xref: /freebsd/sys/netinet/igmp.c (revision 0957b409a90fd597c1e9124cbaf3edd2b488f4ac)
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 VNET_DEFINE_STATIC(int, interface_timers_running);	/* IGMPv3 general
221 							 * query response */
222 VNET_DEFINE_STATIC(int, state_change_timers_running);	/* IGMPv3 state-change
223 							 * retransmit */
224 VNET_DEFINE_STATIC(int, current_state_timers_running);	/* IGMPv1/v2 host
225 							 * report; IGMPv3 g/sg
226 							 * query response */
227 
228 #define	V_interface_timers_running	VNET(interface_timers_running)
229 #define	V_state_change_timers_running	VNET(state_change_timers_running)
230 #define	V_current_state_timers_running	VNET(current_state_timers_running)
231 
232 VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) =
233     LIST_HEAD_INITIALIZER(igi_head);
234 VNET_DEFINE_STATIC(struct igmpstat, igmpstat) = {
235 	.igps_version = IGPS_VERSION_3,
236 	.igps_len = sizeof(struct igmpstat),
237 };
238 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1;
245 VNET_DEFINE_STATIC(int, igmp_sendra) = 1;
246 VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1;
247 VNET_DEFINE_STATIC(int, igmp_v1enable) = 1;
248 VNET_DEFINE_STATIC(int, igmp_v2enable) = 1;
249 VNET_DEFINE_STATIC(int, igmp_legacysupp);
250 VNET_DEFINE_STATIC(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 		CK_STAILQ_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 epoch_tracker	 et;
696 	struct ifmultiaddr	*ifma;
697 	struct igmp_ifsoftc	*igi;
698 	struct in_multi		*inm;
699 
700 	/*
701 	 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
702 	 * 224.0.0.1. They are always treated as General Queries.
703 	 * igmp_group is always ignored. Do not drop it as a userland
704 	 * daemon may wish to see it.
705 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
706 	 */
707 	if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
708 		IGMPSTAT_INC(igps_rcv_badqueries);
709 		return (0);
710 	}
711 	IGMPSTAT_INC(igps_rcv_gen_queries);
712 
713 	IN_MULTI_LIST_LOCK();
714 	IGMP_LOCK();
715 
716 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
717 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
718 
719 	if (igi->igi_flags & IGIF_LOOPBACK) {
720 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
721 		    ifp, ifp->if_xname);
722 		goto out_locked;
723 	}
724 
725 	/*
726 	 * Switch to IGMPv1 host compatibility mode.
727 	 */
728 	igmp_set_version(igi, IGMP_VERSION_1);
729 
730 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
731 
732 	/*
733 	 * Start the timers in all of our group records
734 	 * for the interface on which the query arrived,
735 	 * except those which are already running.
736 	 */
737 	NET_EPOCH_ENTER(et);
738 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
739 		if (ifma->ifma_addr->sa_family != AF_INET ||
740 		    ifma->ifma_protospec == NULL)
741 			continue;
742 		inm = (struct in_multi *)ifma->ifma_protospec;
743 		if (inm->inm_timer != 0)
744 			continue;
745 		switch (inm->inm_state) {
746 		case IGMP_NOT_MEMBER:
747 		case IGMP_SILENT_MEMBER:
748 			break;
749 		case IGMP_G_QUERY_PENDING_MEMBER:
750 		case IGMP_SG_QUERY_PENDING_MEMBER:
751 		case IGMP_REPORTING_MEMBER:
752 		case IGMP_IDLE_MEMBER:
753 		case IGMP_LAZY_MEMBER:
754 		case IGMP_SLEEPING_MEMBER:
755 		case IGMP_AWAKENING_MEMBER:
756 			inm->inm_state = IGMP_REPORTING_MEMBER;
757 			inm->inm_timer = IGMP_RANDOM_DELAY(
758 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
759 			V_current_state_timers_running = 1;
760 			break;
761 		case IGMP_LEAVING_MEMBER:
762 			break;
763 		}
764 	}
765 	NET_EPOCH_EXIT(et);
766 
767 out_locked:
768 	IGMP_UNLOCK();
769 	IN_MULTI_LIST_UNLOCK();
770 
771 	return (0);
772 }
773 
774 /*
775  * Process a received IGMPv2 general or group-specific query.
776  */
777 static int
778 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
779     const struct igmp *igmp)
780 {
781 	struct epoch_tracker	 et;
782 	struct ifmultiaddr	*ifma;
783 	struct igmp_ifsoftc	*igi;
784 	struct in_multi		*inm;
785 	int			 is_general_query;
786 	uint16_t		 timer;
787 
788 	is_general_query = 0;
789 
790 	/*
791 	 * Validate address fields upfront.
792 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
793 	 */
794 	if (in_nullhost(igmp->igmp_group)) {
795 		/*
796 		 * IGMPv2 General Query.
797 		 * If this was not sent to the all-hosts group, ignore it.
798 		 */
799 		if (!in_allhosts(ip->ip_dst))
800 			return (0);
801 		IGMPSTAT_INC(igps_rcv_gen_queries);
802 		is_general_query = 1;
803 	} else {
804 		/* IGMPv2 Group-Specific Query. */
805 		IGMPSTAT_INC(igps_rcv_group_queries);
806 	}
807 
808 	IN_MULTI_LIST_LOCK();
809 	IGMP_LOCK();
810 
811 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
812 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
813 
814 	if (igi->igi_flags & IGIF_LOOPBACK) {
815 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
816 		    ifp, ifp->if_xname);
817 		goto out_locked;
818 	}
819 
820 	/*
821 	 * Ignore v2 query if in v1 Compatibility Mode.
822 	 */
823 	if (igi->igi_version == IGMP_VERSION_1)
824 		goto out_locked;
825 
826 	igmp_set_version(igi, IGMP_VERSION_2);
827 
828 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
829 	if (timer == 0)
830 		timer = 1;
831 
832 	if (is_general_query) {
833 		/*
834 		 * For each reporting group joined on this
835 		 * interface, kick the report timer.
836 		 */
837 		CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
838 		    ifp, ifp->if_xname);
839 		NET_EPOCH_ENTER(et);
840 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
841 			if (ifma->ifma_addr->sa_family != AF_INET ||
842 			    ifma->ifma_protospec == NULL)
843 				continue;
844 			inm = (struct in_multi *)ifma->ifma_protospec;
845 			igmp_v2_update_group(inm, timer);
846 		}
847 		NET_EPOCH_EXIT(et);
848 	} else {
849 		/*
850 		 * Group-specific IGMPv2 query, we need only
851 		 * look up the single group to process it.
852 		 */
853 		inm = inm_lookup(ifp, igmp->igmp_group);
854 		if (inm != NULL) {
855 			CTR3(KTR_IGMPV3,
856 			    "process v2 query 0x%08x on ifp %p(%s)",
857 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
858 			igmp_v2_update_group(inm, timer);
859 		}
860 	}
861 
862 out_locked:
863 	IGMP_UNLOCK();
864 	IN_MULTI_LIST_UNLOCK();
865 
866 	return (0);
867 }
868 
869 /*
870  * Update the report timer on a group in response to an IGMPv2 query.
871  *
872  * If we are becoming the reporting member for this group, start the timer.
873  * If we already are the reporting member for this group, and timer is
874  * below the threshold, reset it.
875  *
876  * We may be updating the group for the first time since we switched
877  * to IGMPv3. If we are, then we must clear any recorded source lists,
878  * and transition to REPORTING state; the group timer is overloaded
879  * for group and group-source query responses.
880  *
881  * Unlike IGMPv3, the delay per group should be jittered
882  * to avoid bursts of IGMPv2 reports.
883  */
884 static void
885 igmp_v2_update_group(struct in_multi *inm, const int timer)
886 {
887 
888 	CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
889 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
890 
891 	IN_MULTI_LIST_LOCK_ASSERT();
892 
893 	switch (inm->inm_state) {
894 	case IGMP_NOT_MEMBER:
895 	case IGMP_SILENT_MEMBER:
896 		break;
897 	case IGMP_REPORTING_MEMBER:
898 		if (inm->inm_timer != 0 &&
899 		    inm->inm_timer <= timer) {
900 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
901 			    "skipping.", __func__);
902 			break;
903 		}
904 		/* FALLTHROUGH */
905 	case IGMP_SG_QUERY_PENDING_MEMBER:
906 	case IGMP_G_QUERY_PENDING_MEMBER:
907 	case IGMP_IDLE_MEMBER:
908 	case IGMP_LAZY_MEMBER:
909 	case IGMP_AWAKENING_MEMBER:
910 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
911 		inm->inm_state = IGMP_REPORTING_MEMBER;
912 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
913 		V_current_state_timers_running = 1;
914 		break;
915 	case IGMP_SLEEPING_MEMBER:
916 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
917 		inm->inm_state = IGMP_AWAKENING_MEMBER;
918 		break;
919 	case IGMP_LEAVING_MEMBER:
920 		break;
921 	}
922 }
923 
924 /*
925  * Process a received IGMPv3 general, group-specific or
926  * group-and-source-specific query.
927  * Assumes m has already been pulled up to the full IGMP message length.
928  * Return 0 if successful, otherwise an appropriate error code is returned.
929  */
930 static int
931 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
932     /*const*/ struct igmpv3 *igmpv3)
933 {
934 	struct igmp_ifsoftc	*igi;
935 	struct in_multi		*inm;
936 	int			 is_general_query;
937 	uint32_t		 maxresp, nsrc, qqi;
938 	uint16_t		 timer;
939 	uint8_t			 qrv;
940 
941 	is_general_query = 0;
942 
943 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
944 
945 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
946 	if (maxresp >= 128) {
947 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
948 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
949 	}
950 
951 	/*
952 	 * Robustness must never be less than 2 for on-wire IGMPv3.
953 	 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
954 	 * an exception for interfaces whose IGMPv3 state changes
955 	 * are redirected to loopback (e.g. MANET).
956 	 */
957 	qrv = IGMP_QRV(igmpv3->igmp_misc);
958 	if (qrv < 2) {
959 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
960 		    qrv, IGMP_RV_INIT);
961 		qrv = IGMP_RV_INIT;
962 	}
963 
964 	qqi = igmpv3->igmp_qqi;
965 	if (qqi >= 128) {
966 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
967 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
968 	}
969 
970 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
971 	if (timer == 0)
972 		timer = 1;
973 
974 	nsrc = ntohs(igmpv3->igmp_numsrc);
975 
976 	/*
977 	 * Validate address fields and versions upfront before
978 	 * accepting v3 query.
979 	 * XXX SMPng: Unlocked access to igmpstat counters here.
980 	 */
981 	if (in_nullhost(igmpv3->igmp_group)) {
982 		/*
983 		 * IGMPv3 General Query.
984 		 *
985 		 * General Queries SHOULD be directed to 224.0.0.1.
986 		 * A general query with a source list has undefined
987 		 * behaviour; discard it.
988 		 */
989 		IGMPSTAT_INC(igps_rcv_gen_queries);
990 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
991 			IGMPSTAT_INC(igps_rcv_badqueries);
992 			return (0);
993 		}
994 		is_general_query = 1;
995 	} else {
996 		/* Group or group-source specific query. */
997 		if (nsrc == 0)
998 			IGMPSTAT_INC(igps_rcv_group_queries);
999 		else
1000 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1001 	}
1002 
1003 	IN_MULTI_LIST_LOCK();
1004 	IGMP_LOCK();
1005 
1006 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1007 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1008 
1009 	if (igi->igi_flags & IGIF_LOOPBACK) {
1010 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1011 		    ifp, ifp->if_xname);
1012 		goto out_locked;
1013 	}
1014 
1015 	/*
1016 	 * Discard the v3 query if we're in Compatibility Mode.
1017 	 * The RFC is not obviously worded that hosts need to stay in
1018 	 * compatibility mode until the Old Version Querier Present
1019 	 * timer expires.
1020 	 */
1021 	if (igi->igi_version != IGMP_VERSION_3) {
1022 		CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1023 		    igi->igi_version, ifp, ifp->if_xname);
1024 		goto out_locked;
1025 	}
1026 
1027 	igmp_set_version(igi, IGMP_VERSION_3);
1028 	igi->igi_rv = qrv;
1029 	igi->igi_qi = qqi;
1030 	igi->igi_qri = maxresp;
1031 
1032 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1033 	    maxresp);
1034 
1035 	if (is_general_query) {
1036 		/*
1037 		 * Schedule a current-state report on this ifp for
1038 		 * all groups, possibly containing source lists.
1039 		 * If there is a pending General Query response
1040 		 * scheduled earlier than the selected delay, do
1041 		 * not schedule any other reports.
1042 		 * Otherwise, reset the interface timer.
1043 		 */
1044 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1045 		    ifp, ifp->if_xname);
1046 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1047 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1048 			V_interface_timers_running = 1;
1049 		}
1050 	} else {
1051 		/*
1052 		 * Group-source-specific queries are throttled on
1053 		 * a per-group basis to defeat denial-of-service attempts.
1054 		 * Queries for groups we are not a member of on this
1055 		 * link are simply ignored.
1056 		 */
1057 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1058 		if (inm == NULL)
1059 			goto out_locked;
1060 		if (nsrc > 0) {
1061 			if (!ratecheck(&inm->inm_lastgsrtv,
1062 			    &V_igmp_gsrdelay)) {
1063 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1064 				    __func__);
1065 				IGMPSTAT_INC(igps_drop_gsr_queries);
1066 				goto out_locked;
1067 			}
1068 		}
1069 		CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1070 		     ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1071 		/*
1072 		 * If there is a pending General Query response
1073 		 * scheduled sooner than the selected delay, no
1074 		 * further report need be scheduled.
1075 		 * Otherwise, prepare to respond to the
1076 		 * group-specific or group-and-source query.
1077 		 */
1078 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1079 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1080 	}
1081 
1082 out_locked:
1083 	IGMP_UNLOCK();
1084 	IN_MULTI_LIST_UNLOCK();
1085 
1086 	return (0);
1087 }
1088 
1089 /*
1090  * Process a received IGMPv3 group-specific or group-and-source-specific
1091  * query.
1092  * Return <0 if any error occurred. Currently this is ignored.
1093  */
1094 static int
1095 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1096     int timer, /*const*/ struct igmpv3 *igmpv3)
1097 {
1098 	int			 retval;
1099 	uint16_t		 nsrc;
1100 
1101 	IN_MULTI_LIST_LOCK_ASSERT();
1102 	IGMP_LOCK_ASSERT();
1103 
1104 	retval = 0;
1105 
1106 	switch (inm->inm_state) {
1107 	case IGMP_NOT_MEMBER:
1108 	case IGMP_SILENT_MEMBER:
1109 	case IGMP_SLEEPING_MEMBER:
1110 	case IGMP_LAZY_MEMBER:
1111 	case IGMP_AWAKENING_MEMBER:
1112 	case IGMP_IDLE_MEMBER:
1113 	case IGMP_LEAVING_MEMBER:
1114 		return (retval);
1115 		break;
1116 	case IGMP_REPORTING_MEMBER:
1117 	case IGMP_G_QUERY_PENDING_MEMBER:
1118 	case IGMP_SG_QUERY_PENDING_MEMBER:
1119 		break;
1120 	}
1121 
1122 	nsrc = ntohs(igmpv3->igmp_numsrc);
1123 
1124 	/*
1125 	 * Deal with group-specific queries upfront.
1126 	 * If any group query is already pending, purge any recorded
1127 	 * source-list state if it exists, and schedule a query response
1128 	 * for this group-specific query.
1129 	 */
1130 	if (nsrc == 0) {
1131 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1132 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1133 			inm_clear_recorded(inm);
1134 			timer = min(inm->inm_timer, timer);
1135 		}
1136 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1137 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1138 		V_current_state_timers_running = 1;
1139 		return (retval);
1140 	}
1141 
1142 	/*
1143 	 * Deal with the case where a group-and-source-specific query has
1144 	 * been received but a group-specific query is already pending.
1145 	 */
1146 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1147 		timer = min(inm->inm_timer, timer);
1148 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1149 		V_current_state_timers_running = 1;
1150 		return (retval);
1151 	}
1152 
1153 	/*
1154 	 * Finally, deal with the case where a group-and-source-specific
1155 	 * query has been received, where a response to a previous g-s-r
1156 	 * query exists, or none exists.
1157 	 * In this case, we need to parse the source-list which the Querier
1158 	 * has provided us with and check if we have any source list filter
1159 	 * entries at T1 for these sources. If we do not, there is no need
1160 	 * schedule a report and the query may be dropped.
1161 	 * If we do, we must record them and schedule a current-state
1162 	 * report for those sources.
1163 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1164 	 * we pass the mbuf chain pointer down to this function, and use
1165 	 * m_getptr() to walk the chain.
1166 	 */
1167 	if (inm->inm_nsrc > 0) {
1168 		const struct in_addr	*ap;
1169 		int			 i, nrecorded;
1170 
1171 		ap = (const struct in_addr *)(igmpv3 + 1);
1172 		nrecorded = 0;
1173 		for (i = 0; i < nsrc; i++, ap++) {
1174 			retval = inm_record_source(inm, ap->s_addr);
1175 			if (retval < 0)
1176 				break;
1177 			nrecorded += retval;
1178 		}
1179 		if (nrecorded > 0) {
1180 			CTR1(KTR_IGMPV3,
1181 			    "%s: schedule response to SG query", __func__);
1182 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1183 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1184 			V_current_state_timers_running = 1;
1185 		}
1186 	}
1187 
1188 	return (retval);
1189 }
1190 
1191 /*
1192  * Process a received IGMPv1 host membership report.
1193  *
1194  * NOTE: 0.0.0.0 workaround breaks const correctness.
1195  */
1196 static int
1197 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1198     /*const*/ struct igmp *igmp)
1199 {
1200 	struct rm_priotracker in_ifa_tracker;
1201 	struct in_ifaddr *ia;
1202 	struct in_multi *inm;
1203 
1204 	IGMPSTAT_INC(igps_rcv_reports);
1205 
1206 	if (ifp->if_flags & IFF_LOOPBACK)
1207 		return (0);
1208 
1209 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1210 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1211 		IGMPSTAT_INC(igps_rcv_badreports);
1212 		return (EINVAL);
1213 	}
1214 
1215 	/*
1216 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1217 	 * Booting clients may use the source address 0.0.0.0. Some
1218 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1219 	 * the interface upon which this message was received.
1220 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1221 	 */
1222 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1223 		struct epoch_tracker et;
1224 
1225 		NET_EPOCH_ENTER(et);
1226 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1227 		if (ia != NULL)
1228 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1229 		NET_EPOCH_EXIT(et);
1230 	}
1231 
1232 	CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1233 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1234 
1235 	/*
1236 	 * IGMPv1 report suppression.
1237 	 * If we are a member of this group, and our membership should be
1238 	 * reported, stop our group timer and transition to the 'lazy' state.
1239 	 */
1240 	IN_MULTI_LIST_LOCK();
1241 	inm = inm_lookup(ifp, igmp->igmp_group);
1242 	if (inm != NULL) {
1243 		struct igmp_ifsoftc *igi;
1244 
1245 		igi = inm->inm_igi;
1246 		if (igi == NULL) {
1247 			KASSERT(igi != NULL,
1248 			    ("%s: no igi for ifp %p", __func__, ifp));
1249 			goto out_locked;
1250 		}
1251 
1252 		IGMPSTAT_INC(igps_rcv_ourreports);
1253 
1254 		/*
1255 		 * If we are in IGMPv3 host mode, do not allow the
1256 		 * other host's IGMPv1 report to suppress our reports
1257 		 * unless explicitly configured to do so.
1258 		 */
1259 		if (igi->igi_version == IGMP_VERSION_3) {
1260 			if (V_igmp_legacysupp)
1261 				igmp_v3_suppress_group_record(inm);
1262 			goto out_locked;
1263 		}
1264 
1265 		inm->inm_timer = 0;
1266 
1267 		switch (inm->inm_state) {
1268 		case IGMP_NOT_MEMBER:
1269 		case IGMP_SILENT_MEMBER:
1270 			break;
1271 		case IGMP_IDLE_MEMBER:
1272 		case IGMP_LAZY_MEMBER:
1273 		case IGMP_AWAKENING_MEMBER:
1274 			CTR3(KTR_IGMPV3,
1275 			    "report suppressed for 0x%08x on ifp %p(%s)",
1276 			    ntohl(igmp->igmp_group.s_addr), ifp,
1277 			    ifp->if_xname);
1278 		case IGMP_SLEEPING_MEMBER:
1279 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1280 			break;
1281 		case IGMP_REPORTING_MEMBER:
1282 			CTR3(KTR_IGMPV3,
1283 			    "report suppressed for 0x%08x on ifp %p(%s)",
1284 			    ntohl(igmp->igmp_group.s_addr), ifp,
1285 			    ifp->if_xname);
1286 			if (igi->igi_version == IGMP_VERSION_1)
1287 				inm->inm_state = IGMP_LAZY_MEMBER;
1288 			else if (igi->igi_version == IGMP_VERSION_2)
1289 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1290 			break;
1291 		case IGMP_G_QUERY_PENDING_MEMBER:
1292 		case IGMP_SG_QUERY_PENDING_MEMBER:
1293 		case IGMP_LEAVING_MEMBER:
1294 			break;
1295 		}
1296 	}
1297 
1298 out_locked:
1299 	IN_MULTI_LIST_UNLOCK();
1300 
1301 	return (0);
1302 }
1303 
1304 /*
1305  * Process a received IGMPv2 host membership report.
1306  *
1307  * NOTE: 0.0.0.0 workaround breaks const correctness.
1308  */
1309 static int
1310 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1311     /*const*/ struct igmp *igmp)
1312 {
1313 	struct rm_priotracker in_ifa_tracker;
1314 	struct epoch_tracker et;
1315 	struct in_ifaddr *ia;
1316 	struct in_multi *inm;
1317 
1318 	/*
1319 	 * Make sure we don't hear our own membership report.  Fast
1320 	 * leave requires knowing that we are the only member of a
1321 	 * group.
1322 	 */
1323 	NET_EPOCH_ENTER(et);
1324 	IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1325 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1326 		NET_EPOCH_EXIT(et);
1327 		return (0);
1328 	}
1329 
1330 	IGMPSTAT_INC(igps_rcv_reports);
1331 
1332 	if (ifp->if_flags & IFF_LOOPBACK) {
1333 		NET_EPOCH_EXIT(et);
1334 		return (0);
1335 	}
1336 
1337 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1338 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1339 		NET_EPOCH_EXIT(et);
1340 		IGMPSTAT_INC(igps_rcv_badreports);
1341 		return (EINVAL);
1342 	}
1343 
1344 	/*
1345 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1346 	 * Booting clients may use the source address 0.0.0.0. Some
1347 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1348 	 * the interface upon which this message was received.
1349 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1350 	 */
1351 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1352 		if (ia != NULL)
1353 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1354 	}
1355 	NET_EPOCH_EXIT(et);
1356 
1357 	CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1358 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1359 
1360 	/*
1361 	 * IGMPv2 report suppression.
1362 	 * If we are a member of this group, and our membership should be
1363 	 * reported, and our group timer is pending or about to be reset,
1364 	 * stop our group timer by transitioning to the 'lazy' state.
1365 	 */
1366 	IN_MULTI_LIST_LOCK();
1367 	inm = inm_lookup(ifp, igmp->igmp_group);
1368 	if (inm != NULL) {
1369 		struct igmp_ifsoftc *igi;
1370 
1371 		igi = inm->inm_igi;
1372 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1373 
1374 		IGMPSTAT_INC(igps_rcv_ourreports);
1375 
1376 		/*
1377 		 * If we are in IGMPv3 host mode, do not allow the
1378 		 * other host's IGMPv1 report to suppress our reports
1379 		 * unless explicitly configured to do so.
1380 		 */
1381 		if (igi->igi_version == IGMP_VERSION_3) {
1382 			if (V_igmp_legacysupp)
1383 				igmp_v3_suppress_group_record(inm);
1384 			goto out_locked;
1385 		}
1386 
1387 		inm->inm_timer = 0;
1388 
1389 		switch (inm->inm_state) {
1390 		case IGMP_NOT_MEMBER:
1391 		case IGMP_SILENT_MEMBER:
1392 		case IGMP_SLEEPING_MEMBER:
1393 			break;
1394 		case IGMP_REPORTING_MEMBER:
1395 		case IGMP_IDLE_MEMBER:
1396 		case IGMP_AWAKENING_MEMBER:
1397 			CTR3(KTR_IGMPV3,
1398 			    "report suppressed for 0x%08x on ifp %p(%s)",
1399 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1400 		case IGMP_LAZY_MEMBER:
1401 			inm->inm_state = IGMP_LAZY_MEMBER;
1402 			break;
1403 		case IGMP_G_QUERY_PENDING_MEMBER:
1404 		case IGMP_SG_QUERY_PENDING_MEMBER:
1405 		case IGMP_LEAVING_MEMBER:
1406 			break;
1407 		}
1408 	}
1409 
1410 out_locked:
1411 	IN_MULTI_LIST_UNLOCK();
1412 
1413 	return (0);
1414 }
1415 
1416 int
1417 igmp_input(struct mbuf **mp, int *offp, int proto)
1418 {
1419 	int iphlen;
1420 	struct ifnet *ifp;
1421 	struct igmp *igmp;
1422 	struct ip *ip;
1423 	struct mbuf *m;
1424 	int igmplen;
1425 	int minlen;
1426 	int queryver;
1427 
1428 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1429 
1430 	m = *mp;
1431 	ifp = m->m_pkthdr.rcvif;
1432 	*mp = NULL;
1433 
1434 	IGMPSTAT_INC(igps_rcv_total);
1435 
1436 	ip = mtod(m, struct ip *);
1437 	iphlen = *offp;
1438 	igmplen = ntohs(ip->ip_len) - iphlen;
1439 
1440 	/*
1441 	 * Validate lengths.
1442 	 */
1443 	if (igmplen < IGMP_MINLEN) {
1444 		IGMPSTAT_INC(igps_rcv_tooshort);
1445 		m_freem(m);
1446 		return (IPPROTO_DONE);
1447 	}
1448 
1449 	/*
1450 	 * Always pullup to the minimum size for v1/v2 or v3
1451 	 * to amortize calls to m_pullup().
1452 	 */
1453 	minlen = iphlen;
1454 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1455 		minlen += IGMP_V3_QUERY_MINLEN;
1456 	else
1457 		minlen += IGMP_MINLEN;
1458 	if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1459 	    (m = m_pullup(m, minlen)) == NULL) {
1460 		IGMPSTAT_INC(igps_rcv_tooshort);
1461 		return (IPPROTO_DONE);
1462 	}
1463 	ip = mtod(m, struct ip *);
1464 
1465 	/*
1466 	 * Validate checksum.
1467 	 */
1468 	m->m_data += iphlen;
1469 	m->m_len -= iphlen;
1470 	igmp = mtod(m, struct igmp *);
1471 	if (in_cksum(m, igmplen)) {
1472 		IGMPSTAT_INC(igps_rcv_badsum);
1473 		m_freem(m);
1474 		return (IPPROTO_DONE);
1475 	}
1476 	m->m_data -= iphlen;
1477 	m->m_len += iphlen;
1478 
1479 	/*
1480 	 * IGMP control traffic is link-scope, and must have a TTL of 1.
1481 	 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1482 	 * probe packets may come from beyond the LAN.
1483 	 */
1484 	if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1485 		IGMPSTAT_INC(igps_rcv_badttl);
1486 		m_freem(m);
1487 		return (IPPROTO_DONE);
1488 	}
1489 
1490 	switch (igmp->igmp_type) {
1491 	case IGMP_HOST_MEMBERSHIP_QUERY:
1492 		if (igmplen == IGMP_MINLEN) {
1493 			if (igmp->igmp_code == 0)
1494 				queryver = IGMP_VERSION_1;
1495 			else
1496 				queryver = IGMP_VERSION_2;
1497 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1498 			queryver = IGMP_VERSION_3;
1499 		} else {
1500 			IGMPSTAT_INC(igps_rcv_tooshort);
1501 			m_freem(m);
1502 			return (IPPROTO_DONE);
1503 		}
1504 
1505 		switch (queryver) {
1506 		case IGMP_VERSION_1:
1507 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1508 			if (!V_igmp_v1enable)
1509 				break;
1510 			if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1511 				m_freem(m);
1512 				return (IPPROTO_DONE);
1513 			}
1514 			break;
1515 
1516 		case IGMP_VERSION_2:
1517 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1518 			if (!V_igmp_v2enable)
1519 				break;
1520 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1521 				m_freem(m);
1522 				return (IPPROTO_DONE);
1523 			}
1524 			break;
1525 
1526 		case IGMP_VERSION_3: {
1527 				struct igmpv3 *igmpv3;
1528 				uint16_t igmpv3len;
1529 				uint16_t nsrc;
1530 
1531 				IGMPSTAT_INC(igps_rcv_v3_queries);
1532 				igmpv3 = (struct igmpv3 *)igmp;
1533 				/*
1534 				 * Validate length based on source count.
1535 				 */
1536 				nsrc = ntohs(igmpv3->igmp_numsrc);
1537 				if (nsrc * sizeof(in_addr_t) >
1538 				    UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1539 					IGMPSTAT_INC(igps_rcv_tooshort);
1540 					return (IPPROTO_DONE);
1541 				}
1542 				/*
1543 				 * m_pullup() may modify m, so pullup in
1544 				 * this scope.
1545 				 */
1546 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1547 				   sizeof(struct in_addr) * nsrc;
1548 				if ((!M_WRITABLE(m) ||
1549 				     m->m_len < igmpv3len) &&
1550 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1551 					IGMPSTAT_INC(igps_rcv_tooshort);
1552 					return (IPPROTO_DONE);
1553 				}
1554 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1555 				    + iphlen);
1556 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1557 					m_freem(m);
1558 					return (IPPROTO_DONE);
1559 				}
1560 			}
1561 			break;
1562 		}
1563 		break;
1564 
1565 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1566 		if (!V_igmp_v1enable)
1567 			break;
1568 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1569 			m_freem(m);
1570 			return (IPPROTO_DONE);
1571 		}
1572 		break;
1573 
1574 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1575 		if (!V_igmp_v2enable)
1576 			break;
1577 		if (!ip_checkrouteralert(m))
1578 			IGMPSTAT_INC(igps_rcv_nora);
1579 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1580 			m_freem(m);
1581 			return (IPPROTO_DONE);
1582 		}
1583 		break;
1584 
1585 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1586 		/*
1587 		 * Hosts do not need to process IGMPv3 membership reports,
1588 		 * as report suppression is no longer required.
1589 		 */
1590 		if (!ip_checkrouteralert(m))
1591 			IGMPSTAT_INC(igps_rcv_nora);
1592 		break;
1593 
1594 	default:
1595 		break;
1596 	}
1597 
1598 	/*
1599 	 * Pass all valid IGMP packets up to any process(es) listening on a
1600 	 * raw IGMP socket.
1601 	 */
1602 	*mp = m;
1603 	return (rip_input(mp, offp, proto));
1604 }
1605 
1606 
1607 /*
1608  * Fast timeout handler (global).
1609  * VIMAGE: Timeout handlers are expected to service all vimages.
1610  */
1611 void
1612 igmp_fasttimo(void)
1613 {
1614 	VNET_ITERATOR_DECL(vnet_iter);
1615 
1616 	VNET_LIST_RLOCK_NOSLEEP();
1617 	VNET_FOREACH(vnet_iter) {
1618 		CURVNET_SET(vnet_iter);
1619 		igmp_fasttimo_vnet();
1620 		CURVNET_RESTORE();
1621 	}
1622 	VNET_LIST_RUNLOCK_NOSLEEP();
1623 }
1624 
1625 /*
1626  * Fast timeout handler (per-vnet).
1627  * Sends are shuffled off to a netisr to deal with Giant.
1628  *
1629  * VIMAGE: Assume caller has set up our curvnet.
1630  */
1631 static void
1632 igmp_fasttimo_vnet(void)
1633 {
1634 	struct mbufq		 scq;	/* State-change packets */
1635 	struct mbufq		 qrq;	/* Query response packets */
1636 	struct ifnet		*ifp;
1637 	struct igmp_ifsoftc	*igi;
1638 	struct ifmultiaddr	*ifma, *next;
1639 	struct in_multi		*inm;
1640 	struct in_multi_head inm_free_tmp;
1641 	int			 loop, uri_fasthz;
1642 
1643 	loop = 0;
1644 	uri_fasthz = 0;
1645 
1646 	/*
1647 	 * Quick check to see if any work needs to be done, in order to
1648 	 * minimize the overhead of fasttimo processing.
1649 	 * SMPng: XXX Unlocked reads.
1650 	 */
1651 	if (!V_current_state_timers_running &&
1652 	    !V_interface_timers_running &&
1653 	    !V_state_change_timers_running)
1654 		return;
1655 
1656 	SLIST_INIT(&inm_free_tmp);
1657 	IN_MULTI_LIST_LOCK();
1658 	IGMP_LOCK();
1659 
1660 	/*
1661 	 * IGMPv3 General Query response timer processing.
1662 	 */
1663 	if (V_interface_timers_running) {
1664 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1665 
1666 		V_interface_timers_running = 0;
1667 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1668 			if (igi->igi_v3_timer == 0) {
1669 				/* Do nothing. */
1670 			} else if (--igi->igi_v3_timer == 0) {
1671 				igmp_v3_dispatch_general_query(igi);
1672 			} else {
1673 				V_interface_timers_running = 1;
1674 			}
1675 		}
1676 	}
1677 
1678 	if (!V_current_state_timers_running &&
1679 	    !V_state_change_timers_running)
1680 		goto out_locked;
1681 
1682 	V_current_state_timers_running = 0;
1683 	V_state_change_timers_running = 0;
1684 
1685 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1686 
1687 	/*
1688 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1689 	 * Note: Processing a v3 group timer may remove a node.
1690 	 */
1691 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1692 		ifp = igi->igi_ifp;
1693 
1694 		if (igi->igi_version == IGMP_VERSION_3) {
1695 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1696 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1697 			    PR_FASTHZ);
1698 			mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1699 			mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1700 		}
1701 
1702 		IF_ADDR_WLOCK(ifp);
1703 	restart:
1704 		CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1705 			if (ifma->ifma_addr->sa_family != AF_INET ||
1706 			    ifma->ifma_protospec == NULL)
1707 				continue;
1708 			inm = (struct in_multi *)ifma->ifma_protospec;
1709 			switch (igi->igi_version) {
1710 			case IGMP_VERSION_1:
1711 			case IGMP_VERSION_2:
1712 				igmp_v1v2_process_group_timer(inm,
1713 				    igi->igi_version);
1714 				break;
1715 			case IGMP_VERSION_3:
1716 				igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1717 				    &scq, inm, uri_fasthz);
1718 				break;
1719 			}
1720 			if (__predict_false(ifma_restart)) {
1721 				ifma_restart = false;
1722 				goto restart;
1723 			}
1724 		}
1725 		IF_ADDR_WUNLOCK(ifp);
1726 
1727 		if (igi->igi_version == IGMP_VERSION_3) {
1728 			igmp_dispatch_queue(&qrq, 0, loop);
1729 			igmp_dispatch_queue(&scq, 0, loop);
1730 
1731 			/*
1732 			 * Free the in_multi reference(s) for this
1733 			 * IGMP lifecycle.
1734 			 */
1735 			inm_release_list_deferred(&inm_free_tmp);
1736 		}
1737 	}
1738 
1739 out_locked:
1740 	IGMP_UNLOCK();
1741 	IN_MULTI_LIST_UNLOCK();
1742 }
1743 
1744 /*
1745  * Update host report group timer for IGMPv1/v2.
1746  * Will update the global pending timer flags.
1747  */
1748 static void
1749 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1750 {
1751 	int report_timer_expired;
1752 
1753 	IN_MULTI_LIST_LOCK_ASSERT();
1754 	IGMP_LOCK_ASSERT();
1755 
1756 	if (inm->inm_timer == 0) {
1757 		report_timer_expired = 0;
1758 	} else if (--inm->inm_timer == 0) {
1759 		report_timer_expired = 1;
1760 	} else {
1761 		V_current_state_timers_running = 1;
1762 		return;
1763 	}
1764 
1765 	switch (inm->inm_state) {
1766 	case IGMP_NOT_MEMBER:
1767 	case IGMP_SILENT_MEMBER:
1768 	case IGMP_IDLE_MEMBER:
1769 	case IGMP_LAZY_MEMBER:
1770 	case IGMP_SLEEPING_MEMBER:
1771 	case IGMP_AWAKENING_MEMBER:
1772 		break;
1773 	case IGMP_REPORTING_MEMBER:
1774 		if (report_timer_expired) {
1775 			inm->inm_state = IGMP_IDLE_MEMBER;
1776 			(void)igmp_v1v2_queue_report(inm,
1777 			    (version == IGMP_VERSION_2) ?
1778 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1779 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1780 		}
1781 		break;
1782 	case IGMP_G_QUERY_PENDING_MEMBER:
1783 	case IGMP_SG_QUERY_PENDING_MEMBER:
1784 	case IGMP_LEAVING_MEMBER:
1785 		break;
1786 	}
1787 }
1788 
1789 /*
1790  * Update a group's timers for IGMPv3.
1791  * Will update the global pending timer flags.
1792  * Note: Unlocked read from igi.
1793  */
1794 static void
1795 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1796     struct mbufq *qrq, struct mbufq *scq,
1797     struct in_multi *inm, const int uri_fasthz)
1798 {
1799 	int query_response_timer_expired;
1800 	int state_change_retransmit_timer_expired;
1801 
1802 	IN_MULTI_LIST_LOCK_ASSERT();
1803 	IGMP_LOCK_ASSERT();
1804 
1805 	query_response_timer_expired = 0;
1806 	state_change_retransmit_timer_expired = 0;
1807 
1808 	/*
1809 	 * During a transition from v1/v2 compatibility mode back to v3,
1810 	 * a group record in REPORTING state may still have its group
1811 	 * timer active. This is a no-op in this function; it is easier
1812 	 * to deal with it here than to complicate the slow-timeout path.
1813 	 */
1814 	if (inm->inm_timer == 0) {
1815 		query_response_timer_expired = 0;
1816 	} else if (--inm->inm_timer == 0) {
1817 		query_response_timer_expired = 1;
1818 	} else {
1819 		V_current_state_timers_running = 1;
1820 	}
1821 
1822 	if (inm->inm_sctimer == 0) {
1823 		state_change_retransmit_timer_expired = 0;
1824 	} else if (--inm->inm_sctimer == 0) {
1825 		state_change_retransmit_timer_expired = 1;
1826 	} else {
1827 		V_state_change_timers_running = 1;
1828 	}
1829 
1830 	/* We are in fasttimo, so be quick about it. */
1831 	if (!state_change_retransmit_timer_expired &&
1832 	    !query_response_timer_expired)
1833 		return;
1834 
1835 	switch (inm->inm_state) {
1836 	case IGMP_NOT_MEMBER:
1837 	case IGMP_SILENT_MEMBER:
1838 	case IGMP_SLEEPING_MEMBER:
1839 	case IGMP_LAZY_MEMBER:
1840 	case IGMP_AWAKENING_MEMBER:
1841 	case IGMP_IDLE_MEMBER:
1842 		break;
1843 	case IGMP_G_QUERY_PENDING_MEMBER:
1844 	case IGMP_SG_QUERY_PENDING_MEMBER:
1845 		/*
1846 		 * Respond to a previously pending Group-Specific
1847 		 * or Group-and-Source-Specific query by enqueueing
1848 		 * the appropriate Current-State report for
1849 		 * immediate transmission.
1850 		 */
1851 		if (query_response_timer_expired) {
1852 			int retval __unused;
1853 
1854 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1855 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1856 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1857 			    __func__, retval);
1858 			inm->inm_state = IGMP_REPORTING_MEMBER;
1859 			/* XXX Clear recorded sources for next time. */
1860 			inm_clear_recorded(inm);
1861 		}
1862 		/* FALLTHROUGH */
1863 	case IGMP_REPORTING_MEMBER:
1864 	case IGMP_LEAVING_MEMBER:
1865 		if (state_change_retransmit_timer_expired) {
1866 			/*
1867 			 * State-change retransmission timer fired.
1868 			 * If there are any further pending retransmissions,
1869 			 * set the global pending state-change flag, and
1870 			 * reset the timer.
1871 			 */
1872 			if (--inm->inm_scrv > 0) {
1873 				inm->inm_sctimer = uri_fasthz;
1874 				V_state_change_timers_running = 1;
1875 			}
1876 			/*
1877 			 * Retransmit the previously computed state-change
1878 			 * report. If there are no further pending
1879 			 * retransmissions, the mbuf queue will be consumed.
1880 			 * Update T0 state to T1 as we have now sent
1881 			 * a state-change.
1882 			 */
1883 			(void)igmp_v3_merge_state_changes(inm, scq);
1884 
1885 			inm_commit(inm);
1886 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1887 			    ntohl(inm->inm_addr.s_addr),
1888 			    inm->inm_ifp->if_xname);
1889 
1890 			/*
1891 			 * If we are leaving the group for good, make sure
1892 			 * we release IGMP's reference to it.
1893 			 * This release must be deferred using a SLIST,
1894 			 * as we are called from a loop which traverses
1895 			 * the in_ifmultiaddr TAILQ.
1896 			 */
1897 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1898 			    inm->inm_scrv == 0) {
1899 				inm->inm_state = IGMP_NOT_MEMBER;
1900 				inm_rele_locked(inmh, inm);
1901 			}
1902 		}
1903 		break;
1904 	}
1905 }
1906 
1907 
1908 /*
1909  * Suppress a group's pending response to a group or source/group query.
1910  *
1911  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1912  * Do NOT update ST1/ST0 as this operation merely suppresses
1913  * the currently pending group record.
1914  * Do NOT suppress the response to a general query. It is possible but
1915  * it would require adding another state or flag.
1916  */
1917 static void
1918 igmp_v3_suppress_group_record(struct in_multi *inm)
1919 {
1920 
1921 	IN_MULTI_LIST_LOCK_ASSERT();
1922 
1923 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1924 		("%s: not IGMPv3 mode on link", __func__));
1925 
1926 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1927 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1928 		return;
1929 
1930 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1931 		inm_clear_recorded(inm);
1932 
1933 	inm->inm_timer = 0;
1934 	inm->inm_state = IGMP_REPORTING_MEMBER;
1935 }
1936 
1937 /*
1938  * Switch to a different IGMP version on the given interface,
1939  * as per Section 7.2.1.
1940  */
1941 static void
1942 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1943 {
1944 	int old_version_timer;
1945 
1946 	IGMP_LOCK_ASSERT();
1947 
1948 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1949 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1950 
1951 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1952 		/*
1953 		 * Compute the "Older Version Querier Present" timer as per
1954 		 * Section 8.12.
1955 		 */
1956 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1957 		old_version_timer *= PR_SLOWHZ;
1958 
1959 		if (version == IGMP_VERSION_1) {
1960 			igi->igi_v1_timer = old_version_timer;
1961 			igi->igi_v2_timer = 0;
1962 		} else if (version == IGMP_VERSION_2) {
1963 			igi->igi_v1_timer = 0;
1964 			igi->igi_v2_timer = old_version_timer;
1965 		}
1966 	}
1967 
1968 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1969 		if (igi->igi_version != IGMP_VERSION_2) {
1970 			igi->igi_version = IGMP_VERSION_2;
1971 			igmp_v3_cancel_link_timers(igi);
1972 		}
1973 	} else if (igi->igi_v1_timer > 0) {
1974 		if (igi->igi_version != IGMP_VERSION_1) {
1975 			igi->igi_version = IGMP_VERSION_1;
1976 			igmp_v3_cancel_link_timers(igi);
1977 		}
1978 	}
1979 }
1980 
1981 /*
1982  * Cancel pending IGMPv3 timers for the given link and all groups
1983  * joined on it; state-change, general-query, and group-query timers.
1984  *
1985  * Only ever called on a transition from v3 to Compatibility mode. Kill
1986  * the timers stone dead (this may be expensive for large N groups), they
1987  * will be restarted if Compatibility Mode deems that they must be due to
1988  * query processing.
1989  */
1990 static void
1991 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
1992 {
1993 	struct ifmultiaddr	*ifma;
1994 	struct ifnet		*ifp;
1995 	struct in_multi		*inm;
1996 	struct in_multi_head inm_free_tmp;
1997 	struct epoch_tracker et;
1998 
1999 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2000 	    igi->igi_ifp, igi->igi_ifp->if_xname);
2001 
2002 	IN_MULTI_LIST_LOCK_ASSERT();
2003 	IGMP_LOCK_ASSERT();
2004 	SLIST_INIT(&inm_free_tmp);
2005 
2006 	/*
2007 	 * Stop the v3 General Query Response on this link stone dead.
2008 	 * If fasttimo is woken up due to V_interface_timers_running,
2009 	 * the flag will be cleared if there are no pending link timers.
2010 	 */
2011 	igi->igi_v3_timer = 0;
2012 
2013 	/*
2014 	 * Now clear the current-state and state-change report timers
2015 	 * for all memberships scoped to this link.
2016 	 */
2017 	ifp = igi->igi_ifp;
2018 	NET_EPOCH_ENTER(et);
2019 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2020 		if (ifma->ifma_addr->sa_family != AF_INET ||
2021 		    ifma->ifma_protospec == NULL)
2022 			continue;
2023 		inm = (struct in_multi *)ifma->ifma_protospec;
2024 		switch (inm->inm_state) {
2025 		case IGMP_NOT_MEMBER:
2026 		case IGMP_SILENT_MEMBER:
2027 		case IGMP_IDLE_MEMBER:
2028 		case IGMP_LAZY_MEMBER:
2029 		case IGMP_SLEEPING_MEMBER:
2030 		case IGMP_AWAKENING_MEMBER:
2031 			/*
2032 			 * These states are either not relevant in v3 mode,
2033 			 * or are unreported. Do nothing.
2034 			 */
2035 			break;
2036 		case IGMP_LEAVING_MEMBER:
2037 			/*
2038 			 * If we are leaving the group and switching to
2039 			 * compatibility mode, we need to release the final
2040 			 * reference held for issuing the INCLUDE {}, and
2041 			 * transition to REPORTING to ensure the host leave
2042 			 * message is sent upstream to the old querier --
2043 			 * transition to NOT would lose the leave and race.
2044 			 */
2045 			inm_rele_locked(&inm_free_tmp, inm);
2046 			/* FALLTHROUGH */
2047 		case IGMP_G_QUERY_PENDING_MEMBER:
2048 		case IGMP_SG_QUERY_PENDING_MEMBER:
2049 			inm_clear_recorded(inm);
2050 			/* FALLTHROUGH */
2051 		case IGMP_REPORTING_MEMBER:
2052 			inm->inm_state = IGMP_REPORTING_MEMBER;
2053 			break;
2054 		}
2055 		/*
2056 		 * Always clear state-change and group report timers.
2057 		 * Free any pending IGMPv3 state-change records.
2058 		 */
2059 		inm->inm_sctimer = 0;
2060 		inm->inm_timer = 0;
2061 		mbufq_drain(&inm->inm_scq);
2062 	}
2063 	NET_EPOCH_EXIT(et);
2064 
2065 	inm_release_list_deferred(&inm_free_tmp);
2066 }
2067 
2068 /*
2069  * Update the Older Version Querier Present timers for a link.
2070  * See Section 7.2.1 of RFC 3376.
2071  */
2072 static void
2073 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2074 {
2075 
2076 	IGMP_LOCK_ASSERT();
2077 
2078 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2079 		/*
2080 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2081 		 *
2082 		 * Revert to IGMPv3.
2083 		 */
2084 		if (igi->igi_version != IGMP_VERSION_3) {
2085 			CTR5(KTR_IGMPV3,
2086 			    "%s: transition from v%d -> v%d on %p(%s)",
2087 			    __func__, igi->igi_version, IGMP_VERSION_3,
2088 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2089 			igi->igi_version = IGMP_VERSION_3;
2090 		}
2091 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2092 		/*
2093 		 * IGMPv1 Querier Present timer expired,
2094 		 * IGMPv2 Querier Present timer running.
2095 		 * If IGMPv2 was disabled since last timeout,
2096 		 * revert to IGMPv3.
2097 		 * If IGMPv2 is enabled, revert to IGMPv2.
2098 		 */
2099 		if (!V_igmp_v2enable) {
2100 			CTR5(KTR_IGMPV3,
2101 			    "%s: transition from v%d -> v%d on %p(%s)",
2102 			    __func__, igi->igi_version, IGMP_VERSION_3,
2103 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2104 			igi->igi_v2_timer = 0;
2105 			igi->igi_version = IGMP_VERSION_3;
2106 		} else {
2107 			--igi->igi_v2_timer;
2108 			if (igi->igi_version != IGMP_VERSION_2) {
2109 				CTR5(KTR_IGMPV3,
2110 				    "%s: transition from v%d -> v%d on %p(%s)",
2111 				    __func__, igi->igi_version, IGMP_VERSION_2,
2112 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2113 				igi->igi_version = IGMP_VERSION_2;
2114 				igmp_v3_cancel_link_timers(igi);
2115 			}
2116 		}
2117 	} else if (igi->igi_v1_timer > 0) {
2118 		/*
2119 		 * IGMPv1 Querier Present timer running.
2120 		 * Stop IGMPv2 timer if running.
2121 		 *
2122 		 * If IGMPv1 was disabled since last timeout,
2123 		 * revert to IGMPv3.
2124 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2125 		 */
2126 		if (!V_igmp_v1enable) {
2127 			CTR5(KTR_IGMPV3,
2128 			    "%s: transition from v%d -> v%d on %p(%s)",
2129 			    __func__, igi->igi_version, IGMP_VERSION_3,
2130 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2131 			igi->igi_v1_timer = 0;
2132 			igi->igi_version = IGMP_VERSION_3;
2133 		} else {
2134 			--igi->igi_v1_timer;
2135 		}
2136 		if (igi->igi_v2_timer > 0) {
2137 			CTR3(KTR_IGMPV3,
2138 			    "%s: cancel v2 timer on %p(%s)",
2139 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2140 			igi->igi_v2_timer = 0;
2141 		}
2142 	}
2143 }
2144 
2145 /*
2146  * Global slowtimo handler.
2147  * VIMAGE: Timeout handlers are expected to service all vimages.
2148  */
2149 void
2150 igmp_slowtimo(void)
2151 {
2152 	VNET_ITERATOR_DECL(vnet_iter);
2153 
2154 	VNET_LIST_RLOCK_NOSLEEP();
2155 	VNET_FOREACH(vnet_iter) {
2156 		CURVNET_SET(vnet_iter);
2157 		igmp_slowtimo_vnet();
2158 		CURVNET_RESTORE();
2159 	}
2160 	VNET_LIST_RUNLOCK_NOSLEEP();
2161 }
2162 
2163 /*
2164  * Per-vnet slowtimo handler.
2165  */
2166 static void
2167 igmp_slowtimo_vnet(void)
2168 {
2169 	struct igmp_ifsoftc *igi;
2170 
2171 	IGMP_LOCK();
2172 
2173 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2174 		igmp_v1v2_process_querier_timers(igi);
2175 	}
2176 
2177 	IGMP_UNLOCK();
2178 }
2179 
2180 /*
2181  * Dispatch an IGMPv1/v2 host report or leave message.
2182  * These are always small enough to fit inside a single mbuf.
2183  */
2184 static int
2185 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2186 {
2187 	struct ifnet		*ifp;
2188 	struct igmp		*igmp;
2189 	struct ip		*ip;
2190 	struct mbuf		*m;
2191 
2192 	IN_MULTI_LIST_LOCK_ASSERT();
2193 	IGMP_LOCK_ASSERT();
2194 
2195 	ifp = inm->inm_ifp;
2196 
2197 	m = m_gethdr(M_NOWAIT, MT_DATA);
2198 	if (m == NULL)
2199 		return (ENOMEM);
2200 	M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2201 
2202 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2203 
2204 	m->m_data += sizeof(struct ip);
2205 	m->m_len = sizeof(struct igmp);
2206 
2207 	igmp = mtod(m, struct igmp *);
2208 	igmp->igmp_type = type;
2209 	igmp->igmp_code = 0;
2210 	igmp->igmp_group = inm->inm_addr;
2211 	igmp->igmp_cksum = 0;
2212 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2213 
2214 	m->m_data -= sizeof(struct ip);
2215 	m->m_len += sizeof(struct ip);
2216 
2217 	ip = mtod(m, struct ip *);
2218 	ip->ip_tos = 0;
2219 	ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2220 	ip->ip_off = 0;
2221 	ip->ip_p = IPPROTO_IGMP;
2222 	ip->ip_src.s_addr = INADDR_ANY;
2223 
2224 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2225 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2226 	else
2227 		ip->ip_dst = inm->inm_addr;
2228 
2229 	igmp_save_context(m, ifp);
2230 
2231 	m->m_flags |= M_IGMPV2;
2232 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2233 		m->m_flags |= M_IGMP_LOOP;
2234 
2235 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2236 	netisr_dispatch(NETISR_IGMP, m);
2237 
2238 	return (0);
2239 }
2240 
2241 /*
2242  * Process a state change from the upper layer for the given IPv4 group.
2243  *
2244  * Each socket holds a reference on the in_multi in its own ip_moptions.
2245  * The socket layer will have made the necessary updates to.the group
2246  * state, it is now up to IGMP to issue a state change report if there
2247  * has been any change between T0 (when the last state-change was issued)
2248  * and T1 (now).
2249  *
2250  * We use the IGMPv3 state machine at group level. The IGMP module
2251  * however makes the decision as to which IGMP protocol version to speak.
2252  * A state change *from* INCLUDE {} always means an initial join.
2253  * A state change *to* INCLUDE {} always means a final leave.
2254  *
2255  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2256  * save ourselves a bunch of work; any exclusive mode groups need not
2257  * compute source filter lists.
2258  *
2259  * VIMAGE: curvnet should have been set by caller, as this routine
2260  * is called from the socket option handlers.
2261  */
2262 int
2263 igmp_change_state(struct in_multi *inm)
2264 {
2265 	struct igmp_ifsoftc *igi;
2266 	struct ifnet *ifp;
2267 	int error;
2268 
2269 	error = 0;
2270 	IN_MULTI_LOCK_ASSERT();
2271 	/*
2272 	 * Try to detect if the upper layer just asked us to change state
2273 	 * for an interface which has now gone away.
2274 	 */
2275 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2276 	ifp = inm->inm_ifma->ifma_ifp;
2277 	/*
2278 	 * Sanity check that netinet's notion of ifp is the
2279 	 * same as net's.
2280 	 */
2281 	KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2282 
2283 	IGMP_LOCK();
2284 
2285 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2286 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2287 
2288 	/*
2289 	 * If we detect a state transition to or from MCAST_UNDEFINED
2290 	 * for this group, then we are starting or finishing an IGMP
2291 	 * life cycle for this group.
2292 	 */
2293 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2294 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2295 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2296 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2297 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2298 			error = igmp_initial_join(inm, igi);
2299 			goto out_locked;
2300 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2301 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2302 			igmp_final_leave(inm, igi);
2303 			goto out_locked;
2304 		}
2305 	} else {
2306 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2307 	}
2308 
2309 	error = igmp_handle_state_change(inm, igi);
2310 
2311 out_locked:
2312 	IGMP_UNLOCK();
2313 	return (error);
2314 }
2315 
2316 /*
2317  * Perform the initial join for an IGMP group.
2318  *
2319  * When joining a group:
2320  *  If the group should have its IGMP traffic suppressed, do nothing.
2321  *  IGMPv1 starts sending IGMPv1 host membership reports.
2322  *  IGMPv2 starts sending IGMPv2 host membership reports.
2323  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2324  *  initial state of the membership.
2325  */
2326 static int
2327 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2328 {
2329 	struct ifnet		*ifp;
2330 	struct mbufq		*mq;
2331 	int			 error, retval, syncstates;
2332 
2333 	CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2334 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2335 
2336 	error = 0;
2337 	syncstates = 1;
2338 
2339 	ifp = inm->inm_ifp;
2340 
2341 	IN_MULTI_LOCK_ASSERT();
2342 	IGMP_LOCK_ASSERT();
2343 
2344 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2345 
2346 	/*
2347 	 * Groups joined on loopback or marked as 'not reported',
2348 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2349 	 * are never reported in any IGMP protocol exchanges.
2350 	 * All other groups enter the appropriate IGMP state machine
2351 	 * for the version in use on this link.
2352 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2353 	 * disabled for the link.
2354 	 */
2355 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2356 	    (igi->igi_flags & IGIF_SILENT) ||
2357 	    !igmp_isgroupreported(inm->inm_addr)) {
2358 		CTR1(KTR_IGMPV3,
2359 "%s: not kicking state machine for silent group", __func__);
2360 		inm->inm_state = IGMP_SILENT_MEMBER;
2361 		inm->inm_timer = 0;
2362 	} else {
2363 		/*
2364 		 * Deal with overlapping in_multi lifecycle.
2365 		 * If this group was LEAVING, then make sure
2366 		 * we drop the reference we picked up to keep the
2367 		 * group around for the final INCLUDE {} enqueue.
2368 		 */
2369 		if (igi->igi_version == IGMP_VERSION_3 &&
2370 		    inm->inm_state == IGMP_LEAVING_MEMBER) {
2371 			MPASS(inm->inm_refcount > 1);
2372 			inm_rele_locked(NULL, inm);
2373 		}
2374 		inm->inm_state = IGMP_REPORTING_MEMBER;
2375 
2376 		switch (igi->igi_version) {
2377 		case IGMP_VERSION_1:
2378 		case IGMP_VERSION_2:
2379 			inm->inm_state = IGMP_IDLE_MEMBER;
2380 			error = igmp_v1v2_queue_report(inm,
2381 			    (igi->igi_version == IGMP_VERSION_2) ?
2382 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2383 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2384 			if (error == 0) {
2385 				inm->inm_timer = IGMP_RANDOM_DELAY(
2386 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2387 				V_current_state_timers_running = 1;
2388 			}
2389 			break;
2390 
2391 		case IGMP_VERSION_3:
2392 			/*
2393 			 * Defer update of T0 to T1, until the first copy
2394 			 * of the state change has been transmitted.
2395 			 */
2396 			syncstates = 0;
2397 
2398 			/*
2399 			 * Immediately enqueue a State-Change Report for
2400 			 * this interface, freeing any previous reports.
2401 			 * Don't kick the timers if there is nothing to do,
2402 			 * or if an error occurred.
2403 			 */
2404 			mq = &inm->inm_scq;
2405 			mbufq_drain(mq);
2406 			retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2407 			    0, 0);
2408 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2409 			    __func__, retval);
2410 			if (retval <= 0) {
2411 				error = retval * -1;
2412 				break;
2413 			}
2414 
2415 			/*
2416 			 * Schedule transmission of pending state-change
2417 			 * report up to RV times for this link. The timer
2418 			 * will fire at the next igmp_fasttimo (~200ms),
2419 			 * giving us an opportunity to merge the reports.
2420 			 */
2421 			if (igi->igi_flags & IGIF_LOOPBACK) {
2422 				inm->inm_scrv = 1;
2423 			} else {
2424 				KASSERT(igi->igi_rv > 1,
2425 				   ("%s: invalid robustness %d", __func__,
2426 				    igi->igi_rv));
2427 				inm->inm_scrv = igi->igi_rv;
2428 			}
2429 			inm->inm_sctimer = 1;
2430 			V_state_change_timers_running = 1;
2431 
2432 			error = 0;
2433 			break;
2434 		}
2435 	}
2436 
2437 	/*
2438 	 * Only update the T0 state if state change is atomic,
2439 	 * i.e. we don't need to wait for a timer to fire before we
2440 	 * can consider the state change to have been communicated.
2441 	 */
2442 	if (syncstates) {
2443 		inm_commit(inm);
2444 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2445 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2446 	}
2447 
2448 	return (error);
2449 }
2450 
2451 /*
2452  * Issue an intermediate state change during the IGMP life-cycle.
2453  */
2454 static int
2455 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2456 {
2457 	struct ifnet		*ifp;
2458 	int			 retval;
2459 
2460 	CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2461 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2462 
2463 	ifp = inm->inm_ifp;
2464 
2465 	IN_MULTI_LIST_LOCK_ASSERT();
2466 	IGMP_LOCK_ASSERT();
2467 
2468 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2469 
2470 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2471 	    (igi->igi_flags & IGIF_SILENT) ||
2472 	    !igmp_isgroupreported(inm->inm_addr) ||
2473 	    (igi->igi_version != IGMP_VERSION_3)) {
2474 		if (!igmp_isgroupreported(inm->inm_addr)) {
2475 			CTR1(KTR_IGMPV3,
2476 "%s: not kicking state machine for silent group", __func__);
2477 		}
2478 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2479 		inm_commit(inm);
2480 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2481 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2482 		return (0);
2483 	}
2484 
2485 	mbufq_drain(&inm->inm_scq);
2486 
2487 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2488 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2489 	if (retval <= 0)
2490 		return (-retval);
2491 
2492 	/*
2493 	 * If record(s) were enqueued, start the state-change
2494 	 * report timer for this group.
2495 	 */
2496 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2497 	inm->inm_sctimer = 1;
2498 	V_state_change_timers_running = 1;
2499 
2500 	return (0);
2501 }
2502 
2503 /*
2504  * Perform the final leave for an IGMP group.
2505  *
2506  * When leaving a group:
2507  *  IGMPv1 does nothing.
2508  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2509  *  IGMPv3 enqueues a state-change report containing a transition
2510  *  to INCLUDE {} for immediate transmission.
2511  */
2512 static void
2513 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2514 {
2515 	int syncstates;
2516 
2517 	syncstates = 1;
2518 
2519 	CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2520 	    __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2521 	    inm->inm_ifp->if_xname);
2522 
2523 	IN_MULTI_LIST_LOCK_ASSERT();
2524 	IGMP_LOCK_ASSERT();
2525 
2526 	switch (inm->inm_state) {
2527 	case IGMP_NOT_MEMBER:
2528 	case IGMP_SILENT_MEMBER:
2529 	case IGMP_LEAVING_MEMBER:
2530 		/* Already leaving or left; do nothing. */
2531 		CTR1(KTR_IGMPV3,
2532 "%s: not kicking state machine for silent group", __func__);
2533 		break;
2534 	case IGMP_REPORTING_MEMBER:
2535 	case IGMP_IDLE_MEMBER:
2536 	case IGMP_G_QUERY_PENDING_MEMBER:
2537 	case IGMP_SG_QUERY_PENDING_MEMBER:
2538 		if (igi->igi_version == IGMP_VERSION_2) {
2539 #ifdef INVARIANTS
2540 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2541 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2542 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2543 			     __func__);
2544 #endif
2545 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2546 			inm->inm_state = IGMP_NOT_MEMBER;
2547 		} else if (igi->igi_version == IGMP_VERSION_3) {
2548 			/*
2549 			 * Stop group timer and all pending reports.
2550 			 * Immediately enqueue a state-change report
2551 			 * TO_IN {} to be sent on the next fast timeout,
2552 			 * giving us an opportunity to merge reports.
2553 			 */
2554 			mbufq_drain(&inm->inm_scq);
2555 			inm->inm_timer = 0;
2556 			if (igi->igi_flags & IGIF_LOOPBACK) {
2557 				inm->inm_scrv = 1;
2558 			} else {
2559 				inm->inm_scrv = igi->igi_rv;
2560 			}
2561 			CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2562 			    "pending retransmissions.", __func__,
2563 			    ntohl(inm->inm_addr.s_addr),
2564 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2565 			if (inm->inm_scrv == 0) {
2566 				inm->inm_state = IGMP_NOT_MEMBER;
2567 				inm->inm_sctimer = 0;
2568 			} else {
2569 				int retval __unused;
2570 
2571 				inm_acquire_locked(inm);
2572 
2573 				retval = igmp_v3_enqueue_group_record(
2574 				    &inm->inm_scq, inm, 1, 0, 0);
2575 				KASSERT(retval != 0,
2576 				    ("%s: enqueue record = %d", __func__,
2577 				     retval));
2578 
2579 				inm->inm_state = IGMP_LEAVING_MEMBER;
2580 				inm->inm_sctimer = 1;
2581 				V_state_change_timers_running = 1;
2582 				syncstates = 0;
2583 			}
2584 			break;
2585 		}
2586 		break;
2587 	case IGMP_LAZY_MEMBER:
2588 	case IGMP_SLEEPING_MEMBER:
2589 	case IGMP_AWAKENING_MEMBER:
2590 		/* Our reports are suppressed; do nothing. */
2591 		break;
2592 	}
2593 
2594 	if (syncstates) {
2595 		inm_commit(inm);
2596 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2597 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2598 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2599 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2600 		    __func__, ntohl(inm->inm_addr.s_addr),
2601 		    inm->inm_ifp->if_xname);
2602 	}
2603 }
2604 
2605 /*
2606  * Enqueue an IGMPv3 group record to the given output queue.
2607  *
2608  * XXX This function could do with having the allocation code
2609  * split out, and the multiple-tree-walks coalesced into a single
2610  * routine as has been done in igmp_v3_enqueue_filter_change().
2611  *
2612  * If is_state_change is zero, a current-state record is appended.
2613  * If is_state_change is non-zero, a state-change report is appended.
2614  *
2615  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2616  * If is_group_query is zero, and if there is a packet with free space
2617  * at the tail of the queue, it will be appended to providing there
2618  * is enough free space.
2619  * Otherwise a new mbuf packet chain is allocated.
2620  *
2621  * If is_source_query is non-zero, each source is checked to see if
2622  * it was recorded for a Group-Source query, and will be omitted if
2623  * it is not both in-mode and recorded.
2624  *
2625  * The function will attempt to allocate leading space in the packet
2626  * for the IP/IGMP header to be prepended without fragmenting the chain.
2627  *
2628  * If successful the size of all data appended to the queue is returned,
2629  * otherwise an error code less than zero is returned, or zero if
2630  * no record(s) were appended.
2631  */
2632 static int
2633 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2634     const int is_state_change, const int is_group_query,
2635     const int is_source_query)
2636 {
2637 	struct igmp_grouprec	 ig;
2638 	struct igmp_grouprec	*pig;
2639 	struct ifnet		*ifp;
2640 	struct ip_msource	*ims, *nims;
2641 	struct mbuf		*m0, *m, *md;
2642 	int			 is_filter_list_change;
2643 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2644 	int			 record_has_sources;
2645 	int			 now;
2646 	int			 type;
2647 	in_addr_t		 naddr;
2648 	uint8_t			 mode;
2649 
2650 	IN_MULTI_LIST_LOCK_ASSERT();
2651 
2652 	ifp = inm->inm_ifp;
2653 	is_filter_list_change = 0;
2654 	m = NULL;
2655 	m0 = NULL;
2656 	m0srcs = 0;
2657 	msrcs = 0;
2658 	nbytes = 0;
2659 	nims = NULL;
2660 	record_has_sources = 1;
2661 	pig = NULL;
2662 	type = IGMP_DO_NOTHING;
2663 	mode = inm->inm_st[1].iss_fmode;
2664 
2665 	/*
2666 	 * If we did not transition out of ASM mode during t0->t1,
2667 	 * and there are no source nodes to process, we can skip
2668 	 * the generation of source records.
2669 	 */
2670 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2671 	    inm->inm_nsrc == 0)
2672 		record_has_sources = 0;
2673 
2674 	if (is_state_change) {
2675 		/*
2676 		 * Queue a state change record.
2677 		 * If the mode did not change, and there are non-ASM
2678 		 * listeners or source filters present,
2679 		 * we potentially need to issue two records for the group.
2680 		 * If we are transitioning to MCAST_UNDEFINED, we need
2681 		 * not send any sources.
2682 		 * If there are ASM listeners, and there was no filter
2683 		 * mode transition of any kind, do nothing.
2684 		 */
2685 		if (mode != inm->inm_st[0].iss_fmode) {
2686 			if (mode == MCAST_EXCLUDE) {
2687 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2688 				    __func__);
2689 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2690 			} else {
2691 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2692 				    __func__);
2693 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2694 				if (mode == MCAST_UNDEFINED)
2695 					record_has_sources = 0;
2696 			}
2697 		} else {
2698 			if (record_has_sources) {
2699 				is_filter_list_change = 1;
2700 			} else {
2701 				type = IGMP_DO_NOTHING;
2702 			}
2703 		}
2704 	} else {
2705 		/*
2706 		 * Queue a current state record.
2707 		 */
2708 		if (mode == MCAST_EXCLUDE) {
2709 			type = IGMP_MODE_IS_EXCLUDE;
2710 		} else if (mode == MCAST_INCLUDE) {
2711 			type = IGMP_MODE_IS_INCLUDE;
2712 			KASSERT(inm->inm_st[1].iss_asm == 0,
2713 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2714 			     __func__, inm, inm->inm_st[1].iss_asm));
2715 		}
2716 	}
2717 
2718 	/*
2719 	 * Generate the filter list changes using a separate function.
2720 	 */
2721 	if (is_filter_list_change)
2722 		return (igmp_v3_enqueue_filter_change(mq, inm));
2723 
2724 	if (type == IGMP_DO_NOTHING) {
2725 		CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2726 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2727 		return (0);
2728 	}
2729 
2730 	/*
2731 	 * If any sources are present, we must be able to fit at least
2732 	 * one in the trailing space of the tail packet's mbuf,
2733 	 * ideally more.
2734 	 */
2735 	minrec0len = sizeof(struct igmp_grouprec);
2736 	if (record_has_sources)
2737 		minrec0len += sizeof(in_addr_t);
2738 
2739 	CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2740 	    igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2741 	    inm->inm_ifp->if_xname);
2742 
2743 	/*
2744 	 * Check if we have a packet in the tail of the queue for this
2745 	 * group into which the first group record for this group will fit.
2746 	 * Otherwise allocate a new packet.
2747 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2748 	 * Note: Group records for G/GSR query responses MUST be sent
2749 	 * in their own packet.
2750 	 */
2751 	m0 = mbufq_last(mq);
2752 	if (!is_group_query &&
2753 	    m0 != NULL &&
2754 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2755 	    (m0->m_pkthdr.len + minrec0len) <
2756 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2757 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2758 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2759 		m = m0;
2760 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2761 	} else {
2762 		if (mbufq_full(mq)) {
2763 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2764 			return (-ENOMEM);
2765 		}
2766 		m = NULL;
2767 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2768 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2769 		if (!is_state_change && !is_group_query) {
2770 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2771 			if (m)
2772 				m->m_data += IGMP_LEADINGSPACE;
2773 		}
2774 		if (m == NULL) {
2775 			m = m_gethdr(M_NOWAIT, MT_DATA);
2776 			if (m)
2777 				M_ALIGN(m, IGMP_LEADINGSPACE);
2778 		}
2779 		if (m == NULL)
2780 			return (-ENOMEM);
2781 
2782 		igmp_save_context(m, ifp);
2783 
2784 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2785 	}
2786 
2787 	/*
2788 	 * Append group record.
2789 	 * If we have sources, we don't know how many yet.
2790 	 */
2791 	ig.ig_type = type;
2792 	ig.ig_datalen = 0;
2793 	ig.ig_numsrc = 0;
2794 	ig.ig_group = inm->inm_addr;
2795 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2796 		if (m != m0)
2797 			m_freem(m);
2798 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2799 		return (-ENOMEM);
2800 	}
2801 	nbytes += sizeof(struct igmp_grouprec);
2802 
2803 	/*
2804 	 * Append as many sources as will fit in the first packet.
2805 	 * If we are appending to a new packet, the chain allocation
2806 	 * may potentially use clusters; use m_getptr() in this case.
2807 	 * If we are appending to an existing packet, we need to obtain
2808 	 * a pointer to the group record after m_append(), in case a new
2809 	 * mbuf was allocated.
2810 	 * Only append sources which are in-mode at t1. If we are
2811 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2812 	 * include source entries.
2813 	 * Only report recorded sources in our filter set when responding
2814 	 * to a group-source query.
2815 	 */
2816 	if (record_has_sources) {
2817 		if (m == m0) {
2818 			md = m_last(m);
2819 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2820 			    md->m_len - nbytes);
2821 		} else {
2822 			md = m_getptr(m, 0, &off);
2823 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2824 			    off);
2825 		}
2826 		msrcs = 0;
2827 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2828 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2829 			    ims->ims_haddr);
2830 			now = ims_get_mode(inm, ims, 1);
2831 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2832 			if ((now != mode) ||
2833 			    (now == mode && mode == MCAST_UNDEFINED)) {
2834 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2835 				continue;
2836 			}
2837 			if (is_source_query && ims->ims_stp == 0) {
2838 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2839 				    __func__);
2840 				continue;
2841 			}
2842 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2843 			naddr = htonl(ims->ims_haddr);
2844 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2845 				if (m != m0)
2846 					m_freem(m);
2847 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2848 				    __func__);
2849 				return (-ENOMEM);
2850 			}
2851 			nbytes += sizeof(in_addr_t);
2852 			++msrcs;
2853 			if (msrcs == m0srcs)
2854 				break;
2855 		}
2856 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2857 		    msrcs);
2858 		pig->ig_numsrc = htons(msrcs);
2859 		nbytes += (msrcs * sizeof(in_addr_t));
2860 	}
2861 
2862 	if (is_source_query && msrcs == 0) {
2863 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2864 		if (m != m0)
2865 			m_freem(m);
2866 		return (0);
2867 	}
2868 
2869 	/*
2870 	 * We are good to go with first packet.
2871 	 */
2872 	if (m != m0) {
2873 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2874 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2875 		mbufq_enqueue(mq, m);
2876 	} else
2877 		m->m_pkthdr.PH_vt.vt_nrecs++;
2878 
2879 	/*
2880 	 * No further work needed if no source list in packet(s).
2881 	 */
2882 	if (!record_has_sources)
2883 		return (nbytes);
2884 
2885 	/*
2886 	 * Whilst sources remain to be announced, we need to allocate
2887 	 * a new packet and fill out as many sources as will fit.
2888 	 * Always try for a cluster first.
2889 	 */
2890 	while (nims != NULL) {
2891 		if (mbufq_full(mq)) {
2892 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2893 			return (-ENOMEM);
2894 		}
2895 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2896 		if (m)
2897 			m->m_data += IGMP_LEADINGSPACE;
2898 		if (m == NULL) {
2899 			m = m_gethdr(M_NOWAIT, MT_DATA);
2900 			if (m)
2901 				M_ALIGN(m, IGMP_LEADINGSPACE);
2902 		}
2903 		if (m == NULL)
2904 			return (-ENOMEM);
2905 		igmp_save_context(m, ifp);
2906 		md = m_getptr(m, 0, &off);
2907 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2908 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2909 
2910 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2911 			if (m != m0)
2912 				m_freem(m);
2913 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2914 			return (-ENOMEM);
2915 		}
2916 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2917 		nbytes += sizeof(struct igmp_grouprec);
2918 
2919 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2920 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2921 
2922 		msrcs = 0;
2923 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2924 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2925 			    ims->ims_haddr);
2926 			now = ims_get_mode(inm, ims, 1);
2927 			if ((now != mode) ||
2928 			    (now == mode && mode == MCAST_UNDEFINED)) {
2929 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2930 				continue;
2931 			}
2932 			if (is_source_query && ims->ims_stp == 0) {
2933 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2934 				    __func__);
2935 				continue;
2936 			}
2937 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2938 			naddr = htonl(ims->ims_haddr);
2939 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2940 				if (m != m0)
2941 					m_freem(m);
2942 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2943 				    __func__);
2944 				return (-ENOMEM);
2945 			}
2946 			++msrcs;
2947 			if (msrcs == m0srcs)
2948 				break;
2949 		}
2950 		pig->ig_numsrc = htons(msrcs);
2951 		nbytes += (msrcs * sizeof(in_addr_t));
2952 
2953 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2954 		mbufq_enqueue(mq, m);
2955 	}
2956 
2957 	return (nbytes);
2958 }
2959 
2960 /*
2961  * Type used to mark record pass completion.
2962  * We exploit the fact we can cast to this easily from the
2963  * current filter modes on each ip_msource node.
2964  */
2965 typedef enum {
2966 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2967 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2968 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2969 	REC_FULL = REC_ALLOW | REC_BLOCK
2970 } rectype_t;
2971 
2972 /*
2973  * Enqueue an IGMPv3 filter list change to the given output queue.
2974  *
2975  * Source list filter state is held in an RB-tree. When the filter list
2976  * for a group is changed without changing its mode, we need to compute
2977  * the deltas between T0 and T1 for each source in the filter set,
2978  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2979  *
2980  * As we may potentially queue two record types, and the entire R-B tree
2981  * needs to be walked at once, we break this out into its own function
2982  * so we can generate a tightly packed queue of packets.
2983  *
2984  * XXX This could be written to only use one tree walk, although that makes
2985  * serializing into the mbuf chains a bit harder. For now we do two walks
2986  * which makes things easier on us, and it may or may not be harder on
2987  * the L2 cache.
2988  *
2989  * If successful the size of all data appended to the queue is returned,
2990  * otherwise an error code less than zero is returned, or zero if
2991  * no record(s) were appended.
2992  */
2993 static int
2994 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
2995 {
2996 	static const int MINRECLEN =
2997 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
2998 	struct ifnet		*ifp;
2999 	struct igmp_grouprec	 ig;
3000 	struct igmp_grouprec	*pig;
3001 	struct ip_msource	*ims, *nims;
3002 	struct mbuf		*m, *m0, *md;
3003 	in_addr_t		 naddr;
3004 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3005 	int			 nallow, nblock;
3006 	uint8_t			 mode, now, then;
3007 	rectype_t		 crt, drt, nrt;
3008 
3009 	IN_MULTI_LIST_LOCK_ASSERT();
3010 
3011 	if (inm->inm_nsrc == 0 ||
3012 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3013 		return (0);
3014 
3015 	ifp = inm->inm_ifp;			/* interface */
3016 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
3017 	crt = REC_NONE;	/* current group record type */
3018 	drt = REC_NONE;	/* mask of completed group record types */
3019 	nrt = REC_NONE;	/* record type for current node */
3020 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3021 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3022 	npbytes = 0;	/* # of bytes appended this packet */
3023 	rsrcs = 0;	/* # sources encoded in current record */
3024 	schanged = 0;	/* # nodes encoded in overall filter change */
3025 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3026 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3027 	nims = NULL;	/* next tree node pointer */
3028 
3029 	/*
3030 	 * For each possible filter record mode.
3031 	 * The first kind of source we encounter tells us which
3032 	 * is the first kind of record we start appending.
3033 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3034 	 * as the inverse of the group's filter mode.
3035 	 */
3036 	while (drt != REC_FULL) {
3037 		do {
3038 			m0 = mbufq_last(mq);
3039 			if (m0 != NULL &&
3040 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3041 			     IGMP_V3_REPORT_MAXRECS) &&
3042 			    (m0->m_pkthdr.len + MINRECLEN) <
3043 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3044 				m = m0;
3045 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3046 					    sizeof(struct igmp_grouprec)) /
3047 				    sizeof(in_addr_t);
3048 				CTR1(KTR_IGMPV3,
3049 				    "%s: use previous packet", __func__);
3050 			} else {
3051 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3052 				if (m)
3053 					m->m_data += IGMP_LEADINGSPACE;
3054 				if (m == NULL) {
3055 					m = m_gethdr(M_NOWAIT, MT_DATA);
3056 					if (m)
3057 						M_ALIGN(m, IGMP_LEADINGSPACE);
3058 				}
3059 				if (m == NULL) {
3060 					CTR1(KTR_IGMPV3,
3061 					    "%s: m_get*() failed", __func__);
3062 					return (-ENOMEM);
3063 				}
3064 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3065 				igmp_save_context(m, ifp);
3066 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3067 				    sizeof(struct igmp_grouprec)) /
3068 				    sizeof(in_addr_t);
3069 				npbytes = 0;
3070 				CTR1(KTR_IGMPV3,
3071 				    "%s: allocated new packet", __func__);
3072 			}
3073 			/*
3074 			 * Append the IGMP group record header to the
3075 			 * current packet's data area.
3076 			 * Recalculate pointer to free space for next
3077 			 * group record, in case m_append() allocated
3078 			 * a new mbuf or cluster.
3079 			 */
3080 			memset(&ig, 0, sizeof(ig));
3081 			ig.ig_group = inm->inm_addr;
3082 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3083 				if (m != m0)
3084 					m_freem(m);
3085 				CTR1(KTR_IGMPV3,
3086 				    "%s: m_append() failed", __func__);
3087 				return (-ENOMEM);
3088 			}
3089 			npbytes += sizeof(struct igmp_grouprec);
3090 			if (m != m0) {
3091 				/* new packet; offset in c hain */
3092 				md = m_getptr(m, npbytes -
3093 				    sizeof(struct igmp_grouprec), &off);
3094 				pig = (struct igmp_grouprec *)(mtod(md,
3095 				    uint8_t *) + off);
3096 			} else {
3097 				/* current packet; offset from last append */
3098 				md = m_last(m);
3099 				pig = (struct igmp_grouprec *)(mtod(md,
3100 				    uint8_t *) + md->m_len -
3101 				    sizeof(struct igmp_grouprec));
3102 			}
3103 			/*
3104 			 * Begin walking the tree for this record type
3105 			 * pass, or continue from where we left off
3106 			 * previously if we had to allocate a new packet.
3107 			 * Only report deltas in-mode at t1.
3108 			 * We need not report included sources as allowed
3109 			 * if we are in inclusive mode on the group,
3110 			 * however the converse is not true.
3111 			 */
3112 			rsrcs = 0;
3113 			if (nims == NULL)
3114 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3115 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3116 				CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3117 				    __func__, ims->ims_haddr);
3118 				now = ims_get_mode(inm, ims, 1);
3119 				then = ims_get_mode(inm, ims, 0);
3120 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3121 				    __func__, then, now);
3122 				if (now == then) {
3123 					CTR1(KTR_IGMPV3,
3124 					    "%s: skip unchanged", __func__);
3125 					continue;
3126 				}
3127 				if (mode == MCAST_EXCLUDE &&
3128 				    now == MCAST_INCLUDE) {
3129 					CTR1(KTR_IGMPV3,
3130 					    "%s: skip IN src on EX group",
3131 					    __func__);
3132 					continue;
3133 				}
3134 				nrt = (rectype_t)now;
3135 				if (nrt == REC_NONE)
3136 					nrt = (rectype_t)(~mode & REC_FULL);
3137 				if (schanged++ == 0) {
3138 					crt = nrt;
3139 				} else if (crt != nrt)
3140 					continue;
3141 				naddr = htonl(ims->ims_haddr);
3142 				if (!m_append(m, sizeof(in_addr_t),
3143 				    (void *)&naddr)) {
3144 					if (m != m0)
3145 						m_freem(m);
3146 					CTR1(KTR_IGMPV3,
3147 					    "%s: m_append() failed", __func__);
3148 					return (-ENOMEM);
3149 				}
3150 				nallow += !!(crt == REC_ALLOW);
3151 				nblock += !!(crt == REC_BLOCK);
3152 				if (++rsrcs == m0srcs)
3153 					break;
3154 			}
3155 			/*
3156 			 * If we did not append any tree nodes on this
3157 			 * pass, back out of allocations.
3158 			 */
3159 			if (rsrcs == 0) {
3160 				npbytes -= sizeof(struct igmp_grouprec);
3161 				if (m != m0) {
3162 					CTR1(KTR_IGMPV3,
3163 					    "%s: m_free(m)", __func__);
3164 					m_freem(m);
3165 				} else {
3166 					CTR1(KTR_IGMPV3,
3167 					    "%s: m_adj(m, -ig)", __func__);
3168 					m_adj(m, -((int)sizeof(
3169 					    struct igmp_grouprec)));
3170 				}
3171 				continue;
3172 			}
3173 			npbytes += (rsrcs * sizeof(in_addr_t));
3174 			if (crt == REC_ALLOW)
3175 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3176 			else if (crt == REC_BLOCK)
3177 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3178 			pig->ig_numsrc = htons(rsrcs);
3179 			/*
3180 			 * Count the new group record, and enqueue this
3181 			 * packet if it wasn't already queued.
3182 			 */
3183 			m->m_pkthdr.PH_vt.vt_nrecs++;
3184 			if (m != m0)
3185 				mbufq_enqueue(mq, m);
3186 			nbytes += npbytes;
3187 		} while (nims != NULL);
3188 		drt |= crt;
3189 		crt = (~crt & REC_FULL);
3190 	}
3191 
3192 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3193 	    nallow, nblock);
3194 
3195 	return (nbytes);
3196 }
3197 
3198 static int
3199 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3200 {
3201 	struct mbufq	*gq;
3202 	struct mbuf	*m;		/* pending state-change */
3203 	struct mbuf	*m0;		/* copy of pending state-change */
3204 	struct mbuf	*mt;		/* last state-change in packet */
3205 	int		 docopy, domerge;
3206 	u_int		 recslen;
3207 
3208 	docopy = 0;
3209 	domerge = 0;
3210 	recslen = 0;
3211 
3212 	IN_MULTI_LIST_LOCK_ASSERT();
3213 	IGMP_LOCK_ASSERT();
3214 
3215 	/*
3216 	 * If there are further pending retransmissions, make a writable
3217 	 * copy of each queued state-change message before merging.
3218 	 */
3219 	if (inm->inm_scrv > 0)
3220 		docopy = 1;
3221 
3222 	gq = &inm->inm_scq;
3223 #ifdef KTR
3224 	if (mbufq_first(gq) == NULL) {
3225 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3226 		    __func__, inm);
3227 	}
3228 #endif
3229 
3230 	m = mbufq_first(gq);
3231 	while (m != NULL) {
3232 		/*
3233 		 * Only merge the report into the current packet if
3234 		 * there is sufficient space to do so; an IGMPv3 report
3235 		 * packet may only contain 65,535 group records.
3236 		 * Always use a simple mbuf chain concatentation to do this,
3237 		 * as large state changes for single groups may have
3238 		 * allocated clusters.
3239 		 */
3240 		domerge = 0;
3241 		mt = mbufq_last(scq);
3242 		if (mt != NULL) {
3243 			recslen = m_length(m, NULL);
3244 
3245 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3246 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3247 			    IGMP_V3_REPORT_MAXRECS) &&
3248 			    (mt->m_pkthdr.len + recslen <=
3249 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3250 				domerge = 1;
3251 		}
3252 
3253 		if (!domerge && mbufq_full(gq)) {
3254 			CTR2(KTR_IGMPV3,
3255 			    "%s: outbound queue full, skipping whole packet %p",
3256 			    __func__, m);
3257 			mt = m->m_nextpkt;
3258 			if (!docopy)
3259 				m_freem(m);
3260 			m = mt;
3261 			continue;
3262 		}
3263 
3264 		if (!docopy) {
3265 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3266 			m0 = mbufq_dequeue(gq);
3267 			m = m0->m_nextpkt;
3268 		} else {
3269 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3270 			m0 = m_dup(m, M_NOWAIT);
3271 			if (m0 == NULL)
3272 				return (ENOMEM);
3273 			m0->m_nextpkt = NULL;
3274 			m = m->m_nextpkt;
3275 		}
3276 
3277 		if (!domerge) {
3278 			CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3279 			    __func__, m0, scq);
3280 			mbufq_enqueue(scq, m0);
3281 		} else {
3282 			struct mbuf *mtl;	/* last mbuf of packet mt */
3283 
3284 			CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3285 			    __func__, m0, mt);
3286 
3287 			mtl = m_last(mt);
3288 			m0->m_flags &= ~M_PKTHDR;
3289 			mt->m_pkthdr.len += recslen;
3290 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3291 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3292 
3293 			mtl->m_next = m0;
3294 		}
3295 	}
3296 
3297 	return (0);
3298 }
3299 
3300 /*
3301  * Respond to a pending IGMPv3 General Query.
3302  */
3303 static void
3304 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3305 {
3306 	struct epoch_tracker	 et;
3307 	struct ifmultiaddr	*ifma;
3308 	struct ifnet		*ifp;
3309 	struct in_multi		*inm;
3310 	int			 retval __unused, loop;
3311 
3312 	IN_MULTI_LIST_LOCK_ASSERT();
3313 	IGMP_LOCK_ASSERT();
3314 
3315 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3316 	    ("%s: called when version %d", __func__, igi->igi_version));
3317 
3318 	/*
3319 	 * Check that there are some packets queued. If so, send them first.
3320 	 * For large number of groups the reply to general query can take
3321 	 * many packets, we should finish sending them before starting of
3322 	 * queuing the new reply.
3323 	 */
3324 	if (mbufq_len(&igi->igi_gq) != 0)
3325 		goto send;
3326 
3327 	ifp = igi->igi_ifp;
3328 
3329 	NET_EPOCH_ENTER(et);
3330 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3331 		if (ifma->ifma_addr->sa_family != AF_INET ||
3332 		    ifma->ifma_protospec == NULL)
3333 			continue;
3334 
3335 		inm = (struct in_multi *)ifma->ifma_protospec;
3336 		KASSERT(ifp == inm->inm_ifp,
3337 		    ("%s: inconsistent ifp", __func__));
3338 
3339 		switch (inm->inm_state) {
3340 		case IGMP_NOT_MEMBER:
3341 		case IGMP_SILENT_MEMBER:
3342 			break;
3343 		case IGMP_REPORTING_MEMBER:
3344 		case IGMP_IDLE_MEMBER:
3345 		case IGMP_LAZY_MEMBER:
3346 		case IGMP_SLEEPING_MEMBER:
3347 		case IGMP_AWAKENING_MEMBER:
3348 			inm->inm_state = IGMP_REPORTING_MEMBER;
3349 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3350 			    inm, 0, 0, 0);
3351 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3352 			    __func__, retval);
3353 			break;
3354 		case IGMP_G_QUERY_PENDING_MEMBER:
3355 		case IGMP_SG_QUERY_PENDING_MEMBER:
3356 		case IGMP_LEAVING_MEMBER:
3357 			break;
3358 		}
3359 	}
3360 	NET_EPOCH_EXIT(et);
3361 
3362 send:
3363 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3364 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3365 
3366 	/*
3367 	 * Slew transmission of bursts over 500ms intervals.
3368 	 */
3369 	if (mbufq_first(&igi->igi_gq) != NULL) {
3370 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3371 		    IGMP_RESPONSE_BURST_INTERVAL);
3372 		V_interface_timers_running = 1;
3373 	}
3374 }
3375 
3376 /*
3377  * Transmit the next pending IGMP message in the output queue.
3378  *
3379  * We get called from netisr_processqueue(). A mutex private to igmpoq
3380  * will be acquired and released around this routine.
3381  *
3382  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3383  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3384  * a link and uses a link-scope multicast address.
3385  */
3386 static void
3387 igmp_intr(struct mbuf *m)
3388 {
3389 	struct ip_moptions	 imo;
3390 	struct ifnet		*ifp;
3391 	struct mbuf		*ipopts, *m0;
3392 	int			 error;
3393 	uint32_t		 ifindex;
3394 
3395 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3396 
3397 	/*
3398 	 * Set VNET image pointer from enqueued mbuf chain
3399 	 * before doing anything else. Whilst we use interface
3400 	 * indexes to guard against interface detach, they are
3401 	 * unique to each VIMAGE and must be retrieved.
3402 	 */
3403 	CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3404 	ifindex = igmp_restore_context(m);
3405 
3406 	/*
3407 	 * Check if the ifnet still exists. This limits the scope of
3408 	 * any race in the absence of a global ifp lock for low cost
3409 	 * (an array lookup).
3410 	 */
3411 	ifp = ifnet_byindex(ifindex);
3412 	if (ifp == NULL) {
3413 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3414 		    __func__, m, ifindex);
3415 		m_freem(m);
3416 		IPSTAT_INC(ips_noroute);
3417 		goto out;
3418 	}
3419 
3420 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3421 
3422 	imo.imo_multicast_ttl  = 1;
3423 	imo.imo_multicast_vif  = -1;
3424 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3425 
3426 	/*
3427 	 * If the user requested that IGMP traffic be explicitly
3428 	 * redirected to the loopback interface (e.g. they are running a
3429 	 * MANET interface and the routing protocol needs to see the
3430 	 * updates), handle this now.
3431 	 */
3432 	if (m->m_flags & M_IGMP_LOOP)
3433 		imo.imo_multicast_ifp = V_loif;
3434 	else
3435 		imo.imo_multicast_ifp = ifp;
3436 
3437 	if (m->m_flags & M_IGMPV2) {
3438 		m0 = m;
3439 	} else {
3440 		m0 = igmp_v3_encap_report(ifp, m);
3441 		if (m0 == NULL) {
3442 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3443 			m_freem(m);
3444 			IPSTAT_INC(ips_odropped);
3445 			goto out;
3446 		}
3447 	}
3448 
3449 	igmp_scrub_context(m0);
3450 	m_clrprotoflags(m);
3451 	m0->m_pkthdr.rcvif = V_loif;
3452 #ifdef MAC
3453 	mac_netinet_igmp_send(ifp, m0);
3454 #endif
3455 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3456 	if (error) {
3457 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3458 		goto out;
3459 	}
3460 
3461 	IGMPSTAT_INC(igps_snd_reports);
3462 
3463 out:
3464 	/*
3465 	 * We must restore the existing vnet pointer before
3466 	 * continuing as we are run from netisr context.
3467 	 */
3468 	CURVNET_RESTORE();
3469 }
3470 
3471 /*
3472  * Encapsulate an IGMPv3 report.
3473  *
3474  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3475  * chain has already had its IP/IGMPv3 header prepended. In this case
3476  * the function will not attempt to prepend; the lengths and checksums
3477  * will however be re-computed.
3478  *
3479  * Returns a pointer to the new mbuf chain head, or NULL if the
3480  * allocation failed.
3481  */
3482 static struct mbuf *
3483 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3484 {
3485 	struct rm_priotracker	in_ifa_tracker;
3486 	struct igmp_report	*igmp;
3487 	struct ip		*ip;
3488 	int			 hdrlen, igmpreclen;
3489 
3490 	KASSERT((m->m_flags & M_PKTHDR),
3491 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3492 
3493 	igmpreclen = m_length(m, NULL);
3494 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3495 
3496 	if (m->m_flags & M_IGMPV3_HDR) {
3497 		igmpreclen -= hdrlen;
3498 	} else {
3499 		M_PREPEND(m, hdrlen, M_NOWAIT);
3500 		if (m == NULL)
3501 			return (NULL);
3502 		m->m_flags |= M_IGMPV3_HDR;
3503 	}
3504 
3505 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3506 
3507 	m->m_data += sizeof(struct ip);
3508 	m->m_len -= sizeof(struct ip);
3509 
3510 	igmp = mtod(m, struct igmp_report *);
3511 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3512 	igmp->ir_rsv1 = 0;
3513 	igmp->ir_rsv2 = 0;
3514 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3515 	igmp->ir_cksum = 0;
3516 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3517 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3518 
3519 	m->m_data -= sizeof(struct ip);
3520 	m->m_len += sizeof(struct ip);
3521 
3522 	ip = mtod(m, struct ip *);
3523 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3524 	ip->ip_len = htons(hdrlen + igmpreclen);
3525 	ip->ip_off = htons(IP_DF);
3526 	ip->ip_p = IPPROTO_IGMP;
3527 	ip->ip_sum = 0;
3528 
3529 	ip->ip_src.s_addr = INADDR_ANY;
3530 
3531 	if (m->m_flags & M_IGMP_LOOP) {
3532 		struct epoch_tracker et;
3533 		struct in_ifaddr *ia;
3534 
3535 		NET_EPOCH_ENTER(et);
3536 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3537 		if (ia != NULL)
3538 			ip->ip_src = ia->ia_addr.sin_addr;
3539 		NET_EPOCH_EXIT(et);
3540 	}
3541 
3542 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3543 
3544 	return (m);
3545 }
3546 
3547 #ifdef KTR
3548 static char *
3549 igmp_rec_type_to_str(const int type)
3550 {
3551 
3552 	switch (type) {
3553 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3554 			return "TO_EX";
3555 			break;
3556 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3557 			return "TO_IN";
3558 			break;
3559 		case IGMP_MODE_IS_EXCLUDE:
3560 			return "MODE_EX";
3561 			break;
3562 		case IGMP_MODE_IS_INCLUDE:
3563 			return "MODE_IN";
3564 			break;
3565 		case IGMP_ALLOW_NEW_SOURCES:
3566 			return "ALLOW_NEW";
3567 			break;
3568 		case IGMP_BLOCK_OLD_SOURCES:
3569 			return "BLOCK_OLD";
3570 			break;
3571 		default:
3572 			break;
3573 	}
3574 	return "unknown";
3575 }
3576 #endif
3577 
3578 #ifdef VIMAGE
3579 static void
3580 vnet_igmp_init(const void *unused __unused)
3581 {
3582 
3583 	netisr_register_vnet(&igmp_nh);
3584 }
3585 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3586     vnet_igmp_init, NULL);
3587 
3588 static void
3589 vnet_igmp_uninit(const void *unused __unused)
3590 {
3591 
3592 	/* This can happen when we shutdown the entire network stack. */
3593 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3594 
3595 	netisr_unregister_vnet(&igmp_nh);
3596 }
3597 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3598     vnet_igmp_uninit, NULL);
3599 #endif
3600 
3601 #ifdef DDB
3602 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3603 {
3604 	struct igmp_ifsoftc *igi, *tigi;
3605 	LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3606 
3607 	if (!have_addr) {
3608 		db_printf("usage: show igi_list <addr>\n");
3609 		return;
3610 	}
3611 	igi_head = (struct _igi_list *)addr;
3612 
3613 	LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3614 		db_printf("igmp_ifsoftc %p:\n", igi);
3615 		db_printf("    ifp %p\n", igi->igi_ifp);
3616 		db_printf("    version %u\n", igi->igi_version);
3617 		db_printf("    v1_timer %u\n", igi->igi_v1_timer);
3618 		db_printf("    v2_timer %u\n", igi->igi_v2_timer);
3619 		db_printf("    v3_timer %u\n", igi->igi_v3_timer);
3620 		db_printf("    flags %#x\n", igi->igi_flags);
3621 		db_printf("    rv %u\n", igi->igi_rv);
3622 		db_printf("    qi %u\n", igi->igi_qi);
3623 		db_printf("    qri %u\n", igi->igi_qri);
3624 		db_printf("    uri %u\n", igi->igi_uri);
3625 		/* struct mbufq    igi_gq; */
3626 		db_printf("\n");
3627 	}
3628 }
3629 #endif
3630 
3631 static int
3632 igmp_modevent(module_t mod, int type, void *unused __unused)
3633 {
3634 
3635 	switch (type) {
3636 	case MOD_LOAD:
3637 		CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3638 		IGMP_LOCK_INIT();
3639 		m_raopt = igmp_ra_alloc();
3640 		netisr_register(&igmp_nh);
3641 		break;
3642 	case MOD_UNLOAD:
3643 		CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3644 		netisr_unregister(&igmp_nh);
3645 		m_free(m_raopt);
3646 		m_raopt = NULL;
3647 		IGMP_LOCK_DESTROY();
3648 		break;
3649 	default:
3650 		return (EOPNOTSUPP);
3651 	}
3652 	return (0);
3653 }
3654 
3655 static moduledata_t igmp_mod = {
3656     "igmp",
3657     igmp_modevent,
3658     0
3659 };
3660 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);
3661