr.rcvif = V_loif;
3110 
3111 	ip6 = mtod(m0, struct ip6_hdr *);
3112 #if 0
3113 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3114 #else
3115 	/*
3116 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3117 	 * occur if we called in6_setscope() at transmission.
3118 	 * See comments at top of file.
3119 	 */
3120 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3121 #endif
3122 
3123 	/*
3124 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3125 	 * so we can bump the stats.
3126 	 */
3127 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3128 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3129 	type = mld->mld_type;
3130 
3131 	oifp = NULL;
3132 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3133 	    &oifp, NULL);
3134 	if (error) {
3135 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3136 		goto out;
3137 	}
3138 	ICMP6STAT_INC(icp6s_outhist[type]);
3139 	if (oifp != NULL) {
3140 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3141 		switch (type) {
3142 		case MLD_LISTENER_REPORT:
3143 		case MLDV2_LISTENER_REPORT:
3144 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3145 			break;
3146 		case MLD_LISTENER_DONE:
3147 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3148 			break;
3149 		}
3150 	}
3151 out:
3152 	return;
3153 }
3154 
3155 /*
3156  * Encapsulate an MLDv2 report.
3157  *
3158  * KAME IPv6 requires that hop-by-hop options be passed separately,
3159  * and that the IPv6 header be prepended in a separate mbuf.
3160  *
3161  * Returns a pointer to the new mbuf chain head, or NULL if the
3162  * allocation failed.
3163  */
3164 static struct mbuf *
3165 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3166 {
3167 	struct mbuf		*mh;
3168 	struct mldv2_report	*mld;
3169 	struct ip6_hdr		*ip6;
3170 	struct in6_ifaddr	*ia;
3171 	int			 mldreclen;
3172 
3173 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3174 	KASSERT((m->m_flags & M_PKTHDR),
3175 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3176 
3177 	/*
3178 	 * RFC3590: OK to send as :: or tentative during DAD.
3179 	 */
3180 	NET_EPOCH_ASSERT();
3181 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3182 	if (ia == NULL)
3183 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3184 
3185 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3186 	if (mh == NULL) {
3187 		if (ia != NULL)
3188 			ifa_free(&ia->ia_ifa);
3189 		m_freem(m);
3190 		return (NULL);
3191 	}
3192 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3193 
3194 	mldreclen = m_length(m, NULL);
3195 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3196 
3197 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3198 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3199 	    sizeof(struct mldv2_report) + mldreclen;
3200 
3201 	ip6 = mtod(mh, struct ip6_hdr *);
3202 	ip6->ip6_flow = 0;
3203 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3204 	ip6->ip6_vfc |= IPV6_VERSION;
3205 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3206 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3207 	if (ia != NULL)
3208 		ifa_free(&ia->ia_ifa);
3209 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3210 	/* scope ID will be set in netisr */
3211 
3212 	mld = (struct mldv2_report *)(ip6 + 1);
3213 	mld->mld_type = MLDV2_LISTENER_REPORT;
3214 	mld->mld_code = 0;
3215 	mld->mld_cksum = 0;
3216 	mld->mld_v2_reserved = 0;
3217 	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3218 	m->m_pkthdr.vt_nrecs = 0;
3219 
3220 	mh->m_next = m;
3221 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3222 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3223 	return (mh);
3224 }
3225 
3226 #ifdef KTR
3227 static char *
3228 mld_rec_type_to_str(const int type)
3229 {
3230 
3231 	switch (type) {
3232 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3233 			return "TO_EX";
3234 			break;
3235 		case MLD_CHANGE_TO_INCLUDE_MODE:
3236 			return "TO_IN";
3237 			break;
3238 		case MLD_MODE_IS_EXCLUDE:
3239 			return "MODE_EX";
3240 			break;
3241 		case MLD_MODE_IS_INCLUDE:
3242 			return "MODE_IN";
3243 			break;
3244 		case MLD_ALLOW_NEW_SOURCES:
3245 			return "ALLOW_NEW";
3246 			break;
3247 		case MLD_BLOCK_OLD_SOURCES:
3248 			return "BLOCK_OLD";
3249 			break;
3250 		default:
3251 			break;
3252 	}
3253 	return "unknown";
3254 }
3255 #endif
3256 
3257 static void
3258 mld_init(void *unused __unused)
3259 {
3260 
3261 	CTR1(KTR_MLD, "%s: initializing", __func__);
3262 	MLD_LOCK_INIT();
3263 
3264 	ip6_initpktopts(&mld_po);
3265 	mld_po.ip6po_hlim = 1;
3266 	mld_po.ip6po_hbh = &mld_ra.hbh;
3267 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3268 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3269 
3270 	callout_init(&mldslow_callout, 1);
3271 	callout_reset(&mldslow_callout, hz / MLD_SLOWHZ, mld_slowtimo, NULL);
3272 	callout_init(&mldfast_callout, 1);
3273 	callout_reset(&mldfast_callout, hz / MLD_FASTHZ, mld_fasttimo, NULL);
3274 }
3275 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3276 
3277 static void
3278 mld_uninit(void *unused __unused)
3279 {
3280 
3281 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3282 	callout_drain(&mldslow_callout);
3283 	callout_drain(&mldfast_callout);
3284 	MLD_LOCK_DESTROY();
3285 }
3286 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3287 
3288 static void
3289 vnet_mld_init(const void *unused __unused)
3290 {
3291 
3292 	CTR1(KTR_MLD, "%s: initializing", __func__);
3293 
3294 	LIST_INIT(&V_mli_head);
3295 }
3296 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3297     NULL);
3298 
3299 static void
3300 vnet_mld_uninit(const void *unused __unused)
3301 {
3302 
3303 	/* This can happen if we shutdown the network stack. */
3304 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3305 }
3306 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3307     NULL);
3308 
3309 static int
3310 mld_modevent(module_t mod, int type, void *unused __unused)
3311 {
3312 
3313     switch (type) {
3314     case MOD_LOAD:
3315     case MOD_UNLOAD:
3316 	break;
3317     default:
3318 	return (EOPNOTSUPP);
3319     }
3320     return (0);
3321 }
3322 
3323 static moduledata_t mld_mod = {
3324     "mld",
3325     mld_modevent,
3326     0
3327 };
3328 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3329