xref: /freebsd/sys/netinet/igmp.c (revision c321779676d0722f5d0a65c30e8b23b8a6547364)
1 /*-
2  * Copyright (c) 2007-2009 Bruce Simpson.
3  * Copyright (c) 1988 Stephen Deering.
4  * Copyright (c) 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Stephen Deering of Stanford University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
35  */
36 
37 /*
38  * Internet Group Management Protocol (IGMP) routines.
39  * [RFC1112, RFC2236, RFC3376]
40  *
41  * Written by Steve Deering, Stanford, May 1988.
42  * Modified by Rosen Sharma, Stanford, Aug 1994.
43  * Modified by Bill Fenner, Xerox PARC, Feb 1995.
44  * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995.
45  * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson.
46  *
47  * MULTICAST Revision: 3.5.1.4
48  */
49 
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
52 
53 #include "opt_ddb.h"
54 
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/module.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/socket.h>
61 #include <sys/protosw.h>
62 #include <sys/kernel.h>
63 #include <sys/lock.h>
64 #include <sys/rmlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/ktr.h>
67 #include <sys/condvar.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif
72 
73 #include <net/if.h>
74 #include <net/if_var.h>
75 #include <net/netisr.h>
76 #include <net/vnet.h>
77 
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_var.h>
83 #include <netinet/ip_options.h>
84 #include <netinet/igmp.h>
85 #include <netinet/igmp_var.h>
86 
87 #include <machine/in_cksum.h>
88 
89 #include <security/mac/mac_framework.h>
90 
91 #ifndef KTR_IGMPV3
92 #define KTR_IGMPV3 KTR_INET
93 #endif
94 
95 static struct igmp_ifsoftc *
96 		igi_alloc_locked(struct ifnet *);
97 static void	igi_delete_locked(const struct ifnet *);
98 static void	igmp_dispatch_queue(struct mbufq *, int, const int);
99 static void	igmp_fasttimo_vnet(void);
100 static void	igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
101 static int	igmp_handle_state_change(struct in_multi *,
102 		    struct igmp_ifsoftc *);
103 static int	igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
104 static int	igmp_input_v1_query(struct ifnet *, const struct ip *,
105 		    const struct igmp *);
106 static int	igmp_input_v2_query(struct ifnet *, const struct ip *,
107 		    const struct igmp *);
108 static int	igmp_input_v3_query(struct ifnet *, const struct ip *,
109 		    /*const*/ struct igmpv3 *);
110 static int	igmp_input_v3_group_query(struct in_multi *,
111 		    struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
112 static int	igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
113 		    /*const*/ struct igmp *);
114 static int	igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
115 		    /*const*/ struct igmp *);
116 static void	igmp_intr(struct mbuf *);
117 static int	igmp_isgroupreported(const struct in_addr);
118 static struct mbuf *
119 		igmp_ra_alloc(void);
120 #ifdef KTR
121 static char *	igmp_rec_type_to_str(const int);
122 #endif
123 static void	igmp_set_version(struct igmp_ifsoftc *, const int);
124 static void	igmp_slowtimo_vnet(void);
125 static int	igmp_v1v2_queue_report(struct in_multi *, const int);
126 static void	igmp_v1v2_process_group_timer(struct in_multi *, const int);
127 static void	igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
128 static void	igmp_v2_update_group(struct in_multi *, const int);
129 static void	igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
130 static void	igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
131 static struct mbuf *
132 		igmp_v3_encap_report(struct ifnet *, struct mbuf *);
133 static int	igmp_v3_enqueue_group_record(struct mbufq *,
134 		    struct in_multi *, const int, const int, const int);
135 static int	igmp_v3_enqueue_filter_change(struct mbufq *,
136 		    struct in_multi *);
137 static void	igmp_v3_process_group_timers(struct igmp_ifsoftc *,
138 		    struct mbufq *, struct mbufq *, struct in_multi *,
139 		    const int);
140 static int	igmp_v3_merge_state_changes(struct in_multi *,
141 		    struct mbufq *);
142 static void	igmp_v3_suppress_group_record(struct in_multi *);
143 static int	sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
144 static int	sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
145 static int	sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
146 
147 static const struct netisr_handler igmp_nh = {
148 	.nh_name = "igmp",
149 	.nh_handler = igmp_intr,
150 	.nh_proto = NETISR_IGMP,
151 	.nh_policy = NETISR_POLICY_SOURCE,
152 };
153 
154 /*
155  * System-wide globals.
156  *
157  * Unlocked access to these is OK, except for the global IGMP output
158  * queue. The IGMP subsystem lock ends up being system-wide for the moment,
159  * because all VIMAGEs have to share a global output queue, as netisrs
160  * themselves are not virtualized.
161  *
162  * Locking:
163  *  * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
164  *    Any may be taken independently; if any are held at the same
165  *    time, the above lock order must be followed.
166  *  * All output is delegated to the netisr.
167  *    Now that Giant has been eliminated, the netisr may be inlined.
168  *  * IN_MULTI_LOCK covers in_multi.
169  *  * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
170  *    including the output queue.
171  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
172  *    per-link state iterators.
173  *  * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
174  *    therefore it is not refcounted.
175  *    We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
176  *
177  * Reference counting
178  *  * IGMP acquires its own reference every time an in_multi is passed to
179  *    it and the group is being joined for the first time.
180  *  * IGMP releases its reference(s) on in_multi in a deferred way,
181  *    because the operations which process the release run as part of
182  *    a loop whose control variables are directly affected by the release
183  *    (that, and not recursing on the IF_ADDR_LOCK).
184  *
185  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
186  * to a vnet in ifp->if_vnet.
187  *
188  * SMPng: XXX We may potentially race operations on ifma_protospec.
189  * The problem is that we currently lack a clean way of taking the
190  * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
191  * as anything which modifies ifma needs to be covered by that lock.
192  * So check for ifma_protospec being NULL before proceeding.
193  */
194 struct mtx		 igmp_mtx;
195 
196 struct mbuf		*m_raopt;		 /* Router Alert option */
197 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
198 
199 /*
200  * VIMAGE-wide globals.
201  *
202  * The IGMPv3 timers themselves need to run per-image, however,
203  * protosw timers run globally (see tcp).
204  * An ifnet can only be in one vimage at a time, and the loopback
205  * ifnet, loif, is itself virtualized.
206  * It would otherwise be possible to seriously hose IGMP state,
207  * and create inconsistencies in upstream multicast routing, if you have
208  * multiple VIMAGEs running on the same link joining different multicast
209  * groups, UNLESS the "primary IP address" is different. This is because
210  * IGMP for IPv4 does not force link-local addresses to be used for each
211  * node, unlike MLD for IPv6.
212  * Obviously the IGMPv3 per-interface state has per-vimage granularity
213  * also as a result.
214  *
215  * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
216  * policy to control the address used by IGMP on the link.
217  */
218 static VNET_DEFINE(int, interface_timers_running);	/* IGMPv3 general
219 							 * query response */
220 static VNET_DEFINE(int, state_change_timers_running);	/* IGMPv3 state-change
221 							 * retransmit */
222 static VNET_DEFINE(int, current_state_timers_running);	/* IGMPv1/v2 host
223 							 * report; IGMPv3 g/sg
224 							 * query response */
225 
226 #define	V_interface_timers_running	VNET(interface_timers_running)
227 #define	V_state_change_timers_running	VNET(state_change_timers_running)
228 #define	V_current_state_timers_running	VNET(current_state_timers_running)
229 
230 static VNET_DEFINE(LIST_HEAD(, igmp_ifsoftc), igi_head) =
231     LIST_HEAD_INITIALIZER(igi_head);
232 static VNET_DEFINE(struct igmpstat, igmpstat) = {
233 	.igps_version = IGPS_VERSION_3,
234 	.igps_len = sizeof(struct igmpstat),
235 };
236 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
237 
238 #define	V_igi_head			VNET(igi_head)
239 #define	V_igmpstat			VNET(igmpstat)
240 #define	V_igmp_gsrdelay			VNET(igmp_gsrdelay)
241 
242 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
243 static VNET_DEFINE(int, igmp_sendra) = 1;
244 static VNET_DEFINE(int, igmp_sendlocal) = 1;
245 static VNET_DEFINE(int, igmp_v1enable) = 1;
246 static VNET_DEFINE(int, igmp_v2enable) = 1;
247 static VNET_DEFINE(int, igmp_legacysupp);
248 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
249 
250 #define	V_igmp_recvifkludge		VNET(igmp_recvifkludge)
251 #define	V_igmp_sendra			VNET(igmp_sendra)
252 #define	V_igmp_sendlocal		VNET(igmp_sendlocal)
253 #define	V_igmp_v1enable			VNET(igmp_v1enable)
254 #define	V_igmp_v2enable			VNET(igmp_v2enable)
255 #define	V_igmp_legacysupp		VNET(igmp_legacysupp)
256 #define	V_igmp_default_version		VNET(igmp_default_version)
257 
258 /*
259  * Virtualized sysctls.
260  */
261 SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW,
262     &VNET_NAME(igmpstat), igmpstat, "");
263 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
264     &VNET_NAME(igmp_recvifkludge), 0,
265     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
266 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
267     &VNET_NAME(igmp_sendra), 0,
268     "Send IP Router Alert option in IGMPv2/v3 messages");
269 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
270     &VNET_NAME(igmp_sendlocal), 0,
271     "Send IGMP membership reports for 224.0.0.0/24 groups");
272 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
273     &VNET_NAME(igmp_v1enable), 0,
274     "Enable backwards compatibility with IGMPv1");
275 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
276     &VNET_NAME(igmp_v2enable), 0,
277     "Enable backwards compatibility with IGMPv2");
278 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
279     &VNET_NAME(igmp_legacysupp), 0,
280     "Allow v1/v2 reports to suppress v3 group responses");
281 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
282     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
283     &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
284     "Default version of IGMP to run on each interface");
285 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
286     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
287     &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
288     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
289 
290 /*
291  * Non-virtualized sysctls.
292  */
293 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
294     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
295     "Per-interface IGMPv3 state");
296 
297 static __inline void
298 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
299 {
300 
301 #ifdef VIMAGE
302 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
303 #endif /* VIMAGE */
304 	m->m_pkthdr.flowid = ifp->if_index;
305 }
306 
307 static __inline void
308 igmp_scrub_context(struct mbuf *m)
309 {
310 
311 	m->m_pkthdr.PH_loc.ptr = NULL;
312 	m->m_pkthdr.flowid = 0;
313 }
314 
315 /*
316  * Restore context from a queued IGMP output chain.
317  * Return saved ifindex.
318  *
319  * VIMAGE: The assertion is there to make sure that we
320  * actually called CURVNET_SET() with what's in the mbuf chain.
321  */
322 static __inline uint32_t
323 igmp_restore_context(struct mbuf *m)
324 {
325 
326 #ifdef notyet
327 #if defined(VIMAGE) && defined(INVARIANTS)
328 	KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
329 	    ("%s: called when curvnet was not restored", __func__));
330 #endif
331 #endif
332 	return (m->m_pkthdr.flowid);
333 }
334 
335 /*
336  * Retrieve or set default IGMP version.
337  *
338  * VIMAGE: Assume curvnet set by caller.
339  * SMPng: NOTE: Serialized by IGMP lock.
340  */
341 static int
342 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
343 {
344 	int	 error;
345 	int	 new;
346 
347 	error = sysctl_wire_old_buffer(req, sizeof(int));
348 	if (error)
349 		return (error);
350 
351 	IGMP_LOCK();
352 
353 	new = V_igmp_default_version;
354 
355 	error = sysctl_handle_int(oidp, &new, 0, req);
356 	if (error || !req->newptr)
357 		goto out_locked;
358 
359 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
360 		error = EINVAL;
361 		goto out_locked;
362 	}
363 
364 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
365 	     V_igmp_default_version, new);
366 
367 	V_igmp_default_version = new;
368 
369 out_locked:
370 	IGMP_UNLOCK();
371 	return (error);
372 }
373 
374 /*
375  * Retrieve or set threshold between group-source queries in seconds.
376  *
377  * VIMAGE: Assume curvnet set by caller.
378  * SMPng: NOTE: Serialized by IGMP lock.
379  */
380 static int
381 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
382 {
383 	int error;
384 	int i;
385 
386 	error = sysctl_wire_old_buffer(req, sizeof(int));
387 	if (error)
388 		return (error);
389 
390 	IGMP_LOCK();
391 
392 	i = V_igmp_gsrdelay.tv_sec;
393 
394 	error = sysctl_handle_int(oidp, &i, 0, req);
395 	if (error || !req->newptr)
396 		goto out_locked;
397 
398 	if (i < -1 || i >= 60) {
399 		error = EINVAL;
400 		goto out_locked;
401 	}
402 
403 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
404 	     V_igmp_gsrdelay.tv_sec, i);
405 	V_igmp_gsrdelay.tv_sec = i;
406 
407 out_locked:
408 	IGMP_UNLOCK();
409 	return (error);
410 }
411 
412 /*
413  * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
414  * For use by ifmcstat(8).
415  *
416  * SMPng: NOTE: Does an unlocked ifindex space read.
417  * VIMAGE: Assume curvnet set by caller. The node handler itself
418  * is not directly virtualized.
419  */
420 static int
421 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
422 {
423 	int			*name;
424 	int			 error;
425 	u_int			 namelen;
426 	struct ifnet		*ifp;
427 	struct igmp_ifsoftc	*igi;
428 
429 	name = (int *)arg1;
430 	namelen = arg2;
431 
432 	if (req->newptr != NULL)
433 		return (EPERM);
434 
435 	if (namelen != 1)
436 		return (EINVAL);
437 
438 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
439 	if (error)
440 		return (error);
441 
442 	IN_MULTI_LOCK();
443 	IGMP_LOCK();
444 
445 	if (name[0] <= 0 || name[0] > V_if_index) {
446 		error = ENOENT;
447 		goto out_locked;
448 	}
449 
450 	error = ENOENT;
451 
452 	ifp = ifnet_byindex(name[0]);
453 	if (ifp == NULL)
454 		goto out_locked;
455 
456 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
457 		if (ifp == igi->igi_ifp) {
458 			struct igmp_ifinfo info;
459 
460 			info.igi_version = igi->igi_version;
461 			info.igi_v1_timer = igi->igi_v1_timer;
462 			info.igi_v2_timer = igi->igi_v2_timer;
463 			info.igi_v3_timer = igi->igi_v3_timer;
464 			info.igi_flags = igi->igi_flags;
465 			info.igi_rv = igi->igi_rv;
466 			info.igi_qi = igi->igi_qi;
467 			info.igi_qri = igi->igi_qri;
468 			info.igi_uri = igi->igi_uri;
469 			error = SYSCTL_OUT(req, &info, sizeof(info));
470 			break;
471 		}
472 	}
473 
474 out_locked:
475 	IGMP_UNLOCK();
476 	IN_MULTI_UNLOCK();
477 	return (error);
478 }
479 
480 /*
481  * Dispatch an entire queue of pending packet chains
482  * using the netisr.
483  * VIMAGE: Assumes the vnet pointer has been set.
484  */
485 static void
486 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
487 {
488 	struct mbuf *m;
489 
490 	while ((m = mbufq_dequeue(mq)) != NULL) {
491 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
492 		if (loop)
493 			m->m_flags |= M_IGMP_LOOP;
494 		netisr_dispatch(NETISR_IGMP, m);
495 		if (--limit == 0)
496 			break;
497 	}
498 }
499 
500 /*
501  * Filter outgoing IGMP report state by group.
502  *
503  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
504  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
505  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
506  * this may break certain IGMP snooping switches which rely on the old
507  * report behaviour.
508  *
509  * Return zero if the given group is one for which IGMP reports
510  * should be suppressed, or non-zero if reports should be issued.
511  */
512 static __inline int
513 igmp_isgroupreported(const struct in_addr addr)
514 {
515 
516 	if (in_allhosts(addr) ||
517 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
518 		return (0);
519 
520 	return (1);
521 }
522 
523 /*
524  * Construct a Router Alert option to use in outgoing packets.
525  */
526 static struct mbuf *
527 igmp_ra_alloc(void)
528 {
529 	struct mbuf	*m;
530 	struct ipoption	*p;
531 
532 	m = m_get(M_WAITOK, MT_DATA);
533 	p = mtod(m, struct ipoption *);
534 	p->ipopt_dst.s_addr = INADDR_ANY;
535 	p->ipopt_list[0] = (char)IPOPT_RA;	/* Router Alert Option */
536 	p->ipopt_list[1] = 0x04;		/* 4 bytes long */
537 	p->ipopt_list[2] = IPOPT_EOL;		/* End of IP option list */
538 	p->ipopt_list[3] = 0x00;		/* pad byte */
539 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
540 
541 	return (m);
542 }
543 
544 /*
545  * Attach IGMP when PF_INET is attached to an interface.
546  */
547 struct igmp_ifsoftc *
548 igmp_domifattach(struct ifnet *ifp)
549 {
550 	struct igmp_ifsoftc *igi;
551 
552 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
553 	    __func__, ifp, ifp->if_xname);
554 
555 	IGMP_LOCK();
556 
557 	igi = igi_alloc_locked(ifp);
558 	if (!(ifp->if_flags & IFF_MULTICAST))
559 		igi->igi_flags |= IGIF_SILENT;
560 
561 	IGMP_UNLOCK();
562 
563 	return (igi);
564 }
565 
566 /*
567  * VIMAGE: assume curvnet set by caller.
568  */
569 static struct igmp_ifsoftc *
570 igi_alloc_locked(/*const*/ struct ifnet *ifp)
571 {
572 	struct igmp_ifsoftc *igi;
573 
574 	IGMP_LOCK_ASSERT();
575 
576 	igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
577 	if (igi == NULL)
578 		goto out;
579 
580 	igi->igi_ifp = ifp;
581 	igi->igi_version = V_igmp_default_version;
582 	igi->igi_flags = 0;
583 	igi->igi_rv = IGMP_RV_INIT;
584 	igi->igi_qi = IGMP_QI_INIT;
585 	igi->igi_qri = IGMP_QRI_INIT;
586 	igi->igi_uri = IGMP_URI_INIT;
587 	SLIST_INIT(&igi->igi_relinmhead);
588 	mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
589 
590 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
591 
592 	CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
593 	     ifp, ifp->if_xname);
594 
595 out:
596 	return (igi);
597 }
598 
599 /*
600  * Hook for ifdetach.
601  *
602  * NOTE: Some finalization tasks need to run before the protocol domain
603  * is detached, but also before the link layer does its cleanup.
604  *
605  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
606  * XXX This is also bitten by unlocked ifma_protospec access.
607  */
608 void
609 igmp_ifdetach(struct ifnet *ifp)
610 {
611 	struct igmp_ifsoftc	*igi;
612 	struct ifmultiaddr	*ifma;
613 	struct in_multi		*inm, *tinm;
614 
615 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
616 	    ifp->if_xname);
617 
618 	IGMP_LOCK();
619 
620 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
621 	if (igi->igi_version == IGMP_VERSION_3) {
622 		IF_ADDR_RLOCK(ifp);
623 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
624 			if (ifma->ifma_addr->sa_family != AF_INET ||
625 			    ifma->ifma_protospec == NULL)
626 				continue;
627 #if 0
628 			KASSERT(ifma->ifma_protospec != NULL,
629 			    ("%s: ifma_protospec is NULL", __func__));
630 #endif
631 			inm = (struct in_multi *)ifma->ifma_protospec;
632 			if (inm->inm_state == IGMP_LEAVING_MEMBER) {
633 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
634 				    inm, inm_nrele);
635 			}
636 			inm_clear_recorded(inm);
637 		}
638 		IF_ADDR_RUNLOCK(ifp);
639 		/*
640 		 * Free the in_multi reference(s) for this IGMP lifecycle.
641 		 */
642 		SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
643 		    tinm) {
644 			SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
645 			inm_release_locked(inm);
646 		}
647 	}
648 
649 	IGMP_UNLOCK();
650 }
651 
652 /*
653  * Hook for domifdetach.
654  */
655 void
656 igmp_domifdetach(struct ifnet *ifp)
657 {
658 
659 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
660 	    __func__, ifp, ifp->if_xname);
661 
662 	IGMP_LOCK();
663 	igi_delete_locked(ifp);
664 	IGMP_UNLOCK();
665 }
666 
667 static void
668 igi_delete_locked(const struct ifnet *ifp)
669 {
670 	struct igmp_ifsoftc *igi, *tigi;
671 
672 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
673 	    __func__, ifp, ifp->if_xname);
674 
675 	IGMP_LOCK_ASSERT();
676 
677 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
678 		if (igi->igi_ifp == ifp) {
679 			/*
680 			 * Free deferred General Query responses.
681 			 */
682 			mbufq_drain(&igi->igi_gq);
683 
684 			LIST_REMOVE(igi, igi_link);
685 
686 			KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
687 			    ("%s: there are dangling in_multi references",
688 			    __func__));
689 
690 			free(igi, M_IGMP);
691 			return;
692 		}
693 	}
694 }
695 
696 /*
697  * Process a received IGMPv1 query.
698  * Return non-zero if the message should be dropped.
699  *
700  * VIMAGE: The curvnet pointer is derived from the input ifp.
701  */
702 static int
703 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
704     const struct igmp *igmp)
705 {
706 	struct ifmultiaddr	*ifma;
707 	struct igmp_ifsoftc	*igi;
708 	struct in_multi		*inm;
709 
710 	/*
711 	 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
712 	 * 224.0.0.1. They are always treated as General Queries.
713 	 * igmp_group is always ignored. Do not drop it as a userland
714 	 * daemon may wish to see it.
715 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
716 	 */
717 	if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
718 		IGMPSTAT_INC(igps_rcv_badqueries);
719 		return (0);
720 	}
721 	IGMPSTAT_INC(igps_rcv_gen_queries);
722 
723 	IN_MULTI_LOCK();
724 	IGMP_LOCK();
725 
726 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
727 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
728 
729 	if (igi->igi_flags & IGIF_LOOPBACK) {
730 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
731 		    ifp, ifp->if_xname);
732 		goto out_locked;
733 	}
734 
735 	/*
736 	 * Switch to IGMPv1 host compatibility mode.
737 	 */
738 	igmp_set_version(igi, IGMP_VERSION_1);
739 
740 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
741 
742 	/*
743 	 * Start the timers in all of our group records
744 	 * for the interface on which the query arrived,
745 	 * except those which are already running.
746 	 */
747 	IF_ADDR_RLOCK(ifp);
748 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
749 		if (ifma->ifma_addr->sa_family != AF_INET ||
750 		    ifma->ifma_protospec == NULL)
751 			continue;
752 		inm = (struct in_multi *)ifma->ifma_protospec;
753 		if (inm->inm_timer != 0)
754 			continue;
755 		switch (inm->inm_state) {
756 		case IGMP_NOT_MEMBER:
757 		case IGMP_SILENT_MEMBER:
758 			break;
759 		case IGMP_G_QUERY_PENDING_MEMBER:
760 		case IGMP_SG_QUERY_PENDING_MEMBER:
761 		case IGMP_REPORTING_MEMBER:
762 		case IGMP_IDLE_MEMBER:
763 		case IGMP_LAZY_MEMBER:
764 		case IGMP_SLEEPING_MEMBER:
765 		case IGMP_AWAKENING_MEMBER:
766 			inm->inm_state = IGMP_REPORTING_MEMBER;
767 			inm->inm_timer = IGMP_RANDOM_DELAY(
768 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
769 			V_current_state_timers_running = 1;
770 			break;
771 		case IGMP_LEAVING_MEMBER:
772 			break;
773 		}
774 	}
775 	IF_ADDR_RUNLOCK(ifp);
776 
777 out_locked:
778 	IGMP_UNLOCK();
779 	IN_MULTI_UNLOCK();
780 
781 	return (0);
782 }
783 
784 /*
785  * Process a received IGMPv2 general or group-specific query.
786  */
787 static int
788 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
789     const struct igmp *igmp)
790 {
791 	struct ifmultiaddr	*ifma;
792 	struct igmp_ifsoftc	*igi;
793 	struct in_multi		*inm;
794 	int			 is_general_query;
795 	uint16_t		 timer;
796 
797 	is_general_query = 0;
798 
799 	/*
800 	 * Validate address fields upfront.
801 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
802 	 */
803 	if (in_nullhost(igmp->igmp_group)) {
804 		/*
805 		 * IGMPv2 General Query.
806 		 * If this was not sent to the all-hosts group, ignore it.
807 		 */
808 		if (!in_allhosts(ip->ip_dst))
809 			return (0);
810 		IGMPSTAT_INC(igps_rcv_gen_queries);
811 		is_general_query = 1;
812 	} else {
813 		/* IGMPv2 Group-Specific Query. */
814 		IGMPSTAT_INC(igps_rcv_group_queries);
815 	}
816 
817 	IN_MULTI_LOCK();
818 	IGMP_LOCK();
819 
820 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
821 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
822 
823 	if (igi->igi_flags & IGIF_LOOPBACK) {
824 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
825 		    ifp, ifp->if_xname);
826 		goto out_locked;
827 	}
828 
829 	/*
830 	 * Ignore v2 query if in v1 Compatibility Mode.
831 	 */
832 	if (igi->igi_version == IGMP_VERSION_1)
833 		goto out_locked;
834 
835 	igmp_set_version(igi, IGMP_VERSION_2);
836 
837 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
838 	if (timer == 0)
839 		timer = 1;
840 
841 	if (is_general_query) {
842 		/*
843 		 * For each reporting group joined on this
844 		 * interface, kick the report timer.
845 		 */
846 		CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
847 		    ifp, ifp->if_xname);
848 		IF_ADDR_RLOCK(ifp);
849 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
850 			if (ifma->ifma_addr->sa_family != AF_INET ||
851 			    ifma->ifma_protospec == NULL)
852 				continue;
853 			inm = (struct in_multi *)ifma->ifma_protospec;
854 			igmp_v2_update_group(inm, timer);
855 		}
856 		IF_ADDR_RUNLOCK(ifp);
857 	} else {
858 		/*
859 		 * Group-specific IGMPv2 query, we need only
860 		 * look up the single group to process it.
861 		 */
862 		inm = inm_lookup(ifp, igmp->igmp_group);
863 		if (inm != NULL) {
864 			CTR3(KTR_IGMPV3,
865 			    "process v2 query 0x%08x on ifp %p(%s)",
866 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
867 			igmp_v2_update_group(inm, timer);
868 		}
869 	}
870 
871 out_locked:
872 	IGMP_UNLOCK();
873 	IN_MULTI_UNLOCK();
874 
875 	return (0);
876 }
877 
878 /*
879  * Update the report timer on a group in response to an IGMPv2 query.
880  *
881  * If we are becoming the reporting member for this group, start the timer.
882  * If we already are the reporting member for this group, and timer is
883  * below the threshold, reset it.
884  *
885  * We may be updating the group for the first time since we switched
886  * to IGMPv3. If we are, then we must clear any recorded source lists,
887  * and transition to REPORTING state; the group timer is overloaded
888  * for group and group-source query responses.
889  *
890  * Unlike IGMPv3, the delay per group should be jittered
891  * to avoid bursts of IGMPv2 reports.
892  */
893 static void
894 igmp_v2_update_group(struct in_multi *inm, const int timer)
895 {
896 
897 	CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
898 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
899 
900 	IN_MULTI_LOCK_ASSERT();
901 
902 	switch (inm->inm_state) {
903 	case IGMP_NOT_MEMBER:
904 	case IGMP_SILENT_MEMBER:
905 		break;
906 	case IGMP_REPORTING_MEMBER:
907 		if (inm->inm_timer != 0 &&
908 		    inm->inm_timer <= timer) {
909 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
910 			    "skipping.", __func__);
911 			break;
912 		}
913 		/* FALLTHROUGH */
914 	case IGMP_SG_QUERY_PENDING_MEMBER:
915 	case IGMP_G_QUERY_PENDING_MEMBER:
916 	case IGMP_IDLE_MEMBER:
917 	case IGMP_LAZY_MEMBER:
918 	case IGMP_AWAKENING_MEMBER:
919 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
920 		inm->inm_state = IGMP_REPORTING_MEMBER;
921 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
922 		V_current_state_timers_running = 1;
923 		break;
924 	case IGMP_SLEEPING_MEMBER:
925 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
926 		inm->inm_state = IGMP_AWAKENING_MEMBER;
927 		break;
928 	case IGMP_LEAVING_MEMBER:
929 		break;
930 	}
931 }
932 
933 /*
934  * Process a received IGMPv3 general, group-specific or
935  * group-and-source-specific query.
936  * Assumes m has already been pulled up to the full IGMP message length.
937  * Return 0 if successful, otherwise an appropriate error code is returned.
938  */
939 static int
940 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
941     /*const*/ struct igmpv3 *igmpv3)
942 {
943 	struct igmp_ifsoftc	*igi;
944 	struct in_multi		*inm;
945 	int			 is_general_query;
946 	uint32_t		 maxresp, nsrc, qqi;
947 	uint16_t		 timer;
948 	uint8_t			 qrv;
949 
950 	is_general_query = 0;
951 
952 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
953 
954 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
955 	if (maxresp >= 128) {
956 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
957 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
958 	}
959 
960 	/*
961 	 * Robustness must never be less than 2 for on-wire IGMPv3.
962 	 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
963 	 * an exception for interfaces whose IGMPv3 state changes
964 	 * are redirected to loopback (e.g. MANET).
965 	 */
966 	qrv = IGMP_QRV(igmpv3->igmp_misc);
967 	if (qrv < 2) {
968 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
969 		    qrv, IGMP_RV_INIT);
970 		qrv = IGMP_RV_INIT;
971 	}
972 
973 	qqi = igmpv3->igmp_qqi;
974 	if (qqi >= 128) {
975 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
976 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
977 	}
978 
979 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
980 	if (timer == 0)
981 		timer = 1;
982 
983 	nsrc = ntohs(igmpv3->igmp_numsrc);
984 
985 	/*
986 	 * Validate address fields and versions upfront before
987 	 * accepting v3 query.
988 	 * XXX SMPng: Unlocked access to igmpstat counters here.
989 	 */
990 	if (in_nullhost(igmpv3->igmp_group)) {
991 		/*
992 		 * IGMPv3 General Query.
993 		 *
994 		 * General Queries SHOULD be directed to 224.0.0.1.
995 		 * A general query with a source list has undefined
996 		 * behaviour; discard it.
997 		 */
998 		IGMPSTAT_INC(igps_rcv_gen_queries);
999 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1000 			IGMPSTAT_INC(igps_rcv_badqueries);
1001 			return (0);
1002 		}
1003 		is_general_query = 1;
1004 	} else {
1005 		/* Group or group-source specific query. */
1006 		if (nsrc == 0)
1007 			IGMPSTAT_INC(igps_rcv_group_queries);
1008 		else
1009 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1010 	}
1011 
1012 	IN_MULTI_LOCK();
1013 	IGMP_LOCK();
1014 
1015 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1016 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1017 
1018 	if (igi->igi_flags & IGIF_LOOPBACK) {
1019 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1020 		    ifp, ifp->if_xname);
1021 		goto out_locked;
1022 	}
1023 
1024 	/*
1025 	 * Discard the v3 query if we're in Compatibility Mode.
1026 	 * The RFC is not obviously worded that hosts need to stay in
1027 	 * compatibility mode until the Old Version Querier Present
1028 	 * timer expires.
1029 	 */
1030 	if (igi->igi_version != IGMP_VERSION_3) {
1031 		CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1032 		    igi->igi_version, ifp, ifp->if_xname);
1033 		goto out_locked;
1034 	}
1035 
1036 	igmp_set_version(igi, IGMP_VERSION_3);
1037 	igi->igi_rv = qrv;
1038 	igi->igi_qi = qqi;
1039 	igi->igi_qri = maxresp;
1040 
1041 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1042 	    maxresp);
1043 
1044 	if (is_general_query) {
1045 		/*
1046 		 * Schedule a current-state report on this ifp for
1047 		 * all groups, possibly containing source lists.
1048 		 * If there is a pending General Query response
1049 		 * scheduled earlier than the selected delay, do
1050 		 * not schedule any other reports.
1051 		 * Otherwise, reset the interface timer.
1052 		 */
1053 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1054 		    ifp, ifp->if_xname);
1055 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1056 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1057 			V_interface_timers_running = 1;
1058 		}
1059 	} else {
1060 		/*
1061 		 * Group-source-specific queries are throttled on
1062 		 * a per-group basis to defeat denial-of-service attempts.
1063 		 * Queries for groups we are not a member of on this
1064 		 * link are simply ignored.
1065 		 */
1066 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1067 		if (inm == NULL)
1068 			goto out_locked;
1069 		if (nsrc > 0) {
1070 			if (!ratecheck(&inm->inm_lastgsrtv,
1071 			    &V_igmp_gsrdelay)) {
1072 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1073 				    __func__);
1074 				IGMPSTAT_INC(igps_drop_gsr_queries);
1075 				goto out_locked;
1076 			}
1077 		}
1078 		CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1079 		     ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1080 		/*
1081 		 * If there is a pending General Query response
1082 		 * scheduled sooner than the selected delay, no
1083 		 * further report need be scheduled.
1084 		 * Otherwise, prepare to respond to the
1085 		 * group-specific or group-and-source query.
1086 		 */
1087 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1088 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1089 	}
1090 
1091 out_locked:
1092 	IGMP_UNLOCK();
1093 	IN_MULTI_UNLOCK();
1094 
1095 	return (0);
1096 }
1097 
1098 /*
1099  * Process a received IGMPv3 group-specific or group-and-source-specific
1100  * query.
1101  * Return <0 if any error occurred. Currently this is ignored.
1102  */
1103 static int
1104 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1105     int timer, /*const*/ struct igmpv3 *igmpv3)
1106 {
1107 	int			 retval;
1108 	uint16_t		 nsrc;
1109 
1110 	IN_MULTI_LOCK_ASSERT();
1111 	IGMP_LOCK_ASSERT();
1112 
1113 	retval = 0;
1114 
1115 	switch (inm->inm_state) {
1116 	case IGMP_NOT_MEMBER:
1117 	case IGMP_SILENT_MEMBER:
1118 	case IGMP_SLEEPING_MEMBER:
1119 	case IGMP_LAZY_MEMBER:
1120 	case IGMP_AWAKENING_MEMBER:
1121 	case IGMP_IDLE_MEMBER:
1122 	case IGMP_LEAVING_MEMBER:
1123 		return (retval);
1124 		break;
1125 	case IGMP_REPORTING_MEMBER:
1126 	case IGMP_G_QUERY_PENDING_MEMBER:
1127 	case IGMP_SG_QUERY_PENDING_MEMBER:
1128 		break;
1129 	}
1130 
1131 	nsrc = ntohs(igmpv3->igmp_numsrc);
1132 
1133 	/*
1134 	 * Deal with group-specific queries upfront.
1135 	 * If any group query is already pending, purge any recorded
1136 	 * source-list state if it exists, and schedule a query response
1137 	 * for this group-specific query.
1138 	 */
1139 	if (nsrc == 0) {
1140 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1141 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1142 			inm_clear_recorded(inm);
1143 			timer = min(inm->inm_timer, timer);
1144 		}
1145 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1146 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1147 		V_current_state_timers_running = 1;
1148 		return (retval);
1149 	}
1150 
1151 	/*
1152 	 * Deal with the case where a group-and-source-specific query has
1153 	 * been received but a group-specific query is already pending.
1154 	 */
1155 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1156 		timer = min(inm->inm_timer, timer);
1157 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1158 		V_current_state_timers_running = 1;
1159 		return (retval);
1160 	}
1161 
1162 	/*
1163 	 * Finally, deal with the case where a group-and-source-specific
1164 	 * query has been received, where a response to a previous g-s-r
1165 	 * query exists, or none exists.
1166 	 * In this case, we need to parse the source-list which the Querier
1167 	 * has provided us with and check if we have any source list filter
1168 	 * entries at T1 for these sources. If we do not, there is no need
1169 	 * schedule a report and the query may be dropped.
1170 	 * If we do, we must record them and schedule a current-state
1171 	 * report for those sources.
1172 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1173 	 * we pass the mbuf chain pointer down to this function, and use
1174 	 * m_getptr() to walk the chain.
1175 	 */
1176 	if (inm->inm_nsrc > 0) {
1177 		const struct in_addr	*ap;
1178 		int			 i, nrecorded;
1179 
1180 		ap = (const struct in_addr *)(igmpv3 + 1);
1181 		nrecorded = 0;
1182 		for (i = 0; i < nsrc; i++, ap++) {
1183 			retval = inm_record_source(inm, ap->s_addr);
1184 			if (retval < 0)
1185 				break;
1186 			nrecorded += retval;
1187 		}
1188 		if (nrecorded > 0) {
1189 			CTR1(KTR_IGMPV3,
1190 			    "%s: schedule response to SG query", __func__);
1191 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1192 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1193 			V_current_state_timers_running = 1;
1194 		}
1195 	}
1196 
1197 	return (retval);
1198 }
1199 
1200 /*
1201  * Process a received IGMPv1 host membership report.
1202  *
1203  * NOTE: 0.0.0.0 workaround breaks const correctness.
1204  */
1205 static int
1206 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1207     /*const*/ struct igmp *igmp)
1208 {
1209 	struct rm_priotracker in_ifa_tracker;
1210 	struct in_ifaddr *ia;
1211 	struct in_multi *inm;
1212 
1213 	IGMPSTAT_INC(igps_rcv_reports);
1214 
1215 	if (ifp->if_flags & IFF_LOOPBACK)
1216 		return (0);
1217 
1218 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1219 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1220 		IGMPSTAT_INC(igps_rcv_badreports);
1221 		return (EINVAL);
1222 	}
1223 
1224 	/*
1225 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1226 	 * Booting clients may use the source address 0.0.0.0. Some
1227 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1228 	 * the interface upon which this message was received.
1229 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1230 	 */
1231 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1232 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1233 		if (ia != NULL) {
1234 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1235 			ifa_free(&ia->ia_ifa);
1236 		}
1237 	}
1238 
1239 	CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1240 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1241 
1242 	/*
1243 	 * IGMPv1 report suppression.
1244 	 * If we are a member of this group, and our membership should be
1245 	 * reported, stop our group timer and transition to the 'lazy' state.
1246 	 */
1247 	IN_MULTI_LOCK();
1248 	inm = inm_lookup(ifp, igmp->igmp_group);
1249 	if (inm != NULL) {
1250 		struct igmp_ifsoftc *igi;
1251 
1252 		igi = inm->inm_igi;
1253 		if (igi == NULL) {
1254 			KASSERT(igi != NULL,
1255 			    ("%s: no igi for ifp %p", __func__, ifp));
1256 			goto out_locked;
1257 		}
1258 
1259 		IGMPSTAT_INC(igps_rcv_ourreports);
1260 
1261 		/*
1262 		 * If we are in IGMPv3 host mode, do not allow the
1263 		 * other host's IGMPv1 report to suppress our reports
1264 		 * unless explicitly configured to do so.
1265 		 */
1266 		if (igi->igi_version == IGMP_VERSION_3) {
1267 			if (V_igmp_legacysupp)
1268 				igmp_v3_suppress_group_record(inm);
1269 			goto out_locked;
1270 		}
1271 
1272 		inm->inm_timer = 0;
1273 
1274 		switch (inm->inm_state) {
1275 		case IGMP_NOT_MEMBER:
1276 		case IGMP_SILENT_MEMBER:
1277 			break;
1278 		case IGMP_IDLE_MEMBER:
1279 		case IGMP_LAZY_MEMBER:
1280 		case IGMP_AWAKENING_MEMBER:
1281 			CTR3(KTR_IGMPV3,
1282 			    "report suppressed for 0x%08x on ifp %p(%s)",
1283 			    ntohl(igmp->igmp_group.s_addr), ifp,
1284 			    ifp->if_xname);
1285 		case IGMP_SLEEPING_MEMBER:
1286 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1287 			break;
1288 		case IGMP_REPORTING_MEMBER:
1289 			CTR3(KTR_IGMPV3,
1290 			    "report suppressed for 0x%08x on ifp %p(%s)",
1291 			    ntohl(igmp->igmp_group.s_addr), ifp,
1292 			    ifp->if_xname);
1293 			if (igi->igi_version == IGMP_VERSION_1)
1294 				inm->inm_state = IGMP_LAZY_MEMBER;
1295 			else if (igi->igi_version == IGMP_VERSION_2)
1296 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1297 			break;
1298 		case IGMP_G_QUERY_PENDING_MEMBER:
1299 		case IGMP_SG_QUERY_PENDING_MEMBER:
1300 		case IGMP_LEAVING_MEMBER:
1301 			break;
1302 		}
1303 	}
1304 
1305 out_locked:
1306 	IN_MULTI_UNLOCK();
1307 
1308 	return (0);
1309 }
1310 
1311 /*
1312  * Process a received IGMPv2 host membership report.
1313  *
1314  * NOTE: 0.0.0.0 workaround breaks const correctness.
1315  */
1316 static int
1317 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1318     /*const*/ struct igmp *igmp)
1319 {
1320 	struct rm_priotracker in_ifa_tracker;
1321 	struct in_ifaddr *ia;
1322 	struct in_multi *inm;
1323 
1324 	/*
1325 	 * Make sure we don't hear our own membership report.  Fast
1326 	 * leave requires knowing that we are the only member of a
1327 	 * group.
1328 	 */
1329 	IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1330 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1331 		ifa_free(&ia->ia_ifa);
1332 		return (0);
1333 	}
1334 
1335 	IGMPSTAT_INC(igps_rcv_reports);
1336 
1337 	if (ifp->if_flags & IFF_LOOPBACK) {
1338 		if (ia != NULL)
1339 			ifa_free(&ia->ia_ifa);
1340 		return (0);
1341 	}
1342 
1343 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1344 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1345 		if (ia != NULL)
1346 			ifa_free(&ia->ia_ifa);
1347 		IGMPSTAT_INC(igps_rcv_badreports);
1348 		return (EINVAL);
1349 	}
1350 
1351 	/*
1352 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1353 	 * Booting clients may use the source address 0.0.0.0. Some
1354 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1355 	 * the interface upon which this message was received.
1356 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1357 	 */
1358 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1359 		if (ia != NULL)
1360 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1361 	}
1362 	if (ia != NULL)
1363 		ifa_free(&ia->ia_ifa);
1364 
1365 	CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1366 	     ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1367 
1368 	/*
1369 	 * IGMPv2 report suppression.
1370 	 * If we are a member of this group, and our membership should be
1371 	 * reported, and our group timer is pending or about to be reset,
1372 	 * stop our group timer by transitioning to the 'lazy' state.
1373 	 */
1374 	IN_MULTI_LOCK();
1375 	inm = inm_lookup(ifp, igmp->igmp_group);
1376 	if (inm != NULL) {
1377 		struct igmp_ifsoftc *igi;
1378 
1379 		igi = inm->inm_igi;
1380 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1381 
1382 		IGMPSTAT_INC(igps_rcv_ourreports);
1383 
1384 		/*
1385 		 * If we are in IGMPv3 host mode, do not allow the
1386 		 * other host's IGMPv1 report to suppress our reports
1387 		 * unless explicitly configured to do so.
1388 		 */
1389 		if (igi->igi_version == IGMP_VERSION_3) {
1390 			if (V_igmp_legacysupp)
1391 				igmp_v3_suppress_group_record(inm);
1392 			goto out_locked;
1393 		}
1394 
1395 		inm->inm_timer = 0;
1396 
1397 		switch (inm->inm_state) {
1398 		case IGMP_NOT_MEMBER:
1399 		case IGMP_SILENT_MEMBER:
1400 		case IGMP_SLEEPING_MEMBER:
1401 			break;
1402 		case IGMP_REPORTING_MEMBER:
1403 		case IGMP_IDLE_MEMBER:
1404 		case IGMP_AWAKENING_MEMBER:
1405 			CTR3(KTR_IGMPV3,
1406 			    "report suppressed for 0x%08x on ifp %p(%s)",
1407 			    ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1408 		case IGMP_LAZY_MEMBER:
1409 			inm->inm_state = IGMP_LAZY_MEMBER;
1410 			break;
1411 		case IGMP_G_QUERY_PENDING_MEMBER:
1412 		case IGMP_SG_QUERY_PENDING_MEMBER:
1413 		case IGMP_LEAVING_MEMBER:
1414 			break;
1415 		}
1416 	}
1417 
1418 out_locked:
1419 	IN_MULTI_UNLOCK();
1420 
1421 	return (0);
1422 }
1423 
1424 int
1425 igmp_input(struct mbuf **mp, int *offp, int proto)
1426 {
1427 	int iphlen;
1428 	struct ifnet *ifp;
1429 	struct igmp *igmp;
1430 	struct ip *ip;
1431 	struct mbuf *m;
1432 	int igmplen;
1433 	int minlen;
1434 	int queryver;
1435 
1436 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1437 
1438 	m = *mp;
1439 	ifp = m->m_pkthdr.rcvif;
1440 	*mp = NULL;
1441 
1442 	IGMPSTAT_INC(igps_rcv_total);
1443 
1444 	ip = mtod(m, struct ip *);
1445 	iphlen = *offp;
1446 	igmplen = ntohs(ip->ip_len) - iphlen;
1447 
1448 	/*
1449 	 * Validate lengths.
1450 	 */
1451 	if (igmplen < IGMP_MINLEN) {
1452 		IGMPSTAT_INC(igps_rcv_tooshort);
1453 		m_freem(m);
1454 		return (IPPROTO_DONE);
1455 	}
1456 
1457 	/*
1458 	 * Always pullup to the minimum size for v1/v2 or v3
1459 	 * to amortize calls to m_pullup().
1460 	 */
1461 	minlen = iphlen;
1462 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1463 		minlen += IGMP_V3_QUERY_MINLEN;
1464 	else
1465 		minlen += IGMP_MINLEN;
1466 	if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1467 	    (m = m_pullup(m, minlen)) == NULL) {
1468 		IGMPSTAT_INC(igps_rcv_tooshort);
1469 		return (IPPROTO_DONE);
1470 	}
1471 	ip = mtod(m, struct ip *);
1472 
1473 	/*
1474 	 * Validate checksum.
1475 	 */
1476 	m->m_data += iphlen;
1477 	m->m_len -= iphlen;
1478 	igmp = mtod(m, struct igmp *);
1479 	if (in_cksum(m, igmplen)) {
1480 		IGMPSTAT_INC(igps_rcv_badsum);
1481 		m_freem(m);
1482 		return (IPPROTO_DONE);
1483 	}
1484 	m->m_data -= iphlen;
1485 	m->m_len += iphlen;
1486 
1487 	/*
1488 	 * IGMP control traffic is link-scope, and must have a TTL of 1.
1489 	 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1490 	 * probe packets may come from beyond the LAN.
1491 	 */
1492 	if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1493 		IGMPSTAT_INC(igps_rcv_badttl);
1494 		m_freem(m);
1495 		return (IPPROTO_DONE);
1496 	}
1497 
1498 	switch (igmp->igmp_type) {
1499 	case IGMP_HOST_MEMBERSHIP_QUERY:
1500 		if (igmplen == IGMP_MINLEN) {
1501 			if (igmp->igmp_code == 0)
1502 				queryver = IGMP_VERSION_1;
1503 			else
1504 				queryver = IGMP_VERSION_2;
1505 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1506 			queryver = IGMP_VERSION_3;
1507 		} else {
1508 			IGMPSTAT_INC(igps_rcv_tooshort);
1509 			m_freem(m);
1510 			return (IPPROTO_DONE);
1511 		}
1512 
1513 		switch (queryver) {
1514 		case IGMP_VERSION_1:
1515 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1516 			if (!V_igmp_v1enable)
1517 				break;
1518 			if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1519 				m_freem(m);
1520 				return (IPPROTO_DONE);
1521 			}
1522 			break;
1523 
1524 		case IGMP_VERSION_2:
1525 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1526 			if (!V_igmp_v2enable)
1527 				break;
1528 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1529 				m_freem(m);
1530 				return (IPPROTO_DONE);
1531 			}
1532 			break;
1533 
1534 		case IGMP_VERSION_3: {
1535 				struct igmpv3 *igmpv3;
1536 				uint16_t igmpv3len;
1537 				uint16_t nsrc;
1538 
1539 				IGMPSTAT_INC(igps_rcv_v3_queries);
1540 				igmpv3 = (struct igmpv3 *)igmp;
1541 				/*
1542 				 * Validate length based on source count.
1543 				 */
1544 				nsrc = ntohs(igmpv3->igmp_numsrc);
1545 				if (nsrc * sizeof(in_addr_t) >
1546 				    UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1547 					IGMPSTAT_INC(igps_rcv_tooshort);
1548 					return (IPPROTO_DONE);
1549 				}
1550 				/*
1551 				 * m_pullup() may modify m, so pullup in
1552 				 * this scope.
1553 				 */
1554 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1555 				   sizeof(struct in_addr) * nsrc;
1556 				if ((!M_WRITABLE(m) ||
1557 				     m->m_len < igmpv3len) &&
1558 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1559 					IGMPSTAT_INC(igps_rcv_tooshort);
1560 					return (IPPROTO_DONE);
1561 				}
1562 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1563 				    + iphlen);
1564 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1565 					m_freem(m);
1566 					return (IPPROTO_DONE);
1567 				}
1568 			}
1569 			break;
1570 		}
1571 		break;
1572 
1573 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1574 		if (!V_igmp_v1enable)
1575 			break;
1576 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1577 			m_freem(m);
1578 			return (IPPROTO_DONE);
1579 		}
1580 		break;
1581 
1582 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1583 		if (!V_igmp_v2enable)
1584 			break;
1585 		if (!ip_checkrouteralert(m))
1586 			IGMPSTAT_INC(igps_rcv_nora);
1587 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1588 			m_freem(m);
1589 			return (IPPROTO_DONE);
1590 		}
1591 		break;
1592 
1593 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1594 		/*
1595 		 * Hosts do not need to process IGMPv3 membership reports,
1596 		 * as report suppression is no longer required.
1597 		 */
1598 		if (!ip_checkrouteralert(m))
1599 			IGMPSTAT_INC(igps_rcv_nora);
1600 		break;
1601 
1602 	default:
1603 		break;
1604 	}
1605 
1606 	/*
1607 	 * Pass all valid IGMP packets up to any process(es) listening on a
1608 	 * raw IGMP socket.
1609 	 */
1610 	*mp = m;
1611 	return (rip_input(mp, offp, proto));
1612 }
1613 
1614 
1615 /*
1616  * Fast timeout handler (global).
1617  * VIMAGE: Timeout handlers are expected to service all vimages.
1618  */
1619 void
1620 igmp_fasttimo(void)
1621 {
1622 	VNET_ITERATOR_DECL(vnet_iter);
1623 
1624 	VNET_LIST_RLOCK_NOSLEEP();
1625 	VNET_FOREACH(vnet_iter) {
1626 		CURVNET_SET(vnet_iter);
1627 		igmp_fasttimo_vnet();
1628 		CURVNET_RESTORE();
1629 	}
1630 	VNET_LIST_RUNLOCK_NOSLEEP();
1631 }
1632 
1633 /*
1634  * Fast timeout handler (per-vnet).
1635  * Sends are shuffled off to a netisr to deal with Giant.
1636  *
1637  * VIMAGE: Assume caller has set up our curvnet.
1638  */
1639 static void
1640 igmp_fasttimo_vnet(void)
1641 {
1642 	struct mbufq		 scq;	/* State-change packets */
1643 	struct mbufq		 qrq;	/* Query response packets */
1644 	struct ifnet		*ifp;
1645 	struct igmp_ifsoftc	*igi;
1646 	struct ifmultiaddr	*ifma;
1647 	struct in_multi		*inm;
1648 	int			 loop, uri_fasthz;
1649 
1650 	loop = 0;
1651 	uri_fasthz = 0;
1652 
1653 	/*
1654 	 * Quick check to see if any work needs to be done, in order to
1655 	 * minimize the overhead of fasttimo processing.
1656 	 * SMPng: XXX Unlocked reads.
1657 	 */
1658 	if (!V_current_state_timers_running &&
1659 	    !V_interface_timers_running &&
1660 	    !V_state_change_timers_running)
1661 		return;
1662 
1663 	IN_MULTI_LOCK();
1664 	IGMP_LOCK();
1665 
1666 	/*
1667 	 * IGMPv3 General Query response timer processing.
1668 	 */
1669 	if (V_interface_timers_running) {
1670 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1671 
1672 		V_interface_timers_running = 0;
1673 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1674 			if (igi->igi_v3_timer == 0) {
1675 				/* Do nothing. */
1676 			} else if (--igi->igi_v3_timer == 0) {
1677 				igmp_v3_dispatch_general_query(igi);
1678 			} else {
1679 				V_interface_timers_running = 1;
1680 			}
1681 		}
1682 	}
1683 
1684 	if (!V_current_state_timers_running &&
1685 	    !V_state_change_timers_running)
1686 		goto out_locked;
1687 
1688 	V_current_state_timers_running = 0;
1689 	V_state_change_timers_running = 0;
1690 
1691 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1692 
1693 	/*
1694 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1695 	 * Note: Processing a v3 group timer may remove a node.
1696 	 */
1697 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1698 		ifp = igi->igi_ifp;
1699 
1700 		if (igi->igi_version == IGMP_VERSION_3) {
1701 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1702 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1703 			    PR_FASTHZ);
1704 			mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1705 			mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1706 		}
1707 
1708 		IF_ADDR_RLOCK(ifp);
1709 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1710 			if (ifma->ifma_addr->sa_family != AF_INET ||
1711 			    ifma->ifma_protospec == NULL)
1712 				continue;
1713 			inm = (struct in_multi *)ifma->ifma_protospec;
1714 			switch (igi->igi_version) {
1715 			case IGMP_VERSION_1:
1716 			case IGMP_VERSION_2:
1717 				igmp_v1v2_process_group_timer(inm,
1718 				    igi->igi_version);
1719 				break;
1720 			case IGMP_VERSION_3:
1721 				igmp_v3_process_group_timers(igi, &qrq,
1722 				    &scq, inm, uri_fasthz);
1723 				break;
1724 			}
1725 		}
1726 		IF_ADDR_RUNLOCK(ifp);
1727 
1728 		if (igi->igi_version == IGMP_VERSION_3) {
1729 			struct in_multi		*tinm;
1730 
1731 			igmp_dispatch_queue(&qrq, 0, loop);
1732 			igmp_dispatch_queue(&scq, 0, loop);
1733 
1734 			/*
1735 			 * Free the in_multi reference(s) for this
1736 			 * IGMP lifecycle.
1737 			 */
1738 			SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1739 			    inm_nrele, tinm) {
1740 				SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1741 				    inm_nrele);
1742 				inm_release_locked(inm);
1743 			}
1744 		}
1745 	}
1746 
1747 out_locked:
1748 	IGMP_UNLOCK();
1749 	IN_MULTI_UNLOCK();
1750 }
1751 
1752 /*
1753  * Update host report group timer for IGMPv1/v2.
1754  * Will update the global pending timer flags.
1755  */
1756 static void
1757 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1758 {
1759 	int report_timer_expired;
1760 
1761 	IN_MULTI_LOCK_ASSERT();
1762 	IGMP_LOCK_ASSERT();
1763 
1764 	if (inm->inm_timer == 0) {
1765 		report_timer_expired = 0;
1766 	} else if (--inm->inm_timer == 0) {
1767 		report_timer_expired = 1;
1768 	} else {
1769 		V_current_state_timers_running = 1;
1770 		return;
1771 	}
1772 
1773 	switch (inm->inm_state) {
1774 	case IGMP_NOT_MEMBER:
1775 	case IGMP_SILENT_MEMBER:
1776 	case IGMP_IDLE_MEMBER:
1777 	case IGMP_LAZY_MEMBER:
1778 	case IGMP_SLEEPING_MEMBER:
1779 	case IGMP_AWAKENING_MEMBER:
1780 		break;
1781 	case IGMP_REPORTING_MEMBER:
1782 		if (report_timer_expired) {
1783 			inm->inm_state = IGMP_IDLE_MEMBER;
1784 			(void)igmp_v1v2_queue_report(inm,
1785 			    (version == IGMP_VERSION_2) ?
1786 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1787 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1788 		}
1789 		break;
1790 	case IGMP_G_QUERY_PENDING_MEMBER:
1791 	case IGMP_SG_QUERY_PENDING_MEMBER:
1792 	case IGMP_LEAVING_MEMBER:
1793 		break;
1794 	}
1795 }
1796 
1797 /*
1798  * Update a group's timers for IGMPv3.
1799  * Will update the global pending timer flags.
1800  * Note: Unlocked read from igi.
1801  */
1802 static void
1803 igmp_v3_process_group_timers(struct igmp_ifsoftc *igi,
1804     struct mbufq *qrq, struct mbufq *scq,
1805     struct in_multi *inm, const int uri_fasthz)
1806 {
1807 	int query_response_timer_expired;
1808 	int state_change_retransmit_timer_expired;
1809 
1810 	IN_MULTI_LOCK_ASSERT();
1811 	IGMP_LOCK_ASSERT();
1812 
1813 	query_response_timer_expired = 0;
1814 	state_change_retransmit_timer_expired = 0;
1815 
1816 	/*
1817 	 * During a transition from v1/v2 compatibility mode back to v3,
1818 	 * a group record in REPORTING state may still have its group
1819 	 * timer active. This is a no-op in this function; it is easier
1820 	 * to deal with it here than to complicate the slow-timeout path.
1821 	 */
1822 	if (inm->inm_timer == 0) {
1823 		query_response_timer_expired = 0;
1824 	} else if (--inm->inm_timer == 0) {
1825 		query_response_timer_expired = 1;
1826 	} else {
1827 		V_current_state_timers_running = 1;
1828 	}
1829 
1830 	if (inm->inm_sctimer == 0) {
1831 		state_change_retransmit_timer_expired = 0;
1832 	} else if (--inm->inm_sctimer == 0) {
1833 		state_change_retransmit_timer_expired = 1;
1834 	} else {
1835 		V_state_change_timers_running = 1;
1836 	}
1837 
1838 	/* We are in fasttimo, so be quick about it. */
1839 	if (!state_change_retransmit_timer_expired &&
1840 	    !query_response_timer_expired)
1841 		return;
1842 
1843 	switch (inm->inm_state) {
1844 	case IGMP_NOT_MEMBER:
1845 	case IGMP_SILENT_MEMBER:
1846 	case IGMP_SLEEPING_MEMBER:
1847 	case IGMP_LAZY_MEMBER:
1848 	case IGMP_AWAKENING_MEMBER:
1849 	case IGMP_IDLE_MEMBER:
1850 		break;
1851 	case IGMP_G_QUERY_PENDING_MEMBER:
1852 	case IGMP_SG_QUERY_PENDING_MEMBER:
1853 		/*
1854 		 * Respond to a previously pending Group-Specific
1855 		 * or Group-and-Source-Specific query by enqueueing
1856 		 * the appropriate Current-State report for
1857 		 * immediate transmission.
1858 		 */
1859 		if (query_response_timer_expired) {
1860 			int retval;
1861 
1862 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1863 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1864 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1865 			    __func__, retval);
1866 			inm->inm_state = IGMP_REPORTING_MEMBER;
1867 			/* XXX Clear recorded sources for next time. */
1868 			inm_clear_recorded(inm);
1869 		}
1870 		/* FALLTHROUGH */
1871 	case IGMP_REPORTING_MEMBER:
1872 	case IGMP_LEAVING_MEMBER:
1873 		if (state_change_retransmit_timer_expired) {
1874 			/*
1875 			 * State-change retransmission timer fired.
1876 			 * If there are any further pending retransmissions,
1877 			 * set the global pending state-change flag, and
1878 			 * reset the timer.
1879 			 */
1880 			if (--inm->inm_scrv > 0) {
1881 				inm->inm_sctimer = uri_fasthz;
1882 				V_state_change_timers_running = 1;
1883 			}
1884 			/*
1885 			 * Retransmit the previously computed state-change
1886 			 * report. If there are no further pending
1887 			 * retransmissions, the mbuf queue will be consumed.
1888 			 * Update T0 state to T1 as we have now sent
1889 			 * a state-change.
1890 			 */
1891 			(void)igmp_v3_merge_state_changes(inm, scq);
1892 
1893 			inm_commit(inm);
1894 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1895 			    ntohl(inm->inm_addr.s_addr),
1896 			    inm->inm_ifp->if_xname);
1897 
1898 			/*
1899 			 * If we are leaving the group for good, make sure
1900 			 * we release IGMP's reference to it.
1901 			 * This release must be deferred using a SLIST,
1902 			 * as we are called from a loop which traverses
1903 			 * the in_ifmultiaddr TAILQ.
1904 			 */
1905 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1906 			    inm->inm_scrv == 0) {
1907 				inm->inm_state = IGMP_NOT_MEMBER;
1908 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1909 				    inm, inm_nrele);
1910 			}
1911 		}
1912 		break;
1913 	}
1914 }
1915 
1916 
1917 /*
1918  * Suppress a group's pending response to a group or source/group query.
1919  *
1920  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1921  * Do NOT update ST1/ST0 as this operation merely suppresses
1922  * the currently pending group record.
1923  * Do NOT suppress the response to a general query. It is possible but
1924  * it would require adding another state or flag.
1925  */
1926 static void
1927 igmp_v3_suppress_group_record(struct in_multi *inm)
1928 {
1929 
1930 	IN_MULTI_LOCK_ASSERT();
1931 
1932 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1933 		("%s: not IGMPv3 mode on link", __func__));
1934 
1935 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1936 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1937 		return;
1938 
1939 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1940 		inm_clear_recorded(inm);
1941 
1942 	inm->inm_timer = 0;
1943 	inm->inm_state = IGMP_REPORTING_MEMBER;
1944 }
1945 
1946 /*
1947  * Switch to a different IGMP version on the given interface,
1948  * as per Section 7.2.1.
1949  */
1950 static void
1951 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1952 {
1953 	int old_version_timer;
1954 
1955 	IGMP_LOCK_ASSERT();
1956 
1957 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1958 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1959 
1960 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1961 		/*
1962 		 * Compute the "Older Version Querier Present" timer as per
1963 		 * Section 8.12.
1964 		 */
1965 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1966 		old_version_timer *= PR_SLOWHZ;
1967 
1968 		if (version == IGMP_VERSION_1) {
1969 			igi->igi_v1_timer = old_version_timer;
1970 			igi->igi_v2_timer = 0;
1971 		} else if (version == IGMP_VERSION_2) {
1972 			igi->igi_v1_timer = 0;
1973 			igi->igi_v2_timer = old_version_timer;
1974 		}
1975 	}
1976 
1977 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1978 		if (igi->igi_version != IGMP_VERSION_2) {
1979 			igi->igi_version = IGMP_VERSION_2;
1980 			igmp_v3_cancel_link_timers(igi);
1981 		}
1982 	} else if (igi->igi_v1_timer > 0) {
1983 		if (igi->igi_version != IGMP_VERSION_1) {
1984 			igi->igi_version = IGMP_VERSION_1;
1985 			igmp_v3_cancel_link_timers(igi);
1986 		}
1987 	}
1988 }
1989 
1990 /*
1991  * Cancel pending IGMPv3 timers for the given link and all groups
1992  * joined on it; state-change, general-query, and group-query timers.
1993  *
1994  * Only ever called on a transition from v3 to Compatibility mode. Kill
1995  * the timers stone dead (this may be expensive for large N groups), they
1996  * will be restarted if Compatibility Mode deems that they must be due to
1997  * query processing.
1998  */
1999 static void
2000 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2001 {
2002 	struct ifmultiaddr	*ifma;
2003 	struct ifnet		*ifp;
2004 	struct in_multi		*inm, *tinm;
2005 
2006 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2007 	    igi->igi_ifp, igi->igi_ifp->if_xname);
2008 
2009 	IN_MULTI_LOCK_ASSERT();
2010 	IGMP_LOCK_ASSERT();
2011 
2012 	/*
2013 	 * Stop the v3 General Query Response on this link stone dead.
2014 	 * If fasttimo is woken up due to V_interface_timers_running,
2015 	 * the flag will be cleared if there are no pending link timers.
2016 	 */
2017 	igi->igi_v3_timer = 0;
2018 
2019 	/*
2020 	 * Now clear the current-state and state-change report timers
2021 	 * for all memberships scoped to this link.
2022 	 */
2023 	ifp = igi->igi_ifp;
2024 	IF_ADDR_RLOCK(ifp);
2025 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2026 		if (ifma->ifma_addr->sa_family != AF_INET ||
2027 		    ifma->ifma_protospec == NULL)
2028 			continue;
2029 		inm = (struct in_multi *)ifma->ifma_protospec;
2030 		switch (inm->inm_state) {
2031 		case IGMP_NOT_MEMBER:
2032 		case IGMP_SILENT_MEMBER:
2033 		case IGMP_IDLE_MEMBER:
2034 		case IGMP_LAZY_MEMBER:
2035 		case IGMP_SLEEPING_MEMBER:
2036 		case IGMP_AWAKENING_MEMBER:
2037 			/*
2038 			 * These states are either not relevant in v3 mode,
2039 			 * or are unreported. Do nothing.
2040 			 */
2041 			break;
2042 		case IGMP_LEAVING_MEMBER:
2043 			/*
2044 			 * If we are leaving the group and switching to
2045 			 * compatibility mode, we need to release the final
2046 			 * reference held for issuing the INCLUDE {}, and
2047 			 * transition to REPORTING to ensure the host leave
2048 			 * message is sent upstream to the old querier --
2049 			 * transition to NOT would lose the leave and race.
2050 			 */
2051 			SLIST_INSERT_HEAD(&igi->igi_relinmhead, inm, inm_nrele);
2052 			/* FALLTHROUGH */
2053 		case IGMP_G_QUERY_PENDING_MEMBER:
2054 		case IGMP_SG_QUERY_PENDING_MEMBER:
2055 			inm_clear_recorded(inm);
2056 			/* FALLTHROUGH */
2057 		case IGMP_REPORTING_MEMBER:
2058 			inm->inm_state = IGMP_REPORTING_MEMBER;
2059 			break;
2060 		}
2061 		/*
2062 		 * Always clear state-change and group report timers.
2063 		 * Free any pending IGMPv3 state-change records.
2064 		 */
2065 		inm->inm_sctimer = 0;
2066 		inm->inm_timer = 0;
2067 		mbufq_drain(&inm->inm_scq);
2068 	}
2069 	IF_ADDR_RUNLOCK(ifp);
2070 	SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, tinm) {
2071 		SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
2072 		inm_release_locked(inm);
2073 	}
2074 }
2075 
2076 /*
2077  * Update the Older Version Querier Present timers for a link.
2078  * See Section 7.2.1 of RFC 3376.
2079  */
2080 static void
2081 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2082 {
2083 
2084 	IGMP_LOCK_ASSERT();
2085 
2086 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2087 		/*
2088 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2089 		 *
2090 		 * Revert to IGMPv3.
2091 		 */
2092 		if (igi->igi_version != IGMP_VERSION_3) {
2093 			CTR5(KTR_IGMPV3,
2094 			    "%s: transition from v%d -> v%d on %p(%s)",
2095 			    __func__, igi->igi_version, IGMP_VERSION_3,
2096 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2097 			igi->igi_version = IGMP_VERSION_3;
2098 		}
2099 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2100 		/*
2101 		 * IGMPv1 Querier Present timer expired,
2102 		 * IGMPv2 Querier Present timer running.
2103 		 * If IGMPv2 was disabled since last timeout,
2104 		 * revert to IGMPv3.
2105 		 * If IGMPv2 is enabled, revert to IGMPv2.
2106 		 */
2107 		if (!V_igmp_v2enable) {
2108 			CTR5(KTR_IGMPV3,
2109 			    "%s: transition from v%d -> v%d on %p(%s)",
2110 			    __func__, igi->igi_version, IGMP_VERSION_3,
2111 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2112 			igi->igi_v2_timer = 0;
2113 			igi->igi_version = IGMP_VERSION_3;
2114 		} else {
2115 			--igi->igi_v2_timer;
2116 			if (igi->igi_version != IGMP_VERSION_2) {
2117 				CTR5(KTR_IGMPV3,
2118 				    "%s: transition from v%d -> v%d on %p(%s)",
2119 				    __func__, igi->igi_version, IGMP_VERSION_2,
2120 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2121 				igi->igi_version = IGMP_VERSION_2;
2122 				igmp_v3_cancel_link_timers(igi);
2123 			}
2124 		}
2125 	} else if (igi->igi_v1_timer > 0) {
2126 		/*
2127 		 * IGMPv1 Querier Present timer running.
2128 		 * Stop IGMPv2 timer if running.
2129 		 *
2130 		 * If IGMPv1 was disabled since last timeout,
2131 		 * revert to IGMPv3.
2132 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2133 		 */
2134 		if (!V_igmp_v1enable) {
2135 			CTR5(KTR_IGMPV3,
2136 			    "%s: transition from v%d -> v%d on %p(%s)",
2137 			    __func__, igi->igi_version, IGMP_VERSION_3,
2138 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2139 			igi->igi_v1_timer = 0;
2140 			igi->igi_version = IGMP_VERSION_3;
2141 		} else {
2142 			--igi->igi_v1_timer;
2143 		}
2144 		if (igi->igi_v2_timer > 0) {
2145 			CTR3(KTR_IGMPV3,
2146 			    "%s: cancel v2 timer on %p(%s)",
2147 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2148 			igi->igi_v2_timer = 0;
2149 		}
2150 	}
2151 }
2152 
2153 /*
2154  * Global slowtimo handler.
2155  * VIMAGE: Timeout handlers are expected to service all vimages.
2156  */
2157 void
2158 igmp_slowtimo(void)
2159 {
2160 	VNET_ITERATOR_DECL(vnet_iter);
2161 
2162 	VNET_LIST_RLOCK_NOSLEEP();
2163 	VNET_FOREACH(vnet_iter) {
2164 		CURVNET_SET(vnet_iter);
2165 		igmp_slowtimo_vnet();
2166 		CURVNET_RESTORE();
2167 	}
2168 	VNET_LIST_RUNLOCK_NOSLEEP();
2169 }
2170 
2171 /*
2172  * Per-vnet slowtimo handler.
2173  */
2174 static void
2175 igmp_slowtimo_vnet(void)
2176 {
2177 	struct igmp_ifsoftc *igi;
2178 
2179 	IGMP_LOCK();
2180 
2181 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2182 		igmp_v1v2_process_querier_timers(igi);
2183 	}
2184 
2185 	IGMP_UNLOCK();
2186 }
2187 
2188 /*
2189  * Dispatch an IGMPv1/v2 host report or leave message.
2190  * These are always small enough to fit inside a single mbuf.
2191  */
2192 static int
2193 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2194 {
2195 	struct ifnet		*ifp;
2196 	struct igmp		*igmp;
2197 	struct ip		*ip;
2198 	struct mbuf		*m;
2199 
2200 	IN_MULTI_LOCK_ASSERT();
2201 	IGMP_LOCK_ASSERT();
2202 
2203 	ifp = inm->inm_ifp;
2204 
2205 	m = m_gethdr(M_NOWAIT, MT_DATA);
2206 	if (m == NULL)
2207 		return (ENOMEM);
2208 	M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2209 
2210 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2211 
2212 	m->m_data += sizeof(struct ip);
2213 	m->m_len = sizeof(struct igmp);
2214 
2215 	igmp = mtod(m, struct igmp *);
2216 	igmp->igmp_type = type;
2217 	igmp->igmp_code = 0;
2218 	igmp->igmp_group = inm->inm_addr;
2219 	igmp->igmp_cksum = 0;
2220 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2221 
2222 	m->m_data -= sizeof(struct ip);
2223 	m->m_len += sizeof(struct ip);
2224 
2225 	ip = mtod(m, struct ip *);
2226 	ip->ip_tos = 0;
2227 	ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2228 	ip->ip_off = 0;
2229 	ip->ip_p = IPPROTO_IGMP;
2230 	ip->ip_src.s_addr = INADDR_ANY;
2231 
2232 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2233 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2234 	else
2235 		ip->ip_dst = inm->inm_addr;
2236 
2237 	igmp_save_context(m, ifp);
2238 
2239 	m->m_flags |= M_IGMPV2;
2240 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2241 		m->m_flags |= M_IGMP_LOOP;
2242 
2243 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2244 	netisr_dispatch(NETISR_IGMP, m);
2245 
2246 	return (0);
2247 }
2248 
2249 /*
2250  * Process a state change from the upper layer for the given IPv4 group.
2251  *
2252  * Each socket holds a reference on the in_multi in its own ip_moptions.
2253  * The socket layer will have made the necessary updates to.the group
2254  * state, it is now up to IGMP to issue a state change report if there
2255  * has been any change between T0 (when the last state-change was issued)
2256  * and T1 (now).
2257  *
2258  * We use the IGMPv3 state machine at group level. The IGMP module
2259  * however makes the decision as to which IGMP protocol version to speak.
2260  * A state change *from* INCLUDE {} always means an initial join.
2261  * A state change *to* INCLUDE {} always means a final leave.
2262  *
2263  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2264  * save ourselves a bunch of work; any exclusive mode groups need not
2265  * compute source filter lists.
2266  *
2267  * VIMAGE: curvnet should have been set by caller, as this routine
2268  * is called from the socket option handlers.
2269  */
2270 int
2271 igmp_change_state(struct in_multi *inm)
2272 {
2273 	struct igmp_ifsoftc *igi;
2274 	struct ifnet *ifp;
2275 	int error;
2276 
2277 	IN_MULTI_LOCK_ASSERT();
2278 
2279 	error = 0;
2280 
2281 	/*
2282 	 * Try to detect if the upper layer just asked us to change state
2283 	 * for an interface which has now gone away.
2284 	 */
2285 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2286 	ifp = inm->inm_ifma->ifma_ifp;
2287 	/*
2288 	 * Sanity check that netinet's notion of ifp is the
2289 	 * same as net's.
2290 	 */
2291 	KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2292 
2293 	IGMP_LOCK();
2294 
2295 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2296 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2297 
2298 	/*
2299 	 * If we detect a state transition to or from MCAST_UNDEFINED
2300 	 * for this group, then we are starting or finishing an IGMP
2301 	 * life cycle for this group.
2302 	 */
2303 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2304 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2305 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2306 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2307 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2308 			error = igmp_initial_join(inm, igi);
2309 			goto out_locked;
2310 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2311 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2312 			igmp_final_leave(inm, igi);
2313 			goto out_locked;
2314 		}
2315 	} else {
2316 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2317 	}
2318 
2319 	error = igmp_handle_state_change(inm, igi);
2320 
2321 out_locked:
2322 	IGMP_UNLOCK();
2323 	return (error);
2324 }
2325 
2326 /*
2327  * Perform the initial join for an IGMP group.
2328  *
2329  * When joining a group:
2330  *  If the group should have its IGMP traffic suppressed, do nothing.
2331  *  IGMPv1 starts sending IGMPv1 host membership reports.
2332  *  IGMPv2 starts sending IGMPv2 host membership reports.
2333  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2334  *  initial state of the membership.
2335  */
2336 static int
2337 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2338 {
2339 	struct ifnet		*ifp;
2340 	struct mbufq		*mq;
2341 	int			 error, retval, syncstates;
2342 
2343 	CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2344 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2345 
2346 	error = 0;
2347 	syncstates = 1;
2348 
2349 	ifp = inm->inm_ifp;
2350 
2351 	IN_MULTI_LOCK_ASSERT();
2352 	IGMP_LOCK_ASSERT();
2353 
2354 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2355 
2356 	/*
2357 	 * Groups joined on loopback or marked as 'not reported',
2358 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2359 	 * are never reported in any IGMP protocol exchanges.
2360 	 * All other groups enter the appropriate IGMP state machine
2361 	 * for the version in use on this link.
2362 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2363 	 * disabled for the link.
2364 	 */
2365 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2366 	    (igi->igi_flags & IGIF_SILENT) ||
2367 	    !igmp_isgroupreported(inm->inm_addr)) {
2368 		CTR1(KTR_IGMPV3,
2369 "%s: not kicking state machine for silent group", __func__);
2370 		inm->inm_state = IGMP_SILENT_MEMBER;
2371 		inm->inm_timer = 0;
2372 	} else {
2373 		/*
2374 		 * Deal with overlapping in_multi lifecycle.
2375 		 * If this group was LEAVING, then make sure
2376 		 * we drop the reference we picked up to keep the
2377 		 * group around for the final INCLUDE {} enqueue.
2378 		 */
2379 		if (igi->igi_version == IGMP_VERSION_3 &&
2380 		    inm->inm_state == IGMP_LEAVING_MEMBER)
2381 			inm_release_locked(inm);
2382 
2383 		inm->inm_state = IGMP_REPORTING_MEMBER;
2384 
2385 		switch (igi->igi_version) {
2386 		case IGMP_VERSION_1:
2387 		case IGMP_VERSION_2:
2388 			inm->inm_state = IGMP_IDLE_MEMBER;
2389 			error = igmp_v1v2_queue_report(inm,
2390 			    (igi->igi_version == IGMP_VERSION_2) ?
2391 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2392 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2393 			if (error == 0) {
2394 				inm->inm_timer = IGMP_RANDOM_DELAY(
2395 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2396 				V_current_state_timers_running = 1;
2397 			}
2398 			break;
2399 
2400 		case IGMP_VERSION_3:
2401 			/*
2402 			 * Defer update of T0 to T1, until the first copy
2403 			 * of the state change has been transmitted.
2404 			 */
2405 			syncstates = 0;
2406 
2407 			/*
2408 			 * Immediately enqueue a State-Change Report for
2409 			 * this interface, freeing any previous reports.
2410 			 * Don't kick the timers if there is nothing to do,
2411 			 * or if an error occurred.
2412 			 */
2413 			mq = &inm->inm_scq;
2414 			mbufq_drain(mq);
2415 			retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2416 			    0, 0);
2417 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2418 			    __func__, retval);
2419 			if (retval <= 0) {
2420 				error = retval * -1;
2421 				break;
2422 			}
2423 
2424 			/*
2425 			 * Schedule transmission of pending state-change
2426 			 * report up to RV times for this link. The timer
2427 			 * will fire at the next igmp_fasttimo (~200ms),
2428 			 * giving us an opportunity to merge the reports.
2429 			 */
2430 			if (igi->igi_flags & IGIF_LOOPBACK) {
2431 				inm->inm_scrv = 1;
2432 			} else {
2433 				KASSERT(igi->igi_rv > 1,
2434 				   ("%s: invalid robustness %d", __func__,
2435 				    igi->igi_rv));
2436 				inm->inm_scrv = igi->igi_rv;
2437 			}
2438 			inm->inm_sctimer = 1;
2439 			V_state_change_timers_running = 1;
2440 
2441 			error = 0;
2442 			break;
2443 		}
2444 	}
2445 
2446 	/*
2447 	 * Only update the T0 state if state change is atomic,
2448 	 * i.e. we don't need to wait for a timer to fire before we
2449 	 * can consider the state change to have been communicated.
2450 	 */
2451 	if (syncstates) {
2452 		inm_commit(inm);
2453 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2454 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2455 	}
2456 
2457 	return (error);
2458 }
2459 
2460 /*
2461  * Issue an intermediate state change during the IGMP life-cycle.
2462  */
2463 static int
2464 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2465 {
2466 	struct ifnet		*ifp;
2467 	int			 retval;
2468 
2469 	CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2470 	    ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2471 
2472 	ifp = inm->inm_ifp;
2473 
2474 	IN_MULTI_LOCK_ASSERT();
2475 	IGMP_LOCK_ASSERT();
2476 
2477 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2478 
2479 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2480 	    (igi->igi_flags & IGIF_SILENT) ||
2481 	    !igmp_isgroupreported(inm->inm_addr) ||
2482 	    (igi->igi_version != IGMP_VERSION_3)) {
2483 		if (!igmp_isgroupreported(inm->inm_addr)) {
2484 			CTR1(KTR_IGMPV3,
2485 "%s: not kicking state machine for silent group", __func__);
2486 		}
2487 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2488 		inm_commit(inm);
2489 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2490 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2491 		return (0);
2492 	}
2493 
2494 	mbufq_drain(&inm->inm_scq);
2495 
2496 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2497 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2498 	if (retval <= 0)
2499 		return (-retval);
2500 
2501 	/*
2502 	 * If record(s) were enqueued, start the state-change
2503 	 * report timer for this group.
2504 	 */
2505 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2506 	inm->inm_sctimer = 1;
2507 	V_state_change_timers_running = 1;
2508 
2509 	return (0);
2510 }
2511 
2512 /*
2513  * Perform the final leave for an IGMP group.
2514  *
2515  * When leaving a group:
2516  *  IGMPv1 does nothing.
2517  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2518  *  IGMPv3 enqueues a state-change report containing a transition
2519  *  to INCLUDE {} for immediate transmission.
2520  */
2521 static void
2522 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2523 {
2524 	int syncstates;
2525 
2526 	syncstates = 1;
2527 
2528 	CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2529 	    __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2530 	    inm->inm_ifp->if_xname);
2531 
2532 	IN_MULTI_LOCK_ASSERT();
2533 	IGMP_LOCK_ASSERT();
2534 
2535 	switch (inm->inm_state) {
2536 	case IGMP_NOT_MEMBER:
2537 	case IGMP_SILENT_MEMBER:
2538 	case IGMP_LEAVING_MEMBER:
2539 		/* Already leaving or left; do nothing. */
2540 		CTR1(KTR_IGMPV3,
2541 "%s: not kicking state machine for silent group", __func__);
2542 		break;
2543 	case IGMP_REPORTING_MEMBER:
2544 	case IGMP_IDLE_MEMBER:
2545 	case IGMP_G_QUERY_PENDING_MEMBER:
2546 	case IGMP_SG_QUERY_PENDING_MEMBER:
2547 		if (igi->igi_version == IGMP_VERSION_2) {
2548 #ifdef INVARIANTS
2549 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2550 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2551 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2552 			     __func__);
2553 #endif
2554 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2555 			inm->inm_state = IGMP_NOT_MEMBER;
2556 		} else if (igi->igi_version == IGMP_VERSION_3) {
2557 			/*
2558 			 * Stop group timer and all pending reports.
2559 			 * Immediately enqueue a state-change report
2560 			 * TO_IN {} to be sent on the next fast timeout,
2561 			 * giving us an opportunity to merge reports.
2562 			 */
2563 			mbufq_drain(&inm->inm_scq);
2564 			inm->inm_timer = 0;
2565 			if (igi->igi_flags & IGIF_LOOPBACK) {
2566 				inm->inm_scrv = 1;
2567 			} else {
2568 				inm->inm_scrv = igi->igi_rv;
2569 			}
2570 			CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2571 			    "pending retransmissions.", __func__,
2572 			    ntohl(inm->inm_addr.s_addr),
2573 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2574 			if (inm->inm_scrv == 0) {
2575 				inm->inm_state = IGMP_NOT_MEMBER;
2576 				inm->inm_sctimer = 0;
2577 			} else {
2578 				int retval;
2579 
2580 				inm_acquire_locked(inm);
2581 
2582 				retval = igmp_v3_enqueue_group_record(
2583 				    &inm->inm_scq, inm, 1, 0, 0);
2584 				KASSERT(retval != 0,
2585 				    ("%s: enqueue record = %d", __func__,
2586 				     retval));
2587 
2588 				inm->inm_state = IGMP_LEAVING_MEMBER;
2589 				inm->inm_sctimer = 1;
2590 				V_state_change_timers_running = 1;
2591 				syncstates = 0;
2592 			}
2593 			break;
2594 		}
2595 		break;
2596 	case IGMP_LAZY_MEMBER:
2597 	case IGMP_SLEEPING_MEMBER:
2598 	case IGMP_AWAKENING_MEMBER:
2599 		/* Our reports are suppressed; do nothing. */
2600 		break;
2601 	}
2602 
2603 	if (syncstates) {
2604 		inm_commit(inm);
2605 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2606 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2607 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2608 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2609 		    __func__, ntohl(inm->inm_addr.s_addr),
2610 		    inm->inm_ifp->if_xname);
2611 	}
2612 }
2613 
2614 /*
2615  * Enqueue an IGMPv3 group record to the given output queue.
2616  *
2617  * XXX This function could do with having the allocation code
2618  * split out, and the multiple-tree-walks coalesced into a single
2619  * routine as has been done in igmp_v3_enqueue_filter_change().
2620  *
2621  * If is_state_change is zero, a current-state record is appended.
2622  * If is_state_change is non-zero, a state-change report is appended.
2623  *
2624  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2625  * If is_group_query is zero, and if there is a packet with free space
2626  * at the tail of the queue, it will be appended to providing there
2627  * is enough free space.
2628  * Otherwise a new mbuf packet chain is allocated.
2629  *
2630  * If is_source_query is non-zero, each source is checked to see if
2631  * it was recorded for a Group-Source query, and will be omitted if
2632  * it is not both in-mode and recorded.
2633  *
2634  * The function will attempt to allocate leading space in the packet
2635  * for the IP/IGMP header to be prepended without fragmenting the chain.
2636  *
2637  * If successful the size of all data appended to the queue is returned,
2638  * otherwise an error code less than zero is returned, or zero if
2639  * no record(s) were appended.
2640  */
2641 static int
2642 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2643     const int is_state_change, const int is_group_query,
2644     const int is_source_query)
2645 {
2646 	struct igmp_grouprec	 ig;
2647 	struct igmp_grouprec	*pig;
2648 	struct ifnet		*ifp;
2649 	struct ip_msource	*ims, *nims;
2650 	struct mbuf		*m0, *m, *md;
2651 	int			 error, is_filter_list_change;
2652 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2653 	int			 record_has_sources;
2654 	int			 now;
2655 	int			 type;
2656 	in_addr_t		 naddr;
2657 	uint8_t			 mode;
2658 
2659 	IN_MULTI_LOCK_ASSERT();
2660 
2661 	error = 0;
2662 	ifp = inm->inm_ifp;
2663 	is_filter_list_change = 0;
2664 	m = NULL;
2665 	m0 = NULL;
2666 	m0srcs = 0;
2667 	msrcs = 0;
2668 	nbytes = 0;
2669 	nims = NULL;
2670 	record_has_sources = 1;
2671 	pig = NULL;
2672 	type = IGMP_DO_NOTHING;
2673 	mode = inm->inm_st[1].iss_fmode;
2674 
2675 	/*
2676 	 * If we did not transition out of ASM mode during t0->t1,
2677 	 * and there are no source nodes to process, we can skip
2678 	 * the generation of source records.
2679 	 */
2680 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2681 	    inm->inm_nsrc == 0)
2682 		record_has_sources = 0;
2683 
2684 	if (is_state_change) {
2685 		/*
2686 		 * Queue a state change record.
2687 		 * If the mode did not change, and there are non-ASM
2688 		 * listeners or source filters present,
2689 		 * we potentially need to issue two records for the group.
2690 		 * If we are transitioning to MCAST_UNDEFINED, we need
2691 		 * not send any sources.
2692 		 * If there are ASM listeners, and there was no filter
2693 		 * mode transition of any kind, do nothing.
2694 		 */
2695 		if (mode != inm->inm_st[0].iss_fmode) {
2696 			if (mode == MCAST_EXCLUDE) {
2697 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2698 				    __func__);
2699 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2700 			} else {
2701 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2702 				    __func__);
2703 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2704 				if (mode == MCAST_UNDEFINED)
2705 					record_has_sources = 0;
2706 			}
2707 		} else {
2708 			if (record_has_sources) {
2709 				is_filter_list_change = 1;
2710 			} else {
2711 				type = IGMP_DO_NOTHING;
2712 			}
2713 		}
2714 	} else {
2715 		/*
2716 		 * Queue a current state record.
2717 		 */
2718 		if (mode == MCAST_EXCLUDE) {
2719 			type = IGMP_MODE_IS_EXCLUDE;
2720 		} else if (mode == MCAST_INCLUDE) {
2721 			type = IGMP_MODE_IS_INCLUDE;
2722 			KASSERT(inm->inm_st[1].iss_asm == 0,
2723 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2724 			     __func__, inm, inm->inm_st[1].iss_asm));
2725 		}
2726 	}
2727 
2728 	/*
2729 	 * Generate the filter list changes using a separate function.
2730 	 */
2731 	if (is_filter_list_change)
2732 		return (igmp_v3_enqueue_filter_change(mq, inm));
2733 
2734 	if (type == IGMP_DO_NOTHING) {
2735 		CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2736 		    ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2737 		return (0);
2738 	}
2739 
2740 	/*
2741 	 * If any sources are present, we must be able to fit at least
2742 	 * one in the trailing space of the tail packet's mbuf,
2743 	 * ideally more.
2744 	 */
2745 	minrec0len = sizeof(struct igmp_grouprec);
2746 	if (record_has_sources)
2747 		minrec0len += sizeof(in_addr_t);
2748 
2749 	CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2750 	    igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2751 	    inm->inm_ifp->if_xname);
2752 
2753 	/*
2754 	 * Check if we have a packet in the tail of the queue for this
2755 	 * group into which the first group record for this group will fit.
2756 	 * Otherwise allocate a new packet.
2757 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2758 	 * Note: Group records for G/GSR query responses MUST be sent
2759 	 * in their own packet.
2760 	 */
2761 	m0 = mbufq_last(mq);
2762 	if (!is_group_query &&
2763 	    m0 != NULL &&
2764 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2765 	    (m0->m_pkthdr.len + minrec0len) <
2766 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2767 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2768 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2769 		m = m0;
2770 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2771 	} else {
2772 		if (mbufq_full(mq)) {
2773 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2774 			return (-ENOMEM);
2775 		}
2776 		m = NULL;
2777 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2778 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2779 		if (!is_state_change && !is_group_query) {
2780 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2781 			if (m)
2782 				m->m_data += IGMP_LEADINGSPACE;
2783 		}
2784 		if (m == NULL) {
2785 			m = m_gethdr(M_NOWAIT, MT_DATA);
2786 			if (m)
2787 				M_ALIGN(m, IGMP_LEADINGSPACE);
2788 		}
2789 		if (m == NULL)
2790 			return (-ENOMEM);
2791 
2792 		igmp_save_context(m, ifp);
2793 
2794 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2795 	}
2796 
2797 	/*
2798 	 * Append group record.
2799 	 * If we have sources, we don't know how many yet.
2800 	 */
2801 	ig.ig_type = type;
2802 	ig.ig_datalen = 0;
2803 	ig.ig_numsrc = 0;
2804 	ig.ig_group = inm->inm_addr;
2805 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2806 		if (m != m0)
2807 			m_freem(m);
2808 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2809 		return (-ENOMEM);
2810 	}
2811 	nbytes += sizeof(struct igmp_grouprec);
2812 
2813 	/*
2814 	 * Append as many sources as will fit in the first packet.
2815 	 * If we are appending to a new packet, the chain allocation
2816 	 * may potentially use clusters; use m_getptr() in this case.
2817 	 * If we are appending to an existing packet, we need to obtain
2818 	 * a pointer to the group record after m_append(), in case a new
2819 	 * mbuf was allocated.
2820 	 * Only append sources which are in-mode at t1. If we are
2821 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2822 	 * include source entries.
2823 	 * Only report recorded sources in our filter set when responding
2824 	 * to a group-source query.
2825 	 */
2826 	if (record_has_sources) {
2827 		if (m == m0) {
2828 			md = m_last(m);
2829 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2830 			    md->m_len - nbytes);
2831 		} else {
2832 			md = m_getptr(m, 0, &off);
2833 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2834 			    off);
2835 		}
2836 		msrcs = 0;
2837 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2838 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2839 			    ims->ims_haddr);
2840 			now = ims_get_mode(inm, ims, 1);
2841 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2842 			if ((now != mode) ||
2843 			    (now == mode && mode == MCAST_UNDEFINED)) {
2844 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2845 				continue;
2846 			}
2847 			if (is_source_query && ims->ims_stp == 0) {
2848 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2849 				    __func__);
2850 				continue;
2851 			}
2852 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2853 			naddr = htonl(ims->ims_haddr);
2854 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2855 				if (m != m0)
2856 					m_freem(m);
2857 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2858 				    __func__);
2859 				return (-ENOMEM);
2860 			}
2861 			nbytes += sizeof(in_addr_t);
2862 			++msrcs;
2863 			if (msrcs == m0srcs)
2864 				break;
2865 		}
2866 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2867 		    msrcs);
2868 		pig->ig_numsrc = htons(msrcs);
2869 		nbytes += (msrcs * sizeof(in_addr_t));
2870 	}
2871 
2872 	if (is_source_query && msrcs == 0) {
2873 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2874 		if (m != m0)
2875 			m_freem(m);
2876 		return (0);
2877 	}
2878 
2879 	/*
2880 	 * We are good to go with first packet.
2881 	 */
2882 	if (m != m0) {
2883 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2884 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2885 		mbufq_enqueue(mq, m);
2886 	} else
2887 		m->m_pkthdr.PH_vt.vt_nrecs++;
2888 
2889 	/*
2890 	 * No further work needed if no source list in packet(s).
2891 	 */
2892 	if (!record_has_sources)
2893 		return (nbytes);
2894 
2895 	/*
2896 	 * Whilst sources remain to be announced, we need to allocate
2897 	 * a new packet and fill out as many sources as will fit.
2898 	 * Always try for a cluster first.
2899 	 */
2900 	while (nims != NULL) {
2901 		if (mbufq_full(mq)) {
2902 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2903 			return (-ENOMEM);
2904 		}
2905 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2906 		if (m)
2907 			m->m_data += IGMP_LEADINGSPACE;
2908 		if (m == NULL) {
2909 			m = m_gethdr(M_NOWAIT, MT_DATA);
2910 			if (m)
2911 				M_ALIGN(m, IGMP_LEADINGSPACE);
2912 		}
2913 		if (m == NULL)
2914 			return (-ENOMEM);
2915 		igmp_save_context(m, ifp);
2916 		md = m_getptr(m, 0, &off);
2917 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2918 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2919 
2920 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2921 			if (m != m0)
2922 				m_freem(m);
2923 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2924 			return (-ENOMEM);
2925 		}
2926 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2927 		nbytes += sizeof(struct igmp_grouprec);
2928 
2929 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2930 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2931 
2932 		msrcs = 0;
2933 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2934 			CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2935 			    ims->ims_haddr);
2936 			now = ims_get_mode(inm, ims, 1);
2937 			if ((now != mode) ||
2938 			    (now == mode && mode == MCAST_UNDEFINED)) {
2939 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2940 				continue;
2941 			}
2942 			if (is_source_query && ims->ims_stp == 0) {
2943 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2944 				    __func__);
2945 				continue;
2946 			}
2947 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2948 			naddr = htonl(ims->ims_haddr);
2949 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2950 				if (m != m0)
2951 					m_freem(m);
2952 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2953 				    __func__);
2954 				return (-ENOMEM);
2955 			}
2956 			++msrcs;
2957 			if (msrcs == m0srcs)
2958 				break;
2959 		}
2960 		pig->ig_numsrc = htons(msrcs);
2961 		nbytes += (msrcs * sizeof(in_addr_t));
2962 
2963 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2964 		mbufq_enqueue(mq, m);
2965 	}
2966 
2967 	return (nbytes);
2968 }
2969 
2970 /*
2971  * Type used to mark record pass completion.
2972  * We exploit the fact we can cast to this easily from the
2973  * current filter modes on each ip_msource node.
2974  */
2975 typedef enum {
2976 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2977 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2978 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2979 	REC_FULL = REC_ALLOW | REC_BLOCK
2980 } rectype_t;
2981 
2982 /*
2983  * Enqueue an IGMPv3 filter list change to the given output queue.
2984  *
2985  * Source list filter state is held in an RB-tree. When the filter list
2986  * for a group is changed without changing its mode, we need to compute
2987  * the deltas between T0 and T1 for each source in the filter set,
2988  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2989  *
2990  * As we may potentially queue two record types, and the entire R-B tree
2991  * needs to be walked at once, we break this out into its own function
2992  * so we can generate a tightly packed queue of packets.
2993  *
2994  * XXX This could be written to only use one tree walk, although that makes
2995  * serializing into the mbuf chains a bit harder. For now we do two walks
2996  * which makes things easier on us, and it may or may not be harder on
2997  * the L2 cache.
2998  *
2999  * If successful the size of all data appended to the queue is returned,
3000  * otherwise an error code less than zero is returned, or zero if
3001  * no record(s) were appended.
3002  */
3003 static int
3004 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3005 {
3006 	static const int MINRECLEN =
3007 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3008 	struct ifnet		*ifp;
3009 	struct igmp_grouprec	 ig;
3010 	struct igmp_grouprec	*pig;
3011 	struct ip_msource	*ims, *nims;
3012 	struct mbuf		*m, *m0, *md;
3013 	in_addr_t		 naddr;
3014 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3015 	int			 nallow, nblock;
3016 	uint8_t			 mode, now, then;
3017 	rectype_t		 crt, drt, nrt;
3018 
3019 	IN_MULTI_LOCK_ASSERT();
3020 
3021 	if (inm->inm_nsrc == 0 ||
3022 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3023 		return (0);
3024 
3025 	ifp = inm->inm_ifp;			/* interface */
3026 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
3027 	crt = REC_NONE;	/* current group record type */
3028 	drt = REC_NONE;	/* mask of completed group record types */
3029 	nrt = REC_NONE;	/* record type for current node */
3030 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3031 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3032 	npbytes = 0;	/* # of bytes appended this packet */
3033 	rsrcs = 0;	/* # sources encoded in current record */
3034 	schanged = 0;	/* # nodes encoded in overall filter change */
3035 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3036 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3037 	nims = NULL;	/* next tree node pointer */
3038 
3039 	/*
3040 	 * For each possible filter record mode.
3041 	 * The first kind of source we encounter tells us which
3042 	 * is the first kind of record we start appending.
3043 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3044 	 * as the inverse of the group's filter mode.
3045 	 */
3046 	while (drt != REC_FULL) {
3047 		do {
3048 			m0 = mbufq_last(mq);
3049 			if (m0 != NULL &&
3050 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3051 			     IGMP_V3_REPORT_MAXRECS) &&
3052 			    (m0->m_pkthdr.len + MINRECLEN) <
3053 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3054 				m = m0;
3055 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3056 					    sizeof(struct igmp_grouprec)) /
3057 				    sizeof(in_addr_t);
3058 				CTR1(KTR_IGMPV3,
3059 				    "%s: use previous packet", __func__);
3060 			} else {
3061 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3062 				if (m)
3063 					m->m_data += IGMP_LEADINGSPACE;
3064 				if (m == NULL) {
3065 					m = m_gethdr(M_NOWAIT, MT_DATA);
3066 					if (m)
3067 						M_ALIGN(m, IGMP_LEADINGSPACE);
3068 				}
3069 				if (m == NULL) {
3070 					CTR1(KTR_IGMPV3,
3071 					    "%s: m_get*() failed", __func__);
3072 					return (-ENOMEM);
3073 				}
3074 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3075 				igmp_save_context(m, ifp);
3076 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3077 				    sizeof(struct igmp_grouprec)) /
3078 				    sizeof(in_addr_t);
3079 				npbytes = 0;
3080 				CTR1(KTR_IGMPV3,
3081 				    "%s: allocated new packet", __func__);
3082 			}
3083 			/*
3084 			 * Append the IGMP group record header to the
3085 			 * current packet's data area.
3086 			 * Recalculate pointer to free space for next
3087 			 * group record, in case m_append() allocated
3088 			 * a new mbuf or cluster.
3089 			 */
3090 			memset(&ig, 0, sizeof(ig));
3091 			ig.ig_group = inm->inm_addr;
3092 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3093 				if (m != m0)
3094 					m_freem(m);
3095 				CTR1(KTR_IGMPV3,
3096 				    "%s: m_append() failed", __func__);
3097 				return (-ENOMEM);
3098 			}
3099 			npbytes += sizeof(struct igmp_grouprec);
3100 			if (m != m0) {
3101 				/* new packet; offset in c hain */
3102 				md = m_getptr(m, npbytes -
3103 				    sizeof(struct igmp_grouprec), &off);
3104 				pig = (struct igmp_grouprec *)(mtod(md,
3105 				    uint8_t *) + off);
3106 			} else {
3107 				/* current packet; offset from last append */
3108 				md = m_last(m);
3109 				pig = (struct igmp_grouprec *)(mtod(md,
3110 				    uint8_t *) + md->m_len -
3111 				    sizeof(struct igmp_grouprec));
3112 			}
3113 			/*
3114 			 * Begin walking the tree for this record type
3115 			 * pass, or continue from where we left off
3116 			 * previously if we had to allocate a new packet.
3117 			 * Only report deltas in-mode at t1.
3118 			 * We need not report included sources as allowed
3119 			 * if we are in inclusive mode on the group,
3120 			 * however the converse is not true.
3121 			 */
3122 			rsrcs = 0;
3123 			if (nims == NULL)
3124 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3125 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3126 				CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3127 				    __func__, ims->ims_haddr);
3128 				now = ims_get_mode(inm, ims, 1);
3129 				then = ims_get_mode(inm, ims, 0);
3130 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3131 				    __func__, then, now);
3132 				if (now == then) {
3133 					CTR1(KTR_IGMPV3,
3134 					    "%s: skip unchanged", __func__);
3135 					continue;
3136 				}
3137 				if (mode == MCAST_EXCLUDE &&
3138 				    now == MCAST_INCLUDE) {
3139 					CTR1(KTR_IGMPV3,
3140 					    "%s: skip IN src on EX group",
3141 					    __func__);
3142 					continue;
3143 				}
3144 				nrt = (rectype_t)now;
3145 				if (nrt == REC_NONE)
3146 					nrt = (rectype_t)(~mode & REC_FULL);
3147 				if (schanged++ == 0) {
3148 					crt = nrt;
3149 				} else if (crt != nrt)
3150 					continue;
3151 				naddr = htonl(ims->ims_haddr);
3152 				if (!m_append(m, sizeof(in_addr_t),
3153 				    (void *)&naddr)) {
3154 					if (m != m0)
3155 						m_freem(m);
3156 					CTR1(KTR_IGMPV3,
3157 					    "%s: m_append() failed", __func__);
3158 					return (-ENOMEM);
3159 				}
3160 				nallow += !!(crt == REC_ALLOW);
3161 				nblock += !!(crt == REC_BLOCK);
3162 				if (++rsrcs == m0srcs)
3163 					break;
3164 			}
3165 			/*
3166 			 * If we did not append any tree nodes on this
3167 			 * pass, back out of allocations.
3168 			 */
3169 			if (rsrcs == 0) {
3170 				npbytes -= sizeof(struct igmp_grouprec);
3171 				if (m != m0) {
3172 					CTR1(KTR_IGMPV3,
3173 					    "%s: m_free(m)", __func__);
3174 					m_freem(m);
3175 				} else {
3176 					CTR1(KTR_IGMPV3,
3177 					    "%s: m_adj(m, -ig)", __func__);
3178 					m_adj(m, -((int)sizeof(
3179 					    struct igmp_grouprec)));
3180 				}
3181 				continue;
3182 			}
3183 			npbytes += (rsrcs * sizeof(in_addr_t));
3184 			if (crt == REC_ALLOW)
3185 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3186 			else if (crt == REC_BLOCK)
3187 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3188 			pig->ig_numsrc = htons(rsrcs);
3189 			/*
3190 			 * Count the new group record, and enqueue this
3191 			 * packet if it wasn't already queued.
3192 			 */
3193 			m->m_pkthdr.PH_vt.vt_nrecs++;
3194 			if (m != m0)
3195 				mbufq_enqueue(mq, m);
3196 			nbytes += npbytes;
3197 		} while (nims != NULL);
3198 		drt |= crt;
3199 		crt = (~crt & REC_FULL);
3200 	}
3201 
3202 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3203 	    nallow, nblock);
3204 
3205 	return (nbytes);
3206 }
3207 
3208 static int
3209 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3210 {
3211 	struct mbufq	*gq;
3212 	struct mbuf	*m;		/* pending state-change */
3213 	struct mbuf	*m0;		/* copy of pending state-change */
3214 	struct mbuf	*mt;		/* last state-change in packet */
3215 	int		 docopy, domerge;
3216 	u_int		 recslen;
3217 
3218 	docopy = 0;
3219 	domerge = 0;
3220 	recslen = 0;
3221 
3222 	IN_MULTI_LOCK_ASSERT();
3223 	IGMP_LOCK_ASSERT();
3224 
3225 	/*
3226 	 * If there are further pending retransmissions, make a writable
3227 	 * copy of each queued state-change message before merging.
3228 	 */
3229 	if (inm->inm_scrv > 0)
3230 		docopy = 1;
3231 
3232 	gq = &inm->inm_scq;
3233 #ifdef KTR
3234 	if (mbufq_first(gq) == NULL) {
3235 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3236 		    __func__, inm);
3237 	}
3238 #endif
3239 
3240 	m = mbufq_first(gq);
3241 	while (m != NULL) {
3242 		/*
3243 		 * Only merge the report into the current packet if
3244 		 * there is sufficient space to do so; an IGMPv3 report
3245 		 * packet may only contain 65,535 group records.
3246 		 * Always use a simple mbuf chain concatentation to do this,
3247 		 * as large state changes for single groups may have
3248 		 * allocated clusters.
3249 		 */
3250 		domerge = 0;
3251 		mt = mbufq_last(scq);
3252 		if (mt != NULL) {
3253 			recslen = m_length(m, NULL);
3254 
3255 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3256 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3257 			    IGMP_V3_REPORT_MAXRECS) &&
3258 			    (mt->m_pkthdr.len + recslen <=
3259 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3260 				domerge = 1;
3261 		}
3262 
3263 		if (!domerge && mbufq_full(gq)) {
3264 			CTR2(KTR_IGMPV3,
3265 			    "%s: outbound queue full, skipping whole packet %p",
3266 			    __func__, m);
3267 			mt = m->m_nextpkt;
3268 			if (!docopy)
3269 				m_freem(m);
3270 			m = mt;
3271 			continue;
3272 		}
3273 
3274 		if (!docopy) {
3275 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3276 			m0 = mbufq_dequeue(gq);
3277 			m = m0->m_nextpkt;
3278 		} else {
3279 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3280 			m0 = m_dup(m, M_NOWAIT);
3281 			if (m0 == NULL)
3282 				return (ENOMEM);
3283 			m0->m_nextpkt = NULL;
3284 			m = m->m_nextpkt;
3285 		}
3286 
3287 		if (!domerge) {
3288 			CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3289 			    __func__, m0, scq);
3290 			mbufq_enqueue(scq, m0);
3291 		} else {
3292 			struct mbuf *mtl;	/* last mbuf of packet mt */
3293 
3294 			CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3295 			    __func__, m0, mt);
3296 
3297 			mtl = m_last(mt);
3298 			m0->m_flags &= ~M_PKTHDR;
3299 			mt->m_pkthdr.len += recslen;
3300 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3301 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3302 
3303 			mtl->m_next = m0;
3304 		}
3305 	}
3306 
3307 	return (0);
3308 }
3309 
3310 /*
3311  * Respond to a pending IGMPv3 General Query.
3312  */
3313 static void
3314 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3315 {
3316 	struct ifmultiaddr	*ifma;
3317 	struct ifnet		*ifp;
3318 	struct in_multi		*inm;
3319 	int			 retval, loop;
3320 
3321 	IN_MULTI_LOCK_ASSERT();
3322 	IGMP_LOCK_ASSERT();
3323 
3324 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3325 	    ("%s: called when version %d", __func__, igi->igi_version));
3326 
3327 	/*
3328 	 * Check that there are some packets queued. If so, send them first.
3329 	 * For large number of groups the reply to general query can take
3330 	 * many packets, we should finish sending them before starting of
3331 	 * queuing the new reply.
3332 	 */
3333 	if (mbufq_len(&igi->igi_gq) != 0)
3334 		goto send;
3335 
3336 	ifp = igi->igi_ifp;
3337 
3338 	IF_ADDR_RLOCK(ifp);
3339 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3340 		if (ifma->ifma_addr->sa_family != AF_INET ||
3341 		    ifma->ifma_protospec == NULL)
3342 			continue;
3343 
3344 		inm = (struct in_multi *)ifma->ifma_protospec;
3345 		KASSERT(ifp == inm->inm_ifp,
3346 		    ("%s: inconsistent ifp", __func__));
3347 
3348 		switch (inm->inm_state) {
3349 		case IGMP_NOT_MEMBER:
3350 		case IGMP_SILENT_MEMBER:
3351 			break;
3352 		case IGMP_REPORTING_MEMBER:
3353 		case IGMP_IDLE_MEMBER:
3354 		case IGMP_LAZY_MEMBER:
3355 		case IGMP_SLEEPING_MEMBER:
3356 		case IGMP_AWAKENING_MEMBER:
3357 			inm->inm_state = IGMP_REPORTING_MEMBER;
3358 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3359 			    inm, 0, 0, 0);
3360 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3361 			    __func__, retval);
3362 			break;
3363 		case IGMP_G_QUERY_PENDING_MEMBER:
3364 		case IGMP_SG_QUERY_PENDING_MEMBER:
3365 		case IGMP_LEAVING_MEMBER:
3366 			break;
3367 		}
3368 	}
3369 	IF_ADDR_RUNLOCK(ifp);
3370 
3371 send:
3372 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3373 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3374 
3375 	/*
3376 	 * Slew transmission of bursts over 500ms intervals.
3377 	 */
3378 	if (mbufq_first(&igi->igi_gq) != NULL) {
3379 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3380 		    IGMP_RESPONSE_BURST_INTERVAL);
3381 		V_interface_timers_running = 1;
3382 	}
3383 }
3384 
3385 /*
3386  * Transmit the next pending IGMP message in the output queue.
3387  *
3388  * We get called from netisr_processqueue(). A mutex private to igmpoq
3389  * will be acquired and released around this routine.
3390  *
3391  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3392  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3393  * a link and uses a link-scope multicast address.
3394  */
3395 static void
3396 igmp_intr(struct mbuf *m)
3397 {
3398 	struct ip_moptions	 imo;
3399 	struct ifnet		*ifp;
3400 	struct mbuf		*ipopts, *m0;
3401 	int			 error;
3402 	uint32_t		 ifindex;
3403 
3404 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3405 
3406 	/*
3407 	 * Set VNET image pointer from enqueued mbuf chain
3408 	 * before doing anything else. Whilst we use interface
3409 	 * indexes to guard against interface detach, they are
3410 	 * unique to each VIMAGE and must be retrieved.
3411 	 */
3412 	CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3413 	ifindex = igmp_restore_context(m);
3414 
3415 	/*
3416 	 * Check if the ifnet still exists. This limits the scope of
3417 	 * any race in the absence of a global ifp lock for low cost
3418 	 * (an array lookup).
3419 	 */
3420 	ifp = ifnet_byindex(ifindex);
3421 	if (ifp == NULL) {
3422 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3423 		    __func__, m, ifindex);
3424 		m_freem(m);
3425 		IPSTAT_INC(ips_noroute);
3426 		goto out;
3427 	}
3428 
3429 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3430 
3431 	imo.imo_multicast_ttl  = 1;
3432 	imo.imo_multicast_vif  = -1;
3433 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3434 
3435 	/*
3436 	 * If the user requested that IGMP traffic be explicitly
3437 	 * redirected to the loopback interface (e.g. they are running a
3438 	 * MANET interface and the routing protocol needs to see the
3439 	 * updates), handle this now.
3440 	 */
3441 	if (m->m_flags & M_IGMP_LOOP)
3442 		imo.imo_multicast_ifp = V_loif;
3443 	else
3444 		imo.imo_multicast_ifp = ifp;
3445 
3446 	if (m->m_flags & M_IGMPV2) {
3447 		m0 = m;
3448 	} else {
3449 		m0 = igmp_v3_encap_report(ifp, m);
3450 		if (m0 == NULL) {
3451 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3452 			m_freem(m);
3453 			IPSTAT_INC(ips_odropped);
3454 			goto out;
3455 		}
3456 	}
3457 
3458 	igmp_scrub_context(m0);
3459 	m_clrprotoflags(m);
3460 	m0->m_pkthdr.rcvif = V_loif;
3461 #ifdef MAC
3462 	mac_netinet_igmp_send(ifp, m0);
3463 #endif
3464 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3465 	if (error) {
3466 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3467 		goto out;
3468 	}
3469 
3470 	IGMPSTAT_INC(igps_snd_reports);
3471 
3472 out:
3473 	/*
3474 	 * We must restore the existing vnet pointer before
3475 	 * continuing as we are run from netisr context.
3476 	 */
3477 	CURVNET_RESTORE();
3478 }
3479 
3480 /*
3481  * Encapsulate an IGMPv3 report.
3482  *
3483  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3484  * chain has already had its IP/IGMPv3 header prepended. In this case
3485  * the function will not attempt to prepend; the lengths and checksums
3486  * will however be re-computed.
3487  *
3488  * Returns a pointer to the new mbuf chain head, or NULL if the
3489  * allocation failed.
3490  */
3491 static struct mbuf *
3492 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3493 {
3494 	struct rm_priotracker	in_ifa_tracker;
3495 	struct igmp_report	*igmp;
3496 	struct ip		*ip;
3497 	int			 hdrlen, igmpreclen;
3498 
3499 	KASSERT((m->m_flags & M_PKTHDR),
3500 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3501 
3502 	igmpreclen = m_length(m, NULL);
3503 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3504 
3505 	if (m->m_flags & M_IGMPV3_HDR) {
3506 		igmpreclen -= hdrlen;
3507 	} else {
3508 		M_PREPEND(m, hdrlen, M_NOWAIT);
3509 		if (m == NULL)
3510 			return (NULL);
3511 		m->m_flags |= M_IGMPV3_HDR;
3512 	}
3513 
3514 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3515 
3516 	m->m_data += sizeof(struct ip);
3517 	m->m_len -= sizeof(struct ip);
3518 
3519 	igmp = mtod(m, struct igmp_report *);
3520 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3521 	igmp->ir_rsv1 = 0;
3522 	igmp->ir_rsv2 = 0;
3523 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3524 	igmp->ir_cksum = 0;
3525 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3526 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3527 
3528 	m->m_data -= sizeof(struct ip);
3529 	m->m_len += sizeof(struct ip);
3530 
3531 	ip = mtod(m, struct ip *);
3532 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3533 	ip->ip_len = htons(hdrlen + igmpreclen);
3534 	ip->ip_off = htons(IP_DF);
3535 	ip->ip_p = IPPROTO_IGMP;
3536 	ip->ip_sum = 0;
3537 
3538 	ip->ip_src.s_addr = INADDR_ANY;
3539 
3540 	if (m->m_flags & M_IGMP_LOOP) {
3541 		struct in_ifaddr *ia;
3542 
3543 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3544 		if (ia != NULL) {
3545 			ip->ip_src = ia->ia_addr.sin_addr;
3546 			ifa_free(&ia->ia_ifa);
3547 		}
3548 	}
3549 
3550 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3551 
3552 	return (m);
3553 }
3554 
3555 #ifdef KTR
3556 static char *
3557 igmp_rec_type_to_str(const int type)
3558 {
3559 
3560 	switch (type) {
3561 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3562 			return "TO_EX";
3563 			break;
3564 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3565 			return "TO_IN";
3566 			break;
3567 		case IGMP_MODE_IS_EXCLUDE:
3568 			return "MODE_EX";
3569 			break;
3570 		case IGMP_MODE_IS_INCLUDE:
3571 			return "MODE_IN";
3572 			break;
3573 		case IGMP_ALLOW_NEW_SOURCES:
3574 			return "ALLOW_NEW";
3575 			break;
3576 		case IGMP_BLOCK_OLD_SOURCES:
3577 			return "BLOCK_OLD";
3578 			break;
3579 		default:
3580 			break;
3581 	}
3582 	return "unknown";
3583 }
3584 #endif
3585 
3586 #ifdef VIMAGE
3587 static void
3588 vnet_igmp_init(const void *unused __unused)
3589 {
3590 
3591 	netisr_register_vnet(&igmp_nh);
3592 }
3593 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3594     vnet_igmp_init, NULL);
3595 
3596 static void
3597 vnet_igmp_uninit(const void *unused __unused)
3598 {
3599 
3600 	/* This can happen when we shutdown the entire network stack. */
3601 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3602 
3603 	netisr_unregister_vnet(&igmp_nh);
3604 }
3605 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3606     vnet_igmp_uninit, NULL);
3607 #endif
3608 
3609 #ifdef DDB
3610 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3611 {
3612 	struct igmp_ifsoftc *igi, *tigi;
3613 	LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3614 
3615 	if (!have_addr) {
3616 		db_printf("usage: show igi_list <addr>\n");
3617 		return;
3618 	}
3619 	igi_head = (struct _igi_list *)addr;
3620 
3621 	LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3622 		db_printf("igmp_ifsoftc %p:\n", igi);
3623 		db_printf("    ifp %p\n", igi->igi_ifp);
3624 		db_printf("    version %u\n", igi->igi_version);
3625 		db_printf("    v1_timer %u\n", igi->igi_v1_timer);
3626 		db_printf("    v2_timer %u\n", igi->igi_v2_timer);
3627 		db_printf("    v3_timer %u\n", igi->igi_v3_timer);
3628 		db_printf("    flags %#x\n", igi->igi_flags);
3629 		db_printf("    rv %u\n", igi->igi_rv);
3630 		db_printf("    qi %u\n", igi->igi_qi);
3631 		db_printf("    qri %u\n", igi->igi_qri);
3632 		db_printf("    uri %u\n", igi->igi_uri);
3633 		/* SLIST_HEAD(,in_multi)   igi_relinmhead */
3634 		/* struct mbufq    igi_gq; */
3635 		db_printf("\n");
3636 	}
3637 }
3638 #endif
3639 
3640 static int
3641 igmp_modevent(module_t mod, int type, void *unused __unused)
3642 {
3643 
3644 	switch (type) {
3645 	case MOD_LOAD:
3646 		CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3647 		IGMP_LOCK_INIT();
3648 		m_raopt = igmp_ra_alloc();
3649 		netisr_register(&igmp_nh);
3650 		break;
3651 	case MOD_UNLOAD:
3652 		CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3653 		netisr_unregister(&igmp_nh);
3654 		m_free(m_raopt);
3655 		m_raopt = NULL;
3656 		IGMP_LOCK_DESTROY();
3657 		break;
3658 	default:
3659 		return (EOPNOTSUPP);
3660 	}
3661 	return (0);
3662 }
3663 
3664 static moduledata_t igmp_mod = {
3665     "igmp",
3666     igmp_modevent,
3667     0
3668 };
3669 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);
3670