xref: /freebsd/sys/netinet/igmp.c (revision c0020399a650364d0134f79f3fa319f84064372d)
1 /*-
2  * Copyright (c) 2007-2009 Bruce Simpson.
3  * Copyright (c) 1988 Stephen Deering.
4  * Copyright (c) 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Stephen Deering of Stanford University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
35  */
36 
37 /*
38  * Internet Group Management Protocol (IGMP) routines.
39  * [RFC1112, RFC2236, RFC3376]
40  *
41  * Written by Steve Deering, Stanford, May 1988.
42  * Modified by Rosen Sharma, Stanford, Aug 1994.
43  * Modified by Bill Fenner, Xerox PARC, Feb 1995.
44  * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995.
45  * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson.
46  *
47  * MULTICAST Revision: 3.5.1.4
48  */
49 
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
52 
53 #include "opt_mac.h"
54 #include "opt_route.h"
55 
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/module.h>
59 #include <sys/malloc.h>
60 #include <sys/mbuf.h>
61 #include <sys/socket.h>
62 #include <sys/protosw.h>
63 #include <sys/kernel.h>
64 #include <sys/sysctl.h>
65 #include <sys/vimage.h>
66 #include <sys/ktr.h>
67 #include <sys/condvar.h>
68 
69 #include <net/if.h>
70 #include <net/netisr.h>
71 #include <net/route.h>
72 #include <net/vnet.h>
73 
74 #include <netinet/in.h>
75 #include <netinet/in_var.h>
76 #include <netinet/in_systm.h>
77 #include <netinet/ip.h>
78 #include <netinet/ip_var.h>
79 #include <netinet/ip_options.h>
80 #include <netinet/igmp.h>
81 #include <netinet/igmp_var.h>
82 #include <netinet/vinet.h>
83 
84 #include <machine/in_cksum.h>
85 
86 #include <security/mac/mac_framework.h>
87 
88 #ifndef KTR_IGMPV3
89 #define KTR_IGMPV3 KTR_SUBSYS
90 #endif
91 
92 static struct igmp_ifinfo *
93 		igi_alloc_locked(struct ifnet *);
94 static void	igi_delete_locked(const struct ifnet *);
95 static void	igmp_dispatch_queue(struct ifqueue *, int, const int);
96 static void	igmp_fasttimo_vnet(void);
97 static void	igmp_final_leave(struct in_multi *, struct igmp_ifinfo *);
98 static int	igmp_handle_state_change(struct in_multi *,
99 		    struct igmp_ifinfo *);
100 static int	igmp_initial_join(struct in_multi *, struct igmp_ifinfo *);
101 static int	igmp_input_v1_query(struct ifnet *, const struct ip *);
102 static int	igmp_input_v2_query(struct ifnet *, const struct ip *,
103 		    const struct igmp *);
104 static int	igmp_input_v3_query(struct ifnet *, const struct ip *,
105 		    /*const*/ struct igmpv3 *);
106 static int	igmp_input_v3_group_query(struct in_multi *,
107 		    struct igmp_ifinfo *, int, /*const*/ struct igmpv3 *);
108 static int	igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
109 		    /*const*/ struct igmp *);
110 static int	igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
111 		    /*const*/ struct igmp *);
112 static void	igmp_intr(struct mbuf *);
113 static int	igmp_isgroupreported(const struct in_addr);
114 static struct mbuf *
115 		igmp_ra_alloc(void);
116 #ifdef KTR
117 static char *	igmp_rec_type_to_str(const int);
118 #endif
119 static void	igmp_set_version(struct igmp_ifinfo *, const int);
120 static void	igmp_slowtimo_vnet(void);
121 static void	igmp_sysinit(void);
122 static int	igmp_v1v2_queue_report(struct in_multi *, const int);
123 static void	igmp_v1v2_process_group_timer(struct in_multi *, const int);
124 static void	igmp_v1v2_process_querier_timers(struct igmp_ifinfo *);
125 static void	igmp_v2_update_group(struct in_multi *, const int);
126 static void	igmp_v3_cancel_link_timers(struct igmp_ifinfo *);
127 static void	igmp_v3_dispatch_general_query(struct igmp_ifinfo *);
128 static struct mbuf *
129 		igmp_v3_encap_report(struct ifnet *, struct mbuf *);
130 static int	igmp_v3_enqueue_group_record(struct ifqueue *,
131 		    struct in_multi *, const int, const int, const int);
132 static int	igmp_v3_enqueue_filter_change(struct ifqueue *,
133 		    struct in_multi *);
134 static void	igmp_v3_process_group_timers(struct igmp_ifinfo *,
135 		    struct ifqueue *, struct ifqueue *, struct in_multi *,
136 		    const int);
137 static int	igmp_v3_merge_state_changes(struct in_multi *,
138 		    struct ifqueue *);
139 static void	igmp_v3_suppress_group_record(struct in_multi *);
140 static int	sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
141 static int	sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
142 static int	sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
143 
144 #ifdef VIMAGE
145 static vnet_attach_fn	vnet_igmp_iattach;
146 static vnet_detach_fn	vnet_igmp_idetach;
147 #else
148 static int	vnet_igmp_iattach(const void *);
149 static int	vnet_igmp_idetach(const void *);
150 #endif /* VIMAGE */
151 
152 /*
153  * System-wide globals.
154  *
155  * Unlocked access to these is OK, except for the global IGMP output
156  * queue. The IGMP subsystem lock ends up being system-wide for the moment,
157  * because all VIMAGEs have to share a global output queue, as netisrs
158  * themselves are not virtualized.
159  *
160  * Locking:
161  *  * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
162  *    Any may be taken independently; if any are held at the same
163  *    time, the above lock order must be followed.
164  *  * All output is delegated to the netisr.
165  *    Now that Giant has been eliminated, the netisr may be inlined.
166  *  * IN_MULTI_LOCK covers in_multi.
167  *  * IGMP_LOCK covers igmp_ifinfo and any global variables in this file,
168  *    including the output queue.
169  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
170  *    per-link state iterators.
171  *  * igmp_ifinfo is valid as long as PF_INET is attached to the interface,
172  *    therefore it is not refcounted.
173  *    We allow unlocked reads of igmp_ifinfo when accessed via in_multi.
174  *
175  * Reference counting
176  *  * IGMP acquires its own reference every time an in_multi is passed to
177  *    it and the group is being joined for the first time.
178  *  * IGMP releases its reference(s) on in_multi in a deferred way,
179  *    because the operations which process the release run as part of
180  *    a loop whose control variables are directly affected by the release
181  *    (that, and not recursing on the IF_ADDR_LOCK).
182  *
183  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
184  * to a vnet in ifp->if_vnet.
185  *
186  * SMPng: XXX We may potentially race operations on ifma_protospec.
187  * The problem is that we currently lack a clean way of taking the
188  * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
189  * as anything which modifies ifma needs to be covered by that lock.
190  * So check for ifma_protospec being NULL before proceeding.
191  */
192 struct mtx		 igmp_mtx;
193 
194 struct mbuf		*m_raopt;		 /* Router Alert option */
195 MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
196 
197 /*
198  * Global netisr output queue.
199  */
200 struct ifqueue		 igmpoq;
201 
202 /*
203  * VIMAGE-wide globals.
204  *
205  * The IGMPv3 timers themselves need to run per-image, however,
206  * protosw timers run globally (see tcp).
207  * An ifnet can only be in one vimage at a time, and the loopback
208  * ifnet, loif, is itself virtualized.
209  * It would otherwise be possible to seriously hose IGMP state,
210  * and create inconsistencies in upstream multicast routing, if you have
211  * multiple VIMAGEs running on the same link joining different multicast
212  * groups, UNLESS the "primary IP address" is different. This is because
213  * IGMP for IPv4 does not force link-local addresses to be used for each
214  * node, unlike MLD for IPv6.
215  * Obviously the IGMPv3 per-interface state has per-vimage granularity
216  * also as a result.
217  *
218  * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
219  * policy to control the address used by IGMP on the link.
220  */
221 #ifdef VIMAGE_GLOBALS
222 int	 interface_timers_running;	 /* IGMPv3 general query response */
223 int	 state_change_timers_running;	 /* IGMPv3 state-change retransmit */
224 int	 current_state_timers_running;	 /* IGMPv1/v2 host report;
225 					  * IGMPv3 g/sg query response */
226 
227 LIST_HEAD(, igmp_ifinfo)	 igi_head;
228 struct igmpstat			 igmpstat;
229 struct timeval			 igmp_gsrdelay;
230 
231 int	 igmp_recvifkludge;
232 int	 igmp_sendra;
233 int	 igmp_sendlocal;
234 int	 igmp_v1enable;
235 int	 igmp_v2enable;
236 int	 igmp_legacysupp;
237 int	 igmp_default_version;
238 #endif /* VIMAGE_GLOBALS */
239 
240 /*
241  * Virtualized sysctls.
242  */
243 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_igmp, IGMPCTL_STATS, stats,
244     CTLFLAG_RW, igmpstat, igmpstat, "");
245 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, recvifkludge,
246     CTLFLAG_RW, igmp_recvifkludge, 0,
247     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
248 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, sendra,
249     CTLFLAG_RW, igmp_sendra, 0,
250     "Send IP Router Alert option in IGMPv2/v3 messages");
251 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, sendlocal,
252     CTLFLAG_RW, igmp_sendlocal, 0,
253     "Send IGMP membership reports for 224.0.0.0/24 groups");
254 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, v1enable,
255     CTLFLAG_RW, igmp_v1enable, 0,
256     "Enable backwards compatibility with IGMPv1");
257 SYSCTL_V_INT(V_NET, vnet_inet,  _net_inet_igmp, OID_AUTO, v2enable,
258     CTLFLAG_RW, igmp_v2enable, 0,
259     "Enable backwards compatibility with IGMPv2");
260 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, legacysupp,
261     CTLFLAG_RW, igmp_legacysupp, 0,
262     "Allow v1/v2 reports to suppress v3 group responses");
263 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, default_version,
264     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, igmp_default_version, 0,
265     sysctl_igmp_default_version, "I",
266     "Default version of IGMP to run on each interface");
267 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, gsrdelay,
268     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, igmp_gsrdelay.tv_sec, 0,
269     sysctl_igmp_gsr, "I",
270     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
271 
272 /*
273  * Non-virtualized sysctls.
274  */
275 SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
276     sysctl_igmp_ifinfo, "Per-interface IGMPv3 state");
277 
278 static __inline void
279 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
280 {
281 
282 #ifdef VIMAGE
283 	m->m_pkthdr.header = ifp->if_vnet;
284 #endif /* VIMAGE */
285 	m->m_pkthdr.flowid = ifp->if_index;
286 }
287 
288 static __inline void
289 igmp_scrub_context(struct mbuf *m)
290 {
291 
292 	m->m_pkthdr.header = NULL;
293 	m->m_pkthdr.flowid = 0;
294 }
295 
296 #ifdef KTR
297 static __inline char *
298 inet_ntoa_haddr(in_addr_t haddr)
299 {
300 	struct in_addr ia;
301 
302 	ia.s_addr = htonl(haddr);
303 	return (inet_ntoa(ia));
304 }
305 #endif
306 
307 /*
308  * Restore context from a queued IGMP output chain.
309  * Return saved ifindex.
310  *
311  * VIMAGE: The assertion is there to make sure that we
312  * actually called CURVNET_SET() with what's in the mbuf chain.
313  */
314 static __inline uint32_t
315 igmp_restore_context(struct mbuf *m)
316 {
317 
318 #ifdef notyet
319 #if defined(VIMAGE) && defined(INVARIANTS)
320 	KASSERT(curvnet == (m->m_pkthdr.header),
321 	    ("%s: called when curvnet was not restored", __func__));
322 #endif
323 #endif
324 	return (m->m_pkthdr.flowid);
325 }
326 
327 /*
328  * Retrieve or set default IGMP version.
329  *
330  * VIMAGE: Assume curvnet set by caller.
331  * SMPng: NOTE: Serialized by IGMP lock.
332  */
333 static int
334 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
335 {
336 	int	 error;
337 	int	 new;
338 
339 	error = sysctl_wire_old_buffer(req, sizeof(int));
340 	if (error)
341 		return (error);
342 
343 	IGMP_LOCK();
344 
345 	new = V_igmp_default_version;
346 
347 	error = sysctl_handle_int(oidp, &new, 0, req);
348 	if (error || !req->newptr)
349 		goto out_locked;
350 
351 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
352 		error = EINVAL;
353 		goto out_locked;
354 	}
355 
356 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
357 	     V_igmp_default_version, new);
358 
359 	V_igmp_default_version = new;
360 
361 out_locked:
362 	IGMP_UNLOCK();
363 	return (error);
364 }
365 
366 /*
367  * Retrieve or set threshold between group-source queries in seconds.
368  *
369  * VIMAGE: Assume curvnet set by caller.
370  * SMPng: NOTE: Serialized by IGMP lock.
371  */
372 static int
373 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
374 {
375 	int error;
376 	int i;
377 
378 	error = sysctl_wire_old_buffer(req, sizeof(int));
379 	if (error)
380 		return (error);
381 
382 	IGMP_LOCK();
383 
384 	i = V_igmp_gsrdelay.tv_sec;
385 
386 	error = sysctl_handle_int(oidp, &i, 0, req);
387 	if (error || !req->newptr)
388 		goto out_locked;
389 
390 	if (i < -1 || i >= 60) {
391 		error = EINVAL;
392 		goto out_locked;
393 	}
394 
395 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
396 	     V_igmp_gsrdelay.tv_sec, i);
397 	V_igmp_gsrdelay.tv_sec = i;
398 
399 out_locked:
400 	IGMP_UNLOCK();
401 	return (error);
402 }
403 
404 /*
405  * Expose struct igmp_ifinfo to userland, keyed by ifindex.
406  * For use by ifmcstat(8).
407  *
408  * SMPng: NOTE: Does an unlocked ifindex space read.
409  * VIMAGE: Assume curvnet set by caller. The node handler itself
410  * is not directly virtualized.
411  */
412 static int
413 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
414 {
415 	INIT_VNET_NET(curvnet);
416 	int			*name;
417 	int			 error;
418 	u_int			 namelen;
419 	struct ifnet		*ifp;
420 	struct igmp_ifinfo	*igi;
421 
422 	name = (int *)arg1;
423 	namelen = arg2;
424 
425 	if (req->newptr != NULL)
426 		return (EPERM);
427 
428 	if (namelen != 1)
429 		return (EINVAL);
430 
431 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
432 	if (error)
433 		return (error);
434 
435 	IN_MULTI_LOCK();
436 	IGMP_LOCK();
437 
438 	if (name[0] <= 0 || name[0] > V_if_index) {
439 		error = ENOENT;
440 		goto out_locked;
441 	}
442 
443 	error = ENOENT;
444 
445 	ifp = ifnet_byindex(name[0]);
446 	if (ifp == NULL)
447 		goto out_locked;
448 
449 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
450 		if (ifp == igi->igi_ifp) {
451 			error = SYSCTL_OUT(req, igi,
452 			    sizeof(struct igmp_ifinfo));
453 			break;
454 		}
455 	}
456 
457 out_locked:
458 	IGMP_UNLOCK();
459 	IN_MULTI_UNLOCK();
460 	return (error);
461 }
462 
463 /*
464  * Dispatch an entire queue of pending packet chains
465  * using the netisr.
466  * VIMAGE: Assumes the vnet pointer has been set.
467  */
468 static void
469 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop)
470 {
471 	struct mbuf *m;
472 
473 	for (;;) {
474 		_IF_DEQUEUE(ifq, m);
475 		if (m == NULL)
476 			break;
477 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m);
478 		if (loop)
479 			m->m_flags |= M_IGMP_LOOP;
480 		netisr_dispatch(NETISR_IGMP, m);
481 		if (--limit == 0)
482 			break;
483 	}
484 }
485 
486 /*
487  * Filter outgoing IGMP report state by group.
488  *
489  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
490  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
491  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
492  * this may break certain IGMP snooping switches which rely on the old
493  * report behaviour.
494  *
495  * Return zero if the given group is one for which IGMP reports
496  * should be suppressed, or non-zero if reports should be issued.
497  */
498 static __inline int
499 igmp_isgroupreported(const struct in_addr addr)
500 {
501 
502 	if (in_allhosts(addr) ||
503 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
504 		return (0);
505 
506 	return (1);
507 }
508 
509 /*
510  * Construct a Router Alert option to use in outgoing packets.
511  */
512 static struct mbuf *
513 igmp_ra_alloc(void)
514 {
515 	struct mbuf	*m;
516 	struct ipoption	*p;
517 
518 	MGET(m, M_DONTWAIT, MT_DATA);
519 	p = mtod(m, struct ipoption *);
520 	p->ipopt_dst.s_addr = INADDR_ANY;
521 	p->ipopt_list[0] = IPOPT_RA;	/* Router Alert Option */
522 	p->ipopt_list[1] = 0x04;	/* 4 bytes long */
523 	p->ipopt_list[2] = IPOPT_EOL;	/* End of IP option list */
524 	p->ipopt_list[3] = 0x00;	/* pad byte */
525 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
526 
527 	return (m);
528 }
529 
530 /*
531  * Attach IGMP when PF_INET is attached to an interface.
532  *
533  * VIMAGE: Currently we set the vnet pointer, although it is
534  * likely that it was already set by our caller.
535  */
536 struct igmp_ifinfo *
537 igmp_domifattach(struct ifnet *ifp)
538 {
539 	struct igmp_ifinfo *igi;
540 
541 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
542 	    __func__, ifp, ifp->if_xname);
543 
544 	CURVNET_SET(ifp->if_vnet);
545 	IGMP_LOCK();
546 
547 	igi = igi_alloc_locked(ifp);
548 	if (!(ifp->if_flags & IFF_MULTICAST))
549 		igi->igi_flags |= IGIF_SILENT;
550 
551 	IGMP_UNLOCK();
552 	CURVNET_RESTORE();
553 
554 	return (igi);
555 }
556 
557 /*
558  * VIMAGE: assume curvnet set by caller.
559  */
560 static struct igmp_ifinfo *
561 igi_alloc_locked(/*const*/ struct ifnet *ifp)
562 {
563 	struct igmp_ifinfo *igi;
564 
565 	IGMP_LOCK_ASSERT();
566 
567 	igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO);
568 	if (igi == NULL)
569 		goto out;
570 
571 	igi->igi_ifp = ifp;
572 	igi->igi_version = V_igmp_default_version;
573 	igi->igi_flags = 0;
574 	igi->igi_rv = IGMP_RV_INIT;
575 	igi->igi_qi = IGMP_QI_INIT;
576 	igi->igi_qri = IGMP_QRI_INIT;
577 	igi->igi_uri = IGMP_URI_INIT;
578 
579 	SLIST_INIT(&igi->igi_relinmhead);
580 
581 	/*
582 	 * Responses to general queries are subject to bounds.
583 	 */
584 	IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
585 
586 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
587 
588 	CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)",
589 	     ifp, ifp->if_xname);
590 
591 out:
592 	return (igi);
593 }
594 
595 /*
596  * Hook for ifdetach.
597  *
598  * NOTE: Some finalization tasks need to run before the protocol domain
599  * is detached, but also before the link layer does its cleanup.
600  *
601  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
602  * XXX This is also bitten by unlocked ifma_protospec access.
603  *
604  * VIMAGE: curvnet should have been set by caller, but let's not assume
605  * that for now.
606  */
607 void
608 igmp_ifdetach(struct ifnet *ifp)
609 {
610 	struct igmp_ifinfo	*igi;
611 	struct ifmultiaddr	*ifma;
612 	struct in_multi		*inm, *tinm;
613 
614 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
615 	    ifp->if_xname);
616 
617 	CURVNET_SET(ifp->if_vnet);
618 
619 	IGMP_LOCK();
620 
621 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
622 	if (igi->igi_version == IGMP_VERSION_3) {
623 		IF_ADDR_LOCK(ifp);
624 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
625 			if (ifma->ifma_addr->sa_family != AF_INET ||
626 			    ifma->ifma_protospec == NULL)
627 				continue;
628 #if 0
629 			KASSERT(ifma->ifma_protospec != NULL,
630 			    ("%s: ifma_protospec is NULL", __func__));
631 #endif
632 			inm = (struct in_multi *)ifma->ifma_protospec;
633 			if (inm->inm_state == IGMP_LEAVING_MEMBER) {
634 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
635 				    inm, inm_nrele);
636 			}
637 			inm_clear_recorded(inm);
638 		}
639 		IF_ADDR_UNLOCK(ifp);
640 		/*
641 		 * Free the in_multi reference(s) for this IGMP lifecycle.
642 		 */
643 		SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
644 		    tinm) {
645 			SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
646 			inm_release_locked(inm);
647 		}
648 	}
649 
650 	IGMP_UNLOCK();
651 
652 	CURVNET_RESTORE();
653 }
654 
655 /*
656  * Hook for domifdetach.
657  *
658  * VIMAGE: curvnet should have been set by caller, but let's not assume
659  * that for now.
660  */
661 void
662 igmp_domifdetach(struct ifnet *ifp)
663 {
664 	struct igmp_ifinfo *igi;
665 
666 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
667 	    __func__, ifp, ifp->if_xname);
668 
669 	CURVNET_SET(ifp->if_vnet);
670 	IGMP_LOCK();
671 
672 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
673 	igi_delete_locked(ifp);
674 
675 	IGMP_UNLOCK();
676 	CURVNET_RESTORE();
677 }
678 
679 static void
680 igi_delete_locked(const struct ifnet *ifp)
681 {
682 	struct igmp_ifinfo *igi, *tigi;
683 
684 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
685 	    __func__, ifp, ifp->if_xname);
686 
687 	IGMP_LOCK_ASSERT();
688 
689 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
690 		if (igi->igi_ifp == ifp) {
691 			/*
692 			 * Free deferred General Query responses.
693 			 */
694 			_IF_DRAIN(&igi->igi_gq);
695 
696 			LIST_REMOVE(igi, igi_link);
697 
698 			KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
699 			    ("%s: there are dangling in_multi references",
700 			    __func__));
701 
702 			free(igi, M_IGMP);
703 			return;
704 		}
705 	}
706 
707 #ifdef INVARIANTS
708 	panic("%s: igmp_ifinfo not found for ifp %p\n", __func__,  ifp);
709 #endif
710 }
711 
712 /*
713  * Process a received IGMPv1 query.
714  * Return non-zero if the message should be dropped.
715  *
716  * VIMAGE: The curvnet pointer is derived from the input ifp.
717  */
718 static int
719 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip)
720 {
721 	INIT_VNET_INET(ifp->if_vnet);
722 	struct ifmultiaddr	*ifma;
723 	struct igmp_ifinfo	*igi;
724 	struct in_multi		*inm;
725 
726 	/*
727 	 * IGMPv1 General Queries SHOULD always addressed to 224.0.0.1.
728 	 * igmp_group is always ignored. Do not drop it as a userland
729 	 * daemon may wish to see it.
730 	 */
731 	if (!in_allhosts(ip->ip_dst)) {
732 		IGMPSTAT_INC(igps_rcv_badqueries);
733 		return (0);
734 	}
735 
736 	IGMPSTAT_INC(igps_rcv_gen_queries);
737 
738 	/*
739 	 * Switch to IGMPv1 host compatibility mode.
740 	 */
741 	IN_MULTI_LOCK();
742 	IGMP_LOCK();
743 
744 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
745 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
746 
747 	if (igi->igi_flags & IGIF_LOOPBACK) {
748 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
749 		    ifp, ifp->if_xname);
750 		goto out_locked;
751 	}
752 
753 	igmp_set_version(igi, IGMP_VERSION_1);
754 
755 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
756 
757 	/*
758 	 * Start the timers in all of our group records
759 	 * for the interface on which the query arrived,
760 	 * except those which are already running.
761 	 */
762 	IF_ADDR_LOCK(ifp);
763 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
764 		if (ifma->ifma_addr->sa_family != AF_INET ||
765 		    ifma->ifma_protospec == NULL)
766 			continue;
767 		inm = (struct in_multi *)ifma->ifma_protospec;
768 		if (inm->inm_timer != 0)
769 			continue;
770 		switch (inm->inm_state) {
771 		case IGMP_NOT_MEMBER:
772 		case IGMP_SILENT_MEMBER:
773 			break;
774 		case IGMP_G_QUERY_PENDING_MEMBER:
775 		case IGMP_SG_QUERY_PENDING_MEMBER:
776 		case IGMP_REPORTING_MEMBER:
777 		case IGMP_IDLE_MEMBER:
778 		case IGMP_LAZY_MEMBER:
779 		case IGMP_SLEEPING_MEMBER:
780 		case IGMP_AWAKENING_MEMBER:
781 			inm->inm_state = IGMP_REPORTING_MEMBER;
782 			inm->inm_timer = IGMP_RANDOM_DELAY(
783 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
784 			V_current_state_timers_running = 1;
785 			break;
786 		case IGMP_LEAVING_MEMBER:
787 			break;
788 		}
789 	}
790 	IF_ADDR_UNLOCK(ifp);
791 
792 out_locked:
793 	IGMP_UNLOCK();
794 	IN_MULTI_UNLOCK();
795 
796 	return (0);
797 }
798 
799 /*
800  * Process a received IGMPv2 general or group-specific query.
801  */
802 static int
803 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
804     const struct igmp *igmp)
805 {
806 	struct ifmultiaddr	*ifma;
807 	struct igmp_ifinfo	*igi;
808 	struct in_multi		*inm;
809 	uint16_t		 timer;
810 
811 	/*
812 	 * Perform lazy allocation of IGMP link info if required,
813 	 * and switch to IGMPv2 host compatibility mode.
814 	 */
815 	IN_MULTI_LOCK();
816 	IGMP_LOCK();
817 
818 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
819 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
820 
821 	if (igi->igi_flags & IGIF_LOOPBACK) {
822 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
823 		    ifp, ifp->if_xname);
824 		goto out_locked;
825 	}
826 
827 	igmp_set_version(igi, IGMP_VERSION_2);
828 
829 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
830 	if (timer == 0)
831 		timer = 1;
832 
833 	if (!in_nullhost(igmp->igmp_group)) {
834 		/*
835 		 * IGMPv2 Group-Specific Query.
836 		 * If this is a group-specific IGMPv2 query, we need only
837 		 * look up the single group to process it.
838 		 */
839 		inm = inm_lookup(ifp, igmp->igmp_group);
840 		if (inm != NULL) {
841 			CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
842 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
843 			igmp_v2_update_group(inm, timer);
844 		}
845 		IGMPSTAT_INC(igps_rcv_group_queries);
846 	} else {
847 		/*
848 		 * IGMPv2 General Query.
849 		 * If this was not sent to the all-hosts group, ignore it.
850 		 */
851 		if (in_allhosts(ip->ip_dst)) {
852 			/*
853 			 * For each reporting group joined on this
854 			 * interface, kick the report timer.
855 			 */
856 			CTR2(KTR_IGMPV3,
857 			    "process v2 general query on ifp %p(%s)",
858 			    ifp, ifp->if_xname);
859 
860 			IF_ADDR_LOCK(ifp);
861 			TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
862 				if (ifma->ifma_addr->sa_family != AF_INET ||
863 				    ifma->ifma_protospec == NULL)
864 					continue;
865 				inm = (struct in_multi *)ifma->ifma_protospec;
866 				igmp_v2_update_group(inm, timer);
867 			}
868 			IF_ADDR_UNLOCK(ifp);
869 		}
870 		IGMPSTAT_INC(igps_rcv_gen_queries);
871 	}
872 
873 out_locked:
874 	IGMP_UNLOCK();
875 	IN_MULTI_UNLOCK();
876 
877 	return (0);
878 }
879 
880 /*
881  * Update the report timer on a group in response to an IGMPv2 query.
882  *
883  * If we are becoming the reporting member for this group, start the timer.
884  * If we already are the reporting member for this group, and timer is
885  * below the threshold, reset it.
886  *
887  * We may be updating the group for the first time since we switched
888  * to IGMPv3. If we are, then we must clear any recorded source lists,
889  * and transition to REPORTING state; the group timer is overloaded
890  * for group and group-source query responses.
891  *
892  * Unlike IGMPv3, the delay per group should be jittered
893  * to avoid bursts of IGMPv2 reports.
894  */
895 static void
896 igmp_v2_update_group(struct in_multi *inm, const int timer)
897 {
898 
899 	CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
900 	    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
901 
902 	IN_MULTI_LOCK_ASSERT();
903 
904 	switch (inm->inm_state) {
905 	case IGMP_NOT_MEMBER:
906 	case IGMP_SILENT_MEMBER:
907 		break;
908 	case IGMP_REPORTING_MEMBER:
909 		if (inm->inm_timer != 0 &&
910 		    inm->inm_timer <= timer) {
911 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
912 			    "skipping.", __func__);
913 			break;
914 		}
915 		/* FALLTHROUGH */
916 	case IGMP_SG_QUERY_PENDING_MEMBER:
917 	case IGMP_G_QUERY_PENDING_MEMBER:
918 	case IGMP_IDLE_MEMBER:
919 	case IGMP_LAZY_MEMBER:
920 	case IGMP_AWAKENING_MEMBER:
921 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
922 		inm->inm_state = IGMP_REPORTING_MEMBER;
923 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
924 		V_current_state_timers_running = 1;
925 		break;
926 	case IGMP_SLEEPING_MEMBER:
927 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
928 		inm->inm_state = IGMP_AWAKENING_MEMBER;
929 		break;
930 	case IGMP_LEAVING_MEMBER:
931 		break;
932 	}
933 }
934 
935 /*
936  * Process a received IGMPv3 general, group-specific or
937  * group-and-source-specific query.
938  * Assumes m has already been pulled up to the full IGMP message length.
939  * Return 0 if successful, otherwise an appropriate error code is returned.
940  */
941 static int
942 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
943     /*const*/ struct igmpv3 *igmpv3)
944 {
945 	struct igmp_ifinfo	*igi;
946 	struct in_multi		*inm;
947 	uint32_t		 maxresp, nsrc, qqi;
948 	uint16_t		 timer;
949 	uint8_t			 qrv;
950 
951 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
952 
953 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
954 	if (maxresp >= 128) {
955 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
956 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
957 	}
958 
959 	/*
960 	 * Robustness must never be less than 2 for on-wire IGMPv3.
961 	 * FIXME: Check if ifp has IGIF_LOOPBACK set, as we make
962 	 * an exception for interfaces whose IGMPv3 state changes
963 	 * are redirected to loopback (e.g. MANET).
964 	 */
965 	qrv = IGMP_QRV(igmpv3->igmp_misc);
966 	if (qrv < 2) {
967 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
968 		    qrv, IGMP_RV_INIT);
969 		qrv = IGMP_RV_INIT;
970 	}
971 
972 	qqi = igmpv3->igmp_qqi;
973 	if (qqi >= 128) {
974 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
975 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
976 	}
977 
978 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
979 	if (timer == 0)
980 		timer = 1;
981 
982 	nsrc = ntohs(igmpv3->igmp_numsrc);
983 
984 	IN_MULTI_LOCK();
985 	IGMP_LOCK();
986 
987 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
988 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
989 
990 	if (igi->igi_flags & IGIF_LOOPBACK) {
991 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
992 		    ifp, ifp->if_xname);
993 		goto out_locked;
994 	}
995 
996 	igmp_set_version(igi, IGMP_VERSION_3);
997 
998 	igi->igi_rv = qrv;
999 	igi->igi_qi = qqi;
1000 	igi->igi_qri = maxresp;
1001 
1002 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1003 	    maxresp);
1004 
1005 	if (in_nullhost(igmpv3->igmp_group)) {
1006 		/*
1007 		 * IGMPv3 General Query.
1008 		 * Schedule a current-state report on this ifp for
1009 		 * all groups, possibly containing source lists.
1010 		 */
1011 		IGMPSTAT_INC(igps_rcv_gen_queries);
1012 
1013 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1014 			/*
1015 			 * General Queries SHOULD be directed to 224.0.0.1.
1016 			 * A general query with a source list has undefined
1017 			 * behaviour; discard it.
1018 			 */
1019 			IGMPSTAT_INC(igps_rcv_badqueries);
1020 			goto out_locked;
1021 		}
1022 
1023 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1024 		    ifp, ifp->if_xname);
1025 
1026 		/*
1027 		 * If there is a pending General Query response
1028 		 * scheduled earlier than the selected delay, do
1029 		 * not schedule any other reports.
1030 		 * Otherwise, reset the interface timer.
1031 		 */
1032 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1033 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1034 			V_interface_timers_running = 1;
1035 		}
1036 	} else {
1037 		/*
1038 		 * IGMPv3 Group-specific or Group-and-source-specific Query.
1039 		 *
1040 		 * Group-source-specific queries are throttled on
1041 		 * a per-group basis to defeat denial-of-service attempts.
1042 		 * Queries for groups we are not a member of on this
1043 		 * link are simply ignored.
1044 		 */
1045 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1046 		if (inm == NULL)
1047 			goto out_locked;
1048 		if (nsrc > 0) {
1049 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1050 			if (!ratecheck(&inm->inm_lastgsrtv,
1051 			    &V_igmp_gsrdelay)) {
1052 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1053 				    __func__);
1054 				IGMPSTAT_INC(igps_drop_gsr_queries);
1055 				goto out_locked;
1056 			}
1057 		} else {
1058 			IGMPSTAT_INC(igps_rcv_group_queries);
1059 		}
1060 		CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1061 		     inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1062 		/*
1063 		 * If there is a pending General Query response
1064 		 * scheduled sooner than the selected delay, no
1065 		 * further report need be scheduled.
1066 		 * Otherwise, prepare to respond to the
1067 		 * group-specific or group-and-source query.
1068 		 */
1069 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1070 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1071 	}
1072 
1073 out_locked:
1074 	IGMP_UNLOCK();
1075 	IN_MULTI_UNLOCK();
1076 
1077 	return (0);
1078 }
1079 
1080 /*
1081  * Process a recieved IGMPv3 group-specific or group-and-source-specific
1082  * query.
1083  * Return <0 if any error occured. Currently this is ignored.
1084  */
1085 static int
1086 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1087     int timer, /*const*/ struct igmpv3 *igmpv3)
1088 {
1089 	int			 retval;
1090 	uint16_t		 nsrc;
1091 
1092 	IN_MULTI_LOCK_ASSERT();
1093 	IGMP_LOCK_ASSERT();
1094 
1095 	retval = 0;
1096 
1097 	switch (inm->inm_state) {
1098 	case IGMP_NOT_MEMBER:
1099 	case IGMP_SILENT_MEMBER:
1100 	case IGMP_SLEEPING_MEMBER:
1101 	case IGMP_LAZY_MEMBER:
1102 	case IGMP_AWAKENING_MEMBER:
1103 	case IGMP_IDLE_MEMBER:
1104 	case IGMP_LEAVING_MEMBER:
1105 		return (retval);
1106 		break;
1107 	case IGMP_REPORTING_MEMBER:
1108 	case IGMP_G_QUERY_PENDING_MEMBER:
1109 	case IGMP_SG_QUERY_PENDING_MEMBER:
1110 		break;
1111 	}
1112 
1113 	nsrc = ntohs(igmpv3->igmp_numsrc);
1114 
1115 	/*
1116 	 * Deal with group-specific queries upfront.
1117 	 * If any group query is already pending, purge any recorded
1118 	 * source-list state if it exists, and schedule a query response
1119 	 * for this group-specific query.
1120 	 */
1121 	if (nsrc == 0) {
1122 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1123 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1124 			inm_clear_recorded(inm);
1125 			timer = min(inm->inm_timer, timer);
1126 		}
1127 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1128 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1129 		V_current_state_timers_running = 1;
1130 		return (retval);
1131 	}
1132 
1133 	/*
1134 	 * Deal with the case where a group-and-source-specific query has
1135 	 * been received but a group-specific query is already pending.
1136 	 */
1137 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1138 		timer = min(inm->inm_timer, timer);
1139 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1140 		V_current_state_timers_running = 1;
1141 		return (retval);
1142 	}
1143 
1144 	/*
1145 	 * Finally, deal with the case where a group-and-source-specific
1146 	 * query has been received, where a response to a previous g-s-r
1147 	 * query exists, or none exists.
1148 	 * In this case, we need to parse the source-list which the Querier
1149 	 * has provided us with and check if we have any source list filter
1150 	 * entries at T1 for these sources. If we do not, there is no need
1151 	 * schedule a report and the query may be dropped.
1152 	 * If we do, we must record them and schedule a current-state
1153 	 * report for those sources.
1154 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1155 	 * we pass the mbuf chain pointer down to this function, and use
1156 	 * m_getptr() to walk the chain.
1157 	 */
1158 	if (inm->inm_nsrc > 0) {
1159 		const struct in_addr	*ap;
1160 		int			 i, nrecorded;
1161 
1162 		ap = (const struct in_addr *)(igmpv3 + 1);
1163 		nrecorded = 0;
1164 		for (i = 0; i < nsrc; i++, ap++) {
1165 			retval = inm_record_source(inm, ap->s_addr);
1166 			if (retval < 0)
1167 				break;
1168 			nrecorded += retval;
1169 		}
1170 		if (nrecorded > 0) {
1171 			CTR1(KTR_IGMPV3,
1172 			    "%s: schedule response to SG query", __func__);
1173 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1174 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1175 			V_current_state_timers_running = 1;
1176 		}
1177 	}
1178 
1179 	return (retval);
1180 }
1181 
1182 /*
1183  * Process a received IGMPv1 host membership report.
1184  *
1185  * NOTE: 0.0.0.0 workaround breaks const correctness.
1186  */
1187 static int
1188 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1189     /*const*/ struct igmp *igmp)
1190 {
1191 	struct in_ifaddr *ia;
1192 	struct in_multi *inm;
1193 
1194 	IGMPSTAT_INC(igps_rcv_reports);
1195 
1196 	if (ifp->if_flags & IFF_LOOPBACK)
1197 		return (0);
1198 
1199 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr) ||
1200 	    !in_hosteq(igmp->igmp_group, ip->ip_dst))) {
1201 		IGMPSTAT_INC(igps_rcv_badreports);
1202 		return (EINVAL);
1203 	}
1204 
1205 	/*
1206 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1207 	 * Booting clients may use the source address 0.0.0.0. Some
1208 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1209 	 * the interface upon which this message was received.
1210 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1211 	 */
1212 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1213 		IFP_TO_IA(ifp, ia);
1214 		if (ia != NULL)
1215 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1216 	}
1217 
1218 	CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1219 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1220 
1221 	/*
1222 	 * IGMPv1 report suppression.
1223 	 * If we are a member of this group, and our membership should be
1224 	 * reported, stop our group timer and transition to the 'lazy' state.
1225 	 */
1226 	IN_MULTI_LOCK();
1227 	inm = inm_lookup(ifp, igmp->igmp_group);
1228 	if (inm != NULL) {
1229 		struct igmp_ifinfo *igi;
1230 
1231 		igi = inm->inm_igi;
1232 		if (igi == NULL) {
1233 			KASSERT(igi != NULL,
1234 			    ("%s: no igi for ifp %p", __func__, ifp));
1235 			goto out_locked;
1236 		}
1237 
1238 		IGMPSTAT_INC(igps_rcv_ourreports);
1239 
1240 		/*
1241 		 * If we are in IGMPv3 host mode, do not allow the
1242 		 * other host's IGMPv1 report to suppress our reports
1243 		 * unless explicitly configured to do so.
1244 		 */
1245 		if (igi->igi_version == IGMP_VERSION_3) {
1246 			if (V_igmp_legacysupp)
1247 				igmp_v3_suppress_group_record(inm);
1248 			goto out_locked;
1249 		}
1250 
1251 		inm->inm_timer = 0;
1252 
1253 		switch (inm->inm_state) {
1254 		case IGMP_NOT_MEMBER:
1255 		case IGMP_SILENT_MEMBER:
1256 			break;
1257 		case IGMP_IDLE_MEMBER:
1258 		case IGMP_LAZY_MEMBER:
1259 		case IGMP_AWAKENING_MEMBER:
1260 			CTR3(KTR_IGMPV3,
1261 			    "report suppressed for %s on ifp %p(%s)",
1262 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1263 		case IGMP_SLEEPING_MEMBER:
1264 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1265 			break;
1266 		case IGMP_REPORTING_MEMBER:
1267 			CTR3(KTR_IGMPV3,
1268 			    "report suppressed for %s on ifp %p(%s)",
1269 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1270 			if (igi->igi_version == IGMP_VERSION_1)
1271 				inm->inm_state = IGMP_LAZY_MEMBER;
1272 			else if (igi->igi_version == IGMP_VERSION_2)
1273 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1274 			break;
1275 		case IGMP_G_QUERY_PENDING_MEMBER:
1276 		case IGMP_SG_QUERY_PENDING_MEMBER:
1277 		case IGMP_LEAVING_MEMBER:
1278 			break;
1279 		}
1280 	}
1281 
1282 out_locked:
1283 	IN_MULTI_UNLOCK();
1284 
1285 	return (0);
1286 }
1287 
1288 /*
1289  * Process a received IGMPv2 host membership report.
1290  *
1291  * NOTE: 0.0.0.0 workaround breaks const correctness.
1292  */
1293 static int
1294 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1295     /*const*/ struct igmp *igmp)
1296 {
1297 	struct in_ifaddr *ia;
1298 	struct in_multi *inm;
1299 
1300 	/*
1301 	 * Make sure we don't hear our own membership report.  Fast
1302 	 * leave requires knowing that we are the only member of a
1303 	 * group.
1304 	 */
1305 	IFP_TO_IA(ifp, ia);
1306 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr))
1307 		return (0);
1308 
1309 	IGMPSTAT_INC(igps_rcv_reports);
1310 
1311 	if (ifp->if_flags & IFF_LOOPBACK)
1312 		return (0);
1313 
1314 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1315 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1316 		IGMPSTAT_INC(igps_rcv_badreports);
1317 		return (EINVAL);
1318 	}
1319 
1320 	/*
1321 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1322 	 * Booting clients may use the source address 0.0.0.0. Some
1323 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1324 	 * the interface upon which this message was received.
1325 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1326 	 */
1327 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1328 		if (ia != NULL)
1329 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1330 	}
1331 
1332 	CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1333 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1334 
1335 	/*
1336 	 * IGMPv2 report suppression.
1337 	 * If we are a member of this group, and our membership should be
1338 	 * reported, and our group timer is pending or about to be reset,
1339 	 * stop our group timer by transitioning to the 'lazy' state.
1340 	 */
1341 	IN_MULTI_LOCK();
1342 	inm = inm_lookup(ifp, igmp->igmp_group);
1343 	if (inm != NULL) {
1344 		struct igmp_ifinfo *igi;
1345 
1346 		igi = inm->inm_igi;
1347 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1348 
1349 		IGMPSTAT_INC(igps_rcv_ourreports);
1350 
1351 		/*
1352 		 * If we are in IGMPv3 host mode, do not allow the
1353 		 * other host's IGMPv1 report to suppress our reports
1354 		 * unless explicitly configured to do so.
1355 		 */
1356 		if (igi->igi_version == IGMP_VERSION_3) {
1357 			if (V_igmp_legacysupp)
1358 				igmp_v3_suppress_group_record(inm);
1359 			goto out_locked;
1360 		}
1361 
1362 		inm->inm_timer = 0;
1363 
1364 		switch (inm->inm_state) {
1365 		case IGMP_NOT_MEMBER:
1366 		case IGMP_SILENT_MEMBER:
1367 		case IGMP_SLEEPING_MEMBER:
1368 			break;
1369 		case IGMP_REPORTING_MEMBER:
1370 		case IGMP_IDLE_MEMBER:
1371 		case IGMP_AWAKENING_MEMBER:
1372 			CTR3(KTR_IGMPV3,
1373 			    "report suppressed for %s on ifp %p(%s)",
1374 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1375 		case IGMP_LAZY_MEMBER:
1376 			inm->inm_state = IGMP_LAZY_MEMBER;
1377 			break;
1378 		case IGMP_G_QUERY_PENDING_MEMBER:
1379 		case IGMP_SG_QUERY_PENDING_MEMBER:
1380 		case IGMP_LEAVING_MEMBER:
1381 			break;
1382 		}
1383 	}
1384 
1385 out_locked:
1386 	IN_MULTI_UNLOCK();
1387 
1388 	return (0);
1389 }
1390 
1391 void
1392 igmp_input(struct mbuf *m, int off)
1393 {
1394 	int iphlen;
1395 	struct ifnet *ifp;
1396 	struct igmp *igmp;
1397 	struct ip *ip;
1398 	int igmplen;
1399 	int minlen;
1400 	int queryver;
1401 
1402 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1403 
1404 	ifp = m->m_pkthdr.rcvif;
1405 	INIT_VNET_INET(ifp->if_vnet);
1406 
1407 	IGMPSTAT_INC(igps_rcv_total);
1408 
1409 	ip = mtod(m, struct ip *);
1410 	iphlen = off;
1411 	igmplen = ip->ip_len;
1412 
1413 	/*
1414 	 * Validate lengths.
1415 	 */
1416 	if (igmplen < IGMP_MINLEN) {
1417 		IGMPSTAT_INC(igps_rcv_tooshort);
1418 		m_freem(m);
1419 		return;
1420 	}
1421 
1422 	/*
1423 	 * Always pullup to the minimum size for v1/v2 or v3
1424 	 * to amortize calls to m_pullup().
1425 	 */
1426 	minlen = iphlen;
1427 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1428 		minlen += IGMP_V3_QUERY_MINLEN;
1429 	else
1430 		minlen += IGMP_MINLEN;
1431 	if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1432 	    (m = m_pullup(m, minlen)) == 0) {
1433 		IGMPSTAT_INC(igps_rcv_tooshort);
1434 		return;
1435 	}
1436 	ip = mtod(m, struct ip *);
1437 
1438 	if (ip->ip_ttl != 1) {
1439 		IGMPSTAT_INC(igps_rcv_badttl);
1440 		m_freem(m);
1441 		return;
1442 	}
1443 
1444 	/*
1445 	 * Validate checksum.
1446 	 */
1447 	m->m_data += iphlen;
1448 	m->m_len -= iphlen;
1449 	igmp = mtod(m, struct igmp *);
1450 	if (in_cksum(m, igmplen)) {
1451 		IGMPSTAT_INC(igps_rcv_badsum);
1452 		m_freem(m);
1453 		return;
1454 	}
1455 	m->m_data -= iphlen;
1456 	m->m_len += iphlen;
1457 
1458 	switch (igmp->igmp_type) {
1459 	case IGMP_HOST_MEMBERSHIP_QUERY:
1460 		if (igmplen == IGMP_MINLEN) {
1461 			if (igmp->igmp_code == 0)
1462 				queryver = IGMP_VERSION_1;
1463 			else
1464 				queryver = IGMP_VERSION_2;
1465 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1466 			queryver = IGMP_VERSION_3;
1467 		} else {
1468 			IGMPSTAT_INC(igps_rcv_tooshort);
1469 			m_freem(m);
1470 			return;
1471 		}
1472 
1473 		switch (queryver) {
1474 		case IGMP_VERSION_1:
1475 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1476 			if (!V_igmp_v1enable)
1477 				break;
1478 			if (igmp_input_v1_query(ifp, ip) != 0) {
1479 				m_freem(m);
1480 				return;
1481 			}
1482 			break;
1483 
1484 		case IGMP_VERSION_2:
1485 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1486 			if (!V_igmp_v2enable)
1487 				break;
1488 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1489 				m_freem(m);
1490 				return;
1491 			}
1492 			break;
1493 
1494 		case IGMP_VERSION_3: {
1495 				struct igmpv3 *igmpv3;
1496 				uint16_t igmpv3len;
1497 				uint16_t srclen;
1498 				int nsrc;
1499 
1500 				IGMPSTAT_INC(igps_rcv_v3_queries);
1501 				igmpv3 = (struct igmpv3 *)igmp;
1502 				/*
1503 				 * Validate length based on source count.
1504 				 */
1505 				nsrc = ntohs(igmpv3->igmp_numsrc);
1506 				srclen = sizeof(struct in_addr) * nsrc;
1507 				if (nsrc * sizeof(in_addr_t) > srclen) {
1508 					IGMPSTAT_INC(igps_rcv_tooshort);
1509 					return;
1510 				}
1511 				/*
1512 				 * m_pullup() may modify m, so pullup in
1513 				 * this scope.
1514 				 */
1515 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1516 				    srclen;
1517 				if ((m->m_flags & M_EXT ||
1518 				     m->m_len < igmpv3len) &&
1519 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1520 					IGMPSTAT_INC(igps_rcv_tooshort);
1521 					return;
1522 				}
1523 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1524 				    + iphlen);
1525 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1526 					m_freem(m);
1527 					return;
1528 				}
1529 			}
1530 			break;
1531 		}
1532 		break;
1533 
1534 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1535 		if (!V_igmp_v1enable)
1536 			break;
1537 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1538 			m_freem(m);
1539 			return;
1540 		}
1541 		break;
1542 
1543 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1544 		if (!V_igmp_v2enable)
1545 			break;
1546 		if (!ip_checkrouteralert(m))
1547 			IGMPSTAT_INC(igps_rcv_nora);
1548 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1549 			m_freem(m);
1550 			return;
1551 		}
1552 		break;
1553 
1554 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1555 		/*
1556 		 * Hosts do not need to process IGMPv3 membership reports,
1557 		 * as report suppression is no longer required.
1558 		 */
1559 		if (!ip_checkrouteralert(m))
1560 			IGMPSTAT_INC(igps_rcv_nora);
1561 		break;
1562 
1563 	default:
1564 		break;
1565 	}
1566 
1567 	/*
1568 	 * Pass all valid IGMP packets up to any process(es) listening on a
1569 	 * raw IGMP socket.
1570 	 */
1571 	rip_input(m, off);
1572 }
1573 
1574 
1575 /*
1576  * Fast timeout handler (global).
1577  * VIMAGE: Timeout handlers are expected to service all vimages.
1578  */
1579 void
1580 igmp_fasttimo(void)
1581 {
1582 #ifdef VIMAGE
1583 	VNET_ITERATOR_DECL(vnet_iter);
1584 
1585 	VNET_LIST_RLOCK();
1586 	VNET_FOREACH(vnet_iter) {
1587 		CURVNET_SET(vnet_iter);
1588 		INIT_VNET_INET(vnet_iter);
1589 		igmp_fasttimo_vnet();
1590 		CURVNET_RESTORE();
1591 	}
1592 	VNET_LIST_RUNLOCK();
1593 #else /* !VIMAGE */
1594 
1595 	igmp_fasttimo_vnet();
1596 #endif /* VIMAGE */
1597 }
1598 
1599 /*
1600  * Fast timeout handler (per-vnet).
1601  * Sends are shuffled off to a netisr to deal with Giant.
1602  *
1603  * VIMAGE: Assume caller has set up our curvnet.
1604  */
1605 static void
1606 igmp_fasttimo_vnet(void)
1607 {
1608 	struct ifqueue		 scq;	/* State-change packets */
1609 	struct ifqueue		 qrq;	/* Query response packets */
1610 	struct ifnet		*ifp;
1611 	struct igmp_ifinfo	*igi;
1612 	struct ifmultiaddr	*ifma, *tifma;
1613 	struct in_multi		*inm;
1614 	int			 loop, uri_fasthz;
1615 
1616 	loop = 0;
1617 	uri_fasthz = 0;
1618 
1619 	/*
1620 	 * Quick check to see if any work needs to be done, in order to
1621 	 * minimize the overhead of fasttimo processing.
1622 	 * SMPng: XXX Unlocked reads.
1623 	 */
1624 	if (!V_current_state_timers_running &&
1625 	    !V_interface_timers_running &&
1626 	    !V_state_change_timers_running)
1627 		return;
1628 
1629 	IN_MULTI_LOCK();
1630 	IGMP_LOCK();
1631 
1632 	/*
1633 	 * IGMPv3 General Query response timer processing.
1634 	 */
1635 	if (V_interface_timers_running) {
1636 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1637 
1638 		V_interface_timers_running = 0;
1639 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1640 			if (igi->igi_v3_timer == 0) {
1641 				/* Do nothing. */
1642 			} else if (--igi->igi_v3_timer == 0) {
1643 				igmp_v3_dispatch_general_query(igi);
1644 			} else {
1645 				V_interface_timers_running = 1;
1646 			}
1647 		}
1648 	}
1649 
1650 	if (!V_current_state_timers_running &&
1651 	    !V_state_change_timers_running)
1652 		goto out_locked;
1653 
1654 	V_current_state_timers_running = 0;
1655 	V_state_change_timers_running = 0;
1656 
1657 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1658 
1659 	/*
1660 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1661 	 * Note: Processing a v3 group timer may remove a node.
1662 	 */
1663 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1664 		ifp = igi->igi_ifp;
1665 
1666 		if (igi->igi_version == IGMP_VERSION_3) {
1667 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1668 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1669 			    PR_FASTHZ);
1670 
1671 			memset(&qrq, 0, sizeof(struct ifqueue));
1672 			IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1673 
1674 			memset(&scq, 0, sizeof(struct ifqueue));
1675 			IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1676 		}
1677 
1678 		IF_ADDR_LOCK(ifp);
1679 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1680 		    tifma) {
1681 			if (ifma->ifma_addr->sa_family != AF_INET ||
1682 			    ifma->ifma_protospec == NULL)
1683 				continue;
1684 			inm = (struct in_multi *)ifma->ifma_protospec;
1685 			switch (igi->igi_version) {
1686 			case IGMP_VERSION_1:
1687 			case IGMP_VERSION_2:
1688 				igmp_v1v2_process_group_timer(inm,
1689 				    igi->igi_version);
1690 				break;
1691 			case IGMP_VERSION_3:
1692 				igmp_v3_process_group_timers(igi, &qrq,
1693 				    &scq, inm, uri_fasthz);
1694 				break;
1695 			}
1696 		}
1697 		IF_ADDR_UNLOCK(ifp);
1698 
1699 		if (igi->igi_version == IGMP_VERSION_3) {
1700 			struct in_multi		*tinm;
1701 
1702 			igmp_dispatch_queue(&qrq, 0, loop);
1703 			igmp_dispatch_queue(&scq, 0, loop);
1704 
1705 			/*
1706 			 * Free the in_multi reference(s) for this
1707 			 * IGMP lifecycle.
1708 			 */
1709 			SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1710 			    inm_nrele, tinm) {
1711 				SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1712 				    inm_nrele);
1713 				inm_release_locked(inm);
1714 			}
1715 		}
1716 	}
1717 
1718 out_locked:
1719 	IGMP_UNLOCK();
1720 	IN_MULTI_UNLOCK();
1721 }
1722 
1723 /*
1724  * Update host report group timer for IGMPv1/v2.
1725  * Will update the global pending timer flags.
1726  */
1727 static void
1728 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1729 {
1730 	int report_timer_expired;
1731 
1732 	IN_MULTI_LOCK_ASSERT();
1733 	IGMP_LOCK_ASSERT();
1734 
1735 	if (inm->inm_timer == 0) {
1736 		report_timer_expired = 0;
1737 	} else if (--inm->inm_timer == 0) {
1738 		report_timer_expired = 1;
1739 	} else {
1740 		V_current_state_timers_running = 1;
1741 		return;
1742 	}
1743 
1744 	switch (inm->inm_state) {
1745 	case IGMP_NOT_MEMBER:
1746 	case IGMP_SILENT_MEMBER:
1747 	case IGMP_IDLE_MEMBER:
1748 	case IGMP_LAZY_MEMBER:
1749 	case IGMP_SLEEPING_MEMBER:
1750 	case IGMP_AWAKENING_MEMBER:
1751 		break;
1752 	case IGMP_REPORTING_MEMBER:
1753 		if (report_timer_expired) {
1754 			inm->inm_state = IGMP_IDLE_MEMBER;
1755 			(void)igmp_v1v2_queue_report(inm,
1756 			    (version == IGMP_VERSION_2) ?
1757 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1758 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1759 		}
1760 		break;
1761 	case IGMP_G_QUERY_PENDING_MEMBER:
1762 	case IGMP_SG_QUERY_PENDING_MEMBER:
1763 	case IGMP_LEAVING_MEMBER:
1764 		break;
1765 	}
1766 }
1767 
1768 /*
1769  * Update a group's timers for IGMPv3.
1770  * Will update the global pending timer flags.
1771  * Note: Unlocked read from igi.
1772  */
1773 static void
1774 igmp_v3_process_group_timers(struct igmp_ifinfo *igi,
1775     struct ifqueue *qrq, struct ifqueue *scq,
1776     struct in_multi *inm, const int uri_fasthz)
1777 {
1778 	int query_response_timer_expired;
1779 	int state_change_retransmit_timer_expired;
1780 
1781 	IN_MULTI_LOCK_ASSERT();
1782 	IGMP_LOCK_ASSERT();
1783 
1784 	query_response_timer_expired = 0;
1785 	state_change_retransmit_timer_expired = 0;
1786 
1787 	/*
1788 	 * During a transition from v1/v2 compatibility mode back to v3,
1789 	 * a group record in REPORTING state may still have its group
1790 	 * timer active. This is a no-op in this function; it is easier
1791 	 * to deal with it here than to complicate the slow-timeout path.
1792 	 */
1793 	if (inm->inm_timer == 0) {
1794 		query_response_timer_expired = 0;
1795 	} else if (--inm->inm_timer == 0) {
1796 		query_response_timer_expired = 1;
1797 	} else {
1798 		V_current_state_timers_running = 1;
1799 	}
1800 
1801 	if (inm->inm_sctimer == 0) {
1802 		state_change_retransmit_timer_expired = 0;
1803 	} else if (--inm->inm_sctimer == 0) {
1804 		state_change_retransmit_timer_expired = 1;
1805 	} else {
1806 		V_state_change_timers_running = 1;
1807 	}
1808 
1809 	/* We are in fasttimo, so be quick about it. */
1810 	if (!state_change_retransmit_timer_expired &&
1811 	    !query_response_timer_expired)
1812 		return;
1813 
1814 	switch (inm->inm_state) {
1815 	case IGMP_NOT_MEMBER:
1816 	case IGMP_SILENT_MEMBER:
1817 	case IGMP_SLEEPING_MEMBER:
1818 	case IGMP_LAZY_MEMBER:
1819 	case IGMP_AWAKENING_MEMBER:
1820 	case IGMP_IDLE_MEMBER:
1821 		break;
1822 	case IGMP_G_QUERY_PENDING_MEMBER:
1823 	case IGMP_SG_QUERY_PENDING_MEMBER:
1824 		/*
1825 		 * Respond to a previously pending Group-Specific
1826 		 * or Group-and-Source-Specific query by enqueueing
1827 		 * the appropriate Current-State report for
1828 		 * immediate transmission.
1829 		 */
1830 		if (query_response_timer_expired) {
1831 			int retval;
1832 
1833 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1834 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1835 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1836 			    __func__, retval);
1837 			inm->inm_state = IGMP_REPORTING_MEMBER;
1838 			/* XXX Clear recorded sources for next time. */
1839 			inm_clear_recorded(inm);
1840 		}
1841 		/* FALLTHROUGH */
1842 	case IGMP_REPORTING_MEMBER:
1843 	case IGMP_LEAVING_MEMBER:
1844 		if (state_change_retransmit_timer_expired) {
1845 			/*
1846 			 * State-change retransmission timer fired.
1847 			 * If there are any further pending retransmissions,
1848 			 * set the global pending state-change flag, and
1849 			 * reset the timer.
1850 			 */
1851 			if (--inm->inm_scrv > 0) {
1852 				inm->inm_sctimer = uri_fasthz;
1853 				V_state_change_timers_running = 1;
1854 			}
1855 			/*
1856 			 * Retransmit the previously computed state-change
1857 			 * report. If there are no further pending
1858 			 * retransmissions, the mbuf queue will be consumed.
1859 			 * Update T0 state to T1 as we have now sent
1860 			 * a state-change.
1861 			 */
1862 			(void)igmp_v3_merge_state_changes(inm, scq);
1863 
1864 			inm_commit(inm);
1865 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1866 			    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1867 
1868 			/*
1869 			 * If we are leaving the group for good, make sure
1870 			 * we release IGMP's reference to it.
1871 			 * This release must be deferred using a SLIST,
1872 			 * as we are called from a loop which traverses
1873 			 * the in_ifmultiaddr TAILQ.
1874 			 */
1875 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1876 			    inm->inm_scrv == 0) {
1877 				inm->inm_state = IGMP_NOT_MEMBER;
1878 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1879 				    inm, inm_nrele);
1880 			}
1881 		}
1882 		break;
1883 	}
1884 }
1885 
1886 
1887 /*
1888  * Suppress a group's pending response to a group or source/group query.
1889  *
1890  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1891  * Do NOT update ST1/ST0 as this operation merely suppresses
1892  * the currently pending group record.
1893  * Do NOT suppress the response to a general query. It is possible but
1894  * it would require adding another state or flag.
1895  */
1896 static void
1897 igmp_v3_suppress_group_record(struct in_multi *inm)
1898 {
1899 
1900 	IN_MULTI_LOCK_ASSERT();
1901 
1902 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1903 		("%s: not IGMPv3 mode on link", __func__));
1904 
1905 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1906 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1907 		return;
1908 
1909 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1910 		inm_clear_recorded(inm);
1911 
1912 	inm->inm_timer = 0;
1913 	inm->inm_state = IGMP_REPORTING_MEMBER;
1914 }
1915 
1916 /*
1917  * Switch to a different IGMP version on the given interface,
1918  * as per Section 7.2.1.
1919  */
1920 static void
1921 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1922 {
1923 
1924 	IGMP_LOCK_ASSERT();
1925 
1926 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1927 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1928 
1929 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1930 		int old_version_timer;
1931 		/*
1932 		 * Compute the "Older Version Querier Present" timer as per
1933 		 * Section 8.12.
1934 		 */
1935 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1936 		old_version_timer *= PR_SLOWHZ;
1937 
1938 		if (version == IGMP_VERSION_1) {
1939 			igi->igi_v1_timer = old_version_timer;
1940 			igi->igi_v2_timer = 0;
1941 		} else if (version == IGMP_VERSION_2) {
1942 			igi->igi_v1_timer = 0;
1943 			igi->igi_v2_timer = old_version_timer;
1944 		}
1945 	}
1946 
1947 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1948 		if (igi->igi_version != IGMP_VERSION_2) {
1949 			igi->igi_version = IGMP_VERSION_2;
1950 			igmp_v3_cancel_link_timers(igi);
1951 		}
1952 	} else if (igi->igi_v1_timer > 0) {
1953 		if (igi->igi_version != IGMP_VERSION_1) {
1954 			igi->igi_version = IGMP_VERSION_1;
1955 			igmp_v3_cancel_link_timers(igi);
1956 		}
1957 	}
1958 }
1959 
1960 /*
1961  * Cancel pending IGMPv3 timers for the given link and all groups
1962  * joined on it; state-change, general-query, and group-query timers.
1963  */
1964 static void
1965 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
1966 {
1967 	struct ifmultiaddr	*ifma;
1968 	struct ifnet		*ifp;
1969 	struct in_multi		*inm;
1970 
1971 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
1972 	    igi->igi_ifp, igi->igi_ifp->if_xname);
1973 
1974 	IN_MULTI_LOCK_ASSERT();
1975 	IGMP_LOCK_ASSERT();
1976 
1977 	/*
1978 	 * Fast-track this potentially expensive operation
1979 	 * by checking all the global 'timer pending' flags.
1980 	 */
1981 	if (!V_interface_timers_running &&
1982 	    !V_state_change_timers_running &&
1983 	    !V_current_state_timers_running)
1984 		return;
1985 
1986 	igi->igi_v3_timer = 0;
1987 
1988 	ifp = igi->igi_ifp;
1989 
1990 	IF_ADDR_LOCK(ifp);
1991 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1992 		if (ifma->ifma_addr->sa_family != AF_INET)
1993 			continue;
1994 		inm = (struct in_multi *)ifma->ifma_protospec;
1995 		switch (inm->inm_state) {
1996 		case IGMP_NOT_MEMBER:
1997 		case IGMP_SILENT_MEMBER:
1998 		case IGMP_IDLE_MEMBER:
1999 		case IGMP_LAZY_MEMBER:
2000 		case IGMP_SLEEPING_MEMBER:
2001 		case IGMP_AWAKENING_MEMBER:
2002 			break;
2003 		case IGMP_LEAVING_MEMBER:
2004 			/*
2005 			 * If we are leaving the group and switching
2006 			 * IGMP version, we need to release the final
2007 			 * reference held for issuing the INCLUDE {}.
2008 			 *
2009 			 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
2010 			 * around inm_release_locked(), as it is not
2011 			 * a recursive mutex.
2012 			 */
2013 			IF_ADDR_UNLOCK(ifp);
2014 			inm_release_locked(inm);
2015 			IF_ADDR_LOCK(ifp);
2016 			/* FALLTHROUGH */
2017 		case IGMP_G_QUERY_PENDING_MEMBER:
2018 		case IGMP_SG_QUERY_PENDING_MEMBER:
2019 			inm_clear_recorded(inm);
2020 			/* FALLTHROUGH */
2021 		case IGMP_REPORTING_MEMBER:
2022 			inm->inm_sctimer = 0;
2023 			inm->inm_timer = 0;
2024 			inm->inm_state = IGMP_REPORTING_MEMBER;
2025 			/*
2026 			 * Free any pending IGMPv3 state-change records.
2027 			 */
2028 			_IF_DRAIN(&inm->inm_scq);
2029 			break;
2030 		}
2031 	}
2032 	IF_ADDR_UNLOCK(ifp);
2033 }
2034 
2035 /*
2036  * Update the Older Version Querier Present timers for a link.
2037  * See Section 7.2.1 of RFC 3376.
2038  */
2039 static void
2040 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2041 {
2042 
2043 	IGMP_LOCK_ASSERT();
2044 
2045 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2046 		/*
2047 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2048 		 *
2049 		 * Revert to IGMPv3.
2050 		 */
2051 		if (igi->igi_version != IGMP_VERSION_3) {
2052 			CTR5(KTR_IGMPV3,
2053 			    "%s: transition from v%d -> v%d on %p(%s)",
2054 			    __func__, igi->igi_version, IGMP_VERSION_3,
2055 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2056 			igi->igi_version = IGMP_VERSION_3;
2057 		}
2058 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2059 		/*
2060 		 * IGMPv1 Querier Present timer expired,
2061 		 * IGMPv2 Querier Present timer running.
2062 		 * If IGMPv2 was disabled since last timeout,
2063 		 * revert to IGMPv3.
2064 		 * If IGMPv2 is enabled, revert to IGMPv2.
2065 		 */
2066 		if (!V_igmp_v2enable) {
2067 			CTR5(KTR_IGMPV3,
2068 			    "%s: transition from v%d -> v%d on %p(%s)",
2069 			    __func__, igi->igi_version, IGMP_VERSION_3,
2070 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2071 			igi->igi_v2_timer = 0;
2072 			igi->igi_version = IGMP_VERSION_3;
2073 		} else {
2074 			--igi->igi_v2_timer;
2075 			if (igi->igi_version != IGMP_VERSION_2) {
2076 				CTR5(KTR_IGMPV3,
2077 				    "%s: transition from v%d -> v%d on %p(%s)",
2078 				    __func__, igi->igi_version, IGMP_VERSION_2,
2079 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2080 				igi->igi_version = IGMP_VERSION_2;
2081 			}
2082 		}
2083 	} else if (igi->igi_v1_timer > 0) {
2084 		/*
2085 		 * IGMPv1 Querier Present timer running.
2086 		 * Stop IGMPv2 timer if running.
2087 		 *
2088 		 * If IGMPv1 was disabled since last timeout,
2089 		 * revert to IGMPv3.
2090 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2091 		 */
2092 		if (!V_igmp_v1enable) {
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_v1_timer = 0;
2098 			igi->igi_version = IGMP_VERSION_3;
2099 		} else {
2100 			--igi->igi_v1_timer;
2101 		}
2102 		if (igi->igi_v2_timer > 0) {
2103 			CTR3(KTR_IGMPV3,
2104 			    "%s: cancel v2 timer on %p(%s)",
2105 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2106 			igi->igi_v2_timer = 0;
2107 		}
2108 	}
2109 }
2110 
2111 /*
2112  * Global slowtimo handler.
2113  * VIMAGE: Timeout handlers are expected to service all vimages.
2114  */
2115 void
2116 igmp_slowtimo(void)
2117 {
2118 #ifdef VIMAGE
2119 	VNET_ITERATOR_DECL(vnet_iter);
2120 
2121 	VNET_LIST_RLOCK();
2122 	VNET_FOREACH(vnet_iter) {
2123 		CURVNET_SET(vnet_iter);
2124 		INIT_VNET_INET(vnet_iter);
2125 		igmp_slowtimo_vnet();
2126 		CURVNET_RESTORE();
2127 	}
2128 	VNET_LIST_RUNLOCK();
2129 #else /* !VIMAGE */
2130 	igmp_slowtimo_vnet();
2131 #endif /* VIMAGE */
2132 }
2133 
2134 /*
2135  * Per-vnet slowtimo handler.
2136  */
2137 static void
2138 igmp_slowtimo_vnet(void)
2139 {
2140 	struct igmp_ifinfo *igi;
2141 
2142 	IGMP_LOCK();
2143 
2144 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2145 		igmp_v1v2_process_querier_timers(igi);
2146 	}
2147 
2148 	IGMP_UNLOCK();
2149 }
2150 
2151 /*
2152  * Dispatch an IGMPv1/v2 host report or leave message.
2153  * These are always small enough to fit inside a single mbuf.
2154  */
2155 static int
2156 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2157 {
2158 	struct ifnet		*ifp;
2159 	struct igmp		*igmp;
2160 	struct ip		*ip;
2161 	struct mbuf		*m;
2162 
2163 	IN_MULTI_LOCK_ASSERT();
2164 	IGMP_LOCK_ASSERT();
2165 
2166 	ifp = inm->inm_ifp;
2167 	/* XXX are these needed ? */
2168 	INIT_VNET_NET(ifp->if_vnet);
2169 	INIT_VNET_INET(ifp->if_vnet);
2170 
2171 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2172 	if (m == NULL)
2173 		return (ENOMEM);
2174 	MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2175 
2176 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2177 
2178 	m->m_data += sizeof(struct ip);
2179 	m->m_len = sizeof(struct igmp);
2180 
2181 	igmp = mtod(m, struct igmp *);
2182 	igmp->igmp_type = type;
2183 	igmp->igmp_code = 0;
2184 	igmp->igmp_group = inm->inm_addr;
2185 	igmp->igmp_cksum = 0;
2186 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2187 
2188 	m->m_data -= sizeof(struct ip);
2189 	m->m_len += sizeof(struct ip);
2190 
2191 	ip = mtod(m, struct ip *);
2192 	ip->ip_tos = 0;
2193 	ip->ip_len = sizeof(struct ip) + sizeof(struct igmp);
2194 	ip->ip_off = 0;
2195 	ip->ip_p = IPPROTO_IGMP;
2196 	ip->ip_src.s_addr = INADDR_ANY;
2197 
2198 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2199 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2200 	else
2201 		ip->ip_dst = inm->inm_addr;
2202 
2203 	igmp_save_context(m, ifp);
2204 
2205 	m->m_flags |= M_IGMPV2;
2206 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2207 		m->m_flags |= M_IGMP_LOOP;
2208 
2209 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2210 	netisr_dispatch(NETISR_IGMP, m);
2211 
2212 	return (0);
2213 }
2214 
2215 /*
2216  * Process a state change from the upper layer for the given IPv4 group.
2217  *
2218  * Each socket holds a reference on the in_multi in its own ip_moptions.
2219  * The socket layer will have made the necessary updates to.the group
2220  * state, it is now up to IGMP to issue a state change report if there
2221  * has been any change between T0 (when the last state-change was issued)
2222  * and T1 (now).
2223  *
2224  * We use the IGMPv3 state machine at group level. The IGMP module
2225  * however makes the decision as to which IGMP protocol version to speak.
2226  * A state change *from* INCLUDE {} always means an initial join.
2227  * A state change *to* INCLUDE {} always means a final leave.
2228  *
2229  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2230  * save ourselves a bunch of work; any exclusive mode groups need not
2231  * compute source filter lists.
2232  *
2233  * VIMAGE: curvnet should have been set by caller, as this routine
2234  * is called from the socket option handlers.
2235  */
2236 int
2237 igmp_change_state(struct in_multi *inm)
2238 {
2239 	struct igmp_ifinfo *igi;
2240 	struct ifnet *ifp;
2241 	int error;
2242 
2243 	IN_MULTI_LOCK_ASSERT();
2244 
2245 	error = 0;
2246 
2247 	/*
2248 	 * Try to detect if the upper layer just asked us to change state
2249 	 * for an interface which has now gone away.
2250 	 */
2251 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2252 	ifp = inm->inm_ifma->ifma_ifp;
2253 	if (ifp != NULL) {
2254 		/*
2255 		 * Sanity check that netinet's notion of ifp is the
2256 		 * same as net's.
2257 		 */
2258 		KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2259 	}
2260 
2261 	IGMP_LOCK();
2262 
2263 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2264 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2265 
2266 	/*
2267 	 * If we detect a state transition to or from MCAST_UNDEFINED
2268 	 * for this group, then we are starting or finishing an IGMP
2269 	 * life cycle for this group.
2270 	 */
2271 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2272 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2273 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2274 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2275 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2276 			error = igmp_initial_join(inm, igi);
2277 			goto out_locked;
2278 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2279 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2280 			igmp_final_leave(inm, igi);
2281 			goto out_locked;
2282 		}
2283 	} else {
2284 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2285 	}
2286 
2287 	error = igmp_handle_state_change(inm, igi);
2288 
2289 out_locked:
2290 	IGMP_UNLOCK();
2291 	return (error);
2292 }
2293 
2294 /*
2295  * Perform the initial join for an IGMP group.
2296  *
2297  * When joining a group:
2298  *  If the group should have its IGMP traffic suppressed, do nothing.
2299  *  IGMPv1 starts sending IGMPv1 host membership reports.
2300  *  IGMPv2 starts sending IGMPv2 host membership reports.
2301  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2302  *  initial state of the membership.
2303  */
2304 static int
2305 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2306 {
2307 	struct ifnet		*ifp;
2308 	struct ifqueue		*ifq;
2309 	int			 error, retval, syncstates;
2310 
2311 	CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2312 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2313 	    inm->inm_ifp->if_xname);
2314 
2315 	error = 0;
2316 	syncstates = 1;
2317 
2318 	ifp = inm->inm_ifp;
2319 
2320 	IN_MULTI_LOCK_ASSERT();
2321 	IGMP_LOCK_ASSERT();
2322 
2323 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2324 
2325 	/*
2326 	 * Groups joined on loopback or marked as 'not reported',
2327 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2328 	 * are never reported in any IGMP protocol exchanges.
2329 	 * All other groups enter the appropriate IGMP state machine
2330 	 * for the version in use on this link.
2331 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2332 	 * disabled for the link.
2333 	 */
2334 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2335 	    (igi->igi_flags & IGIF_SILENT) ||
2336 	    !igmp_isgroupreported(inm->inm_addr)) {
2337 		CTR1(KTR_IGMPV3,
2338 "%s: not kicking state machine for silent group", __func__);
2339 		inm->inm_state = IGMP_SILENT_MEMBER;
2340 		inm->inm_timer = 0;
2341 	} else {
2342 		/*
2343 		 * Deal with overlapping in_multi lifecycle.
2344 		 * If this group was LEAVING, then make sure
2345 		 * we drop the reference we picked up to keep the
2346 		 * group around for the final INCLUDE {} enqueue.
2347 		 */
2348 		if (igi->igi_version == IGMP_VERSION_3 &&
2349 		    inm->inm_state == IGMP_LEAVING_MEMBER)
2350 			inm_release_locked(inm);
2351 
2352 		inm->inm_state = IGMP_REPORTING_MEMBER;
2353 
2354 		switch (igi->igi_version) {
2355 		case IGMP_VERSION_1:
2356 		case IGMP_VERSION_2:
2357 			inm->inm_state = IGMP_IDLE_MEMBER;
2358 			error = igmp_v1v2_queue_report(inm,
2359 			    (igi->igi_version == IGMP_VERSION_2) ?
2360 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2361 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2362 			if (error == 0) {
2363 				inm->inm_timer = IGMP_RANDOM_DELAY(
2364 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2365 				V_current_state_timers_running = 1;
2366 			}
2367 			break;
2368 
2369 		case IGMP_VERSION_3:
2370 			/*
2371 			 * Defer update of T0 to T1, until the first copy
2372 			 * of the state change has been transmitted.
2373 			 */
2374 			syncstates = 0;
2375 
2376 			/*
2377 			 * Immediately enqueue a State-Change Report for
2378 			 * this interface, freeing any previous reports.
2379 			 * Don't kick the timers if there is nothing to do,
2380 			 * or if an error occurred.
2381 			 */
2382 			ifq = &inm->inm_scq;
2383 			_IF_DRAIN(ifq);
2384 			retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2385 			    0, 0);
2386 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2387 			    __func__, retval);
2388 			if (retval <= 0) {
2389 				error = retval * -1;
2390 				break;
2391 			}
2392 
2393 			/*
2394 			 * Schedule transmission of pending state-change
2395 			 * report up to RV times for this link. The timer
2396 			 * will fire at the next igmp_fasttimo (~200ms),
2397 			 * giving us an opportunity to merge the reports.
2398 			 */
2399 			if (igi->igi_flags & IGIF_LOOPBACK) {
2400 				inm->inm_scrv = 1;
2401 			} else {
2402 				KASSERT(igi->igi_rv > 1,
2403 				   ("%s: invalid robustness %d", __func__,
2404 				    igi->igi_rv));
2405 				inm->inm_scrv = igi->igi_rv;
2406 			}
2407 			inm->inm_sctimer = 1;
2408 			V_state_change_timers_running = 1;
2409 
2410 			error = 0;
2411 			break;
2412 		}
2413 	}
2414 
2415 	/*
2416 	 * Only update the T0 state if state change is atomic,
2417 	 * i.e. we don't need to wait for a timer to fire before we
2418 	 * can consider the state change to have been communicated.
2419 	 */
2420 	if (syncstates) {
2421 		inm_commit(inm);
2422 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2423 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2424 	}
2425 
2426 	return (error);
2427 }
2428 
2429 /*
2430  * Issue an intermediate state change during the IGMP life-cycle.
2431  */
2432 static int
2433 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2434 {
2435 	struct ifnet		*ifp;
2436 	int			 retval;
2437 
2438 	CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2439 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2440 	    inm->inm_ifp->if_xname);
2441 
2442 	ifp = inm->inm_ifp;
2443 
2444 	IN_MULTI_LOCK_ASSERT();
2445 	IGMP_LOCK_ASSERT();
2446 
2447 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2448 
2449 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2450 	    (igi->igi_flags & IGIF_SILENT) ||
2451 	    !igmp_isgroupreported(inm->inm_addr) ||
2452 	    (igi->igi_version != IGMP_VERSION_3)) {
2453 		if (!igmp_isgroupreported(inm->inm_addr)) {
2454 			CTR1(KTR_IGMPV3,
2455 "%s: not kicking state machine for silent group", __func__);
2456 		}
2457 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2458 		inm_commit(inm);
2459 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2460 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2461 		return (0);
2462 	}
2463 
2464 	_IF_DRAIN(&inm->inm_scq);
2465 
2466 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2467 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2468 	if (retval <= 0)
2469 		return (-retval);
2470 
2471 	/*
2472 	 * If record(s) were enqueued, start the state-change
2473 	 * report timer for this group.
2474 	 */
2475 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2476 	inm->inm_sctimer = 1;
2477 	V_state_change_timers_running = 1;
2478 
2479 	return (0);
2480 }
2481 
2482 /*
2483  * Perform the final leave for an IGMP group.
2484  *
2485  * When leaving a group:
2486  *  IGMPv1 does nothing.
2487  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2488  *  IGMPv3 enqueues a state-change report containing a transition
2489  *  to INCLUDE {} for immediate transmission.
2490  */
2491 static void
2492 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2493 {
2494 	int syncstates;
2495 
2496 	syncstates = 1;
2497 
2498 	CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2499 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2500 	    inm->inm_ifp->if_xname);
2501 
2502 	IN_MULTI_LOCK_ASSERT();
2503 	IGMP_LOCK_ASSERT();
2504 
2505 	switch (inm->inm_state) {
2506 	case IGMP_NOT_MEMBER:
2507 	case IGMP_SILENT_MEMBER:
2508 	case IGMP_LEAVING_MEMBER:
2509 		/* Already leaving or left; do nothing. */
2510 		CTR1(KTR_IGMPV3,
2511 "%s: not kicking state machine for silent group", __func__);
2512 		break;
2513 	case IGMP_REPORTING_MEMBER:
2514 	case IGMP_IDLE_MEMBER:
2515 	case IGMP_G_QUERY_PENDING_MEMBER:
2516 	case IGMP_SG_QUERY_PENDING_MEMBER:
2517 		if (igi->igi_version == IGMP_VERSION_2) {
2518 #ifdef INVARIANTS
2519 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2520 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2521 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2522 			     __func__);
2523 #endif
2524 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2525 			inm->inm_state = IGMP_NOT_MEMBER;
2526 		} else if (igi->igi_version == IGMP_VERSION_3) {
2527 			/*
2528 			 * Stop group timer and all pending reports.
2529 			 * Immediately enqueue a state-change report
2530 			 * TO_IN {} to be sent on the next fast timeout,
2531 			 * giving us an opportunity to merge reports.
2532 			 */
2533 			_IF_DRAIN(&inm->inm_scq);
2534 			inm->inm_timer = 0;
2535 			if (igi->igi_flags & IGIF_LOOPBACK) {
2536 				inm->inm_scrv = 1;
2537 			} else {
2538 				inm->inm_scrv = igi->igi_rv;
2539 			}
2540 			CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2541 			    "pending retransmissions.", __func__,
2542 			    inet_ntoa(inm->inm_addr),
2543 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2544 			if (inm->inm_scrv == 0) {
2545 				inm->inm_state = IGMP_NOT_MEMBER;
2546 				inm->inm_sctimer = 0;
2547 			} else {
2548 				int retval;
2549 
2550 				inm_acquire_locked(inm);
2551 
2552 				retval = igmp_v3_enqueue_group_record(
2553 				    &inm->inm_scq, inm, 1, 0, 0);
2554 				KASSERT(retval != 0,
2555 				    ("%s: enqueue record = %d", __func__,
2556 				     retval));
2557 
2558 				inm->inm_state = IGMP_LEAVING_MEMBER;
2559 				inm->inm_sctimer = 1;
2560 				V_state_change_timers_running = 1;
2561 				syncstates = 0;
2562 			}
2563 			break;
2564 		}
2565 		break;
2566 	case IGMP_LAZY_MEMBER:
2567 	case IGMP_SLEEPING_MEMBER:
2568 	case IGMP_AWAKENING_MEMBER:
2569 		/* Our reports are suppressed; do nothing. */
2570 		break;
2571 	}
2572 
2573 	if (syncstates) {
2574 		inm_commit(inm);
2575 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2576 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2577 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2578 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2579 		    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2580 	}
2581 }
2582 
2583 /*
2584  * Enqueue an IGMPv3 group record to the given output queue.
2585  *
2586  * XXX This function could do with having the allocation code
2587  * split out, and the multiple-tree-walks coalesced into a single
2588  * routine as has been done in igmp_v3_enqueue_filter_change().
2589  *
2590  * If is_state_change is zero, a current-state record is appended.
2591  * If is_state_change is non-zero, a state-change report is appended.
2592  *
2593  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2594  * If is_group_query is zero, and if there is a packet with free space
2595  * at the tail of the queue, it will be appended to providing there
2596  * is enough free space.
2597  * Otherwise a new mbuf packet chain is allocated.
2598  *
2599  * If is_source_query is non-zero, each source is checked to see if
2600  * it was recorded for a Group-Source query, and will be omitted if
2601  * it is not both in-mode and recorded.
2602  *
2603  * The function will attempt to allocate leading space in the packet
2604  * for the IP/IGMP header to be prepended without fragmenting the chain.
2605  *
2606  * If successful the size of all data appended to the queue is returned,
2607  * otherwise an error code less than zero is returned, or zero if
2608  * no record(s) were appended.
2609  */
2610 static int
2611 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2612     const int is_state_change, const int is_group_query,
2613     const int is_source_query)
2614 {
2615 	struct igmp_grouprec	 ig;
2616 	struct igmp_grouprec	*pig;
2617 	struct ifnet		*ifp;
2618 	struct ip_msource	*ims, *nims;
2619 	struct mbuf		*m0, *m, *md;
2620 	int			 error, is_filter_list_change;
2621 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2622 	int			 record_has_sources;
2623 	int			 now;
2624 	int			 type;
2625 	in_addr_t		 naddr;
2626 	uint8_t			 mode;
2627 
2628 	IN_MULTI_LOCK_ASSERT();
2629 
2630 	error = 0;
2631 	ifp = inm->inm_ifp;
2632 	is_filter_list_change = 0;
2633 	m = NULL;
2634 	m0 = NULL;
2635 	m0srcs = 0;
2636 	msrcs = 0;
2637 	nbytes = 0;
2638 	nims = NULL;
2639 	record_has_sources = 1;
2640 	pig = NULL;
2641 	type = IGMP_DO_NOTHING;
2642 	mode = inm->inm_st[1].iss_fmode;
2643 
2644 	/*
2645 	 * If we did not transition out of ASM mode during t0->t1,
2646 	 * and there are no source nodes to process, we can skip
2647 	 * the generation of source records.
2648 	 */
2649 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2650 	    inm->inm_nsrc == 0)
2651 		record_has_sources = 0;
2652 
2653 	if (is_state_change) {
2654 		/*
2655 		 * Queue a state change record.
2656 		 * If the mode did not change, and there are non-ASM
2657 		 * listeners or source filters present,
2658 		 * we potentially need to issue two records for the group.
2659 		 * If we are transitioning to MCAST_UNDEFINED, we need
2660 		 * not send any sources.
2661 		 * If there are ASM listeners, and there was no filter
2662 		 * mode transition of any kind, do nothing.
2663 		 */
2664 		if (mode != inm->inm_st[0].iss_fmode) {
2665 			if (mode == MCAST_EXCLUDE) {
2666 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2667 				    __func__);
2668 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2669 			} else {
2670 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2671 				    __func__);
2672 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2673 				if (mode == MCAST_UNDEFINED)
2674 					record_has_sources = 0;
2675 			}
2676 		} else {
2677 			if (record_has_sources) {
2678 				is_filter_list_change = 1;
2679 			} else {
2680 				type = IGMP_DO_NOTHING;
2681 			}
2682 		}
2683 	} else {
2684 		/*
2685 		 * Queue a current state record.
2686 		 */
2687 		if (mode == MCAST_EXCLUDE) {
2688 			type = IGMP_MODE_IS_EXCLUDE;
2689 		} else if (mode == MCAST_INCLUDE) {
2690 			type = IGMP_MODE_IS_INCLUDE;
2691 			KASSERT(inm->inm_st[1].iss_asm == 0,
2692 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2693 			     __func__, inm, inm->inm_st[1].iss_asm));
2694 		}
2695 	}
2696 
2697 	/*
2698 	 * Generate the filter list changes using a separate function.
2699 	 */
2700 	if (is_filter_list_change)
2701 		return (igmp_v3_enqueue_filter_change(ifq, inm));
2702 
2703 	if (type == IGMP_DO_NOTHING) {
2704 		CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2705 		    __func__, inet_ntoa(inm->inm_addr),
2706 		    inm->inm_ifp->if_xname);
2707 		return (0);
2708 	}
2709 
2710 	/*
2711 	 * If any sources are present, we must be able to fit at least
2712 	 * one in the trailing space of the tail packet's mbuf,
2713 	 * ideally more.
2714 	 */
2715 	minrec0len = sizeof(struct igmp_grouprec);
2716 	if (record_has_sources)
2717 		minrec0len += sizeof(in_addr_t);
2718 
2719 	CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2720 	    igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2721 	    inm->inm_ifp->if_xname);
2722 
2723 	/*
2724 	 * Check if we have a packet in the tail of the queue for this
2725 	 * group into which the first group record for this group will fit.
2726 	 * Otherwise allocate a new packet.
2727 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2728 	 * Note: Group records for G/GSR query responses MUST be sent
2729 	 * in their own packet.
2730 	 */
2731 	m0 = ifq->ifq_tail;
2732 	if (!is_group_query &&
2733 	    m0 != NULL &&
2734 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2735 	    (m0->m_pkthdr.len + minrec0len) <
2736 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2737 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2738 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2739 		m = m0;
2740 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2741 	} else {
2742 		if (_IF_QFULL(ifq)) {
2743 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2744 			return (-ENOMEM);
2745 		}
2746 		m = NULL;
2747 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2748 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2749 		if (!is_state_change && !is_group_query) {
2750 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2751 			if (m)
2752 				m->m_data += IGMP_LEADINGSPACE;
2753 		}
2754 		if (m == NULL) {
2755 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2756 			if (m)
2757 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2758 		}
2759 		if (m == NULL)
2760 			return (-ENOMEM);
2761 
2762 		igmp_save_context(m, ifp);
2763 
2764 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2765 	}
2766 
2767 	/*
2768 	 * Append group record.
2769 	 * If we have sources, we don't know how many yet.
2770 	 */
2771 	ig.ig_type = type;
2772 	ig.ig_datalen = 0;
2773 	ig.ig_numsrc = 0;
2774 	ig.ig_group = inm->inm_addr;
2775 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2776 		if (m != m0)
2777 			m_freem(m);
2778 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2779 		return (-ENOMEM);
2780 	}
2781 	nbytes += sizeof(struct igmp_grouprec);
2782 
2783 	/*
2784 	 * Append as many sources as will fit in the first packet.
2785 	 * If we are appending to a new packet, the chain allocation
2786 	 * may potentially use clusters; use m_getptr() in this case.
2787 	 * If we are appending to an existing packet, we need to obtain
2788 	 * a pointer to the group record after m_append(), in case a new
2789 	 * mbuf was allocated.
2790 	 * Only append sources which are in-mode at t1. If we are
2791 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2792 	 * include source entries.
2793 	 * Only report recorded sources in our filter set when responding
2794 	 * to a group-source query.
2795 	 */
2796 	if (record_has_sources) {
2797 		if (m == m0) {
2798 			md = m_last(m);
2799 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2800 			    md->m_len - nbytes);
2801 		} else {
2802 			md = m_getptr(m, 0, &off);
2803 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2804 			    off);
2805 		}
2806 		msrcs = 0;
2807 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2808 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2809 			    inet_ntoa_haddr(ims->ims_haddr));
2810 			now = ims_get_mode(inm, ims, 1);
2811 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2812 			if ((now != mode) ||
2813 			    (now == mode && mode == MCAST_UNDEFINED)) {
2814 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2815 				continue;
2816 			}
2817 			if (is_source_query && ims->ims_stp == 0) {
2818 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2819 				    __func__);
2820 				continue;
2821 			}
2822 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2823 			naddr = htonl(ims->ims_haddr);
2824 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2825 				if (m != m0)
2826 					m_freem(m);
2827 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2828 				    __func__);
2829 				return (-ENOMEM);
2830 			}
2831 			nbytes += sizeof(in_addr_t);
2832 			++msrcs;
2833 			if (msrcs == m0srcs)
2834 				break;
2835 		}
2836 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2837 		    msrcs);
2838 		pig->ig_numsrc = htons(msrcs);
2839 		nbytes += (msrcs * sizeof(in_addr_t));
2840 	}
2841 
2842 	if (is_source_query && msrcs == 0) {
2843 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2844 		if (m != m0)
2845 			m_freem(m);
2846 		return (0);
2847 	}
2848 
2849 	/*
2850 	 * We are good to go with first packet.
2851 	 */
2852 	if (m != m0) {
2853 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2854 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2855 		_IF_ENQUEUE(ifq, m);
2856 	} else
2857 		m->m_pkthdr.PH_vt.vt_nrecs++;
2858 
2859 	/*
2860 	 * No further work needed if no source list in packet(s).
2861 	 */
2862 	if (!record_has_sources)
2863 		return (nbytes);
2864 
2865 	/*
2866 	 * Whilst sources remain to be announced, we need to allocate
2867 	 * a new packet and fill out as many sources as will fit.
2868 	 * Always try for a cluster first.
2869 	 */
2870 	while (nims != NULL) {
2871 		if (_IF_QFULL(ifq)) {
2872 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2873 			return (-ENOMEM);
2874 		}
2875 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2876 		if (m)
2877 			m->m_data += IGMP_LEADINGSPACE;
2878 		if (m == NULL) {
2879 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2880 			if (m)
2881 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2882 		}
2883 		if (m == NULL)
2884 			return (-ENOMEM);
2885 		igmp_save_context(m, ifp);
2886 		md = m_getptr(m, 0, &off);
2887 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2888 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2889 
2890 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2891 			if (m != m0)
2892 				m_freem(m);
2893 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2894 			return (-ENOMEM);
2895 		}
2896 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2897 		nbytes += sizeof(struct igmp_grouprec);
2898 
2899 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2900 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2901 
2902 		msrcs = 0;
2903 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2904 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2905 			    inet_ntoa_haddr(ims->ims_haddr));
2906 			now = ims_get_mode(inm, ims, 1);
2907 			if ((now != mode) ||
2908 			    (now == mode && mode == MCAST_UNDEFINED)) {
2909 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2910 				continue;
2911 			}
2912 			if (is_source_query && ims->ims_stp == 0) {
2913 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2914 				    __func__);
2915 				continue;
2916 			}
2917 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2918 			naddr = htonl(ims->ims_haddr);
2919 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2920 				if (m != m0)
2921 					m_freem(m);
2922 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2923 				    __func__);
2924 				return (-ENOMEM);
2925 			}
2926 			++msrcs;
2927 			if (msrcs == m0srcs)
2928 				break;
2929 		}
2930 		pig->ig_numsrc = htons(msrcs);
2931 		nbytes += (msrcs * sizeof(in_addr_t));
2932 
2933 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2934 		_IF_ENQUEUE(ifq, m);
2935 	}
2936 
2937 	return (nbytes);
2938 }
2939 
2940 /*
2941  * Type used to mark record pass completion.
2942  * We exploit the fact we can cast to this easily from the
2943  * current filter modes on each ip_msource node.
2944  */
2945 typedef enum {
2946 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2947 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2948 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2949 	REC_FULL = REC_ALLOW | REC_BLOCK
2950 } rectype_t;
2951 
2952 /*
2953  * Enqueue an IGMPv3 filter list change to the given output queue.
2954  *
2955  * Source list filter state is held in an RB-tree. When the filter list
2956  * for a group is changed without changing its mode, we need to compute
2957  * the deltas between T0 and T1 for each source in the filter set,
2958  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2959  *
2960  * As we may potentially queue two record types, and the entire R-B tree
2961  * needs to be walked at once, we break this out into its own function
2962  * so we can generate a tightly packed queue of packets.
2963  *
2964  * XXX This could be written to only use one tree walk, although that makes
2965  * serializing into the mbuf chains a bit harder. For now we do two walks
2966  * which makes things easier on us, and it may or may not be harder on
2967  * the L2 cache.
2968  *
2969  * If successful the size of all data appended to the queue is returned,
2970  * otherwise an error code less than zero is returned, or zero if
2971  * no record(s) were appended.
2972  */
2973 static int
2974 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
2975 {
2976 	static const int MINRECLEN =
2977 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
2978 	struct ifnet		*ifp;
2979 	struct igmp_grouprec	 ig;
2980 	struct igmp_grouprec	*pig;
2981 	struct ip_msource	*ims, *nims;
2982 	struct mbuf		*m, *m0, *md;
2983 	in_addr_t		 naddr;
2984 	int			 m0srcs, nbytes, off, rsrcs, schanged;
2985 	int			 nallow, nblock;
2986 	uint8_t			 mode, now, then;
2987 	rectype_t		 crt, drt, nrt;
2988 
2989 	IN_MULTI_LOCK_ASSERT();
2990 
2991 	if (inm->inm_nsrc == 0 ||
2992 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
2993 		return (0);
2994 
2995 	ifp = inm->inm_ifp;			/* interface */
2996 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
2997 	crt = REC_NONE;	/* current group record type */
2998 	drt = REC_NONE;	/* mask of completed group record types */
2999 	nrt = REC_NONE;	/* record type for current node */
3000 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3001 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3002 	rsrcs = 0;	/* # sources encoded in current record */
3003 	schanged = 0;	/* # nodes encoded in overall filter change */
3004 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3005 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3006 	nims = NULL;	/* next tree node pointer */
3007 
3008 	/*
3009 	 * For each possible filter record mode.
3010 	 * The first kind of source we encounter tells us which
3011 	 * is the first kind of record we start appending.
3012 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3013 	 * as the inverse of the group's filter mode.
3014 	 */
3015 	while (drt != REC_FULL) {
3016 		do {
3017 			m0 = ifq->ifq_tail;
3018 			if (m0 != NULL &&
3019 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3020 			     IGMP_V3_REPORT_MAXRECS) &&
3021 			    (m0->m_pkthdr.len + MINRECLEN) <
3022 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3023 				m = m0;
3024 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3025 					    sizeof(struct igmp_grouprec)) /
3026 				    sizeof(in_addr_t);
3027 				CTR1(KTR_IGMPV3,
3028 				    "%s: use previous packet", __func__);
3029 			} else {
3030 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3031 				if (m)
3032 					m->m_data += IGMP_LEADINGSPACE;
3033 				if (m == NULL) {
3034 					m = m_gethdr(M_DONTWAIT, MT_DATA);
3035 					if (m)
3036 						MH_ALIGN(m, IGMP_LEADINGSPACE);
3037 				}
3038 				if (m == NULL) {
3039 					CTR1(KTR_IGMPV3,
3040 					    "%s: m_get*() failed", __func__);
3041 					return (-ENOMEM);
3042 				}
3043 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3044 				igmp_save_context(m, ifp);
3045 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3046 				    sizeof(struct igmp_grouprec)) /
3047 				    sizeof(in_addr_t);
3048 				CTR1(KTR_IGMPV3,
3049 				    "%s: allocated new packet", __func__);
3050 			}
3051 			/*
3052 			 * Append the IGMP group record header to the
3053 			 * current packet's data area.
3054 			 * Recalculate pointer to free space for next
3055 			 * group record, in case m_append() allocated
3056 			 * a new mbuf or cluster.
3057 			 */
3058 			memset(&ig, 0, sizeof(ig));
3059 			ig.ig_group = inm->inm_addr;
3060 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3061 				if (m != m0)
3062 					m_freem(m);
3063 				CTR1(KTR_IGMPV3,
3064 				    "%s: m_append() failed", __func__);
3065 				return (-ENOMEM);
3066 			}
3067 			nbytes += sizeof(struct igmp_grouprec);
3068 			if (m == m0) {
3069 				md = m_last(m);
3070 				pig = (struct igmp_grouprec *)(mtod(md,
3071 				    uint8_t *) + md->m_len - nbytes);
3072 			} else {
3073 				md = m_getptr(m, 0, &off);
3074 				pig = (struct igmp_grouprec *)(mtod(md,
3075 				    uint8_t *) + off);
3076 			}
3077 			/*
3078 			 * Begin walking the tree for this record type
3079 			 * pass, or continue from where we left off
3080 			 * previously if we had to allocate a new packet.
3081 			 * Only report deltas in-mode at t1.
3082 			 * We need not report included sources as allowed
3083 			 * if we are in inclusive mode on the group,
3084 			 * however the converse is not true.
3085 			 */
3086 			rsrcs = 0;
3087 			if (nims == NULL)
3088 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3089 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3090 				CTR2(KTR_IGMPV3, "%s: visit node %s",
3091 				    __func__, inet_ntoa_haddr(ims->ims_haddr));
3092 				now = ims_get_mode(inm, ims, 1);
3093 				then = ims_get_mode(inm, ims, 0);
3094 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3095 				    __func__, then, now);
3096 				if (now == then) {
3097 					CTR1(KTR_IGMPV3,
3098 					    "%s: skip unchanged", __func__);
3099 					continue;
3100 				}
3101 				if (mode == MCAST_EXCLUDE &&
3102 				    now == MCAST_INCLUDE) {
3103 					CTR1(KTR_IGMPV3,
3104 					    "%s: skip IN src on EX group",
3105 					    __func__);
3106 					continue;
3107 				}
3108 				nrt = (rectype_t)now;
3109 				if (nrt == REC_NONE)
3110 					nrt = (rectype_t)(~mode & REC_FULL);
3111 				if (schanged++ == 0) {
3112 					crt = nrt;
3113 				} else if (crt != nrt)
3114 					continue;
3115 				naddr = htonl(ims->ims_haddr);
3116 				if (!m_append(m, sizeof(in_addr_t),
3117 				    (void *)&naddr)) {
3118 					if (m != m0)
3119 						m_freem(m);
3120 					CTR1(KTR_IGMPV3,
3121 					    "%s: m_append() failed", __func__);
3122 					return (-ENOMEM);
3123 				}
3124 				nallow += !!(crt == REC_ALLOW);
3125 				nblock += !!(crt == REC_BLOCK);
3126 				if (++rsrcs == m0srcs)
3127 					break;
3128 			}
3129 			/*
3130 			 * If we did not append any tree nodes on this
3131 			 * pass, back out of allocations.
3132 			 */
3133 			if (rsrcs == 0) {
3134 				nbytes -= sizeof(struct igmp_grouprec);
3135 				if (m != m0) {
3136 					CTR1(KTR_IGMPV3,
3137 					    "%s: m_free(m)", __func__);
3138 					m_freem(m);
3139 				} else {
3140 					CTR1(KTR_IGMPV3,
3141 					    "%s: m_adj(m, -ig)", __func__);
3142 					m_adj(m, -((int)sizeof(
3143 					    struct igmp_grouprec)));
3144 				}
3145 				continue;
3146 			}
3147 			nbytes += (rsrcs * sizeof(in_addr_t));
3148 			if (crt == REC_ALLOW)
3149 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3150 			else if (crt == REC_BLOCK)
3151 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3152 			pig->ig_numsrc = htons(rsrcs);
3153 			/*
3154 			 * Count the new group record, and enqueue this
3155 			 * packet if it wasn't already queued.
3156 			 */
3157 			m->m_pkthdr.PH_vt.vt_nrecs++;
3158 			if (m != m0)
3159 				_IF_ENQUEUE(ifq, m);
3160 		} while (nims != NULL);
3161 		drt |= crt;
3162 		crt = (~crt & REC_FULL);
3163 	}
3164 
3165 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3166 	    nallow, nblock);
3167 
3168 	return (nbytes);
3169 }
3170 
3171 static int
3172 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3173 {
3174 	struct ifqueue	*gq;
3175 	struct mbuf	*m;		/* pending state-change */
3176 	struct mbuf	*m0;		/* copy of pending state-change */
3177 	struct mbuf	*mt;		/* last state-change in packet */
3178 	int		 docopy, domerge;
3179 	u_int		 recslen;
3180 
3181 	docopy = 0;
3182 	domerge = 0;
3183 	recslen = 0;
3184 
3185 	IN_MULTI_LOCK_ASSERT();
3186 	IGMP_LOCK_ASSERT();
3187 
3188 	/*
3189 	 * If there are further pending retransmissions, make a writable
3190 	 * copy of each queued state-change message before merging.
3191 	 */
3192 	if (inm->inm_scrv > 0)
3193 		docopy = 1;
3194 
3195 	gq = &inm->inm_scq;
3196 #ifdef KTR
3197 	if (gq->ifq_head == NULL) {
3198 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3199 		    __func__, inm);
3200 	}
3201 #endif
3202 
3203 	m = gq->ifq_head;
3204 	while (m != NULL) {
3205 		/*
3206 		 * Only merge the report into the current packet if
3207 		 * there is sufficient space to do so; an IGMPv3 report
3208 		 * packet may only contain 65,535 group records.
3209 		 * Always use a simple mbuf chain concatentation to do this,
3210 		 * as large state changes for single groups may have
3211 		 * allocated clusters.
3212 		 */
3213 		domerge = 0;
3214 		mt = ifscq->ifq_tail;
3215 		if (mt != NULL) {
3216 			recslen = m_length(m, NULL);
3217 
3218 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3219 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3220 			    IGMP_V3_REPORT_MAXRECS) &&
3221 			    (mt->m_pkthdr.len + recslen <=
3222 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3223 				domerge = 1;
3224 		}
3225 
3226 		if (!domerge && _IF_QFULL(gq)) {
3227 			CTR2(KTR_IGMPV3,
3228 			    "%s: outbound queue full, skipping whole packet %p",
3229 			    __func__, m);
3230 			mt = m->m_nextpkt;
3231 			if (!docopy)
3232 				m_freem(m);
3233 			m = mt;
3234 			continue;
3235 		}
3236 
3237 		if (!docopy) {
3238 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3239 			_IF_DEQUEUE(gq, m0);
3240 			m = m0->m_nextpkt;
3241 		} else {
3242 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3243 			m0 = m_dup(m, M_NOWAIT);
3244 			if (m0 == NULL)
3245 				return (ENOMEM);
3246 			m0->m_nextpkt = NULL;
3247 			m = m->m_nextpkt;
3248 		}
3249 
3250 		if (!domerge) {
3251 			CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3252 			    __func__, m0, ifscq);
3253 			_IF_ENQUEUE(ifscq, m0);
3254 		} else {
3255 			struct mbuf *mtl;	/* last mbuf of packet mt */
3256 
3257 			CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3258 			    __func__, m0, mt);
3259 
3260 			mtl = m_last(mt);
3261 			m0->m_flags &= ~M_PKTHDR;
3262 			mt->m_pkthdr.len += recslen;
3263 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3264 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3265 
3266 			mtl->m_next = m0;
3267 		}
3268 	}
3269 
3270 	return (0);
3271 }
3272 
3273 /*
3274  * Respond to a pending IGMPv3 General Query.
3275  */
3276 static void
3277 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3278 {
3279 	struct ifmultiaddr	*ifma, *tifma;
3280 	struct ifnet		*ifp;
3281 	struct in_multi		*inm;
3282 	int			 retval, loop;
3283 
3284 	IN_MULTI_LOCK_ASSERT();
3285 	IGMP_LOCK_ASSERT();
3286 
3287 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3288 	    ("%s: called when version %d", __func__, igi->igi_version));
3289 
3290 	ifp = igi->igi_ifp;
3291 
3292 	IF_ADDR_LOCK(ifp);
3293 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
3294 		if (ifma->ifma_addr->sa_family != AF_INET ||
3295 		    ifma->ifma_protospec == NULL)
3296 			continue;
3297 
3298 		inm = (struct in_multi *)ifma->ifma_protospec;
3299 		KASSERT(ifp == inm->inm_ifp,
3300 		    ("%s: inconsistent ifp", __func__));
3301 
3302 		switch (inm->inm_state) {
3303 		case IGMP_NOT_MEMBER:
3304 		case IGMP_SILENT_MEMBER:
3305 			break;
3306 		case IGMP_REPORTING_MEMBER:
3307 		case IGMP_IDLE_MEMBER:
3308 		case IGMP_LAZY_MEMBER:
3309 		case IGMP_SLEEPING_MEMBER:
3310 		case IGMP_AWAKENING_MEMBER:
3311 			inm->inm_state = IGMP_REPORTING_MEMBER;
3312 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3313 			    inm, 0, 0, 0);
3314 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3315 			    __func__, retval);
3316 			break;
3317 		case IGMP_G_QUERY_PENDING_MEMBER:
3318 		case IGMP_SG_QUERY_PENDING_MEMBER:
3319 		case IGMP_LEAVING_MEMBER:
3320 			break;
3321 		}
3322 	}
3323 	IF_ADDR_UNLOCK(ifp);
3324 
3325 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3326 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3327 
3328 	/*
3329 	 * Slew transmission of bursts over 500ms intervals.
3330 	 */
3331 	if (igi->igi_gq.ifq_head != NULL) {
3332 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3333 		    IGMP_RESPONSE_BURST_INTERVAL);
3334 		V_interface_timers_running = 1;
3335 	}
3336 }
3337 
3338 /*
3339  * Transmit the next pending IGMP message in the output queue.
3340  *
3341  * We get called from netisr_processqueue(). A mutex private to igmpoq
3342  * will be acquired and released around this routine.
3343  *
3344  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3345  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3346  * a link and uses a link-scope multicast address.
3347  */
3348 static void
3349 igmp_intr(struct mbuf *m)
3350 {
3351 	struct ip_moptions	 imo;
3352 	struct ifnet		*ifp;
3353 	struct mbuf		*ipopts, *m0;
3354 	int			 error;
3355 	uint32_t		 ifindex;
3356 
3357 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3358 
3359 	/*
3360 	 * Restore VNET image pointer from enqueued mbuf chain
3361 	 * before doing anything else. Whilst we use interface
3362 	 * indexes to guard against interface detach, they are
3363 	 * unique to each VIMAGE and must be retrieved.
3364 	 */
3365 	CURVNET_SET(m->m_pkthdr.header);
3366 	ifindex = igmp_restore_context(m);
3367 
3368 	/*
3369 	 * Check if the ifnet still exists. This limits the scope of
3370 	 * any race in the absence of a global ifp lock for low cost
3371 	 * (an array lookup).
3372 	 */
3373 	ifp = ifnet_byindex(ifindex);
3374 	if (ifp == NULL) {
3375 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3376 		    __func__, m, ifindex);
3377 		m_freem(m);
3378 		IPSTAT_INC(ips_noroute);
3379 		goto out;
3380 	}
3381 
3382 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3383 
3384 	imo.imo_multicast_ttl  = 1;
3385 	imo.imo_multicast_vif  = -1;
3386 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3387 
3388 	/*
3389 	 * If the user requested that IGMP traffic be explicitly
3390 	 * redirected to the loopback interface (e.g. they are running a
3391 	 * MANET interface and the routing protocol needs to see the
3392 	 * updates), handle this now.
3393 	 */
3394 	if (m->m_flags & M_IGMP_LOOP)
3395 		imo.imo_multicast_ifp = V_loif;
3396 	else
3397 		imo.imo_multicast_ifp = ifp;
3398 
3399 	if (m->m_flags & M_IGMPV2) {
3400 		m0 = m;
3401 	} else {
3402 		m0 = igmp_v3_encap_report(ifp, m);
3403 		if (m0 == NULL) {
3404 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3405 			m_freem(m);
3406 			IPSTAT_INC(ips_odropped);
3407 			goto out;
3408 		}
3409 	}
3410 
3411 	igmp_scrub_context(m0);
3412 	m->m_flags &= ~(M_PROTOFLAGS);
3413 	m0->m_pkthdr.rcvif = V_loif;
3414 #ifdef MAC
3415 	mac_netinet_igmp_send(ifp, m0);
3416 #endif
3417 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3418 	if (error) {
3419 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3420 		goto out;
3421 	}
3422 
3423 	IGMPSTAT_INC(igps_snd_reports);
3424 
3425 out:
3426 	/*
3427 	 * We must restore the existing vnet pointer before
3428 	 * continuing as we are run from netisr context.
3429 	 */
3430 	CURVNET_RESTORE();
3431 }
3432 
3433 /*
3434  * Encapsulate an IGMPv3 report.
3435  *
3436  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3437  * chain has already had its IP/IGMPv3 header prepended. In this case
3438  * the function will not attempt to prepend; the lengths and checksums
3439  * will however be re-computed.
3440  *
3441  * Returns a pointer to the new mbuf chain head, or NULL if the
3442  * allocation failed.
3443  */
3444 static struct mbuf *
3445 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3446 {
3447 	INIT_VNET_NET(curvnet);
3448 	INIT_VNET_INET(curvnet);
3449 	struct igmp_report	*igmp;
3450 	struct ip		*ip;
3451 	int			 hdrlen, igmpreclen;
3452 
3453 	KASSERT((m->m_flags & M_PKTHDR),
3454 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3455 
3456 	igmpreclen = m_length(m, NULL);
3457 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3458 
3459 	if (m->m_flags & M_IGMPV3_HDR) {
3460 		igmpreclen -= hdrlen;
3461 	} else {
3462 		M_PREPEND(m, hdrlen, M_DONTWAIT);
3463 		if (m == NULL)
3464 			return (NULL);
3465 		m->m_flags |= M_IGMPV3_HDR;
3466 	}
3467 
3468 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3469 
3470 	m->m_data += sizeof(struct ip);
3471 	m->m_len -= sizeof(struct ip);
3472 
3473 	igmp = mtod(m, struct igmp_report *);
3474 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3475 	igmp->ir_rsv1 = 0;
3476 	igmp->ir_rsv2 = 0;
3477 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3478 	igmp->ir_cksum = 0;
3479 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3480 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3481 
3482 	m->m_data -= sizeof(struct ip);
3483 	m->m_len += sizeof(struct ip);
3484 
3485 	ip = mtod(m, struct ip *);
3486 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3487 	ip->ip_len = hdrlen + igmpreclen;
3488 	ip->ip_off = IP_DF;
3489 	ip->ip_p = IPPROTO_IGMP;
3490 	ip->ip_sum = 0;
3491 
3492 	ip->ip_src.s_addr = INADDR_ANY;
3493 
3494 	if (m->m_flags & M_IGMP_LOOP) {
3495 		struct in_ifaddr *ia;
3496 
3497 		IFP_TO_IA(ifp, ia);
3498 		if (ia != NULL)
3499 			ip->ip_src = ia->ia_addr.sin_addr;
3500 	}
3501 
3502 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3503 
3504 	return (m);
3505 }
3506 
3507 #ifdef KTR
3508 static char *
3509 igmp_rec_type_to_str(const int type)
3510 {
3511 
3512 	switch (type) {
3513 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3514 			return "TO_EX";
3515 			break;
3516 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3517 			return "TO_IN";
3518 			break;
3519 		case IGMP_MODE_IS_EXCLUDE:
3520 			return "MODE_EX";
3521 			break;
3522 		case IGMP_MODE_IS_INCLUDE:
3523 			return "MODE_IN";
3524 			break;
3525 		case IGMP_ALLOW_NEW_SOURCES:
3526 			return "ALLOW_NEW";
3527 			break;
3528 		case IGMP_BLOCK_OLD_SOURCES:
3529 			return "BLOCK_OLD";
3530 			break;
3531 		default:
3532 			break;
3533 	}
3534 	return "unknown";
3535 }
3536 #endif
3537 
3538 static void
3539 igmp_sysinit(void)
3540 {
3541 
3542 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3543 
3544 	IGMP_LOCK_INIT();
3545 
3546 	mtx_init(&igmpoq.ifq_mtx, "igmpoq_mtx", NULL, MTX_DEF);
3547 	IFQ_SET_MAXLEN(&igmpoq, IFQ_MAXLEN);
3548 
3549 	m_raopt = igmp_ra_alloc();
3550 
3551 	netisr_register(NETISR_IGMP, igmp_intr, &igmpoq, 0);
3552 }
3553 
3554 static void
3555 igmp_sysuninit(void)
3556 {
3557 
3558 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3559 
3560 	netisr_unregister(NETISR_IGMP);
3561 	mtx_destroy(&igmpoq.ifq_mtx);
3562 
3563 	m_free(m_raopt);
3564 	m_raopt = NULL;
3565 
3566 	IGMP_LOCK_DESTROY();
3567 }
3568 
3569 /*
3570  * Initialize an IGMPv3 instance.
3571  * VIMAGE: Assumes curvnet set by caller and called per vimage.
3572  */
3573 static int
3574 vnet_igmp_iattach(const void *unused __unused)
3575 {
3576 	INIT_VNET_INET(curvnet);
3577 
3578 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3579 
3580 	LIST_INIT(&V_igi_head);
3581 
3582 	V_current_state_timers_running = 0;
3583 	V_state_change_timers_running = 0;
3584 	V_interface_timers_running = 0;
3585 
3586 	/*
3587 	 * Initialize sysctls to default values.
3588 	 */
3589 	V_igmp_recvifkludge = 1;
3590 	V_igmp_sendra = 1;
3591 	V_igmp_sendlocal = 1;
3592 	V_igmp_v1enable = 1;
3593 	V_igmp_v2enable = 1;
3594 	V_igmp_legacysupp = 0;
3595 	V_igmp_default_version = IGMP_VERSION_3;
3596 	V_igmp_gsrdelay.tv_sec = 10;
3597 	V_igmp_gsrdelay.tv_usec = 0;
3598 
3599 	memset(&V_igmpstat, 0, sizeof(struct igmpstat));
3600 	V_igmpstat.igps_version = IGPS_VERSION_3;
3601 	V_igmpstat.igps_len = sizeof(struct igmpstat);
3602 
3603 	return (0);
3604 }
3605 
3606 static int
3607 vnet_igmp_idetach(const void *unused __unused)
3608 {
3609 	INIT_VNET_INET(curvnet);
3610 
3611 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3612 
3613 	KASSERT(LIST_EMPTY(&V_igi_head),
3614 	    ("%s: igi list not empty; ifnets not detached?", __func__));
3615 
3616 	return (0);
3617 }
3618 
3619 #ifdef VIMAGE
3620 static struct vnet_symmap vnet_igmp_symmap[] = {
3621 	VNET_SYMMAP(igmp, igi_head),
3622 	VNET_SYMMAP(igmp, igmpstat),
3623 	VNET_SYMMAP_END
3624 };
3625 VNET_MOD_DECLARE(IGMP, igmp, vnet_igmp_iattach, vnet_igmp_idetach,
3626     vnet_igmp_symmap);
3627 #endif /* VIMAGE */
3628 
3629 static int
3630 igmp_modevent(module_t mod, int type, void *unused __unused)
3631 {
3632 
3633     switch (type) {
3634     case MOD_LOAD:
3635 	igmp_sysinit();
3636 #ifdef VIMAGE
3637 	vnet_mod_register(&vnet_igmp_modinfo);
3638 #else
3639 	(void)vnet_igmp_iattach(NULL);
3640 #endif /* VIMAGE */
3641 	break;
3642     case MOD_UNLOAD:
3643 #ifdef VIMAGE
3644 	/*
3645 	 * TODO: Allow module unload if any VIMAGE instances
3646 	 * are using this module.
3647 	 */
3648 	return (EBUSY);
3649 #else
3650 	(void)vnet_igmp_idetach(NULL);
3651 #endif /* VIMAGE */
3652 	igmp_sysuninit();
3653 	break;
3654     default:
3655 	return (EOPNOTSUPP);
3656     }
3657     return (0);
3658 }
3659 
3660 static moduledata_t igmp_mod = {
3661     "igmp",
3662     igmp_modevent,
3663     0
3664 };
3665 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3666