xref: /freebsd/sys/netinet/igmp.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2007-2009 Bruce Simpson.
5  * Copyright (c) 1988 Stephen Deering.
6  * Copyright (c) 1992, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Stephen Deering of Stanford University.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
37  */
38 
39 /*
40  * Internet Group Management Protocol (IGMP) routines.
41  * [RFC1112, RFC2236, RFC3376]
42  *
43  * Written by Steve Deering, Stanford, May 1988.
44  * Modified by Rosen Sharma, Stanford, Aug 1994.
45  * Modified by Bill Fenner, Xerox PARC, Feb 1995.
46  * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995.
47  * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson.
48  *
49  * MULTICAST Revision: 3.5.1.4
50  */
51 
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
54 
55 #include "opt_ddb.h"
56 
57 #include <sys/param.h>
58 #include <sys/systm.h>
59 #include <sys/module.h>
60 #include <sys/malloc.h>
61 #include <sys/mbuf.h>
62 #include <sys/socket.h>
63 #include <sys/protosw.h>
64 #include <sys/kernel.h>
65 #include <sys/lock.h>
66 #include <sys/sysctl.h>
67 #include <sys/ktr.h>
68 #include <sys/condvar.h>
69 
70 #ifdef DDB
71 #include <ddb/ddb.h>
72 #endif
73 
74 #include <net/if.h>
75 #include <net/if_var.h>
76 #include <net/netisr.h>
77 #include <net/vnet.h>
78 
79 #include <netinet/in.h>
80 #include <netinet/in_var.h>
81 #include <netinet/in_systm.h>
82 #include <netinet/ip.h>
83 #include <netinet/ip_var.h>
84 #include <netinet/ip_options.h>
85 #include <netinet/igmp.h>
86 #include <netinet/igmp_var.h>
87 
88 #include <machine/in_cksum.h>
89 
90 #include <security/mac/mac_framework.h>
91 
92 #ifndef KTR_IGMPV3
93 #define KTR_IGMPV3 KTR_INET
94 #endif
95 
96 static struct igmp_ifsoftc *
97 		igi_alloc_locked(struct ifnet *);
98 static void	igi_delete_locked(const struct ifnet *);
99 static void	igmp_dispatch_queue(struct mbufq *, int, const int);
100 static void	igmp_fasttimo_vnet(void);
101 static void	igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
102 static int	igmp_handle_state_change(struct in_multi *,
103 		    struct igmp_ifsoftc *);
104 static int	igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
105 static int	igmp_input_v1_query(struct ifnet *, const struct ip *,
106 		    const struct igmp *);
107 static int	igmp_input_v2_query(struct ifnet *, const struct ip *,
108 		    const struct igmp *);
109 static int	igmp_input_v3_query(struct ifnet *, const struct ip *,
110 		    /*const*/ struct igmpv3 *);
111 static int	igmp_input_v3_group_query(struct in_multi *,
112 		    struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
113 static int	igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
114 		    /*const*/ struct igmp *);
115 static int	igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
116 		    /*const*/ struct igmp *);
117 static void	igmp_intr(struct mbuf *);
118 static int	igmp_isgroupreported(const struct in_addr);
119 static struct mbuf *
120 		igmp_ra_alloc(void);
121 #ifdef KTR
122 static char *	igmp_rec_type_to_str(const int);
123 #endif
124 static void	igmp_set_version(struct igmp_ifsoftc *, const int);
125 static void	igmp_slowtimo_vnet(void);
126 static int	igmp_v1v2_queue_report(struct in_multi *, const int);
127 static void	igmp_v1v2_process_group_timer(struct in_multi *, const int);
128 static void	igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
129 static void	igmp_v2_update_group(struct in_multi *, const int);
130 static void	igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
131 static void	igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
132 static struct mbuf *
133 		igmp_v3_encap_report(struct ifnet *, struct mbuf *);
134 static int	igmp_v3_enqueue_group_record(struct mbufq *,
135 		    struct in_multi *, const int, const int, const int);
136 static int	igmp_v3_enqueue_filter_change(struct mbufq *,
137 		    struct in_multi *);
138 static void	igmp_v3_process_group_timers(struct in_multi_head *,
139 		    struct mbufq *, struct mbufq *, struct in_multi *,
140 		    const int);
141 static int	igmp_v3_merge_state_changes(struct in_multi *,
142 		    struct mbufq *);
143 static void	igmp_v3_suppress_group_record(struct in_multi *);
144 static int	sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
145 static int	sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
146 static int	sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
147 static int	sysctl_igmp_stat(SYSCTL_HANDLER_ARGS);
148 
149 static const struct netisr_handler igmp_nh = {
150 	.nh_name = "igmp",
151 	.nh_handler = igmp_intr,
152 	.nh_proto = NETISR_IGMP,
153 	.nh_policy = NETISR_POLICY_SOURCE,
154 };
155 
156 /*
157  * System-wide globals.
158  *
159  * Unlocked access to these is OK, except for the global IGMP output
160  * queue. The IGMP subsystem lock ends up being system-wide for the moment,
161  * because all VIMAGEs have to share a global output queue, as netisrs
162  * themselves are not virtualized.
163  *
164  * Locking:
165  *  * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
166  *    Any may be taken independently; if any are held at the same
167  *    time, the above lock order must be followed.
168  *  * All output is delegated to the netisr.
169  *    Now that Giant has been eliminated, the netisr may be inlined.
170  *  * IN_MULTI_LIST_LOCK covers in_multi.
171  *  * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
172  *    including the output queue.
173  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
174  *    per-link state iterators.
175  *  * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
176  *    therefore it is not refcounted.
177  *    We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
178  *
179  * Reference counting
180  *  * IGMP acquires its own reference every time an in_multi is passed to
181  *    it and the group is being joined for the first time.
182  *  * IGMP releases its reference(s) on in_multi in a deferred way,
183  *    because the operations which process the release run as part of
184  *    a loop whose control variables are directly affected by the release
185  *    (that, and not recursing on the IF_ADDR_LOCK).
186  *
187  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
188  * to a vnet in ifp->if_vnet.
189  *
190  * SMPng: XXX We may potentially race operations on ifma_protospec.
191  * The problem is that we currently lack a clean way of taking the
192  * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
193  * as anything which modifies ifma needs to be covered by that lock.
194  * So check for ifma_protospec being NULL before proceeding.
195  */
196 struct mtx		 igmp_mtx;
197 
198 struct mbuf		*m_raopt;		 /* Router Alert option */
199 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
200 
201 /*
202  * VIMAGE-wide globals.
203  *
204  * The IGMPv3 timers themselves need to run per-image, however,
205  * protosw timers run globally (see tcp).
206  * An ifnet can only be in one vimage at a time, and the loopback
207  * ifnet, loif, is itself virtualized.
208  * It would otherwise be possible to seriously hose IGMP state,
209  * and create inconsistencies in upstream multicast routing, if you have
210  * multiple VIMAGEs running on the same link joining different multicast
211  * groups, UNLESS the "primary IP address" is different. This is because
212  * IGMP for IPv4 does not force link-local addresses to be used for each
213  * node, unlike MLD for IPv6.
214  * Obviously the IGMPv3 per-interface state has per-vimage granularity
215  * also as a result.
216  *
217  * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
218  * policy to control the address used by IGMP on the link.
219  */
220 VNET_DEFINE_STATIC(int, interface_timers_running);	/* IGMPv3 general
221 							 * query response */
222 VNET_DEFINE_STATIC(int, state_change_timers_running);	/* IGMPv3 state-change
223 							 * retransmit */
224 VNET_DEFINE_STATIC(int, current_state_timers_running);	/* IGMPv1/v2 host
225 							 * report; IGMPv3 g/sg
226 							 * query response */
227 
228 #define	V_interface_timers_running	VNET(interface_timers_running)
229 #define	V_state_change_timers_running	VNET(state_change_timers_running)
230 #define	V_current_state_timers_running	VNET(current_state_timers_running)
231 
232 VNET_PCPUSTAT_DEFINE(struct igmpstat, igmpstat);
233 VNET_PCPUSTAT_SYSINIT(igmpstat);
234 VNET_PCPUSTAT_SYSUNINIT(igmpstat);
235 
236 VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) =
237     LIST_HEAD_INITIALIZER(igi_head);
238 VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0};
239 
240 #define	V_igi_head			VNET(igi_head)
241 #define	V_igmp_gsrdelay			VNET(igmp_gsrdelay)
242 
243 VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1;
244 VNET_DEFINE_STATIC(int, igmp_sendra) = 1;
245 VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1;
246 VNET_DEFINE_STATIC(int, igmp_v1enable) = 1;
247 VNET_DEFINE_STATIC(int, igmp_v2enable) = 1;
248 VNET_DEFINE_STATIC(int, igmp_legacysupp);
249 VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3;
250 
251 #define	V_igmp_recvifkludge		VNET(igmp_recvifkludge)
252 #define	V_igmp_sendra			VNET(igmp_sendra)
253 #define	V_igmp_sendlocal		VNET(igmp_sendlocal)
254 #define	V_igmp_v1enable			VNET(igmp_v1enable)
255 #define	V_igmp_v2enable			VNET(igmp_v2enable)
256 #define	V_igmp_legacysupp		VNET(igmp_legacysupp)
257 #define	V_igmp_default_version		VNET(igmp_default_version)
258 
259 /*
260  * Virtualized sysctls.
261  */
262 SYSCTL_PROC(_net_inet_igmp, IGMPCTL_STATS, stats,
263     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE,
264     &VNET_NAME(igmpstat), 0, sysctl_igmp_stat, "S,igmpstat",
265     "IGMP statistics (struct igmpstat, netinet/igmp_var.h)");
266 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
267     &VNET_NAME(igmp_recvifkludge), 0,
268     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
269 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
270     &VNET_NAME(igmp_sendra), 0,
271     "Send IP Router Alert option in IGMPv2/v3 messages");
272 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
273     &VNET_NAME(igmp_sendlocal), 0,
274     "Send IGMP membership reports for 224.0.0.0/24 groups");
275 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
276     &VNET_NAME(igmp_v1enable), 0,
277     "Enable backwards compatibility with IGMPv1");
278 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
279     &VNET_NAME(igmp_v2enable), 0,
280     "Enable backwards compatibility with IGMPv2");
281 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
282     &VNET_NAME(igmp_legacysupp), 0,
283     "Allow v1/v2 reports to suppress v3 group responses");
284 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
285     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
286     &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
287     "Default version of IGMP to run on each interface");
288 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
289     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
290     &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
291     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
292 
293 /*
294  * Non-virtualized sysctls.
295  */
296 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
297     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
298     "Per-interface IGMPv3 state");
299 
300 static __inline void
301 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
302 {
303 
304 #ifdef VIMAGE
305 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
306 #endif /* VIMAGE */
307 	m->m_pkthdr.rcvif = ifp;
308 	m->m_pkthdr.flowid = ifp->if_index;
309 }
310 
311 static __inline void
312 igmp_scrub_context(struct mbuf *m)
313 {
314 
315 	m->m_pkthdr.PH_loc.ptr = NULL;
316 	m->m_pkthdr.flowid = 0;
317 }
318 
319 /*
320  * Restore context from a queued IGMP output chain.
321  * Return saved ifindex.
322  *
323  * VIMAGE: The assertion is there to make sure that we
324  * actually called CURVNET_SET() with what's in the mbuf chain.
325  */
326 static __inline uint32_t
327 igmp_restore_context(struct mbuf *m)
328 {
329 
330 #ifdef notyet
331 #if defined(VIMAGE) && defined(INVARIANTS)
332 	KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
333 	    ("%s: called when curvnet was not restored", __func__));
334 #endif
335 #endif
336 	return (m->m_pkthdr.flowid);
337 }
338 
339 /*
340  * IGMP statistics.
341  */
342 static int
343 sysctl_igmp_stat(SYSCTL_HANDLER_ARGS)
344 {
345 	struct igmpstat igps0;
346 	int error;
347 	char *p;
348 
349 	error = sysctl_wire_old_buffer(req, sizeof(struct igmpstat));
350 	if (error)
351 		return (error);
352 
353 	if (req->oldptr != NULL) {
354 		if (req->oldlen < sizeof(struct igmpstat))
355 			error = ENOMEM;
356 		else {
357 			/*
358 			 * Copy the counters, and explicitly set the struct's
359 			 * version and length fields.
360 			 */
361 			COUNTER_ARRAY_COPY(VNET(igmpstat), &igps0,
362 			    sizeof(struct igmpstat) / sizeof(uint64_t));
363 			igps0.igps_version = IGPS_VERSION_3;
364 			igps0.igps_len = IGPS_VERSION3_LEN;
365 			error = SYSCTL_OUT(req, &igps0,
366 			    sizeof(struct igmpstat));
367 		}
368 	} else
369 		req->validlen = sizeof(struct igmpstat);
370 	if (error)
371 		goto out;
372 	if (req->newptr != NULL) {
373 		if (req->newlen < sizeof(struct igmpstat))
374 			error = ENOMEM;
375 		else
376 			error = SYSCTL_IN(req, &igps0,
377 			    sizeof(igps0));
378 		if (error)
379 			goto out;
380 		/*
381 		 * igps0 must be "all zero".
382 		 */
383 		p = (char *)&igps0;
384 		while (p < (char *)&igps0 + sizeof(igps0) && *p == '\0')
385 			p++;
386 		if (p != (char *)&igps0 + sizeof(igps0)) {
387 			error = EINVAL;
388 			goto out;
389 		}
390 		COUNTER_ARRAY_ZERO(VNET(igmpstat),
391 		    sizeof(struct igmpstat) / sizeof(uint64_t));
392 	}
393 out:
394 	return (error);
395 }
396 
397 /*
398  * Retrieve or set default IGMP version.
399  *
400  * VIMAGE: Assume curvnet set by caller.
401  * SMPng: NOTE: Serialized by IGMP lock.
402  */
403 static int
404 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
405 {
406 	int	 error;
407 	int	 new;
408 
409 	error = sysctl_wire_old_buffer(req, sizeof(int));
410 	if (error)
411 		return (error);
412 
413 	IGMP_LOCK();
414 
415 	new = V_igmp_default_version;
416 
417 	error = sysctl_handle_int(oidp, &new, 0, req);
418 	if (error || !req->newptr)
419 		goto out_locked;
420 
421 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
422 		error = EINVAL;
423 		goto out_locked;
424 	}
425 
426 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
427 	     V_igmp_default_version, new);
428 
429 	V_igmp_default_version = new;
430 
431 out_locked:
432 	IGMP_UNLOCK();
433 	return (error);
434 }
435 
436 /*
437  * Retrieve or set threshold between group-source queries in seconds.
438  *
439  * VIMAGE: Assume curvnet set by caller.
440  * SMPng: NOTE: Serialized by IGMP lock.
441  */
442 static int
443 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
444 {
445 	int error;
446 	int i;
447 
448 	error = sysctl_wire_old_buffer(req, sizeof(int));
449 	if (error)
450 		return (error);
451 
452 	IGMP_LOCK();
453 
454 	i = V_igmp_gsrdelay.tv_sec;
455 
456 	error = sysctl_handle_int(oidp, &i, 0, req);
457 	if (error || !req->newptr)
458 		goto out_locked;
459 
460 	if (i < -1 || i >= 60) {
461 		error = EINVAL;
462 		goto out_locked;
463 	}
464 
465 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
466 	     V_igmp_gsrdelay.tv_sec, i);
467 	V_igmp_gsrdelay.tv_sec = i;
468 
469 out_locked:
470 	IGMP_UNLOCK();
471 	return (error);
472 }
473 
474 /*
475  * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
476  * For use by ifmcstat(8).
477  *
478  * SMPng: NOTE: Does an unlocked ifindex space read.
479  * VIMAGE: Assume curvnet set by caller. The node handler itself
480  * is not directly virtualized.
481  */
482 static int
483 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
484 {
485 	struct epoch_tracker	 et;
486 	int			*name;
487 	int			 error;
488 	u_int			 namelen;
489 	struct ifnet		*ifp;
490 	struct igmp_ifsoftc	*igi;
491 
492 	name = (int *)arg1;
493 	namelen = arg2;
494 
495 	if (req->newptr != NULL)
496 		return (EPERM);
497 
498 	if (namelen != 1)
499 		return (EINVAL);
500 
501 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
502 	if (error)
503 		return (error);
504 
505 	IN_MULTI_LIST_LOCK();
506 	IGMP_LOCK();
507 
508 	error = ENOENT;
509 
510 	NET_EPOCH_ENTER(et);
511 	ifp = ifnet_byindex(name[0]);
512 	NET_EPOCH_EXIT(et);
513 	if (ifp == NULL)
514 		goto out_locked;
515 
516 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
517 		if (ifp == igi->igi_ifp) {
518 			struct igmp_ifinfo info;
519 
520 			info.igi_version = igi->igi_version;
521 			info.igi_v1_timer = igi->igi_v1_timer;
522 			info.igi_v2_timer = igi->igi_v2_timer;
523 			info.igi_v3_timer = igi->igi_v3_timer;
524 			info.igi_flags = igi->igi_flags;
525 			info.igi_rv = igi->igi_rv;
526 			info.igi_qi = igi->igi_qi;
527 			info.igi_qri = igi->igi_qri;
528 			info.igi_uri = igi->igi_uri;
529 			error = SYSCTL_OUT(req, &info, sizeof(info));
530 			break;
531 		}
532 	}
533 
534 out_locked:
535 	IGMP_UNLOCK();
536 	IN_MULTI_LIST_UNLOCK();
537 	return (error);
538 }
539 
540 /*
541  * Dispatch an entire queue of pending packet chains
542  * using the netisr.
543  * VIMAGE: Assumes the vnet pointer has been set.
544  */
545 static void
546 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
547 {
548 	struct epoch_tracker et;
549 	struct mbuf *m;
550 
551 	NET_EPOCH_ENTER(et);
552 	while ((m = mbufq_dequeue(mq)) != NULL) {
553 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
554 		if (loop)
555 			m->m_flags |= M_IGMP_LOOP;
556 		netisr_dispatch(NETISR_IGMP, m);
557 		if (--limit == 0)
558 			break;
559 	}
560 	NET_EPOCH_EXIT(et);
561 }
562 
563 /*
564  * Filter outgoing IGMP report state by group.
565  *
566  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
567  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
568  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
569  * this may break certain IGMP snooping switches which rely on the old
570  * report behaviour.
571  *
572  * Return zero if the given group is one for which IGMP reports
573  * should be suppressed, or non-zero if reports should be issued.
574  */
575 static __inline int
576 igmp_isgroupreported(const struct in_addr addr)
577 {
578 
579 	if (in_allhosts(addr) ||
580 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
581 		return (0);
582 
583 	return (1);
584 }
585 
586 /*
587  * Construct a Router Alert option to use in outgoing packets.
588  */
589 static struct mbuf *
590 igmp_ra_alloc(void)
591 {
592 	struct mbuf	*m;
593 	struct ipoption	*p;
594 
595 	m = m_get(M_WAITOK, MT_DATA);
596 	p = mtod(m, struct ipoption *);
597 	p->ipopt_dst.s_addr = INADDR_ANY;
598 	p->ipopt_list[0] = (char)IPOPT_RA;	/* Router Alert Option */
599 	p->ipopt_list[1] = 0x04;		/* 4 bytes long */
600 	p->ipopt_list[2] = IPOPT_EOL;		/* End of IP option list */
601 	p->ipopt_list[3] = 0x00;		/* pad byte */
602 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
603 
604 	return (m);
605 }
606 
607 /*
608  * Attach IGMP when PF_INET is attached to an interface.
609  */
610 struct igmp_ifsoftc *
611 igmp_domifattach(struct ifnet *ifp)
612 {
613 	struct igmp_ifsoftc *igi;
614 
615 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
616 	    __func__, ifp, ifp->if_xname);
617 
618 	IGMP_LOCK();
619 
620 	igi = igi_alloc_locked(ifp);
621 	if (!(ifp->if_flags & IFF_MULTICAST))
622 		igi->igi_flags |= IGIF_SILENT;
623 
624 	IGMP_UNLOCK();
625 
626 	return (igi);
627 }
628 
629 /*
630  * VIMAGE: assume curvnet set by caller.
631  */
632 static struct igmp_ifsoftc *
633 igi_alloc_locked(/*const*/ struct ifnet *ifp)
634 {
635 	struct igmp_ifsoftc *igi;
636 
637 	IGMP_LOCK_ASSERT();
638 
639 	igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
640 	if (igi == NULL)
641 		goto out;
642 
643 	igi->igi_ifp = ifp;
644 	igi->igi_version = V_igmp_default_version;
645 	igi->igi_flags = 0;
646 	igi->igi_rv = IGMP_RV_INIT;
647 	igi->igi_qi = IGMP_QI_INIT;
648 	igi->igi_qri = IGMP_QRI_INIT;
649 	igi->igi_uri = IGMP_URI_INIT;
650 	mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
651 
652 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
653 
654 	CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
655 	     ifp, ifp->if_xname);
656 
657 out:
658 	return (igi);
659 }
660 
661 /*
662  * Hook for ifdetach.
663  *
664  * NOTE: Some finalization tasks need to run before the protocol domain
665  * is detached, but also before the link layer does its cleanup.
666  *
667  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
668  * XXX This is also bitten by unlocked ifma_protospec access.
669  */
670 void
671 igmp_ifdetach(struct ifnet *ifp)
672 {
673 	struct igmp_ifsoftc	*igi;
674 	struct ifmultiaddr	*ifma, *next;
675 	struct in_multi		*inm;
676 	struct in_multi_head inm_free_tmp;
677 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
678 	    ifp->if_xname);
679 
680 	SLIST_INIT(&inm_free_tmp);
681 	IGMP_LOCK();
682 
683 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
684 	if (igi->igi_version == IGMP_VERSION_3) {
685 		IF_ADDR_WLOCK(ifp);
686 	restart:
687 		CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
688 			if (ifma->ifma_addr->sa_family != AF_INET ||
689 			    ifma->ifma_protospec == NULL)
690 				continue;
691 			inm = (struct in_multi *)ifma->ifma_protospec;
692 			if (inm->inm_state == IGMP_LEAVING_MEMBER)
693 				inm_rele_locked(&inm_free_tmp, inm);
694 			inm_clear_recorded(inm);
695 			if (__predict_false(ifma_restart)) {
696 				ifma_restart = false;
697 				goto restart;
698 			}
699 		}
700 		IF_ADDR_WUNLOCK(ifp);
701 		inm_release_list_deferred(&inm_free_tmp);
702 	}
703 	IGMP_UNLOCK();
704 
705 }
706 
707 /*
708  * Hook for domifdetach.
709  */
710 void
711 igmp_domifdetach(struct ifnet *ifp)
712 {
713 
714 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
715 	    __func__, ifp, ifp->if_xname);
716 
717 	IGMP_LOCK();
718 	igi_delete_locked(ifp);
719 	IGMP_UNLOCK();
720 }
721 
722 static void
723 igi_delete_locked(const struct ifnet *ifp)
724 {
725 	struct igmp_ifsoftc *igi, *tigi;
726 
727 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
728 	    __func__, ifp, ifp->if_xname);
729 
730 	IGMP_LOCK_ASSERT();
731 
732 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
733 		if (igi->igi_ifp == ifp) {
734 			/*
735 			 * Free deferred General Query responses.
736 			 */
737 			mbufq_drain(&igi->igi_gq);
738 
739 			LIST_REMOVE(igi, igi_link);
740 			free(igi, M_IGMP);
741 			return;
742 		}
743 	}
744 }
745 
746 /*
747  * Process a received IGMPv1 query.
748  * Return non-zero if the message should be dropped.
749  *
750  * VIMAGE: The curvnet pointer is derived from the input ifp.
751  */
752 static int
753 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
754     const struct igmp *igmp)
755 {
756 	struct ifmultiaddr	*ifma;
757 	struct igmp_ifsoftc	*igi;
758 	struct in_multi		*inm;
759 
760 	NET_EPOCH_ASSERT();
761 
762 	/*
763 	 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
764 	 * 224.0.0.1. They are always treated as General Queries.
765 	 * igmp_group is always ignored. Do not drop it as a userland
766 	 * daemon may wish to see it.
767 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
768 	 */
769 	if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
770 		IGMPSTAT_INC(igps_rcv_badqueries);
771 		return (0);
772 	}
773 	IGMPSTAT_INC(igps_rcv_gen_queries);
774 
775 	IN_MULTI_LIST_LOCK();
776 	IGMP_LOCK();
777 
778 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
779 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
780 
781 	if (igi->igi_flags & IGIF_LOOPBACK) {
782 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
783 		    ifp, ifp->if_xname);
784 		goto out_locked;
785 	}
786 
787 	/*
788 	 * Switch to IGMPv1 host compatibility mode.
789 	 */
790 	igmp_set_version(igi, IGMP_VERSION_1);
791 
792 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
793 
794 	/*
795 	 * Start the timers in all of our group records
796 	 * for the interface on which the query arrived,
797 	 * except those which are already running.
798 	 */
799 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
800 		if (ifma->ifma_addr->sa_family != AF_INET ||
801 		    ifma->ifma_protospec == NULL)
802 			continue;
803 		inm = (struct in_multi *)ifma->ifma_protospec;
804 		if (inm->inm_timer != 0)
805 			continue;
806 		switch (inm->inm_state) {
807 		case IGMP_NOT_MEMBER:
808 		case IGMP_SILENT_MEMBER:
809 			break;
810 		case IGMP_G_QUERY_PENDING_MEMBER:
811 		case IGMP_SG_QUERY_PENDING_MEMBER:
812 		case IGMP_REPORTING_MEMBER:
813 		case IGMP_IDLE_MEMBER:
814 		case IGMP_LAZY_MEMBER:
815 		case IGMP_SLEEPING_MEMBER:
816 		case IGMP_AWAKENING_MEMBER:
817 			inm->inm_state = IGMP_REPORTING_MEMBER;
818 			inm->inm_timer = IGMP_RANDOM_DELAY(
819 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
820 			V_current_state_timers_running = 1;
821 			break;
822 		case IGMP_LEAVING_MEMBER:
823 			break;
824 		}
825 	}
826 
827 out_locked:
828 	IGMP_UNLOCK();
829 	IN_MULTI_LIST_UNLOCK();
830 
831 	return (0);
832 }
833 
834 /*
835  * Process a received IGMPv2 general or group-specific query.
836  */
837 static int
838 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
839     const struct igmp *igmp)
840 {
841 	struct ifmultiaddr	*ifma;
842 	struct igmp_ifsoftc	*igi;
843 	struct in_multi		*inm;
844 	int			 is_general_query;
845 	uint16_t		 timer;
846 
847 	NET_EPOCH_ASSERT();
848 
849 	is_general_query = 0;
850 
851 	/*
852 	 * Validate address fields upfront.
853 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
854 	 */
855 	if (in_nullhost(igmp->igmp_group)) {
856 		/*
857 		 * IGMPv2 General Query.
858 		 * If this was not sent to the all-hosts group, ignore it.
859 		 */
860 		if (!in_allhosts(ip->ip_dst))
861 			return (0);
862 		IGMPSTAT_INC(igps_rcv_gen_queries);
863 		is_general_query = 1;
864 	} else {
865 		/* IGMPv2 Group-Specific Query. */
866 		IGMPSTAT_INC(igps_rcv_group_queries);
867 	}
868 
869 	IN_MULTI_LIST_LOCK();
870 	IGMP_LOCK();
871 
872 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
873 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
874 
875 	if (igi->igi_flags & IGIF_LOOPBACK) {
876 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
877 		    ifp, ifp->if_xname);
878 		goto out_locked;
879 	}
880 
881 	/*
882 	 * Ignore v2 query if in v1 Compatibility Mode.
883 	 */
884 	if (igi->igi_version == IGMP_VERSION_1)
885 		goto out_locked;
886 
887 	igmp_set_version(igi, IGMP_VERSION_2);
888 
889 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
890 	if (timer == 0)
891 		timer = 1;
892 
893 	if (is_general_query) {
894 		/*
895 		 * For each reporting group joined on this
896 		 * interface, kick the report timer.
897 		 */
898 		CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
899 		    ifp, ifp->if_xname);
900 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
901 			if (ifma->ifma_addr->sa_family != AF_INET ||
902 			    ifma->ifma_protospec == NULL)
903 				continue;
904 			inm = (struct in_multi *)ifma->ifma_protospec;
905 			igmp_v2_update_group(inm, timer);
906 		}
907 	} else {
908 		/*
909 		 * Group-specific IGMPv2 query, we need only
910 		 * look up the single group to process it.
911 		 */
912 		inm = inm_lookup(ifp, igmp->igmp_group);
913 		if (inm != NULL) {
914 			CTR3(KTR_IGMPV3,
915 			    "process v2 query 0x%08x on ifp %p(%s)",
916 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
917 			igmp_v2_update_group(inm, timer);
918 		}
919 	}
920 
921 out_locked:
922 	IGMP_UNLOCK();
923 	IN_MULTI_LIST_UNLOCK();
924 
925 	return (0);
926 }
927 
928 /*
929  * Update the report timer on a group in response to an IGMPv2 query.
930  *
931  * If we are becoming the reporting member for this group, start the timer.
932  * If we already are the reporting member for this group, and timer is
933  * below the threshold, reset it.
934  *
935  * We may be updating the group for the first time since we switched
936  * to IGMPv3. If we are, then we must clear any recorded source lists,
937  * and transition to REPORTING state; the group timer is overloaded
938  * for group and group-source query responses.
939  *
940  * Unlike IGMPv3, the delay per group should be jittered
941  * to avoid bursts of IGMPv2 reports.
942  */
943 static void
944 igmp_v2_update_group(struct in_multi *inm, const int timer)
945 {
946 
947 	CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
948 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
949 
950 	IN_MULTI_LIST_LOCK_ASSERT();
951 
952 	switch (inm->inm_state) {
953 	case IGMP_NOT_MEMBER:
954 	case IGMP_SILENT_MEMBER:
955 		break;
956 	case IGMP_REPORTING_MEMBER:
957 		if (inm->inm_timer != 0 &&
958 		    inm->inm_timer <= timer) {
959 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
960 			    "skipping.", __func__);
961 			break;
962 		}
963 		/* FALLTHROUGH */
964 	case IGMP_SG_QUERY_PENDING_MEMBER:
965 	case IGMP_G_QUERY_PENDING_MEMBER:
966 	case IGMP_IDLE_MEMBER:
967 	case IGMP_LAZY_MEMBER:
968 	case IGMP_AWAKENING_MEMBER:
969 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
970 		inm->inm_state = IGMP_REPORTING_MEMBER;
971 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
972 		V_current_state_timers_running = 1;
973 		break;
974 	case IGMP_SLEEPING_MEMBER:
975 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
976 		inm->inm_state = IGMP_AWAKENING_MEMBER;
977 		break;
978 	case IGMP_LEAVING_MEMBER:
979 		break;
980 	}
981 }
982 
983 /*
984  * Process a received IGMPv3 general, group-specific or
985  * group-and-source-specific query.
986  * Assumes m has already been pulled up to the full IGMP message length.
987  * Return 0 if successful, otherwise an appropriate error code is returned.
988  */
989 static int
990 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
991     /*const*/ struct igmpv3 *igmpv3)
992 {
993 	struct igmp_ifsoftc	*igi;
994 	struct in_multi		*inm;
995 	int			 is_general_query;
996 	uint32_t		 maxresp, nsrc, qqi;
997 	uint16_t		 timer;
998 	uint8_t			 qrv;
999 
1000 	is_general_query = 0;
1001 
1002 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
1003 
1004 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
1005 	if (maxresp >= 128) {
1006 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
1007 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
1008 	}
1009 
1010 	/*
1011 	 * Robustness must never be less than 2 for on-wire IGMPv3.
1012 	 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
1013 	 * an exception for interfaces whose IGMPv3 state changes
1014 	 * are redirected to loopback (e.g. MANET).
1015 	 */
1016 	qrv = IGMP_QRV(igmpv3->igmp_misc);
1017 	if (qrv < 2) {
1018 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
1019 		    qrv, IGMP_RV_INIT);
1020 		qrv = IGMP_RV_INIT;
1021 	}
1022 
1023 	qqi = igmpv3->igmp_qqi;
1024 	if (qqi >= 128) {
1025 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
1026 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
1027 	}
1028 
1029 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
1030 	if (timer == 0)
1031 		timer = 1;
1032 
1033 	nsrc = ntohs(igmpv3->igmp_numsrc);
1034 
1035 	/*
1036 	 * Validate address fields and versions upfront before
1037 	 * accepting v3 query.
1038 	 * XXX SMPng: Unlocked access to igmpstat counters here.
1039 	 */
1040 	if (in_nullhost(igmpv3->igmp_group)) {
1041 		/*
1042 		 * IGMPv3 General Query.
1043 		 *
1044 		 * General Queries SHOULD be directed to 224.0.0.1.
1045 		 * A general query with a source list has undefined
1046 		 * behaviour; discard it.
1047 		 */
1048 		IGMPSTAT_INC(igps_rcv_gen_queries);
1049 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1050 			IGMPSTAT_INC(igps_rcv_badqueries);
1051 			return (0);
1052 		}
1053 		is_general_query = 1;
1054 	} else {
1055 		/* Group or group-source specific query. */
1056 		if (nsrc == 0)
1057 			IGMPSTAT_INC(igps_rcv_group_queries);
1058 		else
1059 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1060 	}
1061 
1062 	IN_MULTI_LIST_LOCK();
1063 	IGMP_LOCK();
1064 
1065 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1066 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1067 
1068 	if (igi->igi_flags & IGIF_LOOPBACK) {
1069 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1070 		    ifp, ifp->if_xname);
1071 		goto out_locked;
1072 	}
1073 
1074 	/*
1075 	 * Discard the v3 query if we're in Compatibility Mode.
1076 	 * The RFC is not obviously worded that hosts need to stay in
1077 	 * compatibility mode until the Old Version Querier Present
1078 	 * timer expires.
1079 	 */
1080 	if (igi->igi_version != IGMP_VERSION_3) {
1081 		CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1082 		    igi->igi_version, ifp, ifp->if_xname);
1083 		goto out_locked;
1084 	}
1085 
1086 	igmp_set_version(igi, IGMP_VERSION_3);
1087 	igi->igi_rv = qrv;
1088 	igi->igi_qi = qqi;
1089 	igi->igi_qri = maxresp;
1090 
1091 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1092 	    maxresp);
1093 
1094 	if (is_general_query) {
1095 		/*
1096 		 * Schedule a current-state report on this ifp for
1097 		 * all groups, possibly containing source lists.
1098 		 * If there is a pending General Query response
1099 		 * scheduled earlier than the selected delay, do
1100 		 * not schedule any other reports.
1101 		 * Otherwise, reset the interface timer.
1102 		 */
1103 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1104 		    ifp, ifp->if_xname);
1105 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1106 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1107 			V_interface_timers_running = 1;
1108 		}
1109 	} else {
1110 		/*
1111 		 * Group-source-specific queries are throttled on
1112 		 * a per-group basis to defeat denial-of-service attempts.
1113 		 * Queries for groups we are not a member of on this
1114 		 * link are simply ignored.
1115 		 */
1116 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1117 		if (inm == NULL)
1118 			goto out_locked;
1119 		if (nsrc > 0) {
1120 			if (!ratecheck(&inm->inm_lastgsrtv,
1121 			    &V_igmp_gsrdelay)) {
1122 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1123 				    __func__);
1124 				IGMPSTAT_INC(igps_drop_gsr_queries);
1125 				goto out_locked;
1126 			}
1127 		}
1128 		CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1129 		     ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1130 		/*
1131 		 * If there is a pending General Query response
1132 		 * scheduled sooner than the selected delay, no
1133 		 * further report need be scheduled.
1134 		 * Otherwise, prepare to respond to the
1135 		 * group-specific or group-and-source query.
1136 		 */
1137 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1138 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1139 	}
1140 
1141 out_locked:
1142 	IGMP_UNLOCK();
1143 	IN_MULTI_LIST_UNLOCK();
1144 
1145 	return (0);
1146 }
1147 
1148 /*
1149  * Process a received IGMPv3 group-specific or group-and-source-specific
1150  * query.
1151  * Return <0 if any error occurred. Currently this is ignored.
1152  */
1153 static int
1154 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1155     int timer, /*const*/ struct igmpv3 *igmpv3)
1156 {
1157 	int			 retval;
1158 	uint16_t		 nsrc;
1159 
1160 	IN_MULTI_LIST_LOCK_ASSERT();
1161 	IGMP_LOCK_ASSERT();
1162 
1163 	retval = 0;
1164 
1165 	switch (inm->inm_state) {
1166 	case IGMP_NOT_MEMBER:
1167 	case IGMP_SILENT_MEMBER:
1168 	case IGMP_SLEEPING_MEMBER:
1169 	case IGMP_LAZY_MEMBER:
1170 	case IGMP_AWAKENING_MEMBER:
1171 	case IGMP_IDLE_MEMBER:
1172 	case IGMP_LEAVING_MEMBER:
1173 		return (retval);
1174 		break;
1175 	case IGMP_REPORTING_MEMBER:
1176 	case IGMP_G_QUERY_PENDING_MEMBER:
1177 	case IGMP_SG_QUERY_PENDING_MEMBER:
1178 		break;
1179 	}
1180 
1181 	nsrc = ntohs(igmpv3->igmp_numsrc);
1182 
1183 	/*
1184 	 * Deal with group-specific queries upfront.
1185 	 * If any group query is already pending, purge any recorded
1186 	 * source-list state if it exists, and schedule a query response
1187 	 * for this group-specific query.
1188 	 */
1189 	if (nsrc == 0) {
1190 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1191 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1192 			inm_clear_recorded(inm);
1193 			timer = min(inm->inm_timer, timer);
1194 		}
1195 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1196 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1197 		V_current_state_timers_running = 1;
1198 		return (retval);
1199 	}
1200 
1201 	/*
1202 	 * Deal with the case where a group-and-source-specific query has
1203 	 * been received but a group-specific query is already pending.
1204 	 */
1205 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1206 		timer = min(inm->inm_timer, timer);
1207 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1208 		V_current_state_timers_running = 1;
1209 		return (retval);
1210 	}
1211 
1212 	/*
1213 	 * Finally, deal with the case where a group-and-source-specific
1214 	 * query has been received, where a response to a previous g-s-r
1215 	 * query exists, or none exists.
1216 	 * In this case, we need to parse the source-list which the Querier
1217 	 * has provided us with and check if we have any source list filter
1218 	 * entries at T1 for these sources. If we do not, there is no need
1219 	 * schedule a report and the query may be dropped.
1220 	 * If we do, we must record them and schedule a current-state
1221 	 * report for those sources.
1222 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1223 	 * we pass the mbuf chain pointer down to this function, and use
1224 	 * m_getptr() to walk the chain.
1225 	 */
1226 	if (inm->inm_nsrc > 0) {
1227 		const struct in_addr	*ap;
1228 		int			 i, nrecorded;
1229 
1230 		ap = (const struct in_addr *)(igmpv3 + 1);
1231 		nrecorded = 0;
1232 		for (i = 0; i < nsrc; i++, ap++) {
1233 			retval = inm_record_source(inm, ap->s_addr);
1234 			if (retval < 0)
1235 				break;
1236 			nrecorded += retval;
1237 		}
1238 		if (nrecorded > 0) {
1239 			CTR1(KTR_IGMPV3,
1240 			    "%s: schedule response to SG query", __func__);
1241 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1242 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1243 			V_current_state_timers_running = 1;
1244 		}
1245 	}
1246 
1247 	return (retval);
1248 }
1249 
1250 /*
1251  * Process a received IGMPv1 host membership report.
1252  *
1253  * NOTE: 0.0.0.0 workaround breaks const correctness.
1254  */
1255 static int
1256 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1257     /*const*/ struct igmp *igmp)
1258 {
1259 	struct in_ifaddr *ia;
1260 	struct in_multi *inm;
1261 
1262 	IGMPSTAT_INC(igps_rcv_reports);
1263 
1264 	if (ifp->if_flags & IFF_LOOPBACK)
1265 		return (0);
1266 
1267 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1268 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1269 		IGMPSTAT_INC(igps_rcv_badreports);
1270 		return (EINVAL);
1271 	}
1272 
1273 	/*
1274 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1275 	 * Booting clients may use the source address 0.0.0.0. Some
1276 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1277 	 * the interface upon which this message was received.
1278 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1279 	 */
1280 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1281 		IFP_TO_IA(ifp, ia);
1282 		if (ia != NULL)
1283 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1284 	}
1285 
1286 	CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1287 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1288 
1289 	/*
1290 	 * IGMPv1 report suppression.
1291 	 * If we are a member of this group, and our membership should be
1292 	 * reported, stop our group timer and transition to the 'lazy' state.
1293 	 */
1294 	IN_MULTI_LIST_LOCK();
1295 	inm = inm_lookup(ifp, igmp->igmp_group);
1296 	if (inm != NULL) {
1297 		struct igmp_ifsoftc *igi;
1298 
1299 		igi = inm->inm_igi;
1300 		if (igi == NULL) {
1301 			KASSERT(igi != NULL,
1302 			    ("%s: no igi for ifp %p", __func__, ifp));
1303 			goto out_locked;
1304 		}
1305 
1306 		IGMPSTAT_INC(igps_rcv_ourreports);
1307 
1308 		/*
1309 		 * If we are in IGMPv3 host mode, do not allow the
1310 		 * other host's IGMPv1 report to suppress our reports
1311 		 * unless explicitly configured to do so.
1312 		 */
1313 		if (igi->igi_version == IGMP_VERSION_3) {
1314 			if (V_igmp_legacysupp)
1315 				igmp_v3_suppress_group_record(inm);
1316 			goto out_locked;
1317 		}
1318 
1319 		inm->inm_timer = 0;
1320 
1321 		switch (inm->inm_state) {
1322 		case IGMP_NOT_MEMBER:
1323 		case IGMP_SILENT_MEMBER:
1324 			break;
1325 		case IGMP_IDLE_MEMBER:
1326 		case IGMP_LAZY_MEMBER:
1327 		case IGMP_AWAKENING_MEMBER:
1328 			CTR3(KTR_IGMPV3,
1329 			    "report suppressed for 0x%08x on ifp %p(%s)",
1330 			    ntohl(igmp->igmp_group.s_addr), ifp,
1331 			    ifp->if_xname);
1332 		case IGMP_SLEEPING_MEMBER:
1333 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1334 			break;
1335 		case IGMP_REPORTING_MEMBER:
1336 			CTR3(KTR_IGMPV3,
1337 			    "report suppressed for 0x%08x on ifp %p(%s)",
1338 			    ntohl(igmp->igmp_group.s_addr), ifp,
1339 			    ifp->if_xname);
1340 			if (igi->igi_version == IGMP_VERSION_1)
1341 				inm->inm_state = IGMP_LAZY_MEMBER;
1342 			else if (igi->igi_version == IGMP_VERSION_2)
1343 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1344 			break;
1345 		case IGMP_G_QUERY_PENDING_MEMBER:
1346 		case IGMP_SG_QUERY_PENDING_MEMBER:
1347 		case IGMP_LEAVING_MEMBER:
1348 			break;
1349 		}
1350 	}
1351 
1352 out_locked:
1353 	IN_MULTI_LIST_UNLOCK();
1354 
1355 	return (0);
1356 }
1357 
1358 /*
1359  * Process a received IGMPv2 host membership report.
1360  *
1361  * NOTE: 0.0.0.0 workaround breaks const correctness.
1362  */
1363 static int
1364 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1365     /*const*/ struct igmp *igmp)
1366 {
1367 	struct in_ifaddr *ia;
1368 	struct in_multi *inm;
1369 
1370 	/*
1371 	 * Make sure we don't hear our own membership report.  Fast
1372 	 * leave requires knowing that we are the only member of a
1373 	 * group.
1374 	 */
1375 	IFP_TO_IA(ifp, ia);
1376 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1377 		return (0);
1378 	}
1379 
1380 	IGMPSTAT_INC(igps_rcv_reports);
1381 
1382 	if (ifp->if_flags & IFF_LOOPBACK) {
1383 		return (0);
1384 	}
1385 
1386 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1387 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1388 		IGMPSTAT_INC(igps_rcv_badreports);
1389 		return (EINVAL);
1390 	}
1391 
1392 	/*
1393 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1394 	 * Booting clients may use the source address 0.0.0.0. Some
1395 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1396 	 * the interface upon which this message was received.
1397 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1398 	 */
1399 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1400 		if (ia != NULL)
1401 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1402 	}
1403 
1404 	CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1405 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1406 
1407 	/*
1408 	 * IGMPv2 report suppression.
1409 	 * If we are a member of this group, and our membership should be
1410 	 * reported, and our group timer is pending or about to be reset,
1411 	 * stop our group timer by transitioning to the 'lazy' state.
1412 	 */
1413 	IN_MULTI_LIST_LOCK();
1414 	inm = inm_lookup(ifp, igmp->igmp_group);
1415 	if (inm != NULL) {
1416 		struct igmp_ifsoftc *igi;
1417 
1418 		igi = inm->inm_igi;
1419 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1420 
1421 		IGMPSTAT_INC(igps_rcv_ourreports);
1422 
1423 		/*
1424 		 * If we are in IGMPv3 host mode, do not allow the
1425 		 * other host's IGMPv1 report to suppress our reports
1426 		 * unless explicitly configured to do so.
1427 		 */
1428 		if (igi->igi_version == IGMP_VERSION_3) {
1429 			if (V_igmp_legacysupp)
1430 				igmp_v3_suppress_group_record(inm);
1431 			goto out_locked;
1432 		}
1433 
1434 		inm->inm_timer = 0;
1435 
1436 		switch (inm->inm_state) {
1437 		case IGMP_NOT_MEMBER:
1438 		case IGMP_SILENT_MEMBER:
1439 		case IGMP_SLEEPING_MEMBER:
1440 			break;
1441 		case IGMP_REPORTING_MEMBER:
1442 		case IGMP_IDLE_MEMBER:
1443 		case IGMP_AWAKENING_MEMBER:
1444 			CTR3(KTR_IGMPV3,
1445 			    "report suppressed for 0x%08x on ifp %p(%s)",
1446 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1447 		case IGMP_LAZY_MEMBER:
1448 			inm->inm_state = IGMP_LAZY_MEMBER;
1449 			break;
1450 		case IGMP_G_QUERY_PENDING_MEMBER:
1451 		case IGMP_SG_QUERY_PENDING_MEMBER:
1452 		case IGMP_LEAVING_MEMBER:
1453 			break;
1454 		}
1455 	}
1456 
1457 out_locked:
1458 	IN_MULTI_LIST_UNLOCK();
1459 
1460 	return (0);
1461 }
1462 
1463 int
1464 igmp_input(struct mbuf **mp, int *offp, int proto)
1465 {
1466 	int iphlen;
1467 	struct ifnet *ifp;
1468 	struct igmp *igmp;
1469 	struct ip *ip;
1470 	struct mbuf *m;
1471 	int igmplen;
1472 	int minlen;
1473 	int queryver;
1474 
1475 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1476 
1477 	m = *mp;
1478 	ifp = m->m_pkthdr.rcvif;
1479 	*mp = NULL;
1480 
1481 	IGMPSTAT_INC(igps_rcv_total);
1482 
1483 	ip = mtod(m, struct ip *);
1484 	iphlen = *offp;
1485 	igmplen = ntohs(ip->ip_len) - iphlen;
1486 
1487 	/*
1488 	 * Validate lengths.
1489 	 */
1490 	if (igmplen < IGMP_MINLEN) {
1491 		IGMPSTAT_INC(igps_rcv_tooshort);
1492 		m_freem(m);
1493 		return (IPPROTO_DONE);
1494 	}
1495 
1496 	/*
1497 	 * Always pullup to the minimum size for v1/v2 or v3
1498 	 * to amortize calls to m_pullup().
1499 	 */
1500 	minlen = iphlen;
1501 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1502 		minlen += IGMP_V3_QUERY_MINLEN;
1503 	else
1504 		minlen += IGMP_MINLEN;
1505 	if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1506 	    (m = m_pullup(m, minlen)) == NULL) {
1507 		IGMPSTAT_INC(igps_rcv_tooshort);
1508 		return (IPPROTO_DONE);
1509 	}
1510 	ip = mtod(m, struct ip *);
1511 
1512 	/*
1513 	 * Validate checksum.
1514 	 */
1515 	m->m_data += iphlen;
1516 	m->m_len -= iphlen;
1517 	igmp = mtod(m, struct igmp *);
1518 	if (in_cksum(m, igmplen)) {
1519 		IGMPSTAT_INC(igps_rcv_badsum);
1520 		m_freem(m);
1521 		return (IPPROTO_DONE);
1522 	}
1523 	m->m_data -= iphlen;
1524 	m->m_len += iphlen;
1525 
1526 	/*
1527 	 * IGMP control traffic is link-scope, and must have a TTL of 1.
1528 	 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1529 	 * probe packets may come from beyond the LAN.
1530 	 */
1531 	if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1532 		IGMPSTAT_INC(igps_rcv_badttl);
1533 		m_freem(m);
1534 		return (IPPROTO_DONE);
1535 	}
1536 
1537 	switch (igmp->igmp_type) {
1538 	case IGMP_HOST_MEMBERSHIP_QUERY:
1539 		if (igmplen == IGMP_MINLEN) {
1540 			if (igmp->igmp_code == 0)
1541 				queryver = IGMP_VERSION_1;
1542 			else
1543 				queryver = IGMP_VERSION_2;
1544 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1545 			queryver = IGMP_VERSION_3;
1546 		} else {
1547 			IGMPSTAT_INC(igps_rcv_tooshort);
1548 			m_freem(m);
1549 			return (IPPROTO_DONE);
1550 		}
1551 
1552 		switch (queryver) {
1553 		case IGMP_VERSION_1:
1554 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1555 			if (!V_igmp_v1enable)
1556 				break;
1557 			if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1558 				m_freem(m);
1559 				return (IPPROTO_DONE);
1560 			}
1561 			break;
1562 
1563 		case IGMP_VERSION_2:
1564 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1565 			if (!V_igmp_v2enable)
1566 				break;
1567 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1568 				m_freem(m);
1569 				return (IPPROTO_DONE);
1570 			}
1571 			break;
1572 
1573 		case IGMP_VERSION_3: {
1574 				struct igmpv3 *igmpv3;
1575 				uint16_t igmpv3len;
1576 				uint16_t nsrc;
1577 
1578 				IGMPSTAT_INC(igps_rcv_v3_queries);
1579 				igmpv3 = (struct igmpv3 *)igmp;
1580 				/*
1581 				 * Validate length based on source count.
1582 				 */
1583 				nsrc = ntohs(igmpv3->igmp_numsrc);
1584 				if (nsrc * sizeof(in_addr_t) >
1585 				    UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1586 					IGMPSTAT_INC(igps_rcv_tooshort);
1587 					m_freem(m);
1588 					return (IPPROTO_DONE);
1589 				}
1590 				/*
1591 				 * m_pullup() may modify m, so pullup in
1592 				 * this scope.
1593 				 */
1594 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1595 				   sizeof(struct in_addr) * nsrc;
1596 				if ((!M_WRITABLE(m) ||
1597 				     m->m_len < igmpv3len) &&
1598 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1599 					IGMPSTAT_INC(igps_rcv_tooshort);
1600 					return (IPPROTO_DONE);
1601 				}
1602 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1603 				    + iphlen);
1604 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1605 					m_freem(m);
1606 					return (IPPROTO_DONE);
1607 				}
1608 			}
1609 			break;
1610 		}
1611 		break;
1612 
1613 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1614 		if (!V_igmp_v1enable)
1615 			break;
1616 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1617 			m_freem(m);
1618 			return (IPPROTO_DONE);
1619 		}
1620 		break;
1621 
1622 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1623 		if (!V_igmp_v2enable)
1624 			break;
1625 		if (!ip_checkrouteralert(m))
1626 			IGMPSTAT_INC(igps_rcv_nora);
1627 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1628 			m_freem(m);
1629 			return (IPPROTO_DONE);
1630 		}
1631 		break;
1632 
1633 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1634 		/*
1635 		 * Hosts do not need to process IGMPv3 membership reports,
1636 		 * as report suppression is no longer required.
1637 		 */
1638 		if (!ip_checkrouteralert(m))
1639 			IGMPSTAT_INC(igps_rcv_nora);
1640 		break;
1641 
1642 	default:
1643 		break;
1644 	}
1645 
1646 	/*
1647 	 * Pass all valid IGMP packets up to any process(es) listening on a
1648 	 * raw IGMP socket.
1649 	 */
1650 	*mp = m;
1651 	return (rip_input(mp, offp, proto));
1652 }
1653 
1654 /*
1655  * Fast timeout handler (global).
1656  * VIMAGE: Timeout handlers are expected to service all vimages.
1657  */
1658 void
1659 igmp_fasttimo(void)
1660 {
1661 	VNET_ITERATOR_DECL(vnet_iter);
1662 
1663 	VNET_LIST_RLOCK_NOSLEEP();
1664 	VNET_FOREACH(vnet_iter) {
1665 		CURVNET_SET(vnet_iter);
1666 		igmp_fasttimo_vnet();
1667 		CURVNET_RESTORE();
1668 	}
1669 	VNET_LIST_RUNLOCK_NOSLEEP();
1670 }
1671 
1672 /*
1673  * Fast timeout handler (per-vnet).
1674  * Sends are shuffled off to a netisr to deal with Giant.
1675  *
1676  * VIMAGE: Assume caller has set up our curvnet.
1677  */
1678 static void
1679 igmp_fasttimo_vnet(void)
1680 {
1681 	struct mbufq		 scq;	/* State-change packets */
1682 	struct mbufq		 qrq;	/* Query response packets */
1683 	struct ifnet		*ifp;
1684 	struct igmp_ifsoftc	*igi;
1685 	struct ifmultiaddr	*ifma, *next;
1686 	struct in_multi		*inm;
1687 	struct in_multi_head inm_free_tmp;
1688 	int			 loop, uri_fasthz;
1689 
1690 	loop = 0;
1691 	uri_fasthz = 0;
1692 
1693 	/*
1694 	 * Quick check to see if any work needs to be done, in order to
1695 	 * minimize the overhead of fasttimo processing.
1696 	 * SMPng: XXX Unlocked reads.
1697 	 */
1698 	if (!V_current_state_timers_running &&
1699 	    !V_interface_timers_running &&
1700 	    !V_state_change_timers_running)
1701 		return;
1702 
1703 	SLIST_INIT(&inm_free_tmp);
1704 	IN_MULTI_LIST_LOCK();
1705 	IGMP_LOCK();
1706 
1707 	/*
1708 	 * IGMPv3 General Query response timer processing.
1709 	 */
1710 	if (V_interface_timers_running) {
1711 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1712 
1713 		V_interface_timers_running = 0;
1714 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1715 			if (igi->igi_v3_timer == 0) {
1716 				/* Do nothing. */
1717 			} else if (--igi->igi_v3_timer == 0) {
1718 				igmp_v3_dispatch_general_query(igi);
1719 			} else {
1720 				V_interface_timers_running = 1;
1721 			}
1722 		}
1723 	}
1724 
1725 	if (!V_current_state_timers_running &&
1726 	    !V_state_change_timers_running)
1727 		goto out_locked;
1728 
1729 	V_current_state_timers_running = 0;
1730 	V_state_change_timers_running = 0;
1731 
1732 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1733 
1734 	/*
1735 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1736 	 * Note: Processing a v3 group timer may remove a node.
1737 	 */
1738 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1739 		ifp = igi->igi_ifp;
1740 
1741 		if (igi->igi_version == IGMP_VERSION_3) {
1742 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1743 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1744 			    PR_FASTHZ);
1745 			mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1746 			mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1747 		}
1748 
1749 		IF_ADDR_WLOCK(ifp);
1750 	restart:
1751 		CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1752 			if (ifma->ifma_addr->sa_family != AF_INET ||
1753 			    ifma->ifma_protospec == NULL)
1754 				continue;
1755 			inm = (struct in_multi *)ifma->ifma_protospec;
1756 			switch (igi->igi_version) {
1757 			case IGMP_VERSION_1:
1758 			case IGMP_VERSION_2:
1759 				igmp_v1v2_process_group_timer(inm,
1760 				    igi->igi_version);
1761 				break;
1762 			case IGMP_VERSION_3:
1763 				igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1764 				    &scq, inm, uri_fasthz);
1765 				break;
1766 			}
1767 			if (__predict_false(ifma_restart)) {
1768 				ifma_restart = false;
1769 				goto restart;
1770 			}
1771 		}
1772 		IF_ADDR_WUNLOCK(ifp);
1773 
1774 		if (igi->igi_version == IGMP_VERSION_3) {
1775 			igmp_dispatch_queue(&qrq, 0, loop);
1776 			igmp_dispatch_queue(&scq, 0, loop);
1777 
1778 			/*
1779 			 * Free the in_multi reference(s) for this
1780 			 * IGMP lifecycle.
1781 			 */
1782 			inm_release_list_deferred(&inm_free_tmp);
1783 		}
1784 	}
1785 
1786 out_locked:
1787 	IGMP_UNLOCK();
1788 	IN_MULTI_LIST_UNLOCK();
1789 }
1790 
1791 /*
1792  * Update host report group timer for IGMPv1/v2.
1793  * Will update the global pending timer flags.
1794  */
1795 static void
1796 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1797 {
1798 	int report_timer_expired;
1799 
1800 	IN_MULTI_LIST_LOCK_ASSERT();
1801 	IGMP_LOCK_ASSERT();
1802 
1803 	if (inm->inm_timer == 0) {
1804 		report_timer_expired = 0;
1805 	} else if (--inm->inm_timer == 0) {
1806 		report_timer_expired = 1;
1807 	} else {
1808 		V_current_state_timers_running = 1;
1809 		return;
1810 	}
1811 
1812 	switch (inm->inm_state) {
1813 	case IGMP_NOT_MEMBER:
1814 	case IGMP_SILENT_MEMBER:
1815 	case IGMP_IDLE_MEMBER:
1816 	case IGMP_LAZY_MEMBER:
1817 	case IGMP_SLEEPING_MEMBER:
1818 	case IGMP_AWAKENING_MEMBER:
1819 		break;
1820 	case IGMP_REPORTING_MEMBER:
1821 		if (report_timer_expired) {
1822 			inm->inm_state = IGMP_IDLE_MEMBER;
1823 			(void)igmp_v1v2_queue_report(inm,
1824 			    (version == IGMP_VERSION_2) ?
1825 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1826 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1827 		}
1828 		break;
1829 	case IGMP_G_QUERY_PENDING_MEMBER:
1830 	case IGMP_SG_QUERY_PENDING_MEMBER:
1831 	case IGMP_LEAVING_MEMBER:
1832 		break;
1833 	}
1834 }
1835 
1836 /*
1837  * Update a group's timers for IGMPv3.
1838  * Will update the global pending timer flags.
1839  * Note: Unlocked read from igi.
1840  */
1841 static void
1842 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1843     struct mbufq *qrq, struct mbufq *scq,
1844     struct in_multi *inm, const int uri_fasthz)
1845 {
1846 	int query_response_timer_expired;
1847 	int state_change_retransmit_timer_expired;
1848 
1849 	IN_MULTI_LIST_LOCK_ASSERT();
1850 	IGMP_LOCK_ASSERT();
1851 
1852 	query_response_timer_expired = 0;
1853 	state_change_retransmit_timer_expired = 0;
1854 
1855 	/*
1856 	 * During a transition from v1/v2 compatibility mode back to v3,
1857 	 * a group record in REPORTING state may still have its group
1858 	 * timer active. This is a no-op in this function; it is easier
1859 	 * to deal with it here than to complicate the slow-timeout path.
1860 	 */
1861 	if (inm->inm_timer == 0) {
1862 		query_response_timer_expired = 0;
1863 	} else if (--inm->inm_timer == 0) {
1864 		query_response_timer_expired = 1;
1865 	} else {
1866 		V_current_state_timers_running = 1;
1867 	}
1868 
1869 	if (inm->inm_sctimer == 0) {
1870 		state_change_retransmit_timer_expired = 0;
1871 	} else if (--inm->inm_sctimer == 0) {
1872 		state_change_retransmit_timer_expired = 1;
1873 	} else {
1874 		V_state_change_timers_running = 1;
1875 	}
1876 
1877 	/* We are in fasttimo, so be quick about it. */
1878 	if (!state_change_retransmit_timer_expired &&
1879 	    !query_response_timer_expired)
1880 		return;
1881 
1882 	switch (inm->inm_state) {
1883 	case IGMP_NOT_MEMBER:
1884 	case IGMP_SILENT_MEMBER:
1885 	case IGMP_SLEEPING_MEMBER:
1886 	case IGMP_LAZY_MEMBER:
1887 	case IGMP_AWAKENING_MEMBER:
1888 	case IGMP_IDLE_MEMBER:
1889 		break;
1890 	case IGMP_G_QUERY_PENDING_MEMBER:
1891 	case IGMP_SG_QUERY_PENDING_MEMBER:
1892 		/*
1893 		 * Respond to a previously pending Group-Specific
1894 		 * or Group-and-Source-Specific query by enqueueing
1895 		 * the appropriate Current-State report for
1896 		 * immediate transmission.
1897 		 */
1898 		if (query_response_timer_expired) {
1899 			int retval __unused;
1900 
1901 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1902 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1903 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1904 			    __func__, retval);
1905 			inm->inm_state = IGMP_REPORTING_MEMBER;
1906 			/* XXX Clear recorded sources for next time. */
1907 			inm_clear_recorded(inm);
1908 		}
1909 		/* FALLTHROUGH */
1910 	case IGMP_REPORTING_MEMBER:
1911 	case IGMP_LEAVING_MEMBER:
1912 		if (state_change_retransmit_timer_expired) {
1913 			/*
1914 			 * State-change retransmission timer fired.
1915 			 * If there are any further pending retransmissions,
1916 			 * set the global pending state-change flag, and
1917 			 * reset the timer.
1918 			 */
1919 			if (--inm->inm_scrv > 0) {
1920 				inm->inm_sctimer = uri_fasthz;
1921 				V_state_change_timers_running = 1;
1922 			}
1923 			/*
1924 			 * Retransmit the previously computed state-change
1925 			 * report. If there are no further pending
1926 			 * retransmissions, the mbuf queue will be consumed.
1927 			 * Update T0 state to T1 as we have now sent
1928 			 * a state-change.
1929 			 */
1930 			(void)igmp_v3_merge_state_changes(inm, scq);
1931 
1932 			inm_commit(inm);
1933 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1934 			    ntohl(inm->inm_addr.s_addr),
1935 			    inm->inm_ifp->if_xname);
1936 
1937 			/*
1938 			 * If we are leaving the group for good, make sure
1939 			 * we release IGMP's reference to it.
1940 			 * This release must be deferred using a SLIST,
1941 			 * as we are called from a loop which traverses
1942 			 * the in_ifmultiaddr TAILQ.
1943 			 */
1944 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1945 			    inm->inm_scrv == 0) {
1946 				inm->inm_state = IGMP_NOT_MEMBER;
1947 				inm_rele_locked(inmh, inm);
1948 			}
1949 		}
1950 		break;
1951 	}
1952 }
1953 
1954 /*
1955  * Suppress a group's pending response to a group or source/group query.
1956  *
1957  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1958  * Do NOT update ST1/ST0 as this operation merely suppresses
1959  * the currently pending group record.
1960  * Do NOT suppress the response to a general query. It is possible but
1961  * it would require adding another state or flag.
1962  */
1963 static void
1964 igmp_v3_suppress_group_record(struct in_multi *inm)
1965 {
1966 
1967 	IN_MULTI_LIST_LOCK_ASSERT();
1968 
1969 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1970 		("%s: not IGMPv3 mode on link", __func__));
1971 
1972 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1973 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1974 		return;
1975 
1976 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1977 		inm_clear_recorded(inm);
1978 
1979 	inm->inm_timer = 0;
1980 	inm->inm_state = IGMP_REPORTING_MEMBER;
1981 }
1982 
1983 /*
1984  * Switch to a different IGMP version on the given interface,
1985  * as per Section 7.2.1.
1986  */
1987 static void
1988 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1989 {
1990 	int old_version_timer;
1991 
1992 	IGMP_LOCK_ASSERT();
1993 
1994 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1995 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1996 
1997 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1998 		/*
1999 		 * Compute the "Older Version Querier Present" timer as per
2000 		 * Section 8.12.
2001 		 */
2002 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
2003 		old_version_timer *= PR_SLOWHZ;
2004 
2005 		if (version == IGMP_VERSION_1) {
2006 			igi->igi_v1_timer = old_version_timer;
2007 			igi->igi_v2_timer = 0;
2008 		} else if (version == IGMP_VERSION_2) {
2009 			igi->igi_v1_timer = 0;
2010 			igi->igi_v2_timer = old_version_timer;
2011 		}
2012 	}
2013 
2014 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2015 		if (igi->igi_version != IGMP_VERSION_2) {
2016 			igi->igi_version = IGMP_VERSION_2;
2017 			igmp_v3_cancel_link_timers(igi);
2018 		}
2019 	} else if (igi->igi_v1_timer > 0) {
2020 		if (igi->igi_version != IGMP_VERSION_1) {
2021 			igi->igi_version = IGMP_VERSION_1;
2022 			igmp_v3_cancel_link_timers(igi);
2023 		}
2024 	}
2025 }
2026 
2027 /*
2028  * Cancel pending IGMPv3 timers for the given link and all groups
2029  * joined on it; state-change, general-query, and group-query timers.
2030  *
2031  * Only ever called on a transition from v3 to Compatibility mode. Kill
2032  * the timers stone dead (this may be expensive for large N groups), they
2033  * will be restarted if Compatibility Mode deems that they must be due to
2034  * query processing.
2035  */
2036 static void
2037 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2038 {
2039 	struct ifmultiaddr	*ifma, *ifmatmp;
2040 	struct ifnet		*ifp;
2041 	struct in_multi		*inm;
2042 	struct in_multi_head inm_free_tmp;
2043 
2044 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2045 	    igi->igi_ifp, igi->igi_ifp->if_xname);
2046 
2047 	IN_MULTI_LIST_LOCK_ASSERT();
2048 	IGMP_LOCK_ASSERT();
2049 	NET_EPOCH_ASSERT();
2050 
2051 	SLIST_INIT(&inm_free_tmp);
2052 
2053 	/*
2054 	 * Stop the v3 General Query Response on this link stone dead.
2055 	 * If fasttimo is woken up due to V_interface_timers_running,
2056 	 * the flag will be cleared if there are no pending link timers.
2057 	 */
2058 	igi->igi_v3_timer = 0;
2059 
2060 	/*
2061 	 * Now clear the current-state and state-change report timers
2062 	 * for all memberships scoped to this link.
2063 	 */
2064 	ifp = igi->igi_ifp;
2065 	IF_ADDR_WLOCK(ifp);
2066 	CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, ifmatmp) {
2067 		if (ifma->ifma_addr->sa_family != AF_INET ||
2068 		    ifma->ifma_protospec == NULL)
2069 			continue;
2070 		inm = (struct in_multi *)ifma->ifma_protospec;
2071 		switch (inm->inm_state) {
2072 		case IGMP_NOT_MEMBER:
2073 		case IGMP_SILENT_MEMBER:
2074 		case IGMP_IDLE_MEMBER:
2075 		case IGMP_LAZY_MEMBER:
2076 		case IGMP_SLEEPING_MEMBER:
2077 		case IGMP_AWAKENING_MEMBER:
2078 			/*
2079 			 * These states are either not relevant in v3 mode,
2080 			 * or are unreported. Do nothing.
2081 			 */
2082 			break;
2083 		case IGMP_LEAVING_MEMBER:
2084 			/*
2085 			 * If we are leaving the group and switching to
2086 			 * compatibility mode, we need to release the final
2087 			 * reference held for issuing the INCLUDE {}, and
2088 			 * transition to REPORTING to ensure the host leave
2089 			 * message is sent upstream to the old querier --
2090 			 * transition to NOT would lose the leave and race.
2091 			 */
2092 			inm_rele_locked(&inm_free_tmp, inm);
2093 			/* FALLTHROUGH */
2094 		case IGMP_G_QUERY_PENDING_MEMBER:
2095 		case IGMP_SG_QUERY_PENDING_MEMBER:
2096 			inm_clear_recorded(inm);
2097 			/* FALLTHROUGH */
2098 		case IGMP_REPORTING_MEMBER:
2099 			inm->inm_state = IGMP_REPORTING_MEMBER;
2100 			break;
2101 		}
2102 		/*
2103 		 * Always clear state-change and group report timers.
2104 		 * Free any pending IGMPv3 state-change records.
2105 		 */
2106 		inm->inm_sctimer = 0;
2107 		inm->inm_timer = 0;
2108 		mbufq_drain(&inm->inm_scq);
2109 	}
2110 	IF_ADDR_WUNLOCK(ifp);
2111 
2112 	inm_release_list_deferred(&inm_free_tmp);
2113 }
2114 
2115 /*
2116  * Update the Older Version Querier Present timers for a link.
2117  * See Section 7.2.1 of RFC 3376.
2118  */
2119 static void
2120 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2121 {
2122 
2123 	IGMP_LOCK_ASSERT();
2124 
2125 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2126 		/*
2127 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2128 		 *
2129 		 * Revert to IGMPv3.
2130 		 */
2131 		if (igi->igi_version != IGMP_VERSION_3) {
2132 			CTR5(KTR_IGMPV3,
2133 			    "%s: transition from v%d -> v%d on %p(%s)",
2134 			    __func__, igi->igi_version, IGMP_VERSION_3,
2135 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2136 			igi->igi_version = IGMP_VERSION_3;
2137 		}
2138 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2139 		/*
2140 		 * IGMPv1 Querier Present timer expired,
2141 		 * IGMPv2 Querier Present timer running.
2142 		 * If IGMPv2 was disabled since last timeout,
2143 		 * revert to IGMPv3.
2144 		 * If IGMPv2 is enabled, revert to IGMPv2.
2145 		 */
2146 		if (!V_igmp_v2enable) {
2147 			CTR5(KTR_IGMPV3,
2148 			    "%s: transition from v%d -> v%d on %p(%s)",
2149 			    __func__, igi->igi_version, IGMP_VERSION_3,
2150 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2151 			igi->igi_v2_timer = 0;
2152 			igi->igi_version = IGMP_VERSION_3;
2153 		} else {
2154 			--igi->igi_v2_timer;
2155 			if (igi->igi_version != IGMP_VERSION_2) {
2156 				CTR5(KTR_IGMPV3,
2157 				    "%s: transition from v%d -> v%d on %p(%s)",
2158 				    __func__, igi->igi_version, IGMP_VERSION_2,
2159 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2160 				igi->igi_version = IGMP_VERSION_2;
2161 				igmp_v3_cancel_link_timers(igi);
2162 			}
2163 		}
2164 	} else if (igi->igi_v1_timer > 0) {
2165 		/*
2166 		 * IGMPv1 Querier Present timer running.
2167 		 * Stop IGMPv2 timer if running.
2168 		 *
2169 		 * If IGMPv1 was disabled since last timeout,
2170 		 * revert to IGMPv3.
2171 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2172 		 */
2173 		if (!V_igmp_v1enable) {
2174 			CTR5(KTR_IGMPV3,
2175 			    "%s: transition from v%d -> v%d on %p(%s)",
2176 			    __func__, igi->igi_version, IGMP_VERSION_3,
2177 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2178 			igi->igi_v1_timer = 0;
2179 			igi->igi_version = IGMP_VERSION_3;
2180 		} else {
2181 			--igi->igi_v1_timer;
2182 		}
2183 		if (igi->igi_v2_timer > 0) {
2184 			CTR3(KTR_IGMPV3,
2185 			    "%s: cancel v2 timer on %p(%s)",
2186 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2187 			igi->igi_v2_timer = 0;
2188 		}
2189 	}
2190 }
2191 
2192 /*
2193  * Global slowtimo handler.
2194  * VIMAGE: Timeout handlers are expected to service all vimages.
2195  */
2196 void
2197 igmp_slowtimo(void)
2198 {
2199 	VNET_ITERATOR_DECL(vnet_iter);
2200 
2201 	VNET_LIST_RLOCK_NOSLEEP();
2202 	VNET_FOREACH(vnet_iter) {
2203 		CURVNET_SET(vnet_iter);
2204 		igmp_slowtimo_vnet();
2205 		CURVNET_RESTORE();
2206 	}
2207 	VNET_LIST_RUNLOCK_NOSLEEP();
2208 }
2209 
2210 /*
2211  * Per-vnet slowtimo handler.
2212  */
2213 static void
2214 igmp_slowtimo_vnet(void)
2215 {
2216 	struct igmp_ifsoftc *igi;
2217 
2218 	IGMP_LOCK();
2219 
2220 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2221 		igmp_v1v2_process_querier_timers(igi);
2222 	}
2223 
2224 	IGMP_UNLOCK();
2225 }
2226 
2227 /*
2228  * Dispatch an IGMPv1/v2 host report or leave message.
2229  * These are always small enough to fit inside a single mbuf.
2230  */
2231 static int
2232 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2233 {
2234 	struct epoch_tracker 	et;
2235 	struct ifnet		*ifp;
2236 	struct igmp		*igmp;
2237 	struct ip		*ip;
2238 	struct mbuf		*m;
2239 
2240 	IN_MULTI_LIST_LOCK_ASSERT();
2241 	IGMP_LOCK_ASSERT();
2242 
2243 	ifp = inm->inm_ifp;
2244 
2245 	m = m_gethdr(M_NOWAIT, MT_DATA);
2246 	if (m == NULL)
2247 		return (ENOMEM);
2248 	M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2249 
2250 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2251 
2252 	m->m_data += sizeof(struct ip);
2253 	m->m_len = sizeof(struct igmp);
2254 
2255 	igmp = mtod(m, struct igmp *);
2256 	igmp->igmp_type = type;
2257 	igmp->igmp_code = 0;
2258 	igmp->igmp_group = inm->inm_addr;
2259 	igmp->igmp_cksum = 0;
2260 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2261 
2262 	m->m_data -= sizeof(struct ip);
2263 	m->m_len += sizeof(struct ip);
2264 
2265 	ip = mtod(m, struct ip *);
2266 	ip->ip_tos = 0;
2267 	ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2268 	ip->ip_off = 0;
2269 	ip->ip_p = IPPROTO_IGMP;
2270 	ip->ip_src.s_addr = INADDR_ANY;
2271 
2272 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2273 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2274 	else
2275 		ip->ip_dst = inm->inm_addr;
2276 
2277 	igmp_save_context(m, ifp);
2278 
2279 	m->m_flags |= M_IGMPV2;
2280 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2281 		m->m_flags |= M_IGMP_LOOP;
2282 
2283 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2284 	NET_EPOCH_ENTER(et);
2285 	netisr_dispatch(NETISR_IGMP, m);
2286 	NET_EPOCH_EXIT(et);
2287 
2288 	return (0);
2289 }
2290 
2291 /*
2292  * Process a state change from the upper layer for the given IPv4 group.
2293  *
2294  * Each socket holds a reference on the in_multi in its own ip_moptions.
2295  * The socket layer will have made the necessary updates to.the group
2296  * state, it is now up to IGMP to issue a state change report if there
2297  * has been any change between T0 (when the last state-change was issued)
2298  * and T1 (now).
2299  *
2300  * We use the IGMPv3 state machine at group level. The IGMP module
2301  * however makes the decision as to which IGMP protocol version to speak.
2302  * A state change *from* INCLUDE {} always means an initial join.
2303  * A state change *to* INCLUDE {} always means a final leave.
2304  *
2305  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2306  * save ourselves a bunch of work; any exclusive mode groups need not
2307  * compute source filter lists.
2308  *
2309  * VIMAGE: curvnet should have been set by caller, as this routine
2310  * is called from the socket option handlers.
2311  */
2312 int
2313 igmp_change_state(struct in_multi *inm)
2314 {
2315 	struct igmp_ifsoftc *igi;
2316 	struct ifnet *ifp;
2317 	int error;
2318 
2319 	error = 0;
2320 	IN_MULTI_LOCK_ASSERT();
2321 	/*
2322 	 * Try to detect if the upper layer just asked us to change state
2323 	 * for an interface which has now gone away.
2324 	 */
2325 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2326 	ifp = inm->inm_ifma->ifma_ifp;
2327 	/*
2328 	 * Sanity check that netinet's notion of ifp is the
2329 	 * same as net's.
2330 	 */
2331 	KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2332 
2333 	IGMP_LOCK();
2334 
2335 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2336 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2337 
2338 	/*
2339 	 * If we detect a state transition to or from MCAST_UNDEFINED
2340 	 * for this group, then we are starting or finishing an IGMP
2341 	 * life cycle for this group.
2342 	 */
2343 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2344 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2345 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2346 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2347 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2348 			error = igmp_initial_join(inm, igi);
2349 			goto out_locked;
2350 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2351 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2352 			igmp_final_leave(inm, igi);
2353 			goto out_locked;
2354 		}
2355 	} else {
2356 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2357 	}
2358 
2359 	error = igmp_handle_state_change(inm, igi);
2360 
2361 out_locked:
2362 	IGMP_UNLOCK();
2363 	return (error);
2364 }
2365 
2366 /*
2367  * Perform the initial join for an IGMP group.
2368  *
2369  * When joining a group:
2370  *  If the group should have its IGMP traffic suppressed, do nothing.
2371  *  IGMPv1 starts sending IGMPv1 host membership reports.
2372  *  IGMPv2 starts sending IGMPv2 host membership reports.
2373  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2374  *  initial state of the membership.
2375  */
2376 static int
2377 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2378 {
2379 	struct ifnet		*ifp;
2380 	struct mbufq		*mq;
2381 	int			 error, retval, syncstates;
2382 
2383 	CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2384 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2385 
2386 	error = 0;
2387 	syncstates = 1;
2388 
2389 	ifp = inm->inm_ifp;
2390 
2391 	IN_MULTI_LOCK_ASSERT();
2392 	IGMP_LOCK_ASSERT();
2393 
2394 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2395 
2396 	/*
2397 	 * Groups joined on loopback or marked as 'not reported',
2398 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2399 	 * are never reported in any IGMP protocol exchanges.
2400 	 * All other groups enter the appropriate IGMP state machine
2401 	 * for the version in use on this link.
2402 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2403 	 * disabled for the link.
2404 	 */
2405 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2406 	    (igi->igi_flags & IGIF_SILENT) ||
2407 	    !igmp_isgroupreported(inm->inm_addr)) {
2408 		CTR1(KTR_IGMPV3,
2409 "%s: not kicking state machine for silent group", __func__);
2410 		inm->inm_state = IGMP_SILENT_MEMBER;
2411 		inm->inm_timer = 0;
2412 	} else {
2413 		/*
2414 		 * Deal with overlapping in_multi lifecycle.
2415 		 * If this group was LEAVING, then make sure
2416 		 * we drop the reference we picked up to keep the
2417 		 * group around for the final INCLUDE {} enqueue.
2418 		 */
2419 		if (igi->igi_version == IGMP_VERSION_3 &&
2420 		    inm->inm_state == IGMP_LEAVING_MEMBER) {
2421 			MPASS(inm->inm_refcount > 1);
2422 			inm_rele_locked(NULL, inm);
2423 		}
2424 		inm->inm_state = IGMP_REPORTING_MEMBER;
2425 
2426 		switch (igi->igi_version) {
2427 		case IGMP_VERSION_1:
2428 		case IGMP_VERSION_2:
2429 			inm->inm_state = IGMP_IDLE_MEMBER;
2430 			error = igmp_v1v2_queue_report(inm,
2431 			    (igi->igi_version == IGMP_VERSION_2) ?
2432 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2433 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2434 			if (error == 0) {
2435 				inm->inm_timer = IGMP_RANDOM_DELAY(
2436 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2437 				V_current_state_timers_running = 1;
2438 			}
2439 			break;
2440 
2441 		case IGMP_VERSION_3:
2442 			/*
2443 			 * Defer update of T0 to T1, until the first copy
2444 			 * of the state change has been transmitted.
2445 			 */
2446 			syncstates = 0;
2447 
2448 			/*
2449 			 * Immediately enqueue a State-Change Report for
2450 			 * this interface, freeing any previous reports.
2451 			 * Don't kick the timers if there is nothing to do,
2452 			 * or if an error occurred.
2453 			 */
2454 			mq = &inm->inm_scq;
2455 			mbufq_drain(mq);
2456 			retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2457 			    0, 0);
2458 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2459 			    __func__, retval);
2460 			if (retval <= 0) {
2461 				error = retval * -1;
2462 				break;
2463 			}
2464 
2465 			/*
2466 			 * Schedule transmission of pending state-change
2467 			 * report up to RV times for this link. The timer
2468 			 * will fire at the next igmp_fasttimo (~200ms),
2469 			 * giving us an opportunity to merge the reports.
2470 			 */
2471 			if (igi->igi_flags & IGIF_LOOPBACK) {
2472 				inm->inm_scrv = 1;
2473 			} else {
2474 				KASSERT(igi->igi_rv > 1,
2475 				   ("%s: invalid robustness %d", __func__,
2476 				    igi->igi_rv));
2477 				inm->inm_scrv = igi->igi_rv;
2478 			}
2479 			inm->inm_sctimer = 1;
2480 			V_state_change_timers_running = 1;
2481 
2482 			error = 0;
2483 			break;
2484 		}
2485 	}
2486 
2487 	/*
2488 	 * Only update the T0 state if state change is atomic,
2489 	 * i.e. we don't need to wait for a timer to fire before we
2490 	 * can consider the state change to have been communicated.
2491 	 */
2492 	if (syncstates) {
2493 		inm_commit(inm);
2494 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2495 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2496 	}
2497 
2498 	return (error);
2499 }
2500 
2501 /*
2502  * Issue an intermediate state change during the IGMP life-cycle.
2503  */
2504 static int
2505 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2506 {
2507 	struct ifnet		*ifp;
2508 	int			 retval;
2509 
2510 	CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2511 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2512 
2513 	ifp = inm->inm_ifp;
2514 
2515 	IN_MULTI_LIST_LOCK_ASSERT();
2516 	IGMP_LOCK_ASSERT();
2517 
2518 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2519 
2520 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2521 	    (igi->igi_flags & IGIF_SILENT) ||
2522 	    !igmp_isgroupreported(inm->inm_addr) ||
2523 	    (igi->igi_version != IGMP_VERSION_3)) {
2524 		if (!igmp_isgroupreported(inm->inm_addr)) {
2525 			CTR1(KTR_IGMPV3,
2526 "%s: not kicking state machine for silent group", __func__);
2527 		}
2528 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2529 		inm_commit(inm);
2530 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2531 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2532 		return (0);
2533 	}
2534 
2535 	mbufq_drain(&inm->inm_scq);
2536 
2537 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2538 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2539 	if (retval <= 0)
2540 		return (-retval);
2541 
2542 	/*
2543 	 * If record(s) were enqueued, start the state-change
2544 	 * report timer for this group.
2545 	 */
2546 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2547 	inm->inm_sctimer = 1;
2548 	V_state_change_timers_running = 1;
2549 
2550 	return (0);
2551 }
2552 
2553 /*
2554  * Perform the final leave for an IGMP group.
2555  *
2556  * When leaving a group:
2557  *  IGMPv1 does nothing.
2558  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2559  *  IGMPv3 enqueues a state-change report containing a transition
2560  *  to INCLUDE {} for immediate transmission.
2561  */
2562 static void
2563 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2564 {
2565 	int syncstates;
2566 
2567 	syncstates = 1;
2568 
2569 	CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2570 	    __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2571 	    inm->inm_ifp->if_xname);
2572 
2573 	IN_MULTI_LIST_LOCK_ASSERT();
2574 	IGMP_LOCK_ASSERT();
2575 
2576 	switch (inm->inm_state) {
2577 	case IGMP_NOT_MEMBER:
2578 	case IGMP_SILENT_MEMBER:
2579 	case IGMP_LEAVING_MEMBER:
2580 		/* Already leaving or left; do nothing. */
2581 		CTR1(KTR_IGMPV3,
2582 "%s: not kicking state machine for silent group", __func__);
2583 		break;
2584 	case IGMP_REPORTING_MEMBER:
2585 	case IGMP_IDLE_MEMBER:
2586 	case IGMP_G_QUERY_PENDING_MEMBER:
2587 	case IGMP_SG_QUERY_PENDING_MEMBER:
2588 		if (igi->igi_version == IGMP_VERSION_2) {
2589 #ifdef INVARIANTS
2590 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2591 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2592 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2593 			     __func__);
2594 #endif
2595 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2596 			inm->inm_state = IGMP_NOT_MEMBER;
2597 		} else if (igi->igi_version == IGMP_VERSION_3) {
2598 			/*
2599 			 * Stop group timer and all pending reports.
2600 			 * Immediately enqueue a state-change report
2601 			 * TO_IN {} to be sent on the next fast timeout,
2602 			 * giving us an opportunity to merge reports.
2603 			 */
2604 			mbufq_drain(&inm->inm_scq);
2605 			inm->inm_timer = 0;
2606 			if (igi->igi_flags & IGIF_LOOPBACK) {
2607 				inm->inm_scrv = 1;
2608 			} else {
2609 				inm->inm_scrv = igi->igi_rv;
2610 			}
2611 			CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2612 			    "pending retransmissions.", __func__,
2613 			    ntohl(inm->inm_addr.s_addr),
2614 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2615 			if (inm->inm_scrv == 0) {
2616 				inm->inm_state = IGMP_NOT_MEMBER;
2617 				inm->inm_sctimer = 0;
2618 			} else {
2619 				int retval __unused;
2620 
2621 				inm_acquire_locked(inm);
2622 
2623 				retval = igmp_v3_enqueue_group_record(
2624 				    &inm->inm_scq, inm, 1, 0, 0);
2625 				KASSERT(retval != 0,
2626 				    ("%s: enqueue record = %d", __func__,
2627 				     retval));
2628 
2629 				inm->inm_state = IGMP_LEAVING_MEMBER;
2630 				inm->inm_sctimer = 1;
2631 				V_state_change_timers_running = 1;
2632 				syncstates = 0;
2633 			}
2634 			break;
2635 		}
2636 		break;
2637 	case IGMP_LAZY_MEMBER:
2638 	case IGMP_SLEEPING_MEMBER:
2639 	case IGMP_AWAKENING_MEMBER:
2640 		/* Our reports are suppressed; do nothing. */
2641 		break;
2642 	}
2643 
2644 	if (syncstates) {
2645 		inm_commit(inm);
2646 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2647 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2648 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2649 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2650 		    __func__, ntohl(inm->inm_addr.s_addr),
2651 		    inm->inm_ifp->if_xname);
2652 	}
2653 }
2654 
2655 /*
2656  * Enqueue an IGMPv3 group record to the given output queue.
2657  *
2658  * XXX This function could do with having the allocation code
2659  * split out, and the multiple-tree-walks coalesced into a single
2660  * routine as has been done in igmp_v3_enqueue_filter_change().
2661  *
2662  * If is_state_change is zero, a current-state record is appended.
2663  * If is_state_change is non-zero, a state-change report is appended.
2664  *
2665  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2666  * If is_group_query is zero, and if there is a packet with free space
2667  * at the tail of the queue, it will be appended to providing there
2668  * is enough free space.
2669  * Otherwise a new mbuf packet chain is allocated.
2670  *
2671  * If is_source_query is non-zero, each source is checked to see if
2672  * it was recorded for a Group-Source query, and will be omitted if
2673  * it is not both in-mode and recorded.
2674  *
2675  * The function will attempt to allocate leading space in the packet
2676  * for the IP/IGMP header to be prepended without fragmenting the chain.
2677  *
2678  * If successful the size of all data appended to the queue is returned,
2679  * otherwise an error code less than zero is returned, or zero if
2680  * no record(s) were appended.
2681  */
2682 static int
2683 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2684     const int is_state_change, const int is_group_query,
2685     const int is_source_query)
2686 {
2687 	struct igmp_grouprec	 ig;
2688 	struct igmp_grouprec	*pig;
2689 	struct ifnet		*ifp;
2690 	struct ip_msource	*ims, *nims;
2691 	struct mbuf		*m0, *m, *md;
2692 	int			 is_filter_list_change;
2693 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2694 	int			 record_has_sources;
2695 	int			 now;
2696 	int			 type;
2697 	in_addr_t		 naddr;
2698 	uint8_t			 mode;
2699 
2700 	IN_MULTI_LIST_LOCK_ASSERT();
2701 
2702 	ifp = inm->inm_ifp;
2703 	is_filter_list_change = 0;
2704 	m = NULL;
2705 	m0 = NULL;
2706 	m0srcs = 0;
2707 	msrcs = 0;
2708 	nbytes = 0;
2709 	nims = NULL;
2710 	record_has_sources = 1;
2711 	pig = NULL;
2712 	type = IGMP_DO_NOTHING;
2713 	mode = inm->inm_st[1].iss_fmode;
2714 
2715 	/*
2716 	 * If we did not transition out of ASM mode during t0->t1,
2717 	 * and there are no source nodes to process, we can skip
2718 	 * the generation of source records.
2719 	 */
2720 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2721 	    inm->inm_nsrc == 0)
2722 		record_has_sources = 0;
2723 
2724 	if (is_state_change) {
2725 		/*
2726 		 * Queue a state change record.
2727 		 * If the mode did not change, and there are non-ASM
2728 		 * listeners or source filters present,
2729 		 * we potentially need to issue two records for the group.
2730 		 * If we are transitioning to MCAST_UNDEFINED, we need
2731 		 * not send any sources.
2732 		 * If there are ASM listeners, and there was no filter
2733 		 * mode transition of any kind, do nothing.
2734 		 */
2735 		if (mode != inm->inm_st[0].iss_fmode) {
2736 			if (mode == MCAST_EXCLUDE) {
2737 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2738 				    __func__);
2739 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2740 			} else {
2741 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2742 				    __func__);
2743 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2744 				if (mode == MCAST_UNDEFINED)
2745 					record_has_sources = 0;
2746 			}
2747 		} else {
2748 			if (record_has_sources) {
2749 				is_filter_list_change = 1;
2750 			} else {
2751 				type = IGMP_DO_NOTHING;
2752 			}
2753 		}
2754 	} else {
2755 		/*
2756 		 * Queue a current state record.
2757 		 */
2758 		if (mode == MCAST_EXCLUDE) {
2759 			type = IGMP_MODE_IS_EXCLUDE;
2760 		} else if (mode == MCAST_INCLUDE) {
2761 			type = IGMP_MODE_IS_INCLUDE;
2762 			KASSERT(inm->inm_st[1].iss_asm == 0,
2763 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2764 			     __func__, inm, inm->inm_st[1].iss_asm));
2765 		}
2766 	}
2767 
2768 	/*
2769 	 * Generate the filter list changes using a separate function.
2770 	 */
2771 	if (is_filter_list_change)
2772 		return (igmp_v3_enqueue_filter_change(mq, inm));
2773 
2774 	if (type == IGMP_DO_NOTHING) {
2775 		CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2776 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2777 		return (0);
2778 	}
2779 
2780 	/*
2781 	 * If any sources are present, we must be able to fit at least
2782 	 * one in the trailing space of the tail packet's mbuf,
2783 	 * ideally more.
2784 	 */
2785 	minrec0len = sizeof(struct igmp_grouprec);
2786 	if (record_has_sources)
2787 		minrec0len += sizeof(in_addr_t);
2788 
2789 	CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2790 	    igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2791 	    inm->inm_ifp->if_xname);
2792 
2793 	/*
2794 	 * Check if we have a packet in the tail of the queue for this
2795 	 * group into which the first group record for this group will fit.
2796 	 * Otherwise allocate a new packet.
2797 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2798 	 * Note: Group records for G/GSR query responses MUST be sent
2799 	 * in their own packet.
2800 	 */
2801 	m0 = mbufq_last(mq);
2802 	if (!is_group_query &&
2803 	    m0 != NULL &&
2804 	    (m0->m_pkthdr.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2805 	    (m0->m_pkthdr.len + minrec0len) <
2806 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2807 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2808 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2809 		m = m0;
2810 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2811 	} else {
2812 		if (mbufq_full(mq)) {
2813 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2814 			return (-ENOMEM);
2815 		}
2816 		m = NULL;
2817 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2818 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2819 		if (!is_state_change && !is_group_query) {
2820 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2821 			if (m)
2822 				m->m_data += IGMP_LEADINGSPACE;
2823 		}
2824 		if (m == NULL) {
2825 			m = m_gethdr(M_NOWAIT, MT_DATA);
2826 			if (m)
2827 				M_ALIGN(m, IGMP_LEADINGSPACE);
2828 		}
2829 		if (m == NULL)
2830 			return (-ENOMEM);
2831 
2832 		igmp_save_context(m, ifp);
2833 
2834 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2835 	}
2836 
2837 	/*
2838 	 * Append group record.
2839 	 * If we have sources, we don't know how many yet.
2840 	 */
2841 	ig.ig_type = type;
2842 	ig.ig_datalen = 0;
2843 	ig.ig_numsrc = 0;
2844 	ig.ig_group = inm->inm_addr;
2845 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2846 		if (m != m0)
2847 			m_freem(m);
2848 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2849 		return (-ENOMEM);
2850 	}
2851 	nbytes += sizeof(struct igmp_grouprec);
2852 
2853 	/*
2854 	 * Append as many sources as will fit in the first packet.
2855 	 * If we are appending to a new packet, the chain allocation
2856 	 * may potentially use clusters; use m_getptr() in this case.
2857 	 * If we are appending to an existing packet, we need to obtain
2858 	 * a pointer to the group record after m_append(), in case a new
2859 	 * mbuf was allocated.
2860 	 * Only append sources which are in-mode at t1. If we are
2861 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2862 	 * include source entries.
2863 	 * Only report recorded sources in our filter set when responding
2864 	 * to a group-source query.
2865 	 */
2866 	if (record_has_sources) {
2867 		if (m == m0) {
2868 			md = m_last(m);
2869 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2870 			    md->m_len - nbytes);
2871 		} else {
2872 			md = m_getptr(m, 0, &off);
2873 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2874 			    off);
2875 		}
2876 		msrcs = 0;
2877 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2878 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2879 			    ims->ims_haddr);
2880 			now = ims_get_mode(inm, ims, 1);
2881 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2882 			if ((now != mode) ||
2883 			    (now == mode && mode == MCAST_UNDEFINED)) {
2884 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2885 				continue;
2886 			}
2887 			if (is_source_query && ims->ims_stp == 0) {
2888 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2889 				    __func__);
2890 				continue;
2891 			}
2892 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2893 			naddr = htonl(ims->ims_haddr);
2894 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2895 				if (m != m0)
2896 					m_freem(m);
2897 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2898 				    __func__);
2899 				return (-ENOMEM);
2900 			}
2901 			nbytes += sizeof(in_addr_t);
2902 			++msrcs;
2903 			if (msrcs == m0srcs)
2904 				break;
2905 		}
2906 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2907 		    msrcs);
2908 		pig->ig_numsrc = htons(msrcs);
2909 		nbytes += (msrcs * sizeof(in_addr_t));
2910 	}
2911 
2912 	if (is_source_query && msrcs == 0) {
2913 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2914 		if (m != m0)
2915 			m_freem(m);
2916 		return (0);
2917 	}
2918 
2919 	/*
2920 	 * We are good to go with first packet.
2921 	 */
2922 	if (m != m0) {
2923 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2924 		m->m_pkthdr.vt_nrecs = 1;
2925 		mbufq_enqueue(mq, m);
2926 	} else
2927 		m->m_pkthdr.vt_nrecs++;
2928 
2929 	/*
2930 	 * No further work needed if no source list in packet(s).
2931 	 */
2932 	if (!record_has_sources)
2933 		return (nbytes);
2934 
2935 	/*
2936 	 * Whilst sources remain to be announced, we need to allocate
2937 	 * a new packet and fill out as many sources as will fit.
2938 	 * Always try for a cluster first.
2939 	 */
2940 	while (nims != NULL) {
2941 		if (mbufq_full(mq)) {
2942 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2943 			return (-ENOMEM);
2944 		}
2945 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2946 		if (m)
2947 			m->m_data += IGMP_LEADINGSPACE;
2948 		if (m == NULL) {
2949 			m = m_gethdr(M_NOWAIT, MT_DATA);
2950 			if (m)
2951 				M_ALIGN(m, IGMP_LEADINGSPACE);
2952 		}
2953 		if (m == NULL)
2954 			return (-ENOMEM);
2955 		igmp_save_context(m, ifp);
2956 		md = m_getptr(m, 0, &off);
2957 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2958 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2959 
2960 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2961 			if (m != m0)
2962 				m_freem(m);
2963 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2964 			return (-ENOMEM);
2965 		}
2966 		m->m_pkthdr.vt_nrecs = 1;
2967 		nbytes += sizeof(struct igmp_grouprec);
2968 
2969 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2970 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2971 
2972 		msrcs = 0;
2973 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2974 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2975 			    ims->ims_haddr);
2976 			now = ims_get_mode(inm, ims, 1);
2977 			if ((now != mode) ||
2978 			    (now == mode && mode == MCAST_UNDEFINED)) {
2979 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2980 				continue;
2981 			}
2982 			if (is_source_query && ims->ims_stp == 0) {
2983 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2984 				    __func__);
2985 				continue;
2986 			}
2987 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2988 			naddr = htonl(ims->ims_haddr);
2989 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2990 				if (m != m0)
2991 					m_freem(m);
2992 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2993 				    __func__);
2994 				return (-ENOMEM);
2995 			}
2996 			++msrcs;
2997 			if (msrcs == m0srcs)
2998 				break;
2999 		}
3000 		pig->ig_numsrc = htons(msrcs);
3001 		nbytes += (msrcs * sizeof(in_addr_t));
3002 
3003 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
3004 		mbufq_enqueue(mq, m);
3005 	}
3006 
3007 	return (nbytes);
3008 }
3009 
3010 /*
3011  * Type used to mark record pass completion.
3012  * We exploit the fact we can cast to this easily from the
3013  * current filter modes on each ip_msource node.
3014  */
3015 typedef enum {
3016 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
3017 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
3018 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
3019 	REC_FULL = REC_ALLOW | REC_BLOCK
3020 } rectype_t;
3021 
3022 /*
3023  * Enqueue an IGMPv3 filter list change to the given output queue.
3024  *
3025  * Source list filter state is held in an RB-tree. When the filter list
3026  * for a group is changed without changing its mode, we need to compute
3027  * the deltas between T0 and T1 for each source in the filter set,
3028  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3029  *
3030  * As we may potentially queue two record types, and the entire R-B tree
3031  * needs to be walked at once, we break this out into its own function
3032  * so we can generate a tightly packed queue of packets.
3033  *
3034  * XXX This could be written to only use one tree walk, although that makes
3035  * serializing into the mbuf chains a bit harder. For now we do two walks
3036  * which makes things easier on us, and it may or may not be harder on
3037  * the L2 cache.
3038  *
3039  * If successful the size of all data appended to the queue is returned,
3040  * otherwise an error code less than zero is returned, or zero if
3041  * no record(s) were appended.
3042  */
3043 static int
3044 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3045 {
3046 	static const int MINRECLEN =
3047 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3048 	struct ifnet		*ifp;
3049 	struct igmp_grouprec	 ig;
3050 	struct igmp_grouprec	*pig;
3051 	struct ip_msource	*ims, *nims;
3052 	struct mbuf		*m, *m0, *md;
3053 	in_addr_t		 naddr;
3054 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3055 #ifdef KTR
3056 	int			 nallow, nblock;
3057 #endif
3058 	uint8_t			 mode, now, then;
3059 	rectype_t		 crt, drt, nrt;
3060 
3061 	IN_MULTI_LIST_LOCK_ASSERT();
3062 
3063 	if (inm->inm_nsrc == 0 ||
3064 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3065 		return (0);
3066 
3067 	ifp = inm->inm_ifp;			/* interface */
3068 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
3069 	crt = REC_NONE;	/* current group record type */
3070 	drt = REC_NONE;	/* mask of completed group record types */
3071 	nrt = REC_NONE;	/* record type for current node */
3072 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3073 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3074 	npbytes = 0;	/* # of bytes appended this packet */
3075 	rsrcs = 0;	/* # sources encoded in current record */
3076 	schanged = 0;	/* # nodes encoded in overall filter change */
3077 #ifdef KTR
3078 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3079 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3080 #endif
3081 	nims = NULL;	/* next tree node pointer */
3082 
3083 	/*
3084 	 * For each possible filter record mode.
3085 	 * The first kind of source we encounter tells us which
3086 	 * is the first kind of record we start appending.
3087 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3088 	 * as the inverse of the group's filter mode.
3089 	 */
3090 	while (drt != REC_FULL) {
3091 		do {
3092 			m0 = mbufq_last(mq);
3093 			if (m0 != NULL &&
3094 			    (m0->m_pkthdr.vt_nrecs + 1 <=
3095 			     IGMP_V3_REPORT_MAXRECS) &&
3096 			    (m0->m_pkthdr.len + MINRECLEN) <
3097 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3098 				m = m0;
3099 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3100 					    sizeof(struct igmp_grouprec)) /
3101 				    sizeof(in_addr_t);
3102 				CTR1(KTR_IGMPV3,
3103 				    "%s: use previous packet", __func__);
3104 			} else {
3105 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3106 				if (m)
3107 					m->m_data += IGMP_LEADINGSPACE;
3108 				if (m == NULL) {
3109 					m = m_gethdr(M_NOWAIT, MT_DATA);
3110 					if (m)
3111 						M_ALIGN(m, IGMP_LEADINGSPACE);
3112 				}
3113 				if (m == NULL) {
3114 					CTR1(KTR_IGMPV3,
3115 					    "%s: m_get*() failed", __func__);
3116 					return (-ENOMEM);
3117 				}
3118 				m->m_pkthdr.vt_nrecs = 0;
3119 				igmp_save_context(m, ifp);
3120 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3121 				    sizeof(struct igmp_grouprec)) /
3122 				    sizeof(in_addr_t);
3123 				npbytes = 0;
3124 				CTR1(KTR_IGMPV3,
3125 				    "%s: allocated new packet", __func__);
3126 			}
3127 			/*
3128 			 * Append the IGMP group record header to the
3129 			 * current packet's data area.
3130 			 * Recalculate pointer to free space for next
3131 			 * group record, in case m_append() allocated
3132 			 * a new mbuf or cluster.
3133 			 */
3134 			memset(&ig, 0, sizeof(ig));
3135 			ig.ig_group = inm->inm_addr;
3136 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3137 				if (m != m0)
3138 					m_freem(m);
3139 				CTR1(KTR_IGMPV3,
3140 				    "%s: m_append() failed", __func__);
3141 				return (-ENOMEM);
3142 			}
3143 			npbytes += sizeof(struct igmp_grouprec);
3144 			if (m != m0) {
3145 				/* new packet; offset in c hain */
3146 				md = m_getptr(m, npbytes -
3147 				    sizeof(struct igmp_grouprec), &off);
3148 				pig = (struct igmp_grouprec *)(mtod(md,
3149 				    uint8_t *) + off);
3150 			} else {
3151 				/* current packet; offset from last append */
3152 				md = m_last(m);
3153 				pig = (struct igmp_grouprec *)(mtod(md,
3154 				    uint8_t *) + md->m_len -
3155 				    sizeof(struct igmp_grouprec));
3156 			}
3157 			/*
3158 			 * Begin walking the tree for this record type
3159 			 * pass, or continue from where we left off
3160 			 * previously if we had to allocate a new packet.
3161 			 * Only report deltas in-mode at t1.
3162 			 * We need not report included sources as allowed
3163 			 * if we are in inclusive mode on the group,
3164 			 * however the converse is not true.
3165 			 */
3166 			rsrcs = 0;
3167 			if (nims == NULL)
3168 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3169 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3170 				CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3171 				    __func__, ims->ims_haddr);
3172 				now = ims_get_mode(inm, ims, 1);
3173 				then = ims_get_mode(inm, ims, 0);
3174 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3175 				    __func__, then, now);
3176 				if (now == then) {
3177 					CTR1(KTR_IGMPV3,
3178 					    "%s: skip unchanged", __func__);
3179 					continue;
3180 				}
3181 				if (mode == MCAST_EXCLUDE &&
3182 				    now == MCAST_INCLUDE) {
3183 					CTR1(KTR_IGMPV3,
3184 					    "%s: skip IN src on EX group",
3185 					    __func__);
3186 					continue;
3187 				}
3188 				nrt = (rectype_t)now;
3189 				if (nrt == REC_NONE)
3190 					nrt = (rectype_t)(~mode & REC_FULL);
3191 				if (schanged++ == 0) {
3192 					crt = nrt;
3193 				} else if (crt != nrt)
3194 					continue;
3195 				naddr = htonl(ims->ims_haddr);
3196 				if (!m_append(m, sizeof(in_addr_t),
3197 				    (void *)&naddr)) {
3198 					if (m != m0)
3199 						m_freem(m);
3200 					CTR1(KTR_IGMPV3,
3201 					    "%s: m_append() failed", __func__);
3202 					return (-ENOMEM);
3203 				}
3204 #ifdef KTR
3205 				nallow += !!(crt == REC_ALLOW);
3206 				nblock += !!(crt == REC_BLOCK);
3207 #endif
3208 				if (++rsrcs == m0srcs)
3209 					break;
3210 			}
3211 			/*
3212 			 * If we did not append any tree nodes on this
3213 			 * pass, back out of allocations.
3214 			 */
3215 			if (rsrcs == 0) {
3216 				npbytes -= sizeof(struct igmp_grouprec);
3217 				if (m != m0) {
3218 					CTR1(KTR_IGMPV3,
3219 					    "%s: m_free(m)", __func__);
3220 					m_freem(m);
3221 				} else {
3222 					CTR1(KTR_IGMPV3,
3223 					    "%s: m_adj(m, -ig)", __func__);
3224 					m_adj(m, -((int)sizeof(
3225 					    struct igmp_grouprec)));
3226 				}
3227 				continue;
3228 			}
3229 			npbytes += (rsrcs * sizeof(in_addr_t));
3230 			if (crt == REC_ALLOW)
3231 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3232 			else if (crt == REC_BLOCK)
3233 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3234 			pig->ig_numsrc = htons(rsrcs);
3235 			/*
3236 			 * Count the new group record, and enqueue this
3237 			 * packet if it wasn't already queued.
3238 			 */
3239 			m->m_pkthdr.vt_nrecs++;
3240 			if (m != m0)
3241 				mbufq_enqueue(mq, m);
3242 			nbytes += npbytes;
3243 		} while (nims != NULL);
3244 		drt |= crt;
3245 		crt = (~crt & REC_FULL);
3246 	}
3247 
3248 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3249 	    nallow, nblock);
3250 
3251 	return (nbytes);
3252 }
3253 
3254 static int
3255 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3256 {
3257 	struct mbufq	*gq;
3258 	struct mbuf	*m;		/* pending state-change */
3259 	struct mbuf	*m0;		/* copy of pending state-change */
3260 	struct mbuf	*mt;		/* last state-change in packet */
3261 	int		 docopy, domerge;
3262 	u_int		 recslen;
3263 
3264 	docopy = 0;
3265 	domerge = 0;
3266 	recslen = 0;
3267 
3268 	IN_MULTI_LIST_LOCK_ASSERT();
3269 	IGMP_LOCK_ASSERT();
3270 
3271 	/*
3272 	 * If there are further pending retransmissions, make a writable
3273 	 * copy of each queued state-change message before merging.
3274 	 */
3275 	if (inm->inm_scrv > 0)
3276 		docopy = 1;
3277 
3278 	gq = &inm->inm_scq;
3279 #ifdef KTR
3280 	if (mbufq_first(gq) == NULL) {
3281 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3282 		    __func__, inm);
3283 	}
3284 #endif
3285 
3286 	m = mbufq_first(gq);
3287 	while (m != NULL) {
3288 		/*
3289 		 * Only merge the report into the current packet if
3290 		 * there is sufficient space to do so; an IGMPv3 report
3291 		 * packet may only contain 65,535 group records.
3292 		 * Always use a simple mbuf chain concatentation to do this,
3293 		 * as large state changes for single groups may have
3294 		 * allocated clusters.
3295 		 */
3296 		domerge = 0;
3297 		mt = mbufq_last(scq);
3298 		if (mt != NULL) {
3299 			recslen = m_length(m, NULL);
3300 
3301 			if ((mt->m_pkthdr.vt_nrecs +
3302 			    m->m_pkthdr.vt_nrecs <=
3303 			    IGMP_V3_REPORT_MAXRECS) &&
3304 			    (mt->m_pkthdr.len + recslen <=
3305 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3306 				domerge = 1;
3307 		}
3308 
3309 		if (!domerge && mbufq_full(gq)) {
3310 			CTR2(KTR_IGMPV3,
3311 			    "%s: outbound queue full, skipping whole packet %p",
3312 			    __func__, m);
3313 			mt = m->m_nextpkt;
3314 			if (!docopy)
3315 				m_freem(m);
3316 			m = mt;
3317 			continue;
3318 		}
3319 
3320 		if (!docopy) {
3321 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3322 			m0 = mbufq_dequeue(gq);
3323 			m = m0->m_nextpkt;
3324 		} else {
3325 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3326 			m0 = m_dup(m, M_NOWAIT);
3327 			if (m0 == NULL)
3328 				return (ENOMEM);
3329 			m0->m_nextpkt = NULL;
3330 			m = m->m_nextpkt;
3331 		}
3332 
3333 		if (!domerge) {
3334 			CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3335 			    __func__, m0, scq);
3336 			mbufq_enqueue(scq, m0);
3337 		} else {
3338 			struct mbuf *mtl;	/* last mbuf of packet mt */
3339 
3340 			CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3341 			    __func__, m0, mt);
3342 
3343 			mtl = m_last(mt);
3344 			m0->m_flags &= ~M_PKTHDR;
3345 			mt->m_pkthdr.len += recslen;
3346 			mt->m_pkthdr.vt_nrecs +=
3347 			    m0->m_pkthdr.vt_nrecs;
3348 
3349 			mtl->m_next = m0;
3350 		}
3351 	}
3352 
3353 	return (0);
3354 }
3355 
3356 /*
3357  * Respond to a pending IGMPv3 General Query.
3358  */
3359 static void
3360 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3361 {
3362 	struct ifmultiaddr	*ifma;
3363 	struct ifnet		*ifp;
3364 	struct in_multi		*inm;
3365 	int			 retval __unused, loop;
3366 
3367 	IN_MULTI_LIST_LOCK_ASSERT();
3368 	IGMP_LOCK_ASSERT();
3369 	NET_EPOCH_ASSERT();
3370 
3371 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3372 	    ("%s: called when version %d", __func__, igi->igi_version));
3373 
3374 	/*
3375 	 * Check that there are some packets queued. If so, send them first.
3376 	 * For large number of groups the reply to general query can take
3377 	 * many packets, we should finish sending them before starting of
3378 	 * queuing the new reply.
3379 	 */
3380 	if (mbufq_len(&igi->igi_gq) != 0)
3381 		goto send;
3382 
3383 	ifp = igi->igi_ifp;
3384 
3385 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3386 		if (ifma->ifma_addr->sa_family != AF_INET ||
3387 		    ifma->ifma_protospec == NULL)
3388 			continue;
3389 
3390 		inm = (struct in_multi *)ifma->ifma_protospec;
3391 		KASSERT(ifp == inm->inm_ifp,
3392 		    ("%s: inconsistent ifp", __func__));
3393 
3394 		switch (inm->inm_state) {
3395 		case IGMP_NOT_MEMBER:
3396 		case IGMP_SILENT_MEMBER:
3397 			break;
3398 		case IGMP_REPORTING_MEMBER:
3399 		case IGMP_IDLE_MEMBER:
3400 		case IGMP_LAZY_MEMBER:
3401 		case IGMP_SLEEPING_MEMBER:
3402 		case IGMP_AWAKENING_MEMBER:
3403 			inm->inm_state = IGMP_REPORTING_MEMBER;
3404 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3405 			    inm, 0, 0, 0);
3406 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3407 			    __func__, retval);
3408 			break;
3409 		case IGMP_G_QUERY_PENDING_MEMBER:
3410 		case IGMP_SG_QUERY_PENDING_MEMBER:
3411 		case IGMP_LEAVING_MEMBER:
3412 			break;
3413 		}
3414 	}
3415 
3416 send:
3417 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3418 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3419 
3420 	/*
3421 	 * Slew transmission of bursts over 500ms intervals.
3422 	 */
3423 	if (mbufq_first(&igi->igi_gq) != NULL) {
3424 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3425 		    IGMP_RESPONSE_BURST_INTERVAL);
3426 		V_interface_timers_running = 1;
3427 	}
3428 }
3429 
3430 /*
3431  * Transmit the next pending IGMP message in the output queue.
3432  *
3433  * We get called from netisr_processqueue(). A mutex private to igmpoq
3434  * will be acquired and released around this routine.
3435  *
3436  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3437  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3438  * a link and uses a link-scope multicast address.
3439  */
3440 static void
3441 igmp_intr(struct mbuf *m)
3442 {
3443 	struct ip_moptions	 imo;
3444 	struct ifnet		*ifp;
3445 	struct mbuf		*ipopts, *m0;
3446 	int			 error;
3447 	uint32_t		 ifindex;
3448 
3449 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3450 
3451 	/*
3452 	 * Set VNET image pointer from enqueued mbuf chain
3453 	 * before doing anything else. Whilst we use interface
3454 	 * indexes to guard against interface detach, they are
3455 	 * unique to each VIMAGE and must be retrieved.
3456 	 */
3457 	CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3458 	ifindex = igmp_restore_context(m);
3459 
3460 	/*
3461 	 * Check if the ifnet still exists. This limits the scope of
3462 	 * any race in the absence of a global ifp lock for low cost
3463 	 * (an array lookup).
3464 	 */
3465 	ifp = ifnet_byindex(ifindex);
3466 	if (ifp == NULL) {
3467 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3468 		    __func__, m, ifindex);
3469 		m_freem(m);
3470 		IPSTAT_INC(ips_noroute);
3471 		goto out;
3472 	}
3473 
3474 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3475 
3476 	imo.imo_multicast_ttl  = 1;
3477 	imo.imo_multicast_vif  = -1;
3478 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3479 
3480 	/*
3481 	 * If the user requested that IGMP traffic be explicitly
3482 	 * redirected to the loopback interface (e.g. they are running a
3483 	 * MANET interface and the routing protocol needs to see the
3484 	 * updates), handle this now.
3485 	 */
3486 	if (m->m_flags & M_IGMP_LOOP)
3487 		imo.imo_multicast_ifp = V_loif;
3488 	else
3489 		imo.imo_multicast_ifp = ifp;
3490 
3491 	if (m->m_flags & M_IGMPV2) {
3492 		m0 = m;
3493 	} else {
3494 		m0 = igmp_v3_encap_report(ifp, m);
3495 		if (m0 == NULL) {
3496 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3497 			m_freem(m);
3498 			IPSTAT_INC(ips_odropped);
3499 			goto out;
3500 		}
3501 	}
3502 
3503 	igmp_scrub_context(m0);
3504 	m_clrprotoflags(m);
3505 	m0->m_pkthdr.rcvif = V_loif;
3506 #ifdef MAC
3507 	mac_netinet_igmp_send(ifp, m0);
3508 #endif
3509 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3510 	if (error) {
3511 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3512 		goto out;
3513 	}
3514 
3515 	IGMPSTAT_INC(igps_snd_reports);
3516 
3517 out:
3518 	/*
3519 	 * We must restore the existing vnet pointer before
3520 	 * continuing as we are run from netisr context.
3521 	 */
3522 	CURVNET_RESTORE();
3523 }
3524 
3525 /*
3526  * Encapsulate an IGMPv3 report.
3527  *
3528  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3529  * chain has already had its IP/IGMPv3 header prepended. In this case
3530  * the function will not attempt to prepend; the lengths and checksums
3531  * will however be re-computed.
3532  *
3533  * Returns a pointer to the new mbuf chain head, or NULL if the
3534  * allocation failed.
3535  */
3536 static struct mbuf *
3537 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3538 {
3539 	struct igmp_report	*igmp;
3540 	struct ip		*ip;
3541 	int			 hdrlen, igmpreclen;
3542 
3543 	KASSERT((m->m_flags & M_PKTHDR),
3544 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3545 
3546 	igmpreclen = m_length(m, NULL);
3547 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3548 
3549 	if (m->m_flags & M_IGMPV3_HDR) {
3550 		igmpreclen -= hdrlen;
3551 	} else {
3552 		M_PREPEND(m, hdrlen, M_NOWAIT);
3553 		if (m == NULL)
3554 			return (NULL);
3555 		m->m_flags |= M_IGMPV3_HDR;
3556 	}
3557 
3558 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3559 
3560 	m->m_data += sizeof(struct ip);
3561 	m->m_len -= sizeof(struct ip);
3562 
3563 	igmp = mtod(m, struct igmp_report *);
3564 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3565 	igmp->ir_rsv1 = 0;
3566 	igmp->ir_rsv2 = 0;
3567 	igmp->ir_numgrps = htons(m->m_pkthdr.vt_nrecs);
3568 	igmp->ir_cksum = 0;
3569 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3570 	m->m_pkthdr.vt_nrecs = 0;
3571 
3572 	m->m_data -= sizeof(struct ip);
3573 	m->m_len += sizeof(struct ip);
3574 
3575 	ip = mtod(m, struct ip *);
3576 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3577 	ip->ip_len = htons(hdrlen + igmpreclen);
3578 	ip->ip_off = htons(IP_DF);
3579 	ip->ip_p = IPPROTO_IGMP;
3580 	ip->ip_sum = 0;
3581 
3582 	ip->ip_src.s_addr = INADDR_ANY;
3583 
3584 	if (m->m_flags & M_IGMP_LOOP) {
3585 		struct in_ifaddr *ia;
3586 
3587 		IFP_TO_IA(ifp, ia);
3588 		if (ia != NULL)
3589 			ip->ip_src = ia->ia_addr.sin_addr;
3590 	}
3591 
3592 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3593 
3594 	return (m);
3595 }
3596 
3597 #ifdef KTR
3598 static char *
3599 igmp_rec_type_to_str(const int type)
3600 {
3601 
3602 	switch (type) {
3603 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3604 			return "TO_EX";
3605 			break;
3606 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3607 			return "TO_IN";
3608 			break;
3609 		case IGMP_MODE_IS_EXCLUDE:
3610 			return "MODE_EX";
3611 			break;
3612 		case IGMP_MODE_IS_INCLUDE:
3613 			return "MODE_IN";
3614 			break;
3615 		case IGMP_ALLOW_NEW_SOURCES:
3616 			return "ALLOW_NEW";
3617 			break;
3618 		case IGMP_BLOCK_OLD_SOURCES:
3619 			return "BLOCK_OLD";
3620 			break;
3621 		default:
3622 			break;
3623 	}
3624 	return "unknown";
3625 }
3626 #endif
3627 
3628 #ifdef VIMAGE
3629 static void
3630 vnet_igmp_init(const void *unused __unused)
3631 {
3632 
3633 	netisr_register_vnet(&igmp_nh);
3634 }
3635 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3636     vnet_igmp_init, NULL);
3637 
3638 static void
3639 vnet_igmp_uninit(const void *unused __unused)
3640 {
3641 
3642 	/* This can happen when we shutdown the entire network stack. */
3643 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3644 
3645 	netisr_unregister_vnet(&igmp_nh);
3646 }
3647 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3648     vnet_igmp_uninit, NULL);
3649 #endif
3650 
3651 #ifdef DDB
3652 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3653 {
3654 	struct igmp_ifsoftc *igi, *tigi;
3655 	LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3656 
3657 	if (!have_addr) {
3658 		db_printf("usage: show igi_list <addr>\n");
3659 		return;
3660 	}
3661 	igi_head = (struct _igi_list *)addr;
3662 
3663 	LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3664 		db_printf("igmp_ifsoftc %p:\n", igi);
3665 		db_printf("    ifp %p\n", igi->igi_ifp);
3666 		db_printf("    version %u\n", igi->igi_version);
3667 		db_printf("    v1_timer %u\n", igi->igi_v1_timer);
3668 		db_printf("    v2_timer %u\n", igi->igi_v2_timer);
3669 		db_printf("    v3_timer %u\n", igi->igi_v3_timer);
3670 		db_printf("    flags %#x\n", igi->igi_flags);
3671 		db_printf("    rv %u\n", igi->igi_rv);
3672 		db_printf("    qi %u\n", igi->igi_qi);
3673 		db_printf("    qri %u\n", igi->igi_qri);
3674 		db_printf("    uri %u\n", igi->igi_uri);
3675 		/* struct mbufq    igi_gq; */
3676 		db_printf("\n");
3677 	}
3678 }
3679 #endif
3680 
3681 static int
3682 igmp_modevent(module_t mod, int type, void *unused __unused)
3683 {
3684 
3685 	switch (type) {
3686 	case MOD_LOAD:
3687 		CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3688 		IGMP_LOCK_INIT();
3689 		m_raopt = igmp_ra_alloc();
3690 		netisr_register(&igmp_nh);
3691 		break;
3692 	case MOD_UNLOAD:
3693 		CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3694 		netisr_unregister(&igmp_nh);
3695 		m_free(m_raopt);
3696 		m_raopt = NULL;
3697 		IGMP_LOCK_DESTROY();
3698 		break;
3699 	default:
3700 		return (EOPNOTSUPP);
3701 	}
3702 	return (0);
3703 }
3704 
3705 static moduledata_t igmp_mod = {
3706     "igmp",
3707     igmp_modevent,
3708     0
3709 };
3710 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);
3711