xref: /freebsd/sys/netinet6/mld6.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
1 /*-
2  * Copyright (c) 2009 Bruce Simpson.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. The name of the author may not be used to endorse or promote
13  *    products derived from this software without specific prior written
14  *    permission.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
29  */
30 
31 /*-
32  * Copyright (c) 1988 Stephen Deering.
33  * Copyright (c) 1992, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * Stephen Deering of Stanford University.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
64  */
65 
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
68 
69 #include "opt_inet.h"
70 #include "opt_inet6.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/mbuf.h>
75 #include <sys/socket.h>
76 #include <sys/protosw.h>
77 #include <sys/sysctl.h>
78 #include <sys/kernel.h>
79 #include <sys/callout.h>
80 #include <sys/malloc.h>
81 #include <sys/module.h>
82 #include <sys/ktr.h>
83 
84 #include <net/if.h>
85 #include <net/if_var.h>
86 #include <net/route.h>
87 #include <net/vnet.h>
88 
89 #include <netinet/in.h>
90 #include <netinet/in_var.h>
91 #include <netinet6/in6_var.h>
92 #include <netinet/ip6.h>
93 #include <netinet6/ip6_var.h>
94 #include <netinet6/scope6_var.h>
95 #include <netinet/icmp6.h>
96 #include <netinet6/mld6.h>
97 #include <netinet6/mld6_var.h>
98 
99 #include <security/mac/mac_framework.h>
100 
101 #ifndef KTR_MLD
102 #define KTR_MLD KTR_INET6
103 #endif
104 
105 static struct mld_ifsoftc *
106 		mli_alloc_locked(struct ifnet *);
107 static void	mli_delete_locked(const struct ifnet *);
108 static void	mld_dispatch_packet(struct mbuf *);
109 static void	mld_dispatch_queue(struct mbufq *, int);
110 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
111 static void	mld_fasttimo_vnet(void);
112 static int	mld_handle_state_change(struct in6_multi *,
113 		    struct mld_ifsoftc *);
114 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
115 		    const int);
116 #ifdef KTR
117 static char *	mld_rec_type_to_str(const int);
118 #endif
119 static void	mld_set_version(struct mld_ifsoftc *, const int);
120 static void	mld_slowtimo_vnet(void);
121 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
122 		    /*const*/ struct mld_hdr *);
123 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static void	mld_v1_process_group_timer(struct mld_ifsoftc *,
126 		    struct in6_multi *);
127 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
128 static int	mld_v1_transmit_report(struct in6_multi *, const int);
129 static void	mld_v1_update_group(struct in6_multi *, const int);
130 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
131 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
132 static struct mbuf *
133 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
134 static int	mld_v2_enqueue_filter_change(struct mbufq *,
135 		    struct in6_multi *);
136 static int	mld_v2_enqueue_group_record(struct mbufq *,
137 		    struct in6_multi *, const int, const int, const int,
138 		    const int);
139 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
140 		    struct mbuf *, const int, const int);
141 static int	mld_v2_merge_state_changes(struct in6_multi *,
142 		    struct mbufq *);
143 static void	mld_v2_process_group_timers(struct mld_ifsoftc *,
144 		    struct mbufq *, struct mbufq *,
145 		    struct in6_multi *, const int);
146 static int	mld_v2_process_group_query(struct in6_multi *,
147 		    struct mld_ifsoftc *mli, int, struct mbuf *, const int);
148 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
149 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
150 
151 /*
152  * Normative references: RFC 2710, RFC 3590, RFC 3810.
153  *
154  * Locking:
155  *  * The MLD subsystem lock ends up being system-wide for the moment,
156  *    but could be per-VIMAGE later on.
157  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
158  *    Any may be taken independently; if any are held at the same
159  *    time, the above lock order must be followed.
160  *  * IN6_MULTI_LOCK covers in_multi.
161  *  * MLD_LOCK covers per-link state and any global variables in this file.
162  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
163  *    per-link state iterators.
164  *
165  *  XXX LOR PREVENTION
166  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
167  *  will not accept an ifp; it wants an embedded scope ID, unlike
168  *  ip_output(), which happily takes the ifp given to it. The embedded
169  *  scope ID is only used by MLD to select the outgoing interface.
170  *
171  *  During interface attach and detach, MLD will take MLD_LOCK *after*
172  *  the IF_AFDATA_LOCK.
173  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
174  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
175  *  dispatch could work around this, but we'd rather not do that, as it
176  *  can introduce other races.
177  *
178  *  As such, we exploit the fact that the scope ID is just the interface
179  *  index, and embed it in the IPv6 destination address accordingly.
180  *  This is potentially NOT VALID for MLDv1 reports, as they
181  *  are always sent to the multicast group itself; as MLDv2
182  *  reports are always sent to ff02::16, this is not an issue
183  *  when MLDv2 is in use.
184  *
185  *  This does not however eliminate the LOR when ip6_output() itself
186  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
187  *  trigger a LOR warning in WITNESS when the ifnet is detached.
188  *
189  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
190  *  how it's used across the network stack. Here we're simply exploiting
191  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
192  *
193  * VIMAGE:
194  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
195  *    to a vnet in ifp->if_vnet.
196  */
197 static struct mtx		 mld_mtx;
198 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
199 
200 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
201 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
202 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
203 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
204 
205 /*
206  * VIMAGE-wide globals.
207  */
208 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0};
209 static VNET_DEFINE(LIST_HEAD(, mld_ifsoftc), mli_head);
210 static VNET_DEFINE(int, interface_timers_running6);
211 static VNET_DEFINE(int, state_change_timers_running6);
212 static VNET_DEFINE(int, current_state_timers_running6);
213 
214 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
215 #define	V_mli_head			VNET(mli_head)
216 #define	V_interface_timers_running6	VNET(interface_timers_running6)
217 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
218 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
219 
220 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
221 
222 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
223     "IPv6 Multicast Listener Discovery");
224 
225 /*
226  * Virtualized sysctls.
227  */
228 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
229     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
230     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
231     "Rate limit for MLDv2 Group-and-Source queries in seconds");
232 
233 /*
234  * Non-virtualized sysctls.
235  */
236 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
237     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
238     "Per-interface MLDv2 state");
239 
240 static int	mld_v1enable = 1;
241 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
242     &mld_v1enable, 0, "Enable fallback to MLDv1");
243 
244 static int	mld_use_allow = 1;
245 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
246     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
247 
248 /*
249  * Packed Router Alert option structure declaration.
250  */
251 struct mld_raopt {
252 	struct ip6_hbh		hbh;
253 	struct ip6_opt		pad;
254 	struct ip6_opt_router	ra;
255 } __packed;
256 
257 /*
258  * Router Alert hop-by-hop option header.
259  */
260 static struct mld_raopt mld_ra = {
261 	.hbh = { 0, 0 },
262 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
263 	.ra = {
264 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
265 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
266 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
267 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
268 	}
269 };
270 static struct ip6_pktopts mld_po;
271 
272 static __inline void
273 mld_save_context(struct mbuf *m, struct ifnet *ifp)
274 {
275 
276 #ifdef VIMAGE
277 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
278 #endif /* VIMAGE */
279 	m->m_pkthdr.flowid = ifp->if_index;
280 }
281 
282 static __inline void
283 mld_scrub_context(struct mbuf *m)
284 {
285 
286 	m->m_pkthdr.PH_loc.ptr = NULL;
287 	m->m_pkthdr.flowid = 0;
288 }
289 
290 /*
291  * Restore context from a queued output chain.
292  * Return saved ifindex.
293  *
294  * VIMAGE: The assertion is there to make sure that we
295  * actually called CURVNET_SET() with what's in the mbuf chain.
296  */
297 static __inline uint32_t
298 mld_restore_context(struct mbuf *m)
299 {
300 
301 #if defined(VIMAGE) && defined(INVARIANTS)
302 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
303 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
304 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
305 #endif
306 	return (m->m_pkthdr.flowid);
307 }
308 
309 /*
310  * Retrieve or set threshold between group-source queries in seconds.
311  *
312  * VIMAGE: Assume curvnet set by caller.
313  * SMPng: NOTE: Serialized by MLD lock.
314  */
315 static int
316 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
317 {
318 	int error;
319 	int i;
320 
321 	error = sysctl_wire_old_buffer(req, sizeof(int));
322 	if (error)
323 		return (error);
324 
325 	MLD_LOCK();
326 
327 	i = V_mld_gsrdelay.tv_sec;
328 
329 	error = sysctl_handle_int(oidp, &i, 0, req);
330 	if (error || !req->newptr)
331 		goto out_locked;
332 
333 	if (i < -1 || i >= 60) {
334 		error = EINVAL;
335 		goto out_locked;
336 	}
337 
338 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
339 	     V_mld_gsrdelay.tv_sec, i);
340 	V_mld_gsrdelay.tv_sec = i;
341 
342 out_locked:
343 	MLD_UNLOCK();
344 	return (error);
345 }
346 
347 /*
348  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
349  * For use by ifmcstat(8).
350  *
351  * SMPng: NOTE: Does an unlocked ifindex space read.
352  * VIMAGE: Assume curvnet set by caller. The node handler itself
353  * is not directly virtualized.
354  */
355 static int
356 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
357 {
358 	int			*name;
359 	int			 error;
360 	u_int			 namelen;
361 	struct ifnet		*ifp;
362 	struct mld_ifsoftc	*mli;
363 
364 	name = (int *)arg1;
365 	namelen = arg2;
366 
367 	if (req->newptr != NULL)
368 		return (EPERM);
369 
370 	if (namelen != 1)
371 		return (EINVAL);
372 
373 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
374 	if (error)
375 		return (error);
376 
377 	IN6_MULTI_LOCK();
378 	MLD_LOCK();
379 
380 	if (name[0] <= 0 || name[0] > V_if_index) {
381 		error = ENOENT;
382 		goto out_locked;
383 	}
384 
385 	error = ENOENT;
386 
387 	ifp = ifnet_byindex(name[0]);
388 	if (ifp == NULL)
389 		goto out_locked;
390 
391 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
392 		if (ifp == mli->mli_ifp) {
393 			struct mld_ifinfo info;
394 
395 			info.mli_version = mli->mli_version;
396 			info.mli_v1_timer = mli->mli_v1_timer;
397 			info.mli_v2_timer = mli->mli_v2_timer;
398 			info.mli_flags = mli->mli_flags;
399 			info.mli_rv = mli->mli_rv;
400 			info.mli_qi = mli->mli_qi;
401 			info.mli_qri = mli->mli_qri;
402 			info.mli_uri = mli->mli_uri;
403 			error = SYSCTL_OUT(req, &info, sizeof(info));
404 			break;
405 		}
406 	}
407 
408 out_locked:
409 	MLD_UNLOCK();
410 	IN6_MULTI_UNLOCK();
411 	return (error);
412 }
413 
414 /*
415  * Dispatch an entire queue of pending packet chains.
416  * VIMAGE: Assumes the vnet pointer has been set.
417  */
418 static void
419 mld_dispatch_queue(struct mbufq *mq, int limit)
420 {
421 	struct mbuf *m;
422 
423 	while ((m = mbufq_dequeue(mq)) != NULL) {
424 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
425 		mld_dispatch_packet(m);
426 		if (--limit == 0)
427 			break;
428 	}
429 }
430 
431 /*
432  * Filter outgoing MLD report state by group.
433  *
434  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
435  * and node-local addresses. However, kernel and socket consumers
436  * always embed the KAME scope ID in the address provided, so strip it
437  * when performing comparison.
438  * Note: This is not the same as the *multicast* scope.
439  *
440  * Return zero if the given group is one for which MLD reports
441  * should be suppressed, or non-zero if reports should be issued.
442  */
443 static __inline int
444 mld_is_addr_reported(const struct in6_addr *addr)
445 {
446 
447 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
448 
449 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
450 		return (0);
451 
452 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
453 		struct in6_addr tmp = *addr;
454 		in6_clearscope(&tmp);
455 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
456 			return (0);
457 	}
458 
459 	return (1);
460 }
461 
462 /*
463  * Attach MLD when PF_INET6 is attached to an interface.
464  *
465  * SMPng: Normally called with IF_AFDATA_LOCK held.
466  */
467 struct mld_ifsoftc *
468 mld_domifattach(struct ifnet *ifp)
469 {
470 	struct mld_ifsoftc *mli;
471 
472 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
473 	    __func__, ifp, if_name(ifp));
474 
475 	MLD_LOCK();
476 
477 	mli = mli_alloc_locked(ifp);
478 	if (!(ifp->if_flags & IFF_MULTICAST))
479 		mli->mli_flags |= MLIF_SILENT;
480 	if (mld_use_allow)
481 		mli->mli_flags |= MLIF_USEALLOW;
482 
483 	MLD_UNLOCK();
484 
485 	return (mli);
486 }
487 
488 /*
489  * VIMAGE: assume curvnet set by caller.
490  */
491 static struct mld_ifsoftc *
492 mli_alloc_locked(/*const*/ struct ifnet *ifp)
493 {
494 	struct mld_ifsoftc *mli;
495 
496 	MLD_LOCK_ASSERT();
497 
498 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
499 	if (mli == NULL)
500 		goto out;
501 
502 	mli->mli_ifp = ifp;
503 	mli->mli_version = MLD_VERSION_2;
504 	mli->mli_flags = 0;
505 	mli->mli_rv = MLD_RV_INIT;
506 	mli->mli_qi = MLD_QI_INIT;
507 	mli->mli_qri = MLD_QRI_INIT;
508 	mli->mli_uri = MLD_URI_INIT;
509 	SLIST_INIT(&mli->mli_relinmhead);
510 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
511 
512 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
513 
514 	CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
515 	     ifp, if_name(ifp));
516 
517 out:
518 	return (mli);
519 }
520 
521 /*
522  * Hook for ifdetach.
523  *
524  * NOTE: Some finalization tasks need to run before the protocol domain
525  * is detached, but also before the link layer does its cleanup.
526  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
527  *
528  * SMPng: Caller must hold IN6_MULTI_LOCK().
529  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
530  * XXX This routine is also bitten by unlocked ifma_protospec access.
531  */
532 void
533 mld_ifdetach(struct ifnet *ifp)
534 {
535 	struct mld_ifsoftc	*mli;
536 	struct ifmultiaddr	*ifma;
537 	struct in6_multi	*inm, *tinm;
538 
539 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
540 	    if_name(ifp));
541 
542 	IN6_MULTI_LOCK_ASSERT();
543 	MLD_LOCK();
544 
545 	mli = MLD_IFINFO(ifp);
546 	if (mli->mli_version == MLD_VERSION_2) {
547 		IF_ADDR_RLOCK(ifp);
548 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
549 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
550 			    ifma->ifma_protospec == NULL)
551 				continue;
552 			inm = (struct in6_multi *)ifma->ifma_protospec;
553 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
554 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
555 				    inm, in6m_nrele);
556 			}
557 			in6m_clear_recorded(inm);
558 		}
559 		IF_ADDR_RUNLOCK(ifp);
560 		SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
561 		    tinm) {
562 			SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
563 			in6m_release_locked(inm);
564 		}
565 	}
566 
567 	MLD_UNLOCK();
568 }
569 
570 /*
571  * Hook for domifdetach.
572  * Runs after link-layer cleanup; free MLD state.
573  *
574  * SMPng: Normally called with IF_AFDATA_LOCK held.
575  */
576 void
577 mld_domifdetach(struct ifnet *ifp)
578 {
579 
580 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
581 	    __func__, ifp, if_name(ifp));
582 
583 	MLD_LOCK();
584 	mli_delete_locked(ifp);
585 	MLD_UNLOCK();
586 }
587 
588 static void
589 mli_delete_locked(const struct ifnet *ifp)
590 {
591 	struct mld_ifsoftc *mli, *tmli;
592 
593 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
594 	    __func__, ifp, if_name(ifp));
595 
596 	MLD_LOCK_ASSERT();
597 
598 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
599 		if (mli->mli_ifp == ifp) {
600 			/*
601 			 * Free deferred General Query responses.
602 			 */
603 			mbufq_drain(&mli->mli_gq);
604 
605 			LIST_REMOVE(mli, mli_link);
606 
607 			KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
608 			    ("%s: there are dangling in_multi references",
609 			    __func__));
610 
611 			free(mli, M_MLD);
612 			return;
613 		}
614 	}
615 }
616 
617 /*
618  * Process a received MLDv1 general or address-specific query.
619  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
620  *
621  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
622  * mld_addr. This is OK as we own the mbuf chain.
623  */
624 static int
625 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
626     /*const*/ struct mld_hdr *mld)
627 {
628 	struct ifmultiaddr	*ifma;
629 	struct mld_ifsoftc	*mli;
630 	struct in6_multi	*inm;
631 	int			 is_general_query;
632 	uint16_t		 timer;
633 #ifdef KTR
634 	char			 ip6tbuf[INET6_ADDRSTRLEN];
635 #endif
636 
637 	is_general_query = 0;
638 
639 	if (!mld_v1enable) {
640 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
641 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
642 		    ifp, if_name(ifp));
643 		return (0);
644 	}
645 
646 	/*
647 	 * RFC3810 Section 6.2: MLD queries must originate from
648 	 * a router's link-local address.
649 	 */
650 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
651 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
652 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
653 		    ifp, if_name(ifp));
654 		return (0);
655 	}
656 
657 	/*
658 	 * Do address field validation upfront before we accept
659 	 * the query.
660 	 */
661 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
662 		/*
663 		 * MLDv1 General Query.
664 		 * If this was not sent to the all-nodes group, ignore it.
665 		 */
666 		struct in6_addr		 dst;
667 
668 		dst = ip6->ip6_dst;
669 		in6_clearscope(&dst);
670 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
671 			return (EINVAL);
672 		is_general_query = 1;
673 	} else {
674 		/*
675 		 * Embed scope ID of receiving interface in MLD query for
676 		 * lookup whilst we don't hold other locks.
677 		 */
678 		in6_setscope(&mld->mld_addr, ifp, NULL);
679 	}
680 
681 	IN6_MULTI_LOCK();
682 	MLD_LOCK();
683 
684 	/*
685 	 * Switch to MLDv1 host compatibility mode.
686 	 */
687 	mli = MLD_IFINFO(ifp);
688 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
689 	mld_set_version(mli, MLD_VERSION_1);
690 
691 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
692 	if (timer == 0)
693 		timer = 1;
694 
695 	IF_ADDR_RLOCK(ifp);
696 	if (is_general_query) {
697 		/*
698 		 * For each reporting group joined on this
699 		 * interface, kick the report timer.
700 		 */
701 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
702 		    ifp, if_name(ifp));
703 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
704 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
705 			    ifma->ifma_protospec == NULL)
706 				continue;
707 			inm = (struct in6_multi *)ifma->ifma_protospec;
708 			mld_v1_update_group(inm, timer);
709 		}
710 	} else {
711 		/*
712 		 * MLDv1 Group-Specific Query.
713 		 * If this is a group-specific MLDv1 query, we need only
714 		 * look up the single group to process it.
715 		 */
716 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
717 		if (inm != NULL) {
718 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
719 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
720 			    ifp, if_name(ifp));
721 			mld_v1_update_group(inm, timer);
722 		}
723 		/* XXX Clear embedded scope ID as userland won't expect it. */
724 		in6_clearscope(&mld->mld_addr);
725 	}
726 
727 	IF_ADDR_RUNLOCK(ifp);
728 	MLD_UNLOCK();
729 	IN6_MULTI_UNLOCK();
730 
731 	return (0);
732 }
733 
734 /*
735  * Update the report timer on a group in response to an MLDv1 query.
736  *
737  * If we are becoming the reporting member for this group, start the timer.
738  * If we already are the reporting member for this group, and timer is
739  * below the threshold, reset it.
740  *
741  * We may be updating the group for the first time since we switched
742  * to MLDv2. If we are, then we must clear any recorded source lists,
743  * and transition to REPORTING state; the group timer is overloaded
744  * for group and group-source query responses.
745  *
746  * Unlike MLDv2, the delay per group should be jittered
747  * to avoid bursts of MLDv1 reports.
748  */
749 static void
750 mld_v1_update_group(struct in6_multi *inm, const int timer)
751 {
752 #ifdef KTR
753 	char			 ip6tbuf[INET6_ADDRSTRLEN];
754 #endif
755 
756 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
757 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
758 	    if_name(inm->in6m_ifp), timer);
759 
760 	IN6_MULTI_LOCK_ASSERT();
761 
762 	switch (inm->in6m_state) {
763 	case MLD_NOT_MEMBER:
764 	case MLD_SILENT_MEMBER:
765 		break;
766 	case MLD_REPORTING_MEMBER:
767 		if (inm->in6m_timer != 0 &&
768 		    inm->in6m_timer <= timer) {
769 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
770 			    "skipping.", __func__);
771 			break;
772 		}
773 		/* FALLTHROUGH */
774 	case MLD_SG_QUERY_PENDING_MEMBER:
775 	case MLD_G_QUERY_PENDING_MEMBER:
776 	case MLD_IDLE_MEMBER:
777 	case MLD_LAZY_MEMBER:
778 	case MLD_AWAKENING_MEMBER:
779 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
780 		inm->in6m_state = MLD_REPORTING_MEMBER;
781 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
782 		V_current_state_timers_running6 = 1;
783 		break;
784 	case MLD_SLEEPING_MEMBER:
785 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
786 		inm->in6m_state = MLD_AWAKENING_MEMBER;
787 		break;
788 	case MLD_LEAVING_MEMBER:
789 		break;
790 	}
791 }
792 
793 /*
794  * Process a received MLDv2 general, group-specific or
795  * group-and-source-specific query.
796  *
797  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
798  *
799  * Return 0 if successful, otherwise an appropriate error code is returned.
800  */
801 static int
802 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
803     struct mbuf *m, const int off, const int icmp6len)
804 {
805 	struct mld_ifsoftc	*mli;
806 	struct mldv2_query	*mld;
807 	struct in6_multi	*inm;
808 	uint32_t		 maxdelay, nsrc, qqi;
809 	int			 is_general_query;
810 	uint16_t		 timer;
811 	uint8_t			 qrv;
812 #ifdef KTR
813 	char			 ip6tbuf[INET6_ADDRSTRLEN];
814 #endif
815 
816 	is_general_query = 0;
817 
818 	/*
819 	 * RFC3810 Section 6.2: MLD queries must originate from
820 	 * a router's link-local address.
821 	 */
822 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
823 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
824 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
825 		    ifp, if_name(ifp));
826 		return (0);
827 	}
828 
829 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
830 
831 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
832 
833 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
834 	if (maxdelay >= 32768) {
835 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
836 			   (MLD_MRC_EXP(maxdelay) + 3);
837 	}
838 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
839 	if (timer == 0)
840 		timer = 1;
841 
842 	qrv = MLD_QRV(mld->mld_misc);
843 	if (qrv < 2) {
844 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
845 		    qrv, MLD_RV_INIT);
846 		qrv = MLD_RV_INIT;
847 	}
848 
849 	qqi = mld->mld_qqi;
850 	if (qqi >= 128) {
851 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
852 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
853 	}
854 
855 	nsrc = ntohs(mld->mld_numsrc);
856 	if (nsrc > MLD_MAX_GS_SOURCES)
857 		return (EMSGSIZE);
858 	if (icmp6len < sizeof(struct mldv2_query) +
859 	    (nsrc * sizeof(struct in6_addr)))
860 		return (EMSGSIZE);
861 
862 	/*
863 	 * Do further input validation upfront to avoid resetting timers
864 	 * should we need to discard this query.
865 	 */
866 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
867 		/*
868 		 * A general query with a source list has undefined
869 		 * behaviour; discard it.
870 		 */
871 		if (nsrc > 0)
872 			return (EINVAL);
873 		is_general_query = 1;
874 	} else {
875 		/*
876 		 * Embed scope ID of receiving interface in MLD query for
877 		 * lookup whilst we don't hold other locks (due to KAME
878 		 * locking lameness). We own this mbuf chain just now.
879 		 */
880 		in6_setscope(&mld->mld_addr, ifp, NULL);
881 	}
882 
883 	IN6_MULTI_LOCK();
884 	MLD_LOCK();
885 
886 	mli = MLD_IFINFO(ifp);
887 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
888 
889 	/*
890 	 * Discard the v2 query if we're in Compatibility Mode.
891 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
892 	 * until the Old Version Querier Present timer expires.
893 	 */
894 	if (mli->mli_version != MLD_VERSION_2)
895 		goto out_locked;
896 
897 	mld_set_version(mli, MLD_VERSION_2);
898 	mli->mli_rv = qrv;
899 	mli->mli_qi = qqi;
900 	mli->mli_qri = maxdelay;
901 
902 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
903 	    maxdelay);
904 
905 	if (is_general_query) {
906 		/*
907 		 * MLDv2 General Query.
908 		 *
909 		 * Schedule a current-state report on this ifp for
910 		 * all groups, possibly containing source lists.
911 		 *
912 		 * If there is a pending General Query response
913 		 * scheduled earlier than the selected delay, do
914 		 * not schedule any other reports.
915 		 * Otherwise, reset the interface timer.
916 		 */
917 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
918 		    ifp, if_name(ifp));
919 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
920 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
921 			V_interface_timers_running6 = 1;
922 		}
923 	} else {
924 		/*
925 		 * MLDv2 Group-specific or Group-and-source-specific Query.
926 		 *
927 		 * Group-source-specific queries are throttled on
928 		 * a per-group basis to defeat denial-of-service attempts.
929 		 * Queries for groups we are not a member of on this
930 		 * link are simply ignored.
931 		 */
932 		IF_ADDR_RLOCK(ifp);
933 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
934 		if (inm == NULL) {
935 			IF_ADDR_RUNLOCK(ifp);
936 			goto out_locked;
937 		}
938 		if (nsrc > 0) {
939 			if (!ratecheck(&inm->in6m_lastgsrtv,
940 			    &V_mld_gsrdelay)) {
941 				CTR1(KTR_MLD, "%s: GS query throttled.",
942 				    __func__);
943 				IF_ADDR_RUNLOCK(ifp);
944 				goto out_locked;
945 			}
946 		}
947 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
948 		     ifp, if_name(ifp));
949 		/*
950 		 * If there is a pending General Query response
951 		 * scheduled sooner than the selected delay, no
952 		 * further report need be scheduled.
953 		 * Otherwise, prepare to respond to the
954 		 * group-specific or group-and-source query.
955 		 */
956 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
957 			mld_v2_process_group_query(inm, mli, timer, m, off);
958 
959 		/* XXX Clear embedded scope ID as userland won't expect it. */
960 		in6_clearscope(&mld->mld_addr);
961 		IF_ADDR_RUNLOCK(ifp);
962 	}
963 
964 out_locked:
965 	MLD_UNLOCK();
966 	IN6_MULTI_UNLOCK();
967 
968 	return (0);
969 }
970 
971 /*
972  * Process a received MLDv2 group-specific or group-and-source-specific
973  * query.
974  * Return <0 if any error occurred. Currently this is ignored.
975  */
976 static int
977 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
978     int timer, struct mbuf *m0, const int off)
979 {
980 	struct mldv2_query	*mld;
981 	int			 retval;
982 	uint16_t		 nsrc;
983 
984 	IN6_MULTI_LOCK_ASSERT();
985 	MLD_LOCK_ASSERT();
986 
987 	retval = 0;
988 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
989 
990 	switch (inm->in6m_state) {
991 	case MLD_NOT_MEMBER:
992 	case MLD_SILENT_MEMBER:
993 	case MLD_SLEEPING_MEMBER:
994 	case MLD_LAZY_MEMBER:
995 	case MLD_AWAKENING_MEMBER:
996 	case MLD_IDLE_MEMBER:
997 	case MLD_LEAVING_MEMBER:
998 		return (retval);
999 		break;
1000 	case MLD_REPORTING_MEMBER:
1001 	case MLD_G_QUERY_PENDING_MEMBER:
1002 	case MLD_SG_QUERY_PENDING_MEMBER:
1003 		break;
1004 	}
1005 
1006 	nsrc = ntohs(mld->mld_numsrc);
1007 
1008 	/*
1009 	 * Deal with group-specific queries upfront.
1010 	 * If any group query is already pending, purge any recorded
1011 	 * source-list state if it exists, and schedule a query response
1012 	 * for this group-specific query.
1013 	 */
1014 	if (nsrc == 0) {
1015 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1016 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1017 			in6m_clear_recorded(inm);
1018 			timer = min(inm->in6m_timer, timer);
1019 		}
1020 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1021 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1022 		V_current_state_timers_running6 = 1;
1023 		return (retval);
1024 	}
1025 
1026 	/*
1027 	 * Deal with the case where a group-and-source-specific query has
1028 	 * been received but a group-specific query is already pending.
1029 	 */
1030 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1031 		timer = min(inm->in6m_timer, timer);
1032 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1033 		V_current_state_timers_running6 = 1;
1034 		return (retval);
1035 	}
1036 
1037 	/*
1038 	 * Finally, deal with the case where a group-and-source-specific
1039 	 * query has been received, where a response to a previous g-s-r
1040 	 * query exists, or none exists.
1041 	 * In this case, we need to parse the source-list which the Querier
1042 	 * has provided us with and check if we have any source list filter
1043 	 * entries at T1 for these sources. If we do not, there is no need
1044 	 * schedule a report and the query may be dropped.
1045 	 * If we do, we must record them and schedule a current-state
1046 	 * report for those sources.
1047 	 */
1048 	if (inm->in6m_nsrc > 0) {
1049 		struct mbuf		*m;
1050 		uint8_t			*sp;
1051 		int			 i, nrecorded;
1052 		int			 soff;
1053 
1054 		m = m0;
1055 		soff = off + sizeof(struct mldv2_query);
1056 		nrecorded = 0;
1057 		for (i = 0; i < nsrc; i++) {
1058 			sp = mtod(m, uint8_t *) + soff;
1059 			retval = in6m_record_source(inm,
1060 			    (const struct in6_addr *)sp);
1061 			if (retval < 0)
1062 				break;
1063 			nrecorded += retval;
1064 			soff += sizeof(struct in6_addr);
1065 			if (soff >= m->m_len) {
1066 				soff = soff - m->m_len;
1067 				m = m->m_next;
1068 				if (m == NULL)
1069 					break;
1070 			}
1071 		}
1072 		if (nrecorded > 0) {
1073 			CTR1(KTR_MLD,
1074 			    "%s: schedule response to SG query", __func__);
1075 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1076 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1077 			V_current_state_timers_running6 = 1;
1078 		}
1079 	}
1080 
1081 	return (retval);
1082 }
1083 
1084 /*
1085  * Process a received MLDv1 host membership report.
1086  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1087  *
1088  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1089  * mld_addr. This is OK as we own the mbuf chain.
1090  */
1091 static int
1092 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1093     /*const*/ struct mld_hdr *mld)
1094 {
1095 	struct in6_addr		 src, dst;
1096 	struct in6_ifaddr	*ia;
1097 	struct in6_multi	*inm;
1098 #ifdef KTR
1099 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1100 #endif
1101 
1102 	if (!mld_v1enable) {
1103 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1104 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1105 		    ifp, if_name(ifp));
1106 		return (0);
1107 	}
1108 
1109 	if (ifp->if_flags & IFF_LOOPBACK)
1110 		return (0);
1111 
1112 	/*
1113 	 * MLDv1 reports must originate from a host's link-local address,
1114 	 * or the unspecified address (when booting).
1115 	 */
1116 	src = ip6->ip6_src;
1117 	in6_clearscope(&src);
1118 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1119 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1120 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1121 		    ifp, if_name(ifp));
1122 		return (EINVAL);
1123 	}
1124 
1125 	/*
1126 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1127 	 * group, and must be directed to the group itself.
1128 	 */
1129 	dst = ip6->ip6_dst;
1130 	in6_clearscope(&dst);
1131 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1132 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1133 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1134 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1135 		    ifp, if_name(ifp));
1136 		return (EINVAL);
1137 	}
1138 
1139 	/*
1140 	 * Make sure we don't hear our own membership report, as fast
1141 	 * leave requires knowing that we are the only member of a
1142 	 * group. Assume we used the link-local address if available,
1143 	 * otherwise look for ::.
1144 	 *
1145 	 * XXX Note that scope ID comparison is needed for the address
1146 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1147 	 * performed for the on-wire address.
1148 	 */
1149 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1150 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1151 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1152 		if (ia != NULL)
1153 			ifa_free(&ia->ia_ifa);
1154 		return (0);
1155 	}
1156 	if (ia != NULL)
1157 		ifa_free(&ia->ia_ifa);
1158 
1159 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1160 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1161 
1162 	/*
1163 	 * Embed scope ID of receiving interface in MLD query for lookup
1164 	 * whilst we don't hold other locks (due to KAME locking lameness).
1165 	 */
1166 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1167 		in6_setscope(&mld->mld_addr, ifp, NULL);
1168 
1169 	IN6_MULTI_LOCK();
1170 	MLD_LOCK();
1171 	IF_ADDR_RLOCK(ifp);
1172 
1173 	/*
1174 	 * MLDv1 report suppression.
1175 	 * If we are a member of this group, and our membership should be
1176 	 * reported, and our group timer is pending or about to be reset,
1177 	 * stop our group timer by transitioning to the 'lazy' state.
1178 	 */
1179 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1180 	if (inm != NULL) {
1181 		struct mld_ifsoftc *mli;
1182 
1183 		mli = inm->in6m_mli;
1184 		KASSERT(mli != NULL,
1185 		    ("%s: no mli for ifp %p", __func__, ifp));
1186 
1187 		/*
1188 		 * If we are in MLDv2 host mode, do not allow the
1189 		 * other host's MLDv1 report to suppress our reports.
1190 		 */
1191 		if (mli->mli_version == MLD_VERSION_2)
1192 			goto out_locked;
1193 
1194 		inm->in6m_timer = 0;
1195 
1196 		switch (inm->in6m_state) {
1197 		case MLD_NOT_MEMBER:
1198 		case MLD_SILENT_MEMBER:
1199 		case MLD_SLEEPING_MEMBER:
1200 			break;
1201 		case MLD_REPORTING_MEMBER:
1202 		case MLD_IDLE_MEMBER:
1203 		case MLD_AWAKENING_MEMBER:
1204 			CTR3(KTR_MLD,
1205 			    "report suppressed for %s on ifp %p(%s)",
1206 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1207 			    ifp, if_name(ifp));
1208 		case MLD_LAZY_MEMBER:
1209 			inm->in6m_state = MLD_LAZY_MEMBER;
1210 			break;
1211 		case MLD_G_QUERY_PENDING_MEMBER:
1212 		case MLD_SG_QUERY_PENDING_MEMBER:
1213 		case MLD_LEAVING_MEMBER:
1214 			break;
1215 		}
1216 	}
1217 
1218 out_locked:
1219 	IF_ADDR_RUNLOCK(ifp);
1220 	MLD_UNLOCK();
1221 	IN6_MULTI_UNLOCK();
1222 
1223 	/* XXX Clear embedded scope ID as userland won't expect it. */
1224 	in6_clearscope(&mld->mld_addr);
1225 
1226 	return (0);
1227 }
1228 
1229 /*
1230  * MLD input path.
1231  *
1232  * Assume query messages which fit in a single ICMPv6 message header
1233  * have been pulled up.
1234  * Assume that userland will want to see the message, even if it
1235  * otherwise fails kernel input validation; do not free it.
1236  * Pullup may however free the mbuf chain m if it fails.
1237  *
1238  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1239  */
1240 int
1241 mld_input(struct mbuf *m, int off, int icmp6len)
1242 {
1243 	struct ifnet	*ifp;
1244 	struct ip6_hdr	*ip6;
1245 	struct mld_hdr	*mld;
1246 	int		 mldlen;
1247 
1248 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1249 
1250 	ifp = m->m_pkthdr.rcvif;
1251 
1252 	ip6 = mtod(m, struct ip6_hdr *);
1253 
1254 	/* Pullup to appropriate size. */
1255 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1256 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1257 	    icmp6len >= sizeof(struct mldv2_query)) {
1258 		mldlen = sizeof(struct mldv2_query);
1259 	} else {
1260 		mldlen = sizeof(struct mld_hdr);
1261 	}
1262 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1263 	if (mld == NULL) {
1264 		ICMP6STAT_INC(icp6s_badlen);
1265 		return (IPPROTO_DONE);
1266 	}
1267 
1268 	/*
1269 	 * Userland needs to see all of this traffic for implementing
1270 	 * the endpoint discovery portion of multicast routing.
1271 	 */
1272 	switch (mld->mld_type) {
1273 	case MLD_LISTENER_QUERY:
1274 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1275 		if (icmp6len == sizeof(struct mld_hdr)) {
1276 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1277 				return (0);
1278 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1279 			if (mld_v2_input_query(ifp, ip6, m, off,
1280 			    icmp6len) != 0)
1281 				return (0);
1282 		}
1283 		break;
1284 	case MLD_LISTENER_REPORT:
1285 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1286 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1287 			return (0);
1288 		break;
1289 	case MLDV2_LISTENER_REPORT:
1290 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1291 		break;
1292 	case MLD_LISTENER_DONE:
1293 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1294 		break;
1295 	default:
1296 		break;
1297 	}
1298 
1299 	return (0);
1300 }
1301 
1302 /*
1303  * Fast timeout handler (global).
1304  * VIMAGE: Timeout handlers are expected to service all vimages.
1305  */
1306 void
1307 mld_fasttimo(void)
1308 {
1309 	VNET_ITERATOR_DECL(vnet_iter);
1310 
1311 	VNET_LIST_RLOCK_NOSLEEP();
1312 	VNET_FOREACH(vnet_iter) {
1313 		CURVNET_SET(vnet_iter);
1314 		mld_fasttimo_vnet();
1315 		CURVNET_RESTORE();
1316 	}
1317 	VNET_LIST_RUNLOCK_NOSLEEP();
1318 }
1319 
1320 /*
1321  * Fast timeout handler (per-vnet).
1322  *
1323  * VIMAGE: Assume caller has set up our curvnet.
1324  */
1325 static void
1326 mld_fasttimo_vnet(void)
1327 {
1328 	struct mbufq		 scq;	/* State-change packets */
1329 	struct mbufq		 qrq;	/* Query response packets */
1330 	struct ifnet		*ifp;
1331 	struct mld_ifsoftc	*mli;
1332 	struct ifmultiaddr	*ifma;
1333 	struct in6_multi	*inm, *tinm;
1334 	int			 uri_fasthz;
1335 
1336 	uri_fasthz = 0;
1337 
1338 	/*
1339 	 * Quick check to see if any work needs to be done, in order to
1340 	 * minimize the overhead of fasttimo processing.
1341 	 * SMPng: XXX Unlocked reads.
1342 	 */
1343 	if (!V_current_state_timers_running6 &&
1344 	    !V_interface_timers_running6 &&
1345 	    !V_state_change_timers_running6)
1346 		return;
1347 
1348 	IN6_MULTI_LOCK();
1349 	MLD_LOCK();
1350 
1351 	/*
1352 	 * MLDv2 General Query response timer processing.
1353 	 */
1354 	if (V_interface_timers_running6) {
1355 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1356 
1357 		V_interface_timers_running6 = 0;
1358 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1359 			if (mli->mli_v2_timer == 0) {
1360 				/* Do nothing. */
1361 			} else if (--mli->mli_v2_timer == 0) {
1362 				mld_v2_dispatch_general_query(mli);
1363 			} else {
1364 				V_interface_timers_running6 = 1;
1365 			}
1366 		}
1367 	}
1368 
1369 	if (!V_current_state_timers_running6 &&
1370 	    !V_state_change_timers_running6)
1371 		goto out_locked;
1372 
1373 	V_current_state_timers_running6 = 0;
1374 	V_state_change_timers_running6 = 0;
1375 
1376 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1377 
1378 	/*
1379 	 * MLD host report and state-change timer processing.
1380 	 * Note: Processing a v2 group timer may remove a node.
1381 	 */
1382 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1383 		ifp = mli->mli_ifp;
1384 
1385 		if (mli->mli_version == MLD_VERSION_2) {
1386 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1387 			    PR_FASTHZ);
1388 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1389 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1390 		}
1391 
1392 		IF_ADDR_RLOCK(ifp);
1393 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1394 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1395 			    ifma->ifma_protospec == NULL)
1396 				continue;
1397 			inm = (struct in6_multi *)ifma->ifma_protospec;
1398 			switch (mli->mli_version) {
1399 			case MLD_VERSION_1:
1400 				mld_v1_process_group_timer(mli, inm);
1401 				break;
1402 			case MLD_VERSION_2:
1403 				mld_v2_process_group_timers(mli, &qrq,
1404 				    &scq, inm, uri_fasthz);
1405 				break;
1406 			}
1407 		}
1408 		IF_ADDR_RUNLOCK(ifp);
1409 
1410 		switch (mli->mli_version) {
1411 		case MLD_VERSION_1:
1412 			/*
1413 			 * Transmit reports for this lifecycle.  This
1414 			 * is done while not holding IF_ADDR_LOCK
1415 			 * since this can call
1416 			 * in6ifa_ifpforlinklocal() which locks
1417 			 * IF_ADDR_LOCK internally as well as
1418 			 * ip6_output() to transmit a packet.
1419 			 */
1420 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1421 			    in6m_nrele, tinm) {
1422 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1423 				    in6m_nrele);
1424 				(void)mld_v1_transmit_report(inm,
1425 				    MLD_LISTENER_REPORT);
1426 			}
1427 			break;
1428 		case MLD_VERSION_2:
1429 			mld_dispatch_queue(&qrq, 0);
1430 			mld_dispatch_queue(&scq, 0);
1431 
1432 			/*
1433 			 * Free the in_multi reference(s) for
1434 			 * this lifecycle.
1435 			 */
1436 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1437 			    in6m_nrele, tinm) {
1438 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1439 				    in6m_nrele);
1440 				in6m_release_locked(inm);
1441 			}
1442 			break;
1443 		}
1444 	}
1445 
1446 out_locked:
1447 	MLD_UNLOCK();
1448 	IN6_MULTI_UNLOCK();
1449 }
1450 
1451 /*
1452  * Update host report group timer.
1453  * Will update the global pending timer flags.
1454  */
1455 static void
1456 mld_v1_process_group_timer(struct mld_ifsoftc *mli, struct in6_multi *inm)
1457 {
1458 	int report_timer_expired;
1459 
1460 	IN6_MULTI_LOCK_ASSERT();
1461 	MLD_LOCK_ASSERT();
1462 
1463 	if (inm->in6m_timer == 0) {
1464 		report_timer_expired = 0;
1465 	} else if (--inm->in6m_timer == 0) {
1466 		report_timer_expired = 1;
1467 	} else {
1468 		V_current_state_timers_running6 = 1;
1469 		return;
1470 	}
1471 
1472 	switch (inm->in6m_state) {
1473 	case MLD_NOT_MEMBER:
1474 	case MLD_SILENT_MEMBER:
1475 	case MLD_IDLE_MEMBER:
1476 	case MLD_LAZY_MEMBER:
1477 	case MLD_SLEEPING_MEMBER:
1478 	case MLD_AWAKENING_MEMBER:
1479 		break;
1480 	case MLD_REPORTING_MEMBER:
1481 		if (report_timer_expired) {
1482 			inm->in6m_state = MLD_IDLE_MEMBER;
1483 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
1484 			    in6m_nrele);
1485 		}
1486 		break;
1487 	case MLD_G_QUERY_PENDING_MEMBER:
1488 	case MLD_SG_QUERY_PENDING_MEMBER:
1489 	case MLD_LEAVING_MEMBER:
1490 		break;
1491 	}
1492 }
1493 
1494 /*
1495  * Update a group's timers for MLDv2.
1496  * Will update the global pending timer flags.
1497  * Note: Unlocked read from mli.
1498  */
1499 static void
1500 mld_v2_process_group_timers(struct mld_ifsoftc *mli,
1501     struct mbufq *qrq, struct mbufq *scq,
1502     struct in6_multi *inm, const int uri_fasthz)
1503 {
1504 	int query_response_timer_expired;
1505 	int state_change_retransmit_timer_expired;
1506 #ifdef KTR
1507 	char ip6tbuf[INET6_ADDRSTRLEN];
1508 #endif
1509 
1510 	IN6_MULTI_LOCK_ASSERT();
1511 	MLD_LOCK_ASSERT();
1512 
1513 	query_response_timer_expired = 0;
1514 	state_change_retransmit_timer_expired = 0;
1515 
1516 	/*
1517 	 * During a transition from compatibility mode back to MLDv2,
1518 	 * a group record in REPORTING state may still have its group
1519 	 * timer active. This is a no-op in this function; it is easier
1520 	 * to deal with it here than to complicate the slow-timeout path.
1521 	 */
1522 	if (inm->in6m_timer == 0) {
1523 		query_response_timer_expired = 0;
1524 	} else if (--inm->in6m_timer == 0) {
1525 		query_response_timer_expired = 1;
1526 	} else {
1527 		V_current_state_timers_running6 = 1;
1528 	}
1529 
1530 	if (inm->in6m_sctimer == 0) {
1531 		state_change_retransmit_timer_expired = 0;
1532 	} else if (--inm->in6m_sctimer == 0) {
1533 		state_change_retransmit_timer_expired = 1;
1534 	} else {
1535 		V_state_change_timers_running6 = 1;
1536 	}
1537 
1538 	/* We are in fasttimo, so be quick about it. */
1539 	if (!state_change_retransmit_timer_expired &&
1540 	    !query_response_timer_expired)
1541 		return;
1542 
1543 	switch (inm->in6m_state) {
1544 	case MLD_NOT_MEMBER:
1545 	case MLD_SILENT_MEMBER:
1546 	case MLD_SLEEPING_MEMBER:
1547 	case MLD_LAZY_MEMBER:
1548 	case MLD_AWAKENING_MEMBER:
1549 	case MLD_IDLE_MEMBER:
1550 		break;
1551 	case MLD_G_QUERY_PENDING_MEMBER:
1552 	case MLD_SG_QUERY_PENDING_MEMBER:
1553 		/*
1554 		 * Respond to a previously pending Group-Specific
1555 		 * or Group-and-Source-Specific query by enqueueing
1556 		 * the appropriate Current-State report for
1557 		 * immediate transmission.
1558 		 */
1559 		if (query_response_timer_expired) {
1560 			int retval;
1561 
1562 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1563 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1564 			    0);
1565 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1566 			    __func__, retval);
1567 			inm->in6m_state = MLD_REPORTING_MEMBER;
1568 			in6m_clear_recorded(inm);
1569 		}
1570 		/* FALLTHROUGH */
1571 	case MLD_REPORTING_MEMBER:
1572 	case MLD_LEAVING_MEMBER:
1573 		if (state_change_retransmit_timer_expired) {
1574 			/*
1575 			 * State-change retransmission timer fired.
1576 			 * If there are any further pending retransmissions,
1577 			 * set the global pending state-change flag, and
1578 			 * reset the timer.
1579 			 */
1580 			if (--inm->in6m_scrv > 0) {
1581 				inm->in6m_sctimer = uri_fasthz;
1582 				V_state_change_timers_running6 = 1;
1583 			}
1584 			/*
1585 			 * Retransmit the previously computed state-change
1586 			 * report. If there are no further pending
1587 			 * retransmissions, the mbuf queue will be consumed.
1588 			 * Update T0 state to T1 as we have now sent
1589 			 * a state-change.
1590 			 */
1591 			(void)mld_v2_merge_state_changes(inm, scq);
1592 
1593 			in6m_commit(inm);
1594 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1595 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1596 			    if_name(inm->in6m_ifp));
1597 
1598 			/*
1599 			 * If we are leaving the group for good, make sure
1600 			 * we release MLD's reference to it.
1601 			 * This release must be deferred using a SLIST,
1602 			 * as we are called from a loop which traverses
1603 			 * the in_ifmultiaddr TAILQ.
1604 			 */
1605 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1606 			    inm->in6m_scrv == 0) {
1607 				inm->in6m_state = MLD_NOT_MEMBER;
1608 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1609 				    inm, in6m_nrele);
1610 			}
1611 		}
1612 		break;
1613 	}
1614 }
1615 
1616 /*
1617  * Switch to a different version on the given interface,
1618  * as per Section 9.12.
1619  */
1620 static void
1621 mld_set_version(struct mld_ifsoftc *mli, const int version)
1622 {
1623 	int old_version_timer;
1624 
1625 	MLD_LOCK_ASSERT();
1626 
1627 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1628 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1629 
1630 	if (version == MLD_VERSION_1) {
1631 		/*
1632 		 * Compute the "Older Version Querier Present" timer as per
1633 		 * Section 9.12.
1634 		 */
1635 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1636 		old_version_timer *= PR_SLOWHZ;
1637 		mli->mli_v1_timer = old_version_timer;
1638 	}
1639 
1640 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1641 		mli->mli_version = MLD_VERSION_1;
1642 		mld_v2_cancel_link_timers(mli);
1643 	}
1644 }
1645 
1646 /*
1647  * Cancel pending MLDv2 timers for the given link and all groups
1648  * joined on it; state-change, general-query, and group-query timers.
1649  */
1650 static void
1651 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1652 {
1653 	struct ifmultiaddr	*ifma;
1654 	struct ifnet		*ifp;
1655 	struct in6_multi	*inm, *tinm;
1656 
1657 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1658 	    mli->mli_ifp, if_name(mli->mli_ifp));
1659 
1660 	IN6_MULTI_LOCK_ASSERT();
1661 	MLD_LOCK_ASSERT();
1662 
1663 	/*
1664 	 * Fast-track this potentially expensive operation
1665 	 * by checking all the global 'timer pending' flags.
1666 	 */
1667 	if (!V_interface_timers_running6 &&
1668 	    !V_state_change_timers_running6 &&
1669 	    !V_current_state_timers_running6)
1670 		return;
1671 
1672 	mli->mli_v2_timer = 0;
1673 
1674 	ifp = mli->mli_ifp;
1675 
1676 	IF_ADDR_RLOCK(ifp);
1677 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1678 		if (ifma->ifma_addr->sa_family != AF_INET6)
1679 			continue;
1680 		inm = (struct in6_multi *)ifma->ifma_protospec;
1681 		switch (inm->in6m_state) {
1682 		case MLD_NOT_MEMBER:
1683 		case MLD_SILENT_MEMBER:
1684 		case MLD_IDLE_MEMBER:
1685 		case MLD_LAZY_MEMBER:
1686 		case MLD_SLEEPING_MEMBER:
1687 		case MLD_AWAKENING_MEMBER:
1688 			break;
1689 		case MLD_LEAVING_MEMBER:
1690 			/*
1691 			 * If we are leaving the group and switching
1692 			 * version, we need to release the final
1693 			 * reference held for issuing the INCLUDE {}.
1694 			 */
1695 			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
1696 			    in6m_nrele);
1697 			/* FALLTHROUGH */
1698 		case MLD_G_QUERY_PENDING_MEMBER:
1699 		case MLD_SG_QUERY_PENDING_MEMBER:
1700 			in6m_clear_recorded(inm);
1701 			/* FALLTHROUGH */
1702 		case MLD_REPORTING_MEMBER:
1703 			inm->in6m_sctimer = 0;
1704 			inm->in6m_timer = 0;
1705 			inm->in6m_state = MLD_REPORTING_MEMBER;
1706 			/*
1707 			 * Free any pending MLDv2 state-change records.
1708 			 */
1709 			mbufq_drain(&inm->in6m_scq);
1710 			break;
1711 		}
1712 	}
1713 	IF_ADDR_RUNLOCK(ifp);
1714 	SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) {
1715 		SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
1716 		in6m_release_locked(inm);
1717 	}
1718 }
1719 
1720 /*
1721  * Global slowtimo handler.
1722  * VIMAGE: Timeout handlers are expected to service all vimages.
1723  */
1724 void
1725 mld_slowtimo(void)
1726 {
1727 	VNET_ITERATOR_DECL(vnet_iter);
1728 
1729 	VNET_LIST_RLOCK_NOSLEEP();
1730 	VNET_FOREACH(vnet_iter) {
1731 		CURVNET_SET(vnet_iter);
1732 		mld_slowtimo_vnet();
1733 		CURVNET_RESTORE();
1734 	}
1735 	VNET_LIST_RUNLOCK_NOSLEEP();
1736 }
1737 
1738 /*
1739  * Per-vnet slowtimo handler.
1740  */
1741 static void
1742 mld_slowtimo_vnet(void)
1743 {
1744 	struct mld_ifsoftc *mli;
1745 
1746 	MLD_LOCK();
1747 
1748 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1749 		mld_v1_process_querier_timers(mli);
1750 	}
1751 
1752 	MLD_UNLOCK();
1753 }
1754 
1755 /*
1756  * Update the Older Version Querier Present timers for a link.
1757  * See Section 9.12 of RFC 3810.
1758  */
1759 static void
1760 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1761 {
1762 
1763 	MLD_LOCK_ASSERT();
1764 
1765 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1766 		/*
1767 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1768 		 */
1769 		CTR5(KTR_MLD,
1770 		    "%s: transition from v%d -> v%d on %p(%s)",
1771 		    __func__, mli->mli_version, MLD_VERSION_2,
1772 		    mli->mli_ifp, if_name(mli->mli_ifp));
1773 		mli->mli_version = MLD_VERSION_2;
1774 	}
1775 }
1776 
1777 /*
1778  * Transmit an MLDv1 report immediately.
1779  */
1780 static int
1781 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1782 {
1783 	struct ifnet		*ifp;
1784 	struct in6_ifaddr	*ia;
1785 	struct ip6_hdr		*ip6;
1786 	struct mbuf		*mh, *md;
1787 	struct mld_hdr		*mld;
1788 
1789 	IN6_MULTI_LOCK_ASSERT();
1790 	MLD_LOCK_ASSERT();
1791 
1792 	ifp = in6m->in6m_ifp;
1793 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1794 	/* ia may be NULL if link-local address is tentative. */
1795 
1796 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1797 	if (mh == NULL) {
1798 		if (ia != NULL)
1799 			ifa_free(&ia->ia_ifa);
1800 		return (ENOMEM);
1801 	}
1802 	md = m_get(M_NOWAIT, MT_DATA);
1803 	if (md == NULL) {
1804 		m_free(mh);
1805 		if (ia != NULL)
1806 			ifa_free(&ia->ia_ifa);
1807 		return (ENOMEM);
1808 	}
1809 	mh->m_next = md;
1810 
1811 	/*
1812 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1813 	 * that ether_output() does not need to allocate another mbuf
1814 	 * for the header in the most common case.
1815 	 */
1816 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1817 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1818 	mh->m_len = sizeof(struct ip6_hdr);
1819 
1820 	ip6 = mtod(mh, struct ip6_hdr *);
1821 	ip6->ip6_flow = 0;
1822 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1823 	ip6->ip6_vfc |= IPV6_VERSION;
1824 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1825 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1826 	ip6->ip6_dst = in6m->in6m_addr;
1827 
1828 	md->m_len = sizeof(struct mld_hdr);
1829 	mld = mtod(md, struct mld_hdr *);
1830 	mld->mld_type = type;
1831 	mld->mld_code = 0;
1832 	mld->mld_cksum = 0;
1833 	mld->mld_maxdelay = 0;
1834 	mld->mld_reserved = 0;
1835 	mld->mld_addr = in6m->in6m_addr;
1836 	in6_clearscope(&mld->mld_addr);
1837 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1838 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1839 
1840 	mld_save_context(mh, ifp);
1841 	mh->m_flags |= M_MLDV1;
1842 
1843 	mld_dispatch_packet(mh);
1844 
1845 	if (ia != NULL)
1846 		ifa_free(&ia->ia_ifa);
1847 	return (0);
1848 }
1849 
1850 /*
1851  * Process a state change from the upper layer for the given IPv6 group.
1852  *
1853  * Each socket holds a reference on the in_multi in its own ip_moptions.
1854  * The socket layer will have made the necessary updates to.the group
1855  * state, it is now up to MLD to issue a state change report if there
1856  * has been any change between T0 (when the last state-change was issued)
1857  * and T1 (now).
1858  *
1859  * We use the MLDv2 state machine at group level. The MLd module
1860  * however makes the decision as to which MLD protocol version to speak.
1861  * A state change *from* INCLUDE {} always means an initial join.
1862  * A state change *to* INCLUDE {} always means a final leave.
1863  *
1864  * If delay is non-zero, and the state change is an initial multicast
1865  * join, the state change report will be delayed by 'delay' ticks
1866  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1867  * the initial MLDv2 state change report will be delayed by whichever
1868  * is sooner, a pending state-change timer or delay itself.
1869  *
1870  * VIMAGE: curvnet should have been set by caller, as this routine
1871  * is called from the socket option handlers.
1872  */
1873 int
1874 mld_change_state(struct in6_multi *inm, const int delay)
1875 {
1876 	struct mld_ifsoftc *mli;
1877 	struct ifnet *ifp;
1878 	int error;
1879 
1880 	IN6_MULTI_LOCK_ASSERT();
1881 
1882 	error = 0;
1883 
1884 	/*
1885 	 * Try to detect if the upper layer just asked us to change state
1886 	 * for an interface which has now gone away.
1887 	 */
1888 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1889 	ifp = inm->in6m_ifma->ifma_ifp;
1890 	if (ifp != NULL) {
1891 		/*
1892 		 * Sanity check that netinet6's notion of ifp is the
1893 		 * same as net's.
1894 		 */
1895 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1896 	}
1897 
1898 	MLD_LOCK();
1899 
1900 	mli = MLD_IFINFO(ifp);
1901 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1902 
1903 	/*
1904 	 * If we detect a state transition to or from MCAST_UNDEFINED
1905 	 * for this group, then we are starting or finishing an MLD
1906 	 * life cycle for this group.
1907 	 */
1908 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1909 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1910 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1911 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1912 			CTR1(KTR_MLD, "%s: initial join", __func__);
1913 			error = mld_initial_join(inm, mli, delay);
1914 			goto out_locked;
1915 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1916 			CTR1(KTR_MLD, "%s: final leave", __func__);
1917 			mld_final_leave(inm, mli);
1918 			goto out_locked;
1919 		}
1920 	} else {
1921 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1922 	}
1923 
1924 	error = mld_handle_state_change(inm, mli);
1925 
1926 out_locked:
1927 	MLD_UNLOCK();
1928 	return (error);
1929 }
1930 
1931 /*
1932  * Perform the initial join for an MLD group.
1933  *
1934  * When joining a group:
1935  *  If the group should have its MLD traffic suppressed, do nothing.
1936  *  MLDv1 starts sending MLDv1 host membership reports.
1937  *  MLDv2 will schedule an MLDv2 state-change report containing the
1938  *  initial state of the membership.
1939  *
1940  * If the delay argument is non-zero, then we must delay sending the
1941  * initial state change for delay ticks (in units of PR_FASTHZ).
1942  */
1943 static int
1944 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1945     const int delay)
1946 {
1947 	struct ifnet		*ifp;
1948 	struct mbufq		*mq;
1949 	int			 error, retval, syncstates;
1950 	int			 odelay;
1951 #ifdef KTR
1952 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1953 #endif
1954 
1955 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1956 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1957 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1958 
1959 	error = 0;
1960 	syncstates = 1;
1961 
1962 	ifp = inm->in6m_ifp;
1963 
1964 	IN6_MULTI_LOCK_ASSERT();
1965 	MLD_LOCK_ASSERT();
1966 
1967 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1968 
1969 	/*
1970 	 * Groups joined on loopback or marked as 'not reported',
1971 	 * enter the MLD_SILENT_MEMBER state and
1972 	 * are never reported in any protocol exchanges.
1973 	 * All other groups enter the appropriate state machine
1974 	 * for the version in use on this link.
1975 	 * A link marked as MLIF_SILENT causes MLD to be completely
1976 	 * disabled for the link.
1977 	 */
1978 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1979 	    (mli->mli_flags & MLIF_SILENT) ||
1980 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1981 		CTR1(KTR_MLD,
1982 "%s: not kicking state machine for silent group", __func__);
1983 		inm->in6m_state = MLD_SILENT_MEMBER;
1984 		inm->in6m_timer = 0;
1985 	} else {
1986 		/*
1987 		 * Deal with overlapping in_multi lifecycle.
1988 		 * If this group was LEAVING, then make sure
1989 		 * we drop the reference we picked up to keep the
1990 		 * group around for the final INCLUDE {} enqueue.
1991 		 */
1992 		if (mli->mli_version == MLD_VERSION_2 &&
1993 		    inm->in6m_state == MLD_LEAVING_MEMBER)
1994 			in6m_release_locked(inm);
1995 
1996 		inm->in6m_state = MLD_REPORTING_MEMBER;
1997 
1998 		switch (mli->mli_version) {
1999 		case MLD_VERSION_1:
2000 			/*
2001 			 * If a delay was provided, only use it if
2002 			 * it is greater than the delay normally
2003 			 * used for an MLDv1 state change report,
2004 			 * and delay sending the initial MLDv1 report
2005 			 * by not transitioning to the IDLE state.
2006 			 */
2007 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2008 			if (delay) {
2009 				inm->in6m_timer = max(delay, odelay);
2010 				V_current_state_timers_running6 = 1;
2011 			} else {
2012 				inm->in6m_state = MLD_IDLE_MEMBER;
2013 				error = mld_v1_transmit_report(inm,
2014 				     MLD_LISTENER_REPORT);
2015 				if (error == 0) {
2016 					inm->in6m_timer = odelay;
2017 					V_current_state_timers_running6 = 1;
2018 				}
2019 			}
2020 			break;
2021 
2022 		case MLD_VERSION_2:
2023 			/*
2024 			 * Defer update of T0 to T1, until the first copy
2025 			 * of the state change has been transmitted.
2026 			 */
2027 			syncstates = 0;
2028 
2029 			/*
2030 			 * Immediately enqueue a State-Change Report for
2031 			 * this interface, freeing any previous reports.
2032 			 * Don't kick the timers if there is nothing to do,
2033 			 * or if an error occurred.
2034 			 */
2035 			mq = &inm->in6m_scq;
2036 			mbufq_drain(mq);
2037 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2038 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2039 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2040 			    __func__, retval);
2041 			if (retval <= 0) {
2042 				error = retval * -1;
2043 				break;
2044 			}
2045 
2046 			/*
2047 			 * Schedule transmission of pending state-change
2048 			 * report up to RV times for this link. The timer
2049 			 * will fire at the next mld_fasttimo (~200ms),
2050 			 * giving us an opportunity to merge the reports.
2051 			 *
2052 			 * If a delay was provided to this function, only
2053 			 * use this delay if sooner than the existing one.
2054 			 */
2055 			KASSERT(mli->mli_rv > 1,
2056 			   ("%s: invalid robustness %d", __func__,
2057 			    mli->mli_rv));
2058 			inm->in6m_scrv = mli->mli_rv;
2059 			if (delay) {
2060 				if (inm->in6m_sctimer > 1) {
2061 					inm->in6m_sctimer =
2062 					    min(inm->in6m_sctimer, delay);
2063 				} else
2064 					inm->in6m_sctimer = delay;
2065 			} else
2066 				inm->in6m_sctimer = 1;
2067 			V_state_change_timers_running6 = 1;
2068 
2069 			error = 0;
2070 			break;
2071 		}
2072 	}
2073 
2074 	/*
2075 	 * Only update the T0 state if state change is atomic,
2076 	 * i.e. we don't need to wait for a timer to fire before we
2077 	 * can consider the state change to have been communicated.
2078 	 */
2079 	if (syncstates) {
2080 		in6m_commit(inm);
2081 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2082 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2083 		    if_name(inm->in6m_ifp));
2084 	}
2085 
2086 	return (error);
2087 }
2088 
2089 /*
2090  * Issue an intermediate state change during the life-cycle.
2091  */
2092 static int
2093 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2094 {
2095 	struct ifnet		*ifp;
2096 	int			 retval;
2097 #ifdef KTR
2098 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2099 #endif
2100 
2101 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2102 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2103 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2104 
2105 	ifp = inm->in6m_ifp;
2106 
2107 	IN6_MULTI_LOCK_ASSERT();
2108 	MLD_LOCK_ASSERT();
2109 
2110 	KASSERT(mli && mli->mli_ifp == ifp,
2111 	    ("%s: inconsistent ifp", __func__));
2112 
2113 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2114 	    (mli->mli_flags & MLIF_SILENT) ||
2115 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2116 	    (mli->mli_version != MLD_VERSION_2)) {
2117 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2118 			CTR1(KTR_MLD,
2119 "%s: not kicking state machine for silent group", __func__);
2120 		}
2121 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2122 		in6m_commit(inm);
2123 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2124 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2125 		    if_name(inm->in6m_ifp));
2126 		return (0);
2127 	}
2128 
2129 	mbufq_drain(&inm->in6m_scq);
2130 
2131 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2132 	    (mli->mli_flags & MLIF_USEALLOW));
2133 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2134 	if (retval <= 0)
2135 		return (-retval);
2136 
2137 	/*
2138 	 * If record(s) were enqueued, start the state-change
2139 	 * report timer for this group.
2140 	 */
2141 	inm->in6m_scrv = mli->mli_rv;
2142 	inm->in6m_sctimer = 1;
2143 	V_state_change_timers_running6 = 1;
2144 
2145 	return (0);
2146 }
2147 
2148 /*
2149  * Perform the final leave for a multicast address.
2150  *
2151  * When leaving a group:
2152  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2153  *  MLDv2 enqueues a state-change report containing a transition
2154  *  to INCLUDE {} for immediate transmission.
2155  */
2156 static void
2157 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2158 {
2159 	int syncstates;
2160 #ifdef KTR
2161 	char ip6tbuf[INET6_ADDRSTRLEN];
2162 #endif
2163 
2164 	syncstates = 1;
2165 
2166 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2167 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2168 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2169 
2170 	IN6_MULTI_LOCK_ASSERT();
2171 	MLD_LOCK_ASSERT();
2172 
2173 	switch (inm->in6m_state) {
2174 	case MLD_NOT_MEMBER:
2175 	case MLD_SILENT_MEMBER:
2176 	case MLD_LEAVING_MEMBER:
2177 		/* Already leaving or left; do nothing. */
2178 		CTR1(KTR_MLD,
2179 "%s: not kicking state machine for silent group", __func__);
2180 		break;
2181 	case MLD_REPORTING_MEMBER:
2182 	case MLD_IDLE_MEMBER:
2183 	case MLD_G_QUERY_PENDING_MEMBER:
2184 	case MLD_SG_QUERY_PENDING_MEMBER:
2185 		if (mli->mli_version == MLD_VERSION_1) {
2186 #ifdef INVARIANTS
2187 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2188 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2189 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2190 			     __func__);
2191 #endif
2192 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2193 			inm->in6m_state = MLD_NOT_MEMBER;
2194 			V_current_state_timers_running6 = 1;
2195 		} else if (mli->mli_version == MLD_VERSION_2) {
2196 			/*
2197 			 * Stop group timer and all pending reports.
2198 			 * Immediately enqueue a state-change report
2199 			 * TO_IN {} to be sent on the next fast timeout,
2200 			 * giving us an opportunity to merge reports.
2201 			 */
2202 			mbufq_drain(&inm->in6m_scq);
2203 			inm->in6m_timer = 0;
2204 			inm->in6m_scrv = mli->mli_rv;
2205 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2206 			    "pending retransmissions.", __func__,
2207 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2208 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2209 			if (inm->in6m_scrv == 0) {
2210 				inm->in6m_state = MLD_NOT_MEMBER;
2211 				inm->in6m_sctimer = 0;
2212 			} else {
2213 				int retval;
2214 
2215 				in6m_acquire_locked(inm);
2216 
2217 				retval = mld_v2_enqueue_group_record(
2218 				    &inm->in6m_scq, inm, 1, 0, 0,
2219 				    (mli->mli_flags & MLIF_USEALLOW));
2220 				KASSERT(retval != 0,
2221 				    ("%s: enqueue record = %d", __func__,
2222 				     retval));
2223 
2224 				inm->in6m_state = MLD_LEAVING_MEMBER;
2225 				inm->in6m_sctimer = 1;
2226 				V_state_change_timers_running6 = 1;
2227 				syncstates = 0;
2228 			}
2229 			break;
2230 		}
2231 		break;
2232 	case MLD_LAZY_MEMBER:
2233 	case MLD_SLEEPING_MEMBER:
2234 	case MLD_AWAKENING_MEMBER:
2235 		/* Our reports are suppressed; do nothing. */
2236 		break;
2237 	}
2238 
2239 	if (syncstates) {
2240 		in6m_commit(inm);
2241 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2242 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2243 		    if_name(inm->in6m_ifp));
2244 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2245 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2246 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2247 	}
2248 }
2249 
2250 /*
2251  * Enqueue an MLDv2 group record to the given output queue.
2252  *
2253  * If is_state_change is zero, a current-state record is appended.
2254  * If is_state_change is non-zero, a state-change report is appended.
2255  *
2256  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2257  * If is_group_query is zero, and if there is a packet with free space
2258  * at the tail of the queue, it will be appended to providing there
2259  * is enough free space.
2260  * Otherwise a new mbuf packet chain is allocated.
2261  *
2262  * If is_source_query is non-zero, each source is checked to see if
2263  * it was recorded for a Group-Source query, and will be omitted if
2264  * it is not both in-mode and recorded.
2265  *
2266  * If use_block_allow is non-zero, state change reports for initial join
2267  * and final leave, on an inclusive mode group with a source list, will be
2268  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2269  *
2270  * The function will attempt to allocate leading space in the packet
2271  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2272  *
2273  * If successful the size of all data appended to the queue is returned,
2274  * otherwise an error code less than zero is returned, or zero if
2275  * no record(s) were appended.
2276  */
2277 static int
2278 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2279     const int is_state_change, const int is_group_query,
2280     const int is_source_query, const int use_block_allow)
2281 {
2282 	struct mldv2_record	 mr;
2283 	struct mldv2_record	*pmr;
2284 	struct ifnet		*ifp;
2285 	struct ip6_msource	*ims, *nims;
2286 	struct mbuf		*m0, *m, *md;
2287 	int			 error, is_filter_list_change;
2288 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2289 	int			 record_has_sources;
2290 	int			 now;
2291 	int			 type;
2292 	uint8_t			 mode;
2293 #ifdef KTR
2294 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2295 #endif
2296 
2297 	IN6_MULTI_LOCK_ASSERT();
2298 
2299 	error = 0;
2300 	ifp = inm->in6m_ifp;
2301 	is_filter_list_change = 0;
2302 	m = NULL;
2303 	m0 = NULL;
2304 	m0srcs = 0;
2305 	msrcs = 0;
2306 	nbytes = 0;
2307 	nims = NULL;
2308 	record_has_sources = 1;
2309 	pmr = NULL;
2310 	type = MLD_DO_NOTHING;
2311 	mode = inm->in6m_st[1].iss_fmode;
2312 
2313 	/*
2314 	 * If we did not transition out of ASM mode during t0->t1,
2315 	 * and there are no source nodes to process, we can skip
2316 	 * the generation of source records.
2317 	 */
2318 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2319 	    inm->in6m_nsrc == 0)
2320 		record_has_sources = 0;
2321 
2322 	if (is_state_change) {
2323 		/*
2324 		 * Queue a state change record.
2325 		 * If the mode did not change, and there are non-ASM
2326 		 * listeners or source filters present,
2327 		 * we potentially need to issue two records for the group.
2328 		 * If there are ASM listeners, and there was no filter
2329 		 * mode transition of any kind, do nothing.
2330 		 *
2331 		 * If we are transitioning to MCAST_UNDEFINED, we need
2332 		 * not send any sources. A transition to/from this state is
2333 		 * considered inclusive with some special treatment.
2334 		 *
2335 		 * If we are rewriting initial joins/leaves to use
2336 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2337 		 * we need to send sources in all cases.
2338 		 */
2339 		if (mode != inm->in6m_st[0].iss_fmode) {
2340 			if (mode == MCAST_EXCLUDE) {
2341 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2342 				    __func__);
2343 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2344 			} else {
2345 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2346 				    __func__);
2347 				if (use_block_allow) {
2348 					/*
2349 					 * XXX
2350 					 * Here we're interested in state
2351 					 * edges either direction between
2352 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2353 					 * Perhaps we should just check
2354 					 * the group state, rather than
2355 					 * the filter mode.
2356 					 */
2357 					if (mode == MCAST_UNDEFINED) {
2358 						type = MLD_BLOCK_OLD_SOURCES;
2359 					} else {
2360 						type = MLD_ALLOW_NEW_SOURCES;
2361 					}
2362 				} else {
2363 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2364 					if (mode == MCAST_UNDEFINED)
2365 						record_has_sources = 0;
2366 				}
2367 			}
2368 		} else {
2369 			if (record_has_sources) {
2370 				is_filter_list_change = 1;
2371 			} else {
2372 				type = MLD_DO_NOTHING;
2373 			}
2374 		}
2375 	} else {
2376 		/*
2377 		 * Queue a current state record.
2378 		 */
2379 		if (mode == MCAST_EXCLUDE) {
2380 			type = MLD_MODE_IS_EXCLUDE;
2381 		} else if (mode == MCAST_INCLUDE) {
2382 			type = MLD_MODE_IS_INCLUDE;
2383 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2384 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2385 			     __func__, inm, inm->in6m_st[1].iss_asm));
2386 		}
2387 	}
2388 
2389 	/*
2390 	 * Generate the filter list changes using a separate function.
2391 	 */
2392 	if (is_filter_list_change)
2393 		return (mld_v2_enqueue_filter_change(mq, inm));
2394 
2395 	if (type == MLD_DO_NOTHING) {
2396 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2397 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2398 		    if_name(inm->in6m_ifp));
2399 		return (0);
2400 	}
2401 
2402 	/*
2403 	 * If any sources are present, we must be able to fit at least
2404 	 * one in the trailing space of the tail packet's mbuf,
2405 	 * ideally more.
2406 	 */
2407 	minrec0len = sizeof(struct mldv2_record);
2408 	if (record_has_sources)
2409 		minrec0len += sizeof(struct in6_addr);
2410 
2411 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2412 	    mld_rec_type_to_str(type),
2413 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2414 	    if_name(inm->in6m_ifp));
2415 
2416 	/*
2417 	 * Check if we have a packet in the tail of the queue for this
2418 	 * group into which the first group record for this group will fit.
2419 	 * Otherwise allocate a new packet.
2420 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2421 	 * Note: Group records for G/GSR query responses MUST be sent
2422 	 * in their own packet.
2423 	 */
2424 	m0 = mbufq_last(mq);
2425 	if (!is_group_query &&
2426 	    m0 != NULL &&
2427 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2428 	    (m0->m_pkthdr.len + minrec0len) <
2429 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2430 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2431 			    sizeof(struct mldv2_record)) /
2432 			    sizeof(struct in6_addr);
2433 		m = m0;
2434 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2435 	} else {
2436 		if (mbufq_full(mq)) {
2437 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2438 			return (-ENOMEM);
2439 		}
2440 		m = NULL;
2441 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2442 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2443 		if (!is_state_change && !is_group_query)
2444 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2445 		if (m == NULL)
2446 			m = m_gethdr(M_NOWAIT, MT_DATA);
2447 		if (m == NULL)
2448 			return (-ENOMEM);
2449 
2450 		mld_save_context(m, ifp);
2451 
2452 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2453 	}
2454 
2455 	/*
2456 	 * Append group record.
2457 	 * If we have sources, we don't know how many yet.
2458 	 */
2459 	mr.mr_type = type;
2460 	mr.mr_datalen = 0;
2461 	mr.mr_numsrc = 0;
2462 	mr.mr_addr = inm->in6m_addr;
2463 	in6_clearscope(&mr.mr_addr);
2464 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2465 		if (m != m0)
2466 			m_freem(m);
2467 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2468 		return (-ENOMEM);
2469 	}
2470 	nbytes += sizeof(struct mldv2_record);
2471 
2472 	/*
2473 	 * Append as many sources as will fit in the first packet.
2474 	 * If we are appending to a new packet, the chain allocation
2475 	 * may potentially use clusters; use m_getptr() in this case.
2476 	 * If we are appending to an existing packet, we need to obtain
2477 	 * a pointer to the group record after m_append(), in case a new
2478 	 * mbuf was allocated.
2479 	 *
2480 	 * Only append sources which are in-mode at t1. If we are
2481 	 * transitioning to MCAST_UNDEFINED state on the group, and
2482 	 * use_block_allow is zero, do not include source entries.
2483 	 * Otherwise, we need to include this source in the report.
2484 	 *
2485 	 * Only report recorded sources in our filter set when responding
2486 	 * to a group-source query.
2487 	 */
2488 	if (record_has_sources) {
2489 		if (m == m0) {
2490 			md = m_last(m);
2491 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2492 			    md->m_len - nbytes);
2493 		} else {
2494 			md = m_getptr(m, 0, &off);
2495 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2496 			    off);
2497 		}
2498 		msrcs = 0;
2499 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2500 		    nims) {
2501 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2502 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2503 			now = im6s_get_mode(inm, ims, 1);
2504 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2505 			if ((now != mode) ||
2506 			    (now == mode &&
2507 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2508 				CTR1(KTR_MLD, "%s: skip node", __func__);
2509 				continue;
2510 			}
2511 			if (is_source_query && ims->im6s_stp == 0) {
2512 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2513 				    __func__);
2514 				continue;
2515 			}
2516 			CTR1(KTR_MLD, "%s: append node", __func__);
2517 			if (!m_append(m, sizeof(struct in6_addr),
2518 			    (void *)&ims->im6s_addr)) {
2519 				if (m != m0)
2520 					m_freem(m);
2521 				CTR1(KTR_MLD, "%s: m_append() failed.",
2522 				    __func__);
2523 				return (-ENOMEM);
2524 			}
2525 			nbytes += sizeof(struct in6_addr);
2526 			++msrcs;
2527 			if (msrcs == m0srcs)
2528 				break;
2529 		}
2530 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2531 		    msrcs);
2532 		pmr->mr_numsrc = htons(msrcs);
2533 		nbytes += (msrcs * sizeof(struct in6_addr));
2534 	}
2535 
2536 	if (is_source_query && msrcs == 0) {
2537 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2538 		if (m != m0)
2539 			m_freem(m);
2540 		return (0);
2541 	}
2542 
2543 	/*
2544 	 * We are good to go with first packet.
2545 	 */
2546 	if (m != m0) {
2547 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2548 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2549 		mbufq_enqueue(mq, m);
2550 	} else
2551 		m->m_pkthdr.PH_vt.vt_nrecs++;
2552 
2553 	/*
2554 	 * No further work needed if no source list in packet(s).
2555 	 */
2556 	if (!record_has_sources)
2557 		return (nbytes);
2558 
2559 	/*
2560 	 * Whilst sources remain to be announced, we need to allocate
2561 	 * a new packet and fill out as many sources as will fit.
2562 	 * Always try for a cluster first.
2563 	 */
2564 	while (nims != NULL) {
2565 		if (mbufq_full(mq)) {
2566 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2567 			return (-ENOMEM);
2568 		}
2569 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2570 		if (m == NULL)
2571 			m = m_gethdr(M_NOWAIT, MT_DATA);
2572 		if (m == NULL)
2573 			return (-ENOMEM);
2574 		mld_save_context(m, ifp);
2575 		md = m_getptr(m, 0, &off);
2576 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2577 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2578 
2579 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2580 			if (m != m0)
2581 				m_freem(m);
2582 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2583 			return (-ENOMEM);
2584 		}
2585 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2586 		nbytes += sizeof(struct mldv2_record);
2587 
2588 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2589 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2590 
2591 		msrcs = 0;
2592 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2593 			CTR2(KTR_MLD, "%s: visit node %s",
2594 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2595 			now = im6s_get_mode(inm, ims, 1);
2596 			if ((now != mode) ||
2597 			    (now == mode &&
2598 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2599 				CTR1(KTR_MLD, "%s: skip node", __func__);
2600 				continue;
2601 			}
2602 			if (is_source_query && ims->im6s_stp == 0) {
2603 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2604 				    __func__);
2605 				continue;
2606 			}
2607 			CTR1(KTR_MLD, "%s: append node", __func__);
2608 			if (!m_append(m, sizeof(struct in6_addr),
2609 			    (void *)&ims->im6s_addr)) {
2610 				if (m != m0)
2611 					m_freem(m);
2612 				CTR1(KTR_MLD, "%s: m_append() failed.",
2613 				    __func__);
2614 				return (-ENOMEM);
2615 			}
2616 			++msrcs;
2617 			if (msrcs == m0srcs)
2618 				break;
2619 		}
2620 		pmr->mr_numsrc = htons(msrcs);
2621 		nbytes += (msrcs * sizeof(struct in6_addr));
2622 
2623 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2624 		mbufq_enqueue(mq, m);
2625 	}
2626 
2627 	return (nbytes);
2628 }
2629 
2630 /*
2631  * Type used to mark record pass completion.
2632  * We exploit the fact we can cast to this easily from the
2633  * current filter modes on each ip_msource node.
2634  */
2635 typedef enum {
2636 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2637 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2638 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2639 	REC_FULL = REC_ALLOW | REC_BLOCK
2640 } rectype_t;
2641 
2642 /*
2643  * Enqueue an MLDv2 filter list change to the given output queue.
2644  *
2645  * Source list filter state is held in an RB-tree. When the filter list
2646  * for a group is changed without changing its mode, we need to compute
2647  * the deltas between T0 and T1 for each source in the filter set,
2648  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2649  *
2650  * As we may potentially queue two record types, and the entire R-B tree
2651  * needs to be walked at once, we break this out into its own function
2652  * so we can generate a tightly packed queue of packets.
2653  *
2654  * XXX This could be written to only use one tree walk, although that makes
2655  * serializing into the mbuf chains a bit harder. For now we do two walks
2656  * which makes things easier on us, and it may or may not be harder on
2657  * the L2 cache.
2658  *
2659  * If successful the size of all data appended to the queue is returned,
2660  * otherwise an error code less than zero is returned, or zero if
2661  * no record(s) were appended.
2662  */
2663 static int
2664 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2665 {
2666 	static const int MINRECLEN =
2667 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2668 	struct ifnet		*ifp;
2669 	struct mldv2_record	 mr;
2670 	struct mldv2_record	*pmr;
2671 	struct ip6_msource	*ims, *nims;
2672 	struct mbuf		*m, *m0, *md;
2673 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2674 	int			 nallow, nblock;
2675 	uint8_t			 mode, now, then;
2676 	rectype_t		 crt, drt, nrt;
2677 #ifdef KTR
2678 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2679 #endif
2680 
2681 	IN6_MULTI_LOCK_ASSERT();
2682 
2683 	if (inm->in6m_nsrc == 0 ||
2684 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2685 		return (0);
2686 
2687 	ifp = inm->in6m_ifp;			/* interface */
2688 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2689 	crt = REC_NONE;	/* current group record type */
2690 	drt = REC_NONE;	/* mask of completed group record types */
2691 	nrt = REC_NONE;	/* record type for current node */
2692 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2693 	npbytes = 0;	/* # of bytes appended this packet */
2694 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2695 	rsrcs = 0;	/* # sources encoded in current record */
2696 	schanged = 0;	/* # nodes encoded in overall filter change */
2697 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2698 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2699 	nims = NULL;	/* next tree node pointer */
2700 
2701 	/*
2702 	 * For each possible filter record mode.
2703 	 * The first kind of source we encounter tells us which
2704 	 * is the first kind of record we start appending.
2705 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2706 	 * as the inverse of the group's filter mode.
2707 	 */
2708 	while (drt != REC_FULL) {
2709 		do {
2710 			m0 = mbufq_last(mq);
2711 			if (m0 != NULL &&
2712 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2713 			     MLD_V2_REPORT_MAXRECS) &&
2714 			    (m0->m_pkthdr.len + MINRECLEN) <
2715 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2716 				m = m0;
2717 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2718 					    sizeof(struct mldv2_record)) /
2719 					    sizeof(struct in6_addr);
2720 				CTR1(KTR_MLD,
2721 				    "%s: use previous packet", __func__);
2722 			} else {
2723 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2724 				if (m == NULL)
2725 					m = m_gethdr(M_NOWAIT, MT_DATA);
2726 				if (m == NULL) {
2727 					CTR1(KTR_MLD,
2728 					    "%s: m_get*() failed", __func__);
2729 					return (-ENOMEM);
2730 				}
2731 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2732 				mld_save_context(m, ifp);
2733 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2734 				    sizeof(struct mldv2_record)) /
2735 				    sizeof(struct in6_addr);
2736 				npbytes = 0;
2737 				CTR1(KTR_MLD,
2738 				    "%s: allocated new packet", __func__);
2739 			}
2740 			/*
2741 			 * Append the MLD group record header to the
2742 			 * current packet's data area.
2743 			 * Recalculate pointer to free space for next
2744 			 * group record, in case m_append() allocated
2745 			 * a new mbuf or cluster.
2746 			 */
2747 			memset(&mr, 0, sizeof(mr));
2748 			mr.mr_addr = inm->in6m_addr;
2749 			in6_clearscope(&mr.mr_addr);
2750 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2751 				if (m != m0)
2752 					m_freem(m);
2753 				CTR1(KTR_MLD,
2754 				    "%s: m_append() failed", __func__);
2755 				return (-ENOMEM);
2756 			}
2757 			npbytes += sizeof(struct mldv2_record);
2758 			if (m != m0) {
2759 				/* new packet; offset in chain */
2760 				md = m_getptr(m, npbytes -
2761 				    sizeof(struct mldv2_record), &off);
2762 				pmr = (struct mldv2_record *)(mtod(md,
2763 				    uint8_t *) + off);
2764 			} else {
2765 				/* current packet; offset from last append */
2766 				md = m_last(m);
2767 				pmr = (struct mldv2_record *)(mtod(md,
2768 				    uint8_t *) + md->m_len -
2769 				    sizeof(struct mldv2_record));
2770 			}
2771 			/*
2772 			 * Begin walking the tree for this record type
2773 			 * pass, or continue from where we left off
2774 			 * previously if we had to allocate a new packet.
2775 			 * Only report deltas in-mode at t1.
2776 			 * We need not report included sources as allowed
2777 			 * if we are in inclusive mode on the group,
2778 			 * however the converse is not true.
2779 			 */
2780 			rsrcs = 0;
2781 			if (nims == NULL) {
2782 				nims = RB_MIN(ip6_msource_tree,
2783 				    &inm->in6m_srcs);
2784 			}
2785 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2786 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2787 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2788 				now = im6s_get_mode(inm, ims, 1);
2789 				then = im6s_get_mode(inm, ims, 0);
2790 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2791 				    __func__, then, now);
2792 				if (now == then) {
2793 					CTR1(KTR_MLD,
2794 					    "%s: skip unchanged", __func__);
2795 					continue;
2796 				}
2797 				if (mode == MCAST_EXCLUDE &&
2798 				    now == MCAST_INCLUDE) {
2799 					CTR1(KTR_MLD,
2800 					    "%s: skip IN src on EX group",
2801 					    __func__);
2802 					continue;
2803 				}
2804 				nrt = (rectype_t)now;
2805 				if (nrt == REC_NONE)
2806 					nrt = (rectype_t)(~mode & REC_FULL);
2807 				if (schanged++ == 0) {
2808 					crt = nrt;
2809 				} else if (crt != nrt)
2810 					continue;
2811 				if (!m_append(m, sizeof(struct in6_addr),
2812 				    (void *)&ims->im6s_addr)) {
2813 					if (m != m0)
2814 						m_freem(m);
2815 					CTR1(KTR_MLD,
2816 					    "%s: m_append() failed", __func__);
2817 					return (-ENOMEM);
2818 				}
2819 				nallow += !!(crt == REC_ALLOW);
2820 				nblock += !!(crt == REC_BLOCK);
2821 				if (++rsrcs == m0srcs)
2822 					break;
2823 			}
2824 			/*
2825 			 * If we did not append any tree nodes on this
2826 			 * pass, back out of allocations.
2827 			 */
2828 			if (rsrcs == 0) {
2829 				npbytes -= sizeof(struct mldv2_record);
2830 				if (m != m0) {
2831 					CTR1(KTR_MLD,
2832 					    "%s: m_free(m)", __func__);
2833 					m_freem(m);
2834 				} else {
2835 					CTR1(KTR_MLD,
2836 					    "%s: m_adj(m, -mr)", __func__);
2837 					m_adj(m, -((int)sizeof(
2838 					    struct mldv2_record)));
2839 				}
2840 				continue;
2841 			}
2842 			npbytes += (rsrcs * sizeof(struct in6_addr));
2843 			if (crt == REC_ALLOW)
2844 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2845 			else if (crt == REC_BLOCK)
2846 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2847 			pmr->mr_numsrc = htons(rsrcs);
2848 			/*
2849 			 * Count the new group record, and enqueue this
2850 			 * packet if it wasn't already queued.
2851 			 */
2852 			m->m_pkthdr.PH_vt.vt_nrecs++;
2853 			if (m != m0)
2854 				mbufq_enqueue(mq, m);
2855 			nbytes += npbytes;
2856 		} while (nims != NULL);
2857 		drt |= crt;
2858 		crt = (~crt & REC_FULL);
2859 	}
2860 
2861 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2862 	    nallow, nblock);
2863 
2864 	return (nbytes);
2865 }
2866 
2867 static int
2868 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2869 {
2870 	struct mbufq	*gq;
2871 	struct mbuf	*m;		/* pending state-change */
2872 	struct mbuf	*m0;		/* copy of pending state-change */
2873 	struct mbuf	*mt;		/* last state-change in packet */
2874 	int		 docopy, domerge;
2875 	u_int		 recslen;
2876 
2877 	docopy = 0;
2878 	domerge = 0;
2879 	recslen = 0;
2880 
2881 	IN6_MULTI_LOCK_ASSERT();
2882 	MLD_LOCK_ASSERT();
2883 
2884 	/*
2885 	 * If there are further pending retransmissions, make a writable
2886 	 * copy of each queued state-change message before merging.
2887 	 */
2888 	if (inm->in6m_scrv > 0)
2889 		docopy = 1;
2890 
2891 	gq = &inm->in6m_scq;
2892 #ifdef KTR
2893 	if (mbufq_first(gq) == NULL) {
2894 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2895 		    __func__, inm);
2896 	}
2897 #endif
2898 
2899 	m = mbufq_first(gq);
2900 	while (m != NULL) {
2901 		/*
2902 		 * Only merge the report into the current packet if
2903 		 * there is sufficient space to do so; an MLDv2 report
2904 		 * packet may only contain 65,535 group records.
2905 		 * Always use a simple mbuf chain concatentation to do this,
2906 		 * as large state changes for single groups may have
2907 		 * allocated clusters.
2908 		 */
2909 		domerge = 0;
2910 		mt = mbufq_last(scq);
2911 		if (mt != NULL) {
2912 			recslen = m_length(m, NULL);
2913 
2914 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2915 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2916 			    MLD_V2_REPORT_MAXRECS) &&
2917 			    (mt->m_pkthdr.len + recslen <=
2918 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2919 				domerge = 1;
2920 		}
2921 
2922 		if (!domerge && mbufq_full(gq)) {
2923 			CTR2(KTR_MLD,
2924 			    "%s: outbound queue full, skipping whole packet %p",
2925 			    __func__, m);
2926 			mt = m->m_nextpkt;
2927 			if (!docopy)
2928 				m_freem(m);
2929 			m = mt;
2930 			continue;
2931 		}
2932 
2933 		if (!docopy) {
2934 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2935 			m0 = mbufq_dequeue(gq);
2936 			m = m0->m_nextpkt;
2937 		} else {
2938 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2939 			m0 = m_dup(m, M_NOWAIT);
2940 			if (m0 == NULL)
2941 				return (ENOMEM);
2942 			m0->m_nextpkt = NULL;
2943 			m = m->m_nextpkt;
2944 		}
2945 
2946 		if (!domerge) {
2947 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2948 			    __func__, m0, scq);
2949 			mbufq_enqueue(scq, m0);
2950 		} else {
2951 			struct mbuf *mtl;	/* last mbuf of packet mt */
2952 
2953 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2954 			    __func__, m0, mt);
2955 
2956 			mtl = m_last(mt);
2957 			m0->m_flags &= ~M_PKTHDR;
2958 			mt->m_pkthdr.len += recslen;
2959 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2960 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2961 
2962 			mtl->m_next = m0;
2963 		}
2964 	}
2965 
2966 	return (0);
2967 }
2968 
2969 /*
2970  * Respond to a pending MLDv2 General Query.
2971  */
2972 static void
2973 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2974 {
2975 	struct ifmultiaddr	*ifma;
2976 	struct ifnet		*ifp;
2977 	struct in6_multi	*inm;
2978 	int			 retval;
2979 
2980 	IN6_MULTI_LOCK_ASSERT();
2981 	MLD_LOCK_ASSERT();
2982 
2983 	KASSERT(mli->mli_version == MLD_VERSION_2,
2984 	    ("%s: called when version %d", __func__, mli->mli_version));
2985 
2986 	/*
2987 	 * Check that there are some packets queued. If so, send them first.
2988 	 * For large number of groups the reply to general query can take
2989 	 * many packets, we should finish sending them before starting of
2990 	 * queuing the new reply.
2991 	 */
2992 	if (mbufq_len(&mli->mli_gq) != 0)
2993 		goto send;
2994 
2995 	ifp = mli->mli_ifp;
2996 
2997 	IF_ADDR_RLOCK(ifp);
2998 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2999 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
3000 		    ifma->ifma_protospec == NULL)
3001 			continue;
3002 
3003 		inm = (struct in6_multi *)ifma->ifma_protospec;
3004 		KASSERT(ifp == inm->in6m_ifp,
3005 		    ("%s: inconsistent ifp", __func__));
3006 
3007 		switch (inm->in6m_state) {
3008 		case MLD_NOT_MEMBER:
3009 		case MLD_SILENT_MEMBER:
3010 			break;
3011 		case MLD_REPORTING_MEMBER:
3012 		case MLD_IDLE_MEMBER:
3013 		case MLD_LAZY_MEMBER:
3014 		case MLD_SLEEPING_MEMBER:
3015 		case MLD_AWAKENING_MEMBER:
3016 			inm->in6m_state = MLD_REPORTING_MEMBER;
3017 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3018 			    inm, 0, 0, 0, 0);
3019 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3020 			    __func__, retval);
3021 			break;
3022 		case MLD_G_QUERY_PENDING_MEMBER:
3023 		case MLD_SG_QUERY_PENDING_MEMBER:
3024 		case MLD_LEAVING_MEMBER:
3025 			break;
3026 		}
3027 	}
3028 	IF_ADDR_RUNLOCK(ifp);
3029 
3030 send:
3031 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3032 
3033 	/*
3034 	 * Slew transmission of bursts over 500ms intervals.
3035 	 */
3036 	if (mbufq_first(&mli->mli_gq) != NULL) {
3037 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3038 		    MLD_RESPONSE_BURST_INTERVAL);
3039 		V_interface_timers_running6 = 1;
3040 	}
3041 }
3042 
3043 /*
3044  * Transmit the next pending message in the output queue.
3045  *
3046  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3047  * MRT: Nothing needs to be done, as MLD traffic is always local to
3048  * a link and uses a link-scope multicast address.
3049  */
3050 static void
3051 mld_dispatch_packet(struct mbuf *m)
3052 {
3053 	struct ip6_moptions	 im6o;
3054 	struct ifnet		*ifp;
3055 	struct ifnet		*oifp;
3056 	struct mbuf		*m0;
3057 	struct mbuf		*md;
3058 	struct ip6_hdr		*ip6;
3059 	struct mld_hdr		*mld;
3060 	int			 error;
3061 	int			 off;
3062 	int			 type;
3063 	uint32_t		 ifindex;
3064 
3065 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3066 
3067 	/*
3068 	 * Set VNET image pointer from enqueued mbuf chain
3069 	 * before doing anything else. Whilst we use interface
3070 	 * indexes to guard against interface detach, they are
3071 	 * unique to each VIMAGE and must be retrieved.
3072 	 */
3073 	ifindex = mld_restore_context(m);
3074 
3075 	/*
3076 	 * Check if the ifnet still exists. This limits the scope of
3077 	 * any race in the absence of a global ifp lock for low cost
3078 	 * (an array lookup).
3079 	 */
3080 	ifp = ifnet_byindex(ifindex);
3081 	if (ifp == NULL) {
3082 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3083 		    __func__, m, ifindex);
3084 		m_freem(m);
3085 		IP6STAT_INC(ip6s_noroute);
3086 		goto out;
3087 	}
3088 
3089 	im6o.im6o_multicast_hlim  = 1;
3090 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3091 	im6o.im6o_multicast_ifp = ifp;
3092 
3093 	if (m->m_flags & M_MLDV1) {
3094 		m0 = m;
3095 	} else {
3096 		m0 = mld_v2_encap_report(ifp, m);
3097 		if (m0 == NULL) {
3098 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3099 			IP6STAT_INC(ip6s_odropped);
3100 			goto out;
3101 		}
3102 	}
3103 
3104 	mld_scrub_context(m0);
3105 	m_clrprotoflags(m);
3106 	m0->m_pkthdr.rcvif = V_loif;
3107 
3108 	ip6 = mtod(m0, struct ip6_hdr *);
3109 #if 0
3110 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3111 #else
3112 	/*
3113 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3114 	 * occur if we called in6_setscope() at transmission.
3115 	 * See comments at top of file.
3116 	 */
3117 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3118 #endif
3119 
3120 	/*
3121 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3122 	 * so we can bump the stats.
3123 	 */
3124 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3125 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3126 	type = mld->mld_type;
3127 
3128 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3129 	    &oifp, NULL);
3130 	if (error) {
3131 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3132 		goto out;
3133 	}
3134 	ICMP6STAT_INC(icp6s_outhist[type]);
3135 	if (oifp != NULL) {
3136 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3137 		switch (type) {
3138 		case MLD_LISTENER_REPORT:
3139 		case MLDV2_LISTENER_REPORT:
3140 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3141 			break;
3142 		case MLD_LISTENER_DONE:
3143 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3144 			break;
3145 		}
3146 	}
3147 out:
3148 	return;
3149 }
3150 
3151 /*
3152  * Encapsulate an MLDv2 report.
3153  *
3154  * KAME IPv6 requires that hop-by-hop options be passed separately,
3155  * and that the IPv6 header be prepended in a separate mbuf.
3156  *
3157  * Returns a pointer to the new mbuf chain head, or NULL if the
3158  * allocation failed.
3159  */
3160 static struct mbuf *
3161 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3162 {
3163 	struct mbuf		*mh;
3164 	struct mldv2_report	*mld;
3165 	struct ip6_hdr		*ip6;
3166 	struct in6_ifaddr	*ia;
3167 	int			 mldreclen;
3168 
3169 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3170 	KASSERT((m->m_flags & M_PKTHDR),
3171 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3172 
3173 	/*
3174 	 * RFC3590: OK to send as :: or tentative during DAD.
3175 	 */
3176 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3177 	if (ia == NULL)
3178 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3179 
3180 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3181 	if (mh == NULL) {
3182 		if (ia != NULL)
3183 			ifa_free(&ia->ia_ifa);
3184 		m_freem(m);
3185 		return (NULL);
3186 	}
3187 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3188 
3189 	mldreclen = m_length(m, NULL);
3190 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3191 
3192 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3193 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3194 	    sizeof(struct mldv2_report) + mldreclen;
3195 
3196 	ip6 = mtod(mh, struct ip6_hdr *);
3197 	ip6->ip6_flow = 0;
3198 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3199 	ip6->ip6_vfc |= IPV6_VERSION;
3200 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3201 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3202 	if (ia != NULL)
3203 		ifa_free(&ia->ia_ifa);
3204 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3205 	/* scope ID will be set in netisr */
3206 
3207 	mld = (struct mldv2_report *)(ip6 + 1);
3208 	mld->mld_type = MLDV2_LISTENER_REPORT;
3209 	mld->mld_code = 0;
3210 	mld->mld_cksum = 0;
3211 	mld->mld_v2_reserved = 0;
3212 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3213 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3214 
3215 	mh->m_next = m;
3216 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3217 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3218 	return (mh);
3219 }
3220 
3221 #ifdef KTR
3222 static char *
3223 mld_rec_type_to_str(const int type)
3224 {
3225 
3226 	switch (type) {
3227 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3228 			return "TO_EX";
3229 			break;
3230 		case MLD_CHANGE_TO_INCLUDE_MODE:
3231 			return "TO_IN";
3232 			break;
3233 		case MLD_MODE_IS_EXCLUDE:
3234 			return "MODE_EX";
3235 			break;
3236 		case MLD_MODE_IS_INCLUDE:
3237 			return "MODE_IN";
3238 			break;
3239 		case MLD_ALLOW_NEW_SOURCES:
3240 			return "ALLOW_NEW";
3241 			break;
3242 		case MLD_BLOCK_OLD_SOURCES:
3243 			return "BLOCK_OLD";
3244 			break;
3245 		default:
3246 			break;
3247 	}
3248 	return "unknown";
3249 }
3250 #endif
3251 
3252 static void
3253 mld_init(void *unused __unused)
3254 {
3255 
3256 	CTR1(KTR_MLD, "%s: initializing", __func__);
3257 	MLD_LOCK_INIT();
3258 
3259 	ip6_initpktopts(&mld_po);
3260 	mld_po.ip6po_hlim = 1;
3261 	mld_po.ip6po_hbh = &mld_ra.hbh;
3262 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3263 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3264 }
3265 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3266 
3267 static void
3268 mld_uninit(void *unused __unused)
3269 {
3270 
3271 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3272 	MLD_LOCK_DESTROY();
3273 }
3274 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3275 
3276 static void
3277 vnet_mld_init(const void *unused __unused)
3278 {
3279 
3280 	CTR1(KTR_MLD, "%s: initializing", __func__);
3281 
3282 	LIST_INIT(&V_mli_head);
3283 }
3284 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3285     NULL);
3286 
3287 static void
3288 vnet_mld_uninit(const void *unused __unused)
3289 {
3290 
3291 	/* This can happen if we shutdown the network stack. */
3292 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3293 }
3294 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3295     NULL);
3296 
3297 static int
3298 mld_modevent(module_t mod, int type, void *unused __unused)
3299 {
3300 
3301     switch (type) {
3302     case MOD_LOAD:
3303     case MOD_UNLOAD:
3304 	break;
3305     default:
3306 	return (EOPNOTSUPP);
3307     }
3308     return (0);
3309 }
3310 
3311 static moduledata_t mld_mod = {
3312     "mld",
3313     mld_modevent,
3314     0
3315 };
3316 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3317