xref: /freebsd/sys/netinet6/mld6.c (revision 675be9115aae86ad6b3d877155d4fd7822892105)
1 /*-
2  * Copyright (c) 2009 Bruce Simpson.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. The name of the author may not be used to endorse or promote
13  *    products derived from this software without specific prior written
14  *    permission.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
29  */
30 
31 /*-
32  * Copyright (c) 1988 Stephen Deering.
33  * Copyright (c) 1992, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * Stephen Deering of Stanford University.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 4. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
64  */
65 
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
68 
69 #include "opt_inet.h"
70 #include "opt_inet6.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/mbuf.h>
75 #include <sys/socket.h>
76 #include <sys/protosw.h>
77 #include <sys/sysctl.h>
78 #include <sys/kernel.h>
79 #include <sys/callout.h>
80 #include <sys/malloc.h>
81 #include <sys/module.h>
82 #include <sys/ktr.h>
83 
84 #include <net/if.h>
85 #include <net/route.h>
86 #include <net/vnet.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90 #include <netinet6/in6_var.h>
91 #include <netinet/ip6.h>
92 #include <netinet6/ip6_var.h>
93 #include <netinet6/scope6_var.h>
94 #include <netinet/icmp6.h>
95 #include <netinet6/mld6.h>
96 #include <netinet6/mld6_var.h>
97 
98 #include <security/mac/mac_framework.h>
99 
100 #ifndef KTR_MLD
101 #define KTR_MLD KTR_INET6
102 #endif
103 
104 static struct mld_ifinfo *
105 		mli_alloc_locked(struct ifnet *);
106 static void	mli_delete_locked(const struct ifnet *);
107 static void	mld_dispatch_packet(struct mbuf *);
108 static void	mld_dispatch_queue(struct ifqueue *, int);
109 static void	mld_final_leave(struct in6_multi *, struct mld_ifinfo *);
110 static void	mld_fasttimo_vnet(void);
111 static int	mld_handle_state_change(struct in6_multi *,
112 		    struct mld_ifinfo *);
113 static int	mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
114 		    const int);
115 #ifdef KTR
116 static char *	mld_rec_type_to_str(const int);
117 #endif
118 static void	mld_set_version(struct mld_ifinfo *, const int);
119 static void	mld_slowtimo_vnet(void);
120 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
121 		    /*const*/ struct mld_hdr *);
122 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
123 		    /*const*/ struct mld_hdr *);
124 static void	mld_v1_process_group_timer(struct in6_multi *, const int);
125 static void	mld_v1_process_querier_timers(struct mld_ifinfo *);
126 static int	mld_v1_transmit_report(struct in6_multi *, const int);
127 static void	mld_v1_update_group(struct in6_multi *, const int);
128 static void	mld_v2_cancel_link_timers(struct mld_ifinfo *);
129 static void	mld_v2_dispatch_general_query(struct mld_ifinfo *);
130 static struct mbuf *
131 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
132 static int	mld_v2_enqueue_filter_change(struct ifqueue *,
133 		    struct in6_multi *);
134 static int	mld_v2_enqueue_group_record(struct ifqueue *,
135 		    struct in6_multi *, const int, const int, const int,
136 		    const int);
137 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
138 		    struct mbuf *, const int, const int);
139 static int	mld_v2_merge_state_changes(struct in6_multi *,
140 		    struct ifqueue *);
141 static void	mld_v2_process_group_timers(struct mld_ifinfo *,
142 		    struct ifqueue *, struct ifqueue *,
143 		    struct in6_multi *, const int);
144 static int	mld_v2_process_group_query(struct in6_multi *,
145 		    struct mld_ifinfo *mli, int, struct mbuf *, const int);
146 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
147 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
148 
149 /*
150  * Normative references: RFC 2710, RFC 3590, RFC 3810.
151  *
152  * Locking:
153  *  * The MLD subsystem lock ends up being system-wide for the moment,
154  *    but could be per-VIMAGE later on.
155  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
156  *    Any may be taken independently; if any are held at the same
157  *    time, the above lock order must be followed.
158  *  * IN6_MULTI_LOCK covers in_multi.
159  *  * MLD_LOCK covers per-link state and any global variables in this file.
160  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
161  *    per-link state iterators.
162  *
163  *  XXX LOR PREVENTION
164  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
165  *  will not accept an ifp; it wants an embedded scope ID, unlike
166  *  ip_output(), which happily takes the ifp given to it. The embedded
167  *  scope ID is only used by MLD to select the outgoing interface.
168  *
169  *  During interface attach and detach, MLD will take MLD_LOCK *after*
170  *  the IF_AFDATA_LOCK.
171  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
172  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
173  *  dispatch could work around this, but we'd rather not do that, as it
174  *  can introduce other races.
175  *
176  *  As such, we exploit the fact that the scope ID is just the interface
177  *  index, and embed it in the IPv6 destination address accordingly.
178  *  This is potentially NOT VALID for MLDv1 reports, as they
179  *  are always sent to the multicast group itself; as MLDv2
180  *  reports are always sent to ff02::16, this is not an issue
181  *  when MLDv2 is in use.
182  *
183  *  This does not however eliminate the LOR when ip6_output() itself
184  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
185  *  trigger a LOR warning in WITNESS when the ifnet is detached.
186  *
187  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
188  *  how it's used across the network stack. Here we're simply exploiting
189  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
190  *
191  * VIMAGE:
192  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
193  *    to a vnet in ifp->if_vnet.
194  */
195 static struct mtx		 mld_mtx;
196 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
197 
198 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
199 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
200 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
201 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
202 
203 /*
204  * VIMAGE-wide globals.
205  */
206 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0};
207 static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head);
208 static VNET_DEFINE(int, interface_timers_running6);
209 static VNET_DEFINE(int, state_change_timers_running6);
210 static VNET_DEFINE(int, current_state_timers_running6);
211 
212 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
213 #define	V_mli_head			VNET(mli_head)
214 #define	V_interface_timers_running6	VNET(interface_timers_running6)
215 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
216 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
217 
218 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
219 
220 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
221     "IPv6 Multicast Listener Discovery");
222 
223 /*
224  * Virtualized sysctls.
225  */
226 SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
227     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
228     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
229     "Rate limit for MLDv2 Group-and-Source queries in seconds");
230 
231 /*
232  * Non-virtualized sysctls.
233  */
234 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
235     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
236     "Per-interface MLDv2 state");
237 
238 static int	mld_v1enable = 1;
239 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW,
240     &mld_v1enable, 0, "Enable fallback to MLDv1");
241 TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable);
242 
243 static int	mld_use_allow = 1;
244 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW,
245     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
246 TUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow);
247 
248 /*
249  * Packed Router Alert option structure declaration.
250  */
251 struct mld_raopt {
252 	struct ip6_hbh		hbh;
253 	struct ip6_opt		pad;
254 	struct ip6_opt_router	ra;
255 } __packed;
256 
257 /*
258  * Router Alert hop-by-hop option header.
259  */
260 static struct mld_raopt mld_ra = {
261 	.hbh = { 0, 0 },
262 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
263 	.ra = {
264 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
265 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
266 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
267 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
268 	}
269 };
270 static struct ip6_pktopts mld_po;
271 
272 static __inline void
273 mld_save_context(struct mbuf *m, struct ifnet *ifp)
274 {
275 
276 #ifdef VIMAGE
277 	m->m_pkthdr.header = ifp->if_vnet;
278 #endif /* VIMAGE */
279 	m->m_pkthdr.flowid = ifp->if_index;
280 }
281 
282 static __inline void
283 mld_scrub_context(struct mbuf *m)
284 {
285 
286 	m->m_pkthdr.header = NULL;
287 	m->m_pkthdr.flowid = 0;
288 }
289 
290 /*
291  * Restore context from a queued output chain.
292  * Return saved ifindex.
293  *
294  * VIMAGE: The assertion is there to make sure that we
295  * actually called CURVNET_SET() with what's in the mbuf chain.
296  */
297 static __inline uint32_t
298 mld_restore_context(struct mbuf *m)
299 {
300 
301 #if defined(VIMAGE) && defined(INVARIANTS)
302 	KASSERT(curvnet == m->m_pkthdr.header,
303 	    ("%s: called when curvnet was not restored", __func__));
304 #endif
305 	return (m->m_pkthdr.flowid);
306 }
307 
308 /*
309  * Retrieve or set threshold between group-source queries in seconds.
310  *
311  * VIMAGE: Assume curvnet set by caller.
312  * SMPng: NOTE: Serialized by MLD lock.
313  */
314 static int
315 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
316 {
317 	int error;
318 	int i;
319 
320 	error = sysctl_wire_old_buffer(req, sizeof(int));
321 	if (error)
322 		return (error);
323 
324 	MLD_LOCK();
325 
326 	i = V_mld_gsrdelay.tv_sec;
327 
328 	error = sysctl_handle_int(oidp, &i, 0, req);
329 	if (error || !req->newptr)
330 		goto out_locked;
331 
332 	if (i < -1 || i >= 60) {
333 		error = EINVAL;
334 		goto out_locked;
335 	}
336 
337 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
338 	     V_mld_gsrdelay.tv_sec, i);
339 	V_mld_gsrdelay.tv_sec = i;
340 
341 out_locked:
342 	MLD_UNLOCK();
343 	return (error);
344 }
345 
346 /*
347  * Expose struct mld_ifinfo to userland, keyed by ifindex.
348  * For use by ifmcstat(8).
349  *
350  * SMPng: NOTE: Does an unlocked ifindex space read.
351  * VIMAGE: Assume curvnet set by caller. The node handler itself
352  * is not directly virtualized.
353  */
354 static int
355 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
356 {
357 	int			*name;
358 	int			 error;
359 	u_int			 namelen;
360 	struct ifnet		*ifp;
361 	struct mld_ifinfo	*mli;
362 
363 	name = (int *)arg1;
364 	namelen = arg2;
365 
366 	if (req->newptr != NULL)
367 		return (EPERM);
368 
369 	if (namelen != 1)
370 		return (EINVAL);
371 
372 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
373 	if (error)
374 		return (error);
375 
376 	IN6_MULTI_LOCK();
377 	MLD_LOCK();
378 
379 	if (name[0] <= 0 || name[0] > V_if_index) {
380 		error = ENOENT;
381 		goto out_locked;
382 	}
383 
384 	error = ENOENT;
385 
386 	ifp = ifnet_byindex(name[0]);
387 	if (ifp == NULL)
388 		goto out_locked;
389 
390 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
391 		if (ifp == mli->mli_ifp) {
392 			error = SYSCTL_OUT(req, mli,
393 			    sizeof(struct mld_ifinfo));
394 			break;
395 		}
396 	}
397 
398 out_locked:
399 	MLD_UNLOCK();
400 	IN6_MULTI_UNLOCK();
401 	return (error);
402 }
403 
404 /*
405  * Dispatch an entire queue of pending packet chains.
406  * VIMAGE: Assumes the vnet pointer has been set.
407  */
408 static void
409 mld_dispatch_queue(struct ifqueue *ifq, int limit)
410 {
411 	struct mbuf *m;
412 
413 	for (;;) {
414 		_IF_DEQUEUE(ifq, m);
415 		if (m == NULL)
416 			break;
417 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m);
418 		mld_dispatch_packet(m);
419 		if (--limit == 0)
420 			break;
421 	}
422 }
423 
424 /*
425  * Filter outgoing MLD report state by group.
426  *
427  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
428  * and node-local addresses. However, kernel and socket consumers
429  * always embed the KAME scope ID in the address provided, so strip it
430  * when performing comparison.
431  * Note: This is not the same as the *multicast* scope.
432  *
433  * Return zero if the given group is one for which MLD reports
434  * should be suppressed, or non-zero if reports should be issued.
435  */
436 static __inline int
437 mld_is_addr_reported(const struct in6_addr *addr)
438 {
439 
440 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
441 
442 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
443 		return (0);
444 
445 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
446 		struct in6_addr tmp = *addr;
447 		in6_clearscope(&tmp);
448 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
449 			return (0);
450 	}
451 
452 	return (1);
453 }
454 
455 /*
456  * Attach MLD when PF_INET6 is attached to an interface.
457  *
458  * SMPng: Normally called with IF_AFDATA_LOCK held.
459  */
460 struct mld_ifinfo *
461 mld_domifattach(struct ifnet *ifp)
462 {
463 	struct mld_ifinfo *mli;
464 
465 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
466 	    __func__, ifp, ifp->if_xname);
467 
468 	MLD_LOCK();
469 
470 	mli = mli_alloc_locked(ifp);
471 	if (!(ifp->if_flags & IFF_MULTICAST))
472 		mli->mli_flags |= MLIF_SILENT;
473 	if (mld_use_allow)
474 		mli->mli_flags |= MLIF_USEALLOW;
475 
476 	MLD_UNLOCK();
477 
478 	return (mli);
479 }
480 
481 /*
482  * VIMAGE: assume curvnet set by caller.
483  */
484 static struct mld_ifinfo *
485 mli_alloc_locked(/*const*/ struct ifnet *ifp)
486 {
487 	struct mld_ifinfo *mli;
488 
489 	MLD_LOCK_ASSERT();
490 
491 	mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO);
492 	if (mli == NULL)
493 		goto out;
494 
495 	mli->mli_ifp = ifp;
496 	mli->mli_version = MLD_VERSION_2;
497 	mli->mli_flags = 0;
498 	mli->mli_rv = MLD_RV_INIT;
499 	mli->mli_qi = MLD_QI_INIT;
500 	mli->mli_qri = MLD_QRI_INIT;
501 	mli->mli_uri = MLD_URI_INIT;
502 
503 	SLIST_INIT(&mli->mli_relinmhead);
504 
505 	/*
506 	 * Responses to general queries are subject to bounds.
507 	 */
508 	IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
509 
510 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
511 
512 	CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)",
513 	     ifp, ifp->if_xname);
514 
515 out:
516 	return (mli);
517 }
518 
519 /*
520  * Hook for ifdetach.
521  *
522  * NOTE: Some finalization tasks need to run before the protocol domain
523  * is detached, but also before the link layer does its cleanup.
524  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
525  *
526  * SMPng: Caller must hold IN6_MULTI_LOCK().
527  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
528  * XXX This routine is also bitten by unlocked ifma_protospec access.
529  */
530 void
531 mld_ifdetach(struct ifnet *ifp)
532 {
533 	struct mld_ifinfo	*mli;
534 	struct ifmultiaddr	*ifma;
535 	struct in6_multi	*inm, *tinm;
536 
537 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
538 	    ifp->if_xname);
539 
540 	IN6_MULTI_LOCK_ASSERT();
541 	MLD_LOCK();
542 
543 	mli = MLD_IFINFO(ifp);
544 	if (mli->mli_version == MLD_VERSION_2) {
545 		IF_ADDR_LOCK(ifp);
546 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
547 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
548 			    ifma->ifma_protospec == NULL)
549 				continue;
550 			inm = (struct in6_multi *)ifma->ifma_protospec;
551 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
552 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
553 				    inm, in6m_nrele);
554 			}
555 			in6m_clear_recorded(inm);
556 		}
557 		IF_ADDR_UNLOCK(ifp);
558 		SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele,
559 		    tinm) {
560 			SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
561 			in6m_release_locked(inm);
562 		}
563 	}
564 
565 	MLD_UNLOCK();
566 }
567 
568 /*
569  * Hook for domifdetach.
570  * Runs after link-layer cleanup; free MLD state.
571  *
572  * SMPng: Normally called with IF_AFDATA_LOCK held.
573  */
574 void
575 mld_domifdetach(struct ifnet *ifp)
576 {
577 
578 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
579 	    __func__, ifp, ifp->if_xname);
580 
581 	MLD_LOCK();
582 	mli_delete_locked(ifp);
583 	MLD_UNLOCK();
584 }
585 
586 static void
587 mli_delete_locked(const struct ifnet *ifp)
588 {
589 	struct mld_ifinfo *mli, *tmli;
590 
591 	CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)",
592 	    __func__, ifp, ifp->if_xname);
593 
594 	MLD_LOCK_ASSERT();
595 
596 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
597 		if (mli->mli_ifp == ifp) {
598 			/*
599 			 * Free deferred General Query responses.
600 			 */
601 			_IF_DRAIN(&mli->mli_gq);
602 
603 			LIST_REMOVE(mli, mli_link);
604 
605 			KASSERT(SLIST_EMPTY(&mli->mli_relinmhead),
606 			    ("%s: there are dangling in_multi references",
607 			    __func__));
608 
609 			free(mli, M_MLD);
610 			return;
611 		}
612 	}
613 #ifdef INVARIANTS
614 	panic("%s: mld_ifinfo not found for ifp %p\n", __func__,  ifp);
615 #endif
616 }
617 
618 /*
619  * Process a received MLDv1 general or address-specific query.
620  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
621  *
622  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
623  * mld_addr. This is OK as we own the mbuf chain.
624  */
625 static int
626 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
627     /*const*/ struct mld_hdr *mld)
628 {
629 	struct ifmultiaddr	*ifma;
630 	struct mld_ifinfo	*mli;
631 	struct in6_multi	*inm;
632 	int			 is_general_query;
633 	uint16_t		 timer;
634 #ifdef KTR
635 	char			 ip6tbuf[INET6_ADDRSTRLEN];
636 #endif
637 
638 	is_general_query = 0;
639 
640 	if (!mld_v1enable) {
641 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
642 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
643 		    ifp, ifp->if_xname);
644 		return (0);
645 	}
646 
647 	/*
648 	 * RFC3810 Section 6.2: MLD queries must originate from
649 	 * a router's link-local address.
650 	 */
651 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
652 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
653 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
654 		    ifp, ifp->if_xname);
655 		return (0);
656 	}
657 
658 	/*
659 	 * Do address field validation upfront before we accept
660 	 * the query.
661 	 */
662 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
663 		/*
664 		 * MLDv1 General Query.
665 		 * If this was not sent to the all-nodes group, ignore it.
666 		 */
667 		struct in6_addr		 dst;
668 
669 		dst = ip6->ip6_dst;
670 		in6_clearscope(&dst);
671 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
672 			return (EINVAL);
673 		is_general_query = 1;
674 	} else {
675 		/*
676 		 * Embed scope ID of receiving interface in MLD query for
677 		 * lookup whilst we don't hold other locks.
678 		 */
679 		in6_setscope(&mld->mld_addr, ifp, NULL);
680 	}
681 
682 	IN6_MULTI_LOCK();
683 	MLD_LOCK();
684 
685 	/*
686 	 * Switch to MLDv1 host compatibility mode.
687 	 */
688 	mli = MLD_IFINFO(ifp);
689 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
690 	mld_set_version(mli, MLD_VERSION_1);
691 
692 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
693 	if (timer == 0)
694 		timer = 1;
695 
696 	IF_ADDR_LOCK(ifp);
697 	if (is_general_query) {
698 		/*
699 		 * For each reporting group joined on this
700 		 * interface, kick the report timer.
701 		 */
702 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
703 		    ifp, ifp->if_xname);
704 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
705 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
706 			    ifma->ifma_protospec == NULL)
707 				continue;
708 			inm = (struct in6_multi *)ifma->ifma_protospec;
709 			mld_v1_update_group(inm, timer);
710 		}
711 	} else {
712 		/*
713 		 * MLDv1 Group-Specific Query.
714 		 * If this is a group-specific MLDv1 query, we need only
715 		 * look up the single group to process it.
716 		 */
717 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
718 		if (inm != NULL) {
719 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
720 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
721 			    ifp, ifp->if_xname);
722 			mld_v1_update_group(inm, timer);
723 		}
724 		/* XXX Clear embedded scope ID as userland won't expect it. */
725 		in6_clearscope(&mld->mld_addr);
726 	}
727 
728 	IF_ADDR_UNLOCK(ifp);
729 	MLD_UNLOCK();
730 	IN6_MULTI_UNLOCK();
731 
732 	return (0);
733 }
734 
735 /*
736  * Update the report timer on a group in response to an MLDv1 query.
737  *
738  * If we are becoming the reporting member for this group, start the timer.
739  * If we already are the reporting member for this group, and timer is
740  * below the threshold, reset it.
741  *
742  * We may be updating the group for the first time since we switched
743  * to MLDv2. If we are, then we must clear any recorded source lists,
744  * and transition to REPORTING state; the group timer is overloaded
745  * for group and group-source query responses.
746  *
747  * Unlike MLDv2, the delay per group should be jittered
748  * to avoid bursts of MLDv1 reports.
749  */
750 static void
751 mld_v1_update_group(struct in6_multi *inm, const int timer)
752 {
753 #ifdef KTR
754 	char			 ip6tbuf[INET6_ADDRSTRLEN];
755 #endif
756 
757 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
758 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
759 	    inm->in6m_ifp->if_xname, timer);
760 
761 	IN6_MULTI_LOCK_ASSERT();
762 
763 	switch (inm->in6m_state) {
764 	case MLD_NOT_MEMBER:
765 	case MLD_SILENT_MEMBER:
766 		break;
767 	case MLD_REPORTING_MEMBER:
768 		if (inm->in6m_timer != 0 &&
769 		    inm->in6m_timer <= timer) {
770 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
771 			    "skipping.", __func__);
772 			break;
773 		}
774 		/* FALLTHROUGH */
775 	case MLD_SG_QUERY_PENDING_MEMBER:
776 	case MLD_G_QUERY_PENDING_MEMBER:
777 	case MLD_IDLE_MEMBER:
778 	case MLD_LAZY_MEMBER:
779 	case MLD_AWAKENING_MEMBER:
780 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
781 		inm->in6m_state = MLD_REPORTING_MEMBER;
782 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
783 		V_current_state_timers_running6 = 1;
784 		break;
785 	case MLD_SLEEPING_MEMBER:
786 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
787 		inm->in6m_state = MLD_AWAKENING_MEMBER;
788 		break;
789 	case MLD_LEAVING_MEMBER:
790 		break;
791 	}
792 }
793 
794 /*
795  * Process a received MLDv2 general, group-specific or
796  * group-and-source-specific query.
797  *
798  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
799  *
800  * Return 0 if successful, otherwise an appropriate error code is returned.
801  */
802 static int
803 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
804     struct mbuf *m, const int off, const int icmp6len)
805 {
806 	struct mld_ifinfo	*mli;
807 	struct mldv2_query	*mld;
808 	struct in6_multi	*inm;
809 	uint32_t		 maxdelay, nsrc, qqi;
810 	int			 is_general_query;
811 	uint16_t		 timer;
812 	uint8_t			 qrv;
813 #ifdef KTR
814 	char			 ip6tbuf[INET6_ADDRSTRLEN];
815 #endif
816 
817 	is_general_query = 0;
818 
819 	/*
820 	 * RFC3810 Section 6.2: MLD queries must originate from
821 	 * a router's link-local address.
822 	 */
823 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
824 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
825 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
826 		    ifp, ifp->if_xname);
827 		return (0);
828 	}
829 
830 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname);
831 
832 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
833 
834 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
835 	if (maxdelay >= 32678) {
836 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
837 			   (MLD_MRC_EXP(maxdelay) + 3);
838 	}
839 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
840 	if (timer == 0)
841 		timer = 1;
842 
843 	qrv = MLD_QRV(mld->mld_misc);
844 	if (qrv < 2) {
845 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
846 		    qrv, MLD_RV_INIT);
847 		qrv = MLD_RV_INIT;
848 	}
849 
850 	qqi = mld->mld_qqi;
851 	if (qqi >= 128) {
852 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
853 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
854 	}
855 
856 	nsrc = ntohs(mld->mld_numsrc);
857 	if (nsrc > MLD_MAX_GS_SOURCES)
858 		return (EMSGSIZE);
859 	if (icmp6len < sizeof(struct mldv2_query) +
860 	    (nsrc * sizeof(struct in6_addr)))
861 		return (EMSGSIZE);
862 
863 	/*
864 	 * Do further input validation upfront to avoid resetting timers
865 	 * should we need to discard this query.
866 	 */
867 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
868 		/*
869 		 * General Queries SHOULD be directed to ff02::1.
870 		 * A general query with a source list has undefined
871 		 * behaviour; discard it.
872 		 */
873 		struct in6_addr		 dst;
874 
875 		dst = ip6->ip6_dst;
876 		in6_clearscope(&dst);
877 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) ||
878 		    nsrc > 0)
879 			return (EINVAL);
880 		is_general_query = 1;
881 	} else {
882 		/*
883 		 * Embed scope ID of receiving interface in MLD query for
884 		 * lookup whilst we don't hold other locks (due to KAME
885 		 * locking lameness). We own this mbuf chain just now.
886 		 */
887 		in6_setscope(&mld->mld_addr, ifp, NULL);
888 	}
889 
890 	IN6_MULTI_LOCK();
891 	MLD_LOCK();
892 
893 	mli = MLD_IFINFO(ifp);
894 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
895 
896 	/*
897 	 * Discard the v2 query if we're in Compatibility Mode.
898 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
899 	 * until the Old Version Querier Present timer expires.
900 	 */
901 	if (mli->mli_version != MLD_VERSION_2)
902 		goto out_locked;
903 
904 	mld_set_version(mli, MLD_VERSION_2);
905 	mli->mli_rv = qrv;
906 	mli->mli_qi = qqi;
907 	mli->mli_qri = maxdelay;
908 
909 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
910 	    maxdelay);
911 
912 	if (is_general_query) {
913 		/*
914 		 * MLDv2 General Query.
915 		 *
916 		 * Schedule a current-state report on this ifp for
917 		 * all groups, possibly containing source lists.
918 		 *
919 		 * If there is a pending General Query response
920 		 * scheduled earlier than the selected delay, do
921 		 * not schedule any other reports.
922 		 * Otherwise, reset the interface timer.
923 		 */
924 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
925 		    ifp, ifp->if_xname);
926 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
927 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
928 			V_interface_timers_running6 = 1;
929 		}
930 	} else {
931 		/*
932 		 * MLDv2 Group-specific or Group-and-source-specific Query.
933 		 *
934 		 * Group-source-specific queries are throttled on
935 		 * a per-group basis to defeat denial-of-service attempts.
936 		 * Queries for groups we are not a member of on this
937 		 * link are simply ignored.
938 		 */
939 		IF_ADDR_LOCK(ifp);
940 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
941 		if (inm == NULL) {
942 			IF_ADDR_UNLOCK(ifp);
943 			goto out_locked;
944 		}
945 		if (nsrc > 0) {
946 			if (!ratecheck(&inm->in6m_lastgsrtv,
947 			    &V_mld_gsrdelay)) {
948 				CTR1(KTR_MLD, "%s: GS query throttled.",
949 				    __func__);
950 				IF_ADDR_UNLOCK(ifp);
951 				goto out_locked;
952 			}
953 		}
954 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
955 		     ifp, ifp->if_xname);
956 		/*
957 		 * If there is a pending General Query response
958 		 * scheduled sooner than the selected delay, no
959 		 * further report need be scheduled.
960 		 * Otherwise, prepare to respond to the
961 		 * group-specific or group-and-source query.
962 		 */
963 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
964 			mld_v2_process_group_query(inm, mli, timer, m, off);
965 
966 		/* XXX Clear embedded scope ID as userland won't expect it. */
967 		in6_clearscope(&mld->mld_addr);
968 		IF_ADDR_UNLOCK(ifp);
969 	}
970 
971 out_locked:
972 	MLD_UNLOCK();
973 	IN6_MULTI_UNLOCK();
974 
975 	return (0);
976 }
977 
978 /*
979  * Process a recieved MLDv2 group-specific or group-and-source-specific
980  * query.
981  * Return <0 if any error occured. Currently this is ignored.
982  */
983 static int
984 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli,
985     int timer, struct mbuf *m0, const int off)
986 {
987 	struct mldv2_query	*mld;
988 	int			 retval;
989 	uint16_t		 nsrc;
990 
991 	IN6_MULTI_LOCK_ASSERT();
992 	MLD_LOCK_ASSERT();
993 
994 	retval = 0;
995 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
996 
997 	switch (inm->in6m_state) {
998 	case MLD_NOT_MEMBER:
999 	case MLD_SILENT_MEMBER:
1000 	case MLD_SLEEPING_MEMBER:
1001 	case MLD_LAZY_MEMBER:
1002 	case MLD_AWAKENING_MEMBER:
1003 	case MLD_IDLE_MEMBER:
1004 	case MLD_LEAVING_MEMBER:
1005 		return (retval);
1006 		break;
1007 	case MLD_REPORTING_MEMBER:
1008 	case MLD_G_QUERY_PENDING_MEMBER:
1009 	case MLD_SG_QUERY_PENDING_MEMBER:
1010 		break;
1011 	}
1012 
1013 	nsrc = ntohs(mld->mld_numsrc);
1014 
1015 	/*
1016 	 * Deal with group-specific queries upfront.
1017 	 * If any group query is already pending, purge any recorded
1018 	 * source-list state if it exists, and schedule a query response
1019 	 * for this group-specific query.
1020 	 */
1021 	if (nsrc == 0) {
1022 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1023 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1024 			in6m_clear_recorded(inm);
1025 			timer = min(inm->in6m_timer, timer);
1026 		}
1027 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1028 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1029 		V_current_state_timers_running6 = 1;
1030 		return (retval);
1031 	}
1032 
1033 	/*
1034 	 * Deal with the case where a group-and-source-specific query has
1035 	 * been received but a group-specific query is already pending.
1036 	 */
1037 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1038 		timer = min(inm->in6m_timer, timer);
1039 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1040 		V_current_state_timers_running6 = 1;
1041 		return (retval);
1042 	}
1043 
1044 	/*
1045 	 * Finally, deal with the case where a group-and-source-specific
1046 	 * query has been received, where a response to a previous g-s-r
1047 	 * query exists, or none exists.
1048 	 * In this case, we need to parse the source-list which the Querier
1049 	 * has provided us with and check if we have any source list filter
1050 	 * entries at T1 for these sources. If we do not, there is no need
1051 	 * schedule a report and the query may be dropped.
1052 	 * If we do, we must record them and schedule a current-state
1053 	 * report for those sources.
1054 	 */
1055 	if (inm->in6m_nsrc > 0) {
1056 		struct mbuf		*m;
1057 		uint8_t			*sp;
1058 		int			 i, nrecorded;
1059 		int			 soff;
1060 
1061 		m = m0;
1062 		soff = off + sizeof(struct mldv2_query);
1063 		nrecorded = 0;
1064 		for (i = 0; i < nsrc; i++) {
1065 			sp = mtod(m, uint8_t *) + soff;
1066 			retval = in6m_record_source(inm,
1067 			    (const struct in6_addr *)sp);
1068 			if (retval < 0)
1069 				break;
1070 			nrecorded += retval;
1071 			soff += sizeof(struct in6_addr);
1072 			if (soff >= m->m_len) {
1073 				soff = soff - m->m_len;
1074 				m = m->m_next;
1075 				if (m == NULL)
1076 					break;
1077 			}
1078 		}
1079 		if (nrecorded > 0) {
1080 			CTR1(KTR_MLD,
1081 			    "%s: schedule response to SG query", __func__);
1082 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1083 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1084 			V_current_state_timers_running6 = 1;
1085 		}
1086 	}
1087 
1088 	return (retval);
1089 }
1090 
1091 /*
1092  * Process a received MLDv1 host membership report.
1093  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1094  *
1095  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1096  * mld_addr. This is OK as we own the mbuf chain.
1097  */
1098 static int
1099 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1100     /*const*/ struct mld_hdr *mld)
1101 {
1102 	struct in6_addr		 src, dst;
1103 	struct in6_ifaddr	*ia;
1104 	struct in6_multi	*inm;
1105 #ifdef KTR
1106 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1107 #endif
1108 
1109 	if (!mld_v1enable) {
1110 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1111 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1112 		    ifp, ifp->if_xname);
1113 		return (0);
1114 	}
1115 
1116 	if (ifp->if_flags & IFF_LOOPBACK)
1117 		return (0);
1118 
1119 	/*
1120 	 * MLDv1 reports must originate from a host's link-local address,
1121 	 * or the unspecified address (when booting).
1122 	 */
1123 	src = ip6->ip6_src;
1124 	in6_clearscope(&src);
1125 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1126 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1127 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1128 		    ifp, ifp->if_xname);
1129 		return (EINVAL);
1130 	}
1131 
1132 	/*
1133 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1134 	 * group, and must be directed to the group itself.
1135 	 */
1136 	dst = ip6->ip6_dst;
1137 	in6_clearscope(&dst);
1138 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1139 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1140 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1141 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1142 		    ifp, ifp->if_xname);
1143 		return (EINVAL);
1144 	}
1145 
1146 	/*
1147 	 * Make sure we don't hear our own membership report, as fast
1148 	 * leave requires knowing that we are the only member of a
1149 	 * group. Assume we used the link-local address if available,
1150 	 * otherwise look for ::.
1151 	 *
1152 	 * XXX Note that scope ID comparison is needed for the address
1153 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1154 	 * performed for the on-wire address.
1155 	 */
1156 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1157 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1158 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1159 		if (ia != NULL)
1160 			ifa_free(&ia->ia_ifa);
1161 		return (0);
1162 	}
1163 	if (ia != NULL)
1164 		ifa_free(&ia->ia_ifa);
1165 
1166 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1167 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname);
1168 
1169 	/*
1170 	 * Embed scope ID of receiving interface in MLD query for lookup
1171 	 * whilst we don't hold other locks (due to KAME locking lameness).
1172 	 */
1173 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1174 		in6_setscope(&mld->mld_addr, ifp, NULL);
1175 
1176 	IN6_MULTI_LOCK();
1177 	MLD_LOCK();
1178 	IF_ADDR_LOCK(ifp);
1179 
1180 	/*
1181 	 * MLDv1 report suppression.
1182 	 * If we are a member of this group, and our membership should be
1183 	 * reported, and our group timer is pending or about to be reset,
1184 	 * stop our group timer by transitioning to the 'lazy' state.
1185 	 */
1186 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1187 	if (inm != NULL) {
1188 		struct mld_ifinfo *mli;
1189 
1190 		mli = inm->in6m_mli;
1191 		KASSERT(mli != NULL,
1192 		    ("%s: no mli for ifp %p", __func__, ifp));
1193 
1194 		/*
1195 		 * If we are in MLDv2 host mode, do not allow the
1196 		 * other host's MLDv1 report to suppress our reports.
1197 		 */
1198 		if (mli->mli_version == MLD_VERSION_2)
1199 			goto out_locked;
1200 
1201 		inm->in6m_timer = 0;
1202 
1203 		switch (inm->in6m_state) {
1204 		case MLD_NOT_MEMBER:
1205 		case MLD_SILENT_MEMBER:
1206 		case MLD_SLEEPING_MEMBER:
1207 			break;
1208 		case MLD_REPORTING_MEMBER:
1209 		case MLD_IDLE_MEMBER:
1210 		case MLD_AWAKENING_MEMBER:
1211 			CTR3(KTR_MLD,
1212 			    "report suppressed for %s on ifp %p(%s)",
1213 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1214 			    ifp, ifp->if_xname);
1215 		case MLD_LAZY_MEMBER:
1216 			inm->in6m_state = MLD_LAZY_MEMBER;
1217 			break;
1218 		case MLD_G_QUERY_PENDING_MEMBER:
1219 		case MLD_SG_QUERY_PENDING_MEMBER:
1220 		case MLD_LEAVING_MEMBER:
1221 			break;
1222 		}
1223 	}
1224 
1225 out_locked:
1226 	MLD_UNLOCK();
1227 	IF_ADDR_UNLOCK(ifp);
1228 	IN6_MULTI_UNLOCK();
1229 
1230 	/* XXX Clear embedded scope ID as userland won't expect it. */
1231 	in6_clearscope(&mld->mld_addr);
1232 
1233 	return (0);
1234 }
1235 
1236 /*
1237  * MLD input path.
1238  *
1239  * Assume query messages which fit in a single ICMPv6 message header
1240  * have been pulled up.
1241  * Assume that userland will want to see the message, even if it
1242  * otherwise fails kernel input validation; do not free it.
1243  * Pullup may however free the mbuf chain m if it fails.
1244  *
1245  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1246  */
1247 int
1248 mld_input(struct mbuf *m, int off, int icmp6len)
1249 {
1250 	struct ifnet	*ifp;
1251 	struct ip6_hdr	*ip6;
1252 	struct mld_hdr	*mld;
1253 	int		 mldlen;
1254 
1255 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1256 
1257 	ifp = m->m_pkthdr.rcvif;
1258 
1259 	ip6 = mtod(m, struct ip6_hdr *);
1260 
1261 	/* Pullup to appropriate size. */
1262 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1263 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1264 	    icmp6len >= sizeof(struct mldv2_query)) {
1265 		mldlen = sizeof(struct mldv2_query);
1266 	} else {
1267 		mldlen = sizeof(struct mld_hdr);
1268 	}
1269 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1270 	if (mld == NULL) {
1271 		ICMP6STAT_INC(icp6s_badlen);
1272 		return (IPPROTO_DONE);
1273 	}
1274 
1275 	/*
1276 	 * Userland needs to see all of this traffic for implementing
1277 	 * the endpoint discovery portion of multicast routing.
1278 	 */
1279 	switch (mld->mld_type) {
1280 	case MLD_LISTENER_QUERY:
1281 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1282 		if (icmp6len == sizeof(struct mld_hdr)) {
1283 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1284 				return (0);
1285 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1286 			if (mld_v2_input_query(ifp, ip6, m, off,
1287 			    icmp6len) != 0)
1288 				return (0);
1289 		}
1290 		break;
1291 	case MLD_LISTENER_REPORT:
1292 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1293 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1294 			return (0);
1295 		break;
1296 	case MLDV2_LISTENER_REPORT:
1297 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1298 		break;
1299 	case MLD_LISTENER_DONE:
1300 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1301 		break;
1302 	default:
1303 		break;
1304 	}
1305 
1306 	return (0);
1307 }
1308 
1309 /*
1310  * Fast timeout handler (global).
1311  * VIMAGE: Timeout handlers are expected to service all vimages.
1312  */
1313 void
1314 mld_fasttimo(void)
1315 {
1316 	VNET_ITERATOR_DECL(vnet_iter);
1317 
1318 	VNET_LIST_RLOCK_NOSLEEP();
1319 	VNET_FOREACH(vnet_iter) {
1320 		CURVNET_SET(vnet_iter);
1321 		mld_fasttimo_vnet();
1322 		CURVNET_RESTORE();
1323 	}
1324 	VNET_LIST_RUNLOCK_NOSLEEP();
1325 }
1326 
1327 /*
1328  * Fast timeout handler (per-vnet).
1329  *
1330  * VIMAGE: Assume caller has set up our curvnet.
1331  */
1332 static void
1333 mld_fasttimo_vnet(void)
1334 {
1335 	struct ifqueue		 scq;	/* State-change packets */
1336 	struct ifqueue		 qrq;	/* Query response packets */
1337 	struct ifnet		*ifp;
1338 	struct mld_ifinfo	*mli;
1339 	struct ifmultiaddr	*ifma, *tifma;
1340 	struct in6_multi	*inm;
1341 	int			 uri_fasthz;
1342 
1343 	uri_fasthz = 0;
1344 
1345 	/*
1346 	 * Quick check to see if any work needs to be done, in order to
1347 	 * minimize the overhead of fasttimo processing.
1348 	 * SMPng: XXX Unlocked reads.
1349 	 */
1350 	if (!V_current_state_timers_running6 &&
1351 	    !V_interface_timers_running6 &&
1352 	    !V_state_change_timers_running6)
1353 		return;
1354 
1355 	IN6_MULTI_LOCK();
1356 	MLD_LOCK();
1357 
1358 	/*
1359 	 * MLDv2 General Query response timer processing.
1360 	 */
1361 	if (V_interface_timers_running6) {
1362 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1363 
1364 		V_interface_timers_running6 = 0;
1365 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1366 			if (mli->mli_v2_timer == 0) {
1367 				/* Do nothing. */
1368 			} else if (--mli->mli_v2_timer == 0) {
1369 				mld_v2_dispatch_general_query(mli);
1370 			} else {
1371 				V_interface_timers_running6 = 1;
1372 			}
1373 		}
1374 	}
1375 
1376 	if (!V_current_state_timers_running6 &&
1377 	    !V_state_change_timers_running6)
1378 		goto out_locked;
1379 
1380 	V_current_state_timers_running6 = 0;
1381 	V_state_change_timers_running6 = 0;
1382 
1383 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1384 
1385 	/*
1386 	 * MLD host report and state-change timer processing.
1387 	 * Note: Processing a v2 group timer may remove a node.
1388 	 */
1389 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1390 		ifp = mli->mli_ifp;
1391 
1392 		if (mli->mli_version == MLD_VERSION_2) {
1393 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1394 			    PR_FASTHZ);
1395 
1396 			memset(&qrq, 0, sizeof(struct ifqueue));
1397 			IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS);
1398 
1399 			memset(&scq, 0, sizeof(struct ifqueue));
1400 			IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1401 		}
1402 
1403 		IF_ADDR_LOCK(ifp);
1404 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1405 		    tifma) {
1406 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1407 			    ifma->ifma_protospec == NULL)
1408 				continue;
1409 			inm = (struct in6_multi *)ifma->ifma_protospec;
1410 			switch (mli->mli_version) {
1411 			case MLD_VERSION_1:
1412 				/*
1413 				 * XXX Drop IF_ADDR lock temporarily to
1414 				 * avoid recursion caused by a potential
1415 				 * call by in6ifa_ifpforlinklocal().
1416 				 * rwlock candidate?
1417 				 */
1418 				IF_ADDR_UNLOCK(ifp);
1419 				mld_v1_process_group_timer(inm,
1420 				    mli->mli_version);
1421 				IF_ADDR_LOCK(ifp);
1422 				break;
1423 			case MLD_VERSION_2:
1424 				mld_v2_process_group_timers(mli, &qrq,
1425 				    &scq, inm, uri_fasthz);
1426 				break;
1427 			}
1428 		}
1429 		IF_ADDR_UNLOCK(ifp);
1430 
1431 		if (mli->mli_version == MLD_VERSION_2) {
1432 			struct in6_multi		*tinm;
1433 
1434 			mld_dispatch_queue(&qrq, 0);
1435 			mld_dispatch_queue(&scq, 0);
1436 
1437 			/*
1438 			 * Free the in_multi reference(s) for
1439 			 * this lifecycle.
1440 			 */
1441 			SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead,
1442 			    in6m_nrele, tinm) {
1443 				SLIST_REMOVE_HEAD(&mli->mli_relinmhead,
1444 				    in6m_nrele);
1445 				in6m_release_locked(inm);
1446 			}
1447 		}
1448 	}
1449 
1450 out_locked:
1451 	MLD_UNLOCK();
1452 	IN6_MULTI_UNLOCK();
1453 }
1454 
1455 /*
1456  * Update host report group timer.
1457  * Will update the global pending timer flags.
1458  */
1459 static void
1460 mld_v1_process_group_timer(struct in6_multi *inm, const int version)
1461 {
1462 	int report_timer_expired;
1463 
1464 	IN6_MULTI_LOCK_ASSERT();
1465 	MLD_LOCK_ASSERT();
1466 
1467 	if (inm->in6m_timer == 0) {
1468 		report_timer_expired = 0;
1469 	} else if (--inm->in6m_timer == 0) {
1470 		report_timer_expired = 1;
1471 	} else {
1472 		V_current_state_timers_running6 = 1;
1473 		return;
1474 	}
1475 
1476 	switch (inm->in6m_state) {
1477 	case MLD_NOT_MEMBER:
1478 	case MLD_SILENT_MEMBER:
1479 	case MLD_IDLE_MEMBER:
1480 	case MLD_LAZY_MEMBER:
1481 	case MLD_SLEEPING_MEMBER:
1482 	case MLD_AWAKENING_MEMBER:
1483 		break;
1484 	case MLD_REPORTING_MEMBER:
1485 		if (report_timer_expired) {
1486 			inm->in6m_state = MLD_IDLE_MEMBER;
1487 			(void)mld_v1_transmit_report(inm,
1488 			     MLD_LISTENER_REPORT);
1489 		}
1490 		break;
1491 	case MLD_G_QUERY_PENDING_MEMBER:
1492 	case MLD_SG_QUERY_PENDING_MEMBER:
1493 	case MLD_LEAVING_MEMBER:
1494 		break;
1495 	}
1496 }
1497 
1498 /*
1499  * Update a group's timers for MLDv2.
1500  * Will update the global pending timer flags.
1501  * Note: Unlocked read from mli.
1502  */
1503 static void
1504 mld_v2_process_group_timers(struct mld_ifinfo *mli,
1505     struct ifqueue *qrq, struct ifqueue *scq,
1506     struct in6_multi *inm, const int uri_fasthz)
1507 {
1508 	int query_response_timer_expired;
1509 	int state_change_retransmit_timer_expired;
1510 #ifdef KTR
1511 	char ip6tbuf[INET6_ADDRSTRLEN];
1512 #endif
1513 
1514 	IN6_MULTI_LOCK_ASSERT();
1515 	MLD_LOCK_ASSERT();
1516 
1517 	query_response_timer_expired = 0;
1518 	state_change_retransmit_timer_expired = 0;
1519 
1520 	/*
1521 	 * During a transition from compatibility mode back to MLDv2,
1522 	 * a group record in REPORTING state may still have its group
1523 	 * timer active. This is a no-op in this function; it is easier
1524 	 * to deal with it here than to complicate the slow-timeout path.
1525 	 */
1526 	if (inm->in6m_timer == 0) {
1527 		query_response_timer_expired = 0;
1528 	} else if (--inm->in6m_timer == 0) {
1529 		query_response_timer_expired = 1;
1530 	} else {
1531 		V_current_state_timers_running6 = 1;
1532 	}
1533 
1534 	if (inm->in6m_sctimer == 0) {
1535 		state_change_retransmit_timer_expired = 0;
1536 	} else if (--inm->in6m_sctimer == 0) {
1537 		state_change_retransmit_timer_expired = 1;
1538 	} else {
1539 		V_state_change_timers_running6 = 1;
1540 	}
1541 
1542 	/* We are in fasttimo, so be quick about it. */
1543 	if (!state_change_retransmit_timer_expired &&
1544 	    !query_response_timer_expired)
1545 		return;
1546 
1547 	switch (inm->in6m_state) {
1548 	case MLD_NOT_MEMBER:
1549 	case MLD_SILENT_MEMBER:
1550 	case MLD_SLEEPING_MEMBER:
1551 	case MLD_LAZY_MEMBER:
1552 	case MLD_AWAKENING_MEMBER:
1553 	case MLD_IDLE_MEMBER:
1554 		break;
1555 	case MLD_G_QUERY_PENDING_MEMBER:
1556 	case MLD_SG_QUERY_PENDING_MEMBER:
1557 		/*
1558 		 * Respond to a previously pending Group-Specific
1559 		 * or Group-and-Source-Specific query by enqueueing
1560 		 * the appropriate Current-State report for
1561 		 * immediate transmission.
1562 		 */
1563 		if (query_response_timer_expired) {
1564 			int retval;
1565 
1566 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1567 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1568 			    0);
1569 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1570 			    __func__, retval);
1571 			inm->in6m_state = MLD_REPORTING_MEMBER;
1572 			in6m_clear_recorded(inm);
1573 		}
1574 		/* FALLTHROUGH */
1575 	case MLD_REPORTING_MEMBER:
1576 	case MLD_LEAVING_MEMBER:
1577 		if (state_change_retransmit_timer_expired) {
1578 			/*
1579 			 * State-change retransmission timer fired.
1580 			 * If there are any further pending retransmissions,
1581 			 * set the global pending state-change flag, and
1582 			 * reset the timer.
1583 			 */
1584 			if (--inm->in6m_scrv > 0) {
1585 				inm->in6m_sctimer = uri_fasthz;
1586 				V_state_change_timers_running6 = 1;
1587 			}
1588 			/*
1589 			 * Retransmit the previously computed state-change
1590 			 * report. If there are no further pending
1591 			 * retransmissions, the mbuf queue will be consumed.
1592 			 * Update T0 state to T1 as we have now sent
1593 			 * a state-change.
1594 			 */
1595 			(void)mld_v2_merge_state_changes(inm, scq);
1596 
1597 			in6m_commit(inm);
1598 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1599 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1600 			    inm->in6m_ifp->if_xname);
1601 
1602 			/*
1603 			 * If we are leaving the group for good, make sure
1604 			 * we release MLD's reference to it.
1605 			 * This release must be deferred using a SLIST,
1606 			 * as we are called from a loop which traverses
1607 			 * the in_ifmultiaddr TAILQ.
1608 			 */
1609 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1610 			    inm->in6m_scrv == 0) {
1611 				inm->in6m_state = MLD_NOT_MEMBER;
1612 				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1613 				    inm, in6m_nrele);
1614 			}
1615 		}
1616 		break;
1617 	}
1618 }
1619 
1620 /*
1621  * Switch to a different version on the given interface,
1622  * as per Section 9.12.
1623  */
1624 static void
1625 mld_set_version(struct mld_ifinfo *mli, const int version)
1626 {
1627 	int old_version_timer;
1628 
1629 	MLD_LOCK_ASSERT();
1630 
1631 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1632 	    version, mli->mli_ifp, mli->mli_ifp->if_xname);
1633 
1634 	if (version == MLD_VERSION_1) {
1635 		/*
1636 		 * Compute the "Older Version Querier Present" timer as per
1637 		 * Section 9.12.
1638 		 */
1639 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1640 		old_version_timer *= PR_SLOWHZ;
1641 		mli->mli_v1_timer = old_version_timer;
1642 	}
1643 
1644 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1645 		mli->mli_version = MLD_VERSION_1;
1646 		mld_v2_cancel_link_timers(mli);
1647 	}
1648 }
1649 
1650 /*
1651  * Cancel pending MLDv2 timers for the given link and all groups
1652  * joined on it; state-change, general-query, and group-query timers.
1653  */
1654 static void
1655 mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
1656 {
1657 	struct ifmultiaddr	*ifma;
1658 	struct ifnet		*ifp;
1659 	struct in6_multi		*inm;
1660 
1661 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1662 	    mli->mli_ifp, mli->mli_ifp->if_xname);
1663 
1664 	IN6_MULTI_LOCK_ASSERT();
1665 	MLD_LOCK_ASSERT();
1666 
1667 	/*
1668 	 * Fast-track this potentially expensive operation
1669 	 * by checking all the global 'timer pending' flags.
1670 	 */
1671 	if (!V_interface_timers_running6 &&
1672 	    !V_state_change_timers_running6 &&
1673 	    !V_current_state_timers_running6)
1674 		return;
1675 
1676 	mli->mli_v2_timer = 0;
1677 
1678 	ifp = mli->mli_ifp;
1679 
1680 	IF_ADDR_LOCK(ifp);
1681 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1682 		if (ifma->ifma_addr->sa_family != AF_INET6)
1683 			continue;
1684 		inm = (struct in6_multi *)ifma->ifma_protospec;
1685 		switch (inm->in6m_state) {
1686 		case MLD_NOT_MEMBER:
1687 		case MLD_SILENT_MEMBER:
1688 		case MLD_IDLE_MEMBER:
1689 		case MLD_LAZY_MEMBER:
1690 		case MLD_SLEEPING_MEMBER:
1691 		case MLD_AWAKENING_MEMBER:
1692 			break;
1693 		case MLD_LEAVING_MEMBER:
1694 			/*
1695 			 * If we are leaving the group and switching
1696 			 * version, we need to release the final
1697 			 * reference held for issuing the INCLUDE {}.
1698 			 *
1699 			 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
1700 			 * around in6m_release_locked(), as it is not
1701 			 * a recursive mutex.
1702 			 */
1703 			IF_ADDR_UNLOCK(ifp);
1704 			in6m_release_locked(inm);
1705 			IF_ADDR_LOCK(ifp);
1706 			/* FALLTHROUGH */
1707 		case MLD_G_QUERY_PENDING_MEMBER:
1708 		case MLD_SG_QUERY_PENDING_MEMBER:
1709 			in6m_clear_recorded(inm);
1710 			/* FALLTHROUGH */
1711 		case MLD_REPORTING_MEMBER:
1712 			inm->in6m_sctimer = 0;
1713 			inm->in6m_timer = 0;
1714 			inm->in6m_state = MLD_REPORTING_MEMBER;
1715 			/*
1716 			 * Free any pending MLDv2 state-change records.
1717 			 */
1718 			_IF_DRAIN(&inm->in6m_scq);
1719 			break;
1720 		}
1721 	}
1722 	IF_ADDR_UNLOCK(ifp);
1723 }
1724 
1725 /*
1726  * Global slowtimo handler.
1727  * VIMAGE: Timeout handlers are expected to service all vimages.
1728  */
1729 void
1730 mld_slowtimo(void)
1731 {
1732 	VNET_ITERATOR_DECL(vnet_iter);
1733 
1734 	VNET_LIST_RLOCK_NOSLEEP();
1735 	VNET_FOREACH(vnet_iter) {
1736 		CURVNET_SET(vnet_iter);
1737 		mld_slowtimo_vnet();
1738 		CURVNET_RESTORE();
1739 	}
1740 	VNET_LIST_RUNLOCK_NOSLEEP();
1741 }
1742 
1743 /*
1744  * Per-vnet slowtimo handler.
1745  */
1746 static void
1747 mld_slowtimo_vnet(void)
1748 {
1749 	struct mld_ifinfo *mli;
1750 
1751 	MLD_LOCK();
1752 
1753 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1754 		mld_v1_process_querier_timers(mli);
1755 	}
1756 
1757 	MLD_UNLOCK();
1758 }
1759 
1760 /*
1761  * Update the Older Version Querier Present timers for a link.
1762  * See Section 9.12 of RFC 3810.
1763  */
1764 static void
1765 mld_v1_process_querier_timers(struct mld_ifinfo *mli)
1766 {
1767 
1768 	MLD_LOCK_ASSERT();
1769 
1770 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1771 		/*
1772 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1773 		 */
1774 		CTR5(KTR_MLD,
1775 		    "%s: transition from v%d -> v%d on %p(%s)",
1776 		    __func__, mli->mli_version, MLD_VERSION_2,
1777 		    mli->mli_ifp, mli->mli_ifp->if_xname);
1778 		mli->mli_version = MLD_VERSION_2;
1779 	}
1780 }
1781 
1782 /*
1783  * Transmit an MLDv1 report immediately.
1784  */
1785 static int
1786 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1787 {
1788 	struct ifnet		*ifp;
1789 	struct in6_ifaddr	*ia;
1790 	struct ip6_hdr		*ip6;
1791 	struct mbuf		*mh, *md;
1792 	struct mld_hdr		*mld;
1793 
1794 	IN6_MULTI_LOCK_ASSERT();
1795 	MLD_LOCK_ASSERT();
1796 
1797 	ifp = in6m->in6m_ifp;
1798 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1799 	/* ia may be NULL if link-local address is tentative. */
1800 
1801 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
1802 	if (mh == NULL) {
1803 		if (ia != NULL)
1804 			ifa_free(&ia->ia_ifa);
1805 		return (ENOMEM);
1806 	}
1807 	MGET(md, M_DONTWAIT, MT_DATA);
1808 	if (md == NULL) {
1809 		m_free(mh);
1810 		if (ia != NULL)
1811 			ifa_free(&ia->ia_ifa);
1812 		return (ENOMEM);
1813 	}
1814 	mh->m_next = md;
1815 
1816 	/*
1817 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1818 	 * that ether_output() does not need to allocate another mbuf
1819 	 * for the header in the most common case.
1820 	 */
1821 	MH_ALIGN(mh, sizeof(struct ip6_hdr));
1822 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1823 	mh->m_len = sizeof(struct ip6_hdr);
1824 
1825 	ip6 = mtod(mh, struct ip6_hdr *);
1826 	ip6->ip6_flow = 0;
1827 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1828 	ip6->ip6_vfc |= IPV6_VERSION;
1829 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1830 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1831 	ip6->ip6_dst = in6m->in6m_addr;
1832 
1833 	md->m_len = sizeof(struct mld_hdr);
1834 	mld = mtod(md, struct mld_hdr *);
1835 	mld->mld_type = type;
1836 	mld->mld_code = 0;
1837 	mld->mld_cksum = 0;
1838 	mld->mld_maxdelay = 0;
1839 	mld->mld_reserved = 0;
1840 	mld->mld_addr = in6m->in6m_addr;
1841 	in6_clearscope(&mld->mld_addr);
1842 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1843 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1844 
1845 	mld_save_context(mh, ifp);
1846 	mh->m_flags |= M_MLDV1;
1847 
1848 	mld_dispatch_packet(mh);
1849 
1850 	if (ia != NULL)
1851 		ifa_free(&ia->ia_ifa);
1852 	return (0);
1853 }
1854 
1855 /*
1856  * Process a state change from the upper layer for the given IPv6 group.
1857  *
1858  * Each socket holds a reference on the in_multi in its own ip_moptions.
1859  * The socket layer will have made the necessary updates to.the group
1860  * state, it is now up to MLD to issue a state change report if there
1861  * has been any change between T0 (when the last state-change was issued)
1862  * and T1 (now).
1863  *
1864  * We use the MLDv2 state machine at group level. The MLd module
1865  * however makes the decision as to which MLD protocol version to speak.
1866  * A state change *from* INCLUDE {} always means an initial join.
1867  * A state change *to* INCLUDE {} always means a final leave.
1868  *
1869  * If delay is non-zero, and the state change is an initial multicast
1870  * join, the state change report will be delayed by 'delay' ticks
1871  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1872  * the initial MLDv2 state change report will be delayed by whichever
1873  * is sooner, a pending state-change timer or delay itself.
1874  *
1875  * VIMAGE: curvnet should have been set by caller, as this routine
1876  * is called from the socket option handlers.
1877  */
1878 int
1879 mld_change_state(struct in6_multi *inm, const int delay)
1880 {
1881 	struct mld_ifinfo *mli;
1882 	struct ifnet *ifp;
1883 	int error;
1884 
1885 	IN6_MULTI_LOCK_ASSERT();
1886 
1887 	error = 0;
1888 
1889 	/*
1890 	 * Try to detect if the upper layer just asked us to change state
1891 	 * for an interface which has now gone away.
1892 	 */
1893 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1894 	ifp = inm->in6m_ifma->ifma_ifp;
1895 	if (ifp != NULL) {
1896 		/*
1897 		 * Sanity check that netinet6's notion of ifp is the
1898 		 * same as net's.
1899 		 */
1900 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1901 	}
1902 
1903 	MLD_LOCK();
1904 
1905 	mli = MLD_IFINFO(ifp);
1906 	KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp));
1907 
1908 	/*
1909 	 * If we detect a state transition to or from MCAST_UNDEFINED
1910 	 * for this group, then we are starting or finishing an MLD
1911 	 * life cycle for this group.
1912 	 */
1913 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1914 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1915 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1916 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1917 			CTR1(KTR_MLD, "%s: initial join", __func__);
1918 			error = mld_initial_join(inm, mli, delay);
1919 			goto out_locked;
1920 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1921 			CTR1(KTR_MLD, "%s: final leave", __func__);
1922 			mld_final_leave(inm, mli);
1923 			goto out_locked;
1924 		}
1925 	} else {
1926 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1927 	}
1928 
1929 	error = mld_handle_state_change(inm, mli);
1930 
1931 out_locked:
1932 	MLD_UNLOCK();
1933 	return (error);
1934 }
1935 
1936 /*
1937  * Perform the initial join for an MLD group.
1938  *
1939  * When joining a group:
1940  *  If the group should have its MLD traffic suppressed, do nothing.
1941  *  MLDv1 starts sending MLDv1 host membership reports.
1942  *  MLDv2 will schedule an MLDv2 state-change report containing the
1943  *  initial state of the membership.
1944  *
1945  * If the delay argument is non-zero, then we must delay sending the
1946  * initial state change for delay ticks (in units of PR_FASTHZ).
1947  */
1948 static int
1949 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
1950     const int delay)
1951 {
1952 	struct ifnet		*ifp;
1953 	struct ifqueue		*ifq;
1954 	int			 error, retval, syncstates;
1955 	int			 odelay;
1956 #ifdef KTR
1957 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1958 #endif
1959 
1960 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1961 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1962 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
1963 
1964 	error = 0;
1965 	syncstates = 1;
1966 
1967 	ifp = inm->in6m_ifp;
1968 
1969 	IN6_MULTI_LOCK_ASSERT();
1970 	MLD_LOCK_ASSERT();
1971 
1972 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1973 
1974 	/*
1975 	 * Groups joined on loopback or marked as 'not reported',
1976 	 * enter the MLD_SILENT_MEMBER state and
1977 	 * are never reported in any protocol exchanges.
1978 	 * All other groups enter the appropriate state machine
1979 	 * for the version in use on this link.
1980 	 * A link marked as MLIF_SILENT causes MLD to be completely
1981 	 * disabled for the link.
1982 	 */
1983 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1984 	    (mli->mli_flags & MLIF_SILENT) ||
1985 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1986 		CTR1(KTR_MLD,
1987 "%s: not kicking state machine for silent group", __func__);
1988 		inm->in6m_state = MLD_SILENT_MEMBER;
1989 		inm->in6m_timer = 0;
1990 	} else {
1991 		/*
1992 		 * Deal with overlapping in_multi lifecycle.
1993 		 * If this group was LEAVING, then make sure
1994 		 * we drop the reference we picked up to keep the
1995 		 * group around for the final INCLUDE {} enqueue.
1996 		 */
1997 		if (mli->mli_version == MLD_VERSION_2 &&
1998 		    inm->in6m_state == MLD_LEAVING_MEMBER)
1999 			in6m_release_locked(inm);
2000 
2001 		inm->in6m_state = MLD_REPORTING_MEMBER;
2002 
2003 		switch (mli->mli_version) {
2004 		case MLD_VERSION_1:
2005 			/*
2006 			 * If a delay was provided, only use it if
2007 			 * it is greater than the delay normally
2008 			 * used for an MLDv1 state change report,
2009 			 * and delay sending the initial MLDv1 report
2010 			 * by not transitioning to the IDLE state.
2011 			 */
2012 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2013 			if (delay) {
2014 				inm->in6m_timer = max(delay, odelay);
2015 				V_current_state_timers_running6 = 1;
2016 			} else {
2017 				inm->in6m_state = MLD_IDLE_MEMBER;
2018 				error = mld_v1_transmit_report(inm,
2019 				     MLD_LISTENER_REPORT);
2020 				if (error == 0) {
2021 					inm->in6m_timer = odelay;
2022 					V_current_state_timers_running6 = 1;
2023 				}
2024 			}
2025 			break;
2026 
2027 		case MLD_VERSION_2:
2028 			/*
2029 			 * Defer update of T0 to T1, until the first copy
2030 			 * of the state change has been transmitted.
2031 			 */
2032 			syncstates = 0;
2033 
2034 			/*
2035 			 * Immediately enqueue a State-Change Report for
2036 			 * this interface, freeing any previous reports.
2037 			 * Don't kick the timers if there is nothing to do,
2038 			 * or if an error occurred.
2039 			 */
2040 			ifq = &inm->in6m_scq;
2041 			_IF_DRAIN(ifq);
2042 			retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2043 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2044 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2045 			    __func__, retval);
2046 			if (retval <= 0) {
2047 				error = retval * -1;
2048 				break;
2049 			}
2050 
2051 			/*
2052 			 * Schedule transmission of pending state-change
2053 			 * report up to RV times for this link. The timer
2054 			 * will fire at the next mld_fasttimo (~200ms),
2055 			 * giving us an opportunity to merge the reports.
2056 			 *
2057 			 * If a delay was provided to this function, only
2058 			 * use this delay if sooner than the existing one.
2059 			 */
2060 			KASSERT(mli->mli_rv > 1,
2061 			   ("%s: invalid robustness %d", __func__,
2062 			    mli->mli_rv));
2063 			inm->in6m_scrv = mli->mli_rv;
2064 			if (delay) {
2065 				if (inm->in6m_sctimer > 1) {
2066 					inm->in6m_sctimer =
2067 					    min(inm->in6m_sctimer, delay);
2068 				} else
2069 					inm->in6m_sctimer = delay;
2070 			} else
2071 				inm->in6m_sctimer = 1;
2072 			V_state_change_timers_running6 = 1;
2073 
2074 			error = 0;
2075 			break;
2076 		}
2077 	}
2078 
2079 	/*
2080 	 * Only update the T0 state if state change is atomic,
2081 	 * i.e. we don't need to wait for a timer to fire before we
2082 	 * can consider the state change to have been communicated.
2083 	 */
2084 	if (syncstates) {
2085 		in6m_commit(inm);
2086 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2087 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2088 		    inm->in6m_ifp->if_xname);
2089 	}
2090 
2091 	return (error);
2092 }
2093 
2094 /*
2095  * Issue an intermediate state change during the life-cycle.
2096  */
2097 static int
2098 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli)
2099 {
2100 	struct ifnet		*ifp;
2101 	int			 retval;
2102 #ifdef KTR
2103 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2104 #endif
2105 
2106 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2107 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2108 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
2109 
2110 	ifp = inm->in6m_ifp;
2111 
2112 	IN6_MULTI_LOCK_ASSERT();
2113 	MLD_LOCK_ASSERT();
2114 
2115 	KASSERT(mli && mli->mli_ifp == ifp,
2116 	    ("%s: inconsistent ifp", __func__));
2117 
2118 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2119 	    (mli->mli_flags & MLIF_SILENT) ||
2120 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2121 	    (mli->mli_version != MLD_VERSION_2)) {
2122 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2123 			CTR1(KTR_MLD,
2124 "%s: not kicking state machine for silent group", __func__);
2125 		}
2126 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2127 		in6m_commit(inm);
2128 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2129 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2130 		    inm->in6m_ifp->if_xname);
2131 		return (0);
2132 	}
2133 
2134 	_IF_DRAIN(&inm->in6m_scq);
2135 
2136 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2137 	    (mli->mli_flags & MLIF_USEALLOW));
2138 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2139 	if (retval <= 0)
2140 		return (-retval);
2141 
2142 	/*
2143 	 * If record(s) were enqueued, start the state-change
2144 	 * report timer for this group.
2145 	 */
2146 	inm->in6m_scrv = mli->mli_rv;
2147 	inm->in6m_sctimer = 1;
2148 	V_state_change_timers_running6 = 1;
2149 
2150 	return (0);
2151 }
2152 
2153 /*
2154  * Perform the final leave for a multicast address.
2155  *
2156  * When leaving a group:
2157  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2158  *  MLDv2 enqueues a state-change report containing a transition
2159  *  to INCLUDE {} for immediate transmission.
2160  */
2161 static void
2162 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli)
2163 {
2164 	int syncstates;
2165 #ifdef KTR
2166 	char ip6tbuf[INET6_ADDRSTRLEN];
2167 #endif
2168 
2169 	syncstates = 1;
2170 
2171 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2172 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2173 	    inm->in6m_ifp, inm->in6m_ifp->if_xname);
2174 
2175 	IN6_MULTI_LOCK_ASSERT();
2176 	MLD_LOCK_ASSERT();
2177 
2178 	switch (inm->in6m_state) {
2179 	case MLD_NOT_MEMBER:
2180 	case MLD_SILENT_MEMBER:
2181 	case MLD_LEAVING_MEMBER:
2182 		/* Already leaving or left; do nothing. */
2183 		CTR1(KTR_MLD,
2184 "%s: not kicking state machine for silent group", __func__);
2185 		break;
2186 	case MLD_REPORTING_MEMBER:
2187 	case MLD_IDLE_MEMBER:
2188 	case MLD_G_QUERY_PENDING_MEMBER:
2189 	case MLD_SG_QUERY_PENDING_MEMBER:
2190 		if (mli->mli_version == MLD_VERSION_1) {
2191 #ifdef INVARIANTS
2192 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2193 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2194 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2195 			     __func__);
2196 #endif
2197 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2198 			inm->in6m_state = MLD_NOT_MEMBER;
2199 		} else if (mli->mli_version == MLD_VERSION_2) {
2200 			/*
2201 			 * Stop group timer and all pending reports.
2202 			 * Immediately enqueue a state-change report
2203 			 * TO_IN {} to be sent on the next fast timeout,
2204 			 * giving us an opportunity to merge reports.
2205 			 */
2206 			_IF_DRAIN(&inm->in6m_scq);
2207 			inm->in6m_timer = 0;
2208 			inm->in6m_scrv = mli->mli_rv;
2209 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2210 			    "pending retransmissions.", __func__,
2211 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2212 			    inm->in6m_ifp->if_xname, inm->in6m_scrv);
2213 			if (inm->in6m_scrv == 0) {
2214 				inm->in6m_state = MLD_NOT_MEMBER;
2215 				inm->in6m_sctimer = 0;
2216 			} else {
2217 				int retval;
2218 
2219 				in6m_acquire_locked(inm);
2220 
2221 				retval = mld_v2_enqueue_group_record(
2222 				    &inm->in6m_scq, inm, 1, 0, 0,
2223 				    (mli->mli_flags & MLIF_USEALLOW));
2224 				KASSERT(retval != 0,
2225 				    ("%s: enqueue record = %d", __func__,
2226 				     retval));
2227 
2228 				inm->in6m_state = MLD_LEAVING_MEMBER;
2229 				inm->in6m_sctimer = 1;
2230 				V_state_change_timers_running6 = 1;
2231 				syncstates = 0;
2232 			}
2233 			break;
2234 		}
2235 		break;
2236 	case MLD_LAZY_MEMBER:
2237 	case MLD_SLEEPING_MEMBER:
2238 	case MLD_AWAKENING_MEMBER:
2239 		/* Our reports are suppressed; do nothing. */
2240 		break;
2241 	}
2242 
2243 	if (syncstates) {
2244 		in6m_commit(inm);
2245 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2246 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2247 		    inm->in6m_ifp->if_xname);
2248 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2249 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2250 		    __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname);
2251 	}
2252 }
2253 
2254 /*
2255  * Enqueue an MLDv2 group record to the given output queue.
2256  *
2257  * If is_state_change is zero, a current-state record is appended.
2258  * If is_state_change is non-zero, a state-change report is appended.
2259  *
2260  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2261  * If is_group_query is zero, and if there is a packet with free space
2262  * at the tail of the queue, it will be appended to providing there
2263  * is enough free space.
2264  * Otherwise a new mbuf packet chain is allocated.
2265  *
2266  * If is_source_query is non-zero, each source is checked to see if
2267  * it was recorded for a Group-Source query, and will be omitted if
2268  * it is not both in-mode and recorded.
2269  *
2270  * If use_block_allow is non-zero, state change reports for initial join
2271  * and final leave, on an inclusive mode group with a source list, will be
2272  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2273  *
2274  * The function will attempt to allocate leading space in the packet
2275  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2276  *
2277  * If successful the size of all data appended to the queue is returned,
2278  * otherwise an error code less than zero is returned, or zero if
2279  * no record(s) were appended.
2280  */
2281 static int
2282 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2283     const int is_state_change, const int is_group_query,
2284     const int is_source_query, const int use_block_allow)
2285 {
2286 	struct mldv2_record	 mr;
2287 	struct mldv2_record	*pmr;
2288 	struct ifnet		*ifp;
2289 	struct ip6_msource	*ims, *nims;
2290 	struct mbuf		*m0, *m, *md;
2291 	int			 error, is_filter_list_change;
2292 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2293 	int			 record_has_sources;
2294 	int			 now;
2295 	int			 type;
2296 	uint8_t			 mode;
2297 #ifdef KTR
2298 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2299 #endif
2300 
2301 	IN6_MULTI_LOCK_ASSERT();
2302 
2303 	error = 0;
2304 	ifp = inm->in6m_ifp;
2305 	is_filter_list_change = 0;
2306 	m = NULL;
2307 	m0 = NULL;
2308 	m0srcs = 0;
2309 	msrcs = 0;
2310 	nbytes = 0;
2311 	nims = NULL;
2312 	record_has_sources = 1;
2313 	pmr = NULL;
2314 	type = MLD_DO_NOTHING;
2315 	mode = inm->in6m_st[1].iss_fmode;
2316 
2317 	/*
2318 	 * If we did not transition out of ASM mode during t0->t1,
2319 	 * and there are no source nodes to process, we can skip
2320 	 * the generation of source records.
2321 	 */
2322 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2323 	    inm->in6m_nsrc == 0)
2324 		record_has_sources = 0;
2325 
2326 	if (is_state_change) {
2327 		/*
2328 		 * Queue a state change record.
2329 		 * If the mode did not change, and there are non-ASM
2330 		 * listeners or source filters present,
2331 		 * we potentially need to issue two records for the group.
2332 		 * If there are ASM listeners, and there was no filter
2333 		 * mode transition of any kind, do nothing.
2334 		 *
2335 		 * If we are transitioning to MCAST_UNDEFINED, we need
2336 		 * not send any sources. A transition to/from this state is
2337 		 * considered inclusive with some special treatment.
2338 		 *
2339 		 * If we are rewriting initial joins/leaves to use
2340 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2341 		 * we need to send sources in all cases.
2342 		 */
2343 		if (mode != inm->in6m_st[0].iss_fmode) {
2344 			if (mode == MCAST_EXCLUDE) {
2345 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2346 				    __func__);
2347 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2348 			} else {
2349 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2350 				    __func__);
2351 				if (use_block_allow) {
2352 					/*
2353 					 * XXX
2354 					 * Here we're interested in state
2355 					 * edges either direction between
2356 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2357 					 * Perhaps we should just check
2358 					 * the group state, rather than
2359 					 * the filter mode.
2360 					 */
2361 					if (mode == MCAST_UNDEFINED) {
2362 						type = MLD_BLOCK_OLD_SOURCES;
2363 					} else {
2364 						type = MLD_ALLOW_NEW_SOURCES;
2365 					}
2366 				} else {
2367 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2368 					if (mode == MCAST_UNDEFINED)
2369 						record_has_sources = 0;
2370 				}
2371 			}
2372 		} else {
2373 			if (record_has_sources) {
2374 				is_filter_list_change = 1;
2375 			} else {
2376 				type = MLD_DO_NOTHING;
2377 			}
2378 		}
2379 	} else {
2380 		/*
2381 		 * Queue a current state record.
2382 		 */
2383 		if (mode == MCAST_EXCLUDE) {
2384 			type = MLD_MODE_IS_EXCLUDE;
2385 		} else if (mode == MCAST_INCLUDE) {
2386 			type = MLD_MODE_IS_INCLUDE;
2387 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2388 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2389 			     __func__, inm, inm->in6m_st[1].iss_asm));
2390 		}
2391 	}
2392 
2393 	/*
2394 	 * Generate the filter list changes using a separate function.
2395 	 */
2396 	if (is_filter_list_change)
2397 		return (mld_v2_enqueue_filter_change(ifq, inm));
2398 
2399 	if (type == MLD_DO_NOTHING) {
2400 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2401 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2402 		    inm->in6m_ifp->if_xname);
2403 		return (0);
2404 	}
2405 
2406 	/*
2407 	 * If any sources are present, we must be able to fit at least
2408 	 * one in the trailing space of the tail packet's mbuf,
2409 	 * ideally more.
2410 	 */
2411 	minrec0len = sizeof(struct mldv2_record);
2412 	if (record_has_sources)
2413 		minrec0len += sizeof(struct in6_addr);
2414 
2415 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2416 	    mld_rec_type_to_str(type),
2417 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2418 	    inm->in6m_ifp->if_xname);
2419 
2420 	/*
2421 	 * Check if we have a packet in the tail of the queue for this
2422 	 * group into which the first group record for this group will fit.
2423 	 * Otherwise allocate a new packet.
2424 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2425 	 * Note: Group records for G/GSR query responses MUST be sent
2426 	 * in their own packet.
2427 	 */
2428 	m0 = ifq->ifq_tail;
2429 	if (!is_group_query &&
2430 	    m0 != NULL &&
2431 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2432 	    (m0->m_pkthdr.len + minrec0len) <
2433 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2434 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2435 			    sizeof(struct mldv2_record)) /
2436 			    sizeof(struct in6_addr);
2437 		m = m0;
2438 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2439 	} else {
2440 		if (_IF_QFULL(ifq)) {
2441 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2442 			return (-ENOMEM);
2443 		}
2444 		m = NULL;
2445 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2446 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2447 		if (!is_state_change && !is_group_query)
2448 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2449 		if (m == NULL)
2450 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2451 		if (m == NULL)
2452 			return (-ENOMEM);
2453 
2454 		mld_save_context(m, ifp);
2455 
2456 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2457 	}
2458 
2459 	/*
2460 	 * Append group record.
2461 	 * If we have sources, we don't know how many yet.
2462 	 */
2463 	mr.mr_type = type;
2464 	mr.mr_datalen = 0;
2465 	mr.mr_numsrc = 0;
2466 	mr.mr_addr = inm->in6m_addr;
2467 	in6_clearscope(&mr.mr_addr);
2468 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2469 		if (m != m0)
2470 			m_freem(m);
2471 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2472 		return (-ENOMEM);
2473 	}
2474 	nbytes += sizeof(struct mldv2_record);
2475 
2476 	/*
2477 	 * Append as many sources as will fit in the first packet.
2478 	 * If we are appending to a new packet, the chain allocation
2479 	 * may potentially use clusters; use m_getptr() in this case.
2480 	 * If we are appending to an existing packet, we need to obtain
2481 	 * a pointer to the group record after m_append(), in case a new
2482 	 * mbuf was allocated.
2483 	 *
2484 	 * Only append sources which are in-mode at t1. If we are
2485 	 * transitioning to MCAST_UNDEFINED state on the group, and
2486 	 * use_block_allow is zero, do not include source entries.
2487 	 * Otherwise, we need to include this source in the report.
2488 	 *
2489 	 * Only report recorded sources in our filter set when responding
2490 	 * to a group-source query.
2491 	 */
2492 	if (record_has_sources) {
2493 		if (m == m0) {
2494 			md = m_last(m);
2495 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2496 			    md->m_len - nbytes);
2497 		} else {
2498 			md = m_getptr(m, 0, &off);
2499 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2500 			    off);
2501 		}
2502 		msrcs = 0;
2503 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2504 		    nims) {
2505 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2506 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2507 			now = im6s_get_mode(inm, ims, 1);
2508 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2509 			if ((now != mode) ||
2510 			    (now == mode &&
2511 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2512 				CTR1(KTR_MLD, "%s: skip node", __func__);
2513 				continue;
2514 			}
2515 			if (is_source_query && ims->im6s_stp == 0) {
2516 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2517 				    __func__);
2518 				continue;
2519 			}
2520 			CTR1(KTR_MLD, "%s: append node", __func__);
2521 			if (!m_append(m, sizeof(struct in6_addr),
2522 			    (void *)&ims->im6s_addr)) {
2523 				if (m != m0)
2524 					m_freem(m);
2525 				CTR1(KTR_MLD, "%s: m_append() failed.",
2526 				    __func__);
2527 				return (-ENOMEM);
2528 			}
2529 			nbytes += sizeof(struct in6_addr);
2530 			++msrcs;
2531 			if (msrcs == m0srcs)
2532 				break;
2533 		}
2534 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2535 		    msrcs);
2536 		pmr->mr_numsrc = htons(msrcs);
2537 		nbytes += (msrcs * sizeof(struct in6_addr));
2538 	}
2539 
2540 	if (is_source_query && msrcs == 0) {
2541 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2542 		if (m != m0)
2543 			m_freem(m);
2544 		return (0);
2545 	}
2546 
2547 	/*
2548 	 * We are good to go with first packet.
2549 	 */
2550 	if (m != m0) {
2551 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2552 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2553 		_IF_ENQUEUE(ifq, m);
2554 	} else
2555 		m->m_pkthdr.PH_vt.vt_nrecs++;
2556 
2557 	/*
2558 	 * No further work needed if no source list in packet(s).
2559 	 */
2560 	if (!record_has_sources)
2561 		return (nbytes);
2562 
2563 	/*
2564 	 * Whilst sources remain to be announced, we need to allocate
2565 	 * a new packet and fill out as many sources as will fit.
2566 	 * Always try for a cluster first.
2567 	 */
2568 	while (nims != NULL) {
2569 		if (_IF_QFULL(ifq)) {
2570 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2571 			return (-ENOMEM);
2572 		}
2573 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2574 		if (m == NULL)
2575 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2576 		if (m == NULL)
2577 			return (-ENOMEM);
2578 		mld_save_context(m, ifp);
2579 		md = m_getptr(m, 0, &off);
2580 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2581 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2582 
2583 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2584 			if (m != m0)
2585 				m_freem(m);
2586 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2587 			return (-ENOMEM);
2588 		}
2589 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2590 		nbytes += sizeof(struct mldv2_record);
2591 
2592 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2593 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2594 
2595 		msrcs = 0;
2596 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2597 			CTR2(KTR_MLD, "%s: visit node %s",
2598 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2599 			now = im6s_get_mode(inm, ims, 1);
2600 			if ((now != mode) ||
2601 			    (now == mode &&
2602 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2603 				CTR1(KTR_MLD, "%s: skip node", __func__);
2604 				continue;
2605 			}
2606 			if (is_source_query && ims->im6s_stp == 0) {
2607 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2608 				    __func__);
2609 				continue;
2610 			}
2611 			CTR1(KTR_MLD, "%s: append node", __func__);
2612 			if (!m_append(m, sizeof(struct in6_addr),
2613 			    (void *)&ims->im6s_addr)) {
2614 				if (m != m0)
2615 					m_freem(m);
2616 				CTR1(KTR_MLD, "%s: m_append() failed.",
2617 				    __func__);
2618 				return (-ENOMEM);
2619 			}
2620 			++msrcs;
2621 			if (msrcs == m0srcs)
2622 				break;
2623 		}
2624 		pmr->mr_numsrc = htons(msrcs);
2625 		nbytes += (msrcs * sizeof(struct in6_addr));
2626 
2627 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2628 		_IF_ENQUEUE(ifq, m);
2629 	}
2630 
2631 	return (nbytes);
2632 }
2633 
2634 /*
2635  * Type used to mark record pass completion.
2636  * We exploit the fact we can cast to this easily from the
2637  * current filter modes on each ip_msource node.
2638  */
2639 typedef enum {
2640 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2641 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2642 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2643 	REC_FULL = REC_ALLOW | REC_BLOCK
2644 } rectype_t;
2645 
2646 /*
2647  * Enqueue an MLDv2 filter list change to the given output queue.
2648  *
2649  * Source list filter state is held in an RB-tree. When the filter list
2650  * for a group is changed without changing its mode, we need to compute
2651  * the deltas between T0 and T1 for each source in the filter set,
2652  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2653  *
2654  * As we may potentially queue two record types, and the entire R-B tree
2655  * needs to be walked at once, we break this out into its own function
2656  * so we can generate a tightly packed queue of packets.
2657  *
2658  * XXX This could be written to only use one tree walk, although that makes
2659  * serializing into the mbuf chains a bit harder. For now we do two walks
2660  * which makes things easier on us, and it may or may not be harder on
2661  * the L2 cache.
2662  *
2663  * If successful the size of all data appended to the queue is returned,
2664  * otherwise an error code less than zero is returned, or zero if
2665  * no record(s) were appended.
2666  */
2667 static int
2668 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
2669 {
2670 	static const int MINRECLEN =
2671 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2672 	struct ifnet		*ifp;
2673 	struct mldv2_record	 mr;
2674 	struct mldv2_record	*pmr;
2675 	struct ip6_msource	*ims, *nims;
2676 	struct mbuf		*m, *m0, *md;
2677 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2678 	int			 nallow, nblock;
2679 	uint8_t			 mode, now, then;
2680 	rectype_t		 crt, drt, nrt;
2681 #ifdef KTR
2682 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2683 #endif
2684 
2685 	IN6_MULTI_LOCK_ASSERT();
2686 
2687 	if (inm->in6m_nsrc == 0 ||
2688 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2689 		return (0);
2690 
2691 	ifp = inm->in6m_ifp;			/* interface */
2692 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2693 	crt = REC_NONE;	/* current group record type */
2694 	drt = REC_NONE;	/* mask of completed group record types */
2695 	nrt = REC_NONE;	/* record type for current node */
2696 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2697 	npbytes = 0;	/* # of bytes appended this packet */
2698 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2699 	rsrcs = 0;	/* # sources encoded in current record */
2700 	schanged = 0;	/* # nodes encoded in overall filter change */
2701 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2702 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2703 	nims = NULL;	/* next tree node pointer */
2704 
2705 	/*
2706 	 * For each possible filter record mode.
2707 	 * The first kind of source we encounter tells us which
2708 	 * is the first kind of record we start appending.
2709 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2710 	 * as the inverse of the group's filter mode.
2711 	 */
2712 	while (drt != REC_FULL) {
2713 		do {
2714 			m0 = ifq->ifq_tail;
2715 			if (m0 != NULL &&
2716 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2717 			     MLD_V2_REPORT_MAXRECS) &&
2718 			    (m0->m_pkthdr.len + MINRECLEN) <
2719 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2720 				m = m0;
2721 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2722 					    sizeof(struct mldv2_record)) /
2723 					    sizeof(struct in6_addr);
2724 				CTR1(KTR_MLD,
2725 				    "%s: use previous packet", __func__);
2726 			} else {
2727 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2728 				if (m == NULL)
2729 					m = m_gethdr(M_DONTWAIT, MT_DATA);
2730 				if (m == NULL) {
2731 					CTR1(KTR_MLD,
2732 					    "%s: m_get*() failed", __func__);
2733 					return (-ENOMEM);
2734 				}
2735 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2736 				mld_save_context(m, ifp);
2737 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2738 				    sizeof(struct mldv2_record)) /
2739 				    sizeof(struct in6_addr);
2740 				npbytes = 0;
2741 				CTR1(KTR_MLD,
2742 				    "%s: allocated new packet", __func__);
2743 			}
2744 			/*
2745 			 * Append the MLD group record header to the
2746 			 * current packet's data area.
2747 			 * Recalculate pointer to free space for next
2748 			 * group record, in case m_append() allocated
2749 			 * a new mbuf or cluster.
2750 			 */
2751 			memset(&mr, 0, sizeof(mr));
2752 			mr.mr_addr = inm->in6m_addr;
2753 			in6_clearscope(&mr.mr_addr);
2754 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2755 				if (m != m0)
2756 					m_freem(m);
2757 				CTR1(KTR_MLD,
2758 				    "%s: m_append() failed", __func__);
2759 				return (-ENOMEM);
2760 			}
2761 			npbytes += sizeof(struct mldv2_record);
2762 			if (m != m0) {
2763 				/* new packet; offset in chain */
2764 				md = m_getptr(m, npbytes -
2765 				    sizeof(struct mldv2_record), &off);
2766 				pmr = (struct mldv2_record *)(mtod(md,
2767 				    uint8_t *) + off);
2768 			} else {
2769 				/* current packet; offset from last append */
2770 				md = m_last(m);
2771 				pmr = (struct mldv2_record *)(mtod(md,
2772 				    uint8_t *) + md->m_len -
2773 				    sizeof(struct mldv2_record));
2774 			}
2775 			/*
2776 			 * Begin walking the tree for this record type
2777 			 * pass, or continue from where we left off
2778 			 * previously if we had to allocate a new packet.
2779 			 * Only report deltas in-mode at t1.
2780 			 * We need not report included sources as allowed
2781 			 * if we are in inclusive mode on the group,
2782 			 * however the converse is not true.
2783 			 */
2784 			rsrcs = 0;
2785 			if (nims == NULL) {
2786 				nims = RB_MIN(ip6_msource_tree,
2787 				    &inm->in6m_srcs);
2788 			}
2789 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2790 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2791 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2792 				now = im6s_get_mode(inm, ims, 1);
2793 				then = im6s_get_mode(inm, ims, 0);
2794 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2795 				    __func__, then, now);
2796 				if (now == then) {
2797 					CTR1(KTR_MLD,
2798 					    "%s: skip unchanged", __func__);
2799 					continue;
2800 				}
2801 				if (mode == MCAST_EXCLUDE &&
2802 				    now == MCAST_INCLUDE) {
2803 					CTR1(KTR_MLD,
2804 					    "%s: skip IN src on EX group",
2805 					    __func__);
2806 					continue;
2807 				}
2808 				nrt = (rectype_t)now;
2809 				if (nrt == REC_NONE)
2810 					nrt = (rectype_t)(~mode & REC_FULL);
2811 				if (schanged++ == 0) {
2812 					crt = nrt;
2813 				} else if (crt != nrt)
2814 					continue;
2815 				if (!m_append(m, sizeof(struct in6_addr),
2816 				    (void *)&ims->im6s_addr)) {
2817 					if (m != m0)
2818 						m_freem(m);
2819 					CTR1(KTR_MLD,
2820 					    "%s: m_append() failed", __func__);
2821 					return (-ENOMEM);
2822 				}
2823 				nallow += !!(crt == REC_ALLOW);
2824 				nblock += !!(crt == REC_BLOCK);
2825 				if (++rsrcs == m0srcs)
2826 					break;
2827 			}
2828 			/*
2829 			 * If we did not append any tree nodes on this
2830 			 * pass, back out of allocations.
2831 			 */
2832 			if (rsrcs == 0) {
2833 				npbytes -= sizeof(struct mldv2_record);
2834 				if (m != m0) {
2835 					CTR1(KTR_MLD,
2836 					    "%s: m_free(m)", __func__);
2837 					m_freem(m);
2838 				} else {
2839 					CTR1(KTR_MLD,
2840 					    "%s: m_adj(m, -mr)", __func__);
2841 					m_adj(m, -((int)sizeof(
2842 					    struct mldv2_record)));
2843 				}
2844 				continue;
2845 			}
2846 			npbytes += (rsrcs * sizeof(struct in6_addr));
2847 			if (crt == REC_ALLOW)
2848 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2849 			else if (crt == REC_BLOCK)
2850 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2851 			pmr->mr_numsrc = htons(rsrcs);
2852 			/*
2853 			 * Count the new group record, and enqueue this
2854 			 * packet if it wasn't already queued.
2855 			 */
2856 			m->m_pkthdr.PH_vt.vt_nrecs++;
2857 			if (m != m0)
2858 				_IF_ENQUEUE(ifq, m);
2859 			nbytes += npbytes;
2860 		} while (nims != NULL);
2861 		drt |= crt;
2862 		crt = (~crt & REC_FULL);
2863 	}
2864 
2865 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2866 	    nallow, nblock);
2867 
2868 	return (nbytes);
2869 }
2870 
2871 static int
2872 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
2873 {
2874 	struct ifqueue	*gq;
2875 	struct mbuf	*m;		/* pending state-change */
2876 	struct mbuf	*m0;		/* copy of pending state-change */
2877 	struct mbuf	*mt;		/* last state-change in packet */
2878 	int		 docopy, domerge;
2879 	u_int		 recslen;
2880 
2881 	docopy = 0;
2882 	domerge = 0;
2883 	recslen = 0;
2884 
2885 	IN6_MULTI_LOCK_ASSERT();
2886 	MLD_LOCK_ASSERT();
2887 
2888 	/*
2889 	 * If there are further pending retransmissions, make a writable
2890 	 * copy of each queued state-change message before merging.
2891 	 */
2892 	if (inm->in6m_scrv > 0)
2893 		docopy = 1;
2894 
2895 	gq = &inm->in6m_scq;
2896 #ifdef KTR
2897 	if (gq->ifq_head == NULL) {
2898 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2899 		    __func__, inm);
2900 	}
2901 #endif
2902 
2903 	m = gq->ifq_head;
2904 	while (m != NULL) {
2905 		/*
2906 		 * Only merge the report into the current packet if
2907 		 * there is sufficient space to do so; an MLDv2 report
2908 		 * packet may only contain 65,535 group records.
2909 		 * Always use a simple mbuf chain concatentation to do this,
2910 		 * as large state changes for single groups may have
2911 		 * allocated clusters.
2912 		 */
2913 		domerge = 0;
2914 		mt = ifscq->ifq_tail;
2915 		if (mt != NULL) {
2916 			recslen = m_length(m, NULL);
2917 
2918 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2919 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2920 			    MLD_V2_REPORT_MAXRECS) &&
2921 			    (mt->m_pkthdr.len + recslen <=
2922 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2923 				domerge = 1;
2924 		}
2925 
2926 		if (!domerge && _IF_QFULL(gq)) {
2927 			CTR2(KTR_MLD,
2928 			    "%s: outbound queue full, skipping whole packet %p",
2929 			    __func__, m);
2930 			mt = m->m_nextpkt;
2931 			if (!docopy)
2932 				m_freem(m);
2933 			m = mt;
2934 			continue;
2935 		}
2936 
2937 		if (!docopy) {
2938 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2939 			_IF_DEQUEUE(gq, m0);
2940 			m = m0->m_nextpkt;
2941 		} else {
2942 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2943 			m0 = m_dup(m, M_NOWAIT);
2944 			if (m0 == NULL)
2945 				return (ENOMEM);
2946 			m0->m_nextpkt = NULL;
2947 			m = m->m_nextpkt;
2948 		}
2949 
2950 		if (!domerge) {
2951 			CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)",
2952 			    __func__, m0, ifscq);
2953 			_IF_ENQUEUE(ifscq, m0);
2954 		} else {
2955 			struct mbuf *mtl;	/* last mbuf of packet mt */
2956 
2957 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2958 			    __func__, m0, mt);
2959 
2960 			mtl = m_last(mt);
2961 			m0->m_flags &= ~M_PKTHDR;
2962 			mt->m_pkthdr.len += recslen;
2963 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2964 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2965 
2966 			mtl->m_next = m0;
2967 		}
2968 	}
2969 
2970 	return (0);
2971 }
2972 
2973 /*
2974  * Respond to a pending MLDv2 General Query.
2975  */
2976 static void
2977 mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
2978 {
2979 	struct ifmultiaddr	*ifma, *tifma;
2980 	struct ifnet		*ifp;
2981 	struct in6_multi	*inm;
2982 	int			 retval;
2983 
2984 	IN6_MULTI_LOCK_ASSERT();
2985 	MLD_LOCK_ASSERT();
2986 
2987 	KASSERT(mli->mli_version == MLD_VERSION_2,
2988 	    ("%s: called when version %d", __func__, mli->mli_version));
2989 
2990 	ifp = mli->mli_ifp;
2991 
2992 	IF_ADDR_LOCK(ifp);
2993 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
2994 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2995 		    ifma->ifma_protospec == NULL)
2996 			continue;
2997 
2998 		inm = (struct in6_multi *)ifma->ifma_protospec;
2999 		KASSERT(ifp == inm->in6m_ifp,
3000 		    ("%s: inconsistent ifp", __func__));
3001 
3002 		switch (inm->in6m_state) {
3003 		case MLD_NOT_MEMBER:
3004 		case MLD_SILENT_MEMBER:
3005 			break;
3006 		case MLD_REPORTING_MEMBER:
3007 		case MLD_IDLE_MEMBER:
3008 		case MLD_LAZY_MEMBER:
3009 		case MLD_SLEEPING_MEMBER:
3010 		case MLD_AWAKENING_MEMBER:
3011 			inm->in6m_state = MLD_REPORTING_MEMBER;
3012 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3013 			    inm, 0, 0, 0, 0);
3014 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3015 			    __func__, retval);
3016 			break;
3017 		case MLD_G_QUERY_PENDING_MEMBER:
3018 		case MLD_SG_QUERY_PENDING_MEMBER:
3019 		case MLD_LEAVING_MEMBER:
3020 			break;
3021 		}
3022 	}
3023 	IF_ADDR_UNLOCK(ifp);
3024 
3025 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3026 
3027 	/*
3028 	 * Slew transmission of bursts over 500ms intervals.
3029 	 */
3030 	if (mli->mli_gq.ifq_head != NULL) {
3031 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3032 		    MLD_RESPONSE_BURST_INTERVAL);
3033 		V_interface_timers_running6 = 1;
3034 	}
3035 }
3036 
3037 /*
3038  * Transmit the next pending message in the output queue.
3039  *
3040  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3041  * MRT: Nothing needs to be done, as MLD traffic is always local to
3042  * a link and uses a link-scope multicast address.
3043  */
3044 static void
3045 mld_dispatch_packet(struct mbuf *m)
3046 {
3047 	struct ip6_moptions	 im6o;
3048 	struct ifnet		*ifp;
3049 	struct ifnet		*oifp;
3050 	struct mbuf		*m0;
3051 	struct mbuf		*md;
3052 	struct ip6_hdr		*ip6;
3053 	struct mld_hdr		*mld;
3054 	int			 error;
3055 	int			 off;
3056 	int			 type;
3057 	uint32_t		 ifindex;
3058 
3059 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3060 
3061 	/*
3062 	 * Set VNET image pointer from enqueued mbuf chain
3063 	 * before doing anything else. Whilst we use interface
3064 	 * indexes to guard against interface detach, they are
3065 	 * unique to each VIMAGE and must be retrieved.
3066 	 */
3067 	ifindex = mld_restore_context(m);
3068 
3069 	/*
3070 	 * Check if the ifnet still exists. This limits the scope of
3071 	 * any race in the absence of a global ifp lock for low cost
3072 	 * (an array lookup).
3073 	 */
3074 	ifp = ifnet_byindex(ifindex);
3075 	if (ifp == NULL) {
3076 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3077 		    __func__, m, ifindex);
3078 		m_freem(m);
3079 		IP6STAT_INC(ip6s_noroute);
3080 		goto out;
3081 	}
3082 
3083 	im6o.im6o_multicast_hlim  = 1;
3084 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3085 	im6o.im6o_multicast_ifp = ifp;
3086 
3087 	if (m->m_flags & M_MLDV1) {
3088 		m0 = m;
3089 	} else {
3090 		m0 = mld_v2_encap_report(ifp, m);
3091 		if (m0 == NULL) {
3092 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3093 			IP6STAT_INC(ip6s_odropped);
3094 			goto out;
3095 		}
3096 	}
3097 
3098 	mld_scrub_context(m0);
3099 	m->m_flags &= ~(M_PROTOFLAGS);
3100 	m0->m_pkthdr.rcvif = V_loif;
3101 
3102 	ip6 = mtod(m0, struct ip6_hdr *);
3103 #if 0
3104 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3105 #else
3106 	/*
3107 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3108 	 * occur if we called in6_setscope() at transmission.
3109 	 * See comments at top of file.
3110 	 */
3111 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3112 #endif
3113 
3114 	/*
3115 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3116 	 * so we can bump the stats.
3117 	 */
3118 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3119 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3120 	type = mld->mld_type;
3121 
3122 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3123 	    &oifp, NULL);
3124 	if (error) {
3125 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3126 		goto out;
3127 	}
3128 	ICMP6STAT_INC(icp6s_outhist[type]);
3129 	if (oifp != NULL) {
3130 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3131 		switch (type) {
3132 		case MLD_LISTENER_REPORT:
3133 		case MLDV2_LISTENER_REPORT:
3134 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3135 			break;
3136 		case MLD_LISTENER_DONE:
3137 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3138 			break;
3139 		}
3140 	}
3141 out:
3142 	return;
3143 }
3144 
3145 /*
3146  * Encapsulate an MLDv2 report.
3147  *
3148  * KAME IPv6 requires that hop-by-hop options be passed separately,
3149  * and that the IPv6 header be prepended in a separate mbuf.
3150  *
3151  * Returns a pointer to the new mbuf chain head, or NULL if the
3152  * allocation failed.
3153  */
3154 static struct mbuf *
3155 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3156 {
3157 	struct mbuf		*mh;
3158 	struct mldv2_report	*mld;
3159 	struct ip6_hdr		*ip6;
3160 	struct in6_ifaddr	*ia;
3161 	int			 mldreclen;
3162 
3163 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3164 	KASSERT((m->m_flags & M_PKTHDR),
3165 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3166 
3167 	/*
3168 	 * RFC3590: OK to send as :: or tentative during DAD.
3169 	 */
3170 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3171 	if (ia == NULL)
3172 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3173 
3174 	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3175 	if (mh == NULL) {
3176 		if (ia != NULL)
3177 			ifa_free(&ia->ia_ifa);
3178 		m_freem(m);
3179 		return (NULL);
3180 	}
3181 	MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3182 
3183 	mldreclen = m_length(m, NULL);
3184 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3185 
3186 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3187 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3188 	    sizeof(struct mldv2_report) + mldreclen;
3189 
3190 	ip6 = mtod(mh, struct ip6_hdr *);
3191 	ip6->ip6_flow = 0;
3192 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3193 	ip6->ip6_vfc |= IPV6_VERSION;
3194 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3195 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3196 	if (ia != NULL)
3197 		ifa_free(&ia->ia_ifa);
3198 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3199 	/* scope ID will be set in netisr */
3200 
3201 	mld = (struct mldv2_report *)(ip6 + 1);
3202 	mld->mld_type = MLDV2_LISTENER_REPORT;
3203 	mld->mld_code = 0;
3204 	mld->mld_cksum = 0;
3205 	mld->mld_v2_reserved = 0;
3206 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3207 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3208 
3209 	mh->m_next = m;
3210 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3211 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3212 	return (mh);
3213 }
3214 
3215 #ifdef KTR
3216 static char *
3217 mld_rec_type_to_str(const int type)
3218 {
3219 
3220 	switch (type) {
3221 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3222 			return "TO_EX";
3223 			break;
3224 		case MLD_CHANGE_TO_INCLUDE_MODE:
3225 			return "TO_IN";
3226 			break;
3227 		case MLD_MODE_IS_EXCLUDE:
3228 			return "MODE_EX";
3229 			break;
3230 		case MLD_MODE_IS_INCLUDE:
3231 			return "MODE_IN";
3232 			break;
3233 		case MLD_ALLOW_NEW_SOURCES:
3234 			return "ALLOW_NEW";
3235 			break;
3236 		case MLD_BLOCK_OLD_SOURCES:
3237 			return "BLOCK_OLD";
3238 			break;
3239 		default:
3240 			break;
3241 	}
3242 	return "unknown";
3243 }
3244 #endif
3245 
3246 static void
3247 mld_init(void *unused __unused)
3248 {
3249 
3250 	CTR1(KTR_MLD, "%s: initializing", __func__);
3251 	MLD_LOCK_INIT();
3252 
3253 	ip6_initpktopts(&mld_po);
3254 	mld_po.ip6po_hlim = 1;
3255 	mld_po.ip6po_hbh = &mld_ra.hbh;
3256 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3257 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3258 }
3259 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL);
3260 
3261 static void
3262 mld_uninit(void *unused __unused)
3263 {
3264 
3265 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3266 	MLD_LOCK_DESTROY();
3267 }
3268 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL);
3269 
3270 static void
3271 vnet_mld_init(const void *unused __unused)
3272 {
3273 
3274 	CTR1(KTR_MLD, "%s: initializing", __func__);
3275 
3276 	LIST_INIT(&V_mli_head);
3277 }
3278 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init,
3279     NULL);
3280 
3281 static void
3282 vnet_mld_uninit(const void *unused __unused)
3283 {
3284 
3285 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3286 
3287 	KASSERT(LIST_EMPTY(&V_mli_head),
3288 	    ("%s: mli list not empty; ifnets not detached?", __func__));
3289 }
3290 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit,
3291     NULL);
3292 
3293 static int
3294 mld_modevent(module_t mod, int type, void *unused __unused)
3295 {
3296 
3297     switch (type) {
3298     case MOD_LOAD:
3299     case MOD_UNLOAD:
3300 	break;
3301     default:
3302 	return (EOPNOTSUPP);
3303     }
3304     return (0);
3305 }
3306 
3307 static moduledata_t mld_mod = {
3308     "mld",
3309     mld_modevent,
3310     0
3311 };
3312 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3313