xref: /freebsd/sys/netinet/igmp.c (revision 4f52dfbb8d6c4d446500c5b097e3806ec219fbd4)
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/rmlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/ktr.h>
69 #include <sys/condvar.h>
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif
74 
75 #include <net/if.h>
76 #include <net/if_var.h>
77 #include <net/netisr.h>
78 #include <net/vnet.h>
79 
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 #include <netinet/in_systm.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_var.h>
85 #include <netinet/ip_options.h>
86 #include <netinet/igmp.h>
87 #include <netinet/igmp_var.h>
88 
89 #include <machine/in_cksum.h>
90 
91 #include <security/mac/mac_framework.h>
92 
93 #ifndef KTR_IGMPV3
94 #define KTR_IGMPV3 KTR_INET
95 #endif
96 
97 static struct igmp_ifsoftc *
98 		igi_alloc_locked(struct ifnet *);
99 static void	igi_delete_locked(const struct ifnet *);
100 static void	igmp_dispatch_queue(struct mbufq *, int, const int);
101 static void	igmp_fasttimo_vnet(void);
102 static void	igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
103 static int	igmp_handle_state_change(struct in_multi *,
104 		    struct igmp_ifsoftc *);
105 static int	igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
106 static int	igmp_input_v1_query(struct ifnet *, const struct ip *,
107 		    const struct igmp *);
108 static int	igmp_input_v2_query(struct ifnet *, const struct ip *,
109 		    const struct igmp *);
110 static int	igmp_input_v3_query(struct ifnet *, const struct ip *,
111 		    /*const*/ struct igmpv3 *);
112 static int	igmp_input_v3_group_query(struct in_multi *,
113 		    struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
114 static int	igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
115 		    /*const*/ struct igmp *);
116 static int	igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
117 		    /*const*/ struct igmp *);
118 static void	igmp_intr(struct mbuf *);
119 static int	igmp_isgroupreported(const struct in_addr);
120 static struct mbuf *
121 		igmp_ra_alloc(void);
122 #ifdef KTR
123 static char *	igmp_rec_type_to_str(const int);
124 #endif
125 static void	igmp_set_version(struct igmp_ifsoftc *, const int);
126 static void	igmp_slowtimo_vnet(void);
127 static int	igmp_v1v2_queue_report(struct in_multi *, const int);
128 static void	igmp_v1v2_process_group_timer(struct in_multi *, const int);
129 static void	igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
130 static void	igmp_v2_update_group(struct in_multi *, const int);
131 static void	igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
132 static void	igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
133 static struct mbuf *
134 		igmp_v3_encap_report(struct ifnet *, struct mbuf *);
135 static int	igmp_v3_enqueue_group_record(struct mbufq *,
136 		    struct in_multi *, const int, const int, const int);
137 static int	igmp_v3_enqueue_filter_change(struct mbufq *,
138 		    struct in_multi *);
139 static void	igmp_v3_process_group_timers(struct in_multi_head *,
140 		    struct mbufq *, struct mbufq *, struct in_multi *,
141 		    const int);
142 static int	igmp_v3_merge_state_changes(struct in_multi *,
143 		    struct mbufq *);
144 static void	igmp_v3_suppress_group_record(struct in_multi *);
145 static int	sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
146 static int	sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
147 static int	sysctl_igmp_ifinfo(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 static VNET_DEFINE(int, interface_timers_running);	/* IGMPv3 general
221 							 * query response */
222 static VNET_DEFINE(int, state_change_timers_running);	/* IGMPv3 state-change
223 							 * retransmit */
224 static VNET_DEFINE(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 static VNET_DEFINE(LIST_HEAD(, igmp_ifsoftc), igi_head) =
233     LIST_HEAD_INITIALIZER(igi_head);
234 static VNET_DEFINE(struct igmpstat, igmpstat) = {
235 	.igps_version = IGPS_VERSION_3,
236 	.igps_len = sizeof(struct igmpstat),
237 };
238 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
239 
240 #define	V_igi_head			VNET(igi_head)
241 #define	V_igmpstat			VNET(igmpstat)
242 #define	V_igmp_gsrdelay			VNET(igmp_gsrdelay)
243 
244 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
245 static VNET_DEFINE(int, igmp_sendra) = 1;
246 static VNET_DEFINE(int, igmp_sendlocal) = 1;
247 static VNET_DEFINE(int, igmp_v1enable) = 1;
248 static VNET_DEFINE(int, igmp_v2enable) = 1;
249 static VNET_DEFINE(int, igmp_legacysupp);
250 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
251 
252 #define	V_igmp_recvifkludge		VNET(igmp_recvifkludge)
253 #define	V_igmp_sendra			VNET(igmp_sendra)
254 #define	V_igmp_sendlocal		VNET(igmp_sendlocal)
255 #define	V_igmp_v1enable			VNET(igmp_v1enable)
256 #define	V_igmp_v2enable			VNET(igmp_v2enable)
257 #define	V_igmp_legacysupp		VNET(igmp_legacysupp)
258 #define	V_igmp_default_version		VNET(igmp_default_version)
259 
260 /*
261  * Virtualized sysctls.
262  */
263 SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW,
264     &VNET_NAME(igmpstat), igmpstat, "");
265 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
266     &VNET_NAME(igmp_recvifkludge), 0,
267     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
268 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
269     &VNET_NAME(igmp_sendra), 0,
270     "Send IP Router Alert option in IGMPv2/v3 messages");
271 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
272     &VNET_NAME(igmp_sendlocal), 0,
273     "Send IGMP membership reports for 224.0.0.0/24 groups");
274 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
275     &VNET_NAME(igmp_v1enable), 0,
276     "Enable backwards compatibility with IGMPv1");
277 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
278     &VNET_NAME(igmp_v2enable), 0,
279     "Enable backwards compatibility with IGMPv2");
280 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
281     &VNET_NAME(igmp_legacysupp), 0,
282     "Allow v1/v2 reports to suppress v3 group responses");
283 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
284     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
285     &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
286     "Default version of IGMP to run on each interface");
287 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
288     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
289     &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
290     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
291 
292 /*
293  * Non-virtualized sysctls.
294  */
295 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
296     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
297     "Per-interface IGMPv3 state");
298 
299 static __inline void
300 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
301 {
302 
303 #ifdef VIMAGE
304 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
305 #endif /* VIMAGE */
306 	m->m_pkthdr.flowid = ifp->if_index;
307 }
308 
309 static __inline void
310 igmp_scrub_context(struct mbuf *m)
311 {
312 
313 	m->m_pkthdr.PH_loc.ptr = NULL;
314 	m->m_pkthdr.flowid = 0;
315 }
316 
317 /*
318  * Restore context from a queued IGMP output chain.
319  * Return saved ifindex.
320  *
321  * VIMAGE: The assertion is there to make sure that we
322  * actually called CURVNET_SET() with what's in the mbuf chain.
323  */
324 static __inline uint32_t
325 igmp_restore_context(struct mbuf *m)
326 {
327 
328 #ifdef notyet
329 #if defined(VIMAGE) && defined(INVARIANTS)
330 	KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
331 	    ("%s: called when curvnet was not restored", __func__));
332 #endif
333 #endif
334 	return (m->m_pkthdr.flowid);
335 }
336 
337 /*
338  * Retrieve or set default IGMP version.
339  *
340  * VIMAGE: Assume curvnet set by caller.
341  * SMPng: NOTE: Serialized by IGMP lock.
342  */
343 static int
344 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
345 {
346 	int	 error;
347 	int	 new;
348 
349 	error = sysctl_wire_old_buffer(req, sizeof(int));
350 	if (error)
351 		return (error);
352 
353 	IGMP_LOCK();
354 
355 	new = V_igmp_default_version;
356 
357 	error = sysctl_handle_int(oidp, &new, 0, req);
358 	if (error || !req->newptr)
359 		goto out_locked;
360 
361 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
362 		error = EINVAL;
363 		goto out_locked;
364 	}
365 
366 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
367 	     V_igmp_default_version, new);
368 
369 	V_igmp_default_version = new;
370 
371 out_locked:
372 	IGMP_UNLOCK();
373 	return (error);
374 }
375 
376 /*
377  * Retrieve or set threshold between group-source queries in seconds.
378  *
379  * VIMAGE: Assume curvnet set by caller.
380  * SMPng: NOTE: Serialized by IGMP lock.
381  */
382 static int
383 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
384 {
385 	int error;
386 	int i;
387 
388 	error = sysctl_wire_old_buffer(req, sizeof(int));
389 	if (error)
390 		return (error);
391 
392 	IGMP_LOCK();
393 
394 	i = V_igmp_gsrdelay.tv_sec;
395 
396 	error = sysctl_handle_int(oidp, &i, 0, req);
397 	if (error || !req->newptr)
398 		goto out_locked;
399 
400 	if (i < -1 || i >= 60) {
401 		error = EINVAL;
402 		goto out_locked;
403 	}
404 
405 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
406 	     V_igmp_gsrdelay.tv_sec, i);
407 	V_igmp_gsrdelay.tv_sec = i;
408 
409 out_locked:
410 	IGMP_UNLOCK();
411 	return (error);
412 }
413 
414 /*
415  * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
416  * For use by ifmcstat(8).
417  *
418  * SMPng: NOTE: Does an unlocked ifindex space read.
419  * VIMAGE: Assume curvnet set by caller. The node handler itself
420  * is not directly virtualized.
421  */
422 static int
423 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
424 {
425 	int			*name;
426 	int			 error;
427 	u_int			 namelen;
428 	struct ifnet		*ifp;
429 	struct igmp_ifsoftc	*igi;
430 
431 	name = (int *)arg1;
432 	namelen = arg2;
433 
434 	if (req->newptr != NULL)
435 		return (EPERM);
436 
437 	if (namelen != 1)
438 		return (EINVAL);
439 
440 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
441 	if (error)
442 		return (error);
443 
444 	IN_MULTI_LIST_LOCK();
445 	IGMP_LOCK();
446 
447 	if (name[0] <= 0 || name[0] > V_if_index) {
448 		error = ENOENT;
449 		goto out_locked;
450 	}
451 
452 	error = ENOENT;
453 
454 	ifp = ifnet_byindex(name[0]);
455 	if (ifp == NULL)
456 		goto out_locked;
457 
458 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
459 		if (ifp == igi->igi_ifp) {
460 			struct igmp_ifinfo info;
461 
462 			info.igi_version = igi->igi_version;
463 			info.igi_v1_timer = igi->igi_v1_timer;
464 			info.igi_v2_timer = igi->igi_v2_timer;
465 			info.igi_v3_timer = igi->igi_v3_timer;
466 			info.igi_flags = igi->igi_flags;
467 			info.igi_rv = igi->igi_rv;
468 			info.igi_qi = igi->igi_qi;
469 			info.igi_qri = igi->igi_qri;
470 			info.igi_uri = igi->igi_uri;
471 			error = SYSCTL_OUT(req, &info, sizeof(info));
472 			break;
473 		}
474 	}
475 
476 out_locked:
477 	IGMP_UNLOCK();
478 	IN_MULTI_LIST_UNLOCK();
479 	return (error);
480 }
481 
482 /*
483  * Dispatch an entire queue of pending packet chains
484  * using the netisr.
485  * VIMAGE: Assumes the vnet pointer has been set.
486  */
487 static void
488 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
489 {
490 	struct mbuf *m;
491 
492 	while ((m = mbufq_dequeue(mq)) != NULL) {
493 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
494 		if (loop)
495 			m->m_flags |= M_IGMP_LOOP;
496 		netisr_dispatch(NETISR_IGMP, m);
497 		if (--limit == 0)
498 			break;
499 	}
500 }
501 
502 /*
503  * Filter outgoing IGMP report state by group.
504  *
505  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
506  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
507  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
508  * this may break certain IGMP snooping switches which rely on the old
509  * report behaviour.
510  *
511  * Return zero if the given group is one for which IGMP reports
512  * should be suppressed, or non-zero if reports should be issued.
513  */
514 static __inline int
515 igmp_isgroupreported(const struct in_addr addr)
516 {
517 
518 	if (in_allhosts(addr) ||
519 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
520 		return (0);
521 
522 	return (1);
523 }
524 
525 /*
526  * Construct a Router Alert option to use in outgoing packets.
527  */
528 static struct mbuf *
529 igmp_ra_alloc(void)
530 {
531 	struct mbuf	*m;
532 	struct ipoption	*p;
533 
534 	m = m_get(M_WAITOK, MT_DATA);
535 	p = mtod(m, struct ipoption *);
536 	p->ipopt_dst.s_addr = INADDR_ANY;
537 	p->ipopt_list[0] = (char)IPOPT_RA;	/* Router Alert Option */
538 	p->ipopt_list[1] = 0x04;		/* 4 bytes long */
539 	p->ipopt_list[2] = IPOPT_EOL;		/* End of IP option list */
540 	p->ipopt_list[3] = 0x00;		/* pad byte */
541 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
542 
543 	return (m);
544 }
545 
546 /*
547  * Attach IGMP when PF_INET is attached to an interface.
548  */
549 struct igmp_ifsoftc *
550 igmp_domifattach(struct ifnet *ifp)
551 {
552 	struct igmp_ifsoftc *igi;
553 
554 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
555 	    __func__, ifp, ifp->if_xname);
556 
557 	IGMP_LOCK();
558 
559 	igi = igi_alloc_locked(ifp);
560 	if (!(ifp->if_flags & IFF_MULTICAST))
561 		igi->igi_flags |= IGIF_SILENT;
562 
563 	IGMP_UNLOCK();
564 
565 	return (igi);
566 }
567 
568 /*
569  * VIMAGE: assume curvnet set by caller.
570  */
571 static struct igmp_ifsoftc *
572 igi_alloc_locked(/*const*/ struct ifnet *ifp)
573 {
574 	struct igmp_ifsoftc *igi;
575 
576 	IGMP_LOCK_ASSERT();
577 
578 	igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
579 	if (igi == NULL)
580 		goto out;
581 
582 	igi->igi_ifp = ifp;
583 	igi->igi_version = V_igmp_default_version;
584 	igi->igi_flags = 0;
585 	igi->igi_rv = IGMP_RV_INIT;
586 	igi->igi_qi = IGMP_QI_INIT;
587 	igi->igi_qri = IGMP_QRI_INIT;
588 	igi->igi_uri = IGMP_URI_INIT;
589 	mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
590 
591 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
592 
593 	CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
594 	     ifp, ifp->if_xname);
595 
596 out:
597 	return (igi);
598 }
599 
600 /*
601  * Hook for ifdetach.
602  *
603  * NOTE: Some finalization tasks need to run before the protocol domain
604  * is detached, but also before the link layer does its cleanup.
605  *
606  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
607  * XXX This is also bitten by unlocked ifma_protospec access.
608  */
609 void
610 igmp_ifdetach(struct ifnet *ifp)
611 {
612 	struct igmp_ifsoftc	*igi;
613 	struct ifmultiaddr	*ifma, *next;
614 	struct in_multi		*inm;
615 	struct in_multi_head inm_free_tmp;
616 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
617 	    ifp->if_xname);
618 
619 	SLIST_INIT(&inm_free_tmp);
620 	IGMP_LOCK();
621 
622 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
623 	if (igi->igi_version == IGMP_VERSION_3) {
624 		IF_ADDR_WLOCK(ifp);
625 	restart:
626 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
627 			if (ifma->ifma_addr->sa_family != AF_INET ||
628 			    ifma->ifma_protospec == NULL)
629 				continue;
630 			inm = (struct in_multi *)ifma->ifma_protospec;
631 			if (inm->inm_state == IGMP_LEAVING_MEMBER)
632 				inm_rele_locked(&inm_free_tmp, inm);
633 			inm_clear_recorded(inm);
634 			if (__predict_false(ifma_restart)) {
635 				ifma_restart = false;
636 				goto restart;
637 			}
638 		}
639 		IF_ADDR_WUNLOCK(ifp);
640 		inm_release_list_deferred(&inm_free_tmp);
641 	}
642 	IGMP_UNLOCK();
643 
644 }
645 
646 /*
647  * Hook for domifdetach.
648  */
649 void
650 igmp_domifdetach(struct ifnet *ifp)
651 {
652 
653 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
654 	    __func__, ifp, ifp->if_xname);
655 
656 	IGMP_LOCK();
657 	igi_delete_locked(ifp);
658 	IGMP_UNLOCK();
659 }
660 
661 static void
662 igi_delete_locked(const struct ifnet *ifp)
663 {
664 	struct igmp_ifsoftc *igi, *tigi;
665 
666 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
667 	    __func__, ifp, ifp->if_xname);
668 
669 	IGMP_LOCK_ASSERT();
670 
671 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
672 		if (igi->igi_ifp == ifp) {
673 			/*
674 			 * Free deferred General Query responses.
675 			 */
676 			mbufq_drain(&igi->igi_gq);
677 
678 			LIST_REMOVE(igi, igi_link);
679 			free(igi, M_IGMP);
680 			return;
681 		}
682 	}
683 }
684 
685 /*
686  * Process a received IGMPv1 query.
687  * Return non-zero if the message should be dropped.
688  *
689  * VIMAGE: The curvnet pointer is derived from the input ifp.
690  */
691 static int
692 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
693     const struct igmp *igmp)
694 {
695 	struct ifmultiaddr	*ifma;
696 	struct igmp_ifsoftc	*igi;
697 	struct in_multi		*inm;
698 
699 	/*
700 	 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
701 	 * 224.0.0.1. They are always treated as General Queries.
702 	 * igmp_group is always ignored. Do not drop it as a userland
703 	 * daemon may wish to see it.
704 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
705 	 */
706 	if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
707 		IGMPSTAT_INC(igps_rcv_badqueries);
708 		return (0);
709 	}
710 	IGMPSTAT_INC(igps_rcv_gen_queries);
711 
712 	IN_MULTI_LIST_LOCK();
713 	IGMP_LOCK();
714 
715 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
716 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
717 
718 	if (igi->igi_flags & IGIF_LOOPBACK) {
719 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
720 		    ifp, ifp->if_xname);
721 		goto out_locked;
722 	}
723 
724 	/*
725 	 * Switch to IGMPv1 host compatibility mode.
726 	 */
727 	igmp_set_version(igi, IGMP_VERSION_1);
728 
729 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
730 
731 	/*
732 	 * Start the timers in all of our group records
733 	 * for the interface on which the query arrived,
734 	 * except those which are already running.
735 	 */
736 	IF_ADDR_RLOCK(ifp);
737 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
738 		if (ifma->ifma_addr->sa_family != AF_INET ||
739 		    ifma->ifma_protospec == NULL)
740 			continue;
741 		inm = (struct in_multi *)ifma->ifma_protospec;
742 		if (inm->inm_timer != 0)
743 			continue;
744 		switch (inm->inm_state) {
745 		case IGMP_NOT_MEMBER:
746 		case IGMP_SILENT_MEMBER:
747 			break;
748 		case IGMP_G_QUERY_PENDING_MEMBER:
749 		case IGMP_SG_QUERY_PENDING_MEMBER:
750 		case IGMP_REPORTING_MEMBER:
751 		case IGMP_IDLE_MEMBER:
752 		case IGMP_LAZY_MEMBER:
753 		case IGMP_SLEEPING_MEMBER:
754 		case IGMP_AWAKENING_MEMBER:
755 			inm->inm_state = IGMP_REPORTING_MEMBER;
756 			inm->inm_timer = IGMP_RANDOM_DELAY(
757 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
758 			V_current_state_timers_running = 1;
759 			break;
760 		case IGMP_LEAVING_MEMBER:
761 			break;
762 		}
763 	}
764 	IF_ADDR_RUNLOCK(ifp);
765 
766 out_locked:
767 	IGMP_UNLOCK();
768 	IN_MULTI_LIST_UNLOCK();
769 
770 	return (0);
771 }
772 
773 /*
774  * Process a received IGMPv2 general or group-specific query.
775  */
776 static int
777 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
778     const struct igmp *igmp)
779 {
780 	struct ifmultiaddr	*ifma;
781 	struct igmp_ifsoftc	*igi;
782 	struct in_multi		*inm;
783 	int			 is_general_query;
784 	uint16_t		 timer;
785 
786 	is_general_query = 0;
787 
788 	/*
789 	 * Validate address fields upfront.
790 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
791 	 */
792 	if (in_nullhost(igmp->igmp_group)) {
793 		/*
794 		 * IGMPv2 General Query.
795 		 * If this was not sent to the all-hosts group, ignore it.
796 		 */
797 		if (!in_allhosts(ip->ip_dst))
798 			return (0);
799 		IGMPSTAT_INC(igps_rcv_gen_queries);
800 		is_general_query = 1;
801 	} else {
802 		/* IGMPv2 Group-Specific Query. */
803 		IGMPSTAT_INC(igps_rcv_group_queries);
804 	}
805 
806 	IN_MULTI_LIST_LOCK();
807 	IGMP_LOCK();
808 
809 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
810 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
811 
812 	if (igi->igi_flags & IGIF_LOOPBACK) {
813 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
814 		    ifp, ifp->if_xname);
815 		goto out_locked;
816 	}
817 
818 	/*
819 	 * Ignore v2 query if in v1 Compatibility Mode.
820 	 */
821 	if (igi->igi_version == IGMP_VERSION_1)
822 		goto out_locked;
823 
824 	igmp_set_version(igi, IGMP_VERSION_2);
825 
826 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
827 	if (timer == 0)
828 		timer = 1;
829 
830 	if (is_general_query) {
831 		/*
832 		 * For each reporting group joined on this
833 		 * interface, kick the report timer.
834 		 */
835 		CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
836 		    ifp, ifp->if_xname);
837 		IF_ADDR_RLOCK(ifp);
838 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
839 			if (ifma->ifma_addr->sa_family != AF_INET ||
840 			    ifma->ifma_protospec == NULL)
841 				continue;
842 			inm = (struct in_multi *)ifma->ifma_protospec;
843 			igmp_v2_update_group(inm, timer);
844 		}
845 		IF_ADDR_RUNLOCK(ifp);
846 	} else {
847 		/*
848 		 * Group-specific IGMPv2 query, we need only
849 		 * look up the single group to process it.
850 		 */
851 		inm = inm_lookup(ifp, igmp->igmp_group);
852 		if (inm != NULL) {
853 			CTR3(KTR_IGMPV3,
854 			    "process v2 query 0x%08x on ifp %p(%s)",
855 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
856 			igmp_v2_update_group(inm, timer);
857 		}
858 	}
859 
860 out_locked:
861 	IGMP_UNLOCK();
862 	IN_MULTI_LIST_UNLOCK();
863 
864 	return (0);
865 }
866 
867 /*
868  * Update the report timer on a group in response to an IGMPv2 query.
869  *
870  * If we are becoming the reporting member for this group, start the timer.
871  * If we already are the reporting member for this group, and timer is
872  * below the threshold, reset it.
873  *
874  * We may be updating the group for the first time since we switched
875  * to IGMPv3. If we are, then we must clear any recorded source lists,
876  * and transition to REPORTING state; the group timer is overloaded
877  * for group and group-source query responses.
878  *
879  * Unlike IGMPv3, the delay per group should be jittered
880  * to avoid bursts of IGMPv2 reports.
881  */
882 static void
883 igmp_v2_update_group(struct in_multi *inm, const int timer)
884 {
885 
886 	CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
887 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
888 
889 	IN_MULTI_LIST_LOCK_ASSERT();
890 
891 	switch (inm->inm_state) {
892 	case IGMP_NOT_MEMBER:
893 	case IGMP_SILENT_MEMBER:
894 		break;
895 	case IGMP_REPORTING_MEMBER:
896 		if (inm->inm_timer != 0 &&
897 		    inm->inm_timer <= timer) {
898 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
899 			    "skipping.", __func__);
900 			break;
901 		}
902 		/* FALLTHROUGH */
903 	case IGMP_SG_QUERY_PENDING_MEMBER:
904 	case IGMP_G_QUERY_PENDING_MEMBER:
905 	case IGMP_IDLE_MEMBER:
906 	case IGMP_LAZY_MEMBER:
907 	case IGMP_AWAKENING_MEMBER:
908 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
909 		inm->inm_state = IGMP_REPORTING_MEMBER;
910 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
911 		V_current_state_timers_running = 1;
912 		break;
913 	case IGMP_SLEEPING_MEMBER:
914 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
915 		inm->inm_state = IGMP_AWAKENING_MEMBER;
916 		break;
917 	case IGMP_LEAVING_MEMBER:
918 		break;
919 	}
920 }
921 
922 /*
923  * Process a received IGMPv3 general, group-specific or
924  * group-and-source-specific query.
925  * Assumes m has already been pulled up to the full IGMP message length.
926  * Return 0 if successful, otherwise an appropriate error code is returned.
927  */
928 static int
929 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
930     /*const*/ struct igmpv3 *igmpv3)
931 {
932 	struct igmp_ifsoftc	*igi;
933 	struct in_multi		*inm;
934 	int			 is_general_query;
935 	uint32_t		 maxresp, nsrc, qqi;
936 	uint16_t		 timer;
937 	uint8_t			 qrv;
938 
939 	is_general_query = 0;
940 
941 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
942 
943 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
944 	if (maxresp >= 128) {
945 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
946 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
947 	}
948 
949 	/*
950 	 * Robustness must never be less than 2 for on-wire IGMPv3.
951 	 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
952 	 * an exception for interfaces whose IGMPv3 state changes
953 	 * are redirected to loopback (e.g. MANET).
954 	 */
955 	qrv = IGMP_QRV(igmpv3->igmp_misc);
956 	if (qrv < 2) {
957 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
958 		    qrv, IGMP_RV_INIT);
959 		qrv = IGMP_RV_INIT;
960 	}
961 
962 	qqi = igmpv3->igmp_qqi;
963 	if (qqi >= 128) {
964 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
965 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
966 	}
967 
968 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
969 	if (timer == 0)
970 		timer = 1;
971 
972 	nsrc = ntohs(igmpv3->igmp_numsrc);
973 
974 	/*
975 	 * Validate address fields and versions upfront before
976 	 * accepting v3 query.
977 	 * XXX SMPng: Unlocked access to igmpstat counters here.
978 	 */
979 	if (in_nullhost(igmpv3->igmp_group)) {
980 		/*
981 		 * IGMPv3 General Query.
982 		 *
983 		 * General Queries SHOULD be directed to 224.0.0.1.
984 		 * A general query with a source list has undefined
985 		 * behaviour; discard it.
986 		 */
987 		IGMPSTAT_INC(igps_rcv_gen_queries);
988 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
989 			IGMPSTAT_INC(igps_rcv_badqueries);
990 			return (0);
991 		}
992 		is_general_query = 1;
993 	} else {
994 		/* Group or group-source specific query. */
995 		if (nsrc == 0)
996 			IGMPSTAT_INC(igps_rcv_group_queries);
997 		else
998 			IGMPSTAT_INC(igps_rcv_gsr_queries);
999 	}
1000 
1001 	IN_MULTI_LIST_LOCK();
1002 	IGMP_LOCK();
1003 
1004 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1005 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1006 
1007 	if (igi->igi_flags & IGIF_LOOPBACK) {
1008 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1009 		    ifp, ifp->if_xname);
1010 		goto out_locked;
1011 	}
1012 
1013 	/*
1014 	 * Discard the v3 query if we're in Compatibility Mode.
1015 	 * The RFC is not obviously worded that hosts need to stay in
1016 	 * compatibility mode until the Old Version Querier Present
1017 	 * timer expires.
1018 	 */
1019 	if (igi->igi_version != IGMP_VERSION_3) {
1020 		CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1021 		    igi->igi_version, ifp, ifp->if_xname);
1022 		goto out_locked;
1023 	}
1024 
1025 	igmp_set_version(igi, IGMP_VERSION_3);
1026 	igi->igi_rv = qrv;
1027 	igi->igi_qi = qqi;
1028 	igi->igi_qri = maxresp;
1029 
1030 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1031 	    maxresp);
1032 
1033 	if (is_general_query) {
1034 		/*
1035 		 * Schedule a current-state report on this ifp for
1036 		 * all groups, possibly containing source lists.
1037 		 * If there is a pending General Query response
1038 		 * scheduled earlier than the selected delay, do
1039 		 * not schedule any other reports.
1040 		 * Otherwise, reset the interface timer.
1041 		 */
1042 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1043 		    ifp, ifp->if_xname);
1044 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1045 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1046 			V_interface_timers_running = 1;
1047 		}
1048 	} else {
1049 		/*
1050 		 * Group-source-specific queries are throttled on
1051 		 * a per-group basis to defeat denial-of-service attempts.
1052 		 * Queries for groups we are not a member of on this
1053 		 * link are simply ignored.
1054 		 */
1055 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1056 		if (inm == NULL)
1057 			goto out_locked;
1058 		if (nsrc > 0) {
1059 			if (!ratecheck(&inm->inm_lastgsrtv,
1060 			    &V_igmp_gsrdelay)) {
1061 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1062 				    __func__);
1063 				IGMPSTAT_INC(igps_drop_gsr_queries);
1064 				goto out_locked;
1065 			}
1066 		}
1067 		CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1068 		     ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1069 		/*
1070 		 * If there is a pending General Query response
1071 		 * scheduled sooner than the selected delay, no
1072 		 * further report need be scheduled.
1073 		 * Otherwise, prepare to respond to the
1074 		 * group-specific or group-and-source query.
1075 		 */
1076 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1077 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1078 	}
1079 
1080 out_locked:
1081 	IGMP_UNLOCK();
1082 	IN_MULTI_LIST_UNLOCK();
1083 
1084 	return (0);
1085 }
1086 
1087 /*
1088  * Process a received IGMPv3 group-specific or group-and-source-specific
1089  * query.
1090  * Return <0 if any error occurred. Currently this is ignored.
1091  */
1092 static int
1093 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1094     int timer, /*const*/ struct igmpv3 *igmpv3)
1095 {
1096 	int			 retval;
1097 	uint16_t		 nsrc;
1098 
1099 	IN_MULTI_LIST_LOCK_ASSERT();
1100 	IGMP_LOCK_ASSERT();
1101 
1102 	retval = 0;
1103 
1104 	switch (inm->inm_state) {
1105 	case IGMP_NOT_MEMBER:
1106 	case IGMP_SILENT_MEMBER:
1107 	case IGMP_SLEEPING_MEMBER:
1108 	case IGMP_LAZY_MEMBER:
1109 	case IGMP_AWAKENING_MEMBER:
1110 	case IGMP_IDLE_MEMBER:
1111 	case IGMP_LEAVING_MEMBER:
1112 		return (retval);
1113 		break;
1114 	case IGMP_REPORTING_MEMBER:
1115 	case IGMP_G_QUERY_PENDING_MEMBER:
1116 	case IGMP_SG_QUERY_PENDING_MEMBER:
1117 		break;
1118 	}
1119 
1120 	nsrc = ntohs(igmpv3->igmp_numsrc);
1121 
1122 	/*
1123 	 * Deal with group-specific queries upfront.
1124 	 * If any group query is already pending, purge any recorded
1125 	 * source-list state if it exists, and schedule a query response
1126 	 * for this group-specific query.
1127 	 */
1128 	if (nsrc == 0) {
1129 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1130 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1131 			inm_clear_recorded(inm);
1132 			timer = min(inm->inm_timer, timer);
1133 		}
1134 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1135 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1136 		V_current_state_timers_running = 1;
1137 		return (retval);
1138 	}
1139 
1140 	/*
1141 	 * Deal with the case where a group-and-source-specific query has
1142 	 * been received but a group-specific query is already pending.
1143 	 */
1144 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1145 		timer = min(inm->inm_timer, timer);
1146 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1147 		V_current_state_timers_running = 1;
1148 		return (retval);
1149 	}
1150 
1151 	/*
1152 	 * Finally, deal with the case where a group-and-source-specific
1153 	 * query has been received, where a response to a previous g-s-r
1154 	 * query exists, or none exists.
1155 	 * In this case, we need to parse the source-list which the Querier
1156 	 * has provided us with and check if we have any source list filter
1157 	 * entries at T1 for these sources. If we do not, there is no need
1158 	 * schedule a report and the query may be dropped.
1159 	 * If we do, we must record them and schedule a current-state
1160 	 * report for those sources.
1161 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1162 	 * we pass the mbuf chain pointer down to this function, and use
1163 	 * m_getptr() to walk the chain.
1164 	 */
1165 	if (inm->inm_nsrc > 0) {
1166 		const struct in_addr	*ap;
1167 		int			 i, nrecorded;
1168 
1169 		ap = (const struct in_addr *)(igmpv3 + 1);
1170 		nrecorded = 0;
1171 		for (i = 0; i < nsrc; i++, ap++) {
1172 			retval = inm_record_source(inm, ap->s_addr);
1173 			if (retval < 0)
1174 				break;
1175 			nrecorded += retval;
1176 		}
1177 		if (nrecorded > 0) {
1178 			CTR1(KTR_IGMPV3,
1179 			    "%s: schedule response to SG query", __func__);
1180 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1181 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1182 			V_current_state_timers_running = 1;
1183 		}
1184 	}
1185 
1186 	return (retval);
1187 }
1188 
1189 /*
1190  * Process a received IGMPv1 host membership report.
1191  *
1192  * NOTE: 0.0.0.0 workaround breaks const correctness.
1193  */
1194 static int
1195 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1196     /*const*/ struct igmp *igmp)
1197 {
1198 	struct rm_priotracker in_ifa_tracker;
1199 	struct in_ifaddr *ia;
1200 	struct in_multi *inm;
1201 
1202 	IGMPSTAT_INC(igps_rcv_reports);
1203 
1204 	if (ifp->if_flags & IFF_LOOPBACK)
1205 		return (0);
1206 
1207 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1208 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1209 		IGMPSTAT_INC(igps_rcv_badreports);
1210 		return (EINVAL);
1211 	}
1212 
1213 	/*
1214 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1215 	 * Booting clients may use the source address 0.0.0.0. Some
1216 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1217 	 * the interface upon which this message was received.
1218 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1219 	 */
1220 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1221 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1222 		if (ia != NULL) {
1223 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1224 			ifa_free(&ia->ia_ifa);
1225 		}
1226 	}
1227 
1228 	CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1229 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1230 
1231 	/*
1232 	 * IGMPv1 report suppression.
1233 	 * If we are a member of this group, and our membership should be
1234 	 * reported, stop our group timer and transition to the 'lazy' state.
1235 	 */
1236 	IN_MULTI_LIST_LOCK();
1237 	inm = inm_lookup(ifp, igmp->igmp_group);
1238 	if (inm != NULL) {
1239 		struct igmp_ifsoftc *igi;
1240 
1241 		igi = inm->inm_igi;
1242 		if (igi == NULL) {
1243 			KASSERT(igi != NULL,
1244 			    ("%s: no igi for ifp %p", __func__, ifp));
1245 			goto out_locked;
1246 		}
1247 
1248 		IGMPSTAT_INC(igps_rcv_ourreports);
1249 
1250 		/*
1251 		 * If we are in IGMPv3 host mode, do not allow the
1252 		 * other host's IGMPv1 report to suppress our reports
1253 		 * unless explicitly configured to do so.
1254 		 */
1255 		if (igi->igi_version == IGMP_VERSION_3) {
1256 			if (V_igmp_legacysupp)
1257 				igmp_v3_suppress_group_record(inm);
1258 			goto out_locked;
1259 		}
1260 
1261 		inm->inm_timer = 0;
1262 
1263 		switch (inm->inm_state) {
1264 		case IGMP_NOT_MEMBER:
1265 		case IGMP_SILENT_MEMBER:
1266 			break;
1267 		case IGMP_IDLE_MEMBER:
1268 		case IGMP_LAZY_MEMBER:
1269 		case IGMP_AWAKENING_MEMBER:
1270 			CTR3(KTR_IGMPV3,
1271 			    "report suppressed for 0x%08x on ifp %p(%s)",
1272 			    ntohl(igmp->igmp_group.s_addr), ifp,
1273 			    ifp->if_xname);
1274 		case IGMP_SLEEPING_MEMBER:
1275 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1276 			break;
1277 		case IGMP_REPORTING_MEMBER:
1278 			CTR3(KTR_IGMPV3,
1279 			    "report suppressed for 0x%08x on ifp %p(%s)",
1280 			    ntohl(igmp->igmp_group.s_addr), ifp,
1281 			    ifp->if_xname);
1282 			if (igi->igi_version == IGMP_VERSION_1)
1283 				inm->inm_state = IGMP_LAZY_MEMBER;
1284 			else if (igi->igi_version == IGMP_VERSION_2)
1285 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1286 			break;
1287 		case IGMP_G_QUERY_PENDING_MEMBER:
1288 		case IGMP_SG_QUERY_PENDING_MEMBER:
1289 		case IGMP_LEAVING_MEMBER:
1290 			break;
1291 		}
1292 	}
1293 
1294 out_locked:
1295 	IN_MULTI_LIST_UNLOCK();
1296 
1297 	return (0);
1298 }
1299 
1300 /*
1301  * Process a received IGMPv2 host membership report.
1302  *
1303  * NOTE: 0.0.0.0 workaround breaks const correctness.
1304  */
1305 static int
1306 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1307     /*const*/ struct igmp *igmp)
1308 {
1309 	struct rm_priotracker in_ifa_tracker;
1310 	struct in_ifaddr *ia;
1311 	struct in_multi *inm;
1312 
1313 	/*
1314 	 * Make sure we don't hear our own membership report.  Fast
1315 	 * leave requires knowing that we are the only member of a
1316 	 * group.
1317 	 */
1318 	IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1319 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1320 		ifa_free(&ia->ia_ifa);
1321 		return (0);
1322 	}
1323 
1324 	IGMPSTAT_INC(igps_rcv_reports);
1325 
1326 	if (ifp->if_flags & IFF_LOOPBACK) {
1327 		if (ia != NULL)
1328 			ifa_free(&ia->ia_ifa);
1329 		return (0);
1330 	}
1331 
1332 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1333 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1334 		if (ia != NULL)
1335 			ifa_free(&ia->ia_ifa);
1336 		IGMPSTAT_INC(igps_rcv_badreports);
1337 		return (EINVAL);
1338 	}
1339 
1340 	/*
1341 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1342 	 * Booting clients may use the source address 0.0.0.0. Some
1343 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1344 	 * the interface upon which this message was received.
1345 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1346 	 */
1347 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1348 		if (ia != NULL)
1349 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1350 	}
1351 	if (ia != NULL)
1352 		ifa_free(&ia->ia_ifa);
1353 
1354 	CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1355 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1356 
1357 	/*
1358 	 * IGMPv2 report suppression.
1359 	 * If we are a member of this group, and our membership should be
1360 	 * reported, and our group timer is pending or about to be reset,
1361 	 * stop our group timer by transitioning to the 'lazy' state.
1362 	 */
1363 	IN_MULTI_LIST_LOCK();
1364 	inm = inm_lookup(ifp, igmp->igmp_group);
1365 	if (inm != NULL) {
1366 		struct igmp_ifsoftc *igi;
1367 
1368 		igi = inm->inm_igi;
1369 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1370 
1371 		IGMPSTAT_INC(igps_rcv_ourreports);
1372 
1373 		/*
1374 		 * If we are in IGMPv3 host mode, do not allow the
1375 		 * other host's IGMPv1 report to suppress our reports
1376 		 * unless explicitly configured to do so.
1377 		 */
1378 		if (igi->igi_version == IGMP_VERSION_3) {
1379 			if (V_igmp_legacysupp)
1380 				igmp_v3_suppress_group_record(inm);
1381 			goto out_locked;
1382 		}
1383 
1384 		inm->inm_timer = 0;
1385 
1386 		switch (inm->inm_state) {
1387 		case IGMP_NOT_MEMBER:
1388 		case IGMP_SILENT_MEMBER:
1389 		case IGMP_SLEEPING_MEMBER:
1390 			break;
1391 		case IGMP_REPORTING_MEMBER:
1392 		case IGMP_IDLE_MEMBER:
1393 		case IGMP_AWAKENING_MEMBER:
1394 			CTR3(KTR_IGMPV3,
1395 			    "report suppressed for 0x%08x on ifp %p(%s)",
1396 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1397 		case IGMP_LAZY_MEMBER:
1398 			inm->inm_state = IGMP_LAZY_MEMBER;
1399 			break;
1400 		case IGMP_G_QUERY_PENDING_MEMBER:
1401 		case IGMP_SG_QUERY_PENDING_MEMBER:
1402 		case IGMP_LEAVING_MEMBER:
1403 			break;
1404 		}
1405 	}
1406 
1407 out_locked:
1408 	IN_MULTI_LIST_UNLOCK();
1409 
1410 	return (0);
1411 }
1412 
1413 int
1414 igmp_input(struct mbuf **mp, int *offp, int proto)
1415 {
1416 	int iphlen;
1417 	struct ifnet *ifp;
1418 	struct igmp *igmp;
1419 	struct ip *ip;
1420 	struct mbuf *m;
1421 	int igmplen;
1422 	int minlen;
1423 	int queryver;
1424 
1425 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1426 
1427 	m = *mp;
1428 	ifp = m->m_pkthdr.rcvif;
1429 	*mp = NULL;
1430 
1431 	IGMPSTAT_INC(igps_rcv_total);
1432 
1433 	ip = mtod(m, struct ip *);
1434 	iphlen = *offp;
1435 	igmplen = ntohs(ip->ip_len) - iphlen;
1436 
1437 	/*
1438 	 * Validate lengths.
1439 	 */
1440 	if (igmplen < IGMP_MINLEN) {
1441 		IGMPSTAT_INC(igps_rcv_tooshort);
1442 		m_freem(m);
1443 		return (IPPROTO_DONE);
1444 	}
1445 
1446 	/*
1447 	 * Always pullup to the minimum size for v1/v2 or v3
1448 	 * to amortize calls to m_pullup().
1449 	 */
1450 	minlen = iphlen;
1451 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1452 		minlen += IGMP_V3_QUERY_MINLEN;
1453 	else
1454 		minlen += IGMP_MINLEN;
1455 	if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1456 	    (m = m_pullup(m, minlen)) == NULL) {
1457 		IGMPSTAT_INC(igps_rcv_tooshort);
1458 		return (IPPROTO_DONE);
1459 	}
1460 	ip = mtod(m, struct ip *);
1461 
1462 	/*
1463 	 * Validate checksum.
1464 	 */
1465 	m->m_data += iphlen;
1466 	m->m_len -= iphlen;
1467 	igmp = mtod(m, struct igmp *);
1468 	if (in_cksum(m, igmplen)) {
1469 		IGMPSTAT_INC(igps_rcv_badsum);
1470 		m_freem(m);
1471 		return (IPPROTO_DONE);
1472 	}
1473 	m->m_data -= iphlen;
1474 	m->m_len += iphlen;
1475 
1476 	/*
1477 	 * IGMP control traffic is link-scope, and must have a TTL of 1.
1478 	 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1479 	 * probe packets may come from beyond the LAN.
1480 	 */
1481 	if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1482 		IGMPSTAT_INC(igps_rcv_badttl);
1483 		m_freem(m);
1484 		return (IPPROTO_DONE);
1485 	}
1486 
1487 	switch (igmp->igmp_type) {
1488 	case IGMP_HOST_MEMBERSHIP_QUERY:
1489 		if (igmplen == IGMP_MINLEN) {
1490 			if (igmp->igmp_code == 0)
1491 				queryver = IGMP_VERSION_1;
1492 			else
1493 				queryver = IGMP_VERSION_2;
1494 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1495 			queryver = IGMP_VERSION_3;
1496 		} else {
1497 			IGMPSTAT_INC(igps_rcv_tooshort);
1498 			m_freem(m);
1499 			return (IPPROTO_DONE);
1500 		}
1501 
1502 		switch (queryver) {
1503 		case IGMP_VERSION_1:
1504 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1505 			if (!V_igmp_v1enable)
1506 				break;
1507 			if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1508 				m_freem(m);
1509 				return (IPPROTO_DONE);
1510 			}
1511 			break;
1512 
1513 		case IGMP_VERSION_2:
1514 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1515 			if (!V_igmp_v2enable)
1516 				break;
1517 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1518 				m_freem(m);
1519 				return (IPPROTO_DONE);
1520 			}
1521 			break;
1522 
1523 		case IGMP_VERSION_3: {
1524 				struct igmpv3 *igmpv3;
1525 				uint16_t igmpv3len;
1526 				uint16_t nsrc;
1527 
1528 				IGMPSTAT_INC(igps_rcv_v3_queries);
1529 				igmpv3 = (struct igmpv3 *)igmp;
1530 				/*
1531 				 * Validate length based on source count.
1532 				 */
1533 				nsrc = ntohs(igmpv3->igmp_numsrc);
1534 				if (nsrc * sizeof(in_addr_t) >
1535 				    UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1536 					IGMPSTAT_INC(igps_rcv_tooshort);
1537 					return (IPPROTO_DONE);
1538 				}
1539 				/*
1540 				 * m_pullup() may modify m, so pullup in
1541 				 * this scope.
1542 				 */
1543 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1544 				   sizeof(struct in_addr) * nsrc;
1545 				if ((!M_WRITABLE(m) ||
1546 				     m->m_len < igmpv3len) &&
1547 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1548 					IGMPSTAT_INC(igps_rcv_tooshort);
1549 					return (IPPROTO_DONE);
1550 				}
1551 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1552 				    + iphlen);
1553 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1554 					m_freem(m);
1555 					return (IPPROTO_DONE);
1556 				}
1557 			}
1558 			break;
1559 		}
1560 		break;
1561 
1562 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1563 		if (!V_igmp_v1enable)
1564 			break;
1565 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1566 			m_freem(m);
1567 			return (IPPROTO_DONE);
1568 		}
1569 		break;
1570 
1571 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1572 		if (!V_igmp_v2enable)
1573 			break;
1574 		if (!ip_checkrouteralert(m))
1575 			IGMPSTAT_INC(igps_rcv_nora);
1576 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1577 			m_freem(m);
1578 			return (IPPROTO_DONE);
1579 		}
1580 		break;
1581 
1582 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1583 		/*
1584 		 * Hosts do not need to process IGMPv3 membership reports,
1585 		 * as report suppression is no longer required.
1586 		 */
1587 		if (!ip_checkrouteralert(m))
1588 			IGMPSTAT_INC(igps_rcv_nora);
1589 		break;
1590 
1591 	default:
1592 		break;
1593 	}
1594 
1595 	/*
1596 	 * Pass all valid IGMP packets up to any process(es) listening on a
1597 	 * raw IGMP socket.
1598 	 */
1599 	*mp = m;
1600 	return (rip_input(mp, offp, proto));
1601 }
1602 
1603 
1604 /*
1605  * Fast timeout handler (global).
1606  * VIMAGE: Timeout handlers are expected to service all vimages.
1607  */
1608 void
1609 igmp_fasttimo(void)
1610 {
1611 	VNET_ITERATOR_DECL(vnet_iter);
1612 
1613 	VNET_LIST_RLOCK_NOSLEEP();
1614 	VNET_FOREACH(vnet_iter) {
1615 		CURVNET_SET(vnet_iter);
1616 		igmp_fasttimo_vnet();
1617 		CURVNET_RESTORE();
1618 	}
1619 	VNET_LIST_RUNLOCK_NOSLEEP();
1620 }
1621 
1622 /*
1623  * Fast timeout handler (per-vnet).
1624  * Sends are shuffled off to a netisr to deal with Giant.
1625  *
1626  * VIMAGE: Assume caller has set up our curvnet.
1627  */
1628 static void
1629 igmp_fasttimo_vnet(void)
1630 {
1631 	struct mbufq		 scq;	/* State-change packets */
1632 	struct mbufq		 qrq;	/* Query response packets */
1633 	struct ifnet		*ifp;
1634 	struct igmp_ifsoftc	*igi;
1635 	struct ifmultiaddr	*ifma, *next;
1636 	struct in_multi		*inm;
1637 	struct in_multi_head inm_free_tmp;
1638 	int			 loop, uri_fasthz;
1639 
1640 	loop = 0;
1641 	uri_fasthz = 0;
1642 
1643 	/*
1644 	 * Quick check to see if any work needs to be done, in order to
1645 	 * minimize the overhead of fasttimo processing.
1646 	 * SMPng: XXX Unlocked reads.
1647 	 */
1648 	if (!V_current_state_timers_running &&
1649 	    !V_interface_timers_running &&
1650 	    !V_state_change_timers_running)
1651 		return;
1652 
1653 	SLIST_INIT(&inm_free_tmp);
1654 	IN_MULTI_LIST_LOCK();
1655 	IGMP_LOCK();
1656 
1657 	/*
1658 	 * IGMPv3 General Query response timer processing.
1659 	 */
1660 	if (V_interface_timers_running) {
1661 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1662 
1663 		V_interface_timers_running = 0;
1664 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1665 			if (igi->igi_v3_timer == 0) {
1666 				/* Do nothing. */
1667 			} else if (--igi->igi_v3_timer == 0) {
1668 				igmp_v3_dispatch_general_query(igi);
1669 			} else {
1670 				V_interface_timers_running = 1;
1671 			}
1672 		}
1673 	}
1674 
1675 	if (!V_current_state_timers_running &&
1676 	    !V_state_change_timers_running)
1677 		goto out_locked;
1678 
1679 	V_current_state_timers_running = 0;
1680 	V_state_change_timers_running = 0;
1681 
1682 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1683 
1684 	/*
1685 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1686 	 * Note: Processing a v3 group timer may remove a node.
1687 	 */
1688 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1689 		ifp = igi->igi_ifp;
1690 
1691 		if (igi->igi_version == IGMP_VERSION_3) {
1692 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1693 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1694 			    PR_FASTHZ);
1695 			mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1696 			mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1697 		}
1698 
1699 		IF_ADDR_WLOCK(ifp);
1700 	restart:
1701 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1702 			if (ifma->ifma_addr->sa_family != AF_INET ||
1703 			    ifma->ifma_protospec == NULL)
1704 				continue;
1705 			inm = (struct in_multi *)ifma->ifma_protospec;
1706 			switch (igi->igi_version) {
1707 			case IGMP_VERSION_1:
1708 			case IGMP_VERSION_2:
1709 				igmp_v1v2_process_group_timer(inm,
1710 				    igi->igi_version);
1711 				break;
1712 			case IGMP_VERSION_3:
1713 				igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1714 				    &scq, inm, uri_fasthz);
1715 				break;
1716 			}
1717 			if (__predict_false(ifma_restart)) {
1718 				ifma_restart = false;
1719 				goto restart;
1720 			}
1721 		}
1722 		IF_ADDR_WUNLOCK(ifp);
1723 
1724 		if (igi->igi_version == IGMP_VERSION_3) {
1725 			igmp_dispatch_queue(&qrq, 0, loop);
1726 			igmp_dispatch_queue(&scq, 0, loop);
1727 
1728 			/*
1729 			 * Free the in_multi reference(s) for this
1730 			 * IGMP lifecycle.
1731 			 */
1732 			inm_release_list_deferred(&inm_free_tmp);
1733 		}
1734 	}
1735 
1736 out_locked:
1737 	IGMP_UNLOCK();
1738 	IN_MULTI_LIST_UNLOCK();
1739 }
1740 
1741 /*
1742  * Update host report group timer for IGMPv1/v2.
1743  * Will update the global pending timer flags.
1744  */
1745 static void
1746 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1747 {
1748 	int report_timer_expired;
1749 
1750 	IN_MULTI_LIST_LOCK_ASSERT();
1751 	IGMP_LOCK_ASSERT();
1752 
1753 	if (inm->inm_timer == 0) {
1754 		report_timer_expired = 0;
1755 	} else if (--inm->inm_timer == 0) {
1756 		report_timer_expired = 1;
1757 	} else {
1758 		V_current_state_timers_running = 1;
1759 		return;
1760 	}
1761 
1762 	switch (inm->inm_state) {
1763 	case IGMP_NOT_MEMBER:
1764 	case IGMP_SILENT_MEMBER:
1765 	case IGMP_IDLE_MEMBER:
1766 	case IGMP_LAZY_MEMBER:
1767 	case IGMP_SLEEPING_MEMBER:
1768 	case IGMP_AWAKENING_MEMBER:
1769 		break;
1770 	case IGMP_REPORTING_MEMBER:
1771 		if (report_timer_expired) {
1772 			inm->inm_state = IGMP_IDLE_MEMBER;
1773 			(void)igmp_v1v2_queue_report(inm,
1774 			    (version == IGMP_VERSION_2) ?
1775 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1776 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1777 		}
1778 		break;
1779 	case IGMP_G_QUERY_PENDING_MEMBER:
1780 	case IGMP_SG_QUERY_PENDING_MEMBER:
1781 	case IGMP_LEAVING_MEMBER:
1782 		break;
1783 	}
1784 }
1785 
1786 /*
1787  * Update a group's timers for IGMPv3.
1788  * Will update the global pending timer flags.
1789  * Note: Unlocked read from igi.
1790  */
1791 static void
1792 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1793     struct mbufq *qrq, struct mbufq *scq,
1794     struct in_multi *inm, const int uri_fasthz)
1795 {
1796 	int query_response_timer_expired;
1797 	int state_change_retransmit_timer_expired;
1798 
1799 	IN_MULTI_LIST_LOCK_ASSERT();
1800 	IGMP_LOCK_ASSERT();
1801 
1802 	query_response_timer_expired = 0;
1803 	state_change_retransmit_timer_expired = 0;
1804 
1805 	/*
1806 	 * During a transition from v1/v2 compatibility mode back to v3,
1807 	 * a group record in REPORTING state may still have its group
1808 	 * timer active. This is a no-op in this function; it is easier
1809 	 * to deal with it here than to complicate the slow-timeout path.
1810 	 */
1811 	if (inm->inm_timer == 0) {
1812 		query_response_timer_expired = 0;
1813 	} else if (--inm->inm_timer == 0) {
1814 		query_response_timer_expired = 1;
1815 	} else {
1816 		V_current_state_timers_running = 1;
1817 	}
1818 
1819 	if (inm->inm_sctimer == 0) {
1820 		state_change_retransmit_timer_expired = 0;
1821 	} else if (--inm->inm_sctimer == 0) {
1822 		state_change_retransmit_timer_expired = 1;
1823 	} else {
1824 		V_state_change_timers_running = 1;
1825 	}
1826 
1827 	/* We are in fasttimo, so be quick about it. */
1828 	if (!state_change_retransmit_timer_expired &&
1829 	    !query_response_timer_expired)
1830 		return;
1831 
1832 	switch (inm->inm_state) {
1833 	case IGMP_NOT_MEMBER:
1834 	case IGMP_SILENT_MEMBER:
1835 	case IGMP_SLEEPING_MEMBER:
1836 	case IGMP_LAZY_MEMBER:
1837 	case IGMP_AWAKENING_MEMBER:
1838 	case IGMP_IDLE_MEMBER:
1839 		break;
1840 	case IGMP_G_QUERY_PENDING_MEMBER:
1841 	case IGMP_SG_QUERY_PENDING_MEMBER:
1842 		/*
1843 		 * Respond to a previously pending Group-Specific
1844 		 * or Group-and-Source-Specific query by enqueueing
1845 		 * the appropriate Current-State report for
1846 		 * immediate transmission.
1847 		 */
1848 		if (query_response_timer_expired) {
1849 			int retval;
1850 
1851 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1852 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1853 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1854 			    __func__, retval);
1855 			inm->inm_state = IGMP_REPORTING_MEMBER;
1856 			/* XXX Clear recorded sources for next time. */
1857 			inm_clear_recorded(inm);
1858 		}
1859 		/* FALLTHROUGH */
1860 	case IGMP_REPORTING_MEMBER:
1861 	case IGMP_LEAVING_MEMBER:
1862 		if (state_change_retransmit_timer_expired) {
1863 			/*
1864 			 * State-change retransmission timer fired.
1865 			 * If there are any further pending retransmissions,
1866 			 * set the global pending state-change flag, and
1867 			 * reset the timer.
1868 			 */
1869 			if (--inm->inm_scrv > 0) {
1870 				inm->inm_sctimer = uri_fasthz;
1871 				V_state_change_timers_running = 1;
1872 			}
1873 			/*
1874 			 * Retransmit the previously computed state-change
1875 			 * report. If there are no further pending
1876 			 * retransmissions, the mbuf queue will be consumed.
1877 			 * Update T0 state to T1 as we have now sent
1878 			 * a state-change.
1879 			 */
1880 			(void)igmp_v3_merge_state_changes(inm, scq);
1881 
1882 			inm_commit(inm);
1883 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1884 			    ntohl(inm->inm_addr.s_addr),
1885 			    inm->inm_ifp->if_xname);
1886 
1887 			/*
1888 			 * If we are leaving the group for good, make sure
1889 			 * we release IGMP's reference to it.
1890 			 * This release must be deferred using a SLIST,
1891 			 * as we are called from a loop which traverses
1892 			 * the in_ifmultiaddr TAILQ.
1893 			 */
1894 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1895 			    inm->inm_scrv == 0) {
1896 				inm->inm_state = IGMP_NOT_MEMBER;
1897 				inm_rele_locked(inmh, inm);
1898 			}
1899 		}
1900 		break;
1901 	}
1902 }
1903 
1904 
1905 /*
1906  * Suppress a group's pending response to a group or source/group query.
1907  *
1908  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1909  * Do NOT update ST1/ST0 as this operation merely suppresses
1910  * the currently pending group record.
1911  * Do NOT suppress the response to a general query. It is possible but
1912  * it would require adding another state or flag.
1913  */
1914 static void
1915 igmp_v3_suppress_group_record(struct in_multi *inm)
1916 {
1917 
1918 	IN_MULTI_LIST_LOCK_ASSERT();
1919 
1920 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1921 		("%s: not IGMPv3 mode on link", __func__));
1922 
1923 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1924 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1925 		return;
1926 
1927 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1928 		inm_clear_recorded(inm);
1929 
1930 	inm->inm_timer = 0;
1931 	inm->inm_state = IGMP_REPORTING_MEMBER;
1932 }
1933 
1934 /*
1935  * Switch to a different IGMP version on the given interface,
1936  * as per Section 7.2.1.
1937  */
1938 static void
1939 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1940 {
1941 	int old_version_timer;
1942 
1943 	IGMP_LOCK_ASSERT();
1944 
1945 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1946 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1947 
1948 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1949 		/*
1950 		 * Compute the "Older Version Querier Present" timer as per
1951 		 * Section 8.12.
1952 		 */
1953 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1954 		old_version_timer *= PR_SLOWHZ;
1955 
1956 		if (version == IGMP_VERSION_1) {
1957 			igi->igi_v1_timer = old_version_timer;
1958 			igi->igi_v2_timer = 0;
1959 		} else if (version == IGMP_VERSION_2) {
1960 			igi->igi_v1_timer = 0;
1961 			igi->igi_v2_timer = old_version_timer;
1962 		}
1963 	}
1964 
1965 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1966 		if (igi->igi_version != IGMP_VERSION_2) {
1967 			igi->igi_version = IGMP_VERSION_2;
1968 			igmp_v3_cancel_link_timers(igi);
1969 		}
1970 	} else if (igi->igi_v1_timer > 0) {
1971 		if (igi->igi_version != IGMP_VERSION_1) {
1972 			igi->igi_version = IGMP_VERSION_1;
1973 			igmp_v3_cancel_link_timers(igi);
1974 		}
1975 	}
1976 }
1977 
1978 /*
1979  * Cancel pending IGMPv3 timers for the given link and all groups
1980  * joined on it; state-change, general-query, and group-query timers.
1981  *
1982  * Only ever called on a transition from v3 to Compatibility mode. Kill
1983  * the timers stone dead (this may be expensive for large N groups), they
1984  * will be restarted if Compatibility Mode deems that they must be due to
1985  * query processing.
1986  */
1987 static void
1988 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
1989 {
1990 	struct ifmultiaddr	*ifma;
1991 	struct ifnet		*ifp;
1992 	struct in_multi		*inm;
1993 	struct in_multi_head inm_free_tmp;
1994 
1995 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
1996 	    igi->igi_ifp, igi->igi_ifp->if_xname);
1997 
1998 	IN_MULTI_LIST_LOCK_ASSERT();
1999 	IGMP_LOCK_ASSERT();
2000 	SLIST_INIT(&inm_free_tmp);
2001 
2002 	/*
2003 	 * Stop the v3 General Query Response on this link stone dead.
2004 	 * If fasttimo is woken up due to V_interface_timers_running,
2005 	 * the flag will be cleared if there are no pending link timers.
2006 	 */
2007 	igi->igi_v3_timer = 0;
2008 
2009 	/*
2010 	 * Now clear the current-state and state-change report timers
2011 	 * for all memberships scoped to this link.
2012 	 */
2013 	ifp = igi->igi_ifp;
2014 	IF_ADDR_RLOCK(ifp);
2015 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2016 		if (ifma->ifma_addr->sa_family != AF_INET ||
2017 		    ifma->ifma_protospec == NULL)
2018 			continue;
2019 		inm = (struct in_multi *)ifma->ifma_protospec;
2020 		switch (inm->inm_state) {
2021 		case IGMP_NOT_MEMBER:
2022 		case IGMP_SILENT_MEMBER:
2023 		case IGMP_IDLE_MEMBER:
2024 		case IGMP_LAZY_MEMBER:
2025 		case IGMP_SLEEPING_MEMBER:
2026 		case IGMP_AWAKENING_MEMBER:
2027 			/*
2028 			 * These states are either not relevant in v3 mode,
2029 			 * or are unreported. Do nothing.
2030 			 */
2031 			break;
2032 		case IGMP_LEAVING_MEMBER:
2033 			/*
2034 			 * If we are leaving the group and switching to
2035 			 * compatibility mode, we need to release the final
2036 			 * reference held for issuing the INCLUDE {}, and
2037 			 * transition to REPORTING to ensure the host leave
2038 			 * message is sent upstream to the old querier --
2039 			 * transition to NOT would lose the leave and race.
2040 			 */
2041 			inm_rele_locked(&inm_free_tmp, inm);
2042 			/* FALLTHROUGH */
2043 		case IGMP_G_QUERY_PENDING_MEMBER:
2044 		case IGMP_SG_QUERY_PENDING_MEMBER:
2045 			inm_clear_recorded(inm);
2046 			/* FALLTHROUGH */
2047 		case IGMP_REPORTING_MEMBER:
2048 			inm->inm_state = IGMP_REPORTING_MEMBER;
2049 			break;
2050 		}
2051 		/*
2052 		 * Always clear state-change and group report timers.
2053 		 * Free any pending IGMPv3 state-change records.
2054 		 */
2055 		inm->inm_sctimer = 0;
2056 		inm->inm_timer = 0;
2057 		mbufq_drain(&inm->inm_scq);
2058 	}
2059 	IF_ADDR_RUNLOCK(ifp);
2060 
2061 	inm_release_list_deferred(&inm_free_tmp);
2062 }
2063 
2064 /*
2065  * Update the Older Version Querier Present timers for a link.
2066  * See Section 7.2.1 of RFC 3376.
2067  */
2068 static void
2069 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2070 {
2071 
2072 	IGMP_LOCK_ASSERT();
2073 
2074 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2075 		/*
2076 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2077 		 *
2078 		 * Revert to IGMPv3.
2079 		 */
2080 		if (igi->igi_version != IGMP_VERSION_3) {
2081 			CTR5(KTR_IGMPV3,
2082 			    "%s: transition from v%d -> v%d on %p(%s)",
2083 			    __func__, igi->igi_version, IGMP_VERSION_3,
2084 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2085 			igi->igi_version = IGMP_VERSION_3;
2086 		}
2087 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2088 		/*
2089 		 * IGMPv1 Querier Present timer expired,
2090 		 * IGMPv2 Querier Present timer running.
2091 		 * If IGMPv2 was disabled since last timeout,
2092 		 * revert to IGMPv3.
2093 		 * If IGMPv2 is enabled, revert to IGMPv2.
2094 		 */
2095 		if (!V_igmp_v2enable) {
2096 			CTR5(KTR_IGMPV3,
2097 			    "%s: transition from v%d -> v%d on %p(%s)",
2098 			    __func__, igi->igi_version, IGMP_VERSION_3,
2099 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2100 			igi->igi_v2_timer = 0;
2101 			igi->igi_version = IGMP_VERSION_3;
2102 		} else {
2103 			--igi->igi_v2_timer;
2104 			if (igi->igi_version != IGMP_VERSION_2) {
2105 				CTR5(KTR_IGMPV3,
2106 				    "%s: transition from v%d -> v%d on %p(%s)",
2107 				    __func__, igi->igi_version, IGMP_VERSION_2,
2108 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2109 				igi->igi_version = IGMP_VERSION_2;
2110 				igmp_v3_cancel_link_timers(igi);
2111 			}
2112 		}
2113 	} else if (igi->igi_v1_timer > 0) {
2114 		/*
2115 		 * IGMPv1 Querier Present timer running.
2116 		 * Stop IGMPv2 timer if running.
2117 		 *
2118 		 * If IGMPv1 was disabled since last timeout,
2119 		 * revert to IGMPv3.
2120 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2121 		 */
2122 		if (!V_igmp_v1enable) {
2123 			CTR5(KTR_IGMPV3,
2124 			    "%s: transition from v%d -> v%d on %p(%s)",
2125 			    __func__, igi->igi_version, IGMP_VERSION_3,
2126 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2127 			igi->igi_v1_timer = 0;
2128 			igi->igi_version = IGMP_VERSION_3;
2129 		} else {
2130 			--igi->igi_v1_timer;
2131 		}
2132 		if (igi->igi_v2_timer > 0) {
2133 			CTR3(KTR_IGMPV3,
2134 			    "%s: cancel v2 timer on %p(%s)",
2135 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2136 			igi->igi_v2_timer = 0;
2137 		}
2138 	}
2139 }
2140 
2141 /*
2142  * Global slowtimo handler.
2143  * VIMAGE: Timeout handlers are expected to service all vimages.
2144  */
2145 void
2146 igmp_slowtimo(void)
2147 {
2148 	VNET_ITERATOR_DECL(vnet_iter);
2149 
2150 	VNET_LIST_RLOCK_NOSLEEP();
2151 	VNET_FOREACH(vnet_iter) {
2152 		CURVNET_SET(vnet_iter);
2153 		igmp_slowtimo_vnet();
2154 		CURVNET_RESTORE();
2155 	}
2156 	VNET_LIST_RUNLOCK_NOSLEEP();
2157 }
2158 
2159 /*
2160  * Per-vnet slowtimo handler.
2161  */
2162 static void
2163 igmp_slowtimo_vnet(void)
2164 {
2165 	struct igmp_ifsoftc *igi;
2166 
2167 	IGMP_LOCK();
2168 
2169 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2170 		igmp_v1v2_process_querier_timers(igi);
2171 	}
2172 
2173 	IGMP_UNLOCK();
2174 }
2175 
2176 /*
2177  * Dispatch an IGMPv1/v2 host report or leave message.
2178  * These are always small enough to fit inside a single mbuf.
2179  */
2180 static int
2181 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2182 {
2183 	struct ifnet		*ifp;
2184 	struct igmp		*igmp;
2185 	struct ip		*ip;
2186 	struct mbuf		*m;
2187 
2188 	IN_MULTI_LIST_LOCK_ASSERT();
2189 	IGMP_LOCK_ASSERT();
2190 
2191 	ifp = inm->inm_ifp;
2192 
2193 	m = m_gethdr(M_NOWAIT, MT_DATA);
2194 	if (m == NULL)
2195 		return (ENOMEM);
2196 	M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2197 
2198 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2199 
2200 	m->m_data += sizeof(struct ip);
2201 	m->m_len = sizeof(struct igmp);
2202 
2203 	igmp = mtod(m, struct igmp *);
2204 	igmp->igmp_type = type;
2205 	igmp->igmp_code = 0;
2206 	igmp->igmp_group = inm->inm_addr;
2207 	igmp->igmp_cksum = 0;
2208 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2209 
2210 	m->m_data -= sizeof(struct ip);
2211 	m->m_len += sizeof(struct ip);
2212 
2213 	ip = mtod(m, struct ip *);
2214 	ip->ip_tos = 0;
2215 	ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2216 	ip->ip_off = 0;
2217 	ip->ip_p = IPPROTO_IGMP;
2218 	ip->ip_src.s_addr = INADDR_ANY;
2219 
2220 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2221 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2222 	else
2223 		ip->ip_dst = inm->inm_addr;
2224 
2225 	igmp_save_context(m, ifp);
2226 
2227 	m->m_flags |= M_IGMPV2;
2228 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2229 		m->m_flags |= M_IGMP_LOOP;
2230 
2231 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2232 	netisr_dispatch(NETISR_IGMP, m);
2233 
2234 	return (0);
2235 }
2236 
2237 /*
2238  * Process a state change from the upper layer for the given IPv4 group.
2239  *
2240  * Each socket holds a reference on the in_multi in its own ip_moptions.
2241  * The socket layer will have made the necessary updates to.the group
2242  * state, it is now up to IGMP to issue a state change report if there
2243  * has been any change between T0 (when the last state-change was issued)
2244  * and T1 (now).
2245  *
2246  * We use the IGMPv3 state machine at group level. The IGMP module
2247  * however makes the decision as to which IGMP protocol version to speak.
2248  * A state change *from* INCLUDE {} always means an initial join.
2249  * A state change *to* INCLUDE {} always means a final leave.
2250  *
2251  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2252  * save ourselves a bunch of work; any exclusive mode groups need not
2253  * compute source filter lists.
2254  *
2255  * VIMAGE: curvnet should have been set by caller, as this routine
2256  * is called from the socket option handlers.
2257  */
2258 int
2259 igmp_change_state(struct in_multi *inm)
2260 {
2261 	struct igmp_ifsoftc *igi;
2262 	struct ifnet *ifp;
2263 	int error;
2264 
2265 	error = 0;
2266 	IN_MULTI_LOCK_ASSERT();
2267 	/*
2268 	 * Try to detect if the upper layer just asked us to change state
2269 	 * for an interface which has now gone away.
2270 	 */
2271 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2272 	ifp = inm->inm_ifma->ifma_ifp;
2273 	/*
2274 	 * Sanity check that netinet's notion of ifp is the
2275 	 * same as net's.
2276 	 */
2277 	KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2278 
2279 	IGMP_LOCK();
2280 
2281 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2282 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2283 
2284 	/*
2285 	 * If we detect a state transition to or from MCAST_UNDEFINED
2286 	 * for this group, then we are starting or finishing an IGMP
2287 	 * life cycle for this group.
2288 	 */
2289 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2290 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2291 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2292 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2293 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2294 			error = igmp_initial_join(inm, igi);
2295 			goto out_locked;
2296 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2297 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2298 			igmp_final_leave(inm, igi);
2299 			goto out_locked;
2300 		}
2301 	} else {
2302 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2303 	}
2304 
2305 	error = igmp_handle_state_change(inm, igi);
2306 
2307 out_locked:
2308 	IGMP_UNLOCK();
2309 	return (error);
2310 }
2311 
2312 /*
2313  * Perform the initial join for an IGMP group.
2314  *
2315  * When joining a group:
2316  *  If the group should have its IGMP traffic suppressed, do nothing.
2317  *  IGMPv1 starts sending IGMPv1 host membership reports.
2318  *  IGMPv2 starts sending IGMPv2 host membership reports.
2319  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2320  *  initial state of the membership.
2321  */
2322 static int
2323 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2324 {
2325 	struct ifnet		*ifp;
2326 	struct mbufq		*mq;
2327 	int			 error, retval, syncstates;
2328 
2329 	CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2330 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2331 
2332 	error = 0;
2333 	syncstates = 1;
2334 
2335 	ifp = inm->inm_ifp;
2336 
2337 	IN_MULTI_LOCK_ASSERT();
2338 	IGMP_LOCK_ASSERT();
2339 
2340 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2341 
2342 	/*
2343 	 * Groups joined on loopback or marked as 'not reported',
2344 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2345 	 * are never reported in any IGMP protocol exchanges.
2346 	 * All other groups enter the appropriate IGMP state machine
2347 	 * for the version in use on this link.
2348 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2349 	 * disabled for the link.
2350 	 */
2351 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2352 	    (igi->igi_flags & IGIF_SILENT) ||
2353 	    !igmp_isgroupreported(inm->inm_addr)) {
2354 		CTR1(KTR_IGMPV3,
2355 "%s: not kicking state machine for silent group", __func__);
2356 		inm->inm_state = IGMP_SILENT_MEMBER;
2357 		inm->inm_timer = 0;
2358 	} else {
2359 		/*
2360 		 * Deal with overlapping in_multi lifecycle.
2361 		 * If this group was LEAVING, then make sure
2362 		 * we drop the reference we picked up to keep the
2363 		 * group around for the final INCLUDE {} enqueue.
2364 		 */
2365 		if (igi->igi_version == IGMP_VERSION_3 &&
2366 		    inm->inm_state == IGMP_LEAVING_MEMBER) {
2367 			MPASS(inm->inm_refcount > 1);
2368 			inm_rele_locked(NULL, inm);
2369 		}
2370 		inm->inm_state = IGMP_REPORTING_MEMBER;
2371 
2372 		switch (igi->igi_version) {
2373 		case IGMP_VERSION_1:
2374 		case IGMP_VERSION_2:
2375 			inm->inm_state = IGMP_IDLE_MEMBER;
2376 			error = igmp_v1v2_queue_report(inm,
2377 			    (igi->igi_version == IGMP_VERSION_2) ?
2378 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2379 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2380 			if (error == 0) {
2381 				inm->inm_timer = IGMP_RANDOM_DELAY(
2382 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2383 				V_current_state_timers_running = 1;
2384 			}
2385 			break;
2386 
2387 		case IGMP_VERSION_3:
2388 			/*
2389 			 * Defer update of T0 to T1, until the first copy
2390 			 * of the state change has been transmitted.
2391 			 */
2392 			syncstates = 0;
2393 
2394 			/*
2395 			 * Immediately enqueue a State-Change Report for
2396 			 * this interface, freeing any previous reports.
2397 			 * Don't kick the timers if there is nothing to do,
2398 			 * or if an error occurred.
2399 			 */
2400 			mq = &inm->inm_scq;
2401 			mbufq_drain(mq);
2402 			retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2403 			    0, 0);
2404 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2405 			    __func__, retval);
2406 			if (retval <= 0) {
2407 				error = retval * -1;
2408 				break;
2409 			}
2410 
2411 			/*
2412 			 * Schedule transmission of pending state-change
2413 			 * report up to RV times for this link. The timer
2414 			 * will fire at the next igmp_fasttimo (~200ms),
2415 			 * giving us an opportunity to merge the reports.
2416 			 */
2417 			if (igi->igi_flags & IGIF_LOOPBACK) {
2418 				inm->inm_scrv = 1;
2419 			} else {
2420 				KASSERT(igi->igi_rv > 1,
2421 				   ("%s: invalid robustness %d", __func__,
2422 				    igi->igi_rv));
2423 				inm->inm_scrv = igi->igi_rv;
2424 			}
2425 			inm->inm_sctimer = 1;
2426 			V_state_change_timers_running = 1;
2427 
2428 			error = 0;
2429 			break;
2430 		}
2431 	}
2432 
2433 	/*
2434 	 * Only update the T0 state if state change is atomic,
2435 	 * i.e. we don't need to wait for a timer to fire before we
2436 	 * can consider the state change to have been communicated.
2437 	 */
2438 	if (syncstates) {
2439 		inm_commit(inm);
2440 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2441 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2442 	}
2443 
2444 	return (error);
2445 }
2446 
2447 /*
2448  * Issue an intermediate state change during the IGMP life-cycle.
2449  */
2450 static int
2451 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2452 {
2453 	struct ifnet		*ifp;
2454 	int			 retval;
2455 
2456 	CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2457 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2458 
2459 	ifp = inm->inm_ifp;
2460 
2461 	IN_MULTI_LIST_LOCK_ASSERT();
2462 	IGMP_LOCK_ASSERT();
2463 
2464 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2465 
2466 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2467 	    (igi->igi_flags & IGIF_SILENT) ||
2468 	    !igmp_isgroupreported(inm->inm_addr) ||
2469 	    (igi->igi_version != IGMP_VERSION_3)) {
2470 		if (!igmp_isgroupreported(inm->inm_addr)) {
2471 			CTR1(KTR_IGMPV3,
2472 "%s: not kicking state machine for silent group", __func__);
2473 		}
2474 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2475 		inm_commit(inm);
2476 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2477 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2478 		return (0);
2479 	}
2480 
2481 	mbufq_drain(&inm->inm_scq);
2482 
2483 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2484 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2485 	if (retval <= 0)
2486 		return (-retval);
2487 
2488 	/*
2489 	 * If record(s) were enqueued, start the state-change
2490 	 * report timer for this group.
2491 	 */
2492 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2493 	inm->inm_sctimer = 1;
2494 	V_state_change_timers_running = 1;
2495 
2496 	return (0);
2497 }
2498 
2499 /*
2500  * Perform the final leave for an IGMP group.
2501  *
2502  * When leaving a group:
2503  *  IGMPv1 does nothing.
2504  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2505  *  IGMPv3 enqueues a state-change report containing a transition
2506  *  to INCLUDE {} for immediate transmission.
2507  */
2508 static void
2509 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2510 {
2511 	int syncstates;
2512 
2513 	syncstates = 1;
2514 
2515 	CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2516 	    __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2517 	    inm->inm_ifp->if_xname);
2518 
2519 	IN_MULTI_LIST_LOCK_ASSERT();
2520 	IGMP_LOCK_ASSERT();
2521 
2522 	switch (inm->inm_state) {
2523 	case IGMP_NOT_MEMBER:
2524 	case IGMP_SILENT_MEMBER:
2525 	case IGMP_LEAVING_MEMBER:
2526 		/* Already leaving or left; do nothing. */
2527 		CTR1(KTR_IGMPV3,
2528 "%s: not kicking state machine for silent group", __func__);
2529 		break;
2530 	case IGMP_REPORTING_MEMBER:
2531 	case IGMP_IDLE_MEMBER:
2532 	case IGMP_G_QUERY_PENDING_MEMBER:
2533 	case IGMP_SG_QUERY_PENDING_MEMBER:
2534 		if (igi->igi_version == IGMP_VERSION_2) {
2535 #ifdef INVARIANTS
2536 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2537 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2538 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2539 			     __func__);
2540 #endif
2541 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2542 			inm->inm_state = IGMP_NOT_MEMBER;
2543 		} else if (igi->igi_version == IGMP_VERSION_3) {
2544 			/*
2545 			 * Stop group timer and all pending reports.
2546 			 * Immediately enqueue a state-change report
2547 			 * TO_IN {} to be sent on the next fast timeout,
2548 			 * giving us an opportunity to merge reports.
2549 			 */
2550 			mbufq_drain(&inm->inm_scq);
2551 			inm->inm_timer = 0;
2552 			if (igi->igi_flags & IGIF_LOOPBACK) {
2553 				inm->inm_scrv = 1;
2554 			} else {
2555 				inm->inm_scrv = igi->igi_rv;
2556 			}
2557 			CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2558 			    "pending retransmissions.", __func__,
2559 			    ntohl(inm->inm_addr.s_addr),
2560 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2561 			if (inm->inm_scrv == 0) {
2562 				inm->inm_state = IGMP_NOT_MEMBER;
2563 				inm->inm_sctimer = 0;
2564 			} else {
2565 				int retval;
2566 
2567 				inm_acquire_locked(inm);
2568 
2569 				retval = igmp_v3_enqueue_group_record(
2570 				    &inm->inm_scq, inm, 1, 0, 0);
2571 				KASSERT(retval != 0,
2572 				    ("%s: enqueue record = %d", __func__,
2573 				     retval));
2574 
2575 				inm->inm_state = IGMP_LEAVING_MEMBER;
2576 				inm->inm_sctimer = 1;
2577 				V_state_change_timers_running = 1;
2578 				syncstates = 0;
2579 			}
2580 			break;
2581 		}
2582 		break;
2583 	case IGMP_LAZY_MEMBER:
2584 	case IGMP_SLEEPING_MEMBER:
2585 	case IGMP_AWAKENING_MEMBER:
2586 		/* Our reports are suppressed; do nothing. */
2587 		break;
2588 	}
2589 
2590 	if (syncstates) {
2591 		inm_commit(inm);
2592 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2593 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2594 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2595 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2596 		    __func__, ntohl(inm->inm_addr.s_addr),
2597 		    inm->inm_ifp->if_xname);
2598 	}
2599 }
2600 
2601 /*
2602  * Enqueue an IGMPv3 group record to the given output queue.
2603  *
2604  * XXX This function could do with having the allocation code
2605  * split out, and the multiple-tree-walks coalesced into a single
2606  * routine as has been done in igmp_v3_enqueue_filter_change().
2607  *
2608  * If is_state_change is zero, a current-state record is appended.
2609  * If is_state_change is non-zero, a state-change report is appended.
2610  *
2611  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2612  * If is_group_query is zero, and if there is a packet with free space
2613  * at the tail of the queue, it will be appended to providing there
2614  * is enough free space.
2615  * Otherwise a new mbuf packet chain is allocated.
2616  *
2617  * If is_source_query is non-zero, each source is checked to see if
2618  * it was recorded for a Group-Source query, and will be omitted if
2619  * it is not both in-mode and recorded.
2620  *
2621  * The function will attempt to allocate leading space in the packet
2622  * for the IP/IGMP header to be prepended without fragmenting the chain.
2623  *
2624  * If successful the size of all data appended to the queue is returned,
2625  * otherwise an error code less than zero is returned, or zero if
2626  * no record(s) were appended.
2627  */
2628 static int
2629 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2630     const int is_state_change, const int is_group_query,
2631     const int is_source_query)
2632 {
2633 	struct igmp_grouprec	 ig;
2634 	struct igmp_grouprec	*pig;
2635 	struct ifnet		*ifp;
2636 	struct ip_msource	*ims, *nims;
2637 	struct mbuf		*m0, *m, *md;
2638 	int			 error, is_filter_list_change;
2639 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2640 	int			 record_has_sources;
2641 	int			 now;
2642 	int			 type;
2643 	in_addr_t		 naddr;
2644 	uint8_t			 mode;
2645 
2646 	IN_MULTI_LIST_LOCK_ASSERT();
2647 
2648 	error = 0;
2649 	ifp = inm->inm_ifp;
2650 	is_filter_list_change = 0;
2651 	m = NULL;
2652 	m0 = NULL;
2653 	m0srcs = 0;
2654 	msrcs = 0;
2655 	nbytes = 0;
2656 	nims = NULL;
2657 	record_has_sources = 1;
2658 	pig = NULL;
2659 	type = IGMP_DO_NOTHING;
2660 	mode = inm->inm_st[1].iss_fmode;
2661 
2662 	/*
2663 	 * If we did not transition out of ASM mode during t0->t1,
2664 	 * and there are no source nodes to process, we can skip
2665 	 * the generation of source records.
2666 	 */
2667 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2668 	    inm->inm_nsrc == 0)
2669 		record_has_sources = 0;
2670 
2671 	if (is_state_change) {
2672 		/*
2673 		 * Queue a state change record.
2674 		 * If the mode did not change, and there are non-ASM
2675 		 * listeners or source filters present,
2676 		 * we potentially need to issue two records for the group.
2677 		 * If we are transitioning to MCAST_UNDEFINED, we need
2678 		 * not send any sources.
2679 		 * If there are ASM listeners, and there was no filter
2680 		 * mode transition of any kind, do nothing.
2681 		 */
2682 		if (mode != inm->inm_st[0].iss_fmode) {
2683 			if (mode == MCAST_EXCLUDE) {
2684 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2685 				    __func__);
2686 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2687 			} else {
2688 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2689 				    __func__);
2690 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2691 				if (mode == MCAST_UNDEFINED)
2692 					record_has_sources = 0;
2693 			}
2694 		} else {
2695 			if (record_has_sources) {
2696 				is_filter_list_change = 1;
2697 			} else {
2698 				type = IGMP_DO_NOTHING;
2699 			}
2700 		}
2701 	} else {
2702 		/*
2703 		 * Queue a current state record.
2704 		 */
2705 		if (mode == MCAST_EXCLUDE) {
2706 			type = IGMP_MODE_IS_EXCLUDE;
2707 		} else if (mode == MCAST_INCLUDE) {
2708 			type = IGMP_MODE_IS_INCLUDE;
2709 			KASSERT(inm->inm_st[1].iss_asm == 0,
2710 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2711 			     __func__, inm, inm->inm_st[1].iss_asm));
2712 		}
2713 	}
2714 
2715 	/*
2716 	 * Generate the filter list changes using a separate function.
2717 	 */
2718 	if (is_filter_list_change)
2719 		return (igmp_v3_enqueue_filter_change(mq, inm));
2720 
2721 	if (type == IGMP_DO_NOTHING) {
2722 		CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2723 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2724 		return (0);
2725 	}
2726 
2727 	/*
2728 	 * If any sources are present, we must be able to fit at least
2729 	 * one in the trailing space of the tail packet's mbuf,
2730 	 * ideally more.
2731 	 */
2732 	minrec0len = sizeof(struct igmp_grouprec);
2733 	if (record_has_sources)
2734 		minrec0len += sizeof(in_addr_t);
2735 
2736 	CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2737 	    igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2738 	    inm->inm_ifp->if_xname);
2739 
2740 	/*
2741 	 * Check if we have a packet in the tail of the queue for this
2742 	 * group into which the first group record for this group will fit.
2743 	 * Otherwise allocate a new packet.
2744 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2745 	 * Note: Group records for G/GSR query responses MUST be sent
2746 	 * in their own packet.
2747 	 */
2748 	m0 = mbufq_last(mq);
2749 	if (!is_group_query &&
2750 	    m0 != NULL &&
2751 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2752 	    (m0->m_pkthdr.len + minrec0len) <
2753 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2754 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2755 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2756 		m = m0;
2757 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2758 	} else {
2759 		if (mbufq_full(mq)) {
2760 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2761 			return (-ENOMEM);
2762 		}
2763 		m = NULL;
2764 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2765 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2766 		if (!is_state_change && !is_group_query) {
2767 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2768 			if (m)
2769 				m->m_data += IGMP_LEADINGSPACE;
2770 		}
2771 		if (m == NULL) {
2772 			m = m_gethdr(M_NOWAIT, MT_DATA);
2773 			if (m)
2774 				M_ALIGN(m, IGMP_LEADINGSPACE);
2775 		}
2776 		if (m == NULL)
2777 			return (-ENOMEM);
2778 
2779 		igmp_save_context(m, ifp);
2780 
2781 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2782 	}
2783 
2784 	/*
2785 	 * Append group record.
2786 	 * If we have sources, we don't know how many yet.
2787 	 */
2788 	ig.ig_type = type;
2789 	ig.ig_datalen = 0;
2790 	ig.ig_numsrc = 0;
2791 	ig.ig_group = inm->inm_addr;
2792 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2793 		if (m != m0)
2794 			m_freem(m);
2795 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2796 		return (-ENOMEM);
2797 	}
2798 	nbytes += sizeof(struct igmp_grouprec);
2799 
2800 	/*
2801 	 * Append as many sources as will fit in the first packet.
2802 	 * If we are appending to a new packet, the chain allocation
2803 	 * may potentially use clusters; use m_getptr() in this case.
2804 	 * If we are appending to an existing packet, we need to obtain
2805 	 * a pointer to the group record after m_append(), in case a new
2806 	 * mbuf was allocated.
2807 	 * Only append sources which are in-mode at t1. If we are
2808 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2809 	 * include source entries.
2810 	 * Only report recorded sources in our filter set when responding
2811 	 * to a group-source query.
2812 	 */
2813 	if (record_has_sources) {
2814 		if (m == m0) {
2815 			md = m_last(m);
2816 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2817 			    md->m_len - nbytes);
2818 		} else {
2819 			md = m_getptr(m, 0, &off);
2820 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2821 			    off);
2822 		}
2823 		msrcs = 0;
2824 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2825 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2826 			    ims->ims_haddr);
2827 			now = ims_get_mode(inm, ims, 1);
2828 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2829 			if ((now != mode) ||
2830 			    (now == mode && mode == MCAST_UNDEFINED)) {
2831 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2832 				continue;
2833 			}
2834 			if (is_source_query && ims->ims_stp == 0) {
2835 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2836 				    __func__);
2837 				continue;
2838 			}
2839 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2840 			naddr = htonl(ims->ims_haddr);
2841 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2842 				if (m != m0)
2843 					m_freem(m);
2844 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2845 				    __func__);
2846 				return (-ENOMEM);
2847 			}
2848 			nbytes += sizeof(in_addr_t);
2849 			++msrcs;
2850 			if (msrcs == m0srcs)
2851 				break;
2852 		}
2853 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2854 		    msrcs);
2855 		pig->ig_numsrc = htons(msrcs);
2856 		nbytes += (msrcs * sizeof(in_addr_t));
2857 	}
2858 
2859 	if (is_source_query && msrcs == 0) {
2860 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2861 		if (m != m0)
2862 			m_freem(m);
2863 		return (0);
2864 	}
2865 
2866 	/*
2867 	 * We are good to go with first packet.
2868 	 */
2869 	if (m != m0) {
2870 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2871 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2872 		mbufq_enqueue(mq, m);
2873 	} else
2874 		m->m_pkthdr.PH_vt.vt_nrecs++;
2875 
2876 	/*
2877 	 * No further work needed if no source list in packet(s).
2878 	 */
2879 	if (!record_has_sources)
2880 		return (nbytes);
2881 
2882 	/*
2883 	 * Whilst sources remain to be announced, we need to allocate
2884 	 * a new packet and fill out as many sources as will fit.
2885 	 * Always try for a cluster first.
2886 	 */
2887 	while (nims != NULL) {
2888 		if (mbufq_full(mq)) {
2889 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2890 			return (-ENOMEM);
2891 		}
2892 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2893 		if (m)
2894 			m->m_data += IGMP_LEADINGSPACE;
2895 		if (m == NULL) {
2896 			m = m_gethdr(M_NOWAIT, MT_DATA);
2897 			if (m)
2898 				M_ALIGN(m, IGMP_LEADINGSPACE);
2899 		}
2900 		if (m == NULL)
2901 			return (-ENOMEM);
2902 		igmp_save_context(m, ifp);
2903 		md = m_getptr(m, 0, &off);
2904 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2905 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2906 
2907 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2908 			if (m != m0)
2909 				m_freem(m);
2910 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2911 			return (-ENOMEM);
2912 		}
2913 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2914 		nbytes += sizeof(struct igmp_grouprec);
2915 
2916 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2917 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2918 
2919 		msrcs = 0;
2920 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2921 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2922 			    ims->ims_haddr);
2923 			now = ims_get_mode(inm, ims, 1);
2924 			if ((now != mode) ||
2925 			    (now == mode && mode == MCAST_UNDEFINED)) {
2926 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2927 				continue;
2928 			}
2929 			if (is_source_query && ims->ims_stp == 0) {
2930 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2931 				    __func__);
2932 				continue;
2933 			}
2934 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2935 			naddr = htonl(ims->ims_haddr);
2936 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2937 				if (m != m0)
2938 					m_freem(m);
2939 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2940 				    __func__);
2941 				return (-ENOMEM);
2942 			}
2943 			++msrcs;
2944 			if (msrcs == m0srcs)
2945 				break;
2946 		}
2947 		pig->ig_numsrc = htons(msrcs);
2948 		nbytes += (msrcs * sizeof(in_addr_t));
2949 
2950 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2951 		mbufq_enqueue(mq, m);
2952 	}
2953 
2954 	return (nbytes);
2955 }
2956 
2957 /*
2958  * Type used to mark record pass completion.
2959  * We exploit the fact we can cast to this easily from the
2960  * current filter modes on each ip_msource node.
2961  */
2962 typedef enum {
2963 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2964 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2965 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2966 	REC_FULL = REC_ALLOW | REC_BLOCK
2967 } rectype_t;
2968 
2969 /*
2970  * Enqueue an IGMPv3 filter list change to the given output queue.
2971  *
2972  * Source list filter state is held in an RB-tree. When the filter list
2973  * for a group is changed without changing its mode, we need to compute
2974  * the deltas between T0 and T1 for each source in the filter set,
2975  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2976  *
2977  * As we may potentially queue two record types, and the entire R-B tree
2978  * needs to be walked at once, we break this out into its own function
2979  * so we can generate a tightly packed queue of packets.
2980  *
2981  * XXX This could be written to only use one tree walk, although that makes
2982  * serializing into the mbuf chains a bit harder. For now we do two walks
2983  * which makes things easier on us, and it may or may not be harder on
2984  * the L2 cache.
2985  *
2986  * If successful the size of all data appended to the queue is returned,
2987  * otherwise an error code less than zero is returned, or zero if
2988  * no record(s) were appended.
2989  */
2990 static int
2991 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
2992 {
2993 	static const int MINRECLEN =
2994 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
2995 	struct ifnet		*ifp;
2996 	struct igmp_grouprec	 ig;
2997 	struct igmp_grouprec	*pig;
2998 	struct ip_msource	*ims, *nims;
2999 	struct mbuf		*m, *m0, *md;
3000 	in_addr_t		 naddr;
3001 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3002 	int			 nallow, nblock;
3003 	uint8_t			 mode, now, then;
3004 	rectype_t		 crt, drt, nrt;
3005 
3006 	IN_MULTI_LIST_LOCK_ASSERT();
3007 
3008 	if (inm->inm_nsrc == 0 ||
3009 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3010 		return (0);
3011 
3012 	ifp = inm->inm_ifp;			/* interface */
3013 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
3014 	crt = REC_NONE;	/* current group record type */
3015 	drt = REC_NONE;	/* mask of completed group record types */
3016 	nrt = REC_NONE;	/* record type for current node */
3017 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3018 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3019 	npbytes = 0;	/* # of bytes appended this packet */
3020 	rsrcs = 0;	/* # sources encoded in current record */
3021 	schanged = 0;	/* # nodes encoded in overall filter change */
3022 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3023 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3024 	nims = NULL;	/* next tree node pointer */
3025 
3026 	/*
3027 	 * For each possible filter record mode.
3028 	 * The first kind of source we encounter tells us which
3029 	 * is the first kind of record we start appending.
3030 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3031 	 * as the inverse of the group's filter mode.
3032 	 */
3033 	while (drt != REC_FULL) {
3034 		do {
3035 			m0 = mbufq_last(mq);
3036 			if (m0 != NULL &&
3037 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3038 			     IGMP_V3_REPORT_MAXRECS) &&
3039 			    (m0->m_pkthdr.len + MINRECLEN) <
3040 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3041 				m = m0;
3042 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3043 					    sizeof(struct igmp_grouprec)) /
3044 				    sizeof(in_addr_t);
3045 				CTR1(KTR_IGMPV3,
3046 				    "%s: use previous packet", __func__);
3047 			} else {
3048 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3049 				if (m)
3050 					m->m_data += IGMP_LEADINGSPACE;
3051 				if (m == NULL) {
3052 					m = m_gethdr(M_NOWAIT, MT_DATA);
3053 					if (m)
3054 						M_ALIGN(m, IGMP_LEADINGSPACE);
3055 				}
3056 				if (m == NULL) {
3057 					CTR1(KTR_IGMPV3,
3058 					    "%s: m_get*() failed", __func__);
3059 					return (-ENOMEM);
3060 				}
3061 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3062 				igmp_save_context(m, ifp);
3063 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3064 				    sizeof(struct igmp_grouprec)) /
3065 				    sizeof(in_addr_t);
3066 				npbytes = 0;
3067 				CTR1(KTR_IGMPV3,
3068 				    "%s: allocated new packet", __func__);
3069 			}
3070 			/*
3071 			 * Append the IGMP group record header to the
3072 			 * current packet's data area.
3073 			 * Recalculate pointer to free space for next
3074 			 * group record, in case m_append() allocated
3075 			 * a new mbuf or cluster.
3076 			 */
3077 			memset(&ig, 0, sizeof(ig));
3078 			ig.ig_group = inm->inm_addr;
3079 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3080 				if (m != m0)
3081 					m_freem(m);
3082 				CTR1(KTR_IGMPV3,
3083 				    "%s: m_append() failed", __func__);
3084 				return (-ENOMEM);
3085 			}
3086 			npbytes += sizeof(struct igmp_grouprec);
3087 			if (m != m0) {
3088 				/* new packet; offset in c hain */
3089 				md = m_getptr(m, npbytes -
3090 				    sizeof(struct igmp_grouprec), &off);
3091 				pig = (struct igmp_grouprec *)(mtod(md,
3092 				    uint8_t *) + off);
3093 			} else {
3094 				/* current packet; offset from last append */
3095 				md = m_last(m);
3096 				pig = (struct igmp_grouprec *)(mtod(md,
3097 				    uint8_t *) + md->m_len -
3098 				    sizeof(struct igmp_grouprec));
3099 			}
3100 			/*
3101 			 * Begin walking the tree for this record type
3102 			 * pass, or continue from where we left off
3103 			 * previously if we had to allocate a new packet.
3104 			 * Only report deltas in-mode at t1.
3105 			 * We need not report included sources as allowed
3106 			 * if we are in inclusive mode on the group,
3107 			 * however the converse is not true.
3108 			 */
3109 			rsrcs = 0;
3110 			if (nims == NULL)
3111 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3112 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3113 				CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3114 				    __func__, ims->ims_haddr);
3115 				now = ims_get_mode(inm, ims, 1);
3116 				then = ims_get_mode(inm, ims, 0);
3117 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3118 				    __func__, then, now);
3119 				if (now == then) {
3120 					CTR1(KTR_IGMPV3,
3121 					    "%s: skip unchanged", __func__);
3122 					continue;
3123 				}
3124 				if (mode == MCAST_EXCLUDE &&
3125 				    now == MCAST_INCLUDE) {
3126 					CTR1(KTR_IGMPV3,
3127 					    "%s: skip IN src on EX group",
3128 					    __func__);
3129 					continue;
3130 				}
3131 				nrt = (rectype_t)now;
3132 				if (nrt == REC_NONE)
3133 					nrt = (rectype_t)(~mode & REC_FULL);
3134 				if (schanged++ == 0) {
3135 					crt = nrt;
3136 				} else if (crt != nrt)
3137 					continue;
3138 				naddr = htonl(ims->ims_haddr);
3139 				if (!m_append(m, sizeof(in_addr_t),
3140 				    (void *)&naddr)) {
3141 					if (m != m0)
3142 						m_freem(m);
3143 					CTR1(KTR_IGMPV3,
3144 					    "%s: m_append() failed", __func__);
3145 					return (-ENOMEM);
3146 				}
3147 				nallow += !!(crt == REC_ALLOW);
3148 				nblock += !!(crt == REC_BLOCK);
3149 				if (++rsrcs == m0srcs)
3150 					break;
3151 			}
3152 			/*
3153 			 * If we did not append any tree nodes on this
3154 			 * pass, back out of allocations.
3155 			 */
3156 			if (rsrcs == 0) {
3157 				npbytes -= sizeof(struct igmp_grouprec);
3158 				if (m != m0) {
3159 					CTR1(KTR_IGMPV3,
3160 					    "%s: m_free(m)", __func__);
3161 					m_freem(m);
3162 				} else {
3163 					CTR1(KTR_IGMPV3,
3164 					    "%s: m_adj(m, -ig)", __func__);
3165 					m_adj(m, -((int)sizeof(
3166 					    struct igmp_grouprec)));
3167 				}
3168 				continue;
3169 			}
3170 			npbytes += (rsrcs * sizeof(in_addr_t));
3171 			if (crt == REC_ALLOW)
3172 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3173 			else if (crt == REC_BLOCK)
3174 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3175 			pig->ig_numsrc = htons(rsrcs);
3176 			/*
3177 			 * Count the new group record, and enqueue this
3178 			 * packet if it wasn't already queued.
3179 			 */
3180 			m->m_pkthdr.PH_vt.vt_nrecs++;
3181 			if (m != m0)
3182 				mbufq_enqueue(mq, m);
3183 			nbytes += npbytes;
3184 		} while (nims != NULL);
3185 		drt |= crt;
3186 		crt = (~crt & REC_FULL);
3187 	}
3188 
3189 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3190 	    nallow, nblock);
3191 
3192 	return (nbytes);
3193 }
3194 
3195 static int
3196 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3197 {
3198 	struct mbufq	*gq;
3199 	struct mbuf	*m;		/* pending state-change */
3200 	struct mbuf	*m0;		/* copy of pending state-change */
3201 	struct mbuf	*mt;		/* last state-change in packet */
3202 	int		 docopy, domerge;
3203 	u_int		 recslen;
3204 
3205 	docopy = 0;
3206 	domerge = 0;
3207 	recslen = 0;
3208 
3209 	IN_MULTI_LIST_LOCK_ASSERT();
3210 	IGMP_LOCK_ASSERT();
3211 
3212 	/*
3213 	 * If there are further pending retransmissions, make a writable
3214 	 * copy of each queued state-change message before merging.
3215 	 */
3216 	if (inm->inm_scrv > 0)
3217 		docopy = 1;
3218 
3219 	gq = &inm->inm_scq;
3220 #ifdef KTR
3221 	if (mbufq_first(gq) == NULL) {
3222 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3223 		    __func__, inm);
3224 	}
3225 #endif
3226 
3227 	m = mbufq_first(gq);
3228 	while (m != NULL) {
3229 		/*
3230 		 * Only merge the report into the current packet if
3231 		 * there is sufficient space to do so; an IGMPv3 report
3232 		 * packet may only contain 65,535 group records.
3233 		 * Always use a simple mbuf chain concatentation to do this,
3234 		 * as large state changes for single groups may have
3235 		 * allocated clusters.
3236 		 */
3237 		domerge = 0;
3238 		mt = mbufq_last(scq);
3239 		if (mt != NULL) {
3240 			recslen = m_length(m, NULL);
3241 
3242 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3243 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3244 			    IGMP_V3_REPORT_MAXRECS) &&
3245 			    (mt->m_pkthdr.len + recslen <=
3246 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3247 				domerge = 1;
3248 		}
3249 
3250 		if (!domerge && mbufq_full(gq)) {
3251 			CTR2(KTR_IGMPV3,
3252 			    "%s: outbound queue full, skipping whole packet %p",
3253 			    __func__, m);
3254 			mt = m->m_nextpkt;
3255 			if (!docopy)
3256 				m_freem(m);
3257 			m = mt;
3258 			continue;
3259 		}
3260 
3261 		if (!docopy) {
3262 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3263 			m0 = mbufq_dequeue(gq);
3264 			m = m0->m_nextpkt;
3265 		} else {
3266 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3267 			m0 = m_dup(m, M_NOWAIT);
3268 			if (m0 == NULL)
3269 				return (ENOMEM);
3270 			m0->m_nextpkt = NULL;
3271 			m = m->m_nextpkt;
3272 		}
3273 
3274 		if (!domerge) {
3275 			CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3276 			    __func__, m0, scq);
3277 			mbufq_enqueue(scq, m0);
3278 		} else {
3279 			struct mbuf *mtl;	/* last mbuf of packet mt */
3280 
3281 			CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3282 			    __func__, m0, mt);
3283 
3284 			mtl = m_last(mt);
3285 			m0->m_flags &= ~M_PKTHDR;
3286 			mt->m_pkthdr.len += recslen;
3287 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3288 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3289 
3290 			mtl->m_next = m0;
3291 		}
3292 	}
3293 
3294 	return (0);
3295 }
3296 
3297 /*
3298  * Respond to a pending IGMPv3 General Query.
3299  */
3300 static void
3301 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3302 {
3303 	struct ifmultiaddr	*ifma;
3304 	struct ifnet		*ifp;
3305 	struct in_multi		*inm;
3306 	int			 retval, loop;
3307 
3308 	IN_MULTI_LIST_LOCK_ASSERT();
3309 	IGMP_LOCK_ASSERT();
3310 
3311 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3312 	    ("%s: called when version %d", __func__, igi->igi_version));
3313 
3314 	/*
3315 	 * Check that there are some packets queued. If so, send them first.
3316 	 * For large number of groups the reply to general query can take
3317 	 * many packets, we should finish sending them before starting of
3318 	 * queuing the new reply.
3319 	 */
3320 	if (mbufq_len(&igi->igi_gq) != 0)
3321 		goto send;
3322 
3323 	ifp = igi->igi_ifp;
3324 
3325 	IF_ADDR_RLOCK(ifp);
3326 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3327 		if (ifma->ifma_addr->sa_family != AF_INET ||
3328 		    ifma->ifma_protospec == NULL)
3329 			continue;
3330 
3331 		inm = (struct in_multi *)ifma->ifma_protospec;
3332 		KASSERT(ifp == inm->inm_ifp,
3333 		    ("%s: inconsistent ifp", __func__));
3334 
3335 		switch (inm->inm_state) {
3336 		case IGMP_NOT_MEMBER:
3337 		case IGMP_SILENT_MEMBER:
3338 			break;
3339 		case IGMP_REPORTING_MEMBER:
3340 		case IGMP_IDLE_MEMBER:
3341 		case IGMP_LAZY_MEMBER:
3342 		case IGMP_SLEEPING_MEMBER:
3343 		case IGMP_AWAKENING_MEMBER:
3344 			inm->inm_state = IGMP_REPORTING_MEMBER;
3345 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3346 			    inm, 0, 0, 0);
3347 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3348 			    __func__, retval);
3349 			break;
3350 		case IGMP_G_QUERY_PENDING_MEMBER:
3351 		case IGMP_SG_QUERY_PENDING_MEMBER:
3352 		case IGMP_LEAVING_MEMBER:
3353 			break;
3354 		}
3355 	}
3356 	IF_ADDR_RUNLOCK(ifp);
3357 
3358 send:
3359 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3360 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3361 
3362 	/*
3363 	 * Slew transmission of bursts over 500ms intervals.
3364 	 */
3365 	if (mbufq_first(&igi->igi_gq) != NULL) {
3366 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3367 		    IGMP_RESPONSE_BURST_INTERVAL);
3368 		V_interface_timers_running = 1;
3369 	}
3370 }
3371 
3372 /*
3373  * Transmit the next pending IGMP message in the output queue.
3374  *
3375  * We get called from netisr_processqueue(). A mutex private to igmpoq
3376  * will be acquired and released around this routine.
3377  *
3378  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3379  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3380  * a link and uses a link-scope multicast address.
3381  */
3382 static void
3383 igmp_intr(struct mbuf *m)
3384 {
3385 	struct ip_moptions	 imo;
3386 	struct ifnet		*ifp;
3387 	struct mbuf		*ipopts, *m0;
3388 	int			 error;
3389 	uint32_t		 ifindex;
3390 
3391 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3392 
3393 	/*
3394 	 * Set VNET image pointer from enqueued mbuf chain
3395 	 * before doing anything else. Whilst we use interface
3396 	 * indexes to guard against interface detach, they are
3397 	 * unique to each VIMAGE and must be retrieved.
3398 	 */
3399 	CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3400 	ifindex = igmp_restore_context(m);
3401 
3402 	/*
3403 	 * Check if the ifnet still exists. This limits the scope of
3404 	 * any race in the absence of a global ifp lock for low cost
3405 	 * (an array lookup).
3406 	 */
3407 	ifp = ifnet_byindex(ifindex);
3408 	if (ifp == NULL) {
3409 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3410 		    __func__, m, ifindex);
3411 		m_freem(m);
3412 		IPSTAT_INC(ips_noroute);
3413 		goto out;
3414 	}
3415 
3416 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3417 
3418 	imo.imo_multicast_ttl  = 1;
3419 	imo.imo_multicast_vif  = -1;
3420 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3421 
3422 	/*
3423 	 * If the user requested that IGMP traffic be explicitly
3424 	 * redirected to the loopback interface (e.g. they are running a
3425 	 * MANET interface and the routing protocol needs to see the
3426 	 * updates), handle this now.
3427 	 */
3428 	if (m->m_flags & M_IGMP_LOOP)
3429 		imo.imo_multicast_ifp = V_loif;
3430 	else
3431 		imo.imo_multicast_ifp = ifp;
3432 
3433 	if (m->m_flags & M_IGMPV2) {
3434 		m0 = m;
3435 	} else {
3436 		m0 = igmp_v3_encap_report(ifp, m);
3437 		if (m0 == NULL) {
3438 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3439 			m_freem(m);
3440 			IPSTAT_INC(ips_odropped);
3441 			goto out;
3442 		}
3443 	}
3444 
3445 	igmp_scrub_context(m0);
3446 	m_clrprotoflags(m);
3447 	m0->m_pkthdr.rcvif = V_loif;
3448 #ifdef MAC
3449 	mac_netinet_igmp_send(ifp, m0);
3450 #endif
3451 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3452 	if (error) {
3453 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3454 		goto out;
3455 	}
3456 
3457 	IGMPSTAT_INC(igps_snd_reports);
3458 
3459 out:
3460 	/*
3461 	 * We must restore the existing vnet pointer before
3462 	 * continuing as we are run from netisr context.
3463 	 */
3464 	CURVNET_RESTORE();
3465 }
3466 
3467 /*
3468  * Encapsulate an IGMPv3 report.
3469  *
3470  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3471  * chain has already had its IP/IGMPv3 header prepended. In this case
3472  * the function will not attempt to prepend; the lengths and checksums
3473  * will however be re-computed.
3474  *
3475  * Returns a pointer to the new mbuf chain head, or NULL if the
3476  * allocation failed.
3477  */
3478 static struct mbuf *
3479 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3480 {
3481 	struct rm_priotracker	in_ifa_tracker;
3482 	struct igmp_report	*igmp;
3483 	struct ip		*ip;
3484 	int			 hdrlen, igmpreclen;
3485 
3486 	KASSERT((m->m_flags & M_PKTHDR),
3487 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3488 
3489 	igmpreclen = m_length(m, NULL);
3490 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3491 
3492 	if (m->m_flags & M_IGMPV3_HDR) {
3493 		igmpreclen -= hdrlen;
3494 	} else {
3495 		M_PREPEND(m, hdrlen, M_NOWAIT);
3496 		if (m == NULL)
3497 			return (NULL);
3498 		m->m_flags |= M_IGMPV3_HDR;
3499 	}
3500 
3501 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3502 
3503 	m->m_data += sizeof(struct ip);
3504 	m->m_len -= sizeof(struct ip);
3505 
3506 	igmp = mtod(m, struct igmp_report *);
3507 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3508 	igmp->ir_rsv1 = 0;
3509 	igmp->ir_rsv2 = 0;
3510 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3511 	igmp->ir_cksum = 0;
3512 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3513 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3514 
3515 	m->m_data -= sizeof(struct ip);
3516 	m->m_len += sizeof(struct ip);
3517 
3518 	ip = mtod(m, struct ip *);
3519 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3520 	ip->ip_len = htons(hdrlen + igmpreclen);
3521 	ip->ip_off = htons(IP_DF);
3522 	ip->ip_p = IPPROTO_IGMP;
3523 	ip->ip_sum = 0;
3524 
3525 	ip->ip_src.s_addr = INADDR_ANY;
3526 
3527 	if (m->m_flags & M_IGMP_LOOP) {
3528 		struct in_ifaddr *ia;
3529 
3530 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3531 		if (ia != NULL) {
3532 			ip->ip_src = ia->ia_addr.sin_addr;
3533 			ifa_free(&ia->ia_ifa);
3534 		}
3535 	}
3536 
3537 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3538 
3539 	return (m);
3540 }
3541 
3542 #ifdef KTR
3543 static char *
3544 igmp_rec_type_to_str(const int type)
3545 {
3546 
3547 	switch (type) {
3548 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3549 			return "TO_EX";
3550 			break;
3551 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3552 			return "TO_IN";
3553 			break;
3554 		case IGMP_MODE_IS_EXCLUDE:
3555 			return "MODE_EX";
3556 			break;
3557 		case IGMP_MODE_IS_INCLUDE:
3558 			return "MODE_IN";
3559 			break;
3560 		case IGMP_ALLOW_NEW_SOURCES:
3561 			return "ALLOW_NEW";
3562 			break;
3563 		case IGMP_BLOCK_OLD_SOURCES:
3564 			return "BLOCK_OLD";
3565 			break;
3566 		default:
3567 			break;
3568 	}
3569 	return "unknown";
3570 }
3571 #endif
3572 
3573 #ifdef VIMAGE
3574 static void
3575 vnet_igmp_init(const void *unused __unused)
3576 {
3577 
3578 	netisr_register_vnet(&igmp_nh);
3579 }
3580 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3581     vnet_igmp_init, NULL);
3582 
3583 static void
3584 vnet_igmp_uninit(const void *unused __unused)
3585 {
3586 
3587 	/* This can happen when we shutdown the entire network stack. */
3588 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3589 
3590 	netisr_unregister_vnet(&igmp_nh);
3591 }
3592 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3593     vnet_igmp_uninit, NULL);
3594 #endif
3595 
3596 #ifdef DDB
3597 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3598 {
3599 	struct igmp_ifsoftc *igi, *tigi;
3600 	LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3601 
3602 	if (!have_addr) {
3603 		db_printf("usage: show igi_list <addr>\n");
3604 		return;
3605 	}
3606 	igi_head = (struct _igi_list *)addr;
3607 
3608 	LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3609 		db_printf("igmp_ifsoftc %p:\n", igi);
3610 		db_printf("    ifp %p\n", igi->igi_ifp);
3611 		db_printf("    version %u\n", igi->igi_version);
3612 		db_printf("    v1_timer %u\n", igi->igi_v1_timer);
3613 		db_printf("    v2_timer %u\n", igi->igi_v2_timer);
3614 		db_printf("    v3_timer %u\n", igi->igi_v3_timer);
3615 		db_printf("    flags %#x\n", igi->igi_flags);
3616 		db_printf("    rv %u\n", igi->igi_rv);
3617 		db_printf("    qi %u\n", igi->igi_qi);
3618 		db_printf("    qri %u\n", igi->igi_qri);
3619 		db_printf("    uri %u\n", igi->igi_uri);
3620 		/* struct mbufq    igi_gq; */
3621 		db_printf("\n");
3622 	}
3623 }
3624 #endif
3625 
3626 static int
3627 igmp_modevent(module_t mod, int type, void *unused __unused)
3628 {
3629 
3630 	switch (type) {
3631 	case MOD_LOAD:
3632 		CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3633 		IGMP_LOCK_INIT();
3634 		m_raopt = igmp_ra_alloc();
3635 		netisr_register(&igmp_nh);
3636 		break;
3637 	case MOD_UNLOAD:
3638 		CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3639 		netisr_unregister(&igmp_nh);
3640 		m_free(m_raopt);
3641 		m_raopt = NULL;
3642 		IGMP_LOCK_DESTROY();
3643 		break;
3644 	default:
3645 		return (EOPNOTSUPP);
3646 	}
3647 	return (0);
3648 }
3649 
3650 static moduledata_t igmp_mod = {
3651     "igmp",
3652     igmp_modevent,
3653     0
3654 };
3655 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);
3656