xref: /freebsd/sys/netinet6/mld6.c (revision 389e4940069316fe667ffa263fa7d6390d0a960f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote
15  *    products derived from this software without specific prior written
16  *    permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
31  */
32 
33 /*-
34  * Copyright (c) 1988 Stephen Deering.
35  * Copyright (c) 1992, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * This code is derived from software contributed to Berkeley by
39  * Stephen Deering of Stanford University.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
66  */
67 
68 #include <sys/cdefs.h>
69 __FBSDID("$FreeBSD$");
70 
71 #include "opt_inet.h"
72 #include "opt_inet6.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/protosw.h>
79 #include <sys/sysctl.h>
80 #include <sys/kernel.h>
81 #include <sys/callout.h>
82 #include <sys/malloc.h>
83 #include <sys/module.h>
84 #include <sys/ktr.h>
85 
86 #include <net/if.h>
87 #include <net/if_var.h>
88 #include <net/route.h>
89 #include <net/vnet.h>
90 
91 #include <netinet/in.h>
92 #include <netinet/in_var.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/mld6.h>
99 #include <netinet6/mld6_var.h>
100 
101 #include <security/mac/mac_framework.h>
102 
103 #ifndef KTR_MLD
104 #define KTR_MLD KTR_INET6
105 #endif
106 
107 static struct mld_ifsoftc *
108 		mli_alloc_locked(struct ifnet *);
109 static void	mli_delete_locked(const struct ifnet *);
110 static void	mld_dispatch_packet(struct mbuf *);
111 static void	mld_dispatch_queue(struct mbufq *, int);
112 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
113 static void	mld_fasttimo_vnet(void);
114 static int	mld_handle_state_change(struct in6_multi *,
115 		    struct mld_ifsoftc *);
116 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
117 		    const int);
118 #ifdef KTR
119 static char *	mld_rec_type_to_str(const int);
120 #endif
121 static void	mld_set_version(struct mld_ifsoftc *, const int);
122 static void	mld_slowtimo_vnet(void);
123 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
126 		    /*const*/ struct mld_hdr *);
127 static void	mld_v1_process_group_timer(struct in6_multi_head *,
128 		    struct in6_multi *);
129 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
130 static int	mld_v1_transmit_report(struct in6_multi *, const int);
131 static void	mld_v1_update_group(struct in6_multi *, const int);
132 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
133 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
134 static struct mbuf *
135 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
136 static int	mld_v2_enqueue_filter_change(struct mbufq *,
137 		    struct in6_multi *);
138 static int	mld_v2_enqueue_group_record(struct mbufq *,
139 		    struct in6_multi *, const int, const int, const int,
140 		    const int);
141 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
142 		    struct mbuf *, const int, const int);
143 static int	mld_v2_merge_state_changes(struct in6_multi *,
144 		    struct mbufq *);
145 static void	mld_v2_process_group_timers(struct in6_multi_head *,
146 		    struct mbufq *, struct mbufq *,
147 		    struct in6_multi *, const int);
148 static int	mld_v2_process_group_query(struct in6_multi *,
149 		    struct mld_ifsoftc *mli, int, struct mbuf *, const int);
150 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
151 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
152 
153 /*
154  * Normative references: RFC 2710, RFC 3590, RFC 3810.
155  *
156  * Locking:
157  *  * The MLD subsystem lock ends up being system-wide for the moment,
158  *    but could be per-VIMAGE later on.
159  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
160  *    Any may be taken independently; if any are held at the same
161  *    time, the above lock order must be followed.
162  *  * IN6_MULTI_LOCK covers in_multi.
163  *  * MLD_LOCK covers per-link state and any global variables in this file.
164  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
165  *    per-link state iterators.
166  *
167  *  XXX LOR PREVENTION
168  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
169  *  will not accept an ifp; it wants an embedded scope ID, unlike
170  *  ip_output(), which happily takes the ifp given to it. The embedded
171  *  scope ID is only used by MLD to select the outgoing interface.
172  *
173  *  During interface attach and detach, MLD will take MLD_LOCK *after*
174  *  the IF_AFDATA_LOCK.
175  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
176  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
177  *  dispatch could work around this, but we'd rather not do that, as it
178  *  can introduce other races.
179  *
180  *  As such, we exploit the fact that the scope ID is just the interface
181  *  index, and embed it in the IPv6 destination address accordingly.
182  *  This is potentially NOT VALID for MLDv1 reports, as they
183  *  are always sent to the multicast group itself; as MLDv2
184  *  reports are always sent to ff02::16, this is not an issue
185  *  when MLDv2 is in use.
186  *
187  *  This does not however eliminate the LOR when ip6_output() itself
188  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
189  *  trigger a LOR warning in WITNESS when the ifnet is detached.
190  *
191  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
192  *  how it's used across the network stack. Here we're simply exploiting
193  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
194  *
195  * VIMAGE:
196  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
197  *    to a vnet in ifp->if_vnet.
198  */
199 static struct mtx		 mld_mtx;
200 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
201 
202 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
203 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
204 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
205 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
206 
207 /*
208  * VIMAGE-wide globals.
209  */
210 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0};
211 static VNET_DEFINE(LIST_HEAD(, mld_ifsoftc), mli_head);
212 static VNET_DEFINE(int, interface_timers_running6);
213 static VNET_DEFINE(int, state_change_timers_running6);
214 static VNET_DEFINE(int, current_state_timers_running6);
215 
216 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
217 #define	V_mli_head			VNET(mli_head)
218 #define	V_interface_timers_running6	VNET(interface_timers_running6)
219 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
220 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
221 
222 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
223 
224 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
225     "IPv6 Multicast Listener Discovery");
226 
227 /*
228  * Virtualized sysctls.
229  */
230 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
231     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
232     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
233     "Rate limit for MLDv2 Group-and-Source queries in seconds");
234 
235 /*
236  * Non-virtualized sysctls.
237  */
238 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
239     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
240     "Per-interface MLDv2 state");
241 
242 static int	mld_v1enable = 1;
243 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
244     &mld_v1enable, 0, "Enable fallback to MLDv1");
245 
246 static int	mld_use_allow = 1;
247 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
248     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
249 
250 /*
251  * Packed Router Alert option structure declaration.
252  */
253 struct mld_raopt {
254 	struct ip6_hbh		hbh;
255 	struct ip6_opt		pad;
256 	struct ip6_opt_router	ra;
257 } __packed;
258 
259 /*
260  * Router Alert hop-by-hop option header.
261  */
262 static struct mld_raopt mld_ra = {
263 	.hbh = { 0, 0 },
264 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
265 	.ra = {
266 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
267 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
268 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
269 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
270 	}
271 };
272 static struct ip6_pktopts mld_po;
273 
274 static __inline void
275 mld_save_context(struct mbuf *m, struct ifnet *ifp)
276 {
277 
278 #ifdef VIMAGE
279 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
280 #endif /* VIMAGE */
281 	m->m_pkthdr.flowid = ifp->if_index;
282 }
283 
284 static __inline void
285 mld_scrub_context(struct mbuf *m)
286 {
287 
288 	m->m_pkthdr.PH_loc.ptr = NULL;
289 	m->m_pkthdr.flowid = 0;
290 }
291 
292 /*
293  * Restore context from a queued output chain.
294  * Return saved ifindex.
295  *
296  * VIMAGE: The assertion is there to make sure that we
297  * actually called CURVNET_SET() with what's in the mbuf chain.
298  */
299 static __inline uint32_t
300 mld_restore_context(struct mbuf *m)
301 {
302 
303 #if defined(VIMAGE) && defined(INVARIANTS)
304 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
305 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
306 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
307 #endif
308 	return (m->m_pkthdr.flowid);
309 }
310 
311 /*
312  * Retrieve or set threshold between group-source queries in seconds.
313  *
314  * VIMAGE: Assume curvnet set by caller.
315  * SMPng: NOTE: Serialized by MLD lock.
316  */
317 static int
318 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
319 {
320 	int error;
321 	int i;
322 
323 	error = sysctl_wire_old_buffer(req, sizeof(int));
324 	if (error)
325 		return (error);
326 
327 	MLD_LOCK();
328 
329 	i = V_mld_gsrdelay.tv_sec;
330 
331 	error = sysctl_handle_int(oidp, &i, 0, req);
332 	if (error || !req->newptr)
333 		goto out_locked;
334 
335 	if (i < -1 || i >= 60) {
336 		error = EINVAL;
337 		goto out_locked;
338 	}
339 
340 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
341 	     V_mld_gsrdelay.tv_sec, i);
342 	V_mld_gsrdelay.tv_sec = i;
343 
344 out_locked:
345 	MLD_UNLOCK();
346 	return (error);
347 }
348 
349 /*
350  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
351  * For use by ifmcstat(8).
352  *
353  * SMPng: NOTE: Does an unlocked ifindex space read.
354  * VIMAGE: Assume curvnet set by caller. The node handler itself
355  * is not directly virtualized.
356  */
357 static int
358 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
359 {
360 	int			*name;
361 	int			 error;
362 	u_int			 namelen;
363 	struct ifnet		*ifp;
364 	struct mld_ifsoftc	*mli;
365 
366 	name = (int *)arg1;
367 	namelen = arg2;
368 
369 	if (req->newptr != NULL)
370 		return (EPERM);
371 
372 	if (namelen != 1)
373 		return (EINVAL);
374 
375 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
376 	if (error)
377 		return (error);
378 
379 	IN6_MULTI_LOCK();
380 	IN6_MULTI_LIST_LOCK();
381 	MLD_LOCK();
382 
383 	if (name[0] <= 0 || name[0] > V_if_index) {
384 		error = ENOENT;
385 		goto out_locked;
386 	}
387 
388 	error = ENOENT;
389 
390 	ifp = ifnet_byindex(name[0]);
391 	if (ifp == NULL)
392 		goto out_locked;
393 
394 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
395 		if (ifp == mli->mli_ifp) {
396 			struct mld_ifinfo info;
397 
398 			info.mli_version = mli->mli_version;
399 			info.mli_v1_timer = mli->mli_v1_timer;
400 			info.mli_v2_timer = mli->mli_v2_timer;
401 			info.mli_flags = mli->mli_flags;
402 			info.mli_rv = mli->mli_rv;
403 			info.mli_qi = mli->mli_qi;
404 			info.mli_qri = mli->mli_qri;
405 			info.mli_uri = mli->mli_uri;
406 			error = SYSCTL_OUT(req, &info, sizeof(info));
407 			break;
408 		}
409 	}
410 
411 out_locked:
412 	MLD_UNLOCK();
413 	IN6_MULTI_LIST_UNLOCK();
414 	IN6_MULTI_UNLOCK();
415 	return (error);
416 }
417 
418 /*
419  * Dispatch an entire queue of pending packet chains.
420  * VIMAGE: Assumes the vnet pointer has been set.
421  */
422 static void
423 mld_dispatch_queue(struct mbufq *mq, int limit)
424 {
425 	struct mbuf *m;
426 
427 	while ((m = mbufq_dequeue(mq)) != NULL) {
428 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
429 		mld_dispatch_packet(m);
430 		if (--limit == 0)
431 			break;
432 	}
433 }
434 
435 /*
436  * Filter outgoing MLD report state by group.
437  *
438  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
439  * and node-local addresses. However, kernel and socket consumers
440  * always embed the KAME scope ID in the address provided, so strip it
441  * when performing comparison.
442  * Note: This is not the same as the *multicast* scope.
443  *
444  * Return zero if the given group is one for which MLD reports
445  * should be suppressed, or non-zero if reports should be issued.
446  */
447 static __inline int
448 mld_is_addr_reported(const struct in6_addr *addr)
449 {
450 
451 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
452 
453 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
454 		return (0);
455 
456 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
457 		struct in6_addr tmp = *addr;
458 		in6_clearscope(&tmp);
459 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
460 			return (0);
461 	}
462 
463 	return (1);
464 }
465 
466 /*
467  * Attach MLD when PF_INET6 is attached to an interface.
468  *
469  * SMPng: Normally called with IF_AFDATA_LOCK held.
470  */
471 struct mld_ifsoftc *
472 mld_domifattach(struct ifnet *ifp)
473 {
474 	struct mld_ifsoftc *mli;
475 
476 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
477 	    __func__, ifp, if_name(ifp));
478 
479 	MLD_LOCK();
480 
481 	mli = mli_alloc_locked(ifp);
482 	if (!(ifp->if_flags & IFF_MULTICAST))
483 		mli->mli_flags |= MLIF_SILENT;
484 	if (mld_use_allow)
485 		mli->mli_flags |= MLIF_USEALLOW;
486 
487 	MLD_UNLOCK();
488 
489 	return (mli);
490 }
491 
492 /*
493  * VIMAGE: assume curvnet set by caller.
494  */
495 static struct mld_ifsoftc *
496 mli_alloc_locked(/*const*/ struct ifnet *ifp)
497 {
498 	struct mld_ifsoftc *mli;
499 
500 	MLD_LOCK_ASSERT();
501 
502 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
503 	if (mli == NULL)
504 		goto out;
505 
506 	mli->mli_ifp = ifp;
507 	mli->mli_version = MLD_VERSION_2;
508 	mli->mli_flags = 0;
509 	mli->mli_rv = MLD_RV_INIT;
510 	mli->mli_qi = MLD_QI_INIT;
511 	mli->mli_qri = MLD_QRI_INIT;
512 	mli->mli_uri = MLD_URI_INIT;
513 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
514 
515 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
516 
517 	CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
518 	     ifp, if_name(ifp));
519 
520 out:
521 	return (mli);
522 }
523 
524 /*
525  * Hook for ifdetach.
526  *
527  * NOTE: Some finalization tasks need to run before the protocol domain
528  * is detached, but also before the link layer does its cleanup.
529  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
530  *
531  * SMPng: Caller must hold IN6_MULTI_LOCK().
532  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
533  * XXX This routine is also bitten by unlocked ifma_protospec access.
534  */
535 void
536 mld_ifdetach(struct ifnet *ifp)
537 {
538 	struct mld_ifsoftc	*mli;
539 	struct ifmultiaddr	*ifma, *next;
540 	struct in6_multi	*inm;
541 	struct in6_multi_head inmh;
542 
543 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
544 	    if_name(ifp));
545 
546 	SLIST_INIT(&inmh);
547 	IN6_MULTI_LIST_LOCK_ASSERT();
548 	MLD_LOCK();
549 
550 	mli = MLD_IFINFO(ifp);
551 	if (mli->mli_version == MLD_VERSION_2) {
552 		IF_ADDR_WLOCK(ifp);
553 	restart:
554 		CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
555 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
556 			    ifma->ifma_protospec == NULL)
557 				continue;
558 			inm = (struct in6_multi *)ifma->ifma_protospec;
559 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
560 				in6m_rele_locked(&inmh, inm);
561 				ifma->ifma_protospec = NULL;
562 			}
563 			in6m_clear_recorded(inm);
564 			if (__predict_false(ifma6_restart)) {
565 				ifma6_restart = false;
566 				goto restart;
567 			}
568 		}
569 		IF_ADDR_WUNLOCK(ifp);
570 	}
571 
572 	MLD_UNLOCK();
573 	in6m_release_list_deferred(&inmh);
574 }
575 
576 /*
577  * Hook for domifdetach.
578  * Runs after link-layer cleanup; free MLD state.
579  *
580  * SMPng: Normally called with IF_AFDATA_LOCK held.
581  */
582 void
583 mld_domifdetach(struct ifnet *ifp)
584 {
585 
586 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
587 	    __func__, ifp, if_name(ifp));
588 
589 	MLD_LOCK();
590 	mli_delete_locked(ifp);
591 	MLD_UNLOCK();
592 }
593 
594 static void
595 mli_delete_locked(const struct ifnet *ifp)
596 {
597 	struct mld_ifsoftc *mli, *tmli;
598 
599 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
600 	    __func__, ifp, if_name(ifp));
601 
602 	MLD_LOCK_ASSERT();
603 
604 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
605 		if (mli->mli_ifp == ifp) {
606 			/*
607 			 * Free deferred General Query responses.
608 			 */
609 			mbufq_drain(&mli->mli_gq);
610 
611 			LIST_REMOVE(mli, mli_link);
612 
613 			free(mli, M_MLD);
614 			return;
615 		}
616 	}
617 }
618 
619 /*
620  * Process a received MLDv1 general or address-specific query.
621  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
622  *
623  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
624  * mld_addr. This is OK as we own the mbuf chain.
625  */
626 static int
627 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
628     /*const*/ struct mld_hdr *mld)
629 {
630 	struct ifmultiaddr	*ifma;
631 	struct mld_ifsoftc	*mli;
632 	struct in6_multi	*inm;
633 	int			 is_general_query;
634 	uint16_t		 timer;
635 #ifdef KTR
636 	char			 ip6tbuf[INET6_ADDRSTRLEN];
637 #endif
638 
639 	is_general_query = 0;
640 
641 	if (!mld_v1enable) {
642 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
643 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
644 		    ifp, if_name(ifp));
645 		return (0);
646 	}
647 
648 	/*
649 	 * RFC3810 Section 6.2: MLD queries must originate from
650 	 * a router's link-local address.
651 	 */
652 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
653 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
654 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
655 		    ifp, if_name(ifp));
656 		return (0);
657 	}
658 
659 	/*
660 	 * Do address field validation upfront before we accept
661 	 * the query.
662 	 */
663 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
664 		/*
665 		 * MLDv1 General Query.
666 		 * If this was not sent to the all-nodes group, ignore it.
667 		 */
668 		struct in6_addr		 dst;
669 
670 		dst = ip6->ip6_dst;
671 		in6_clearscope(&dst);
672 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
673 			return (EINVAL);
674 		is_general_query = 1;
675 	} else {
676 		/*
677 		 * Embed scope ID of receiving interface in MLD query for
678 		 * lookup whilst we don't hold other locks.
679 		 */
680 		in6_setscope(&mld->mld_addr, ifp, NULL);
681 	}
682 
683 	IN6_MULTI_LIST_LOCK();
684 	MLD_LOCK();
685 
686 	/*
687 	 * Switch to MLDv1 host compatibility mode.
688 	 */
689 	mli = MLD_IFINFO(ifp);
690 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
691 	mld_set_version(mli, MLD_VERSION_1);
692 
693 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
694 	if (timer == 0)
695 		timer = 1;
696 
697 	IF_ADDR_RLOCK(ifp);
698 	if (is_general_query) {
699 		/*
700 		 * For each reporting group joined on this
701 		 * interface, kick the report timer.
702 		 */
703 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
704 			 ifp, if_name(ifp));
705 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
706 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
707 			    ifma->ifma_protospec == NULL)
708 				continue;
709 			inm = (struct in6_multi *)ifma->ifma_protospec;
710 			mld_v1_update_group(inm, timer);
711 		}
712 	} else {
713 		/*
714 		 * MLDv1 Group-Specific Query.
715 		 * If this is a group-specific MLDv1 query, we need only
716 		 * look up the single group to process it.
717 		 */
718 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
719 		if (inm != NULL) {
720 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
721 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
722 			    ifp, if_name(ifp));
723 			mld_v1_update_group(inm, timer);
724 		}
725 		/* XXX Clear embedded scope ID as userland won't expect it. */
726 		in6_clearscope(&mld->mld_addr);
727 	}
728 
729 	IF_ADDR_RUNLOCK(ifp);
730 	MLD_UNLOCK();
731 	IN6_MULTI_LIST_UNLOCK();
732 
733 	return (0);
734 }
735 
736 /*
737  * Update the report timer on a group in response to an MLDv1 query.
738  *
739  * If we are becoming the reporting member for this group, start the timer.
740  * If we already are the reporting member for this group, and timer is
741  * below the threshold, reset it.
742  *
743  * We may be updating the group for the first time since we switched
744  * to MLDv2. If we are, then we must clear any recorded source lists,
745  * and transition to REPORTING state; the group timer is overloaded
746  * for group and group-source query responses.
747  *
748  * Unlike MLDv2, the delay per group should be jittered
749  * to avoid bursts of MLDv1 reports.
750  */
751 static void
752 mld_v1_update_group(struct in6_multi *inm, const int timer)
753 {
754 #ifdef KTR
755 	char			 ip6tbuf[INET6_ADDRSTRLEN];
756 #endif
757 
758 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
759 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
760 	    if_name(inm->in6m_ifp), timer);
761 
762 	IN6_MULTI_LIST_LOCK_ASSERT();
763 
764 	switch (inm->in6m_state) {
765 	case MLD_NOT_MEMBER:
766 	case MLD_SILENT_MEMBER:
767 		break;
768 	case MLD_REPORTING_MEMBER:
769 		if (inm->in6m_timer != 0 &&
770 		    inm->in6m_timer <= timer) {
771 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
772 			    "skipping.", __func__);
773 			break;
774 		}
775 		/* FALLTHROUGH */
776 	case MLD_SG_QUERY_PENDING_MEMBER:
777 	case MLD_G_QUERY_PENDING_MEMBER:
778 	case MLD_IDLE_MEMBER:
779 	case MLD_LAZY_MEMBER:
780 	case MLD_AWAKENING_MEMBER:
781 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
782 		inm->in6m_state = MLD_REPORTING_MEMBER;
783 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
784 		V_current_state_timers_running6 = 1;
785 		break;
786 	case MLD_SLEEPING_MEMBER:
787 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
788 		inm->in6m_state = MLD_AWAKENING_MEMBER;
789 		break;
790 	case MLD_LEAVING_MEMBER:
791 		break;
792 	}
793 }
794 
795 /*
796  * Process a received MLDv2 general, group-specific or
797  * group-and-source-specific query.
798  *
799  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
800  *
801  * Return 0 if successful, otherwise an appropriate error code is returned.
802  */
803 static int
804 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
805     struct mbuf *m, const int off, const int icmp6len)
806 {
807 	struct mld_ifsoftc	*mli;
808 	struct mldv2_query	*mld;
809 	struct in6_multi	*inm;
810 	uint32_t		 maxdelay, nsrc, qqi;
811 	int			 is_general_query;
812 	uint16_t		 timer;
813 	uint8_t			 qrv;
814 #ifdef KTR
815 	char			 ip6tbuf[INET6_ADDRSTRLEN];
816 #endif
817 
818 	is_general_query = 0;
819 
820 	/*
821 	 * RFC3810 Section 6.2: MLD queries must originate from
822 	 * a router's link-local address.
823 	 */
824 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
825 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
826 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
827 		    ifp, if_name(ifp));
828 		return (0);
829 	}
830 
831 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
832 
833 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
834 
835 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
836 	if (maxdelay >= 32768) {
837 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
838 			   (MLD_MRC_EXP(maxdelay) + 3);
839 	}
840 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
841 	if (timer == 0)
842 		timer = 1;
843 
844 	qrv = MLD_QRV(mld->mld_misc);
845 	if (qrv < 2) {
846 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
847 		    qrv, MLD_RV_INIT);
848 		qrv = MLD_RV_INIT;
849 	}
850 
851 	qqi = mld->mld_qqi;
852 	if (qqi >= 128) {
853 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
854 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
855 	}
856 
857 	nsrc = ntohs(mld->mld_numsrc);
858 	if (nsrc > MLD_MAX_GS_SOURCES)
859 		return (EMSGSIZE);
860 	if (icmp6len < sizeof(struct mldv2_query) +
861 	    (nsrc * sizeof(struct in6_addr)))
862 		return (EMSGSIZE);
863 
864 	/*
865 	 * Do further input validation upfront to avoid resetting timers
866 	 * should we need to discard this query.
867 	 */
868 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
869 		/*
870 		 * A general query with a source list has undefined
871 		 * behaviour; discard it.
872 		 */
873 		if (nsrc > 0)
874 			return (EINVAL);
875 		is_general_query = 1;
876 	} else {
877 		/*
878 		 * Embed scope ID of receiving interface in MLD query for
879 		 * lookup whilst we don't hold other locks (due to KAME
880 		 * locking lameness). We own this mbuf chain just now.
881 		 */
882 		in6_setscope(&mld->mld_addr, ifp, NULL);
883 	}
884 
885 	IN6_MULTI_LIST_LOCK();
886 	MLD_LOCK();
887 
888 	mli = MLD_IFINFO(ifp);
889 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
890 
891 	/*
892 	 * Discard the v2 query if we're in Compatibility Mode.
893 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
894 	 * until the Old Version Querier Present timer expires.
895 	 */
896 	if (mli->mli_version != MLD_VERSION_2)
897 		goto out_locked;
898 
899 	mld_set_version(mli, MLD_VERSION_2);
900 	mli->mli_rv = qrv;
901 	mli->mli_qi = qqi;
902 	mli->mli_qri = maxdelay;
903 
904 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
905 	    maxdelay);
906 
907 	if (is_general_query) {
908 		/*
909 		 * MLDv2 General Query.
910 		 *
911 		 * Schedule a current-state report on this ifp for
912 		 * all groups, possibly containing source lists.
913 		 *
914 		 * If there is a pending General Query response
915 		 * scheduled earlier than the selected delay, do
916 		 * not schedule any other reports.
917 		 * Otherwise, reset the interface timer.
918 		 */
919 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
920 		    ifp, if_name(ifp));
921 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
922 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
923 			V_interface_timers_running6 = 1;
924 		}
925 	} else {
926 		/*
927 		 * MLDv2 Group-specific or Group-and-source-specific Query.
928 		 *
929 		 * Group-source-specific queries are throttled on
930 		 * a per-group basis to defeat denial-of-service attempts.
931 		 * Queries for groups we are not a member of on this
932 		 * link are simply ignored.
933 		 */
934 		IF_ADDR_RLOCK(ifp);
935 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
936 		if (inm == NULL) {
937 			IF_ADDR_RUNLOCK(ifp);
938 			goto out_locked;
939 		}
940 		if (nsrc > 0) {
941 			if (!ratecheck(&inm->in6m_lastgsrtv,
942 			    &V_mld_gsrdelay)) {
943 				CTR1(KTR_MLD, "%s: GS query throttled.",
944 				    __func__);
945 				IF_ADDR_RUNLOCK(ifp);
946 				goto out_locked;
947 			}
948 		}
949 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
950 		     ifp, if_name(ifp));
951 		/*
952 		 * If there is a pending General Query response
953 		 * scheduled sooner than the selected delay, no
954 		 * further report need be scheduled.
955 		 * Otherwise, prepare to respond to the
956 		 * group-specific or group-and-source query.
957 		 */
958 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
959 			mld_v2_process_group_query(inm, mli, timer, m, off);
960 
961 		/* XXX Clear embedded scope ID as userland won't expect it. */
962 		in6_clearscope(&mld->mld_addr);
963 		IF_ADDR_RUNLOCK(ifp);
964 	}
965 
966 out_locked:
967 	MLD_UNLOCK();
968 	IN6_MULTI_LIST_UNLOCK();
969 
970 	return (0);
971 }
972 
973 /*
974  * Process a received MLDv2 group-specific or group-and-source-specific
975  * query.
976  * Return <0 if any error occurred. Currently this is ignored.
977  */
978 static int
979 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
980     int timer, struct mbuf *m0, const int off)
981 {
982 	struct mldv2_query	*mld;
983 	int			 retval;
984 	uint16_t		 nsrc;
985 
986 	IN6_MULTI_LIST_LOCK_ASSERT();
987 	MLD_LOCK_ASSERT();
988 
989 	retval = 0;
990 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
991 
992 	switch (inm->in6m_state) {
993 	case MLD_NOT_MEMBER:
994 	case MLD_SILENT_MEMBER:
995 	case MLD_SLEEPING_MEMBER:
996 	case MLD_LAZY_MEMBER:
997 	case MLD_AWAKENING_MEMBER:
998 	case MLD_IDLE_MEMBER:
999 	case MLD_LEAVING_MEMBER:
1000 		return (retval);
1001 		break;
1002 	case MLD_REPORTING_MEMBER:
1003 	case MLD_G_QUERY_PENDING_MEMBER:
1004 	case MLD_SG_QUERY_PENDING_MEMBER:
1005 		break;
1006 	}
1007 
1008 	nsrc = ntohs(mld->mld_numsrc);
1009 
1010 	/*
1011 	 * Deal with group-specific queries upfront.
1012 	 * If any group query is already pending, purge any recorded
1013 	 * source-list state if it exists, and schedule a query response
1014 	 * for this group-specific query.
1015 	 */
1016 	if (nsrc == 0) {
1017 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1018 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1019 			in6m_clear_recorded(inm);
1020 			timer = min(inm->in6m_timer, timer);
1021 		}
1022 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1023 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1024 		V_current_state_timers_running6 = 1;
1025 		return (retval);
1026 	}
1027 
1028 	/*
1029 	 * Deal with the case where a group-and-source-specific query has
1030 	 * been received but a group-specific query is already pending.
1031 	 */
1032 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1033 		timer = min(inm->in6m_timer, timer);
1034 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1035 		V_current_state_timers_running6 = 1;
1036 		return (retval);
1037 	}
1038 
1039 	/*
1040 	 * Finally, deal with the case where a group-and-source-specific
1041 	 * query has been received, where a response to a previous g-s-r
1042 	 * query exists, or none exists.
1043 	 * In this case, we need to parse the source-list which the Querier
1044 	 * has provided us with and check if we have any source list filter
1045 	 * entries at T1 for these sources. If we do not, there is no need
1046 	 * schedule a report and the query may be dropped.
1047 	 * If we do, we must record them and schedule a current-state
1048 	 * report for those sources.
1049 	 */
1050 	if (inm->in6m_nsrc > 0) {
1051 		struct mbuf		*m;
1052 		uint8_t			*sp;
1053 		int			 i, nrecorded;
1054 		int			 soff;
1055 
1056 		m = m0;
1057 		soff = off + sizeof(struct mldv2_query);
1058 		nrecorded = 0;
1059 		for (i = 0; i < nsrc; i++) {
1060 			sp = mtod(m, uint8_t *) + soff;
1061 			retval = in6m_record_source(inm,
1062 			    (const struct in6_addr *)sp);
1063 			if (retval < 0)
1064 				break;
1065 			nrecorded += retval;
1066 			soff += sizeof(struct in6_addr);
1067 			if (soff >= m->m_len) {
1068 				soff = soff - m->m_len;
1069 				m = m->m_next;
1070 				if (m == NULL)
1071 					break;
1072 			}
1073 		}
1074 		if (nrecorded > 0) {
1075 			CTR1(KTR_MLD,
1076 			    "%s: schedule response to SG query", __func__);
1077 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1078 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1079 			V_current_state_timers_running6 = 1;
1080 		}
1081 	}
1082 
1083 	return (retval);
1084 }
1085 
1086 /*
1087  * Process a received MLDv1 host membership report.
1088  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1089  *
1090  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1091  * mld_addr. This is OK as we own the mbuf chain.
1092  */
1093 static int
1094 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1095     /*const*/ struct mld_hdr *mld)
1096 {
1097 	struct in6_addr		 src, dst;
1098 	struct in6_ifaddr	*ia;
1099 	struct in6_multi	*inm;
1100 #ifdef KTR
1101 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1102 #endif
1103 
1104 	if (!mld_v1enable) {
1105 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1106 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1107 		    ifp, if_name(ifp));
1108 		return (0);
1109 	}
1110 
1111 	if (ifp->if_flags & IFF_LOOPBACK)
1112 		return (0);
1113 
1114 	/*
1115 	 * MLDv1 reports must originate from a host's link-local address,
1116 	 * or the unspecified address (when booting).
1117 	 */
1118 	src = ip6->ip6_src;
1119 	in6_clearscope(&src);
1120 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1121 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1122 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1123 		    ifp, if_name(ifp));
1124 		return (EINVAL);
1125 	}
1126 
1127 	/*
1128 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1129 	 * group, and must be directed to the group itself.
1130 	 */
1131 	dst = ip6->ip6_dst;
1132 	in6_clearscope(&dst);
1133 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1134 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1135 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1136 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1137 		    ifp, if_name(ifp));
1138 		return (EINVAL);
1139 	}
1140 
1141 	/*
1142 	 * Make sure we don't hear our own membership report, as fast
1143 	 * leave requires knowing that we are the only member of a
1144 	 * group. Assume we used the link-local address if available,
1145 	 * otherwise look for ::.
1146 	 *
1147 	 * XXX Note that scope ID comparison is needed for the address
1148 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1149 	 * performed for the on-wire address.
1150 	 */
1151 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1152 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1153 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1154 		if (ia != NULL)
1155 			ifa_free(&ia->ia_ifa);
1156 		return (0);
1157 	}
1158 	if (ia != NULL)
1159 		ifa_free(&ia->ia_ifa);
1160 
1161 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1162 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1163 
1164 	/*
1165 	 * Embed scope ID of receiving interface in MLD query for lookup
1166 	 * whilst we don't hold other locks (due to KAME locking lameness).
1167 	 */
1168 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1169 		in6_setscope(&mld->mld_addr, ifp, NULL);
1170 
1171 	IN6_MULTI_LIST_LOCK();
1172 	MLD_LOCK();
1173 	IF_ADDR_RLOCK(ifp);
1174 
1175 	/*
1176 	 * MLDv1 report suppression.
1177 	 * If we are a member of this group, and our membership should be
1178 	 * reported, and our group timer is pending or about to be reset,
1179 	 * stop our group timer by transitioning to the 'lazy' state.
1180 	 */
1181 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1182 	if (inm != NULL) {
1183 		struct mld_ifsoftc *mli;
1184 
1185 		mli = inm->in6m_mli;
1186 		KASSERT(mli != NULL,
1187 		    ("%s: no mli for ifp %p", __func__, ifp));
1188 
1189 		/*
1190 		 * If we are in MLDv2 host mode, do not allow the
1191 		 * other host's MLDv1 report to suppress our reports.
1192 		 */
1193 		if (mli->mli_version == MLD_VERSION_2)
1194 			goto out_locked;
1195 
1196 		inm->in6m_timer = 0;
1197 
1198 		switch (inm->in6m_state) {
1199 		case MLD_NOT_MEMBER:
1200 		case MLD_SILENT_MEMBER:
1201 		case MLD_SLEEPING_MEMBER:
1202 			break;
1203 		case MLD_REPORTING_MEMBER:
1204 		case MLD_IDLE_MEMBER:
1205 		case MLD_AWAKENING_MEMBER:
1206 			CTR3(KTR_MLD,
1207 			    "report suppressed for %s on ifp %p(%s)",
1208 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1209 			    ifp, if_name(ifp));
1210 		case MLD_LAZY_MEMBER:
1211 			inm->in6m_state = MLD_LAZY_MEMBER;
1212 			break;
1213 		case MLD_G_QUERY_PENDING_MEMBER:
1214 		case MLD_SG_QUERY_PENDING_MEMBER:
1215 		case MLD_LEAVING_MEMBER:
1216 			break;
1217 		}
1218 	}
1219 
1220 out_locked:
1221 	IF_ADDR_RUNLOCK(ifp);
1222 	MLD_UNLOCK();
1223 	IN6_MULTI_LIST_UNLOCK();
1224 
1225 	/* XXX Clear embedded scope ID as userland won't expect it. */
1226 	in6_clearscope(&mld->mld_addr);
1227 
1228 	return (0);
1229 }
1230 
1231 /*
1232  * MLD input path.
1233  *
1234  * Assume query messages which fit in a single ICMPv6 message header
1235  * have been pulled up.
1236  * Assume that userland will want to see the message, even if it
1237  * otherwise fails kernel input validation; do not free it.
1238  * Pullup may however free the mbuf chain m if it fails.
1239  *
1240  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1241  */
1242 int
1243 mld_input(struct mbuf *m, int off, int icmp6len)
1244 {
1245 	struct ifnet	*ifp;
1246 	struct ip6_hdr	*ip6;
1247 	struct mld_hdr	*mld;
1248 	int		 mldlen;
1249 
1250 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1251 
1252 	ifp = m->m_pkthdr.rcvif;
1253 
1254 	ip6 = mtod(m, struct ip6_hdr *);
1255 
1256 	/* Pullup to appropriate size. */
1257 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1258 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1259 	    icmp6len >= sizeof(struct mldv2_query)) {
1260 		mldlen = sizeof(struct mldv2_query);
1261 	} else {
1262 		mldlen = sizeof(struct mld_hdr);
1263 	}
1264 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1265 	if (mld == NULL) {
1266 		ICMP6STAT_INC(icp6s_badlen);
1267 		return (IPPROTO_DONE);
1268 	}
1269 
1270 	/*
1271 	 * Userland needs to see all of this traffic for implementing
1272 	 * the endpoint discovery portion of multicast routing.
1273 	 */
1274 	switch (mld->mld_type) {
1275 	case MLD_LISTENER_QUERY:
1276 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1277 		if (icmp6len == sizeof(struct mld_hdr)) {
1278 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1279 				return (0);
1280 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1281 			if (mld_v2_input_query(ifp, ip6, m, off,
1282 			    icmp6len) != 0)
1283 				return (0);
1284 		}
1285 		break;
1286 	case MLD_LISTENER_REPORT:
1287 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1288 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1289 			return (0);
1290 		break;
1291 	case MLDV2_LISTENER_REPORT:
1292 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1293 		break;
1294 	case MLD_LISTENER_DONE:
1295 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1296 		break;
1297 	default:
1298 		break;
1299 	}
1300 
1301 	return (0);
1302 }
1303 
1304 /*
1305  * Fast timeout handler (global).
1306  * VIMAGE: Timeout handlers are expected to service all vimages.
1307  */
1308 void
1309 mld_fasttimo(void)
1310 {
1311 	VNET_ITERATOR_DECL(vnet_iter);
1312 
1313 	VNET_LIST_RLOCK_NOSLEEP();
1314 	VNET_FOREACH(vnet_iter) {
1315 		CURVNET_SET(vnet_iter);
1316 		mld_fasttimo_vnet();
1317 		CURVNET_RESTORE();
1318 	}
1319 	VNET_LIST_RUNLOCK_NOSLEEP();
1320 }
1321 
1322 /*
1323  * Fast timeout handler (per-vnet).
1324  *
1325  * VIMAGE: Assume caller has set up our curvnet.
1326  */
1327 static void
1328 mld_fasttimo_vnet(void)
1329 {
1330 	struct mbufq		 scq;	/* State-change packets */
1331 	struct mbufq		 qrq;	/* Query response packets */
1332 	struct ifnet		*ifp;
1333 	struct mld_ifsoftc	*mli;
1334 	struct ifmultiaddr	*ifma, *next;
1335 	struct in6_multi	*inm, *tinm;
1336 	struct in6_multi_head inmh;
1337 	int			 uri_fasthz;
1338 
1339 	uri_fasthz = 0;
1340 
1341 	/*
1342 	 * Quick check to see if any work needs to be done, in order to
1343 	 * minimize the overhead of fasttimo processing.
1344 	 * SMPng: XXX Unlocked reads.
1345 	 */
1346 	if (!V_current_state_timers_running6 &&
1347 	    !V_interface_timers_running6 &&
1348 	    !V_state_change_timers_running6)
1349 		return;
1350 
1351 	SLIST_INIT(&inmh);
1352 	IN6_MULTI_LIST_LOCK();
1353 	MLD_LOCK();
1354 
1355 	/*
1356 	 * MLDv2 General Query response timer processing.
1357 	 */
1358 	if (V_interface_timers_running6) {
1359 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1360 
1361 		V_interface_timers_running6 = 0;
1362 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1363 			if (mli->mli_v2_timer == 0) {
1364 				/* Do nothing. */
1365 			} else if (--mli->mli_v2_timer == 0) {
1366 				mld_v2_dispatch_general_query(mli);
1367 			} else {
1368 				V_interface_timers_running6 = 1;
1369 			}
1370 		}
1371 	}
1372 
1373 	if (!V_current_state_timers_running6 &&
1374 	    !V_state_change_timers_running6)
1375 		goto out_locked;
1376 
1377 	V_current_state_timers_running6 = 0;
1378 	V_state_change_timers_running6 = 0;
1379 
1380 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1381 
1382 	/*
1383 	 * MLD host report and state-change timer processing.
1384 	 * Note: Processing a v2 group timer may remove a node.
1385 	 */
1386 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1387 		ifp = mli->mli_ifp;
1388 
1389 		if (mli->mli_version == MLD_VERSION_2) {
1390 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1391 			    PR_FASTHZ);
1392 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1393 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1394 		}
1395 
1396 		IF_ADDR_WLOCK(ifp);
1397 	restart:
1398 		CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1399 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1400 			    ifma->ifma_protospec == NULL)
1401 				continue;
1402 			inm = (struct in6_multi *)ifma->ifma_protospec;
1403 			switch (mli->mli_version) {
1404 			case MLD_VERSION_1:
1405 				mld_v1_process_group_timer(&inmh, inm);
1406 				break;
1407 			case MLD_VERSION_2:
1408 				mld_v2_process_group_timers(&inmh, &qrq,
1409 				    &scq, inm, uri_fasthz);
1410 				break;
1411 			}
1412 			if (__predict_false(ifma6_restart)) {
1413 				ifma6_restart = false;
1414 				goto restart;
1415 			}
1416 		}
1417 		IF_ADDR_WUNLOCK(ifp);
1418 
1419 		switch (mli->mli_version) {
1420 		case MLD_VERSION_1:
1421 			/*
1422 			 * Transmit reports for this lifecycle.  This
1423 			 * is done while not holding IF_ADDR_LOCK
1424 			 * since this can call
1425 			 * in6ifa_ifpforlinklocal() which locks
1426 			 * IF_ADDR_LOCK internally as well as
1427 			 * ip6_output() to transmit a packet.
1428 			 */
1429 			SLIST_FOREACH_SAFE(inm, &inmh, in6m_nrele, tinm) {
1430 				SLIST_REMOVE_HEAD(&inmh,
1431 				    in6m_nrele);
1432 				(void)mld_v1_transmit_report(inm,
1433 				    MLD_LISTENER_REPORT);
1434 			}
1435 			break;
1436 		case MLD_VERSION_2:
1437 			mld_dispatch_queue(&qrq, 0);
1438 			mld_dispatch_queue(&scq, 0);
1439 
1440 			/*
1441 			 * Free the in_multi reference(s) for
1442 			 * this lifecycle.
1443 			 */
1444 			in6m_release_list_deferred(&inmh);
1445 			break;
1446 		}
1447 	}
1448 
1449 out_locked:
1450 	MLD_UNLOCK();
1451 	IN6_MULTI_LIST_UNLOCK();
1452 }
1453 
1454 /*
1455  * Update host report group timer.
1456  * Will update the global pending timer flags.
1457  */
1458 static void
1459 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1460 {
1461 	int report_timer_expired;
1462 
1463 	IN6_MULTI_LIST_LOCK_ASSERT();
1464 	MLD_LOCK_ASSERT();
1465 
1466 	if (inm->in6m_timer == 0) {
1467 		report_timer_expired = 0;
1468 	} else if (--inm->in6m_timer == 0) {
1469 		report_timer_expired = 1;
1470 	} else {
1471 		V_current_state_timers_running6 = 1;
1472 		return;
1473 	}
1474 
1475 	switch (inm->in6m_state) {
1476 	case MLD_NOT_MEMBER:
1477 	case MLD_SILENT_MEMBER:
1478 	case MLD_IDLE_MEMBER:
1479 	case MLD_LAZY_MEMBER:
1480 	case MLD_SLEEPING_MEMBER:
1481 	case MLD_AWAKENING_MEMBER:
1482 		break;
1483 	case MLD_REPORTING_MEMBER:
1484 		if (report_timer_expired) {
1485 			inm->in6m_state = MLD_IDLE_MEMBER;
1486 			in6m_rele_locked(inmh, inm);
1487 		}
1488 		break;
1489 	case MLD_G_QUERY_PENDING_MEMBER:
1490 	case MLD_SG_QUERY_PENDING_MEMBER:
1491 	case MLD_LEAVING_MEMBER:
1492 		break;
1493 	}
1494 }
1495 
1496 /*
1497  * Update a group's timers for MLDv2.
1498  * Will update the global pending timer flags.
1499  * Note: Unlocked read from mli.
1500  */
1501 static void
1502 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1503     struct mbufq *qrq, struct mbufq *scq,
1504     struct in6_multi *inm, const int uri_fasthz)
1505 {
1506 	int query_response_timer_expired;
1507 	int state_change_retransmit_timer_expired;
1508 #ifdef KTR
1509 	char ip6tbuf[INET6_ADDRSTRLEN];
1510 #endif
1511 
1512 	IN6_MULTI_LIST_LOCK_ASSERT();
1513 	MLD_LOCK_ASSERT();
1514 
1515 	query_response_timer_expired = 0;
1516 	state_change_retransmit_timer_expired = 0;
1517 
1518 	/*
1519 	 * During a transition from compatibility mode back to MLDv2,
1520 	 * a group record in REPORTING state may still have its group
1521 	 * timer active. This is a no-op in this function; it is easier
1522 	 * to deal with it here than to complicate the slow-timeout path.
1523 	 */
1524 	if (inm->in6m_timer == 0) {
1525 		query_response_timer_expired = 0;
1526 	} else if (--inm->in6m_timer == 0) {
1527 		query_response_timer_expired = 1;
1528 	} else {
1529 		V_current_state_timers_running6 = 1;
1530 	}
1531 
1532 	if (inm->in6m_sctimer == 0) {
1533 		state_change_retransmit_timer_expired = 0;
1534 	} else if (--inm->in6m_sctimer == 0) {
1535 		state_change_retransmit_timer_expired = 1;
1536 	} else {
1537 		V_state_change_timers_running6 = 1;
1538 	}
1539 
1540 	/* We are in fasttimo, so be quick about it. */
1541 	if (!state_change_retransmit_timer_expired &&
1542 	    !query_response_timer_expired)
1543 		return;
1544 
1545 	switch (inm->in6m_state) {
1546 	case MLD_NOT_MEMBER:
1547 	case MLD_SILENT_MEMBER:
1548 	case MLD_SLEEPING_MEMBER:
1549 	case MLD_LAZY_MEMBER:
1550 	case MLD_AWAKENING_MEMBER:
1551 	case MLD_IDLE_MEMBER:
1552 		break;
1553 	case MLD_G_QUERY_PENDING_MEMBER:
1554 	case MLD_SG_QUERY_PENDING_MEMBER:
1555 		/*
1556 		 * Respond to a previously pending Group-Specific
1557 		 * or Group-and-Source-Specific query by enqueueing
1558 		 * the appropriate Current-State report for
1559 		 * immediate transmission.
1560 		 */
1561 		if (query_response_timer_expired) {
1562 			int retval;
1563 
1564 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1565 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1566 			    0);
1567 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1568 			    __func__, retval);
1569 			inm->in6m_state = MLD_REPORTING_MEMBER;
1570 			in6m_clear_recorded(inm);
1571 		}
1572 		/* FALLTHROUGH */
1573 	case MLD_REPORTING_MEMBER:
1574 	case MLD_LEAVING_MEMBER:
1575 		if (state_change_retransmit_timer_expired) {
1576 			/*
1577 			 * State-change retransmission timer fired.
1578 			 * If there are any further pending retransmissions,
1579 			 * set the global pending state-change flag, and
1580 			 * reset the timer.
1581 			 */
1582 			if (--inm->in6m_scrv > 0) {
1583 				inm->in6m_sctimer = uri_fasthz;
1584 				V_state_change_timers_running6 = 1;
1585 			}
1586 			/*
1587 			 * Retransmit the previously computed state-change
1588 			 * report. If there are no further pending
1589 			 * retransmissions, the mbuf queue will be consumed.
1590 			 * Update T0 state to T1 as we have now sent
1591 			 * a state-change.
1592 			 */
1593 			(void)mld_v2_merge_state_changes(inm, scq);
1594 
1595 			in6m_commit(inm);
1596 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1597 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1598 			    if_name(inm->in6m_ifp));
1599 
1600 			/*
1601 			 * If we are leaving the group for good, make sure
1602 			 * we release MLD's reference to it.
1603 			 * This release must be deferred using a SLIST,
1604 			 * as we are called from a loop which traverses
1605 			 * the in_ifmultiaddr TAILQ.
1606 			 */
1607 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1608 			    inm->in6m_scrv == 0) {
1609 				inm->in6m_state = MLD_NOT_MEMBER;
1610 				in6m_rele_locked(inmh, inm);
1611 			}
1612 		}
1613 		break;
1614 	}
1615 }
1616 
1617 /*
1618  * Switch to a different version on the given interface,
1619  * as per Section 9.12.
1620  */
1621 static void
1622 mld_set_version(struct mld_ifsoftc *mli, const int version)
1623 {
1624 	int old_version_timer;
1625 
1626 	MLD_LOCK_ASSERT();
1627 
1628 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1629 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1630 
1631 	if (version == MLD_VERSION_1) {
1632 		/*
1633 		 * Compute the "Older Version Querier Present" timer as per
1634 		 * Section 9.12.
1635 		 */
1636 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1637 		old_version_timer *= PR_SLOWHZ;
1638 		mli->mli_v1_timer = old_version_timer;
1639 	}
1640 
1641 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1642 		mli->mli_version = MLD_VERSION_1;
1643 		mld_v2_cancel_link_timers(mli);
1644 	}
1645 }
1646 
1647 /*
1648  * Cancel pending MLDv2 timers for the given link and all groups
1649  * joined on it; state-change, general-query, and group-query timers.
1650  */
1651 static void
1652 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1653 {
1654 	struct ifmultiaddr	*ifma, *next;
1655 	struct ifnet		*ifp;
1656 	struct in6_multi	*inm;
1657 	struct in6_multi_head inmh;
1658 
1659 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1660 	    mli->mli_ifp, if_name(mli->mli_ifp));
1661 
1662 	SLIST_INIT(&inmh);
1663 	IN6_MULTI_LIST_LOCK_ASSERT();
1664 	MLD_LOCK_ASSERT();
1665 
1666 	/*
1667 	 * Fast-track this potentially expensive operation
1668 	 * by checking all the global 'timer pending' flags.
1669 	 */
1670 	if (!V_interface_timers_running6 &&
1671 	    !V_state_change_timers_running6 &&
1672 	    !V_current_state_timers_running6)
1673 		return;
1674 
1675 	mli->mli_v2_timer = 0;
1676 
1677 	ifp = mli->mli_ifp;
1678 
1679 	IF_ADDR_WLOCK(ifp);
1680  restart:
1681 	CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1682 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
1683 		    ifma->ifma_protospec == NULL)
1684 			continue;
1685 		inm = (struct in6_multi *)ifma->ifma_protospec;
1686 		switch (inm->in6m_state) {
1687 		case MLD_NOT_MEMBER:
1688 		case MLD_SILENT_MEMBER:
1689 		case MLD_IDLE_MEMBER:
1690 		case MLD_LAZY_MEMBER:
1691 		case MLD_SLEEPING_MEMBER:
1692 		case MLD_AWAKENING_MEMBER:
1693 			break;
1694 		case MLD_LEAVING_MEMBER:
1695 			/*
1696 			 * If we are leaving the group and switching
1697 			 * version, we need to release the final
1698 			 * reference held for issuing the INCLUDE {}.
1699 			 */
1700 			in6m_rele_locked(&inmh, inm);
1701 			ifma->ifma_protospec = NULL;
1702 			/* FALLTHROUGH */
1703 		case MLD_G_QUERY_PENDING_MEMBER:
1704 		case MLD_SG_QUERY_PENDING_MEMBER:
1705 			in6m_clear_recorded(inm);
1706 			/* FALLTHROUGH */
1707 		case MLD_REPORTING_MEMBER:
1708 			inm->in6m_sctimer = 0;
1709 			inm->in6m_timer = 0;
1710 			inm->in6m_state = MLD_REPORTING_MEMBER;
1711 			/*
1712 			 * Free any pending MLDv2 state-change records.
1713 			 */
1714 			mbufq_drain(&inm->in6m_scq);
1715 			break;
1716 		}
1717 		if (__predict_false(ifma6_restart)) {
1718 			ifma6_restart = false;
1719 			goto restart;
1720 		}
1721 	}
1722 	IF_ADDR_WUNLOCK(ifp);
1723 	in6m_release_list_deferred(&inmh);
1724 }
1725 
1726 /*
1727  * Global slowtimo handler.
1728  * VIMAGE: Timeout handlers are expected to service all vimages.
1729  */
1730 void
1731 mld_slowtimo(void)
1732 {
1733 	VNET_ITERATOR_DECL(vnet_iter);
1734 
1735 	VNET_LIST_RLOCK_NOSLEEP();
1736 	VNET_FOREACH(vnet_iter) {
1737 		CURVNET_SET(vnet_iter);
1738 		mld_slowtimo_vnet();
1739 		CURVNET_RESTORE();
1740 	}
1741 	VNET_LIST_RUNLOCK_NOSLEEP();
1742 }
1743 
1744 /*
1745  * Per-vnet slowtimo handler.
1746  */
1747 static void
1748 mld_slowtimo_vnet(void)
1749 {
1750 	struct mld_ifsoftc *mli;
1751 
1752 	MLD_LOCK();
1753 
1754 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1755 		mld_v1_process_querier_timers(mli);
1756 	}
1757 
1758 	MLD_UNLOCK();
1759 }
1760 
1761 /*
1762  * Update the Older Version Querier Present timers for a link.
1763  * See Section 9.12 of RFC 3810.
1764  */
1765 static void
1766 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1767 {
1768 
1769 	MLD_LOCK_ASSERT();
1770 
1771 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1772 		/*
1773 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1774 		 */
1775 		CTR5(KTR_MLD,
1776 		    "%s: transition from v%d -> v%d on %p(%s)",
1777 		    __func__, mli->mli_version, MLD_VERSION_2,
1778 		    mli->mli_ifp, if_name(mli->mli_ifp));
1779 		mli->mli_version = MLD_VERSION_2;
1780 	}
1781 }
1782 
1783 /*
1784  * Transmit an MLDv1 report immediately.
1785  */
1786 static int
1787 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1788 {
1789 	struct ifnet		*ifp;
1790 	struct in6_ifaddr	*ia;
1791 	struct ip6_hdr		*ip6;
1792 	struct mbuf		*mh, *md;
1793 	struct mld_hdr		*mld;
1794 
1795 	IN6_MULTI_LIST_LOCK_ASSERT();
1796 	MLD_LOCK_ASSERT();
1797 
1798 	ifp = in6m->in6m_ifp;
1799 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1800 	/* ia may be NULL if link-local address is tentative. */
1801 
1802 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1803 	if (mh == NULL) {
1804 		if (ia != NULL)
1805 			ifa_free(&ia->ia_ifa);
1806 		return (ENOMEM);
1807 	}
1808 	md = m_get(M_NOWAIT, MT_DATA);
1809 	if (md == NULL) {
1810 		m_free(mh);
1811 		if (ia != NULL)
1812 			ifa_free(&ia->ia_ifa);
1813 		return (ENOMEM);
1814 	}
1815 	mh->m_next = md;
1816 
1817 	/*
1818 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1819 	 * that ether_output() does not need to allocate another mbuf
1820 	 * for the header in the most common case.
1821 	 */
1822 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1823 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1824 	mh->m_len = sizeof(struct ip6_hdr);
1825 
1826 	ip6 = mtod(mh, struct ip6_hdr *);
1827 	ip6->ip6_flow = 0;
1828 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1829 	ip6->ip6_vfc |= IPV6_VERSION;
1830 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1831 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1832 	ip6->ip6_dst = in6m->in6m_addr;
1833 
1834 	md->m_len = sizeof(struct mld_hdr);
1835 	mld = mtod(md, struct mld_hdr *);
1836 	mld->mld_type = type;
1837 	mld->mld_code = 0;
1838 	mld->mld_cksum = 0;
1839 	mld->mld_maxdelay = 0;
1840 	mld->mld_reserved = 0;
1841 	mld->mld_addr = in6m->in6m_addr;
1842 	in6_clearscope(&mld->mld_addr);
1843 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1844 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1845 
1846 	mld_save_context(mh, ifp);
1847 	mh->m_flags |= M_MLDV1;
1848 
1849 	mld_dispatch_packet(mh);
1850 
1851 	if (ia != NULL)
1852 		ifa_free(&ia->ia_ifa);
1853 	return (0);
1854 }
1855 
1856 /*
1857  * Process a state change from the upper layer for the given IPv6 group.
1858  *
1859  * Each socket holds a reference on the in_multi in its own ip_moptions.
1860  * The socket layer will have made the necessary updates to.the group
1861  * state, it is now up to MLD to issue a state change report if there
1862  * has been any change between T0 (when the last state-change was issued)
1863  * and T1 (now).
1864  *
1865  * We use the MLDv2 state machine at group level. The MLd module
1866  * however makes the decision as to which MLD protocol version to speak.
1867  * A state change *from* INCLUDE {} always means an initial join.
1868  * A state change *to* INCLUDE {} always means a final leave.
1869  *
1870  * If delay is non-zero, and the state change is an initial multicast
1871  * join, the state change report will be delayed by 'delay' ticks
1872  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1873  * the initial MLDv2 state change report will be delayed by whichever
1874  * is sooner, a pending state-change timer or delay itself.
1875  *
1876  * VIMAGE: curvnet should have been set by caller, as this routine
1877  * is called from the socket option handlers.
1878  */
1879 int
1880 mld_change_state(struct in6_multi *inm, const int delay)
1881 {
1882 	struct mld_ifsoftc *mli;
1883 	struct ifnet *ifp;
1884 	int error;
1885 
1886 	IN6_MULTI_LIST_LOCK_ASSERT();
1887 
1888 	error = 0;
1889 
1890 	/*
1891 	 * Try to detect if the upper layer just asked us to change state
1892 	 * for an interface which has now gone away.
1893 	 */
1894 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1895 	ifp = inm->in6m_ifma->ifma_ifp;
1896 	if (ifp != NULL) {
1897 		/*
1898 		 * Sanity check that netinet6's notion of ifp is the
1899 		 * same as net's.
1900 		 */
1901 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1902 	}
1903 
1904 	MLD_LOCK();
1905 
1906 	mli = MLD_IFINFO(ifp);
1907 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1908 
1909 	/*
1910 	 * If we detect a state transition to or from MCAST_UNDEFINED
1911 	 * for this group, then we are starting or finishing an MLD
1912 	 * life cycle for this group.
1913 	 */
1914 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1915 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1916 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1917 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1918 			CTR1(KTR_MLD, "%s: initial join", __func__);
1919 			error = mld_initial_join(inm, mli, delay);
1920 			goto out_locked;
1921 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1922 			CTR1(KTR_MLD, "%s: final leave", __func__);
1923 			mld_final_leave(inm, mli);
1924 			goto out_locked;
1925 		}
1926 	} else {
1927 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1928 	}
1929 
1930 	error = mld_handle_state_change(inm, mli);
1931 
1932 out_locked:
1933 	MLD_UNLOCK();
1934 	return (error);
1935 }
1936 
1937 /*
1938  * Perform the initial join for an MLD group.
1939  *
1940  * When joining a group:
1941  *  If the group should have its MLD traffic suppressed, do nothing.
1942  *  MLDv1 starts sending MLDv1 host membership reports.
1943  *  MLDv2 will schedule an MLDv2 state-change report containing the
1944  *  initial state of the membership.
1945  *
1946  * If the delay argument is non-zero, then we must delay sending the
1947  * initial state change for delay ticks (in units of PR_FASTHZ).
1948  */
1949 static int
1950 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1951     const int delay)
1952 {
1953 	struct ifnet		*ifp;
1954 	struct mbufq		*mq;
1955 	int			 error, retval, syncstates;
1956 	int			 odelay;
1957 #ifdef KTR
1958 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1959 #endif
1960 
1961 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1962 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1963 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1964 
1965 	error = 0;
1966 	syncstates = 1;
1967 
1968 	ifp = inm->in6m_ifp;
1969 
1970 	IN6_MULTI_LIST_LOCK_ASSERT();
1971 	MLD_LOCK_ASSERT();
1972 
1973 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1974 
1975 	/*
1976 	 * Groups joined on loopback or marked as 'not reported',
1977 	 * enter the MLD_SILENT_MEMBER state and
1978 	 * are never reported in any protocol exchanges.
1979 	 * All other groups enter the appropriate state machine
1980 	 * for the version in use on this link.
1981 	 * A link marked as MLIF_SILENT causes MLD to be completely
1982 	 * disabled for the link.
1983 	 */
1984 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1985 	    (mli->mli_flags & MLIF_SILENT) ||
1986 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1987 		CTR1(KTR_MLD,
1988 "%s: not kicking state machine for silent group", __func__);
1989 		inm->in6m_state = MLD_SILENT_MEMBER;
1990 		inm->in6m_timer = 0;
1991 	} else {
1992 		/*
1993 		 * Deal with overlapping in_multi lifecycle.
1994 		 * If this group was LEAVING, then make sure
1995 		 * we drop the reference we picked up to keep the
1996 		 * group around for the final INCLUDE {} enqueue.
1997 		 */
1998 		if (mli->mli_version == MLD_VERSION_2 &&
1999 		    inm->in6m_state == MLD_LEAVING_MEMBER)
2000 			in6m_release_deferred(inm);
2001 
2002 		inm->in6m_state = MLD_REPORTING_MEMBER;
2003 
2004 		switch (mli->mli_version) {
2005 		case MLD_VERSION_1:
2006 			/*
2007 			 * If a delay was provided, only use it if
2008 			 * it is greater than the delay normally
2009 			 * used for an MLDv1 state change report,
2010 			 * and delay sending the initial MLDv1 report
2011 			 * by not transitioning to the IDLE state.
2012 			 */
2013 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2014 			if (delay) {
2015 				inm->in6m_timer = max(delay, odelay);
2016 				V_current_state_timers_running6 = 1;
2017 			} else {
2018 				inm->in6m_state = MLD_IDLE_MEMBER;
2019 				error = mld_v1_transmit_report(inm,
2020 				     MLD_LISTENER_REPORT);
2021 				if (error == 0) {
2022 					inm->in6m_timer = odelay;
2023 					V_current_state_timers_running6 = 1;
2024 				}
2025 			}
2026 			break;
2027 
2028 		case MLD_VERSION_2:
2029 			/*
2030 			 * Defer update of T0 to T1, until the first copy
2031 			 * of the state change has been transmitted.
2032 			 */
2033 			syncstates = 0;
2034 
2035 			/*
2036 			 * Immediately enqueue a State-Change Report for
2037 			 * this interface, freeing any previous reports.
2038 			 * Don't kick the timers if there is nothing to do,
2039 			 * or if an error occurred.
2040 			 */
2041 			mq = &inm->in6m_scq;
2042 			mbufq_drain(mq);
2043 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2044 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2045 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2046 			    __func__, retval);
2047 			if (retval <= 0) {
2048 				error = retval * -1;
2049 				break;
2050 			}
2051 
2052 			/*
2053 			 * Schedule transmission of pending state-change
2054 			 * report up to RV times for this link. The timer
2055 			 * will fire at the next mld_fasttimo (~200ms),
2056 			 * giving us an opportunity to merge the reports.
2057 			 *
2058 			 * If a delay was provided to this function, only
2059 			 * use this delay if sooner than the existing one.
2060 			 */
2061 			KASSERT(mli->mli_rv > 1,
2062 			   ("%s: invalid robustness %d", __func__,
2063 			    mli->mli_rv));
2064 			inm->in6m_scrv = mli->mli_rv;
2065 			if (delay) {
2066 				if (inm->in6m_sctimer > 1) {
2067 					inm->in6m_sctimer =
2068 					    min(inm->in6m_sctimer, delay);
2069 				} else
2070 					inm->in6m_sctimer = delay;
2071 			} else
2072 				inm->in6m_sctimer = 1;
2073 			V_state_change_timers_running6 = 1;
2074 
2075 			error = 0;
2076 			break;
2077 		}
2078 	}
2079 
2080 	/*
2081 	 * Only update the T0 state if state change is atomic,
2082 	 * i.e. we don't need to wait for a timer to fire before we
2083 	 * can consider the state change to have been communicated.
2084 	 */
2085 	if (syncstates) {
2086 		in6m_commit(inm);
2087 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2088 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2089 		    if_name(inm->in6m_ifp));
2090 	}
2091 
2092 	return (error);
2093 }
2094 
2095 /*
2096  * Issue an intermediate state change during the life-cycle.
2097  */
2098 static int
2099 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2100 {
2101 	struct ifnet		*ifp;
2102 	int			 retval;
2103 #ifdef KTR
2104 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2105 #endif
2106 
2107 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2108 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2109 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2110 
2111 	ifp = inm->in6m_ifp;
2112 
2113 	IN6_MULTI_LIST_LOCK_ASSERT();
2114 	MLD_LOCK_ASSERT();
2115 
2116 	KASSERT(mli && mli->mli_ifp == ifp,
2117 	    ("%s: inconsistent ifp", __func__));
2118 
2119 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2120 	    (mli->mli_flags & MLIF_SILENT) ||
2121 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2122 	    (mli->mli_version != MLD_VERSION_2)) {
2123 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2124 			CTR1(KTR_MLD,
2125 "%s: not kicking state machine for silent group", __func__);
2126 		}
2127 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2128 		in6m_commit(inm);
2129 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2130 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2131 		    if_name(inm->in6m_ifp));
2132 		return (0);
2133 	}
2134 
2135 	mbufq_drain(&inm->in6m_scq);
2136 
2137 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2138 	    (mli->mli_flags & MLIF_USEALLOW));
2139 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2140 	if (retval <= 0)
2141 		return (-retval);
2142 
2143 	/*
2144 	 * If record(s) were enqueued, start the state-change
2145 	 * report timer for this group.
2146 	 */
2147 	inm->in6m_scrv = mli->mli_rv;
2148 	inm->in6m_sctimer = 1;
2149 	V_state_change_timers_running6 = 1;
2150 
2151 	return (0);
2152 }
2153 
2154 /*
2155  * Perform the final leave for a multicast address.
2156  *
2157  * When leaving a group:
2158  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2159  *  MLDv2 enqueues a state-change report containing a transition
2160  *  to INCLUDE {} for immediate transmission.
2161  */
2162 static void
2163 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2164 {
2165 	int syncstates;
2166 #ifdef KTR
2167 	char ip6tbuf[INET6_ADDRSTRLEN];
2168 #endif
2169 
2170 	syncstates = 1;
2171 
2172 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2173 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2174 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2175 
2176 	IN6_MULTI_LIST_LOCK_ASSERT();
2177 	MLD_LOCK_ASSERT();
2178 
2179 	switch (inm->in6m_state) {
2180 	case MLD_NOT_MEMBER:
2181 	case MLD_SILENT_MEMBER:
2182 	case MLD_LEAVING_MEMBER:
2183 		/* Already leaving or left; do nothing. */
2184 		CTR1(KTR_MLD,
2185 "%s: not kicking state machine for silent group", __func__);
2186 		break;
2187 	case MLD_REPORTING_MEMBER:
2188 	case MLD_IDLE_MEMBER:
2189 	case MLD_G_QUERY_PENDING_MEMBER:
2190 	case MLD_SG_QUERY_PENDING_MEMBER:
2191 		if (mli->mli_version == MLD_VERSION_1) {
2192 #ifdef INVARIANTS
2193 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2194 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2195 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2196 			     __func__);
2197 #endif
2198 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2199 			inm->in6m_state = MLD_NOT_MEMBER;
2200 			V_current_state_timers_running6 = 1;
2201 		} else if (mli->mli_version == MLD_VERSION_2) {
2202 			/*
2203 			 * Stop group timer and all pending reports.
2204 			 * Immediately enqueue a state-change report
2205 			 * TO_IN {} to be sent on the next fast timeout,
2206 			 * giving us an opportunity to merge reports.
2207 			 */
2208 			mbufq_drain(&inm->in6m_scq);
2209 			inm->in6m_timer = 0;
2210 			inm->in6m_scrv = mli->mli_rv;
2211 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2212 			    "pending retransmissions.", __func__,
2213 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2214 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2215 			if (inm->in6m_scrv == 0) {
2216 				inm->in6m_state = MLD_NOT_MEMBER;
2217 				inm->in6m_sctimer = 0;
2218 			} else {
2219 				int retval;
2220 
2221 				in6m_acquire_locked(inm);
2222 
2223 				retval = mld_v2_enqueue_group_record(
2224 				    &inm->in6m_scq, inm, 1, 0, 0,
2225 				    (mli->mli_flags & MLIF_USEALLOW));
2226 				KASSERT(retval != 0,
2227 				    ("%s: enqueue record = %d", __func__,
2228 				     retval));
2229 
2230 				inm->in6m_state = MLD_LEAVING_MEMBER;
2231 				inm->in6m_sctimer = 1;
2232 				V_state_change_timers_running6 = 1;
2233 				syncstates = 0;
2234 			}
2235 			break;
2236 		}
2237 		break;
2238 	case MLD_LAZY_MEMBER:
2239 	case MLD_SLEEPING_MEMBER:
2240 	case MLD_AWAKENING_MEMBER:
2241 		/* Our reports are suppressed; do nothing. */
2242 		break;
2243 	}
2244 
2245 	if (syncstates) {
2246 		in6m_commit(inm);
2247 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2248 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2249 		    if_name(inm->in6m_ifp));
2250 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2251 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2252 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2253 	}
2254 }
2255 
2256 /*
2257  * Enqueue an MLDv2 group record to the given output queue.
2258  *
2259  * If is_state_change is zero, a current-state record is appended.
2260  * If is_state_change is non-zero, a state-change report is appended.
2261  *
2262  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2263  * If is_group_query is zero, and if there is a packet with free space
2264  * at the tail of the queue, it will be appended to providing there
2265  * is enough free space.
2266  * Otherwise a new mbuf packet chain is allocated.
2267  *
2268  * If is_source_query is non-zero, each source is checked to see if
2269  * it was recorded for a Group-Source query, and will be omitted if
2270  * it is not both in-mode and recorded.
2271  *
2272  * If use_block_allow is non-zero, state change reports for initial join
2273  * and final leave, on an inclusive mode group with a source list, will be
2274  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2275  *
2276  * The function will attempt to allocate leading space in the packet
2277  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2278  *
2279  * If successful the size of all data appended to the queue is returned,
2280  * otherwise an error code less than zero is returned, or zero if
2281  * no record(s) were appended.
2282  */
2283 static int
2284 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2285     const int is_state_change, const int is_group_query,
2286     const int is_source_query, const int use_block_allow)
2287 {
2288 	struct mldv2_record	 mr;
2289 	struct mldv2_record	*pmr;
2290 	struct ifnet		*ifp;
2291 	struct ip6_msource	*ims, *nims;
2292 	struct mbuf		*m0, *m, *md;
2293 	int			 is_filter_list_change;
2294 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2295 	int			 record_has_sources;
2296 	int			 now;
2297 	int			 type;
2298 	uint8_t			 mode;
2299 #ifdef KTR
2300 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2301 #endif
2302 
2303 	IN6_MULTI_LIST_LOCK_ASSERT();
2304 
2305 	ifp = inm->in6m_ifp;
2306 	is_filter_list_change = 0;
2307 	m = NULL;
2308 	m0 = NULL;
2309 	m0srcs = 0;
2310 	msrcs = 0;
2311 	nbytes = 0;
2312 	nims = NULL;
2313 	record_has_sources = 1;
2314 	pmr = NULL;
2315 	type = MLD_DO_NOTHING;
2316 	mode = inm->in6m_st[1].iss_fmode;
2317 
2318 	/*
2319 	 * If we did not transition out of ASM mode during t0->t1,
2320 	 * and there are no source nodes to process, we can skip
2321 	 * the generation of source records.
2322 	 */
2323 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2324 	    inm->in6m_nsrc == 0)
2325 		record_has_sources = 0;
2326 
2327 	if (is_state_change) {
2328 		/*
2329 		 * Queue a state change record.
2330 		 * If the mode did not change, and there are non-ASM
2331 		 * listeners or source filters present,
2332 		 * we potentially need to issue two records for the group.
2333 		 * If there are ASM listeners, and there was no filter
2334 		 * mode transition of any kind, do nothing.
2335 		 *
2336 		 * If we are transitioning to MCAST_UNDEFINED, we need
2337 		 * not send any sources. A transition to/from this state is
2338 		 * considered inclusive with some special treatment.
2339 		 *
2340 		 * If we are rewriting initial joins/leaves to use
2341 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2342 		 * we need to send sources in all cases.
2343 		 */
2344 		if (mode != inm->in6m_st[0].iss_fmode) {
2345 			if (mode == MCAST_EXCLUDE) {
2346 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2347 				    __func__);
2348 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2349 			} else {
2350 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2351 				    __func__);
2352 				if (use_block_allow) {
2353 					/*
2354 					 * XXX
2355 					 * Here we're interested in state
2356 					 * edges either direction between
2357 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2358 					 * Perhaps we should just check
2359 					 * the group state, rather than
2360 					 * the filter mode.
2361 					 */
2362 					if (mode == MCAST_UNDEFINED) {
2363 						type = MLD_BLOCK_OLD_SOURCES;
2364 					} else {
2365 						type = MLD_ALLOW_NEW_SOURCES;
2366 					}
2367 				} else {
2368 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2369 					if (mode == MCAST_UNDEFINED)
2370 						record_has_sources = 0;
2371 				}
2372 			}
2373 		} else {
2374 			if (record_has_sources) {
2375 				is_filter_list_change = 1;
2376 			} else {
2377 				type = MLD_DO_NOTHING;
2378 			}
2379 		}
2380 	} else {
2381 		/*
2382 		 * Queue a current state record.
2383 		 */
2384 		if (mode == MCAST_EXCLUDE) {
2385 			type = MLD_MODE_IS_EXCLUDE;
2386 		} else if (mode == MCAST_INCLUDE) {
2387 			type = MLD_MODE_IS_INCLUDE;
2388 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2389 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2390 			     __func__, inm, inm->in6m_st[1].iss_asm));
2391 		}
2392 	}
2393 
2394 	/*
2395 	 * Generate the filter list changes using a separate function.
2396 	 */
2397 	if (is_filter_list_change)
2398 		return (mld_v2_enqueue_filter_change(mq, inm));
2399 
2400 	if (type == MLD_DO_NOTHING) {
2401 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2402 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2403 		    if_name(inm->in6m_ifp));
2404 		return (0);
2405 	}
2406 
2407 	/*
2408 	 * If any sources are present, we must be able to fit at least
2409 	 * one in the trailing space of the tail packet's mbuf,
2410 	 * ideally more.
2411 	 */
2412 	minrec0len = sizeof(struct mldv2_record);
2413 	if (record_has_sources)
2414 		minrec0len += sizeof(struct in6_addr);
2415 
2416 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2417 	    mld_rec_type_to_str(type),
2418 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2419 	    if_name(inm->in6m_ifp));
2420 
2421 	/*
2422 	 * Check if we have a packet in the tail of the queue for this
2423 	 * group into which the first group record for this group will fit.
2424 	 * Otherwise allocate a new packet.
2425 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2426 	 * Note: Group records for G/GSR query responses MUST be sent
2427 	 * in their own packet.
2428 	 */
2429 	m0 = mbufq_last(mq);
2430 	if (!is_group_query &&
2431 	    m0 != NULL &&
2432 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2433 	    (m0->m_pkthdr.len + minrec0len) <
2434 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2435 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2436 			    sizeof(struct mldv2_record)) /
2437 			    sizeof(struct in6_addr);
2438 		m = m0;
2439 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2440 	} else {
2441 		if (mbufq_full(mq)) {
2442 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2443 			return (-ENOMEM);
2444 		}
2445 		m = NULL;
2446 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2447 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2448 		if (!is_state_change && !is_group_query)
2449 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2450 		if (m == NULL)
2451 			m = m_gethdr(M_NOWAIT, MT_DATA);
2452 		if (m == NULL)
2453 			return (-ENOMEM);
2454 
2455 		mld_save_context(m, ifp);
2456 
2457 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2458 	}
2459 
2460 	/*
2461 	 * Append group record.
2462 	 * If we have sources, we don't know how many yet.
2463 	 */
2464 	mr.mr_type = type;
2465 	mr.mr_datalen = 0;
2466 	mr.mr_numsrc = 0;
2467 	mr.mr_addr = inm->in6m_addr;
2468 	in6_clearscope(&mr.mr_addr);
2469 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2470 		if (m != m0)
2471 			m_freem(m);
2472 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2473 		return (-ENOMEM);
2474 	}
2475 	nbytes += sizeof(struct mldv2_record);
2476 
2477 	/*
2478 	 * Append as many sources as will fit in the first packet.
2479 	 * If we are appending to a new packet, the chain allocation
2480 	 * may potentially use clusters; use m_getptr() in this case.
2481 	 * If we are appending to an existing packet, we need to obtain
2482 	 * a pointer to the group record after m_append(), in case a new
2483 	 * mbuf was allocated.
2484 	 *
2485 	 * Only append sources which are in-mode at t1. If we are
2486 	 * transitioning to MCAST_UNDEFINED state on the group, and
2487 	 * use_block_allow is zero, do not include source entries.
2488 	 * Otherwise, we need to include this source in the report.
2489 	 *
2490 	 * Only report recorded sources in our filter set when responding
2491 	 * to a group-source query.
2492 	 */
2493 	if (record_has_sources) {
2494 		if (m == m0) {
2495 			md = m_last(m);
2496 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2497 			    md->m_len - nbytes);
2498 		} else {
2499 			md = m_getptr(m, 0, &off);
2500 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2501 			    off);
2502 		}
2503 		msrcs = 0;
2504 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2505 		    nims) {
2506 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2507 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2508 			now = im6s_get_mode(inm, ims, 1);
2509 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2510 			if ((now != mode) ||
2511 			    (now == mode &&
2512 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2513 				CTR1(KTR_MLD, "%s: skip node", __func__);
2514 				continue;
2515 			}
2516 			if (is_source_query && ims->im6s_stp == 0) {
2517 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2518 				    __func__);
2519 				continue;
2520 			}
2521 			CTR1(KTR_MLD, "%s: append node", __func__);
2522 			if (!m_append(m, sizeof(struct in6_addr),
2523 			    (void *)&ims->im6s_addr)) {
2524 				if (m != m0)
2525 					m_freem(m);
2526 				CTR1(KTR_MLD, "%s: m_append() failed.",
2527 				    __func__);
2528 				return (-ENOMEM);
2529 			}
2530 			nbytes += sizeof(struct in6_addr);
2531 			++msrcs;
2532 			if (msrcs == m0srcs)
2533 				break;
2534 		}
2535 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2536 		    msrcs);
2537 		pmr->mr_numsrc = htons(msrcs);
2538 		nbytes += (msrcs * sizeof(struct in6_addr));
2539 	}
2540 
2541 	if (is_source_query && msrcs == 0) {
2542 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2543 		if (m != m0)
2544 			m_freem(m);
2545 		return (0);
2546 	}
2547 
2548 	/*
2549 	 * We are good to go with first packet.
2550 	 */
2551 	if (m != m0) {
2552 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2553 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2554 		mbufq_enqueue(mq, m);
2555 	} else
2556 		m->m_pkthdr.PH_vt.vt_nrecs++;
2557 
2558 	/*
2559 	 * No further work needed if no source list in packet(s).
2560 	 */
2561 	if (!record_has_sources)
2562 		return (nbytes);
2563 
2564 	/*
2565 	 * Whilst sources remain to be announced, we need to allocate
2566 	 * a new packet and fill out as many sources as will fit.
2567 	 * Always try for a cluster first.
2568 	 */
2569 	while (nims != NULL) {
2570 		if (mbufq_full(mq)) {
2571 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2572 			return (-ENOMEM);
2573 		}
2574 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2575 		if (m == NULL)
2576 			m = m_gethdr(M_NOWAIT, MT_DATA);
2577 		if (m == NULL)
2578 			return (-ENOMEM);
2579 		mld_save_context(m, ifp);
2580 		md = m_getptr(m, 0, &off);
2581 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2582 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2583 
2584 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2585 			if (m != m0)
2586 				m_freem(m);
2587 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2588 			return (-ENOMEM);
2589 		}
2590 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2591 		nbytes += sizeof(struct mldv2_record);
2592 
2593 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2594 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2595 
2596 		msrcs = 0;
2597 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2598 			CTR2(KTR_MLD, "%s: visit node %s",
2599 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2600 			now = im6s_get_mode(inm, ims, 1);
2601 			if ((now != mode) ||
2602 			    (now == mode &&
2603 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2604 				CTR1(KTR_MLD, "%s: skip node", __func__);
2605 				continue;
2606 			}
2607 			if (is_source_query && ims->im6s_stp == 0) {
2608 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2609 				    __func__);
2610 				continue;
2611 			}
2612 			CTR1(KTR_MLD, "%s: append node", __func__);
2613 			if (!m_append(m, sizeof(struct in6_addr),
2614 			    (void *)&ims->im6s_addr)) {
2615 				if (m != m0)
2616 					m_freem(m);
2617 				CTR1(KTR_MLD, "%s: m_append() failed.",
2618 				    __func__);
2619 				return (-ENOMEM);
2620 			}
2621 			++msrcs;
2622 			if (msrcs == m0srcs)
2623 				break;
2624 		}
2625 		pmr->mr_numsrc = htons(msrcs);
2626 		nbytes += (msrcs * sizeof(struct in6_addr));
2627 
2628 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2629 		mbufq_enqueue(mq, m);
2630 	}
2631 
2632 	return (nbytes);
2633 }
2634 
2635 /*
2636  * Type used to mark record pass completion.
2637  * We exploit the fact we can cast to this easily from the
2638  * current filter modes on each ip_msource node.
2639  */
2640 typedef enum {
2641 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2642 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2643 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2644 	REC_FULL = REC_ALLOW | REC_BLOCK
2645 } rectype_t;
2646 
2647 /*
2648  * Enqueue an MLDv2 filter list change to the given output queue.
2649  *
2650  * Source list filter state is held in an RB-tree. When the filter list
2651  * for a group is changed without changing its mode, we need to compute
2652  * the deltas between T0 and T1 for each source in the filter set,
2653  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2654  *
2655  * As we may potentially queue two record types, and the entire R-B tree
2656  * needs to be walked at once, we break this out into its own function
2657  * so we can generate a tightly packed queue of packets.
2658  *
2659  * XXX This could be written to only use one tree walk, although that makes
2660  * serializing into the mbuf chains a bit harder. For now we do two walks
2661  * which makes things easier on us, and it may or may not be harder on
2662  * the L2 cache.
2663  *
2664  * If successful the size of all data appended to the queue is returned,
2665  * otherwise an error code less than zero is returned, or zero if
2666  * no record(s) were appended.
2667  */
2668 static int
2669 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2670 {
2671 	static const int MINRECLEN =
2672 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2673 	struct ifnet		*ifp;
2674 	struct mldv2_record	 mr;
2675 	struct mldv2_record	*pmr;
2676 	struct ip6_msource	*ims, *nims;
2677 	struct mbuf		*m, *m0, *md;
2678 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2679 	int			 nallow, nblock;
2680 	uint8_t			 mode, now, then;
2681 	rectype_t		 crt, drt, nrt;
2682 #ifdef KTR
2683 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2684 #endif
2685 
2686 	IN6_MULTI_LIST_LOCK_ASSERT();
2687 
2688 	if (inm->in6m_nsrc == 0 ||
2689 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2690 		return (0);
2691 
2692 	ifp = inm->in6m_ifp;			/* interface */
2693 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2694 	crt = REC_NONE;	/* current group record type */
2695 	drt = REC_NONE;	/* mask of completed group record types */
2696 	nrt = REC_NONE;	/* record type for current node */
2697 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2698 	npbytes = 0;	/* # of bytes appended this packet */
2699 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2700 	rsrcs = 0;	/* # sources encoded in current record */
2701 	schanged = 0;	/* # nodes encoded in overall filter change */
2702 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2703 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2704 	nims = NULL;	/* next tree node pointer */
2705 
2706 	/*
2707 	 * For each possible filter record mode.
2708 	 * The first kind of source we encounter tells us which
2709 	 * is the first kind of record we start appending.
2710 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2711 	 * as the inverse of the group's filter mode.
2712 	 */
2713 	while (drt != REC_FULL) {
2714 		do {
2715 			m0 = mbufq_last(mq);
2716 			if (m0 != NULL &&
2717 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2718 			     MLD_V2_REPORT_MAXRECS) &&
2719 			    (m0->m_pkthdr.len + MINRECLEN) <
2720 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2721 				m = m0;
2722 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2723 					    sizeof(struct mldv2_record)) /
2724 					    sizeof(struct in6_addr);
2725 				CTR1(KTR_MLD,
2726 				    "%s: use previous packet", __func__);
2727 			} else {
2728 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2729 				if (m == NULL)
2730 					m = m_gethdr(M_NOWAIT, MT_DATA);
2731 				if (m == NULL) {
2732 					CTR1(KTR_MLD,
2733 					    "%s: m_get*() failed", __func__);
2734 					return (-ENOMEM);
2735 				}
2736 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2737 				mld_save_context(m, ifp);
2738 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2739 				    sizeof(struct mldv2_record)) /
2740 				    sizeof(struct in6_addr);
2741 				npbytes = 0;
2742 				CTR1(KTR_MLD,
2743 				    "%s: allocated new packet", __func__);
2744 			}
2745 			/*
2746 			 * Append the MLD group record header to the
2747 			 * current packet's data area.
2748 			 * Recalculate pointer to free space for next
2749 			 * group record, in case m_append() allocated
2750 			 * a new mbuf or cluster.
2751 			 */
2752 			memset(&mr, 0, sizeof(mr));
2753 			mr.mr_addr = inm->in6m_addr;
2754 			in6_clearscope(&mr.mr_addr);
2755 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2756 				if (m != m0)
2757 					m_freem(m);
2758 				CTR1(KTR_MLD,
2759 				    "%s: m_append() failed", __func__);
2760 				return (-ENOMEM);
2761 			}
2762 			npbytes += sizeof(struct mldv2_record);
2763 			if (m != m0) {
2764 				/* new packet; offset in chain */
2765 				md = m_getptr(m, npbytes -
2766 				    sizeof(struct mldv2_record), &off);
2767 				pmr = (struct mldv2_record *)(mtod(md,
2768 				    uint8_t *) + off);
2769 			} else {
2770 				/* current packet; offset from last append */
2771 				md = m_last(m);
2772 				pmr = (struct mldv2_record *)(mtod(md,
2773 				    uint8_t *) + md->m_len -
2774 				    sizeof(struct mldv2_record));
2775 			}
2776 			/*
2777 			 * Begin walking the tree for this record type
2778 			 * pass, or continue from where we left off
2779 			 * previously if we had to allocate a new packet.
2780 			 * Only report deltas in-mode at t1.
2781 			 * We need not report included sources as allowed
2782 			 * if we are in inclusive mode on the group,
2783 			 * however the converse is not true.
2784 			 */
2785 			rsrcs = 0;
2786 			if (nims == NULL) {
2787 				nims = RB_MIN(ip6_msource_tree,
2788 				    &inm->in6m_srcs);
2789 			}
2790 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2791 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2792 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2793 				now = im6s_get_mode(inm, ims, 1);
2794 				then = im6s_get_mode(inm, ims, 0);
2795 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2796 				    __func__, then, now);
2797 				if (now == then) {
2798 					CTR1(KTR_MLD,
2799 					    "%s: skip unchanged", __func__);
2800 					continue;
2801 				}
2802 				if (mode == MCAST_EXCLUDE &&
2803 				    now == MCAST_INCLUDE) {
2804 					CTR1(KTR_MLD,
2805 					    "%s: skip IN src on EX group",
2806 					    __func__);
2807 					continue;
2808 				}
2809 				nrt = (rectype_t)now;
2810 				if (nrt == REC_NONE)
2811 					nrt = (rectype_t)(~mode & REC_FULL);
2812 				if (schanged++ == 0) {
2813 					crt = nrt;
2814 				} else if (crt != nrt)
2815 					continue;
2816 				if (!m_append(m, sizeof(struct in6_addr),
2817 				    (void *)&ims->im6s_addr)) {
2818 					if (m != m0)
2819 						m_freem(m);
2820 					CTR1(KTR_MLD,
2821 					    "%s: m_append() failed", __func__);
2822 					return (-ENOMEM);
2823 				}
2824 				nallow += !!(crt == REC_ALLOW);
2825 				nblock += !!(crt == REC_BLOCK);
2826 				if (++rsrcs == m0srcs)
2827 					break;
2828 			}
2829 			/*
2830 			 * If we did not append any tree nodes on this
2831 			 * pass, back out of allocations.
2832 			 */
2833 			if (rsrcs == 0) {
2834 				npbytes -= sizeof(struct mldv2_record);
2835 				if (m != m0) {
2836 					CTR1(KTR_MLD,
2837 					    "%s: m_free(m)", __func__);
2838 					m_freem(m);
2839 				} else {
2840 					CTR1(KTR_MLD,
2841 					    "%s: m_adj(m, -mr)", __func__);
2842 					m_adj(m, -((int)sizeof(
2843 					    struct mldv2_record)));
2844 				}
2845 				continue;
2846 			}
2847 			npbytes += (rsrcs * sizeof(struct in6_addr));
2848 			if (crt == REC_ALLOW)
2849 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2850 			else if (crt == REC_BLOCK)
2851 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2852 			pmr->mr_numsrc = htons(rsrcs);
2853 			/*
2854 			 * Count the new group record, and enqueue this
2855 			 * packet if it wasn't already queued.
2856 			 */
2857 			m->m_pkthdr.PH_vt.vt_nrecs++;
2858 			if (m != m0)
2859 				mbufq_enqueue(mq, m);
2860 			nbytes += npbytes;
2861 		} while (nims != NULL);
2862 		drt |= crt;
2863 		crt = (~crt & REC_FULL);
2864 	}
2865 
2866 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2867 	    nallow, nblock);
2868 
2869 	return (nbytes);
2870 }
2871 
2872 static int
2873 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2874 {
2875 	struct mbufq	*gq;
2876 	struct mbuf	*m;		/* pending state-change */
2877 	struct mbuf	*m0;		/* copy of pending state-change */
2878 	struct mbuf	*mt;		/* last state-change in packet */
2879 	int		 docopy, domerge;
2880 	u_int		 recslen;
2881 
2882 	docopy = 0;
2883 	domerge = 0;
2884 	recslen = 0;
2885 
2886 	IN6_MULTI_LIST_LOCK_ASSERT();
2887 	MLD_LOCK_ASSERT();
2888 
2889 	/*
2890 	 * If there are further pending retransmissions, make a writable
2891 	 * copy of each queued state-change message before merging.
2892 	 */
2893 	if (inm->in6m_scrv > 0)
2894 		docopy = 1;
2895 
2896 	gq = &inm->in6m_scq;
2897 #ifdef KTR
2898 	if (mbufq_first(gq) == NULL) {
2899 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2900 		    __func__, inm);
2901 	}
2902 #endif
2903 
2904 	m = mbufq_first(gq);
2905 	while (m != NULL) {
2906 		/*
2907 		 * Only merge the report into the current packet if
2908 		 * there is sufficient space to do so; an MLDv2 report
2909 		 * packet may only contain 65,535 group records.
2910 		 * Always use a simple mbuf chain concatentation to do this,
2911 		 * as large state changes for single groups may have
2912 		 * allocated clusters.
2913 		 */
2914 		domerge = 0;
2915 		mt = mbufq_last(scq);
2916 		if (mt != NULL) {
2917 			recslen = m_length(m, NULL);
2918 
2919 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2920 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2921 			    MLD_V2_REPORT_MAXRECS) &&
2922 			    (mt->m_pkthdr.len + recslen <=
2923 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2924 				domerge = 1;
2925 		}
2926 
2927 		if (!domerge && mbufq_full(gq)) {
2928 			CTR2(KTR_MLD,
2929 			    "%s: outbound queue full, skipping whole packet %p",
2930 			    __func__, m);
2931 			mt = m->m_nextpkt;
2932 			if (!docopy)
2933 				m_freem(m);
2934 			m = mt;
2935 			continue;
2936 		}
2937 
2938 		if (!docopy) {
2939 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2940 			m0 = mbufq_dequeue(gq);
2941 			m = m0->m_nextpkt;
2942 		} else {
2943 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2944 			m0 = m_dup(m, M_NOWAIT);
2945 			if (m0 == NULL)
2946 				return (ENOMEM);
2947 			m0->m_nextpkt = NULL;
2948 			m = m->m_nextpkt;
2949 		}
2950 
2951 		if (!domerge) {
2952 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2953 			    __func__, m0, scq);
2954 			mbufq_enqueue(scq, m0);
2955 		} else {
2956 			struct mbuf *mtl;	/* last mbuf of packet mt */
2957 
2958 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2959 			    __func__, m0, mt);
2960 
2961 			mtl = m_last(mt);
2962 			m0->m_flags &= ~M_PKTHDR;
2963 			mt->m_pkthdr.len += recslen;
2964 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2965 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2966 
2967 			mtl->m_next = m0;
2968 		}
2969 	}
2970 
2971 	return (0);
2972 }
2973 
2974 /*
2975  * Respond to a pending MLDv2 General Query.
2976  */
2977 static void
2978 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2979 {
2980 	struct ifmultiaddr	*ifma;
2981 	struct ifnet		*ifp;
2982 	struct in6_multi	*inm;
2983 	int			 retval;
2984 
2985 	IN6_MULTI_LIST_LOCK_ASSERT();
2986 	MLD_LOCK_ASSERT();
2987 
2988 	KASSERT(mli->mli_version == MLD_VERSION_2,
2989 	    ("%s: called when version %d", __func__, mli->mli_version));
2990 
2991 	/*
2992 	 * Check that there are some packets queued. If so, send them first.
2993 	 * For large number of groups the reply to general query can take
2994 	 * many packets, we should finish sending them before starting of
2995 	 * queuing the new reply.
2996 	 */
2997 	if (mbufq_len(&mli->mli_gq) != 0)
2998 		goto send;
2999 
3000 	ifp = mli->mli_ifp;
3001 
3002 	IF_ADDR_RLOCK(ifp);
3003 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3004 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
3005 		    ifma->ifma_protospec == NULL)
3006 			continue;
3007 
3008 		inm = (struct in6_multi *)ifma->ifma_protospec;
3009 		KASSERT(ifp == inm->in6m_ifp,
3010 		    ("%s: inconsistent ifp", __func__));
3011 
3012 		switch (inm->in6m_state) {
3013 		case MLD_NOT_MEMBER:
3014 		case MLD_SILENT_MEMBER:
3015 			break;
3016 		case MLD_REPORTING_MEMBER:
3017 		case MLD_IDLE_MEMBER:
3018 		case MLD_LAZY_MEMBER:
3019 		case MLD_SLEEPING_MEMBER:
3020 		case MLD_AWAKENING_MEMBER:
3021 			inm->in6m_state = MLD_REPORTING_MEMBER;
3022 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3023 			    inm, 0, 0, 0, 0);
3024 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3025 			    __func__, retval);
3026 			break;
3027 		case MLD_G_QUERY_PENDING_MEMBER:
3028 		case MLD_SG_QUERY_PENDING_MEMBER:
3029 		case MLD_LEAVING_MEMBER:
3030 			break;
3031 		}
3032 	}
3033 	IF_ADDR_RUNLOCK(ifp);
3034 
3035 send:
3036 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3037 
3038 	/*
3039 	 * Slew transmission of bursts over 500ms intervals.
3040 	 */
3041 	if (mbufq_first(&mli->mli_gq) != NULL) {
3042 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3043 		    MLD_RESPONSE_BURST_INTERVAL);
3044 		V_interface_timers_running6 = 1;
3045 	}
3046 }
3047 
3048 /*
3049  * Transmit the next pending message in the output queue.
3050  *
3051  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3052  * MRT: Nothing needs to be done, as MLD traffic is always local to
3053  * a link and uses a link-scope multicast address.
3054  */
3055 static void
3056 mld_dispatch_packet(struct mbuf *m)
3057 {
3058 	struct ip6_moptions	 im6o;
3059 	struct ifnet		*ifp;
3060 	struct ifnet		*oifp;
3061 	struct mbuf		*m0;
3062 	struct mbuf		*md;
3063 	struct ip6_hdr		*ip6;
3064 	struct mld_hdr		*mld;
3065 	int			 error;
3066 	int			 off;
3067 	int			 type;
3068 	uint32_t		 ifindex;
3069 
3070 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3071 
3072 	/*
3073 	 * Set VNET image pointer from enqueued mbuf chain
3074 	 * before doing anything else. Whilst we use interface
3075 	 * indexes to guard against interface detach, they are
3076 	 * unique to each VIMAGE and must be retrieved.
3077 	 */
3078 	ifindex = mld_restore_context(m);
3079 
3080 	/*
3081 	 * Check if the ifnet still exists. This limits the scope of
3082 	 * any race in the absence of a global ifp lock for low cost
3083 	 * (an array lookup).
3084 	 */
3085 	ifp = ifnet_byindex(ifindex);
3086 	if (ifp == NULL) {
3087 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3088 		    __func__, m, ifindex);
3089 		m_freem(m);
3090 		IP6STAT_INC(ip6s_noroute);
3091 		goto out;
3092 	}
3093 
3094 	im6o.im6o_multicast_hlim  = 1;
3095 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3096 	im6o.im6o_multicast_ifp = ifp;
3097 
3098 	if (m->m_flags & M_MLDV1) {
3099 		m0 = m;
3100 	} else {
3101 		m0 = mld_v2_encap_report(ifp, m);
3102 		if (m0 == NULL) {
3103 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3104 			IP6STAT_INC(ip6s_odropped);
3105 			goto out;
3106 		}
3107 	}
3108 
3109 	mld_scrub_context(m0);
3110 	m_clrprotoflags(m);
3111 	m0->m_pkthdr.rcvif = V_loif;
3112 
3113 	ip6 = mtod(m0, struct ip6_hdr *);
3114 #if 0
3115 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3116 #else
3117 	/*
3118 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3119 	 * occur if we called in6_setscope() at transmission.
3120 	 * See comments at top of file.
3121 	 */
3122 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3123 #endif
3124 
3125 	/*
3126 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3127 	 * so we can bump the stats.
3128 	 */
3129 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3130 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3131 	type = mld->mld_type;
3132 
3133 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3134 	    &oifp, NULL);
3135 	if (error) {
3136 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3137 		goto out;
3138 	}
3139 	ICMP6STAT_INC(icp6s_outhist[type]);
3140 	if (oifp != NULL) {
3141 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3142 		switch (type) {
3143 		case MLD_LISTENER_REPORT:
3144 		case MLDV2_LISTENER_REPORT:
3145 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3146 			break;
3147 		case MLD_LISTENER_DONE:
3148 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3149 			break;
3150 		}
3151 	}
3152 out:
3153 	return;
3154 }
3155 
3156 /*
3157  * Encapsulate an MLDv2 report.
3158  *
3159  * KAME IPv6 requires that hop-by-hop options be passed separately,
3160  * and that the IPv6 header be prepended in a separate mbuf.
3161  *
3162  * Returns a pointer to the new mbuf chain head, or NULL if the
3163  * allocation failed.
3164  */
3165 static struct mbuf *
3166 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3167 {
3168 	struct mbuf		*mh;
3169 	struct mldv2_report	*mld;
3170 	struct ip6_hdr		*ip6;
3171 	struct in6_ifaddr	*ia;
3172 	int			 mldreclen;
3173 
3174 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3175 	KASSERT((m->m_flags & M_PKTHDR),
3176 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3177 
3178 	/*
3179 	 * RFC3590: OK to send as :: or tentative during DAD.
3180 	 */
3181 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3182 	if (ia == NULL)
3183 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3184 
3185 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3186 	if (mh == NULL) {
3187 		if (ia != NULL)
3188 			ifa_free(&ia->ia_ifa);
3189 		m_freem(m);
3190 		return (NULL);
3191 	}
3192 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3193 
3194 	mldreclen = m_length(m, NULL);
3195 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3196 
3197 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3198 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3199 	    sizeof(struct mldv2_report) + mldreclen;
3200 
3201 	ip6 = mtod(mh, struct ip6_hdr *);
3202 	ip6->ip6_flow = 0;
3203 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3204 	ip6->ip6_vfc |= IPV6_VERSION;
3205 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3206 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3207 	if (ia != NULL)
3208 		ifa_free(&ia->ia_ifa);
3209 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3210 	/* scope ID will be set in netisr */
3211 
3212 	mld = (struct mldv2_report *)(ip6 + 1);
3213 	mld->mld_type = MLDV2_LISTENER_REPORT;
3214 	mld->mld_code = 0;
3215 	mld->mld_cksum = 0;
3216 	mld->mld_v2_reserved = 0;
3217 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3218 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3219 
3220 	mh->m_next = m;
3221 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3222 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3223 	return (mh);
3224 }
3225 
3226 #ifdef KTR
3227 static char *
3228 mld_rec_type_to_str(const int type)
3229 {
3230 
3231 	switch (type) {
3232 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3233 			return "TO_EX";
3234 			break;
3235 		case MLD_CHANGE_TO_INCLUDE_MODE:
3236 			return "TO_IN";
3237 			break;
3238 		case MLD_MODE_IS_EXCLUDE:
3239 			return "MODE_EX";
3240 			break;
3241 		case MLD_MODE_IS_INCLUDE:
3242 			return "MODE_IN";
3243 			break;
3244 		case MLD_ALLOW_NEW_SOURCES:
3245 			return "ALLOW_NEW";
3246 			break;
3247 		case MLD_BLOCK_OLD_SOURCES:
3248 			return "BLOCK_OLD";
3249 			break;
3250 		default:
3251 			break;
3252 	}
3253 	return "unknown";
3254 }
3255 #endif
3256 
3257 static void
3258 mld_init(void *unused __unused)
3259 {
3260 
3261 	CTR1(KTR_MLD, "%s: initializing", __func__);
3262 	MLD_LOCK_INIT();
3263 
3264 	ip6_initpktopts(&mld_po);
3265 	mld_po.ip6po_hlim = 1;
3266 	mld_po.ip6po_hbh = &mld_ra.hbh;
3267 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3268 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3269 }
3270 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3271 
3272 static void
3273 mld_uninit(void *unused __unused)
3274 {
3275 
3276 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3277 	MLD_LOCK_DESTROY();
3278 }
3279 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3280 
3281 static void
3282 vnet_mld_init(const void *unused __unused)
3283 {
3284 
3285 	CTR1(KTR_MLD, "%s: initializing", __func__);
3286 
3287 	LIST_INIT(&V_mli_head);
3288 }
3289 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3290     NULL);
3291 
3292 static void
3293 vnet_mld_uninit(const void *unused __unused)
3294 {
3295 
3296 	/* This can happen if we shutdown the network stack. */
3297 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3298 }
3299 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3300     NULL);
3301 
3302 static int
3303 mld_modevent(module_t mod, int type, void *unused __unused)
3304 {
3305 
3306     switch (type) {
3307     case MOD_LOAD:
3308     case MOD_UNLOAD:
3309 	break;
3310     default:
3311 	return (EOPNOTSUPP);
3312     }
3313     return (0);
3314 }
3315 
3316 static moduledata_t mld_mod = {
3317     "mld",
3318     mld_modevent,
3319     0
3320 };
3321 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3322