xref: /freebsd/sys/netinet6/mld6.c (revision a0409676120c1e558d0ade943019934e0f15118d)
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(struct in6_multi_head *inmh);
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 *, struct mldv2_query *, 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 *,
150 		    struct mldv2_query *, const int);
151 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
152 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
153 
154 /*
155  * Normative references: RFC 2710, RFC 3590, RFC 3810.
156  *
157  * Locking:
158  *  * The MLD subsystem lock ends up being system-wide for the moment,
159  *    but could be per-VIMAGE later on.
160  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
161  *    Any may be taken independently; if any are held at the same
162  *    time, the above lock order must be followed.
163  *  * IN6_MULTI_LOCK covers in_multi.
164  *  * MLD_LOCK covers per-link state and any global variables in this file.
165  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
166  *    per-link state iterators.
167  *
168  *  XXX LOR PREVENTION
169  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
170  *  will not accept an ifp; it wants an embedded scope ID, unlike
171  *  ip_output(), which happily takes the ifp given to it. The embedded
172  *  scope ID is only used by MLD to select the outgoing interface.
173  *
174  *  During interface attach and detach, MLD will take MLD_LOCK *after*
175  *  the IF_AFDATA_LOCK.
176  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
177  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
178  *  dispatch could work around this, but we'd rather not do that, as it
179  *  can introduce other races.
180  *
181  *  As such, we exploit the fact that the scope ID is just the interface
182  *  index, and embed it in the IPv6 destination address accordingly.
183  *  This is potentially NOT VALID for MLDv1 reports, as they
184  *  are always sent to the multicast group itself; as MLDv2
185  *  reports are always sent to ff02::16, this is not an issue
186  *  when MLDv2 is in use.
187  *
188  *  This does not however eliminate the LOR when ip6_output() itself
189  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
190  *  trigger a LOR warning in WITNESS when the ifnet is detached.
191  *
192  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
193  *  how it's used across the network stack. Here we're simply exploiting
194  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
195  *
196  * VIMAGE:
197  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
198  *    to a vnet in ifp->if_vnet.
199  */
200 static struct mtx		 mld_mtx;
201 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
202 
203 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
204 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
205 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
206 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
207 
208 /*
209  * VIMAGE-wide globals.
210  */
211 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
212 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
213 VNET_DEFINE_STATIC(int, interface_timers_running6);
214 VNET_DEFINE_STATIC(int, state_change_timers_running6);
215 VNET_DEFINE_STATIC(int, current_state_timers_running6);
216 
217 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
218 #define	V_mli_head			VNET(mli_head)
219 #define	V_interface_timers_running6	VNET(interface_timers_running6)
220 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
221 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
222 
223 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
224 
225 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
226     "IPv6 Multicast Listener Discovery");
227 
228 /*
229  * Virtualized sysctls.
230  */
231 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
232     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
233     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
234     "Rate limit for MLDv2 Group-and-Source queries in seconds");
235 
236 /*
237  * Non-virtualized sysctls.
238  */
239 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
240     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
241     "Per-interface MLDv2 state");
242 
243 static int	mld_v1enable = 1;
244 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
245     &mld_v1enable, 0, "Enable fallback to MLDv1");
246 
247 static int	mld_v2enable = 1;
248 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
249     &mld_v2enable, 0, "Enable MLDv2");
250 
251 static int	mld_use_allow = 1;
252 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
253     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
254 
255 /*
256  * Packed Router Alert option structure declaration.
257  */
258 struct mld_raopt {
259 	struct ip6_hbh		hbh;
260 	struct ip6_opt		pad;
261 	struct ip6_opt_router	ra;
262 } __packed;
263 
264 /*
265  * Router Alert hop-by-hop option header.
266  */
267 static struct mld_raopt mld_ra = {
268 	.hbh = { 0, 0 },
269 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
270 	.ra = {
271 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
272 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
273 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
274 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
275 	}
276 };
277 static struct ip6_pktopts mld_po;
278 
279 static __inline void
280 mld_save_context(struct mbuf *m, struct ifnet *ifp)
281 {
282 
283 #ifdef VIMAGE
284 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
285 #endif /* VIMAGE */
286 	m->m_pkthdr.rcvif = ifp;
287 	m->m_pkthdr.flowid = ifp->if_index;
288 }
289 
290 static __inline void
291 mld_scrub_context(struct mbuf *m)
292 {
293 
294 	m->m_pkthdr.PH_loc.ptr = NULL;
295 	m->m_pkthdr.flowid = 0;
296 }
297 
298 /*
299  * Restore context from a queued output chain.
300  * Return saved ifindex.
301  *
302  * VIMAGE: The assertion is there to make sure that we
303  * actually called CURVNET_SET() with what's in the mbuf chain.
304  */
305 static __inline uint32_t
306 mld_restore_context(struct mbuf *m)
307 {
308 
309 #if defined(VIMAGE) && defined(INVARIANTS)
310 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
311 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
312 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
313 #endif
314 	return (m->m_pkthdr.flowid);
315 }
316 
317 /*
318  * Retrieve or set threshold between group-source queries in seconds.
319  *
320  * VIMAGE: Assume curvnet set by caller.
321  * SMPng: NOTE: Serialized by MLD lock.
322  */
323 static int
324 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
325 {
326 	int error;
327 	int i;
328 
329 	error = sysctl_wire_old_buffer(req, sizeof(int));
330 	if (error)
331 		return (error);
332 
333 	MLD_LOCK();
334 
335 	i = V_mld_gsrdelay.tv_sec;
336 
337 	error = sysctl_handle_int(oidp, &i, 0, req);
338 	if (error || !req->newptr)
339 		goto out_locked;
340 
341 	if (i < -1 || i >= 60) {
342 		error = EINVAL;
343 		goto out_locked;
344 	}
345 
346 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
347 	     V_mld_gsrdelay.tv_sec, i);
348 	V_mld_gsrdelay.tv_sec = i;
349 
350 out_locked:
351 	MLD_UNLOCK();
352 	return (error);
353 }
354 
355 /*
356  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
357  * For use by ifmcstat(8).
358  *
359  * SMPng: NOTE: Does an unlocked ifindex space read.
360  * VIMAGE: Assume curvnet set by caller. The node handler itself
361  * is not directly virtualized.
362  */
363 static int
364 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
365 {
366 	int			*name;
367 	int			 error;
368 	u_int			 namelen;
369 	struct ifnet		*ifp;
370 	struct mld_ifsoftc	*mli;
371 
372 	name = (int *)arg1;
373 	namelen = arg2;
374 
375 	if (req->newptr != NULL)
376 		return (EPERM);
377 
378 	if (namelen != 1)
379 		return (EINVAL);
380 
381 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
382 	if (error)
383 		return (error);
384 
385 	IN6_MULTI_LOCK();
386 	IN6_MULTI_LIST_LOCK();
387 	MLD_LOCK();
388 
389 	if (name[0] <= 0 || name[0] > V_if_index) {
390 		error = ENOENT;
391 		goto out_locked;
392 	}
393 
394 	error = ENOENT;
395 
396 	ifp = ifnet_byindex(name[0]);
397 	if (ifp == NULL)
398 		goto out_locked;
399 
400 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
401 		if (ifp == mli->mli_ifp) {
402 			struct mld_ifinfo info;
403 
404 			info.mli_version = mli->mli_version;
405 			info.mli_v1_timer = mli->mli_v1_timer;
406 			info.mli_v2_timer = mli->mli_v2_timer;
407 			info.mli_flags = mli->mli_flags;
408 			info.mli_rv = mli->mli_rv;
409 			info.mli_qi = mli->mli_qi;
410 			info.mli_qri = mli->mli_qri;
411 			info.mli_uri = mli->mli_uri;
412 			error = SYSCTL_OUT(req, &info, sizeof(info));
413 			break;
414 		}
415 	}
416 
417 out_locked:
418 	MLD_UNLOCK();
419 	IN6_MULTI_LIST_UNLOCK();
420 	IN6_MULTI_UNLOCK();
421 	return (error);
422 }
423 
424 /*
425  * Dispatch an entire queue of pending packet chains.
426  * VIMAGE: Assumes the vnet pointer has been set.
427  */
428 static void
429 mld_dispatch_queue(struct mbufq *mq, int limit)
430 {
431 	struct mbuf *m;
432 
433 	while ((m = mbufq_dequeue(mq)) != NULL) {
434 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
435 		mld_dispatch_packet(m);
436 		if (--limit == 0)
437 			break;
438 	}
439 }
440 
441 /*
442  * Filter outgoing MLD report state by group.
443  *
444  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
445  * and node-local addresses. However, kernel and socket consumers
446  * always embed the KAME scope ID in the address provided, so strip it
447  * when performing comparison.
448  * Note: This is not the same as the *multicast* scope.
449  *
450  * Return zero if the given group is one for which MLD reports
451  * should be suppressed, or non-zero if reports should be issued.
452  */
453 static __inline int
454 mld_is_addr_reported(const struct in6_addr *addr)
455 {
456 
457 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
458 
459 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
460 		return (0);
461 
462 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
463 		struct in6_addr tmp = *addr;
464 		in6_clearscope(&tmp);
465 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
466 			return (0);
467 	}
468 
469 	return (1);
470 }
471 
472 /*
473  * Attach MLD when PF_INET6 is attached to an interface.
474  *
475  * SMPng: Normally called with IF_AFDATA_LOCK held.
476  */
477 struct mld_ifsoftc *
478 mld_domifattach(struct ifnet *ifp)
479 {
480 	struct mld_ifsoftc *mli;
481 
482 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
483 	    __func__, ifp, if_name(ifp));
484 
485 	MLD_LOCK();
486 
487 	mli = mli_alloc_locked(ifp);
488 	if (!(ifp->if_flags & IFF_MULTICAST))
489 		mli->mli_flags |= MLIF_SILENT;
490 	if (mld_use_allow)
491 		mli->mli_flags |= MLIF_USEALLOW;
492 
493 	MLD_UNLOCK();
494 
495 	return (mli);
496 }
497 
498 /*
499  * VIMAGE: assume curvnet set by caller.
500  */
501 static struct mld_ifsoftc *
502 mli_alloc_locked(/*const*/ struct ifnet *ifp)
503 {
504 	struct mld_ifsoftc *mli;
505 
506 	MLD_LOCK_ASSERT();
507 
508 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
509 	if (mli == NULL)
510 		goto out;
511 
512 	mli->mli_ifp = ifp;
513 	mli->mli_version = MLD_VERSION_2;
514 	mli->mli_flags = 0;
515 	mli->mli_rv = MLD_RV_INIT;
516 	mli->mli_qi = MLD_QI_INIT;
517 	mli->mli_qri = MLD_QRI_INIT;
518 	mli->mli_uri = MLD_URI_INIT;
519 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
520 
521 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
522 
523 	CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
524 	     ifp, if_name(ifp));
525 
526 out:
527 	return (mli);
528 }
529 
530 /*
531  * Hook for ifdetach.
532  *
533  * NOTE: Some finalization tasks need to run before the protocol domain
534  * is detached, but also before the link layer does its cleanup.
535  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
536  *
537  * SMPng: Caller must hold IN6_MULTI_LOCK().
538  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
539  * XXX This routine is also bitten by unlocked ifma_protospec access.
540  */
541 void
542 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
543 {
544 	struct epoch_tracker     et;
545 	struct mld_ifsoftc	*mli;
546 	struct ifmultiaddr	*ifma;
547 	struct in6_multi	*inm;
548 
549 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
550 	    if_name(ifp));
551 
552 	IN6_MULTI_LIST_LOCK_ASSERT();
553 	MLD_LOCK();
554 
555 	mli = MLD_IFINFO(ifp);
556 	IF_ADDR_WLOCK(ifp);
557 	/*
558 	 * Extract list of in6_multi associated with the detaching ifp
559 	 * which the PF_INET6 layer is about to release.
560 	 */
561 	NET_EPOCH_ENTER(et);
562 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
563 		inm = in6m_ifmultiaddr_get_inm(ifma);
564 		if (inm == NULL)
565 			continue;
566 		in6m_disconnect_locked(inmh, inm);
567 
568 		if (mli->mli_version == MLD_VERSION_2) {
569 			in6m_clear_recorded(inm);
570 
571 			/*
572 			 * We need to release the final reference held
573 			 * for issuing the INCLUDE {}.
574 			 */
575 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
576 				inm->in6m_state = MLD_NOT_MEMBER;
577 				in6m_rele_locked(inmh, inm);
578 			}
579 		}
580 	}
581 	NET_EPOCH_EXIT(et);
582 	IF_ADDR_WUNLOCK(ifp);
583 	MLD_UNLOCK();
584 }
585 
586 /*
587  * Hook for domifdetach.
588  * Runs after link-layer cleanup; free MLD state.
589  *
590  * SMPng: Normally called with IF_AFDATA_LOCK held.
591  */
592 void
593 mld_domifdetach(struct ifnet *ifp)
594 {
595 
596 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
597 	    __func__, ifp, if_name(ifp));
598 
599 	MLD_LOCK();
600 	mli_delete_locked(ifp);
601 	MLD_UNLOCK();
602 }
603 
604 static void
605 mli_delete_locked(const struct ifnet *ifp)
606 {
607 	struct mld_ifsoftc *mli, *tmli;
608 
609 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
610 	    __func__, ifp, if_name(ifp));
611 
612 	MLD_LOCK_ASSERT();
613 
614 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
615 		if (mli->mli_ifp == ifp) {
616 			/*
617 			 * Free deferred General Query responses.
618 			 */
619 			mbufq_drain(&mli->mli_gq);
620 
621 			LIST_REMOVE(mli, mli_link);
622 
623 			free(mli, M_MLD);
624 			return;
625 		}
626 	}
627 }
628 
629 /*
630  * Process a received MLDv1 general or address-specific query.
631  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
632  *
633  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
634  * mld_addr. This is OK as we own the mbuf chain.
635  */
636 static int
637 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
638     /*const*/ struct mld_hdr *mld)
639 {
640 	struct ifmultiaddr	*ifma;
641 	struct mld_ifsoftc	*mli;
642 	struct in6_multi	*inm;
643 	int			 is_general_query;
644 	uint16_t		 timer;
645 #ifdef KTR
646 	char			 ip6tbuf[INET6_ADDRSTRLEN];
647 #endif
648 
649 	NET_EPOCH_ASSERT();
650 
651 	is_general_query = 0;
652 
653 	if (!mld_v1enable) {
654 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
655 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
656 		    ifp, if_name(ifp));
657 		return (0);
658 	}
659 
660 	/*
661 	 * RFC3810 Section 6.2: MLD queries must originate from
662 	 * a router's link-local address.
663 	 */
664 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
665 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
666 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
667 		    ifp, if_name(ifp));
668 		return (0);
669 	}
670 
671 	/*
672 	 * Do address field validation upfront before we accept
673 	 * the query.
674 	 */
675 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
676 		/*
677 		 * MLDv1 General Query.
678 		 * If this was not sent to the all-nodes group, ignore it.
679 		 */
680 		struct in6_addr		 dst;
681 
682 		dst = ip6->ip6_dst;
683 		in6_clearscope(&dst);
684 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
685 			return (EINVAL);
686 		is_general_query = 1;
687 	} else {
688 		/*
689 		 * Embed scope ID of receiving interface in MLD query for
690 		 * lookup whilst we don't hold other locks.
691 		 */
692 		in6_setscope(&mld->mld_addr, ifp, NULL);
693 	}
694 
695 	IN6_MULTI_LIST_LOCK();
696 	MLD_LOCK();
697 
698 	/*
699 	 * Switch to MLDv1 host compatibility mode.
700 	 */
701 	mli = MLD_IFINFO(ifp);
702 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
703 	mld_set_version(mli, MLD_VERSION_1);
704 
705 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
706 	if (timer == 0)
707 		timer = 1;
708 
709 	if (is_general_query) {
710 		/*
711 		 * For each reporting group joined on this
712 		 * interface, kick the report timer.
713 		 */
714 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
715 			 ifp, if_name(ifp));
716 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
717 			inm = in6m_ifmultiaddr_get_inm(ifma);
718 			if (inm == NULL)
719 				continue;
720 			mld_v1_update_group(inm, timer);
721 		}
722 	} else {
723 		/*
724 		 * MLDv1 Group-Specific Query.
725 		 * If this is a group-specific MLDv1 query, we need only
726 		 * look up the single group to process it.
727 		 */
728 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
729 		if (inm != NULL) {
730 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
731 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
732 			    ifp, if_name(ifp));
733 			mld_v1_update_group(inm, timer);
734 		}
735 		/* XXX Clear embedded scope ID as userland won't expect it. */
736 		in6_clearscope(&mld->mld_addr);
737 	}
738 
739 	MLD_UNLOCK();
740 	IN6_MULTI_LIST_UNLOCK();
741 
742 	return (0);
743 }
744 
745 /*
746  * Update the report timer on a group in response to an MLDv1 query.
747  *
748  * If we are becoming the reporting member for this group, start the timer.
749  * If we already are the reporting member for this group, and timer is
750  * below the threshold, reset it.
751  *
752  * We may be updating the group for the first time since we switched
753  * to MLDv2. If we are, then we must clear any recorded source lists,
754  * and transition to REPORTING state; the group timer is overloaded
755  * for group and group-source query responses.
756  *
757  * Unlike MLDv2, the delay per group should be jittered
758  * to avoid bursts of MLDv1 reports.
759  */
760 static void
761 mld_v1_update_group(struct in6_multi *inm, const int timer)
762 {
763 #ifdef KTR
764 	char			 ip6tbuf[INET6_ADDRSTRLEN];
765 #endif
766 
767 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
768 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
769 	    if_name(inm->in6m_ifp), timer);
770 
771 	IN6_MULTI_LIST_LOCK_ASSERT();
772 
773 	switch (inm->in6m_state) {
774 	case MLD_NOT_MEMBER:
775 	case MLD_SILENT_MEMBER:
776 		break;
777 	case MLD_REPORTING_MEMBER:
778 		if (inm->in6m_timer != 0 &&
779 		    inm->in6m_timer <= timer) {
780 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
781 			    "skipping.", __func__);
782 			break;
783 		}
784 		/* FALLTHROUGH */
785 	case MLD_SG_QUERY_PENDING_MEMBER:
786 	case MLD_G_QUERY_PENDING_MEMBER:
787 	case MLD_IDLE_MEMBER:
788 	case MLD_LAZY_MEMBER:
789 	case MLD_AWAKENING_MEMBER:
790 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
791 		inm->in6m_state = MLD_REPORTING_MEMBER;
792 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
793 		V_current_state_timers_running6 = 1;
794 		break;
795 	case MLD_SLEEPING_MEMBER:
796 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
797 		inm->in6m_state = MLD_AWAKENING_MEMBER;
798 		break;
799 	case MLD_LEAVING_MEMBER:
800 		break;
801 	}
802 }
803 
804 /*
805  * Process a received MLDv2 general, group-specific or
806  * group-and-source-specific query.
807  *
808  * Assumes that mld points to a struct mldv2_query which is stored in
809  * contiguous memory.
810  *
811  * Return 0 if successful, otherwise an appropriate error code is returned.
812  */
813 static int
814 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
815     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
816 {
817 	struct mld_ifsoftc	*mli;
818 	struct in6_multi	*inm;
819 	uint32_t		 maxdelay, nsrc, qqi;
820 	int			 is_general_query;
821 	uint16_t		 timer;
822 	uint8_t			 qrv;
823 #ifdef KTR
824 	char			 ip6tbuf[INET6_ADDRSTRLEN];
825 #endif
826 
827 	NET_EPOCH_ASSERT();
828 
829 	if (!mld_v2enable) {
830 		CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
831 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
832 		    ifp, if_name(ifp));
833 		return (0);
834 	}
835 
836 	/*
837 	 * RFC3810 Section 6.2: MLD queries must originate from
838 	 * a router's link-local address.
839 	 */
840 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
841 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
842 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
843 		    ifp, if_name(ifp));
844 		return (0);
845 	}
846 
847 	is_general_query = 0;
848 
849 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
850 
851 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
852 	if (maxdelay >= 32768) {
853 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
854 			   (MLD_MRC_EXP(maxdelay) + 3);
855 	}
856 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
857 	if (timer == 0)
858 		timer = 1;
859 
860 	qrv = MLD_QRV(mld->mld_misc);
861 	if (qrv < 2) {
862 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
863 		    qrv, MLD_RV_INIT);
864 		qrv = MLD_RV_INIT;
865 	}
866 
867 	qqi = mld->mld_qqi;
868 	if (qqi >= 128) {
869 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
870 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
871 	}
872 
873 	nsrc = ntohs(mld->mld_numsrc);
874 	if (nsrc > MLD_MAX_GS_SOURCES)
875 		return (EMSGSIZE);
876 	if (icmp6len < sizeof(struct mldv2_query) +
877 	    (nsrc * sizeof(struct in6_addr)))
878 		return (EMSGSIZE);
879 
880 	/*
881 	 * Do further input validation upfront to avoid resetting timers
882 	 * should we need to discard this query.
883 	 */
884 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
885 		/*
886 		 * A general query with a source list has undefined
887 		 * behaviour; discard it.
888 		 */
889 		if (nsrc > 0)
890 			return (EINVAL);
891 		is_general_query = 1;
892 	} else {
893 		/*
894 		 * Embed scope ID of receiving interface in MLD query for
895 		 * lookup whilst we don't hold other locks (due to KAME
896 		 * locking lameness). We own this mbuf chain just now.
897 		 */
898 		in6_setscope(&mld->mld_addr, ifp, NULL);
899 	}
900 
901 	IN6_MULTI_LIST_LOCK();
902 	MLD_LOCK();
903 
904 	mli = MLD_IFINFO(ifp);
905 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
906 
907 	/*
908 	 * Discard the v2 query if we're in Compatibility Mode.
909 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
910 	 * until the Old Version Querier Present timer expires.
911 	 */
912 	if (mli->mli_version != MLD_VERSION_2)
913 		goto out_locked;
914 
915 	mld_set_version(mli, MLD_VERSION_2);
916 	mli->mli_rv = qrv;
917 	mli->mli_qi = qqi;
918 	mli->mli_qri = maxdelay;
919 
920 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
921 	    maxdelay);
922 
923 	if (is_general_query) {
924 		/*
925 		 * MLDv2 General Query.
926 		 *
927 		 * Schedule a current-state report on this ifp for
928 		 * all groups, possibly containing source lists.
929 		 *
930 		 * If there is a pending General Query response
931 		 * scheduled earlier than the selected delay, do
932 		 * not schedule any other reports.
933 		 * Otherwise, reset the interface timer.
934 		 */
935 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
936 		    ifp, if_name(ifp));
937 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
938 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
939 			V_interface_timers_running6 = 1;
940 		}
941 	} else {
942 		/*
943 		 * MLDv2 Group-specific or Group-and-source-specific Query.
944 		 *
945 		 * Group-source-specific queries are throttled on
946 		 * a per-group basis to defeat denial-of-service attempts.
947 		 * Queries for groups we are not a member of on this
948 		 * link are simply ignored.
949 		 */
950 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
951 		if (inm == NULL)
952 			goto out_locked;
953 		if (nsrc > 0) {
954 			if (!ratecheck(&inm->in6m_lastgsrtv,
955 			    &V_mld_gsrdelay)) {
956 				CTR1(KTR_MLD, "%s: GS query throttled.",
957 				    __func__);
958 				goto out_locked;
959 			}
960 		}
961 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
962 		     ifp, if_name(ifp));
963 		/*
964 		 * If there is a pending General Query response
965 		 * scheduled sooner than the selected delay, no
966 		 * further report need be scheduled.
967 		 * Otherwise, prepare to respond to the
968 		 * group-specific or group-and-source query.
969 		 */
970 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
971 			mld_v2_process_group_query(inm, mli, timer, m, mld, off);
972 
973 		/* XXX Clear embedded scope ID as userland won't expect it. */
974 		in6_clearscope(&mld->mld_addr);
975 	}
976 
977 out_locked:
978 	MLD_UNLOCK();
979 	IN6_MULTI_LIST_UNLOCK();
980 
981 	return (0);
982 }
983 
984 /*
985  * Process a received MLDv2 group-specific or group-and-source-specific
986  * query.
987  * Return <0 if any error occurred. Currently this is ignored.
988  */
989 static int
990 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
991     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
992 {
993 	int			 retval;
994 	uint16_t		 nsrc;
995 
996 	IN6_MULTI_LIST_LOCK_ASSERT();
997 	MLD_LOCK_ASSERT();
998 
999 	retval = 0;
1000 
1001 	switch (inm->in6m_state) {
1002 	case MLD_NOT_MEMBER:
1003 	case MLD_SILENT_MEMBER:
1004 	case MLD_SLEEPING_MEMBER:
1005 	case MLD_LAZY_MEMBER:
1006 	case MLD_AWAKENING_MEMBER:
1007 	case MLD_IDLE_MEMBER:
1008 	case MLD_LEAVING_MEMBER:
1009 		return (retval);
1010 		break;
1011 	case MLD_REPORTING_MEMBER:
1012 	case MLD_G_QUERY_PENDING_MEMBER:
1013 	case MLD_SG_QUERY_PENDING_MEMBER:
1014 		break;
1015 	}
1016 
1017 	nsrc = ntohs(mld->mld_numsrc);
1018 
1019 	/* Length should be checked by calling function. */
1020 	KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
1021 	    m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
1022 	    nsrc * sizeof(struct in6_addr),
1023 	    ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
1024 	    m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
1025 	    nsrc * sizeof(struct in6_addr), m0));
1026 
1027 	/*
1028 	 * Deal with group-specific queries upfront.
1029 	 * If any group query is already pending, purge any recorded
1030 	 * source-list state if it exists, and schedule a query response
1031 	 * for this group-specific query.
1032 	 */
1033 	if (nsrc == 0) {
1034 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1035 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1036 			in6m_clear_recorded(inm);
1037 			timer = min(inm->in6m_timer, timer);
1038 		}
1039 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1040 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1041 		V_current_state_timers_running6 = 1;
1042 		return (retval);
1043 	}
1044 
1045 	/*
1046 	 * Deal with the case where a group-and-source-specific query has
1047 	 * been received but a group-specific query is already pending.
1048 	 */
1049 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1050 		timer = min(inm->in6m_timer, timer);
1051 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1052 		V_current_state_timers_running6 = 1;
1053 		return (retval);
1054 	}
1055 
1056 	/*
1057 	 * Finally, deal with the case where a group-and-source-specific
1058 	 * query has been received, where a response to a previous g-s-r
1059 	 * query exists, or none exists.
1060 	 * In this case, we need to parse the source-list which the Querier
1061 	 * has provided us with and check if we have any source list filter
1062 	 * entries at T1 for these sources. If we do not, there is no need
1063 	 * schedule a report and the query may be dropped.
1064 	 * If we do, we must record them and schedule a current-state
1065 	 * report for those sources.
1066 	 */
1067 	if (inm->in6m_nsrc > 0) {
1068 		struct in6_addr		 srcaddr;
1069 		int			 i, nrecorded;
1070 		int			 soff;
1071 
1072 		soff = off + sizeof(struct mldv2_query);
1073 		nrecorded = 0;
1074 		for (i = 0; i < nsrc; i++) {
1075 			m_copydata(m0, soff, sizeof(struct in6_addr),
1076 			    (caddr_t)&srcaddr);
1077 			retval = in6m_record_source(inm, &srcaddr);
1078 			if (retval < 0)
1079 				break;
1080 			nrecorded += retval;
1081 			soff += sizeof(struct in6_addr);
1082 		}
1083 		if (nrecorded > 0) {
1084 			CTR1(KTR_MLD,
1085 			    "%s: schedule response to SG query", __func__);
1086 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1087 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1088 			V_current_state_timers_running6 = 1;
1089 		}
1090 	}
1091 
1092 	return (retval);
1093 }
1094 
1095 /*
1096  * Process a received MLDv1 host membership report.
1097  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1098  *
1099  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1100  * mld_addr. This is OK as we own the mbuf chain.
1101  */
1102 static int
1103 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1104     /*const*/ struct mld_hdr *mld)
1105 {
1106 	struct in6_addr		 src, dst;
1107 	struct in6_ifaddr	*ia;
1108 	struct in6_multi	*inm;
1109 #ifdef KTR
1110 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1111 #endif
1112 
1113 	NET_EPOCH_ASSERT();
1114 
1115 	if (!mld_v1enable) {
1116 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1117 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1118 		    ifp, if_name(ifp));
1119 		return (0);
1120 	}
1121 
1122 	if (ifp->if_flags & IFF_LOOPBACK)
1123 		return (0);
1124 
1125 	/*
1126 	 * MLDv1 reports must originate from a host's link-local address,
1127 	 * or the unspecified address (when booting).
1128 	 */
1129 	src = ip6->ip6_src;
1130 	in6_clearscope(&src);
1131 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1132 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1133 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1134 		    ifp, if_name(ifp));
1135 		return (EINVAL);
1136 	}
1137 
1138 	/*
1139 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1140 	 * group, and must be directed to the group itself.
1141 	 */
1142 	dst = ip6->ip6_dst;
1143 	in6_clearscope(&dst);
1144 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1145 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1146 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1147 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1148 		    ifp, if_name(ifp));
1149 		return (EINVAL);
1150 	}
1151 
1152 	/*
1153 	 * Make sure we don't hear our own membership report, as fast
1154 	 * leave requires knowing that we are the only member of a
1155 	 * group. Assume we used the link-local address if available,
1156 	 * otherwise look for ::.
1157 	 *
1158 	 * XXX Note that scope ID comparison is needed for the address
1159 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1160 	 * performed for the on-wire address.
1161 	 */
1162 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1163 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1164 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1165 		if (ia != NULL)
1166 			ifa_free(&ia->ia_ifa);
1167 		return (0);
1168 	}
1169 	if (ia != NULL)
1170 		ifa_free(&ia->ia_ifa);
1171 
1172 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1173 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1174 
1175 	/*
1176 	 * Embed scope ID of receiving interface in MLD query for lookup
1177 	 * whilst we don't hold other locks (due to KAME locking lameness).
1178 	 */
1179 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1180 		in6_setscope(&mld->mld_addr, ifp, NULL);
1181 
1182 	IN6_MULTI_LIST_LOCK();
1183 	MLD_LOCK();
1184 
1185 	/*
1186 	 * MLDv1 report suppression.
1187 	 * If we are a member of this group, and our membership should be
1188 	 * reported, and our group timer is pending or about to be reset,
1189 	 * stop our group timer by transitioning to the 'lazy' state.
1190 	 */
1191 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1192 	if (inm != NULL) {
1193 		struct mld_ifsoftc *mli;
1194 
1195 		mli = inm->in6m_mli;
1196 		KASSERT(mli != NULL,
1197 		    ("%s: no mli for ifp %p", __func__, ifp));
1198 
1199 		/*
1200 		 * If we are in MLDv2 host mode, do not allow the
1201 		 * other host's MLDv1 report to suppress our reports.
1202 		 */
1203 		if (mli->mli_version == MLD_VERSION_2)
1204 			goto out_locked;
1205 
1206 		inm->in6m_timer = 0;
1207 
1208 		switch (inm->in6m_state) {
1209 		case MLD_NOT_MEMBER:
1210 		case MLD_SILENT_MEMBER:
1211 		case MLD_SLEEPING_MEMBER:
1212 			break;
1213 		case MLD_REPORTING_MEMBER:
1214 		case MLD_IDLE_MEMBER:
1215 		case MLD_AWAKENING_MEMBER:
1216 			CTR3(KTR_MLD,
1217 			    "report suppressed for %s on ifp %p(%s)",
1218 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1219 			    ifp, if_name(ifp));
1220 		case MLD_LAZY_MEMBER:
1221 			inm->in6m_state = MLD_LAZY_MEMBER;
1222 			break;
1223 		case MLD_G_QUERY_PENDING_MEMBER:
1224 		case MLD_SG_QUERY_PENDING_MEMBER:
1225 		case MLD_LEAVING_MEMBER:
1226 			break;
1227 		}
1228 	}
1229 
1230 out_locked:
1231 	MLD_UNLOCK();
1232 	IN6_MULTI_LIST_UNLOCK();
1233 
1234 	/* XXX Clear embedded scope ID as userland won't expect it. */
1235 	in6_clearscope(&mld->mld_addr);
1236 
1237 	return (0);
1238 }
1239 
1240 /*
1241  * MLD input path.
1242  *
1243  * Assume query messages which fit in a single ICMPv6 message header
1244  * have been pulled up.
1245  * Assume that userland will want to see the message, even if it
1246  * otherwise fails kernel input validation; do not free it.
1247  * Pullup may however free the mbuf chain m if it fails.
1248  *
1249  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1250  */
1251 int
1252 mld_input(struct mbuf **mp, int off, int icmp6len)
1253 {
1254 	struct ifnet	*ifp;
1255 	struct ip6_hdr	*ip6;
1256 	struct mbuf	*m;
1257 	struct mld_hdr	*mld;
1258 	int		 mldlen;
1259 
1260 	m = *mp;
1261 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1262 
1263 	ifp = m->m_pkthdr.rcvif;
1264 
1265 	/* Pullup to appropriate size. */
1266 	if (m->m_len < off + sizeof(*mld)) {
1267 		m = m_pullup(m, off + sizeof(*mld));
1268 		if (m == NULL) {
1269 			ICMP6STAT_INC(icp6s_badlen);
1270 			return (IPPROTO_DONE);
1271 		}
1272 	}
1273 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1274 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1275 	    icmp6len >= sizeof(struct mldv2_query)) {
1276 		mldlen = sizeof(struct mldv2_query);
1277 	} else {
1278 		mldlen = sizeof(struct mld_hdr);
1279 	}
1280 	if (m->m_len < off + mldlen) {
1281 		m = m_pullup(m, off + mldlen);
1282 		if (m == NULL) {
1283 			ICMP6STAT_INC(icp6s_badlen);
1284 			return (IPPROTO_DONE);
1285 		}
1286 	}
1287 	*mp = m;
1288 	ip6 = mtod(m, struct ip6_hdr *);
1289 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1290 
1291 	/*
1292 	 * Userland needs to see all of this traffic for implementing
1293 	 * the endpoint discovery portion of multicast routing.
1294 	 */
1295 	switch (mld->mld_type) {
1296 	case MLD_LISTENER_QUERY:
1297 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1298 		if (icmp6len == sizeof(struct mld_hdr)) {
1299 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1300 				return (0);
1301 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1302 			if (mld_v2_input_query(ifp, ip6, m,
1303 			    (struct mldv2_query *)mld, off, icmp6len) != 0)
1304 				return (0);
1305 		}
1306 		break;
1307 	case MLD_LISTENER_REPORT:
1308 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1309 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1310 			return (0);
1311 		break;
1312 	case MLDV2_LISTENER_REPORT:
1313 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1314 		break;
1315 	case MLD_LISTENER_DONE:
1316 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1317 		break;
1318 	default:
1319 		break;
1320 	}
1321 
1322 	return (0);
1323 }
1324 
1325 /*
1326  * Fast timeout handler (global).
1327  * VIMAGE: Timeout handlers are expected to service all vimages.
1328  */
1329 void
1330 mld_fasttimo(void)
1331 {
1332 	struct in6_multi_head inmh;
1333 	VNET_ITERATOR_DECL(vnet_iter);
1334 
1335 	SLIST_INIT(&inmh);
1336 
1337 	VNET_LIST_RLOCK_NOSLEEP();
1338 	VNET_FOREACH(vnet_iter) {
1339 		CURVNET_SET(vnet_iter);
1340 		mld_fasttimo_vnet(&inmh);
1341 		CURVNET_RESTORE();
1342 	}
1343 	VNET_LIST_RUNLOCK_NOSLEEP();
1344 	in6m_release_list_deferred(&inmh);
1345 }
1346 
1347 /*
1348  * Fast timeout handler (per-vnet).
1349  *
1350  * VIMAGE: Assume caller has set up our curvnet.
1351  */
1352 static void
1353 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1354 {
1355 	struct epoch_tracker     et;
1356 	struct mbufq		 scq;	/* State-change packets */
1357 	struct mbufq		 qrq;	/* Query response packets */
1358 	struct ifnet		*ifp;
1359 	struct mld_ifsoftc	*mli;
1360 	struct ifmultiaddr	*ifma;
1361 	struct in6_multi	*inm;
1362 	int			 uri_fasthz;
1363 
1364 	uri_fasthz = 0;
1365 
1366 	/*
1367 	 * Quick check to see if any work needs to be done, in order to
1368 	 * minimize the overhead of fasttimo processing.
1369 	 * SMPng: XXX Unlocked reads.
1370 	 */
1371 	if (!V_current_state_timers_running6 &&
1372 	    !V_interface_timers_running6 &&
1373 	    !V_state_change_timers_running6)
1374 		return;
1375 
1376 	IN6_MULTI_LIST_LOCK();
1377 	MLD_LOCK();
1378 
1379 	/*
1380 	 * MLDv2 General Query response timer processing.
1381 	 */
1382 	if (V_interface_timers_running6) {
1383 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1384 
1385 		V_interface_timers_running6 = 0;
1386 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1387 			if (mli->mli_v2_timer == 0) {
1388 				/* Do nothing. */
1389 			} else if (--mli->mli_v2_timer == 0) {
1390 				mld_v2_dispatch_general_query(mli);
1391 			} else {
1392 				V_interface_timers_running6 = 1;
1393 			}
1394 		}
1395 	}
1396 
1397 	if (!V_current_state_timers_running6 &&
1398 	    !V_state_change_timers_running6)
1399 		goto out_locked;
1400 
1401 	V_current_state_timers_running6 = 0;
1402 	V_state_change_timers_running6 = 0;
1403 
1404 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1405 
1406 	/*
1407 	 * MLD host report and state-change timer processing.
1408 	 * Note: Processing a v2 group timer may remove a node.
1409 	 */
1410 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1411 		ifp = mli->mli_ifp;
1412 
1413 		if (mli->mli_version == MLD_VERSION_2) {
1414 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1415 			    PR_FASTHZ);
1416 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1417 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1418 		}
1419 
1420 		NET_EPOCH_ENTER(et);
1421 		IF_ADDR_WLOCK(ifp);
1422 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1423 			inm = in6m_ifmultiaddr_get_inm(ifma);
1424 			if (inm == NULL)
1425 				continue;
1426 			switch (mli->mli_version) {
1427 			case MLD_VERSION_1:
1428 				mld_v1_process_group_timer(inmh, inm);
1429 				break;
1430 			case MLD_VERSION_2:
1431 				mld_v2_process_group_timers(inmh, &qrq,
1432 				    &scq, inm, uri_fasthz);
1433 				break;
1434 			}
1435 		}
1436 		IF_ADDR_WUNLOCK(ifp);
1437 
1438 		switch (mli->mli_version) {
1439 		case MLD_VERSION_1:
1440 			/*
1441 			 * Transmit reports for this lifecycle.  This
1442 			 * is done while not holding IF_ADDR_LOCK
1443 			 * since this can call
1444 			 * in6ifa_ifpforlinklocal() which locks
1445 			 * IF_ADDR_LOCK internally as well as
1446 			 * ip6_output() to transmit a packet.
1447 			 */
1448 			while ((inm = SLIST_FIRST(inmh)) != NULL) {
1449 				SLIST_REMOVE_HEAD(inmh, in6m_defer);
1450 				(void)mld_v1_transmit_report(inm,
1451 				    MLD_LISTENER_REPORT);
1452 			}
1453 			break;
1454 		case MLD_VERSION_2:
1455 			mld_dispatch_queue(&qrq, 0);
1456 			mld_dispatch_queue(&scq, 0);
1457 			break;
1458 		}
1459 		NET_EPOCH_EXIT(et);
1460 	}
1461 
1462 out_locked:
1463 	MLD_UNLOCK();
1464 	IN6_MULTI_LIST_UNLOCK();
1465 }
1466 
1467 /*
1468  * Update host report group timer.
1469  * Will update the global pending timer flags.
1470  */
1471 static void
1472 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1473 {
1474 	int report_timer_expired;
1475 
1476 	IN6_MULTI_LIST_LOCK_ASSERT();
1477 	MLD_LOCK_ASSERT();
1478 
1479 	if (inm->in6m_timer == 0) {
1480 		report_timer_expired = 0;
1481 	} else if (--inm->in6m_timer == 0) {
1482 		report_timer_expired = 1;
1483 	} else {
1484 		V_current_state_timers_running6 = 1;
1485 		return;
1486 	}
1487 
1488 	switch (inm->in6m_state) {
1489 	case MLD_NOT_MEMBER:
1490 	case MLD_SILENT_MEMBER:
1491 	case MLD_IDLE_MEMBER:
1492 	case MLD_LAZY_MEMBER:
1493 	case MLD_SLEEPING_MEMBER:
1494 	case MLD_AWAKENING_MEMBER:
1495 		break;
1496 	case MLD_REPORTING_MEMBER:
1497 		if (report_timer_expired) {
1498 			inm->in6m_state = MLD_IDLE_MEMBER;
1499 			SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1500 		}
1501 		break;
1502 	case MLD_G_QUERY_PENDING_MEMBER:
1503 	case MLD_SG_QUERY_PENDING_MEMBER:
1504 	case MLD_LEAVING_MEMBER:
1505 		break;
1506 	}
1507 }
1508 
1509 /*
1510  * Update a group's timers for MLDv2.
1511  * Will update the global pending timer flags.
1512  * Note: Unlocked read from mli.
1513  */
1514 static void
1515 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1516     struct mbufq *qrq, struct mbufq *scq,
1517     struct in6_multi *inm, const int uri_fasthz)
1518 {
1519 	int query_response_timer_expired;
1520 	int state_change_retransmit_timer_expired;
1521 #ifdef KTR
1522 	char ip6tbuf[INET6_ADDRSTRLEN];
1523 #endif
1524 
1525 	IN6_MULTI_LIST_LOCK_ASSERT();
1526 	MLD_LOCK_ASSERT();
1527 
1528 	query_response_timer_expired = 0;
1529 	state_change_retransmit_timer_expired = 0;
1530 
1531 	/*
1532 	 * During a transition from compatibility mode back to MLDv2,
1533 	 * a group record in REPORTING state may still have its group
1534 	 * timer active. This is a no-op in this function; it is easier
1535 	 * to deal with it here than to complicate the slow-timeout path.
1536 	 */
1537 	if (inm->in6m_timer == 0) {
1538 		query_response_timer_expired = 0;
1539 	} else if (--inm->in6m_timer == 0) {
1540 		query_response_timer_expired = 1;
1541 	} else {
1542 		V_current_state_timers_running6 = 1;
1543 	}
1544 
1545 	if (inm->in6m_sctimer == 0) {
1546 		state_change_retransmit_timer_expired = 0;
1547 	} else if (--inm->in6m_sctimer == 0) {
1548 		state_change_retransmit_timer_expired = 1;
1549 	} else {
1550 		V_state_change_timers_running6 = 1;
1551 	}
1552 
1553 	/* We are in fasttimo, so be quick about it. */
1554 	if (!state_change_retransmit_timer_expired &&
1555 	    !query_response_timer_expired)
1556 		return;
1557 
1558 	switch (inm->in6m_state) {
1559 	case MLD_NOT_MEMBER:
1560 	case MLD_SILENT_MEMBER:
1561 	case MLD_SLEEPING_MEMBER:
1562 	case MLD_LAZY_MEMBER:
1563 	case MLD_AWAKENING_MEMBER:
1564 	case MLD_IDLE_MEMBER:
1565 		break;
1566 	case MLD_G_QUERY_PENDING_MEMBER:
1567 	case MLD_SG_QUERY_PENDING_MEMBER:
1568 		/*
1569 		 * Respond to a previously pending Group-Specific
1570 		 * or Group-and-Source-Specific query by enqueueing
1571 		 * the appropriate Current-State report for
1572 		 * immediate transmission.
1573 		 */
1574 		if (query_response_timer_expired) {
1575 			int retval;
1576 
1577 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1578 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1579 			    0);
1580 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1581 			    __func__, retval);
1582 			inm->in6m_state = MLD_REPORTING_MEMBER;
1583 			in6m_clear_recorded(inm);
1584 		}
1585 		/* FALLTHROUGH */
1586 	case MLD_REPORTING_MEMBER:
1587 	case MLD_LEAVING_MEMBER:
1588 		if (state_change_retransmit_timer_expired) {
1589 			/*
1590 			 * State-change retransmission timer fired.
1591 			 * If there are any further pending retransmissions,
1592 			 * set the global pending state-change flag, and
1593 			 * reset the timer.
1594 			 */
1595 			if (--inm->in6m_scrv > 0) {
1596 				inm->in6m_sctimer = uri_fasthz;
1597 				V_state_change_timers_running6 = 1;
1598 			}
1599 			/*
1600 			 * Retransmit the previously computed state-change
1601 			 * report. If there are no further pending
1602 			 * retransmissions, the mbuf queue will be consumed.
1603 			 * Update T0 state to T1 as we have now sent
1604 			 * a state-change.
1605 			 */
1606 			(void)mld_v2_merge_state_changes(inm, scq);
1607 
1608 			in6m_commit(inm);
1609 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1610 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1611 			    if_name(inm->in6m_ifp));
1612 
1613 			/*
1614 			 * If we are leaving the group for good, make sure
1615 			 * we release MLD's reference to it.
1616 			 * This release must be deferred using a SLIST,
1617 			 * as we are called from a loop which traverses
1618 			 * the in_ifmultiaddr TAILQ.
1619 			 */
1620 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1621 			    inm->in6m_scrv == 0) {
1622 				inm->in6m_state = MLD_NOT_MEMBER;
1623 				in6m_disconnect_locked(inmh, inm);
1624 				in6m_rele_locked(inmh, inm);
1625 			}
1626 		}
1627 		break;
1628 	}
1629 }
1630 
1631 /*
1632  * Switch to a different version on the given interface,
1633  * as per Section 9.12.
1634  */
1635 static void
1636 mld_set_version(struct mld_ifsoftc *mli, const int version)
1637 {
1638 	int old_version_timer;
1639 
1640 	MLD_LOCK_ASSERT();
1641 
1642 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1643 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1644 
1645 	if (version == MLD_VERSION_1) {
1646 		/*
1647 		 * Compute the "Older Version Querier Present" timer as per
1648 		 * Section 9.12.
1649 		 */
1650 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1651 		old_version_timer *= PR_SLOWHZ;
1652 		mli->mli_v1_timer = old_version_timer;
1653 	}
1654 
1655 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1656 		mli->mli_version = MLD_VERSION_1;
1657 		mld_v2_cancel_link_timers(mli);
1658 	}
1659 }
1660 
1661 /*
1662  * Cancel pending MLDv2 timers for the given link and all groups
1663  * joined on it; state-change, general-query, and group-query timers.
1664  */
1665 static void
1666 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1667 {
1668 	struct epoch_tracker	 et;
1669 	struct in6_multi_head	 inmh;
1670 	struct ifmultiaddr	*ifma;
1671 	struct ifnet		*ifp;
1672 	struct in6_multi	*inm;
1673 
1674 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1675 	    mli->mli_ifp, if_name(mli->mli_ifp));
1676 
1677 	SLIST_INIT(&inmh);
1678 	IN6_MULTI_LIST_LOCK_ASSERT();
1679 	MLD_LOCK_ASSERT();
1680 
1681 	/*
1682 	 * Fast-track this potentially expensive operation
1683 	 * by checking all the global 'timer pending' flags.
1684 	 */
1685 	if (!V_interface_timers_running6 &&
1686 	    !V_state_change_timers_running6 &&
1687 	    !V_current_state_timers_running6)
1688 		return;
1689 
1690 	mli->mli_v2_timer = 0;
1691 
1692 	ifp = mli->mli_ifp;
1693 
1694 	IF_ADDR_WLOCK(ifp);
1695 	NET_EPOCH_ENTER(et);
1696 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1697 		inm = in6m_ifmultiaddr_get_inm(ifma);
1698 		if (inm == NULL)
1699 			continue;
1700 		switch (inm->in6m_state) {
1701 		case MLD_NOT_MEMBER:
1702 		case MLD_SILENT_MEMBER:
1703 		case MLD_IDLE_MEMBER:
1704 		case MLD_LAZY_MEMBER:
1705 		case MLD_SLEEPING_MEMBER:
1706 		case MLD_AWAKENING_MEMBER:
1707 			break;
1708 		case MLD_LEAVING_MEMBER:
1709 			/*
1710 			 * If we are leaving the group and switching
1711 			 * version, we need to release the final
1712 			 * reference held for issuing the INCLUDE {}.
1713 			 */
1714 			if (inm->in6m_refcount == 1)
1715 				in6m_disconnect_locked(&inmh, inm);
1716 			in6m_rele_locked(&inmh, inm);
1717 			/* FALLTHROUGH */
1718 		case MLD_G_QUERY_PENDING_MEMBER:
1719 		case MLD_SG_QUERY_PENDING_MEMBER:
1720 			in6m_clear_recorded(inm);
1721 			/* FALLTHROUGH */
1722 		case MLD_REPORTING_MEMBER:
1723 			inm->in6m_sctimer = 0;
1724 			inm->in6m_timer = 0;
1725 			inm->in6m_state = MLD_REPORTING_MEMBER;
1726 			/*
1727 			 * Free any pending MLDv2 state-change records.
1728 			 */
1729 			mbufq_drain(&inm->in6m_scq);
1730 			break;
1731 		}
1732 	}
1733 	NET_EPOCH_EXIT(et);
1734 	IF_ADDR_WUNLOCK(ifp);
1735 	in6m_release_list_deferred(&inmh);
1736 }
1737 
1738 /*
1739  * Global slowtimo handler.
1740  * VIMAGE: Timeout handlers are expected to service all vimages.
1741  */
1742 void
1743 mld_slowtimo(void)
1744 {
1745 	VNET_ITERATOR_DECL(vnet_iter);
1746 
1747 	VNET_LIST_RLOCK_NOSLEEP();
1748 	VNET_FOREACH(vnet_iter) {
1749 		CURVNET_SET(vnet_iter);
1750 		mld_slowtimo_vnet();
1751 		CURVNET_RESTORE();
1752 	}
1753 	VNET_LIST_RUNLOCK_NOSLEEP();
1754 }
1755 
1756 /*
1757  * Per-vnet slowtimo handler.
1758  */
1759 static void
1760 mld_slowtimo_vnet(void)
1761 {
1762 	struct mld_ifsoftc *mli;
1763 
1764 	MLD_LOCK();
1765 
1766 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1767 		mld_v1_process_querier_timers(mli);
1768 	}
1769 
1770 	MLD_UNLOCK();
1771 }
1772 
1773 /*
1774  * Update the Older Version Querier Present timers for a link.
1775  * See Section 9.12 of RFC 3810.
1776  */
1777 static void
1778 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1779 {
1780 
1781 	MLD_LOCK_ASSERT();
1782 
1783 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1784 		/*
1785 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1786 		 */
1787 		CTR5(KTR_MLD,
1788 		    "%s: transition from v%d -> v%d on %p(%s)",
1789 		    __func__, mli->mli_version, MLD_VERSION_2,
1790 		    mli->mli_ifp, if_name(mli->mli_ifp));
1791 		mli->mli_version = MLD_VERSION_2;
1792 	}
1793 }
1794 
1795 /*
1796  * Transmit an MLDv1 report immediately.
1797  */
1798 static int
1799 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1800 {
1801 	struct ifnet		*ifp;
1802 	struct in6_ifaddr	*ia;
1803 	struct ip6_hdr		*ip6;
1804 	struct mbuf		*mh, *md;
1805 	struct mld_hdr		*mld;
1806 
1807 	NET_EPOCH_ASSERT();
1808 	IN6_MULTI_LIST_LOCK_ASSERT();
1809 	MLD_LOCK_ASSERT();
1810 
1811 	ifp = in6m->in6m_ifp;
1812 	/* in process of being freed */
1813 	if (ifp == NULL)
1814 		return (0);
1815 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1816 	/* ia may be NULL if link-local address is tentative. */
1817 
1818 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1819 	if (mh == NULL) {
1820 		if (ia != NULL)
1821 			ifa_free(&ia->ia_ifa);
1822 		return (ENOMEM);
1823 	}
1824 	md = m_get(M_NOWAIT, MT_DATA);
1825 	if (md == NULL) {
1826 		m_free(mh);
1827 		if (ia != NULL)
1828 			ifa_free(&ia->ia_ifa);
1829 		return (ENOMEM);
1830 	}
1831 	mh->m_next = md;
1832 
1833 	/*
1834 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1835 	 * that ether_output() does not need to allocate another mbuf
1836 	 * for the header in the most common case.
1837 	 */
1838 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1839 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1840 	mh->m_len = sizeof(struct ip6_hdr);
1841 
1842 	ip6 = mtod(mh, struct ip6_hdr *);
1843 	ip6->ip6_flow = 0;
1844 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1845 	ip6->ip6_vfc |= IPV6_VERSION;
1846 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1847 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1848 	ip6->ip6_dst = in6m->in6m_addr;
1849 
1850 	md->m_len = sizeof(struct mld_hdr);
1851 	mld = mtod(md, struct mld_hdr *);
1852 	mld->mld_type = type;
1853 	mld->mld_code = 0;
1854 	mld->mld_cksum = 0;
1855 	mld->mld_maxdelay = 0;
1856 	mld->mld_reserved = 0;
1857 	mld->mld_addr = in6m->in6m_addr;
1858 	in6_clearscope(&mld->mld_addr);
1859 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1860 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1861 
1862 	mld_save_context(mh, ifp);
1863 	mh->m_flags |= M_MLDV1;
1864 
1865 	mld_dispatch_packet(mh);
1866 
1867 	if (ia != NULL)
1868 		ifa_free(&ia->ia_ifa);
1869 	return (0);
1870 }
1871 
1872 /*
1873  * Process a state change from the upper layer for the given IPv6 group.
1874  *
1875  * Each socket holds a reference on the in_multi in its own ip_moptions.
1876  * The socket layer will have made the necessary updates to.the group
1877  * state, it is now up to MLD to issue a state change report if there
1878  * has been any change between T0 (when the last state-change was issued)
1879  * and T1 (now).
1880  *
1881  * We use the MLDv2 state machine at group level. The MLd module
1882  * however makes the decision as to which MLD protocol version to speak.
1883  * A state change *from* INCLUDE {} always means an initial join.
1884  * A state change *to* INCLUDE {} always means a final leave.
1885  *
1886  * If delay is non-zero, and the state change is an initial multicast
1887  * join, the state change report will be delayed by 'delay' ticks
1888  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1889  * the initial MLDv2 state change report will be delayed by whichever
1890  * is sooner, a pending state-change timer or delay itself.
1891  *
1892  * VIMAGE: curvnet should have been set by caller, as this routine
1893  * is called from the socket option handlers.
1894  */
1895 int
1896 mld_change_state(struct in6_multi *inm, const int delay)
1897 {
1898 	struct mld_ifsoftc *mli;
1899 	struct ifnet *ifp;
1900 	int error;
1901 
1902 	IN6_MULTI_LIST_LOCK_ASSERT();
1903 
1904 	error = 0;
1905 
1906 	/*
1907 	 * Check if the in6_multi has already been disconnected.
1908 	 */
1909 	if (inm->in6m_ifp == NULL) {
1910 		CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1911 		return (0);
1912 	}
1913 
1914 	/*
1915 	 * Try to detect if the upper layer just asked us to change state
1916 	 * for an interface which has now gone away.
1917 	 */
1918 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1919 	ifp = inm->in6m_ifma->ifma_ifp;
1920 	if (ifp == NULL)
1921 		return (0);
1922 	/*
1923 	 * Sanity check that netinet6's notion of ifp is the
1924 	 * same as net's.
1925 	 */
1926 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1927 
1928 	MLD_LOCK();
1929 	mli = MLD_IFINFO(ifp);
1930 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1931 
1932 	/*
1933 	 * If we detect a state transition to or from MCAST_UNDEFINED
1934 	 * for this group, then we are starting or finishing an MLD
1935 	 * life cycle for this group.
1936 	 */
1937 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1938 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1939 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1940 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1941 			CTR1(KTR_MLD, "%s: initial join", __func__);
1942 			error = mld_initial_join(inm, mli, delay);
1943 			goto out_locked;
1944 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1945 			CTR1(KTR_MLD, "%s: final leave", __func__);
1946 			mld_final_leave(inm, mli);
1947 			goto out_locked;
1948 		}
1949 	} else {
1950 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1951 	}
1952 
1953 	error = mld_handle_state_change(inm, mli);
1954 
1955 out_locked:
1956 	MLD_UNLOCK();
1957 	return (error);
1958 }
1959 
1960 /*
1961  * Perform the initial join for an MLD group.
1962  *
1963  * When joining a group:
1964  *  If the group should have its MLD traffic suppressed, do nothing.
1965  *  MLDv1 starts sending MLDv1 host membership reports.
1966  *  MLDv2 will schedule an MLDv2 state-change report containing the
1967  *  initial state of the membership.
1968  *
1969  * If the delay argument is non-zero, then we must delay sending the
1970  * initial state change for delay ticks (in units of PR_FASTHZ).
1971  */
1972 static int
1973 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1974     const int delay)
1975 {
1976 	struct epoch_tracker     et;
1977 	struct ifnet		*ifp;
1978 	struct mbufq		*mq;
1979 	int			 error, retval, syncstates;
1980 	int			 odelay;
1981 #ifdef KTR
1982 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1983 #endif
1984 
1985 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1986 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1987 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1988 
1989 	error = 0;
1990 	syncstates = 1;
1991 
1992 	ifp = inm->in6m_ifp;
1993 
1994 	IN6_MULTI_LIST_LOCK_ASSERT();
1995 	MLD_LOCK_ASSERT();
1996 
1997 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1998 
1999 	/*
2000 	 * Groups joined on loopback or marked as 'not reported',
2001 	 * enter the MLD_SILENT_MEMBER state and
2002 	 * are never reported in any protocol exchanges.
2003 	 * All other groups enter the appropriate state machine
2004 	 * for the version in use on this link.
2005 	 * A link marked as MLIF_SILENT causes MLD to be completely
2006 	 * disabled for the link.
2007 	 */
2008 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2009 	    (mli->mli_flags & MLIF_SILENT) ||
2010 	    !mld_is_addr_reported(&inm->in6m_addr)) {
2011 		CTR1(KTR_MLD,
2012 "%s: not kicking state machine for silent group", __func__);
2013 		inm->in6m_state = MLD_SILENT_MEMBER;
2014 		inm->in6m_timer = 0;
2015 	} else {
2016 		/*
2017 		 * Deal with overlapping in_multi lifecycle.
2018 		 * If this group was LEAVING, then make sure
2019 		 * we drop the reference we picked up to keep the
2020 		 * group around for the final INCLUDE {} enqueue.
2021 		 */
2022 		if (mli->mli_version == MLD_VERSION_2 &&
2023 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
2024 			inm->in6m_refcount--;
2025 			MPASS(inm->in6m_refcount > 0);
2026 		}
2027 		inm->in6m_state = MLD_REPORTING_MEMBER;
2028 
2029 		switch (mli->mli_version) {
2030 		case MLD_VERSION_1:
2031 			/*
2032 			 * If a delay was provided, only use it if
2033 			 * it is greater than the delay normally
2034 			 * used for an MLDv1 state change report,
2035 			 * and delay sending the initial MLDv1 report
2036 			 * by not transitioning to the IDLE state.
2037 			 */
2038 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2039 			if (delay) {
2040 				inm->in6m_timer = max(delay, odelay);
2041 				V_current_state_timers_running6 = 1;
2042 			} else {
2043 				inm->in6m_state = MLD_IDLE_MEMBER;
2044 				NET_EPOCH_ENTER(et);
2045 				error = mld_v1_transmit_report(inm,
2046 				     MLD_LISTENER_REPORT);
2047 				NET_EPOCH_EXIT(et);
2048 				if (error == 0) {
2049 					inm->in6m_timer = odelay;
2050 					V_current_state_timers_running6 = 1;
2051 				}
2052 			}
2053 			break;
2054 
2055 		case MLD_VERSION_2:
2056 			/*
2057 			 * Defer update of T0 to T1, until the first copy
2058 			 * of the state change has been transmitted.
2059 			 */
2060 			syncstates = 0;
2061 
2062 			/*
2063 			 * Immediately enqueue a State-Change Report for
2064 			 * this interface, freeing any previous reports.
2065 			 * Don't kick the timers if there is nothing to do,
2066 			 * or if an error occurred.
2067 			 */
2068 			mq = &inm->in6m_scq;
2069 			mbufq_drain(mq);
2070 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2071 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2072 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2073 			    __func__, retval);
2074 			if (retval <= 0) {
2075 				error = retval * -1;
2076 				break;
2077 			}
2078 
2079 			/*
2080 			 * Schedule transmission of pending state-change
2081 			 * report up to RV times for this link. The timer
2082 			 * will fire at the next mld_fasttimo (~200ms),
2083 			 * giving us an opportunity to merge the reports.
2084 			 *
2085 			 * If a delay was provided to this function, only
2086 			 * use this delay if sooner than the existing one.
2087 			 */
2088 			KASSERT(mli->mli_rv > 1,
2089 			   ("%s: invalid robustness %d", __func__,
2090 			    mli->mli_rv));
2091 			inm->in6m_scrv = mli->mli_rv;
2092 			if (delay) {
2093 				if (inm->in6m_sctimer > 1) {
2094 					inm->in6m_sctimer =
2095 					    min(inm->in6m_sctimer, delay);
2096 				} else
2097 					inm->in6m_sctimer = delay;
2098 			} else
2099 				inm->in6m_sctimer = 1;
2100 			V_state_change_timers_running6 = 1;
2101 
2102 			error = 0;
2103 			break;
2104 		}
2105 	}
2106 
2107 	/*
2108 	 * Only update the T0 state if state change is atomic,
2109 	 * i.e. we don't need to wait for a timer to fire before we
2110 	 * can consider the state change to have been communicated.
2111 	 */
2112 	if (syncstates) {
2113 		in6m_commit(inm);
2114 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2115 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2116 		    if_name(inm->in6m_ifp));
2117 	}
2118 
2119 	return (error);
2120 }
2121 
2122 /*
2123  * Issue an intermediate state change during the life-cycle.
2124  */
2125 static int
2126 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2127 {
2128 	struct ifnet		*ifp;
2129 	int			 retval;
2130 #ifdef KTR
2131 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2132 #endif
2133 
2134 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2135 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2136 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2137 
2138 	ifp = inm->in6m_ifp;
2139 
2140 	IN6_MULTI_LIST_LOCK_ASSERT();
2141 	MLD_LOCK_ASSERT();
2142 
2143 	KASSERT(mli && mli->mli_ifp == ifp,
2144 	    ("%s: inconsistent ifp", __func__));
2145 
2146 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2147 	    (mli->mli_flags & MLIF_SILENT) ||
2148 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2149 	    (mli->mli_version != MLD_VERSION_2)) {
2150 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2151 			CTR1(KTR_MLD,
2152 "%s: not kicking state machine for silent group", __func__);
2153 		}
2154 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2155 		in6m_commit(inm);
2156 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2157 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2158 		    if_name(inm->in6m_ifp));
2159 		return (0);
2160 	}
2161 
2162 	mbufq_drain(&inm->in6m_scq);
2163 
2164 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2165 	    (mli->mli_flags & MLIF_USEALLOW));
2166 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2167 	if (retval <= 0)
2168 		return (-retval);
2169 
2170 	/*
2171 	 * If record(s) were enqueued, start the state-change
2172 	 * report timer for this group.
2173 	 */
2174 	inm->in6m_scrv = mli->mli_rv;
2175 	inm->in6m_sctimer = 1;
2176 	V_state_change_timers_running6 = 1;
2177 
2178 	return (0);
2179 }
2180 
2181 /*
2182  * Perform the final leave for a multicast address.
2183  *
2184  * When leaving a group:
2185  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2186  *  MLDv2 enqueues a state-change report containing a transition
2187  *  to INCLUDE {} for immediate transmission.
2188  */
2189 static void
2190 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2191 {
2192 	struct epoch_tracker     et;
2193 	int syncstates;
2194 #ifdef KTR
2195 	char ip6tbuf[INET6_ADDRSTRLEN];
2196 #endif
2197 
2198 	syncstates = 1;
2199 
2200 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2201 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2202 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2203 
2204 	IN6_MULTI_LIST_LOCK_ASSERT();
2205 	MLD_LOCK_ASSERT();
2206 
2207 	switch (inm->in6m_state) {
2208 	case MLD_NOT_MEMBER:
2209 	case MLD_SILENT_MEMBER:
2210 	case MLD_LEAVING_MEMBER:
2211 		/* Already leaving or left; do nothing. */
2212 		CTR1(KTR_MLD,
2213 "%s: not kicking state machine for silent group", __func__);
2214 		break;
2215 	case MLD_REPORTING_MEMBER:
2216 	case MLD_IDLE_MEMBER:
2217 	case MLD_G_QUERY_PENDING_MEMBER:
2218 	case MLD_SG_QUERY_PENDING_MEMBER:
2219 		if (mli->mli_version == MLD_VERSION_1) {
2220 #ifdef INVARIANTS
2221 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2222 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2223 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2224 			     __func__);
2225 #endif
2226 			NET_EPOCH_ENTER(et);
2227 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2228 			NET_EPOCH_EXIT(et);
2229 			inm->in6m_state = MLD_NOT_MEMBER;
2230 			V_current_state_timers_running6 = 1;
2231 		} else if (mli->mli_version == MLD_VERSION_2) {
2232 			/*
2233 			 * Stop group timer and all pending reports.
2234 			 * Immediately enqueue a state-change report
2235 			 * TO_IN {} to be sent on the next fast timeout,
2236 			 * giving us an opportunity to merge reports.
2237 			 */
2238 			mbufq_drain(&inm->in6m_scq);
2239 			inm->in6m_timer = 0;
2240 			inm->in6m_scrv = mli->mli_rv;
2241 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2242 			    "pending retransmissions.", __func__,
2243 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2244 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2245 			if (inm->in6m_scrv == 0) {
2246 				inm->in6m_state = MLD_NOT_MEMBER;
2247 				inm->in6m_sctimer = 0;
2248 			} else {
2249 				int retval;
2250 
2251 				in6m_acquire_locked(inm);
2252 
2253 				retval = mld_v2_enqueue_group_record(
2254 				    &inm->in6m_scq, inm, 1, 0, 0,
2255 				    (mli->mli_flags & MLIF_USEALLOW));
2256 				KASSERT(retval != 0,
2257 				    ("%s: enqueue record = %d", __func__,
2258 				     retval));
2259 
2260 				inm->in6m_state = MLD_LEAVING_MEMBER;
2261 				inm->in6m_sctimer = 1;
2262 				V_state_change_timers_running6 = 1;
2263 				syncstates = 0;
2264 			}
2265 			break;
2266 		}
2267 		break;
2268 	case MLD_LAZY_MEMBER:
2269 	case MLD_SLEEPING_MEMBER:
2270 	case MLD_AWAKENING_MEMBER:
2271 		/* Our reports are suppressed; do nothing. */
2272 		break;
2273 	}
2274 
2275 	if (syncstates) {
2276 		in6m_commit(inm);
2277 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2278 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2279 		    if_name(inm->in6m_ifp));
2280 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2281 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2282 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2283 	}
2284 }
2285 
2286 /*
2287  * Enqueue an MLDv2 group record to the given output queue.
2288  *
2289  * If is_state_change is zero, a current-state record is appended.
2290  * If is_state_change is non-zero, a state-change report is appended.
2291  *
2292  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2293  * If is_group_query is zero, and if there is a packet with free space
2294  * at the tail of the queue, it will be appended to providing there
2295  * is enough free space.
2296  * Otherwise a new mbuf packet chain is allocated.
2297  *
2298  * If is_source_query is non-zero, each source is checked to see if
2299  * it was recorded for a Group-Source query, and will be omitted if
2300  * it is not both in-mode and recorded.
2301  *
2302  * If use_block_allow is non-zero, state change reports for initial join
2303  * and final leave, on an inclusive mode group with a source list, will be
2304  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2305  *
2306  * The function will attempt to allocate leading space in the packet
2307  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2308  *
2309  * If successful the size of all data appended to the queue is returned,
2310  * otherwise an error code less than zero is returned, or zero if
2311  * no record(s) were appended.
2312  */
2313 static int
2314 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2315     const int is_state_change, const int is_group_query,
2316     const int is_source_query, const int use_block_allow)
2317 {
2318 	struct mldv2_record	 mr;
2319 	struct mldv2_record	*pmr;
2320 	struct ifnet		*ifp;
2321 	struct ip6_msource	*ims, *nims;
2322 	struct mbuf		*m0, *m, *md;
2323 	int			 is_filter_list_change;
2324 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2325 	int			 record_has_sources;
2326 	int			 now;
2327 	int			 type;
2328 	uint8_t			 mode;
2329 #ifdef KTR
2330 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2331 #endif
2332 
2333 	IN6_MULTI_LIST_LOCK_ASSERT();
2334 
2335 	ifp = inm->in6m_ifp;
2336 	is_filter_list_change = 0;
2337 	m = NULL;
2338 	m0 = NULL;
2339 	m0srcs = 0;
2340 	msrcs = 0;
2341 	nbytes = 0;
2342 	nims = NULL;
2343 	record_has_sources = 1;
2344 	pmr = NULL;
2345 	type = MLD_DO_NOTHING;
2346 	mode = inm->in6m_st[1].iss_fmode;
2347 
2348 	/*
2349 	 * If we did not transition out of ASM mode during t0->t1,
2350 	 * and there are no source nodes to process, we can skip
2351 	 * the generation of source records.
2352 	 */
2353 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2354 	    inm->in6m_nsrc == 0)
2355 		record_has_sources = 0;
2356 
2357 	if (is_state_change) {
2358 		/*
2359 		 * Queue a state change record.
2360 		 * If the mode did not change, and there are non-ASM
2361 		 * listeners or source filters present,
2362 		 * we potentially need to issue two records for the group.
2363 		 * If there are ASM listeners, and there was no filter
2364 		 * mode transition of any kind, do nothing.
2365 		 *
2366 		 * If we are transitioning to MCAST_UNDEFINED, we need
2367 		 * not send any sources. A transition to/from this state is
2368 		 * considered inclusive with some special treatment.
2369 		 *
2370 		 * If we are rewriting initial joins/leaves to use
2371 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2372 		 * we need to send sources in all cases.
2373 		 */
2374 		if (mode != inm->in6m_st[0].iss_fmode) {
2375 			if (mode == MCAST_EXCLUDE) {
2376 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2377 				    __func__);
2378 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2379 			} else {
2380 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2381 				    __func__);
2382 				if (use_block_allow) {
2383 					/*
2384 					 * XXX
2385 					 * Here we're interested in state
2386 					 * edges either direction between
2387 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2388 					 * Perhaps we should just check
2389 					 * the group state, rather than
2390 					 * the filter mode.
2391 					 */
2392 					if (mode == MCAST_UNDEFINED) {
2393 						type = MLD_BLOCK_OLD_SOURCES;
2394 					} else {
2395 						type = MLD_ALLOW_NEW_SOURCES;
2396 					}
2397 				} else {
2398 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2399 					if (mode == MCAST_UNDEFINED)
2400 						record_has_sources = 0;
2401 				}
2402 			}
2403 		} else {
2404 			if (record_has_sources) {
2405 				is_filter_list_change = 1;
2406 			} else {
2407 				type = MLD_DO_NOTHING;
2408 			}
2409 		}
2410 	} else {
2411 		/*
2412 		 * Queue a current state record.
2413 		 */
2414 		if (mode == MCAST_EXCLUDE) {
2415 			type = MLD_MODE_IS_EXCLUDE;
2416 		} else if (mode == MCAST_INCLUDE) {
2417 			type = MLD_MODE_IS_INCLUDE;
2418 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2419 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2420 			     __func__, inm, inm->in6m_st[1].iss_asm));
2421 		}
2422 	}
2423 
2424 	/*
2425 	 * Generate the filter list changes using a separate function.
2426 	 */
2427 	if (is_filter_list_change)
2428 		return (mld_v2_enqueue_filter_change(mq, inm));
2429 
2430 	if (type == MLD_DO_NOTHING) {
2431 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2432 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2433 		    if_name(inm->in6m_ifp));
2434 		return (0);
2435 	}
2436 
2437 	/*
2438 	 * If any sources are present, we must be able to fit at least
2439 	 * one in the trailing space of the tail packet's mbuf,
2440 	 * ideally more.
2441 	 */
2442 	minrec0len = sizeof(struct mldv2_record);
2443 	if (record_has_sources)
2444 		minrec0len += sizeof(struct in6_addr);
2445 
2446 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2447 	    mld_rec_type_to_str(type),
2448 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2449 	    if_name(inm->in6m_ifp));
2450 
2451 	/*
2452 	 * Check if we have a packet in the tail of the queue for this
2453 	 * group into which the first group record for this group will fit.
2454 	 * Otherwise allocate a new packet.
2455 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2456 	 * Note: Group records for G/GSR query responses MUST be sent
2457 	 * in their own packet.
2458 	 */
2459 	m0 = mbufq_last(mq);
2460 	if (!is_group_query &&
2461 	    m0 != NULL &&
2462 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2463 	    (m0->m_pkthdr.len + minrec0len) <
2464 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2465 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2466 			    sizeof(struct mldv2_record)) /
2467 			    sizeof(struct in6_addr);
2468 		m = m0;
2469 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2470 	} else {
2471 		if (mbufq_full(mq)) {
2472 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2473 			return (-ENOMEM);
2474 		}
2475 		m = NULL;
2476 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2477 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2478 		if (!is_state_change && !is_group_query)
2479 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2480 		if (m == NULL)
2481 			m = m_gethdr(M_NOWAIT, MT_DATA);
2482 		if (m == NULL)
2483 			return (-ENOMEM);
2484 
2485 		mld_save_context(m, ifp);
2486 
2487 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2488 	}
2489 
2490 	/*
2491 	 * Append group record.
2492 	 * If we have sources, we don't know how many yet.
2493 	 */
2494 	mr.mr_type = type;
2495 	mr.mr_datalen = 0;
2496 	mr.mr_numsrc = 0;
2497 	mr.mr_addr = inm->in6m_addr;
2498 	in6_clearscope(&mr.mr_addr);
2499 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2500 		if (m != m0)
2501 			m_freem(m);
2502 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2503 		return (-ENOMEM);
2504 	}
2505 	nbytes += sizeof(struct mldv2_record);
2506 
2507 	/*
2508 	 * Append as many sources as will fit in the first packet.
2509 	 * If we are appending to a new packet, the chain allocation
2510 	 * may potentially use clusters; use m_getptr() in this case.
2511 	 * If we are appending to an existing packet, we need to obtain
2512 	 * a pointer to the group record after m_append(), in case a new
2513 	 * mbuf was allocated.
2514 	 *
2515 	 * Only append sources which are in-mode at t1. If we are
2516 	 * transitioning to MCAST_UNDEFINED state on the group, and
2517 	 * use_block_allow is zero, do not include source entries.
2518 	 * Otherwise, we need to include this source in the report.
2519 	 *
2520 	 * Only report recorded sources in our filter set when responding
2521 	 * to a group-source query.
2522 	 */
2523 	if (record_has_sources) {
2524 		if (m == m0) {
2525 			md = m_last(m);
2526 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2527 			    md->m_len - nbytes);
2528 		} else {
2529 			md = m_getptr(m, 0, &off);
2530 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2531 			    off);
2532 		}
2533 		msrcs = 0;
2534 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2535 		    nims) {
2536 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2537 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2538 			now = im6s_get_mode(inm, ims, 1);
2539 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2540 			if ((now != mode) ||
2541 			    (now == mode &&
2542 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2543 				CTR1(KTR_MLD, "%s: skip node", __func__);
2544 				continue;
2545 			}
2546 			if (is_source_query && ims->im6s_stp == 0) {
2547 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2548 				    __func__);
2549 				continue;
2550 			}
2551 			CTR1(KTR_MLD, "%s: append node", __func__);
2552 			if (!m_append(m, sizeof(struct in6_addr),
2553 			    (void *)&ims->im6s_addr)) {
2554 				if (m != m0)
2555 					m_freem(m);
2556 				CTR1(KTR_MLD, "%s: m_append() failed.",
2557 				    __func__);
2558 				return (-ENOMEM);
2559 			}
2560 			nbytes += sizeof(struct in6_addr);
2561 			++msrcs;
2562 			if (msrcs == m0srcs)
2563 				break;
2564 		}
2565 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2566 		    msrcs);
2567 		pmr->mr_numsrc = htons(msrcs);
2568 		nbytes += (msrcs * sizeof(struct in6_addr));
2569 	}
2570 
2571 	if (is_source_query && msrcs == 0) {
2572 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2573 		if (m != m0)
2574 			m_freem(m);
2575 		return (0);
2576 	}
2577 
2578 	/*
2579 	 * We are good to go with first packet.
2580 	 */
2581 	if (m != m0) {
2582 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2583 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2584 		mbufq_enqueue(mq, m);
2585 	} else
2586 		m->m_pkthdr.PH_vt.vt_nrecs++;
2587 
2588 	/*
2589 	 * No further work needed if no source list in packet(s).
2590 	 */
2591 	if (!record_has_sources)
2592 		return (nbytes);
2593 
2594 	/*
2595 	 * Whilst sources remain to be announced, we need to allocate
2596 	 * a new packet and fill out as many sources as will fit.
2597 	 * Always try for a cluster first.
2598 	 */
2599 	while (nims != NULL) {
2600 		if (mbufq_full(mq)) {
2601 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2602 			return (-ENOMEM);
2603 		}
2604 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2605 		if (m == NULL)
2606 			m = m_gethdr(M_NOWAIT, MT_DATA);
2607 		if (m == NULL)
2608 			return (-ENOMEM);
2609 		mld_save_context(m, ifp);
2610 		md = m_getptr(m, 0, &off);
2611 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2612 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2613 
2614 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2615 			if (m != m0)
2616 				m_freem(m);
2617 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2618 			return (-ENOMEM);
2619 		}
2620 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2621 		nbytes += sizeof(struct mldv2_record);
2622 
2623 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2624 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2625 
2626 		msrcs = 0;
2627 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2628 			CTR2(KTR_MLD, "%s: visit node %s",
2629 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2630 			now = im6s_get_mode(inm, ims, 1);
2631 			if ((now != mode) ||
2632 			    (now == mode &&
2633 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2634 				CTR1(KTR_MLD, "%s: skip node", __func__);
2635 				continue;
2636 			}
2637 			if (is_source_query && ims->im6s_stp == 0) {
2638 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2639 				    __func__);
2640 				continue;
2641 			}
2642 			CTR1(KTR_MLD, "%s: append node", __func__);
2643 			if (!m_append(m, sizeof(struct in6_addr),
2644 			    (void *)&ims->im6s_addr)) {
2645 				if (m != m0)
2646 					m_freem(m);
2647 				CTR1(KTR_MLD, "%s: m_append() failed.",
2648 				    __func__);
2649 				return (-ENOMEM);
2650 			}
2651 			++msrcs;
2652 			if (msrcs == m0srcs)
2653 				break;
2654 		}
2655 		pmr->mr_numsrc = htons(msrcs);
2656 		nbytes += (msrcs * sizeof(struct in6_addr));
2657 
2658 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2659 		mbufq_enqueue(mq, m);
2660 	}
2661 
2662 	return (nbytes);
2663 }
2664 
2665 /*
2666  * Type used to mark record pass completion.
2667  * We exploit the fact we can cast to this easily from the
2668  * current filter modes on each ip_msource node.
2669  */
2670 typedef enum {
2671 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2672 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2673 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2674 	REC_FULL = REC_ALLOW | REC_BLOCK
2675 } rectype_t;
2676 
2677 /*
2678  * Enqueue an MLDv2 filter list change to the given output queue.
2679  *
2680  * Source list filter state is held in an RB-tree. When the filter list
2681  * for a group is changed without changing its mode, we need to compute
2682  * the deltas between T0 and T1 for each source in the filter set,
2683  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2684  *
2685  * As we may potentially queue two record types, and the entire R-B tree
2686  * needs to be walked at once, we break this out into its own function
2687  * so we can generate a tightly packed queue of packets.
2688  *
2689  * XXX This could be written to only use one tree walk, although that makes
2690  * serializing into the mbuf chains a bit harder. For now we do two walks
2691  * which makes things easier on us, and it may or may not be harder on
2692  * the L2 cache.
2693  *
2694  * If successful the size of all data appended to the queue is returned,
2695  * otherwise an error code less than zero is returned, or zero if
2696  * no record(s) were appended.
2697  */
2698 static int
2699 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2700 {
2701 	static const int MINRECLEN =
2702 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2703 	struct ifnet		*ifp;
2704 	struct mldv2_record	 mr;
2705 	struct mldv2_record	*pmr;
2706 	struct ip6_msource	*ims, *nims;
2707 	struct mbuf		*m, *m0, *md;
2708 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2709 	int			 nallow, nblock;
2710 	uint8_t			 mode, now, then;
2711 	rectype_t		 crt, drt, nrt;
2712 #ifdef KTR
2713 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2714 #endif
2715 
2716 	IN6_MULTI_LIST_LOCK_ASSERT();
2717 
2718 	if (inm->in6m_nsrc == 0 ||
2719 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2720 		return (0);
2721 
2722 	ifp = inm->in6m_ifp;			/* interface */
2723 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2724 	crt = REC_NONE;	/* current group record type */
2725 	drt = REC_NONE;	/* mask of completed group record types */
2726 	nrt = REC_NONE;	/* record type for current node */
2727 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2728 	npbytes = 0;	/* # of bytes appended this packet */
2729 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2730 	rsrcs = 0;	/* # sources encoded in current record */
2731 	schanged = 0;	/* # nodes encoded in overall filter change */
2732 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2733 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2734 	nims = NULL;	/* next tree node pointer */
2735 
2736 	/*
2737 	 * For each possible filter record mode.
2738 	 * The first kind of source we encounter tells us which
2739 	 * is the first kind of record we start appending.
2740 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2741 	 * as the inverse of the group's filter mode.
2742 	 */
2743 	while (drt != REC_FULL) {
2744 		do {
2745 			m0 = mbufq_last(mq);
2746 			if (m0 != NULL &&
2747 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2748 			     MLD_V2_REPORT_MAXRECS) &&
2749 			    (m0->m_pkthdr.len + MINRECLEN) <
2750 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2751 				m = m0;
2752 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2753 					    sizeof(struct mldv2_record)) /
2754 					    sizeof(struct in6_addr);
2755 				CTR1(KTR_MLD,
2756 				    "%s: use previous packet", __func__);
2757 			} else {
2758 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2759 				if (m == NULL)
2760 					m = m_gethdr(M_NOWAIT, MT_DATA);
2761 				if (m == NULL) {
2762 					CTR1(KTR_MLD,
2763 					    "%s: m_get*() failed", __func__);
2764 					return (-ENOMEM);
2765 				}
2766 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2767 				mld_save_context(m, ifp);
2768 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2769 				    sizeof(struct mldv2_record)) /
2770 				    sizeof(struct in6_addr);
2771 				npbytes = 0;
2772 				CTR1(KTR_MLD,
2773 				    "%s: allocated new packet", __func__);
2774 			}
2775 			/*
2776 			 * Append the MLD group record header to the
2777 			 * current packet's data area.
2778 			 * Recalculate pointer to free space for next
2779 			 * group record, in case m_append() allocated
2780 			 * a new mbuf or cluster.
2781 			 */
2782 			memset(&mr, 0, sizeof(mr));
2783 			mr.mr_addr = inm->in6m_addr;
2784 			in6_clearscope(&mr.mr_addr);
2785 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2786 				if (m != m0)
2787 					m_freem(m);
2788 				CTR1(KTR_MLD,
2789 				    "%s: m_append() failed", __func__);
2790 				return (-ENOMEM);
2791 			}
2792 			npbytes += sizeof(struct mldv2_record);
2793 			if (m != m0) {
2794 				/* new packet; offset in chain */
2795 				md = m_getptr(m, npbytes -
2796 				    sizeof(struct mldv2_record), &off);
2797 				pmr = (struct mldv2_record *)(mtod(md,
2798 				    uint8_t *) + off);
2799 			} else {
2800 				/* current packet; offset from last append */
2801 				md = m_last(m);
2802 				pmr = (struct mldv2_record *)(mtod(md,
2803 				    uint8_t *) + md->m_len -
2804 				    sizeof(struct mldv2_record));
2805 			}
2806 			/*
2807 			 * Begin walking the tree for this record type
2808 			 * pass, or continue from where we left off
2809 			 * previously if we had to allocate a new packet.
2810 			 * Only report deltas in-mode at t1.
2811 			 * We need not report included sources as allowed
2812 			 * if we are in inclusive mode on the group,
2813 			 * however the converse is not true.
2814 			 */
2815 			rsrcs = 0;
2816 			if (nims == NULL) {
2817 				nims = RB_MIN(ip6_msource_tree,
2818 				    &inm->in6m_srcs);
2819 			}
2820 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2821 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2822 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2823 				now = im6s_get_mode(inm, ims, 1);
2824 				then = im6s_get_mode(inm, ims, 0);
2825 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2826 				    __func__, then, now);
2827 				if (now == then) {
2828 					CTR1(KTR_MLD,
2829 					    "%s: skip unchanged", __func__);
2830 					continue;
2831 				}
2832 				if (mode == MCAST_EXCLUDE &&
2833 				    now == MCAST_INCLUDE) {
2834 					CTR1(KTR_MLD,
2835 					    "%s: skip IN src on EX group",
2836 					    __func__);
2837 					continue;
2838 				}
2839 				nrt = (rectype_t)now;
2840 				if (nrt == REC_NONE)
2841 					nrt = (rectype_t)(~mode & REC_FULL);
2842 				if (schanged++ == 0) {
2843 					crt = nrt;
2844 				} else if (crt != nrt)
2845 					continue;
2846 				if (!m_append(m, sizeof(struct in6_addr),
2847 				    (void *)&ims->im6s_addr)) {
2848 					if (m != m0)
2849 						m_freem(m);
2850 					CTR1(KTR_MLD,
2851 					    "%s: m_append() failed", __func__);
2852 					return (-ENOMEM);
2853 				}
2854 				nallow += !!(crt == REC_ALLOW);
2855 				nblock += !!(crt == REC_BLOCK);
2856 				if (++rsrcs == m0srcs)
2857 					break;
2858 			}
2859 			/*
2860 			 * If we did not append any tree nodes on this
2861 			 * pass, back out of allocations.
2862 			 */
2863 			if (rsrcs == 0) {
2864 				npbytes -= sizeof(struct mldv2_record);
2865 				if (m != m0) {
2866 					CTR1(KTR_MLD,
2867 					    "%s: m_free(m)", __func__);
2868 					m_freem(m);
2869 				} else {
2870 					CTR1(KTR_MLD,
2871 					    "%s: m_adj(m, -mr)", __func__);
2872 					m_adj(m, -((int)sizeof(
2873 					    struct mldv2_record)));
2874 				}
2875 				continue;
2876 			}
2877 			npbytes += (rsrcs * sizeof(struct in6_addr));
2878 			if (crt == REC_ALLOW)
2879 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2880 			else if (crt == REC_BLOCK)
2881 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2882 			pmr->mr_numsrc = htons(rsrcs);
2883 			/*
2884 			 * Count the new group record, and enqueue this
2885 			 * packet if it wasn't already queued.
2886 			 */
2887 			m->m_pkthdr.PH_vt.vt_nrecs++;
2888 			if (m != m0)
2889 				mbufq_enqueue(mq, m);
2890 			nbytes += npbytes;
2891 		} while (nims != NULL);
2892 		drt |= crt;
2893 		crt = (~crt & REC_FULL);
2894 	}
2895 
2896 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2897 	    nallow, nblock);
2898 
2899 	return (nbytes);
2900 }
2901 
2902 static int
2903 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2904 {
2905 	struct mbufq	*gq;
2906 	struct mbuf	*m;		/* pending state-change */
2907 	struct mbuf	*m0;		/* copy of pending state-change */
2908 	struct mbuf	*mt;		/* last state-change in packet */
2909 	int		 docopy, domerge;
2910 	u_int		 recslen;
2911 
2912 	docopy = 0;
2913 	domerge = 0;
2914 	recslen = 0;
2915 
2916 	IN6_MULTI_LIST_LOCK_ASSERT();
2917 	MLD_LOCK_ASSERT();
2918 
2919 	/*
2920 	 * If there are further pending retransmissions, make a writable
2921 	 * copy of each queued state-change message before merging.
2922 	 */
2923 	if (inm->in6m_scrv > 0)
2924 		docopy = 1;
2925 
2926 	gq = &inm->in6m_scq;
2927 #ifdef KTR
2928 	if (mbufq_first(gq) == NULL) {
2929 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2930 		    __func__, inm);
2931 	}
2932 #endif
2933 
2934 	m = mbufq_first(gq);
2935 	while (m != NULL) {
2936 		/*
2937 		 * Only merge the report into the current packet if
2938 		 * there is sufficient space to do so; an MLDv2 report
2939 		 * packet may only contain 65,535 group records.
2940 		 * Always use a simple mbuf chain concatentation to do this,
2941 		 * as large state changes for single groups may have
2942 		 * allocated clusters.
2943 		 */
2944 		domerge = 0;
2945 		mt = mbufq_last(scq);
2946 		if (mt != NULL) {
2947 			recslen = m_length(m, NULL);
2948 
2949 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2950 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2951 			    MLD_V2_REPORT_MAXRECS) &&
2952 			    (mt->m_pkthdr.len + recslen <=
2953 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2954 				domerge = 1;
2955 		}
2956 
2957 		if (!domerge && mbufq_full(gq)) {
2958 			CTR2(KTR_MLD,
2959 			    "%s: outbound queue full, skipping whole packet %p",
2960 			    __func__, m);
2961 			mt = m->m_nextpkt;
2962 			if (!docopy)
2963 				m_freem(m);
2964 			m = mt;
2965 			continue;
2966 		}
2967 
2968 		if (!docopy) {
2969 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2970 			m0 = mbufq_dequeue(gq);
2971 			m = m0->m_nextpkt;
2972 		} else {
2973 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2974 			m0 = m_dup(m, M_NOWAIT);
2975 			if (m0 == NULL)
2976 				return (ENOMEM);
2977 			m0->m_nextpkt = NULL;
2978 			m = m->m_nextpkt;
2979 		}
2980 
2981 		if (!domerge) {
2982 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2983 			    __func__, m0, scq);
2984 			mbufq_enqueue(scq, m0);
2985 		} else {
2986 			struct mbuf *mtl;	/* last mbuf of packet mt */
2987 
2988 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2989 			    __func__, m0, mt);
2990 
2991 			mtl = m_last(mt);
2992 			m0->m_flags &= ~M_PKTHDR;
2993 			mt->m_pkthdr.len += recslen;
2994 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2995 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2996 
2997 			mtl->m_next = m0;
2998 		}
2999 	}
3000 
3001 	return (0);
3002 }
3003 
3004 /*
3005  * Respond to a pending MLDv2 General Query.
3006  */
3007 static void
3008 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
3009 {
3010 	struct ifmultiaddr	*ifma;
3011 	struct ifnet		*ifp;
3012 	struct in6_multi	*inm;
3013 	int			 retval;
3014 
3015 	NET_EPOCH_ASSERT();
3016 	IN6_MULTI_LIST_LOCK_ASSERT();
3017 	MLD_LOCK_ASSERT();
3018 
3019 	KASSERT(mli->mli_version == MLD_VERSION_2,
3020 	    ("%s: called when version %d", __func__, mli->mli_version));
3021 
3022 	/*
3023 	 * Check that there are some packets queued. If so, send them first.
3024 	 * For large number of groups the reply to general query can take
3025 	 * many packets, we should finish sending them before starting of
3026 	 * queuing the new reply.
3027 	 */
3028 	if (mbufq_len(&mli->mli_gq) != 0)
3029 		goto send;
3030 
3031 	ifp = mli->mli_ifp;
3032 
3033 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3034 		inm = in6m_ifmultiaddr_get_inm(ifma);
3035 		if (inm == NULL)
3036 			continue;
3037 		KASSERT(ifp == inm->in6m_ifp,
3038 		    ("%s: inconsistent ifp", __func__));
3039 
3040 		switch (inm->in6m_state) {
3041 		case MLD_NOT_MEMBER:
3042 		case MLD_SILENT_MEMBER:
3043 			break;
3044 		case MLD_REPORTING_MEMBER:
3045 		case MLD_IDLE_MEMBER:
3046 		case MLD_LAZY_MEMBER:
3047 		case MLD_SLEEPING_MEMBER:
3048 		case MLD_AWAKENING_MEMBER:
3049 			inm->in6m_state = MLD_REPORTING_MEMBER;
3050 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3051 			    inm, 0, 0, 0, 0);
3052 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3053 			    __func__, retval);
3054 			break;
3055 		case MLD_G_QUERY_PENDING_MEMBER:
3056 		case MLD_SG_QUERY_PENDING_MEMBER:
3057 		case MLD_LEAVING_MEMBER:
3058 			break;
3059 		}
3060 	}
3061 
3062 send:
3063 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3064 
3065 	/*
3066 	 * Slew transmission of bursts over 500ms intervals.
3067 	 */
3068 	if (mbufq_first(&mli->mli_gq) != NULL) {
3069 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3070 		    MLD_RESPONSE_BURST_INTERVAL);
3071 		V_interface_timers_running6 = 1;
3072 	}
3073 }
3074 
3075 /*
3076  * Transmit the next pending message in the output queue.
3077  *
3078  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3079  * MRT: Nothing needs to be done, as MLD traffic is always local to
3080  * a link and uses a link-scope multicast address.
3081  */
3082 static void
3083 mld_dispatch_packet(struct mbuf *m)
3084 {
3085 	struct ip6_moptions	 im6o;
3086 	struct ifnet		*ifp;
3087 	struct ifnet		*oifp;
3088 	struct mbuf		*m0;
3089 	struct mbuf		*md;
3090 	struct ip6_hdr		*ip6;
3091 	struct mld_hdr		*mld;
3092 	int			 error;
3093 	int			 off;
3094 	int			 type;
3095 	uint32_t		 ifindex;
3096 
3097 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3098 	NET_EPOCH_ASSERT();
3099 
3100 	/*
3101 	 * Set VNET image pointer from enqueued mbuf chain
3102 	 * before doing anything else. Whilst we use interface
3103 	 * indexes to guard against interface detach, they are
3104 	 * unique to each VIMAGE and must be retrieved.
3105 	 */
3106 	ifindex = mld_restore_context(m);
3107 
3108 	/*
3109 	 * Check if the ifnet still exists. This limits the scope of
3110 	 * any race in the absence of a global ifp lock for low cost
3111 	 * (an array lookup).
3112 	 */
3113 	ifp = ifnet_byindex(ifindex);
3114 	if (ifp == NULL) {
3115 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3116 		    __func__, m, ifindex);
3117 		m_freem(m);
3118 		IP6STAT_INC(ip6s_noroute);
3119 		goto out;
3120 	}
3121 
3122 	im6o.im6o_multicast_hlim  = 1;
3123 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3124 	im6o.im6o_multicast_ifp = ifp;
3125 
3126 	if (m->m_flags & M_MLDV1) {
3127 		m0 = m;
3128 	} else {
3129 		m0 = mld_v2_encap_report(ifp, m);
3130 		if (m0 == NULL) {
3131 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3132 			IP6STAT_INC(ip6s_odropped);
3133 			goto out;
3134 		}
3135 	}
3136 
3137 	mld_scrub_context(m0);
3138 	m_clrprotoflags(m);
3139 	m0->m_pkthdr.rcvif = V_loif;
3140 
3141 	ip6 = mtod(m0, struct ip6_hdr *);
3142 #if 0
3143 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3144 #else
3145 	/*
3146 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3147 	 * occur if we called in6_setscope() at transmission.
3148 	 * See comments at top of file.
3149 	 */
3150 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3151 #endif
3152 
3153 	/*
3154 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3155 	 * so we can bump the stats.
3156 	 */
3157 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3158 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3159 	type = mld->mld_type;
3160 
3161 	oifp = NULL;
3162 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3163 	    &oifp, NULL);
3164 	if (error) {
3165 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3166 		goto out;
3167 	}
3168 	ICMP6STAT_INC(icp6s_outhist[type]);
3169 	if (oifp != NULL) {
3170 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3171 		switch (type) {
3172 		case MLD_LISTENER_REPORT:
3173 		case MLDV2_LISTENER_REPORT:
3174 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3175 			break;
3176 		case MLD_LISTENER_DONE:
3177 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3178 			break;
3179 		}
3180 	}
3181 out:
3182 	return;
3183 }
3184 
3185 /*
3186  * Encapsulate an MLDv2 report.
3187  *
3188  * KAME IPv6 requires that hop-by-hop options be passed separately,
3189  * and that the IPv6 header be prepended in a separate mbuf.
3190  *
3191  * Returns a pointer to the new mbuf chain head, or NULL if the
3192  * allocation failed.
3193  */
3194 static struct mbuf *
3195 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3196 {
3197 	struct mbuf		*mh;
3198 	struct mldv2_report	*mld;
3199 	struct ip6_hdr		*ip6;
3200 	struct in6_ifaddr	*ia;
3201 	int			 mldreclen;
3202 
3203 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3204 	KASSERT((m->m_flags & M_PKTHDR),
3205 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3206 
3207 	/*
3208 	 * RFC3590: OK to send as :: or tentative during DAD.
3209 	 */
3210 	NET_EPOCH_ASSERT();
3211 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3212 	if (ia == NULL)
3213 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3214 
3215 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3216 	if (mh == NULL) {
3217 		if (ia != NULL)
3218 			ifa_free(&ia->ia_ifa);
3219 		m_freem(m);
3220 		return (NULL);
3221 	}
3222 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3223 
3224 	mldreclen = m_length(m, NULL);
3225 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3226 
3227 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3228 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3229 	    sizeof(struct mldv2_report) + mldreclen;
3230 
3231 	ip6 = mtod(mh, struct ip6_hdr *);
3232 	ip6->ip6_flow = 0;
3233 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3234 	ip6->ip6_vfc |= IPV6_VERSION;
3235 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3236 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3237 	if (ia != NULL)
3238 		ifa_free(&ia->ia_ifa);
3239 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3240 	/* scope ID will be set in netisr */
3241 
3242 	mld = (struct mldv2_report *)(ip6 + 1);
3243 	mld->mld_type = MLDV2_LISTENER_REPORT;
3244 	mld->mld_code = 0;
3245 	mld->mld_cksum = 0;
3246 	mld->mld_v2_reserved = 0;
3247 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3248 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3249 
3250 	mh->m_next = m;
3251 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3252 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3253 	return (mh);
3254 }
3255 
3256 #ifdef KTR
3257 static char *
3258 mld_rec_type_to_str(const int type)
3259 {
3260 
3261 	switch (type) {
3262 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3263 			return "TO_EX";
3264 			break;
3265 		case MLD_CHANGE_TO_INCLUDE_MODE:
3266 			return "TO_IN";
3267 			break;
3268 		case MLD_MODE_IS_EXCLUDE:
3269 			return "MODE_EX";
3270 			break;
3271 		case MLD_MODE_IS_INCLUDE:
3272 			return "MODE_IN";
3273 			break;
3274 		case MLD_ALLOW_NEW_SOURCES:
3275 			return "ALLOW_NEW";
3276 			break;
3277 		case MLD_BLOCK_OLD_SOURCES:
3278 			return "BLOCK_OLD";
3279 			break;
3280 		default:
3281 			break;
3282 	}
3283 	return "unknown";
3284 }
3285 #endif
3286 
3287 static void
3288 mld_init(void *unused __unused)
3289 {
3290 
3291 	CTR1(KTR_MLD, "%s: initializing", __func__);
3292 	MLD_LOCK_INIT();
3293 
3294 	ip6_initpktopts(&mld_po);
3295 	mld_po.ip6po_hlim = 1;
3296 	mld_po.ip6po_hbh = &mld_ra.hbh;
3297 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3298 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3299 }
3300 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3301 
3302 static void
3303 mld_uninit(void *unused __unused)
3304 {
3305 
3306 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3307 	MLD_LOCK_DESTROY();
3308 }
3309 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3310 
3311 static void
3312 vnet_mld_init(const void *unused __unused)
3313 {
3314 
3315 	CTR1(KTR_MLD, "%s: initializing", __func__);
3316 
3317 	LIST_INIT(&V_mli_head);
3318 }
3319 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3320     NULL);
3321 
3322 static void
3323 vnet_mld_uninit(const void *unused __unused)
3324 {
3325 
3326 	/* This can happen if we shutdown the network stack. */
3327 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3328 }
3329 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3330     NULL);
3331 
3332 static int
3333 mld_modevent(module_t mod, int type, void *unused __unused)
3334 {
3335 
3336     switch (type) {
3337     case MOD_LOAD:
3338     case MOD_UNLOAD:
3339 	break;
3340     default:
3341 	return (EOPNOTSUPP);
3342     }
3343     return (0);
3344 }
3345 
3346 static moduledata_t mld_mod = {
3347     "mld",
3348     mld_modevent,
3349     0
3350 };
3351 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3352