xref: /freebsd/sys/netinet/igmp.c (revision bb15ca603fa442c72dde3f3cb8b46db6970e3950)
1 /*-
2  * Copyright (c) 2007-2009 Bruce Simpson.
3  * Copyright (c) 1988 Stephen Deering.
4  * Copyright (c) 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Stephen Deering of Stanford University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
35  */
36 
37 /*
38  * Internet Group Management Protocol (IGMP) routines.
39  * [RFC1112, RFC2236, RFC3376]
40  *
41  * Written by Steve Deering, Stanford, May 1988.
42  * Modified by Rosen Sharma, Stanford, Aug 1994.
43  * Modified by Bill Fenner, Xerox PARC, Feb 1995.
44  * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995.
45  * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson.
46  *
47  * MULTICAST Revision: 3.5.1.4
48  */
49 
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/module.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/socket.h>
59 #include <sys/protosw.h>
60 #include <sys/kernel.h>
61 #include <sys/sysctl.h>
62 #include <sys/ktr.h>
63 #include <sys/condvar.h>
64 
65 #include <net/if.h>
66 #include <net/netisr.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_var.h>
71 #include <netinet/in_systm.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_options.h>
75 #include <netinet/igmp.h>
76 #include <netinet/igmp_var.h>
77 
78 #include <machine/in_cksum.h>
79 
80 #include <security/mac/mac_framework.h>
81 
82 #ifndef KTR_IGMPV3
83 #define KTR_IGMPV3 KTR_INET
84 #endif
85 
86 static struct igmp_ifinfo *
87 		igi_alloc_locked(struct ifnet *);
88 static void	igi_delete_locked(const struct ifnet *);
89 static void	igmp_dispatch_queue(struct ifqueue *, int, const int);
90 static void	igmp_fasttimo_vnet(void);
91 static void	igmp_final_leave(struct in_multi *, struct igmp_ifinfo *);
92 static int	igmp_handle_state_change(struct in_multi *,
93 		    struct igmp_ifinfo *);
94 static int	igmp_initial_join(struct in_multi *, struct igmp_ifinfo *);
95 static int	igmp_input_v1_query(struct ifnet *, const struct ip *,
96 		    const struct igmp *);
97 static int	igmp_input_v2_query(struct ifnet *, const struct ip *,
98 		    const struct igmp *);
99 static int	igmp_input_v3_query(struct ifnet *, const struct ip *,
100 		    /*const*/ struct igmpv3 *);
101 static int	igmp_input_v3_group_query(struct in_multi *,
102 		    struct igmp_ifinfo *, int, /*const*/ struct igmpv3 *);
103 static int	igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
104 		    /*const*/ struct igmp *);
105 static int	igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
106 		    /*const*/ struct igmp *);
107 static void	igmp_intr(struct mbuf *);
108 static int	igmp_isgroupreported(const struct in_addr);
109 static struct mbuf *
110 		igmp_ra_alloc(void);
111 #ifdef KTR
112 static char *	igmp_rec_type_to_str(const int);
113 #endif
114 static void	igmp_set_version(struct igmp_ifinfo *, const int);
115 static void	igmp_slowtimo_vnet(void);
116 static int	igmp_v1v2_queue_report(struct in_multi *, const int);
117 static void	igmp_v1v2_process_group_timer(struct in_multi *, const int);
118 static void	igmp_v1v2_process_querier_timers(struct igmp_ifinfo *);
119 static void	igmp_v2_update_group(struct in_multi *, const int);
120 static void	igmp_v3_cancel_link_timers(struct igmp_ifinfo *);
121 static void	igmp_v3_dispatch_general_query(struct igmp_ifinfo *);
122 static struct mbuf *
123 		igmp_v3_encap_report(struct ifnet *, struct mbuf *);
124 static int	igmp_v3_enqueue_group_record(struct ifqueue *,
125 		    struct in_multi *, const int, const int, const int);
126 static int	igmp_v3_enqueue_filter_change(struct ifqueue *,
127 		    struct in_multi *);
128 static void	igmp_v3_process_group_timers(struct igmp_ifinfo *,
129 		    struct ifqueue *, struct ifqueue *, struct in_multi *,
130 		    const int);
131 static int	igmp_v3_merge_state_changes(struct in_multi *,
132 		    struct ifqueue *);
133 static void	igmp_v3_suppress_group_record(struct in_multi *);
134 static int	sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
135 static int	sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
136 static int	sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
137 
138 static const struct netisr_handler igmp_nh = {
139 	.nh_name = "igmp",
140 	.nh_handler = igmp_intr,
141 	.nh_proto = NETISR_IGMP,
142 	.nh_policy = NETISR_POLICY_SOURCE,
143 };
144 
145 /*
146  * System-wide globals.
147  *
148  * Unlocked access to these is OK, except for the global IGMP output
149  * queue. The IGMP subsystem lock ends up being system-wide for the moment,
150  * because all VIMAGEs have to share a global output queue, as netisrs
151  * themselves are not virtualized.
152  *
153  * Locking:
154  *  * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
155  *    Any may be taken independently; if any are held at the same
156  *    time, the above lock order must be followed.
157  *  * All output is delegated to the netisr.
158  *    Now that Giant has been eliminated, the netisr may be inlined.
159  *  * IN_MULTI_LOCK covers in_multi.
160  *  * IGMP_LOCK covers igmp_ifinfo and any global variables in this file,
161  *    including the output queue.
162  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
163  *    per-link state iterators.
164  *  * igmp_ifinfo is valid as long as PF_INET is attached to the interface,
165  *    therefore it is not refcounted.
166  *    We allow unlocked reads of igmp_ifinfo when accessed via in_multi.
167  *
168  * Reference counting
169  *  * IGMP acquires its own reference every time an in_multi is passed to
170  *    it and the group is being joined for the first time.
171  *  * IGMP releases its reference(s) on in_multi in a deferred way,
172  *    because the operations which process the release run as part of
173  *    a loop whose control variables are directly affected by the release
174  *    (that, and not recursing on the IF_ADDR_LOCK).
175  *
176  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
177  * to a vnet in ifp->if_vnet.
178  *
179  * SMPng: XXX We may potentially race operations on ifma_protospec.
180  * The problem is that we currently lack a clean way of taking the
181  * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
182  * as anything which modifies ifma needs to be covered by that lock.
183  * So check for ifma_protospec being NULL before proceeding.
184  */
185 struct mtx		 igmp_mtx;
186 
187 struct mbuf		*m_raopt;		 /* Router Alert option */
188 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
189 
190 /*
191  * VIMAGE-wide globals.
192  *
193  * The IGMPv3 timers themselves need to run per-image, however,
194  * protosw timers run globally (see tcp).
195  * An ifnet can only be in one vimage at a time, and the loopback
196  * ifnet, loif, is itself virtualized.
197  * It would otherwise be possible to seriously hose IGMP state,
198  * and create inconsistencies in upstream multicast routing, if you have
199  * multiple VIMAGEs running on the same link joining different multicast
200  * groups, UNLESS the "primary IP address" is different. This is because
201  * IGMP for IPv4 does not force link-local addresses to be used for each
202  * node, unlike MLD for IPv6.
203  * Obviously the IGMPv3 per-interface state has per-vimage granularity
204  * also as a result.
205  *
206  * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
207  * policy to control the address used by IGMP on the link.
208  */
209 static VNET_DEFINE(int, interface_timers_running);	/* IGMPv3 general
210 							 * query response */
211 static VNET_DEFINE(int, state_change_timers_running);	/* IGMPv3 state-change
212 							 * retransmit */
213 static VNET_DEFINE(int, current_state_timers_running);	/* IGMPv1/v2 host
214 							 * report; IGMPv3 g/sg
215 							 * query response */
216 
217 #define	V_interface_timers_running	VNET(interface_timers_running)
218 #define	V_state_change_timers_running	VNET(state_change_timers_running)
219 #define	V_current_state_timers_running	VNET(current_state_timers_running)
220 
221 static VNET_DEFINE(LIST_HEAD(, igmp_ifinfo), igi_head);
222 static VNET_DEFINE(struct igmpstat, igmpstat) = {
223 	.igps_version = IGPS_VERSION_3,
224 	.igps_len = sizeof(struct igmpstat),
225 };
226 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
227 
228 #define	V_igi_head			VNET(igi_head)
229 #define	V_igmpstat			VNET(igmpstat)
230 #define	V_igmp_gsrdelay			VNET(igmp_gsrdelay)
231 
232 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
233 static VNET_DEFINE(int, igmp_sendra) = 1;
234 static VNET_DEFINE(int, igmp_sendlocal) = 1;
235 static VNET_DEFINE(int, igmp_v1enable) = 1;
236 static VNET_DEFINE(int, igmp_v2enable) = 1;
237 static VNET_DEFINE(int, igmp_legacysupp);
238 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
239 
240 #define	V_igmp_recvifkludge		VNET(igmp_recvifkludge)
241 #define	V_igmp_sendra			VNET(igmp_sendra)
242 #define	V_igmp_sendlocal		VNET(igmp_sendlocal)
243 #define	V_igmp_v1enable			VNET(igmp_v1enable)
244 #define	V_igmp_v2enable			VNET(igmp_v2enable)
245 #define	V_igmp_legacysupp		VNET(igmp_legacysupp)
246 #define	V_igmp_default_version		VNET(igmp_default_version)
247 
248 /*
249  * Virtualized sysctls.
250  */
251 SYSCTL_VNET_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_RW,
252     &VNET_NAME(igmpstat), igmpstat, "");
253 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_RW,
254     &VNET_NAME(igmp_recvifkludge), 0,
255     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
256 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_RW,
257     &VNET_NAME(igmp_sendra), 0,
258     "Send IP Router Alert option in IGMPv2/v3 messages");
259 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_RW,
260     &VNET_NAME(igmp_sendlocal), 0,
261     "Send IGMP membership reports for 224.0.0.0/24 groups");
262 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_RW,
263     &VNET_NAME(igmp_v1enable), 0,
264     "Enable backwards compatibility with IGMPv1");
265 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_RW,
266     &VNET_NAME(igmp_v2enable), 0,
267     "Enable backwards compatibility with IGMPv2");
268 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_RW,
269     &VNET_NAME(igmp_legacysupp), 0,
270     "Allow v1/v2 reports to suppress v3 group responses");
271 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, default_version,
272     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
273     &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
274     "Default version of IGMP to run on each interface");
275 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
276     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
277     &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
278     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
279 
280 /*
281  * Non-virtualized sysctls.
282  */
283 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
284     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
285     "Per-interface IGMPv3 state");
286 
287 static __inline void
288 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
289 {
290 
291 #ifdef VIMAGE
292 	m->m_pkthdr.header = ifp->if_vnet;
293 #endif /* VIMAGE */
294 	m->m_pkthdr.flowid = ifp->if_index;
295 }
296 
297 static __inline void
298 igmp_scrub_context(struct mbuf *m)
299 {
300 
301 	m->m_pkthdr.header = NULL;
302 	m->m_pkthdr.flowid = 0;
303 }
304 
305 #ifdef KTR
306 static __inline char *
307 inet_ntoa_haddr(in_addr_t haddr)
308 {
309 	struct in_addr ia;
310 
311 	ia.s_addr = htonl(haddr);
312 	return (inet_ntoa(ia));
313 }
314 #endif
315 
316 /*
317  * Restore context from a queued IGMP output chain.
318  * Return saved ifindex.
319  *
320  * VIMAGE: The assertion is there to make sure that we
321  * actually called CURVNET_SET() with what's in the mbuf chain.
322  */
323 static __inline uint32_t
324 igmp_restore_context(struct mbuf *m)
325 {
326 
327 #ifdef notyet
328 #if defined(VIMAGE) && defined(INVARIANTS)
329 	KASSERT(curvnet == (m->m_pkthdr.header),
330 	    ("%s: called when curvnet was not restored", __func__));
331 #endif
332 #endif
333 	return (m->m_pkthdr.flowid);
334 }
335 
336 /*
337  * Retrieve or set default IGMP version.
338  *
339  * VIMAGE: Assume curvnet set by caller.
340  * SMPng: NOTE: Serialized by IGMP lock.
341  */
342 static int
343 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
344 {
345 	int	 error;
346 	int	 new;
347 
348 	error = sysctl_wire_old_buffer(req, sizeof(int));
349 	if (error)
350 		return (error);
351 
352 	IGMP_LOCK();
353 
354 	new = V_igmp_default_version;
355 
356 	error = sysctl_handle_int(oidp, &new, 0, req);
357 	if (error || !req->newptr)
358 		goto out_locked;
359 
360 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
361 		error = EINVAL;
362 		goto out_locked;
363 	}
364 
365 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
366 	     V_igmp_default_version, new);
367 
368 	V_igmp_default_version = new;
369 
370 out_locked:
371 	IGMP_UNLOCK();
372 	return (error);
373 }
374 
375 /*
376  * Retrieve or set threshold between group-source queries in seconds.
377  *
378  * VIMAGE: Assume curvnet set by caller.
379  * SMPng: NOTE: Serialized by IGMP lock.
380  */
381 static int
382 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
383 {
384 	int error;
385 	int i;
386 
387 	error = sysctl_wire_old_buffer(req, sizeof(int));
388 	if (error)
389 		return (error);
390 
391 	IGMP_LOCK();
392 
393 	i = V_igmp_gsrdelay.tv_sec;
394 
395 	error = sysctl_handle_int(oidp, &i, 0, req);
396 	if (error || !req->newptr)
397 		goto out_locked;
398 
399 	if (i < -1 || i >= 60) {
400 		error = EINVAL;
401 		goto out_locked;
402 	}
403 
404 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
405 	     V_igmp_gsrdelay.tv_sec, i);
406 	V_igmp_gsrdelay.tv_sec = i;
407 
408 out_locked:
409 	IGMP_UNLOCK();
410 	return (error);
411 }
412 
413 /*
414  * Expose struct igmp_ifinfo to userland, keyed by ifindex.
415  * For use by ifmcstat(8).
416  *
417  * SMPng: NOTE: Does an unlocked ifindex space read.
418  * VIMAGE: Assume curvnet set by caller. The node handler itself
419  * is not directly virtualized.
420  */
421 static int
422 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
423 {
424 	int			*name;
425 	int			 error;
426 	u_int			 namelen;
427 	struct ifnet		*ifp;
428 	struct igmp_ifinfo	*igi;
429 
430 	name = (int *)arg1;
431 	namelen = arg2;
432 
433 	if (req->newptr != NULL)
434 		return (EPERM);
435 
436 	if (namelen != 1)
437 		return (EINVAL);
438 
439 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
440 	if (error)
441 		return (error);
442 
443 	IN_MULTI_LOCK();
444 	IGMP_LOCK();
445 
446 	if (name[0] <= 0 || name[0] > V_if_index) {
447 		error = ENOENT;
448 		goto out_locked;
449 	}
450 
451 	error = ENOENT;
452 
453 	ifp = ifnet_byindex(name[0]);
454 	if (ifp == NULL)
455 		goto out_locked;
456 
457 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
458 		if (ifp == igi->igi_ifp) {
459 			error = SYSCTL_OUT(req, igi,
460 			    sizeof(struct igmp_ifinfo));
461 			break;
462 		}
463 	}
464 
465 out_locked:
466 	IGMP_UNLOCK();
467 	IN_MULTI_UNLOCK();
468 	return (error);
469 }
470 
471 /*
472  * Dispatch an entire queue of pending packet chains
473  * using the netisr.
474  * VIMAGE: Assumes the vnet pointer has been set.
475  */
476 static void
477 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop)
478 {
479 	struct mbuf *m;
480 
481 	for (;;) {
482 		_IF_DEQUEUE(ifq, m);
483 		if (m == NULL)
484 			break;
485 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m);
486 		if (loop)
487 			m->m_flags |= M_IGMP_LOOP;
488 		netisr_dispatch(NETISR_IGMP, m);
489 		if (--limit == 0)
490 			break;
491 	}
492 }
493 
494 /*
495  * Filter outgoing IGMP report state by group.
496  *
497  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
498  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
499  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
500  * this may break certain IGMP snooping switches which rely on the old
501  * report behaviour.
502  *
503  * Return zero if the given group is one for which IGMP reports
504  * should be suppressed, or non-zero if reports should be issued.
505  */
506 static __inline int
507 igmp_isgroupreported(const struct in_addr addr)
508 {
509 
510 	if (in_allhosts(addr) ||
511 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
512 		return (0);
513 
514 	return (1);
515 }
516 
517 /*
518  * Construct a Router Alert option to use in outgoing packets.
519  */
520 static struct mbuf *
521 igmp_ra_alloc(void)
522 {
523 	struct mbuf	*m;
524 	struct ipoption	*p;
525 
526 	MGET(m, M_DONTWAIT, MT_DATA);
527 	p = mtod(m, struct ipoption *);
528 	p->ipopt_dst.s_addr = INADDR_ANY;
529 	p->ipopt_list[0] = IPOPT_RA;	/* Router Alert Option */
530 	p->ipopt_list[1] = 0x04;	/* 4 bytes long */
531 	p->ipopt_list[2] = IPOPT_EOL;	/* End of IP option list */
532 	p->ipopt_list[3] = 0x00;	/* pad byte */
533 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
534 
535 	return (m);
536 }
537 
538 /*
539  * Attach IGMP when PF_INET is attached to an interface.
540  */
541 struct igmp_ifinfo *
542 igmp_domifattach(struct ifnet *ifp)
543 {
544 	struct igmp_ifinfo *igi;
545 
546 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
547 	    __func__, ifp, ifp->if_xname);
548 
549 	IGMP_LOCK();
550 
551 	igi = igi_alloc_locked(ifp);
552 	if (!(ifp->if_flags & IFF_MULTICAST))
553 		igi->igi_flags |= IGIF_SILENT;
554 
555 	IGMP_UNLOCK();
556 
557 	return (igi);
558 }
559 
560 /*
561  * VIMAGE: assume curvnet set by caller.
562  */
563 static struct igmp_ifinfo *
564 igi_alloc_locked(/*const*/ struct ifnet *ifp)
565 {
566 	struct igmp_ifinfo *igi;
567 
568 	IGMP_LOCK_ASSERT();
569 
570 	igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO);
571 	if (igi == NULL)
572 		goto out;
573 
574 	igi->igi_ifp = ifp;
575 	igi->igi_version = V_igmp_default_version;
576 	igi->igi_flags = 0;
577 	igi->igi_rv = IGMP_RV_INIT;
578 	igi->igi_qi = IGMP_QI_INIT;
579 	igi->igi_qri = IGMP_QRI_INIT;
580 	igi->igi_uri = IGMP_URI_INIT;
581 
582 	SLIST_INIT(&igi->igi_relinmhead);
583 
584 	/*
585 	 * Responses to general queries are subject to bounds.
586 	 */
587 	IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
588 
589 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
590 
591 	CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)",
592 	     ifp, ifp->if_xname);
593 
594 out:
595 	return (igi);
596 }
597 
598 /*
599  * Hook for ifdetach.
600  *
601  * NOTE: Some finalization tasks need to run before the protocol domain
602  * is detached, but also before the link layer does its cleanup.
603  *
604  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
605  * XXX This is also bitten by unlocked ifma_protospec access.
606  */
607 void
608 igmp_ifdetach(struct ifnet *ifp)
609 {
610 	struct igmp_ifinfo	*igi;
611 	struct ifmultiaddr	*ifma;
612 	struct in_multi		*inm, *tinm;
613 
614 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
615 	    ifp->if_xname);
616 
617 	IGMP_LOCK();
618 
619 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
620 	if (igi->igi_version == IGMP_VERSION_3) {
621 		IF_ADDR_LOCK(ifp);
622 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
623 			if (ifma->ifma_addr->sa_family != AF_INET ||
624 			    ifma->ifma_protospec == NULL)
625 				continue;
626 #if 0
627 			KASSERT(ifma->ifma_protospec != NULL,
628 			    ("%s: ifma_protospec is NULL", __func__));
629 #endif
630 			inm = (struct in_multi *)ifma->ifma_protospec;
631 			if (inm->inm_state == IGMP_LEAVING_MEMBER) {
632 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
633 				    inm, inm_nrele);
634 			}
635 			inm_clear_recorded(inm);
636 		}
637 		IF_ADDR_UNLOCK(ifp);
638 		/*
639 		 * Free the in_multi reference(s) for this IGMP lifecycle.
640 		 */
641 		SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
642 		    tinm) {
643 			SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
644 			inm_release_locked(inm);
645 		}
646 	}
647 
648 	IGMP_UNLOCK();
649 }
650 
651 /*
652  * Hook for domifdetach.
653  */
654 void
655 igmp_domifdetach(struct ifnet *ifp)
656 {
657 	struct igmp_ifinfo *igi;
658 
659 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
660 	    __func__, ifp, ifp->if_xname);
661 
662 	IGMP_LOCK();
663 
664 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
665 	igi_delete_locked(ifp);
666 
667 	IGMP_UNLOCK();
668 }
669 
670 static void
671 igi_delete_locked(const struct ifnet *ifp)
672 {
673 	struct igmp_ifinfo *igi, *tigi;
674 
675 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
676 	    __func__, ifp, ifp->if_xname);
677 
678 	IGMP_LOCK_ASSERT();
679 
680 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
681 		if (igi->igi_ifp == ifp) {
682 			/*
683 			 * Free deferred General Query responses.
684 			 */
685 			_IF_DRAIN(&igi->igi_gq);
686 
687 			LIST_REMOVE(igi, igi_link);
688 
689 			KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
690 			    ("%s: there are dangling in_multi references",
691 			    __func__));
692 
693 			free(igi, M_IGMP);
694 			return;
695 		}
696 	}
697 
698 #ifdef INVARIANTS
699 	panic("%s: igmp_ifinfo not found for ifp %p\n", __func__,  ifp);
700 #endif
701 }
702 
703 /*
704  * Process a received IGMPv1 query.
705  * Return non-zero if the message should be dropped.
706  *
707  * VIMAGE: The curvnet pointer is derived from the input ifp.
708  */
709 static int
710 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
711     const struct igmp *igmp)
712 {
713 	struct ifmultiaddr	*ifma;
714 	struct igmp_ifinfo	*igi;
715 	struct in_multi		*inm;
716 
717 	/*
718 	 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
719 	 * 224.0.0.1. They are always treated as General Queries.
720 	 * igmp_group is always ignored. Do not drop it as a userland
721 	 * daemon may wish to see it.
722 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
723 	 */
724 	if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
725 		IGMPSTAT_INC(igps_rcv_badqueries);
726 		return (0);
727 	}
728 	IGMPSTAT_INC(igps_rcv_gen_queries);
729 
730 	IN_MULTI_LOCK();
731 	IGMP_LOCK();
732 
733 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
734 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
735 
736 	if (igi->igi_flags & IGIF_LOOPBACK) {
737 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
738 		    ifp, ifp->if_xname);
739 		goto out_locked;
740 	}
741 
742 	/*
743 	 * Switch to IGMPv1 host compatibility mode.
744 	 */
745 	igmp_set_version(igi, IGMP_VERSION_1);
746 
747 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
748 
749 	/*
750 	 * Start the timers in all of our group records
751 	 * for the interface on which the query arrived,
752 	 * except those which are already running.
753 	 */
754 	IF_ADDR_LOCK(ifp);
755 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
756 		if (ifma->ifma_addr->sa_family != AF_INET ||
757 		    ifma->ifma_protospec == NULL)
758 			continue;
759 		inm = (struct in_multi *)ifma->ifma_protospec;
760 		if (inm->inm_timer != 0)
761 			continue;
762 		switch (inm->inm_state) {
763 		case IGMP_NOT_MEMBER:
764 		case IGMP_SILENT_MEMBER:
765 			break;
766 		case IGMP_G_QUERY_PENDING_MEMBER:
767 		case IGMP_SG_QUERY_PENDING_MEMBER:
768 		case IGMP_REPORTING_MEMBER:
769 		case IGMP_IDLE_MEMBER:
770 		case IGMP_LAZY_MEMBER:
771 		case IGMP_SLEEPING_MEMBER:
772 		case IGMP_AWAKENING_MEMBER:
773 			inm->inm_state = IGMP_REPORTING_MEMBER;
774 			inm->inm_timer = IGMP_RANDOM_DELAY(
775 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
776 			V_current_state_timers_running = 1;
777 			break;
778 		case IGMP_LEAVING_MEMBER:
779 			break;
780 		}
781 	}
782 	IF_ADDR_UNLOCK(ifp);
783 
784 out_locked:
785 	IGMP_UNLOCK();
786 	IN_MULTI_UNLOCK();
787 
788 	return (0);
789 }
790 
791 /*
792  * Process a received IGMPv2 general or group-specific query.
793  */
794 static int
795 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
796     const struct igmp *igmp)
797 {
798 	struct ifmultiaddr	*ifma;
799 	struct igmp_ifinfo	*igi;
800 	struct in_multi		*inm;
801 	int			 is_general_query;
802 	uint16_t		 timer;
803 
804 	is_general_query = 0;
805 
806 	/*
807 	 * Validate address fields upfront.
808 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
809 	 */
810 	if (in_nullhost(igmp->igmp_group)) {
811 		/*
812 		 * IGMPv2 General Query.
813 		 * If this was not sent to the all-hosts group, ignore it.
814 		 */
815 		if (!in_allhosts(ip->ip_dst))
816 			return (0);
817 		IGMPSTAT_INC(igps_rcv_gen_queries);
818 		is_general_query = 1;
819 	} else {
820 		/* IGMPv2 Group-Specific Query. */
821 		IGMPSTAT_INC(igps_rcv_group_queries);
822 	}
823 
824 	IN_MULTI_LOCK();
825 	IGMP_LOCK();
826 
827 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
828 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
829 
830 	if (igi->igi_flags & IGIF_LOOPBACK) {
831 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
832 		    ifp, ifp->if_xname);
833 		goto out_locked;
834 	}
835 
836 	/*
837 	 * Ignore v2 query if in v1 Compatibility Mode.
838 	 */
839 	if (igi->igi_version == IGMP_VERSION_1)
840 		goto out_locked;
841 
842 	igmp_set_version(igi, IGMP_VERSION_2);
843 
844 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
845 	if (timer == 0)
846 		timer = 1;
847 
848 	if (is_general_query) {
849 		/*
850 		 * For each reporting group joined on this
851 		 * interface, kick the report timer.
852 		 */
853 		CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
854 		    ifp, ifp->if_xname);
855 		IF_ADDR_LOCK(ifp);
856 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
857 			if (ifma->ifma_addr->sa_family != AF_INET ||
858 			    ifma->ifma_protospec == NULL)
859 				continue;
860 			inm = (struct in_multi *)ifma->ifma_protospec;
861 			igmp_v2_update_group(inm, timer);
862 		}
863 		IF_ADDR_UNLOCK(ifp);
864 	} else {
865 		/*
866 		 * Group-specific IGMPv2 query, we need only
867 		 * look up the single group to process it.
868 		 */
869 		inm = inm_lookup(ifp, igmp->igmp_group);
870 		if (inm != NULL) {
871 			CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
872 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
873 			igmp_v2_update_group(inm, timer);
874 		}
875 	}
876 
877 out_locked:
878 	IGMP_UNLOCK();
879 	IN_MULTI_UNLOCK();
880 
881 	return (0);
882 }
883 
884 /*
885  * Update the report timer on a group in response to an IGMPv2 query.
886  *
887  * If we are becoming the reporting member for this group, start the timer.
888  * If we already are the reporting member for this group, and timer is
889  * below the threshold, reset it.
890  *
891  * We may be updating the group for the first time since we switched
892  * to IGMPv3. If we are, then we must clear any recorded source lists,
893  * and transition to REPORTING state; the group timer is overloaded
894  * for group and group-source query responses.
895  *
896  * Unlike IGMPv3, the delay per group should be jittered
897  * to avoid bursts of IGMPv2 reports.
898  */
899 static void
900 igmp_v2_update_group(struct in_multi *inm, const int timer)
901 {
902 
903 	CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
904 	    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
905 
906 	IN_MULTI_LOCK_ASSERT();
907 
908 	switch (inm->inm_state) {
909 	case IGMP_NOT_MEMBER:
910 	case IGMP_SILENT_MEMBER:
911 		break;
912 	case IGMP_REPORTING_MEMBER:
913 		if (inm->inm_timer != 0 &&
914 		    inm->inm_timer <= timer) {
915 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
916 			    "skipping.", __func__);
917 			break;
918 		}
919 		/* FALLTHROUGH */
920 	case IGMP_SG_QUERY_PENDING_MEMBER:
921 	case IGMP_G_QUERY_PENDING_MEMBER:
922 	case IGMP_IDLE_MEMBER:
923 	case IGMP_LAZY_MEMBER:
924 	case IGMP_AWAKENING_MEMBER:
925 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
926 		inm->inm_state = IGMP_REPORTING_MEMBER;
927 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
928 		V_current_state_timers_running = 1;
929 		break;
930 	case IGMP_SLEEPING_MEMBER:
931 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
932 		inm->inm_state = IGMP_AWAKENING_MEMBER;
933 		break;
934 	case IGMP_LEAVING_MEMBER:
935 		break;
936 	}
937 }
938 
939 /*
940  * Process a received IGMPv3 general, group-specific or
941  * group-and-source-specific query.
942  * Assumes m has already been pulled up to the full IGMP message length.
943  * Return 0 if successful, otherwise an appropriate error code is returned.
944  */
945 static int
946 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
947     /*const*/ struct igmpv3 *igmpv3)
948 {
949 	struct igmp_ifinfo	*igi;
950 	struct in_multi		*inm;
951 	int			 is_general_query;
952 	uint32_t		 maxresp, nsrc, qqi;
953 	uint16_t		 timer;
954 	uint8_t			 qrv;
955 
956 	is_general_query = 0;
957 
958 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
959 
960 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
961 	if (maxresp >= 128) {
962 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
963 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
964 	}
965 
966 	/*
967 	 * Robustness must never be less than 2 for on-wire IGMPv3.
968 	 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
969 	 * an exception for interfaces whose IGMPv3 state changes
970 	 * are redirected to loopback (e.g. MANET).
971 	 */
972 	qrv = IGMP_QRV(igmpv3->igmp_misc);
973 	if (qrv < 2) {
974 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
975 		    qrv, IGMP_RV_INIT);
976 		qrv = IGMP_RV_INIT;
977 	}
978 
979 	qqi = igmpv3->igmp_qqi;
980 	if (qqi >= 128) {
981 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
982 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
983 	}
984 
985 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
986 	if (timer == 0)
987 		timer = 1;
988 
989 	nsrc = ntohs(igmpv3->igmp_numsrc);
990 
991 	/*
992 	 * Validate address fields and versions upfront before
993 	 * accepting v3 query.
994 	 * XXX SMPng: Unlocked access to igmpstat counters here.
995 	 */
996 	if (in_nullhost(igmpv3->igmp_group)) {
997 		/*
998 		 * IGMPv3 General Query.
999 		 *
1000 		 * General Queries SHOULD be directed to 224.0.0.1.
1001 		 * A general query with a source list has undefined
1002 		 * behaviour; discard it.
1003 		 */
1004 		IGMPSTAT_INC(igps_rcv_gen_queries);
1005 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1006 			IGMPSTAT_INC(igps_rcv_badqueries);
1007 			return (0);
1008 		}
1009 		is_general_query = 1;
1010 	} else {
1011 		/* Group or group-source specific query. */
1012 		if (nsrc == 0)
1013 			IGMPSTAT_INC(igps_rcv_group_queries);
1014 		else
1015 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1016 	}
1017 
1018 	IN_MULTI_LOCK();
1019 	IGMP_LOCK();
1020 
1021 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1022 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
1023 
1024 	if (igi->igi_flags & IGIF_LOOPBACK) {
1025 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1026 		    ifp, ifp->if_xname);
1027 		goto out_locked;
1028 	}
1029 
1030 	/*
1031 	 * Discard the v3 query if we're in Compatibility Mode.
1032 	 * The RFC is not obviously worded that hosts need to stay in
1033 	 * compatibility mode until the Old Version Querier Present
1034 	 * timer expires.
1035 	 */
1036 	if (igi->igi_version != IGMP_VERSION_3) {
1037 		CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1038 		    igi->igi_version, ifp, ifp->if_xname);
1039 		goto out_locked;
1040 	}
1041 
1042 	igmp_set_version(igi, IGMP_VERSION_3);
1043 	igi->igi_rv = qrv;
1044 	igi->igi_qi = qqi;
1045 	igi->igi_qri = maxresp;
1046 
1047 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1048 	    maxresp);
1049 
1050 	if (is_general_query) {
1051 		/*
1052 		 * Schedule a current-state report on this ifp for
1053 		 * all groups, possibly containing source lists.
1054 		 * If there is a pending General Query response
1055 		 * scheduled earlier than the selected delay, do
1056 		 * not schedule any other reports.
1057 		 * Otherwise, reset the interface timer.
1058 		 */
1059 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1060 		    ifp, ifp->if_xname);
1061 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1062 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1063 			V_interface_timers_running = 1;
1064 		}
1065 	} else {
1066 		/*
1067 		 * Group-source-specific queries are throttled on
1068 		 * a per-group basis to defeat denial-of-service attempts.
1069 		 * Queries for groups we are not a member of on this
1070 		 * link are simply ignored.
1071 		 */
1072 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1073 		if (inm == NULL)
1074 			goto out_locked;
1075 		if (nsrc > 0) {
1076 			if (!ratecheck(&inm->inm_lastgsrtv,
1077 			    &V_igmp_gsrdelay)) {
1078 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1079 				    __func__);
1080 				IGMPSTAT_INC(igps_drop_gsr_queries);
1081 				goto out_locked;
1082 			}
1083 		}
1084 		CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1085 		     inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1086 		/*
1087 		 * If there is a pending General Query response
1088 		 * scheduled sooner than the selected delay, no
1089 		 * further report need be scheduled.
1090 		 * Otherwise, prepare to respond to the
1091 		 * group-specific or group-and-source query.
1092 		 */
1093 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1094 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1095 	}
1096 
1097 out_locked:
1098 	IGMP_UNLOCK();
1099 	IN_MULTI_UNLOCK();
1100 
1101 	return (0);
1102 }
1103 
1104 /*
1105  * Process a recieved IGMPv3 group-specific or group-and-source-specific
1106  * query.
1107  * Return <0 if any error occured. Currently this is ignored.
1108  */
1109 static int
1110 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1111     int timer, /*const*/ struct igmpv3 *igmpv3)
1112 {
1113 	int			 retval;
1114 	uint16_t		 nsrc;
1115 
1116 	IN_MULTI_LOCK_ASSERT();
1117 	IGMP_LOCK_ASSERT();
1118 
1119 	retval = 0;
1120 
1121 	switch (inm->inm_state) {
1122 	case IGMP_NOT_MEMBER:
1123 	case IGMP_SILENT_MEMBER:
1124 	case IGMP_SLEEPING_MEMBER:
1125 	case IGMP_LAZY_MEMBER:
1126 	case IGMP_AWAKENING_MEMBER:
1127 	case IGMP_IDLE_MEMBER:
1128 	case IGMP_LEAVING_MEMBER:
1129 		return (retval);
1130 		break;
1131 	case IGMP_REPORTING_MEMBER:
1132 	case IGMP_G_QUERY_PENDING_MEMBER:
1133 	case IGMP_SG_QUERY_PENDING_MEMBER:
1134 		break;
1135 	}
1136 
1137 	nsrc = ntohs(igmpv3->igmp_numsrc);
1138 
1139 	/*
1140 	 * Deal with group-specific queries upfront.
1141 	 * If any group query is already pending, purge any recorded
1142 	 * source-list state if it exists, and schedule a query response
1143 	 * for this group-specific query.
1144 	 */
1145 	if (nsrc == 0) {
1146 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1147 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1148 			inm_clear_recorded(inm);
1149 			timer = min(inm->inm_timer, timer);
1150 		}
1151 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1152 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1153 		V_current_state_timers_running = 1;
1154 		return (retval);
1155 	}
1156 
1157 	/*
1158 	 * Deal with the case where a group-and-source-specific query has
1159 	 * been received but a group-specific query is already pending.
1160 	 */
1161 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1162 		timer = min(inm->inm_timer, timer);
1163 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1164 		V_current_state_timers_running = 1;
1165 		return (retval);
1166 	}
1167 
1168 	/*
1169 	 * Finally, deal with the case where a group-and-source-specific
1170 	 * query has been received, where a response to a previous g-s-r
1171 	 * query exists, or none exists.
1172 	 * In this case, we need to parse the source-list which the Querier
1173 	 * has provided us with and check if we have any source list filter
1174 	 * entries at T1 for these sources. If we do not, there is no need
1175 	 * schedule a report and the query may be dropped.
1176 	 * If we do, we must record them and schedule a current-state
1177 	 * report for those sources.
1178 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1179 	 * we pass the mbuf chain pointer down to this function, and use
1180 	 * m_getptr() to walk the chain.
1181 	 */
1182 	if (inm->inm_nsrc > 0) {
1183 		const struct in_addr	*ap;
1184 		int			 i, nrecorded;
1185 
1186 		ap = (const struct in_addr *)(igmpv3 + 1);
1187 		nrecorded = 0;
1188 		for (i = 0; i < nsrc; i++, ap++) {
1189 			retval = inm_record_source(inm, ap->s_addr);
1190 			if (retval < 0)
1191 				break;
1192 			nrecorded += retval;
1193 		}
1194 		if (nrecorded > 0) {
1195 			CTR1(KTR_IGMPV3,
1196 			    "%s: schedule response to SG query", __func__);
1197 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1198 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1199 			V_current_state_timers_running = 1;
1200 		}
1201 	}
1202 
1203 	return (retval);
1204 }
1205 
1206 /*
1207  * Process a received IGMPv1 host membership report.
1208  *
1209  * NOTE: 0.0.0.0 workaround breaks const correctness.
1210  */
1211 static int
1212 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1213     /*const*/ struct igmp *igmp)
1214 {
1215 	struct in_ifaddr *ia;
1216 	struct in_multi *inm;
1217 
1218 	IGMPSTAT_INC(igps_rcv_reports);
1219 
1220 	if (ifp->if_flags & IFF_LOOPBACK)
1221 		return (0);
1222 
1223 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1224 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1225 		IGMPSTAT_INC(igps_rcv_badreports);
1226 		return (EINVAL);
1227 	}
1228 
1229 	/*
1230 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1231 	 * Booting clients may use the source address 0.0.0.0. Some
1232 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1233 	 * the interface upon which this message was received.
1234 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1235 	 */
1236 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1237 		IFP_TO_IA(ifp, ia);
1238 		if (ia != NULL) {
1239 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1240 			ifa_free(&ia->ia_ifa);
1241 		}
1242 	}
1243 
1244 	CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1245 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1246 
1247 	/*
1248 	 * IGMPv1 report suppression.
1249 	 * If we are a member of this group, and our membership should be
1250 	 * reported, stop our group timer and transition to the 'lazy' state.
1251 	 */
1252 	IN_MULTI_LOCK();
1253 	inm = inm_lookup(ifp, igmp->igmp_group);
1254 	if (inm != NULL) {
1255 		struct igmp_ifinfo *igi;
1256 
1257 		igi = inm->inm_igi;
1258 		if (igi == NULL) {
1259 			KASSERT(igi != NULL,
1260 			    ("%s: no igi for ifp %p", __func__, ifp));
1261 			goto out_locked;
1262 		}
1263 
1264 		IGMPSTAT_INC(igps_rcv_ourreports);
1265 
1266 		/*
1267 		 * If we are in IGMPv3 host mode, do not allow the
1268 		 * other host's IGMPv1 report to suppress our reports
1269 		 * unless explicitly configured to do so.
1270 		 */
1271 		if (igi->igi_version == IGMP_VERSION_3) {
1272 			if (V_igmp_legacysupp)
1273 				igmp_v3_suppress_group_record(inm);
1274 			goto out_locked;
1275 		}
1276 
1277 		inm->inm_timer = 0;
1278 
1279 		switch (inm->inm_state) {
1280 		case IGMP_NOT_MEMBER:
1281 		case IGMP_SILENT_MEMBER:
1282 			break;
1283 		case IGMP_IDLE_MEMBER:
1284 		case IGMP_LAZY_MEMBER:
1285 		case IGMP_AWAKENING_MEMBER:
1286 			CTR3(KTR_IGMPV3,
1287 			    "report suppressed for %s on ifp %p(%s)",
1288 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1289 		case IGMP_SLEEPING_MEMBER:
1290 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1291 			break;
1292 		case IGMP_REPORTING_MEMBER:
1293 			CTR3(KTR_IGMPV3,
1294 			    "report suppressed for %s on ifp %p(%s)",
1295 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1296 			if (igi->igi_version == IGMP_VERSION_1)
1297 				inm->inm_state = IGMP_LAZY_MEMBER;
1298 			else if (igi->igi_version == IGMP_VERSION_2)
1299 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1300 			break;
1301 		case IGMP_G_QUERY_PENDING_MEMBER:
1302 		case IGMP_SG_QUERY_PENDING_MEMBER:
1303 		case IGMP_LEAVING_MEMBER:
1304 			break;
1305 		}
1306 	}
1307 
1308 out_locked:
1309 	IN_MULTI_UNLOCK();
1310 
1311 	return (0);
1312 }
1313 
1314 /*
1315  * Process a received IGMPv2 host membership report.
1316  *
1317  * NOTE: 0.0.0.0 workaround breaks const correctness.
1318  */
1319 static int
1320 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1321     /*const*/ struct igmp *igmp)
1322 {
1323 	struct in_ifaddr *ia;
1324 	struct in_multi *inm;
1325 
1326 	/*
1327 	 * Make sure we don't hear our own membership report.  Fast
1328 	 * leave requires knowing that we are the only member of a
1329 	 * group.
1330 	 */
1331 	IFP_TO_IA(ifp, ia);
1332 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1333 		ifa_free(&ia->ia_ifa);
1334 		return (0);
1335 	}
1336 
1337 	IGMPSTAT_INC(igps_rcv_reports);
1338 
1339 	if (ifp->if_flags & IFF_LOOPBACK) {
1340 		if (ia != NULL)
1341 			ifa_free(&ia->ia_ifa);
1342 		return (0);
1343 	}
1344 
1345 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1346 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1347 		if (ia != NULL)
1348 			ifa_free(&ia->ia_ifa);
1349 		IGMPSTAT_INC(igps_rcv_badreports);
1350 		return (EINVAL);
1351 	}
1352 
1353 	/*
1354 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1355 	 * Booting clients may use the source address 0.0.0.0. Some
1356 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1357 	 * the interface upon which this message was received.
1358 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1359 	 */
1360 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1361 		if (ia != NULL)
1362 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1363 	}
1364 	if (ia != NULL)
1365 		ifa_free(&ia->ia_ifa);
1366 
1367 	CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1368 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1369 
1370 	/*
1371 	 * IGMPv2 report suppression.
1372 	 * If we are a member of this group, and our membership should be
1373 	 * reported, and our group timer is pending or about to be reset,
1374 	 * stop our group timer by transitioning to the 'lazy' state.
1375 	 */
1376 	IN_MULTI_LOCK();
1377 	inm = inm_lookup(ifp, igmp->igmp_group);
1378 	if (inm != NULL) {
1379 		struct igmp_ifinfo *igi;
1380 
1381 		igi = inm->inm_igi;
1382 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1383 
1384 		IGMPSTAT_INC(igps_rcv_ourreports);
1385 
1386 		/*
1387 		 * If we are in IGMPv3 host mode, do not allow the
1388 		 * other host's IGMPv1 report to suppress our reports
1389 		 * unless explicitly configured to do so.
1390 		 */
1391 		if (igi->igi_version == IGMP_VERSION_3) {
1392 			if (V_igmp_legacysupp)
1393 				igmp_v3_suppress_group_record(inm);
1394 			goto out_locked;
1395 		}
1396 
1397 		inm->inm_timer = 0;
1398 
1399 		switch (inm->inm_state) {
1400 		case IGMP_NOT_MEMBER:
1401 		case IGMP_SILENT_MEMBER:
1402 		case IGMP_SLEEPING_MEMBER:
1403 			break;
1404 		case IGMP_REPORTING_MEMBER:
1405 		case IGMP_IDLE_MEMBER:
1406 		case IGMP_AWAKENING_MEMBER:
1407 			CTR3(KTR_IGMPV3,
1408 			    "report suppressed for %s on ifp %p(%s)",
1409 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1410 		case IGMP_LAZY_MEMBER:
1411 			inm->inm_state = IGMP_LAZY_MEMBER;
1412 			break;
1413 		case IGMP_G_QUERY_PENDING_MEMBER:
1414 		case IGMP_SG_QUERY_PENDING_MEMBER:
1415 		case IGMP_LEAVING_MEMBER:
1416 			break;
1417 		}
1418 	}
1419 
1420 out_locked:
1421 	IN_MULTI_UNLOCK();
1422 
1423 	return (0);
1424 }
1425 
1426 void
1427 igmp_input(struct mbuf *m, int off)
1428 {
1429 	int iphlen;
1430 	struct ifnet *ifp;
1431 	struct igmp *igmp;
1432 	struct ip *ip;
1433 	int igmplen;
1434 	int minlen;
1435 	int queryver;
1436 
1437 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1438 
1439 	ifp = m->m_pkthdr.rcvif;
1440 
1441 	IGMPSTAT_INC(igps_rcv_total);
1442 
1443 	ip = mtod(m, struct ip *);
1444 	iphlen = off;
1445 	igmplen = ip->ip_len;
1446 
1447 	/*
1448 	 * Validate lengths.
1449 	 */
1450 	if (igmplen < IGMP_MINLEN) {
1451 		IGMPSTAT_INC(igps_rcv_tooshort);
1452 		m_freem(m);
1453 		return;
1454 	}
1455 
1456 	/*
1457 	 * Always pullup to the minimum size for v1/v2 or v3
1458 	 * to amortize calls to m_pullup().
1459 	 */
1460 	minlen = iphlen;
1461 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1462 		minlen += IGMP_V3_QUERY_MINLEN;
1463 	else
1464 		minlen += IGMP_MINLEN;
1465 	if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1466 	    (m = m_pullup(m, minlen)) == 0) {
1467 		IGMPSTAT_INC(igps_rcv_tooshort);
1468 		return;
1469 	}
1470 	ip = mtod(m, struct ip *);
1471 
1472 	/*
1473 	 * Validate checksum.
1474 	 */
1475 	m->m_data += iphlen;
1476 	m->m_len -= iphlen;
1477 	igmp = mtod(m, struct igmp *);
1478 	if (in_cksum(m, igmplen)) {
1479 		IGMPSTAT_INC(igps_rcv_badsum);
1480 		m_freem(m);
1481 		return;
1482 	}
1483 	m->m_data -= iphlen;
1484 	m->m_len += iphlen;
1485 
1486 	/*
1487 	 * IGMP control traffic is link-scope, and must have a TTL of 1.
1488 	 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1489 	 * probe packets may come from beyond the LAN.
1490 	 */
1491 	if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1492 		IGMPSTAT_INC(igps_rcv_badttl);
1493 		m_freem(m);
1494 		return;
1495 	}
1496 
1497 	switch (igmp->igmp_type) {
1498 	case IGMP_HOST_MEMBERSHIP_QUERY:
1499 		if (igmplen == IGMP_MINLEN) {
1500 			if (igmp->igmp_code == 0)
1501 				queryver = IGMP_VERSION_1;
1502 			else
1503 				queryver = IGMP_VERSION_2;
1504 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1505 			queryver = IGMP_VERSION_3;
1506 		} else {
1507 			IGMPSTAT_INC(igps_rcv_tooshort);
1508 			m_freem(m);
1509 			return;
1510 		}
1511 
1512 		switch (queryver) {
1513 		case IGMP_VERSION_1:
1514 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1515 			if (!V_igmp_v1enable)
1516 				break;
1517 			if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1518 				m_freem(m);
1519 				return;
1520 			}
1521 			break;
1522 
1523 		case IGMP_VERSION_2:
1524 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1525 			if (!V_igmp_v2enable)
1526 				break;
1527 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1528 				m_freem(m);
1529 				return;
1530 			}
1531 			break;
1532 
1533 		case IGMP_VERSION_3: {
1534 				struct igmpv3 *igmpv3;
1535 				uint16_t igmpv3len;
1536 				uint16_t srclen;
1537 				int nsrc;
1538 
1539 				IGMPSTAT_INC(igps_rcv_v3_queries);
1540 				igmpv3 = (struct igmpv3 *)igmp;
1541 				/*
1542 				 * Validate length based on source count.
1543 				 */
1544 				nsrc = ntohs(igmpv3->igmp_numsrc);
1545 				srclen = sizeof(struct in_addr) * nsrc;
1546 				if (nsrc * sizeof(in_addr_t) > srclen) {
1547 					IGMPSTAT_INC(igps_rcv_tooshort);
1548 					return;
1549 				}
1550 				/*
1551 				 * m_pullup() may modify m, so pullup in
1552 				 * this scope.
1553 				 */
1554 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1555 				    srclen;
1556 				if ((m->m_flags & M_EXT ||
1557 				     m->m_len < igmpv3len) &&
1558 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1559 					IGMPSTAT_INC(igps_rcv_tooshort);
1560 					return;
1561 				}
1562 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1563 				    + iphlen);
1564 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1565 					m_freem(m);
1566 					return;
1567 				}
1568 			}
1569 			break;
1570 		}
1571 		break;
1572 
1573 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1574 		if (!V_igmp_v1enable)
1575 			break;
1576 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1577 			m_freem(m);
1578 			return;
1579 		}
1580 		break;
1581 
1582 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1583 		if (!V_igmp_v2enable)
1584 			break;
1585 		if (!ip_checkrouteralert(m))
1586 			IGMPSTAT_INC(igps_rcv_nora);
1587 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1588 			m_freem(m);
1589 			return;
1590 		}
1591 		break;
1592 
1593 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1594 		/*
1595 		 * Hosts do not need to process IGMPv3 membership reports,
1596 		 * as report suppression is no longer required.
1597 		 */
1598 		if (!ip_checkrouteralert(m))
1599 			IGMPSTAT_INC(igps_rcv_nora);
1600 		break;
1601 
1602 	default:
1603 		break;
1604 	}
1605 
1606 	/*
1607 	 * Pass all valid IGMP packets up to any process(es) listening on a
1608 	 * raw IGMP socket.
1609 	 */
1610 	rip_input(m, off);
1611 }
1612 
1613 
1614 /*
1615  * Fast timeout handler (global).
1616  * VIMAGE: Timeout handlers are expected to service all vimages.
1617  */
1618 void
1619 igmp_fasttimo(void)
1620 {
1621 	VNET_ITERATOR_DECL(vnet_iter);
1622 
1623 	VNET_LIST_RLOCK_NOSLEEP();
1624 	VNET_FOREACH(vnet_iter) {
1625 		CURVNET_SET(vnet_iter);
1626 		igmp_fasttimo_vnet();
1627 		CURVNET_RESTORE();
1628 	}
1629 	VNET_LIST_RUNLOCK_NOSLEEP();
1630 }
1631 
1632 /*
1633  * Fast timeout handler (per-vnet).
1634  * Sends are shuffled off to a netisr to deal with Giant.
1635  *
1636  * VIMAGE: Assume caller has set up our curvnet.
1637  */
1638 static void
1639 igmp_fasttimo_vnet(void)
1640 {
1641 	struct ifqueue		 scq;	/* State-change packets */
1642 	struct ifqueue		 qrq;	/* Query response packets */
1643 	struct ifnet		*ifp;
1644 	struct igmp_ifinfo	*igi;
1645 	struct ifmultiaddr	*ifma, *tifma;
1646 	struct in_multi		*inm;
1647 	int			 loop, uri_fasthz;
1648 
1649 	loop = 0;
1650 	uri_fasthz = 0;
1651 
1652 	/*
1653 	 * Quick check to see if any work needs to be done, in order to
1654 	 * minimize the overhead of fasttimo processing.
1655 	 * SMPng: XXX Unlocked reads.
1656 	 */
1657 	if (!V_current_state_timers_running &&
1658 	    !V_interface_timers_running &&
1659 	    !V_state_change_timers_running)
1660 		return;
1661 
1662 	IN_MULTI_LOCK();
1663 	IGMP_LOCK();
1664 
1665 	/*
1666 	 * IGMPv3 General Query response timer processing.
1667 	 */
1668 	if (V_interface_timers_running) {
1669 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1670 
1671 		V_interface_timers_running = 0;
1672 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1673 			if (igi->igi_v3_timer == 0) {
1674 				/* Do nothing. */
1675 			} else if (--igi->igi_v3_timer == 0) {
1676 				igmp_v3_dispatch_general_query(igi);
1677 			} else {
1678 				V_interface_timers_running = 1;
1679 			}
1680 		}
1681 	}
1682 
1683 	if (!V_current_state_timers_running &&
1684 	    !V_state_change_timers_running)
1685 		goto out_locked;
1686 
1687 	V_current_state_timers_running = 0;
1688 	V_state_change_timers_running = 0;
1689 
1690 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1691 
1692 	/*
1693 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1694 	 * Note: Processing a v3 group timer may remove a node.
1695 	 */
1696 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1697 		ifp = igi->igi_ifp;
1698 
1699 		if (igi->igi_version == IGMP_VERSION_3) {
1700 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1701 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1702 			    PR_FASTHZ);
1703 
1704 			memset(&qrq, 0, sizeof(struct ifqueue));
1705 			IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1706 
1707 			memset(&scq, 0, sizeof(struct ifqueue));
1708 			IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1709 		}
1710 
1711 		IF_ADDR_LOCK(ifp);
1712 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1713 		    tifma) {
1714 			if (ifma->ifma_addr->sa_family != AF_INET ||
1715 			    ifma->ifma_protospec == NULL)
1716 				continue;
1717 			inm = (struct in_multi *)ifma->ifma_protospec;
1718 			switch (igi->igi_version) {
1719 			case IGMP_VERSION_1:
1720 			case IGMP_VERSION_2:
1721 				igmp_v1v2_process_group_timer(inm,
1722 				    igi->igi_version);
1723 				break;
1724 			case IGMP_VERSION_3:
1725 				igmp_v3_process_group_timers(igi, &qrq,
1726 				    &scq, inm, uri_fasthz);
1727 				break;
1728 			}
1729 		}
1730 		IF_ADDR_UNLOCK(ifp);
1731 
1732 		if (igi->igi_version == IGMP_VERSION_3) {
1733 			struct in_multi		*tinm;
1734 
1735 			igmp_dispatch_queue(&qrq, 0, loop);
1736 			igmp_dispatch_queue(&scq, 0, loop);
1737 
1738 			/*
1739 			 * Free the in_multi reference(s) for this
1740 			 * IGMP lifecycle.
1741 			 */
1742 			SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1743 			    inm_nrele, tinm) {
1744 				SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1745 				    inm_nrele);
1746 				inm_release_locked(inm);
1747 			}
1748 		}
1749 	}
1750 
1751 out_locked:
1752 	IGMP_UNLOCK();
1753 	IN_MULTI_UNLOCK();
1754 }
1755 
1756 /*
1757  * Update host report group timer for IGMPv1/v2.
1758  * Will update the global pending timer flags.
1759  */
1760 static void
1761 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1762 {
1763 	int report_timer_expired;
1764 
1765 	IN_MULTI_LOCK_ASSERT();
1766 	IGMP_LOCK_ASSERT();
1767 
1768 	if (inm->inm_timer == 0) {
1769 		report_timer_expired = 0;
1770 	} else if (--inm->inm_timer == 0) {
1771 		report_timer_expired = 1;
1772 	} else {
1773 		V_current_state_timers_running = 1;
1774 		return;
1775 	}
1776 
1777 	switch (inm->inm_state) {
1778 	case IGMP_NOT_MEMBER:
1779 	case IGMP_SILENT_MEMBER:
1780 	case IGMP_IDLE_MEMBER:
1781 	case IGMP_LAZY_MEMBER:
1782 	case IGMP_SLEEPING_MEMBER:
1783 	case IGMP_AWAKENING_MEMBER:
1784 		break;
1785 	case IGMP_REPORTING_MEMBER:
1786 		if (report_timer_expired) {
1787 			inm->inm_state = IGMP_IDLE_MEMBER;
1788 			(void)igmp_v1v2_queue_report(inm,
1789 			    (version == IGMP_VERSION_2) ?
1790 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1791 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1792 		}
1793 		break;
1794 	case IGMP_G_QUERY_PENDING_MEMBER:
1795 	case IGMP_SG_QUERY_PENDING_MEMBER:
1796 	case IGMP_LEAVING_MEMBER:
1797 		break;
1798 	}
1799 }
1800 
1801 /*
1802  * Update a group's timers for IGMPv3.
1803  * Will update the global pending timer flags.
1804  * Note: Unlocked read from igi.
1805  */
1806 static void
1807 igmp_v3_process_group_timers(struct igmp_ifinfo *igi,
1808     struct ifqueue *qrq, struct ifqueue *scq,
1809     struct in_multi *inm, const int uri_fasthz)
1810 {
1811 	int query_response_timer_expired;
1812 	int state_change_retransmit_timer_expired;
1813 
1814 	IN_MULTI_LOCK_ASSERT();
1815 	IGMP_LOCK_ASSERT();
1816 
1817 	query_response_timer_expired = 0;
1818 	state_change_retransmit_timer_expired = 0;
1819 
1820 	/*
1821 	 * During a transition from v1/v2 compatibility mode back to v3,
1822 	 * a group record in REPORTING state may still have its group
1823 	 * timer active. This is a no-op in this function; it is easier
1824 	 * to deal with it here than to complicate the slow-timeout path.
1825 	 */
1826 	if (inm->inm_timer == 0) {
1827 		query_response_timer_expired = 0;
1828 	} else if (--inm->inm_timer == 0) {
1829 		query_response_timer_expired = 1;
1830 	} else {
1831 		V_current_state_timers_running = 1;
1832 	}
1833 
1834 	if (inm->inm_sctimer == 0) {
1835 		state_change_retransmit_timer_expired = 0;
1836 	} else if (--inm->inm_sctimer == 0) {
1837 		state_change_retransmit_timer_expired = 1;
1838 	} else {
1839 		V_state_change_timers_running = 1;
1840 	}
1841 
1842 	/* We are in fasttimo, so be quick about it. */
1843 	if (!state_change_retransmit_timer_expired &&
1844 	    !query_response_timer_expired)
1845 		return;
1846 
1847 	switch (inm->inm_state) {
1848 	case IGMP_NOT_MEMBER:
1849 	case IGMP_SILENT_MEMBER:
1850 	case IGMP_SLEEPING_MEMBER:
1851 	case IGMP_LAZY_MEMBER:
1852 	case IGMP_AWAKENING_MEMBER:
1853 	case IGMP_IDLE_MEMBER:
1854 		break;
1855 	case IGMP_G_QUERY_PENDING_MEMBER:
1856 	case IGMP_SG_QUERY_PENDING_MEMBER:
1857 		/*
1858 		 * Respond to a previously pending Group-Specific
1859 		 * or Group-and-Source-Specific query by enqueueing
1860 		 * the appropriate Current-State report for
1861 		 * immediate transmission.
1862 		 */
1863 		if (query_response_timer_expired) {
1864 			int retval;
1865 
1866 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1867 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1868 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1869 			    __func__, retval);
1870 			inm->inm_state = IGMP_REPORTING_MEMBER;
1871 			/* XXX Clear recorded sources for next time. */
1872 			inm_clear_recorded(inm);
1873 		}
1874 		/* FALLTHROUGH */
1875 	case IGMP_REPORTING_MEMBER:
1876 	case IGMP_LEAVING_MEMBER:
1877 		if (state_change_retransmit_timer_expired) {
1878 			/*
1879 			 * State-change retransmission timer fired.
1880 			 * If there are any further pending retransmissions,
1881 			 * set the global pending state-change flag, and
1882 			 * reset the timer.
1883 			 */
1884 			if (--inm->inm_scrv > 0) {
1885 				inm->inm_sctimer = uri_fasthz;
1886 				V_state_change_timers_running = 1;
1887 			}
1888 			/*
1889 			 * Retransmit the previously computed state-change
1890 			 * report. If there are no further pending
1891 			 * retransmissions, the mbuf queue will be consumed.
1892 			 * Update T0 state to T1 as we have now sent
1893 			 * a state-change.
1894 			 */
1895 			(void)igmp_v3_merge_state_changes(inm, scq);
1896 
1897 			inm_commit(inm);
1898 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1899 			    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1900 
1901 			/*
1902 			 * If we are leaving the group for good, make sure
1903 			 * we release IGMP's reference to it.
1904 			 * This release must be deferred using a SLIST,
1905 			 * as we are called from a loop which traverses
1906 			 * the in_ifmultiaddr TAILQ.
1907 			 */
1908 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1909 			    inm->inm_scrv == 0) {
1910 				inm->inm_state = IGMP_NOT_MEMBER;
1911 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1912 				    inm, inm_nrele);
1913 			}
1914 		}
1915 		break;
1916 	}
1917 }
1918 
1919 
1920 /*
1921  * Suppress a group's pending response to a group or source/group query.
1922  *
1923  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1924  * Do NOT update ST1/ST0 as this operation merely suppresses
1925  * the currently pending group record.
1926  * Do NOT suppress the response to a general query. It is possible but
1927  * it would require adding another state or flag.
1928  */
1929 static void
1930 igmp_v3_suppress_group_record(struct in_multi *inm)
1931 {
1932 
1933 	IN_MULTI_LOCK_ASSERT();
1934 
1935 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1936 		("%s: not IGMPv3 mode on link", __func__));
1937 
1938 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1939 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1940 		return;
1941 
1942 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1943 		inm_clear_recorded(inm);
1944 
1945 	inm->inm_timer = 0;
1946 	inm->inm_state = IGMP_REPORTING_MEMBER;
1947 }
1948 
1949 /*
1950  * Switch to a different IGMP version on the given interface,
1951  * as per Section 7.2.1.
1952  */
1953 static void
1954 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1955 {
1956 	int old_version_timer;
1957 
1958 	IGMP_LOCK_ASSERT();
1959 
1960 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1961 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1962 
1963 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1964 		/*
1965 		 * Compute the "Older Version Querier Present" timer as per
1966 		 * Section 8.12.
1967 		 */
1968 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1969 		old_version_timer *= PR_SLOWHZ;
1970 
1971 		if (version == IGMP_VERSION_1) {
1972 			igi->igi_v1_timer = old_version_timer;
1973 			igi->igi_v2_timer = 0;
1974 		} else if (version == IGMP_VERSION_2) {
1975 			igi->igi_v1_timer = 0;
1976 			igi->igi_v2_timer = old_version_timer;
1977 		}
1978 	}
1979 
1980 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1981 		if (igi->igi_version != IGMP_VERSION_2) {
1982 			igi->igi_version = IGMP_VERSION_2;
1983 			igmp_v3_cancel_link_timers(igi);
1984 		}
1985 	} else if (igi->igi_v1_timer > 0) {
1986 		if (igi->igi_version != IGMP_VERSION_1) {
1987 			igi->igi_version = IGMP_VERSION_1;
1988 			igmp_v3_cancel_link_timers(igi);
1989 		}
1990 	}
1991 }
1992 
1993 /*
1994  * Cancel pending IGMPv3 timers for the given link and all groups
1995  * joined on it; state-change, general-query, and group-query timers.
1996  *
1997  * Only ever called on a transition from v3 to Compatibility mode. Kill
1998  * the timers stone dead (this may be expensive for large N groups), they
1999  * will be restarted if Compatibility Mode deems that they must be due to
2000  * query processing.
2001  */
2002 static void
2003 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
2004 {
2005 	struct ifmultiaddr	*ifma;
2006 	struct ifnet		*ifp;
2007 	struct in_multi		*inm;
2008 
2009 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2010 	    igi->igi_ifp, igi->igi_ifp->if_xname);
2011 
2012 	IN_MULTI_LOCK_ASSERT();
2013 	IGMP_LOCK_ASSERT();
2014 
2015 	/*
2016 	 * Stop the v3 General Query Response on this link stone dead.
2017 	 * If fasttimo is woken up due to V_interface_timers_running,
2018 	 * the flag will be cleared if there are no pending link timers.
2019 	 */
2020 	igi->igi_v3_timer = 0;
2021 
2022 	/*
2023 	 * Now clear the current-state and state-change report timers
2024 	 * for all memberships scoped to this link.
2025 	 */
2026 	ifp = igi->igi_ifp;
2027 	IF_ADDR_LOCK(ifp);
2028 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2029 		if (ifma->ifma_addr->sa_family != AF_INET ||
2030 		    ifma->ifma_protospec == NULL)
2031 			continue;
2032 		inm = (struct in_multi *)ifma->ifma_protospec;
2033 		switch (inm->inm_state) {
2034 		case IGMP_NOT_MEMBER:
2035 		case IGMP_SILENT_MEMBER:
2036 		case IGMP_IDLE_MEMBER:
2037 		case IGMP_LAZY_MEMBER:
2038 		case IGMP_SLEEPING_MEMBER:
2039 		case IGMP_AWAKENING_MEMBER:
2040 			/*
2041 			 * These states are either not relevant in v3 mode,
2042 			 * or are unreported. Do nothing.
2043 			 */
2044 			break;
2045 		case IGMP_LEAVING_MEMBER:
2046 			/*
2047 			 * If we are leaving the group and switching to
2048 			 * compatibility mode, we need to release the final
2049 			 * reference held for issuing the INCLUDE {}, and
2050 			 * transition to REPORTING to ensure the host leave
2051 			 * message is sent upstream to the old querier --
2052 			 * transition to NOT would lose the leave and race.
2053 			 *
2054 			 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
2055 			 * around inm_release_locked(), as it is not
2056 			 * a recursive mutex.
2057 			 */
2058 			IF_ADDR_UNLOCK(ifp);
2059 			inm_release_locked(inm);
2060 			IF_ADDR_LOCK(ifp);
2061 			/* FALLTHROUGH */
2062 		case IGMP_G_QUERY_PENDING_MEMBER:
2063 		case IGMP_SG_QUERY_PENDING_MEMBER:
2064 			inm_clear_recorded(inm);
2065 			/* FALLTHROUGH */
2066 		case IGMP_REPORTING_MEMBER:
2067 			inm->inm_state = IGMP_REPORTING_MEMBER;
2068 			break;
2069 		}
2070 		/*
2071 		 * Always clear state-change and group report timers.
2072 		 * Free any pending IGMPv3 state-change records.
2073 		 */
2074 		inm->inm_sctimer = 0;
2075 		inm->inm_timer = 0;
2076 		_IF_DRAIN(&inm->inm_scq);
2077 	}
2078 	IF_ADDR_UNLOCK(ifp);
2079 }
2080 
2081 /*
2082  * Update the Older Version Querier Present timers for a link.
2083  * See Section 7.2.1 of RFC 3376.
2084  */
2085 static void
2086 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2087 {
2088 
2089 	IGMP_LOCK_ASSERT();
2090 
2091 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2092 		/*
2093 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2094 		 *
2095 		 * Revert to IGMPv3.
2096 		 */
2097 		if (igi->igi_version != IGMP_VERSION_3) {
2098 			CTR5(KTR_IGMPV3,
2099 			    "%s: transition from v%d -> v%d on %p(%s)",
2100 			    __func__, igi->igi_version, IGMP_VERSION_3,
2101 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2102 			igi->igi_version = IGMP_VERSION_3;
2103 		}
2104 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2105 		/*
2106 		 * IGMPv1 Querier Present timer expired,
2107 		 * IGMPv2 Querier Present timer running.
2108 		 * If IGMPv2 was disabled since last timeout,
2109 		 * revert to IGMPv3.
2110 		 * If IGMPv2 is enabled, revert to IGMPv2.
2111 		 */
2112 		if (!V_igmp_v2enable) {
2113 			CTR5(KTR_IGMPV3,
2114 			    "%s: transition from v%d -> v%d on %p(%s)",
2115 			    __func__, igi->igi_version, IGMP_VERSION_3,
2116 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2117 			igi->igi_v2_timer = 0;
2118 			igi->igi_version = IGMP_VERSION_3;
2119 		} else {
2120 			--igi->igi_v2_timer;
2121 			if (igi->igi_version != IGMP_VERSION_2) {
2122 				CTR5(KTR_IGMPV3,
2123 				    "%s: transition from v%d -> v%d on %p(%s)",
2124 				    __func__, igi->igi_version, IGMP_VERSION_2,
2125 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2126 				igi->igi_version = IGMP_VERSION_2;
2127 			}
2128 		}
2129 	} else if (igi->igi_v1_timer > 0) {
2130 		/*
2131 		 * IGMPv1 Querier Present timer running.
2132 		 * Stop IGMPv2 timer if running.
2133 		 *
2134 		 * If IGMPv1 was disabled since last timeout,
2135 		 * revert to IGMPv3.
2136 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2137 		 */
2138 		if (!V_igmp_v1enable) {
2139 			CTR5(KTR_IGMPV3,
2140 			    "%s: transition from v%d -> v%d on %p(%s)",
2141 			    __func__, igi->igi_version, IGMP_VERSION_3,
2142 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2143 			igi->igi_v1_timer = 0;
2144 			igi->igi_version = IGMP_VERSION_3;
2145 		} else {
2146 			--igi->igi_v1_timer;
2147 		}
2148 		if (igi->igi_v2_timer > 0) {
2149 			CTR3(KTR_IGMPV3,
2150 			    "%s: cancel v2 timer on %p(%s)",
2151 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2152 			igi->igi_v2_timer = 0;
2153 		}
2154 	}
2155 }
2156 
2157 /*
2158  * Global slowtimo handler.
2159  * VIMAGE: Timeout handlers are expected to service all vimages.
2160  */
2161 void
2162 igmp_slowtimo(void)
2163 {
2164 	VNET_ITERATOR_DECL(vnet_iter);
2165 
2166 	VNET_LIST_RLOCK_NOSLEEP();
2167 	VNET_FOREACH(vnet_iter) {
2168 		CURVNET_SET(vnet_iter);
2169 		igmp_slowtimo_vnet();
2170 		CURVNET_RESTORE();
2171 	}
2172 	VNET_LIST_RUNLOCK_NOSLEEP();
2173 }
2174 
2175 /*
2176  * Per-vnet slowtimo handler.
2177  */
2178 static void
2179 igmp_slowtimo_vnet(void)
2180 {
2181 	struct igmp_ifinfo *igi;
2182 
2183 	IGMP_LOCK();
2184 
2185 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2186 		igmp_v1v2_process_querier_timers(igi);
2187 	}
2188 
2189 	IGMP_UNLOCK();
2190 }
2191 
2192 /*
2193  * Dispatch an IGMPv1/v2 host report or leave message.
2194  * These are always small enough to fit inside a single mbuf.
2195  */
2196 static int
2197 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2198 {
2199 	struct ifnet		*ifp;
2200 	struct igmp		*igmp;
2201 	struct ip		*ip;
2202 	struct mbuf		*m;
2203 
2204 	IN_MULTI_LOCK_ASSERT();
2205 	IGMP_LOCK_ASSERT();
2206 
2207 	ifp = inm->inm_ifp;
2208 
2209 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2210 	if (m == NULL)
2211 		return (ENOMEM);
2212 	MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2213 
2214 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2215 
2216 	m->m_data += sizeof(struct ip);
2217 	m->m_len = sizeof(struct igmp);
2218 
2219 	igmp = mtod(m, struct igmp *);
2220 	igmp->igmp_type = type;
2221 	igmp->igmp_code = 0;
2222 	igmp->igmp_group = inm->inm_addr;
2223 	igmp->igmp_cksum = 0;
2224 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2225 
2226 	m->m_data -= sizeof(struct ip);
2227 	m->m_len += sizeof(struct ip);
2228 
2229 	ip = mtod(m, struct ip *);
2230 	ip->ip_tos = 0;
2231 	ip->ip_len = sizeof(struct ip) + sizeof(struct igmp);
2232 	ip->ip_off = 0;
2233 	ip->ip_p = IPPROTO_IGMP;
2234 	ip->ip_src.s_addr = INADDR_ANY;
2235 
2236 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2237 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2238 	else
2239 		ip->ip_dst = inm->inm_addr;
2240 
2241 	igmp_save_context(m, ifp);
2242 
2243 	m->m_flags |= M_IGMPV2;
2244 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2245 		m->m_flags |= M_IGMP_LOOP;
2246 
2247 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2248 	netisr_dispatch(NETISR_IGMP, m);
2249 
2250 	return (0);
2251 }
2252 
2253 /*
2254  * Process a state change from the upper layer for the given IPv4 group.
2255  *
2256  * Each socket holds a reference on the in_multi in its own ip_moptions.
2257  * The socket layer will have made the necessary updates to.the group
2258  * state, it is now up to IGMP to issue a state change report if there
2259  * has been any change between T0 (when the last state-change was issued)
2260  * and T1 (now).
2261  *
2262  * We use the IGMPv3 state machine at group level. The IGMP module
2263  * however makes the decision as to which IGMP protocol version to speak.
2264  * A state change *from* INCLUDE {} always means an initial join.
2265  * A state change *to* INCLUDE {} always means a final leave.
2266  *
2267  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2268  * save ourselves a bunch of work; any exclusive mode groups need not
2269  * compute source filter lists.
2270  *
2271  * VIMAGE: curvnet should have been set by caller, as this routine
2272  * is called from the socket option handlers.
2273  */
2274 int
2275 igmp_change_state(struct in_multi *inm)
2276 {
2277 	struct igmp_ifinfo *igi;
2278 	struct ifnet *ifp;
2279 	int error;
2280 
2281 	IN_MULTI_LOCK_ASSERT();
2282 
2283 	error = 0;
2284 
2285 	/*
2286 	 * Try to detect if the upper layer just asked us to change state
2287 	 * for an interface which has now gone away.
2288 	 */
2289 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2290 	ifp = inm->inm_ifma->ifma_ifp;
2291 	if (ifp != NULL) {
2292 		/*
2293 		 * Sanity check that netinet's notion of ifp is the
2294 		 * same as net's.
2295 		 */
2296 		KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2297 	}
2298 
2299 	IGMP_LOCK();
2300 
2301 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2302 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2303 
2304 	/*
2305 	 * If we detect a state transition to or from MCAST_UNDEFINED
2306 	 * for this group, then we are starting or finishing an IGMP
2307 	 * life cycle for this group.
2308 	 */
2309 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2310 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2311 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2312 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2313 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2314 			error = igmp_initial_join(inm, igi);
2315 			goto out_locked;
2316 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2317 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2318 			igmp_final_leave(inm, igi);
2319 			goto out_locked;
2320 		}
2321 	} else {
2322 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2323 	}
2324 
2325 	error = igmp_handle_state_change(inm, igi);
2326 
2327 out_locked:
2328 	IGMP_UNLOCK();
2329 	return (error);
2330 }
2331 
2332 /*
2333  * Perform the initial join for an IGMP group.
2334  *
2335  * When joining a group:
2336  *  If the group should have its IGMP traffic suppressed, do nothing.
2337  *  IGMPv1 starts sending IGMPv1 host membership reports.
2338  *  IGMPv2 starts sending IGMPv2 host membership reports.
2339  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2340  *  initial state of the membership.
2341  */
2342 static int
2343 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2344 {
2345 	struct ifnet		*ifp;
2346 	struct ifqueue		*ifq;
2347 	int			 error, retval, syncstates;
2348 
2349 	CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2350 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2351 	    inm->inm_ifp->if_xname);
2352 
2353 	error = 0;
2354 	syncstates = 1;
2355 
2356 	ifp = inm->inm_ifp;
2357 
2358 	IN_MULTI_LOCK_ASSERT();
2359 	IGMP_LOCK_ASSERT();
2360 
2361 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2362 
2363 	/*
2364 	 * Groups joined on loopback or marked as 'not reported',
2365 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2366 	 * are never reported in any IGMP protocol exchanges.
2367 	 * All other groups enter the appropriate IGMP state machine
2368 	 * for the version in use on this link.
2369 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2370 	 * disabled for the link.
2371 	 */
2372 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2373 	    (igi->igi_flags & IGIF_SILENT) ||
2374 	    !igmp_isgroupreported(inm->inm_addr)) {
2375 		CTR1(KTR_IGMPV3,
2376 "%s: not kicking state machine for silent group", __func__);
2377 		inm->inm_state = IGMP_SILENT_MEMBER;
2378 		inm->inm_timer = 0;
2379 	} else {
2380 		/*
2381 		 * Deal with overlapping in_multi lifecycle.
2382 		 * If this group was LEAVING, then make sure
2383 		 * we drop the reference we picked up to keep the
2384 		 * group around for the final INCLUDE {} enqueue.
2385 		 */
2386 		if (igi->igi_version == IGMP_VERSION_3 &&
2387 		    inm->inm_state == IGMP_LEAVING_MEMBER)
2388 			inm_release_locked(inm);
2389 
2390 		inm->inm_state = IGMP_REPORTING_MEMBER;
2391 
2392 		switch (igi->igi_version) {
2393 		case IGMP_VERSION_1:
2394 		case IGMP_VERSION_2:
2395 			inm->inm_state = IGMP_IDLE_MEMBER;
2396 			error = igmp_v1v2_queue_report(inm,
2397 			    (igi->igi_version == IGMP_VERSION_2) ?
2398 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2399 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2400 			if (error == 0) {
2401 				inm->inm_timer = IGMP_RANDOM_DELAY(
2402 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2403 				V_current_state_timers_running = 1;
2404 			}
2405 			break;
2406 
2407 		case IGMP_VERSION_3:
2408 			/*
2409 			 * Defer update of T0 to T1, until the first copy
2410 			 * of the state change has been transmitted.
2411 			 */
2412 			syncstates = 0;
2413 
2414 			/*
2415 			 * Immediately enqueue a State-Change Report for
2416 			 * this interface, freeing any previous reports.
2417 			 * Don't kick the timers if there is nothing to do,
2418 			 * or if an error occurred.
2419 			 */
2420 			ifq = &inm->inm_scq;
2421 			_IF_DRAIN(ifq);
2422 			retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2423 			    0, 0);
2424 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2425 			    __func__, retval);
2426 			if (retval <= 0) {
2427 				error = retval * -1;
2428 				break;
2429 			}
2430 
2431 			/*
2432 			 * Schedule transmission of pending state-change
2433 			 * report up to RV times for this link. The timer
2434 			 * will fire at the next igmp_fasttimo (~200ms),
2435 			 * giving us an opportunity to merge the reports.
2436 			 */
2437 			if (igi->igi_flags & IGIF_LOOPBACK) {
2438 				inm->inm_scrv = 1;
2439 			} else {
2440 				KASSERT(igi->igi_rv > 1,
2441 				   ("%s: invalid robustness %d", __func__,
2442 				    igi->igi_rv));
2443 				inm->inm_scrv = igi->igi_rv;
2444 			}
2445 			inm->inm_sctimer = 1;
2446 			V_state_change_timers_running = 1;
2447 
2448 			error = 0;
2449 			break;
2450 		}
2451 	}
2452 
2453 	/*
2454 	 * Only update the T0 state if state change is atomic,
2455 	 * i.e. we don't need to wait for a timer to fire before we
2456 	 * can consider the state change to have been communicated.
2457 	 */
2458 	if (syncstates) {
2459 		inm_commit(inm);
2460 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2461 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2462 	}
2463 
2464 	return (error);
2465 }
2466 
2467 /*
2468  * Issue an intermediate state change during the IGMP life-cycle.
2469  */
2470 static int
2471 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2472 {
2473 	struct ifnet		*ifp;
2474 	int			 retval;
2475 
2476 	CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2477 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2478 	    inm->inm_ifp->if_xname);
2479 
2480 	ifp = inm->inm_ifp;
2481 
2482 	IN_MULTI_LOCK_ASSERT();
2483 	IGMP_LOCK_ASSERT();
2484 
2485 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2486 
2487 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2488 	    (igi->igi_flags & IGIF_SILENT) ||
2489 	    !igmp_isgroupreported(inm->inm_addr) ||
2490 	    (igi->igi_version != IGMP_VERSION_3)) {
2491 		if (!igmp_isgroupreported(inm->inm_addr)) {
2492 			CTR1(KTR_IGMPV3,
2493 "%s: not kicking state machine for silent group", __func__);
2494 		}
2495 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2496 		inm_commit(inm);
2497 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2498 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2499 		return (0);
2500 	}
2501 
2502 	_IF_DRAIN(&inm->inm_scq);
2503 
2504 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2505 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2506 	if (retval <= 0)
2507 		return (-retval);
2508 
2509 	/*
2510 	 * If record(s) were enqueued, start the state-change
2511 	 * report timer for this group.
2512 	 */
2513 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2514 	inm->inm_sctimer = 1;
2515 	V_state_change_timers_running = 1;
2516 
2517 	return (0);
2518 }
2519 
2520 /*
2521  * Perform the final leave for an IGMP group.
2522  *
2523  * When leaving a group:
2524  *  IGMPv1 does nothing.
2525  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2526  *  IGMPv3 enqueues a state-change report containing a transition
2527  *  to INCLUDE {} for immediate transmission.
2528  */
2529 static void
2530 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2531 {
2532 	int syncstates;
2533 
2534 	syncstates = 1;
2535 
2536 	CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2537 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2538 	    inm->inm_ifp->if_xname);
2539 
2540 	IN_MULTI_LOCK_ASSERT();
2541 	IGMP_LOCK_ASSERT();
2542 
2543 	switch (inm->inm_state) {
2544 	case IGMP_NOT_MEMBER:
2545 	case IGMP_SILENT_MEMBER:
2546 	case IGMP_LEAVING_MEMBER:
2547 		/* Already leaving or left; do nothing. */
2548 		CTR1(KTR_IGMPV3,
2549 "%s: not kicking state machine for silent group", __func__);
2550 		break;
2551 	case IGMP_REPORTING_MEMBER:
2552 	case IGMP_IDLE_MEMBER:
2553 	case IGMP_G_QUERY_PENDING_MEMBER:
2554 	case IGMP_SG_QUERY_PENDING_MEMBER:
2555 		if (igi->igi_version == IGMP_VERSION_2) {
2556 #ifdef INVARIANTS
2557 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2558 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2559 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2560 			     __func__);
2561 #endif
2562 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2563 			inm->inm_state = IGMP_NOT_MEMBER;
2564 		} else if (igi->igi_version == IGMP_VERSION_3) {
2565 			/*
2566 			 * Stop group timer and all pending reports.
2567 			 * Immediately enqueue a state-change report
2568 			 * TO_IN {} to be sent on the next fast timeout,
2569 			 * giving us an opportunity to merge reports.
2570 			 */
2571 			_IF_DRAIN(&inm->inm_scq);
2572 			inm->inm_timer = 0;
2573 			if (igi->igi_flags & IGIF_LOOPBACK) {
2574 				inm->inm_scrv = 1;
2575 			} else {
2576 				inm->inm_scrv = igi->igi_rv;
2577 			}
2578 			CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2579 			    "pending retransmissions.", __func__,
2580 			    inet_ntoa(inm->inm_addr),
2581 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2582 			if (inm->inm_scrv == 0) {
2583 				inm->inm_state = IGMP_NOT_MEMBER;
2584 				inm->inm_sctimer = 0;
2585 			} else {
2586 				int retval;
2587 
2588 				inm_acquire_locked(inm);
2589 
2590 				retval = igmp_v3_enqueue_group_record(
2591 				    &inm->inm_scq, inm, 1, 0, 0);
2592 				KASSERT(retval != 0,
2593 				    ("%s: enqueue record = %d", __func__,
2594 				     retval));
2595 
2596 				inm->inm_state = IGMP_LEAVING_MEMBER;
2597 				inm->inm_sctimer = 1;
2598 				V_state_change_timers_running = 1;
2599 				syncstates = 0;
2600 			}
2601 			break;
2602 		}
2603 		break;
2604 	case IGMP_LAZY_MEMBER:
2605 	case IGMP_SLEEPING_MEMBER:
2606 	case IGMP_AWAKENING_MEMBER:
2607 		/* Our reports are suppressed; do nothing. */
2608 		break;
2609 	}
2610 
2611 	if (syncstates) {
2612 		inm_commit(inm);
2613 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2614 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2615 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2616 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2617 		    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2618 	}
2619 }
2620 
2621 /*
2622  * Enqueue an IGMPv3 group record to the given output queue.
2623  *
2624  * XXX This function could do with having the allocation code
2625  * split out, and the multiple-tree-walks coalesced into a single
2626  * routine as has been done in igmp_v3_enqueue_filter_change().
2627  *
2628  * If is_state_change is zero, a current-state record is appended.
2629  * If is_state_change is non-zero, a state-change report is appended.
2630  *
2631  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2632  * If is_group_query is zero, and if there is a packet with free space
2633  * at the tail of the queue, it will be appended to providing there
2634  * is enough free space.
2635  * Otherwise a new mbuf packet chain is allocated.
2636  *
2637  * If is_source_query is non-zero, each source is checked to see if
2638  * it was recorded for a Group-Source query, and will be omitted if
2639  * it is not both in-mode and recorded.
2640  *
2641  * The function will attempt to allocate leading space in the packet
2642  * for the IP/IGMP header to be prepended without fragmenting the chain.
2643  *
2644  * If successful the size of all data appended to the queue is returned,
2645  * otherwise an error code less than zero is returned, or zero if
2646  * no record(s) were appended.
2647  */
2648 static int
2649 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2650     const int is_state_change, const int is_group_query,
2651     const int is_source_query)
2652 {
2653 	struct igmp_grouprec	 ig;
2654 	struct igmp_grouprec	*pig;
2655 	struct ifnet		*ifp;
2656 	struct ip_msource	*ims, *nims;
2657 	struct mbuf		*m0, *m, *md;
2658 	int			 error, is_filter_list_change;
2659 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2660 	int			 record_has_sources;
2661 	int			 now;
2662 	int			 type;
2663 	in_addr_t		 naddr;
2664 	uint8_t			 mode;
2665 
2666 	IN_MULTI_LOCK_ASSERT();
2667 
2668 	error = 0;
2669 	ifp = inm->inm_ifp;
2670 	is_filter_list_change = 0;
2671 	m = NULL;
2672 	m0 = NULL;
2673 	m0srcs = 0;
2674 	msrcs = 0;
2675 	nbytes = 0;
2676 	nims = NULL;
2677 	record_has_sources = 1;
2678 	pig = NULL;
2679 	type = IGMP_DO_NOTHING;
2680 	mode = inm->inm_st[1].iss_fmode;
2681 
2682 	/*
2683 	 * If we did not transition out of ASM mode during t0->t1,
2684 	 * and there are no source nodes to process, we can skip
2685 	 * the generation of source records.
2686 	 */
2687 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2688 	    inm->inm_nsrc == 0)
2689 		record_has_sources = 0;
2690 
2691 	if (is_state_change) {
2692 		/*
2693 		 * Queue a state change record.
2694 		 * If the mode did not change, and there are non-ASM
2695 		 * listeners or source filters present,
2696 		 * we potentially need to issue two records for the group.
2697 		 * If we are transitioning to MCAST_UNDEFINED, we need
2698 		 * not send any sources.
2699 		 * If there are ASM listeners, and there was no filter
2700 		 * mode transition of any kind, do nothing.
2701 		 */
2702 		if (mode != inm->inm_st[0].iss_fmode) {
2703 			if (mode == MCAST_EXCLUDE) {
2704 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2705 				    __func__);
2706 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2707 			} else {
2708 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2709 				    __func__);
2710 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2711 				if (mode == MCAST_UNDEFINED)
2712 					record_has_sources = 0;
2713 			}
2714 		} else {
2715 			if (record_has_sources) {
2716 				is_filter_list_change = 1;
2717 			} else {
2718 				type = IGMP_DO_NOTHING;
2719 			}
2720 		}
2721 	} else {
2722 		/*
2723 		 * Queue a current state record.
2724 		 */
2725 		if (mode == MCAST_EXCLUDE) {
2726 			type = IGMP_MODE_IS_EXCLUDE;
2727 		} else if (mode == MCAST_INCLUDE) {
2728 			type = IGMP_MODE_IS_INCLUDE;
2729 			KASSERT(inm->inm_st[1].iss_asm == 0,
2730 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2731 			     __func__, inm, inm->inm_st[1].iss_asm));
2732 		}
2733 	}
2734 
2735 	/*
2736 	 * Generate the filter list changes using a separate function.
2737 	 */
2738 	if (is_filter_list_change)
2739 		return (igmp_v3_enqueue_filter_change(ifq, inm));
2740 
2741 	if (type == IGMP_DO_NOTHING) {
2742 		CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2743 		    __func__, inet_ntoa(inm->inm_addr),
2744 		    inm->inm_ifp->if_xname);
2745 		return (0);
2746 	}
2747 
2748 	/*
2749 	 * If any sources are present, we must be able to fit at least
2750 	 * one in the trailing space of the tail packet's mbuf,
2751 	 * ideally more.
2752 	 */
2753 	minrec0len = sizeof(struct igmp_grouprec);
2754 	if (record_has_sources)
2755 		minrec0len += sizeof(in_addr_t);
2756 
2757 	CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2758 	    igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2759 	    inm->inm_ifp->if_xname);
2760 
2761 	/*
2762 	 * Check if we have a packet in the tail of the queue for this
2763 	 * group into which the first group record for this group will fit.
2764 	 * Otherwise allocate a new packet.
2765 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2766 	 * Note: Group records for G/GSR query responses MUST be sent
2767 	 * in their own packet.
2768 	 */
2769 	m0 = ifq->ifq_tail;
2770 	if (!is_group_query &&
2771 	    m0 != NULL &&
2772 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2773 	    (m0->m_pkthdr.len + minrec0len) <
2774 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2775 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2776 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2777 		m = m0;
2778 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2779 	} else {
2780 		if (_IF_QFULL(ifq)) {
2781 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2782 			return (-ENOMEM);
2783 		}
2784 		m = NULL;
2785 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2786 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2787 		if (!is_state_change && !is_group_query) {
2788 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2789 			if (m)
2790 				m->m_data += IGMP_LEADINGSPACE;
2791 		}
2792 		if (m == NULL) {
2793 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2794 			if (m)
2795 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2796 		}
2797 		if (m == NULL)
2798 			return (-ENOMEM);
2799 
2800 		igmp_save_context(m, ifp);
2801 
2802 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2803 	}
2804 
2805 	/*
2806 	 * Append group record.
2807 	 * If we have sources, we don't know how many yet.
2808 	 */
2809 	ig.ig_type = type;
2810 	ig.ig_datalen = 0;
2811 	ig.ig_numsrc = 0;
2812 	ig.ig_group = inm->inm_addr;
2813 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2814 		if (m != m0)
2815 			m_freem(m);
2816 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2817 		return (-ENOMEM);
2818 	}
2819 	nbytes += sizeof(struct igmp_grouprec);
2820 
2821 	/*
2822 	 * Append as many sources as will fit in the first packet.
2823 	 * If we are appending to a new packet, the chain allocation
2824 	 * may potentially use clusters; use m_getptr() in this case.
2825 	 * If we are appending to an existing packet, we need to obtain
2826 	 * a pointer to the group record after m_append(), in case a new
2827 	 * mbuf was allocated.
2828 	 * Only append sources which are in-mode at t1. If we are
2829 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2830 	 * include source entries.
2831 	 * Only report recorded sources in our filter set when responding
2832 	 * to a group-source query.
2833 	 */
2834 	if (record_has_sources) {
2835 		if (m == m0) {
2836 			md = m_last(m);
2837 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2838 			    md->m_len - nbytes);
2839 		} else {
2840 			md = m_getptr(m, 0, &off);
2841 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2842 			    off);
2843 		}
2844 		msrcs = 0;
2845 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2846 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2847 			    inet_ntoa_haddr(ims->ims_haddr));
2848 			now = ims_get_mode(inm, ims, 1);
2849 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2850 			if ((now != mode) ||
2851 			    (now == mode && mode == MCAST_UNDEFINED)) {
2852 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2853 				continue;
2854 			}
2855 			if (is_source_query && ims->ims_stp == 0) {
2856 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2857 				    __func__);
2858 				continue;
2859 			}
2860 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2861 			naddr = htonl(ims->ims_haddr);
2862 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2863 				if (m != m0)
2864 					m_freem(m);
2865 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2866 				    __func__);
2867 				return (-ENOMEM);
2868 			}
2869 			nbytes += sizeof(in_addr_t);
2870 			++msrcs;
2871 			if (msrcs == m0srcs)
2872 				break;
2873 		}
2874 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2875 		    msrcs);
2876 		pig->ig_numsrc = htons(msrcs);
2877 		nbytes += (msrcs * sizeof(in_addr_t));
2878 	}
2879 
2880 	if (is_source_query && msrcs == 0) {
2881 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2882 		if (m != m0)
2883 			m_freem(m);
2884 		return (0);
2885 	}
2886 
2887 	/*
2888 	 * We are good to go with first packet.
2889 	 */
2890 	if (m != m0) {
2891 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2892 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2893 		_IF_ENQUEUE(ifq, m);
2894 	} else
2895 		m->m_pkthdr.PH_vt.vt_nrecs++;
2896 
2897 	/*
2898 	 * No further work needed if no source list in packet(s).
2899 	 */
2900 	if (!record_has_sources)
2901 		return (nbytes);
2902 
2903 	/*
2904 	 * Whilst sources remain to be announced, we need to allocate
2905 	 * a new packet and fill out as many sources as will fit.
2906 	 * Always try for a cluster first.
2907 	 */
2908 	while (nims != NULL) {
2909 		if (_IF_QFULL(ifq)) {
2910 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2911 			return (-ENOMEM);
2912 		}
2913 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2914 		if (m)
2915 			m->m_data += IGMP_LEADINGSPACE;
2916 		if (m == NULL) {
2917 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2918 			if (m)
2919 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2920 		}
2921 		if (m == NULL)
2922 			return (-ENOMEM);
2923 		igmp_save_context(m, ifp);
2924 		md = m_getptr(m, 0, &off);
2925 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2926 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2927 
2928 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2929 			if (m != m0)
2930 				m_freem(m);
2931 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2932 			return (-ENOMEM);
2933 		}
2934 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2935 		nbytes += sizeof(struct igmp_grouprec);
2936 
2937 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2938 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2939 
2940 		msrcs = 0;
2941 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2942 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2943 			    inet_ntoa_haddr(ims->ims_haddr));
2944 			now = ims_get_mode(inm, ims, 1);
2945 			if ((now != mode) ||
2946 			    (now == mode && mode == MCAST_UNDEFINED)) {
2947 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2948 				continue;
2949 			}
2950 			if (is_source_query && ims->ims_stp == 0) {
2951 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2952 				    __func__);
2953 				continue;
2954 			}
2955 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2956 			naddr = htonl(ims->ims_haddr);
2957 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2958 				if (m != m0)
2959 					m_freem(m);
2960 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2961 				    __func__);
2962 				return (-ENOMEM);
2963 			}
2964 			++msrcs;
2965 			if (msrcs == m0srcs)
2966 				break;
2967 		}
2968 		pig->ig_numsrc = htons(msrcs);
2969 		nbytes += (msrcs * sizeof(in_addr_t));
2970 
2971 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2972 		_IF_ENQUEUE(ifq, m);
2973 	}
2974 
2975 	return (nbytes);
2976 }
2977 
2978 /*
2979  * Type used to mark record pass completion.
2980  * We exploit the fact we can cast to this easily from the
2981  * current filter modes on each ip_msource node.
2982  */
2983 typedef enum {
2984 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2985 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2986 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2987 	REC_FULL = REC_ALLOW | REC_BLOCK
2988 } rectype_t;
2989 
2990 /*
2991  * Enqueue an IGMPv3 filter list change to the given output queue.
2992  *
2993  * Source list filter state is held in an RB-tree. When the filter list
2994  * for a group is changed without changing its mode, we need to compute
2995  * the deltas between T0 and T1 for each source in the filter set,
2996  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2997  *
2998  * As we may potentially queue two record types, and the entire R-B tree
2999  * needs to be walked at once, we break this out into its own function
3000  * so we can generate a tightly packed queue of packets.
3001  *
3002  * XXX This could be written to only use one tree walk, although that makes
3003  * serializing into the mbuf chains a bit harder. For now we do two walks
3004  * which makes things easier on us, and it may or may not be harder on
3005  * the L2 cache.
3006  *
3007  * If successful the size of all data appended to the queue is returned,
3008  * otherwise an error code less than zero is returned, or zero if
3009  * no record(s) were appended.
3010  */
3011 static int
3012 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
3013 {
3014 	static const int MINRECLEN =
3015 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3016 	struct ifnet		*ifp;
3017 	struct igmp_grouprec	 ig;
3018 	struct igmp_grouprec	*pig;
3019 	struct ip_msource	*ims, *nims;
3020 	struct mbuf		*m, *m0, *md;
3021 	in_addr_t		 naddr;
3022 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3023 	int			 nallow, nblock;
3024 	uint8_t			 mode, now, then;
3025 	rectype_t		 crt, drt, nrt;
3026 
3027 	IN_MULTI_LOCK_ASSERT();
3028 
3029 	if (inm->inm_nsrc == 0 ||
3030 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3031 		return (0);
3032 
3033 	ifp = inm->inm_ifp;			/* interface */
3034 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
3035 	crt = REC_NONE;	/* current group record type */
3036 	drt = REC_NONE;	/* mask of completed group record types */
3037 	nrt = REC_NONE;	/* record type for current node */
3038 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3039 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3040 	npbytes = 0;	/* # of bytes appended this packet */
3041 	rsrcs = 0;	/* # sources encoded in current record */
3042 	schanged = 0;	/* # nodes encoded in overall filter change */
3043 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3044 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3045 	nims = NULL;	/* next tree node pointer */
3046 
3047 	/*
3048 	 * For each possible filter record mode.
3049 	 * The first kind of source we encounter tells us which
3050 	 * is the first kind of record we start appending.
3051 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3052 	 * as the inverse of the group's filter mode.
3053 	 */
3054 	while (drt != REC_FULL) {
3055 		do {
3056 			m0 = ifq->ifq_tail;
3057 			if (m0 != NULL &&
3058 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3059 			     IGMP_V3_REPORT_MAXRECS) &&
3060 			    (m0->m_pkthdr.len + MINRECLEN) <
3061 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3062 				m = m0;
3063 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3064 					    sizeof(struct igmp_grouprec)) /
3065 				    sizeof(in_addr_t);
3066 				CTR1(KTR_IGMPV3,
3067 				    "%s: use previous packet", __func__);
3068 			} else {
3069 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3070 				if (m)
3071 					m->m_data += IGMP_LEADINGSPACE;
3072 				if (m == NULL) {
3073 					m = m_gethdr(M_DONTWAIT, MT_DATA);
3074 					if (m)
3075 						MH_ALIGN(m, IGMP_LEADINGSPACE);
3076 				}
3077 				if (m == NULL) {
3078 					CTR1(KTR_IGMPV3,
3079 					    "%s: m_get*() failed", __func__);
3080 					return (-ENOMEM);
3081 				}
3082 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3083 				igmp_save_context(m, ifp);
3084 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3085 				    sizeof(struct igmp_grouprec)) /
3086 				    sizeof(in_addr_t);
3087 				npbytes = 0;
3088 				CTR1(KTR_IGMPV3,
3089 				    "%s: allocated new packet", __func__);
3090 			}
3091 			/*
3092 			 * Append the IGMP group record header to the
3093 			 * current packet's data area.
3094 			 * Recalculate pointer to free space for next
3095 			 * group record, in case m_append() allocated
3096 			 * a new mbuf or cluster.
3097 			 */
3098 			memset(&ig, 0, sizeof(ig));
3099 			ig.ig_group = inm->inm_addr;
3100 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3101 				if (m != m0)
3102 					m_freem(m);
3103 				CTR1(KTR_IGMPV3,
3104 				    "%s: m_append() failed", __func__);
3105 				return (-ENOMEM);
3106 			}
3107 			npbytes += sizeof(struct igmp_grouprec);
3108 			if (m != m0) {
3109 				/* new packet; offset in c hain */
3110 				md = m_getptr(m, npbytes -
3111 				    sizeof(struct igmp_grouprec), &off);
3112 				pig = (struct igmp_grouprec *)(mtod(md,
3113 				    uint8_t *) + off);
3114 			} else {
3115 				/* current packet; offset from last append */
3116 				md = m_last(m);
3117 				pig = (struct igmp_grouprec *)(mtod(md,
3118 				    uint8_t *) + md->m_len -
3119 				    sizeof(struct igmp_grouprec));
3120 			}
3121 			/*
3122 			 * Begin walking the tree for this record type
3123 			 * pass, or continue from where we left off
3124 			 * previously if we had to allocate a new packet.
3125 			 * Only report deltas in-mode at t1.
3126 			 * We need not report included sources as allowed
3127 			 * if we are in inclusive mode on the group,
3128 			 * however the converse is not true.
3129 			 */
3130 			rsrcs = 0;
3131 			if (nims == NULL)
3132 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3133 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3134 				CTR2(KTR_IGMPV3, "%s: visit node %s",
3135 				    __func__, inet_ntoa_haddr(ims->ims_haddr));
3136 				now = ims_get_mode(inm, ims, 1);
3137 				then = ims_get_mode(inm, ims, 0);
3138 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3139 				    __func__, then, now);
3140 				if (now == then) {
3141 					CTR1(KTR_IGMPV3,
3142 					    "%s: skip unchanged", __func__);
3143 					continue;
3144 				}
3145 				if (mode == MCAST_EXCLUDE &&
3146 				    now == MCAST_INCLUDE) {
3147 					CTR1(KTR_IGMPV3,
3148 					    "%s: skip IN src on EX group",
3149 					    __func__);
3150 					continue;
3151 				}
3152 				nrt = (rectype_t)now;
3153 				if (nrt == REC_NONE)
3154 					nrt = (rectype_t)(~mode & REC_FULL);
3155 				if (schanged++ == 0) {
3156 					crt = nrt;
3157 				} else if (crt != nrt)
3158 					continue;
3159 				naddr = htonl(ims->ims_haddr);
3160 				if (!m_append(m, sizeof(in_addr_t),
3161 				    (void *)&naddr)) {
3162 					if (m != m0)
3163 						m_freem(m);
3164 					CTR1(KTR_IGMPV3,
3165 					    "%s: m_append() failed", __func__);
3166 					return (-ENOMEM);
3167 				}
3168 				nallow += !!(crt == REC_ALLOW);
3169 				nblock += !!(crt == REC_BLOCK);
3170 				if (++rsrcs == m0srcs)
3171 					break;
3172 			}
3173 			/*
3174 			 * If we did not append any tree nodes on this
3175 			 * pass, back out of allocations.
3176 			 */
3177 			if (rsrcs == 0) {
3178 				npbytes -= sizeof(struct igmp_grouprec);
3179 				if (m != m0) {
3180 					CTR1(KTR_IGMPV3,
3181 					    "%s: m_free(m)", __func__);
3182 					m_freem(m);
3183 				} else {
3184 					CTR1(KTR_IGMPV3,
3185 					    "%s: m_adj(m, -ig)", __func__);
3186 					m_adj(m, -((int)sizeof(
3187 					    struct igmp_grouprec)));
3188 				}
3189 				continue;
3190 			}
3191 			npbytes += (rsrcs * sizeof(in_addr_t));
3192 			if (crt == REC_ALLOW)
3193 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3194 			else if (crt == REC_BLOCK)
3195 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3196 			pig->ig_numsrc = htons(rsrcs);
3197 			/*
3198 			 * Count the new group record, and enqueue this
3199 			 * packet if it wasn't already queued.
3200 			 */
3201 			m->m_pkthdr.PH_vt.vt_nrecs++;
3202 			if (m != m0)
3203 				_IF_ENQUEUE(ifq, m);
3204 			nbytes += npbytes;
3205 		} while (nims != NULL);
3206 		drt |= crt;
3207 		crt = (~crt & REC_FULL);
3208 	}
3209 
3210 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3211 	    nallow, nblock);
3212 
3213 	return (nbytes);
3214 }
3215 
3216 static int
3217 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3218 {
3219 	struct ifqueue	*gq;
3220 	struct mbuf	*m;		/* pending state-change */
3221 	struct mbuf	*m0;		/* copy of pending state-change */
3222 	struct mbuf	*mt;		/* last state-change in packet */
3223 	int		 docopy, domerge;
3224 	u_int		 recslen;
3225 
3226 	docopy = 0;
3227 	domerge = 0;
3228 	recslen = 0;
3229 
3230 	IN_MULTI_LOCK_ASSERT();
3231 	IGMP_LOCK_ASSERT();
3232 
3233 	/*
3234 	 * If there are further pending retransmissions, make a writable
3235 	 * copy of each queued state-change message before merging.
3236 	 */
3237 	if (inm->inm_scrv > 0)
3238 		docopy = 1;
3239 
3240 	gq = &inm->inm_scq;
3241 #ifdef KTR
3242 	if (gq->ifq_head == NULL) {
3243 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3244 		    __func__, inm);
3245 	}
3246 #endif
3247 
3248 	m = gq->ifq_head;
3249 	while (m != NULL) {
3250 		/*
3251 		 * Only merge the report into the current packet if
3252 		 * there is sufficient space to do so; an IGMPv3 report
3253 		 * packet may only contain 65,535 group records.
3254 		 * Always use a simple mbuf chain concatentation to do this,
3255 		 * as large state changes for single groups may have
3256 		 * allocated clusters.
3257 		 */
3258 		domerge = 0;
3259 		mt = ifscq->ifq_tail;
3260 		if (mt != NULL) {
3261 			recslen = m_length(m, NULL);
3262 
3263 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3264 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3265 			    IGMP_V3_REPORT_MAXRECS) &&
3266 			    (mt->m_pkthdr.len + recslen <=
3267 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3268 				domerge = 1;
3269 		}
3270 
3271 		if (!domerge && _IF_QFULL(gq)) {
3272 			CTR2(KTR_IGMPV3,
3273 			    "%s: outbound queue full, skipping whole packet %p",
3274 			    __func__, m);
3275 			mt = m->m_nextpkt;
3276 			if (!docopy)
3277 				m_freem(m);
3278 			m = mt;
3279 			continue;
3280 		}
3281 
3282 		if (!docopy) {
3283 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3284 			_IF_DEQUEUE(gq, m0);
3285 			m = m0->m_nextpkt;
3286 		} else {
3287 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3288 			m0 = m_dup(m, M_NOWAIT);
3289 			if (m0 == NULL)
3290 				return (ENOMEM);
3291 			m0->m_nextpkt = NULL;
3292 			m = m->m_nextpkt;
3293 		}
3294 
3295 		if (!domerge) {
3296 			CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3297 			    __func__, m0, ifscq);
3298 			_IF_ENQUEUE(ifscq, m0);
3299 		} else {
3300 			struct mbuf *mtl;	/* last mbuf of packet mt */
3301 
3302 			CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3303 			    __func__, m0, mt);
3304 
3305 			mtl = m_last(mt);
3306 			m0->m_flags &= ~M_PKTHDR;
3307 			mt->m_pkthdr.len += recslen;
3308 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3309 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3310 
3311 			mtl->m_next = m0;
3312 		}
3313 	}
3314 
3315 	return (0);
3316 }
3317 
3318 /*
3319  * Respond to a pending IGMPv3 General Query.
3320  */
3321 static void
3322 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3323 {
3324 	struct ifmultiaddr	*ifma, *tifma;
3325 	struct ifnet		*ifp;
3326 	struct in_multi		*inm;
3327 	int			 retval, loop;
3328 
3329 	IN_MULTI_LOCK_ASSERT();
3330 	IGMP_LOCK_ASSERT();
3331 
3332 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3333 	    ("%s: called when version %d", __func__, igi->igi_version));
3334 
3335 	ifp = igi->igi_ifp;
3336 
3337 	IF_ADDR_LOCK(ifp);
3338 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
3339 		if (ifma->ifma_addr->sa_family != AF_INET ||
3340 		    ifma->ifma_protospec == NULL)
3341 			continue;
3342 
3343 		inm = (struct in_multi *)ifma->ifma_protospec;
3344 		KASSERT(ifp == inm->inm_ifp,
3345 		    ("%s: inconsistent ifp", __func__));
3346 
3347 		switch (inm->inm_state) {
3348 		case IGMP_NOT_MEMBER:
3349 		case IGMP_SILENT_MEMBER:
3350 			break;
3351 		case IGMP_REPORTING_MEMBER:
3352 		case IGMP_IDLE_MEMBER:
3353 		case IGMP_LAZY_MEMBER:
3354 		case IGMP_SLEEPING_MEMBER:
3355 		case IGMP_AWAKENING_MEMBER:
3356 			inm->inm_state = IGMP_REPORTING_MEMBER;
3357 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3358 			    inm, 0, 0, 0);
3359 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3360 			    __func__, retval);
3361 			break;
3362 		case IGMP_G_QUERY_PENDING_MEMBER:
3363 		case IGMP_SG_QUERY_PENDING_MEMBER:
3364 		case IGMP_LEAVING_MEMBER:
3365 			break;
3366 		}
3367 	}
3368 	IF_ADDR_UNLOCK(ifp);
3369 
3370 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3371 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3372 
3373 	/*
3374 	 * Slew transmission of bursts over 500ms intervals.
3375 	 */
3376 	if (igi->igi_gq.ifq_head != NULL) {
3377 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3378 		    IGMP_RESPONSE_BURST_INTERVAL);
3379 		V_interface_timers_running = 1;
3380 	}
3381 }
3382 
3383 /*
3384  * Transmit the next pending IGMP message in the output queue.
3385  *
3386  * We get called from netisr_processqueue(). A mutex private to igmpoq
3387  * will be acquired and released around this routine.
3388  *
3389  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3390  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3391  * a link and uses a link-scope multicast address.
3392  */
3393 static void
3394 igmp_intr(struct mbuf *m)
3395 {
3396 	struct ip_moptions	 imo;
3397 	struct ifnet		*ifp;
3398 	struct mbuf		*ipopts, *m0;
3399 	int			 error;
3400 	uint32_t		 ifindex;
3401 
3402 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3403 
3404 	/*
3405 	 * Set VNET image pointer from enqueued mbuf chain
3406 	 * before doing anything else. Whilst we use interface
3407 	 * indexes to guard against interface detach, they are
3408 	 * unique to each VIMAGE and must be retrieved.
3409 	 */
3410 	CURVNET_SET((struct vnet *)(m->m_pkthdr.header));
3411 	ifindex = igmp_restore_context(m);
3412 
3413 	/*
3414 	 * Check if the ifnet still exists. This limits the scope of
3415 	 * any race in the absence of a global ifp lock for low cost
3416 	 * (an array lookup).
3417 	 */
3418 	ifp = ifnet_byindex(ifindex);
3419 	if (ifp == NULL) {
3420 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3421 		    __func__, m, ifindex);
3422 		m_freem(m);
3423 		IPSTAT_INC(ips_noroute);
3424 		goto out;
3425 	}
3426 
3427 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3428 
3429 	imo.imo_multicast_ttl  = 1;
3430 	imo.imo_multicast_vif  = -1;
3431 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3432 
3433 	/*
3434 	 * If the user requested that IGMP traffic be explicitly
3435 	 * redirected to the loopback interface (e.g. they are running a
3436 	 * MANET interface and the routing protocol needs to see the
3437 	 * updates), handle this now.
3438 	 */
3439 	if (m->m_flags & M_IGMP_LOOP)
3440 		imo.imo_multicast_ifp = V_loif;
3441 	else
3442 		imo.imo_multicast_ifp = ifp;
3443 
3444 	if (m->m_flags & M_IGMPV2) {
3445 		m0 = m;
3446 	} else {
3447 		m0 = igmp_v3_encap_report(ifp, m);
3448 		if (m0 == NULL) {
3449 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3450 			m_freem(m);
3451 			IPSTAT_INC(ips_odropped);
3452 			goto out;
3453 		}
3454 	}
3455 
3456 	igmp_scrub_context(m0);
3457 	m->m_flags &= ~(M_PROTOFLAGS);
3458 	m0->m_pkthdr.rcvif = V_loif;
3459 #ifdef MAC
3460 	mac_netinet_igmp_send(ifp, m0);
3461 #endif
3462 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3463 	if (error) {
3464 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3465 		goto out;
3466 	}
3467 
3468 	IGMPSTAT_INC(igps_snd_reports);
3469 
3470 out:
3471 	/*
3472 	 * We must restore the existing vnet pointer before
3473 	 * continuing as we are run from netisr context.
3474 	 */
3475 	CURVNET_RESTORE();
3476 }
3477 
3478 /*
3479  * Encapsulate an IGMPv3 report.
3480  *
3481  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3482  * chain has already had its IP/IGMPv3 header prepended. In this case
3483  * the function will not attempt to prepend; the lengths and checksums
3484  * will however be re-computed.
3485  *
3486  * Returns a pointer to the new mbuf chain head, or NULL if the
3487  * allocation failed.
3488  */
3489 static struct mbuf *
3490 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3491 {
3492 	struct igmp_report	*igmp;
3493 	struct ip		*ip;
3494 	int			 hdrlen, igmpreclen;
3495 
3496 	KASSERT((m->m_flags & M_PKTHDR),
3497 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3498 
3499 	igmpreclen = m_length(m, NULL);
3500 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3501 
3502 	if (m->m_flags & M_IGMPV3_HDR) {
3503 		igmpreclen -= hdrlen;
3504 	} else {
3505 		M_PREPEND(m, hdrlen, M_DONTWAIT);
3506 		if (m == NULL)
3507 			return (NULL);
3508 		m->m_flags |= M_IGMPV3_HDR;
3509 	}
3510 
3511 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3512 
3513 	m->m_data += sizeof(struct ip);
3514 	m->m_len -= sizeof(struct ip);
3515 
3516 	igmp = mtod(m, struct igmp_report *);
3517 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3518 	igmp->ir_rsv1 = 0;
3519 	igmp->ir_rsv2 = 0;
3520 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3521 	igmp->ir_cksum = 0;
3522 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3523 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3524 
3525 	m->m_data -= sizeof(struct ip);
3526 	m->m_len += sizeof(struct ip);
3527 
3528 	ip = mtod(m, struct ip *);
3529 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3530 	ip->ip_len = hdrlen + igmpreclen;
3531 	ip->ip_off = IP_DF;
3532 	ip->ip_p = IPPROTO_IGMP;
3533 	ip->ip_sum = 0;
3534 
3535 	ip->ip_src.s_addr = INADDR_ANY;
3536 
3537 	if (m->m_flags & M_IGMP_LOOP) {
3538 		struct in_ifaddr *ia;
3539 
3540 		IFP_TO_IA(ifp, ia);
3541 		if (ia != NULL) {
3542 			ip->ip_src = ia->ia_addr.sin_addr;
3543 			ifa_free(&ia->ia_ifa);
3544 		}
3545 	}
3546 
3547 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3548 
3549 	return (m);
3550 }
3551 
3552 #ifdef KTR
3553 static char *
3554 igmp_rec_type_to_str(const int type)
3555 {
3556 
3557 	switch (type) {
3558 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3559 			return "TO_EX";
3560 			break;
3561 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3562 			return "TO_IN";
3563 			break;
3564 		case IGMP_MODE_IS_EXCLUDE:
3565 			return "MODE_EX";
3566 			break;
3567 		case IGMP_MODE_IS_INCLUDE:
3568 			return "MODE_IN";
3569 			break;
3570 		case IGMP_ALLOW_NEW_SOURCES:
3571 			return "ALLOW_NEW";
3572 			break;
3573 		case IGMP_BLOCK_OLD_SOURCES:
3574 			return "BLOCK_OLD";
3575 			break;
3576 		default:
3577 			break;
3578 	}
3579 	return "unknown";
3580 }
3581 #endif
3582 
3583 static void
3584 igmp_init(void *unused __unused)
3585 {
3586 
3587 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3588 
3589 	IGMP_LOCK_INIT();
3590 
3591 	m_raopt = igmp_ra_alloc();
3592 
3593 	netisr_register(&igmp_nh);
3594 }
3595 SYSINIT(igmp_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_init, NULL);
3596 
3597 static void
3598 igmp_uninit(void *unused __unused)
3599 {
3600 
3601 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3602 
3603 	netisr_unregister(&igmp_nh);
3604 
3605 	m_free(m_raopt);
3606 	m_raopt = NULL;
3607 
3608 	IGMP_LOCK_DESTROY();
3609 }
3610 SYSUNINIT(igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_uninit, NULL);
3611 
3612 static void
3613 vnet_igmp_init(const void *unused __unused)
3614 {
3615 
3616 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3617 
3618 	LIST_INIT(&V_igi_head);
3619 }
3620 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_igmp_init,
3621     NULL);
3622 
3623 static void
3624 vnet_igmp_uninit(const void *unused __unused)
3625 {
3626 
3627 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3628 
3629 	KASSERT(LIST_EMPTY(&V_igi_head),
3630 	    ("%s: igi list not empty; ifnets not detached?", __func__));
3631 }
3632 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY,
3633     vnet_igmp_uninit, NULL);
3634 
3635 static int
3636 igmp_modevent(module_t mod, int type, void *unused __unused)
3637 {
3638 
3639     switch (type) {
3640     case MOD_LOAD:
3641     case MOD_UNLOAD:
3642 	break;
3643     default:
3644 	return (EOPNOTSUPP);
3645     }
3646     return (0);
3647 }
3648 
3649 static moduledata_t igmp_mod = {
3650     "igmp",
3651     igmp_modevent,
3652     0
3653 };
3654 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3655