xref: /freebsd/sys/netinet/igmp.c (revision 82431678fce5c893ef9c7418ad6d998ad4187de6)
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_INET
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 static vnet_attach_fn	vnet_igmp_iattach;
145 static vnet_detach_fn	vnet_igmp_idetach;
146 
147 /*
148  * System-wide globals.
149  *
150  * Unlocked access to these is OK, except for the global IGMP output
151  * queue. The IGMP subsystem lock ends up being system-wide for the moment,
152  * because all VIMAGEs have to share a global output queue, as netisrs
153  * themselves are not virtualized.
154  *
155  * Locking:
156  *  * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
157  *    Any may be taken independently; if any are held at the same
158  *    time, the above lock order must be followed.
159  *  * All output is delegated to the netisr.
160  *    Now that Giant has been eliminated, the netisr may be inlined.
161  *  * IN_MULTI_LOCK covers in_multi.
162  *  * IGMP_LOCK covers igmp_ifinfo and any global variables in this file,
163  *    including the output queue.
164  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
165  *    per-link state iterators.
166  *  * igmp_ifinfo is valid as long as PF_INET is attached to the interface,
167  *    therefore it is not refcounted.
168  *    We allow unlocked reads of igmp_ifinfo when accessed via in_multi.
169  *
170  * Reference counting
171  *  * IGMP acquires its own reference every time an in_multi is passed to
172  *    it and the group is being joined for the first time.
173  *  * IGMP releases its reference(s) on in_multi in a deferred way,
174  *    because the operations which process the release run as part of
175  *    a loop whose control variables are directly affected by the release
176  *    (that, and not recursing on the IF_ADDR_LOCK).
177  *
178  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
179  * to a vnet in ifp->if_vnet.
180  *
181  * SMPng: XXX We may potentially race operations on ifma_protospec.
182  * The problem is that we currently lack a clean way of taking the
183  * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
184  * as anything which modifies ifma needs to be covered by that lock.
185  * So check for ifma_protospec being NULL before proceeding.
186  */
187 struct mtx		 igmp_mtx;
188 
189 struct mbuf		*m_raopt;		 /* Router Alert option */
190 MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
191 
192 /*
193  * Global netisr output queue.
194  */
195 struct ifqueue		 igmpoq;
196 
197 /*
198  * VIMAGE-wide globals.
199  *
200  * The IGMPv3 timers themselves need to run per-image, however,
201  * protosw timers run globally (see tcp).
202  * An ifnet can only be in one vimage at a time, and the loopback
203  * ifnet, loif, is itself virtualized.
204  * It would otherwise be possible to seriously hose IGMP state,
205  * and create inconsistencies in upstream multicast routing, if you have
206  * multiple VIMAGEs running on the same link joining different multicast
207  * groups, UNLESS the "primary IP address" is different. This is because
208  * IGMP for IPv4 does not force link-local addresses to be used for each
209  * node, unlike MLD for IPv6.
210  * Obviously the IGMPv3 per-interface state has per-vimage granularity
211  * also as a result.
212  *
213  * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
214  * policy to control the address used by IGMP on the link.
215  */
216 #ifdef VIMAGE_GLOBALS
217 int	 interface_timers_running;	 /* IGMPv3 general query response */
218 int	 state_change_timers_running;	 /* IGMPv3 state-change retransmit */
219 int	 current_state_timers_running;	 /* IGMPv1/v2 host report;
220 					  * IGMPv3 g/sg query response */
221 
222 LIST_HEAD(, igmp_ifinfo)	 igi_head;
223 struct igmpstat			 igmpstat;
224 struct timeval			 igmp_gsrdelay;
225 
226 int	 igmp_recvifkludge;
227 int	 igmp_sendra;
228 int	 igmp_sendlocal;
229 int	 igmp_v1enable;
230 int	 igmp_v2enable;
231 int	 igmp_legacysupp;
232 int	 igmp_default_version;
233 #endif /* VIMAGE_GLOBALS */
234 
235 /*
236  * Virtualized sysctls.
237  */
238 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_igmp, IGMPCTL_STATS, stats,
239     CTLFLAG_RW, igmpstat, igmpstat, "");
240 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, recvifkludge,
241     CTLFLAG_RW, igmp_recvifkludge, 0,
242     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
243 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, sendra,
244     CTLFLAG_RW, igmp_sendra, 0,
245     "Send IP Router Alert option in IGMPv2/v3 messages");
246 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, sendlocal,
247     CTLFLAG_RW, igmp_sendlocal, 0,
248     "Send IGMP membership reports for 224.0.0.0/24 groups");
249 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, v1enable,
250     CTLFLAG_RW, igmp_v1enable, 0,
251     "Enable backwards compatibility with IGMPv1");
252 SYSCTL_V_INT(V_NET, vnet_inet,  _net_inet_igmp, OID_AUTO, v2enable,
253     CTLFLAG_RW, igmp_v2enable, 0,
254     "Enable backwards compatibility with IGMPv2");
255 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, legacysupp,
256     CTLFLAG_RW, igmp_legacysupp, 0,
257     "Allow v1/v2 reports to suppress v3 group responses");
258 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, default_version,
259     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, igmp_default_version, 0,
260     sysctl_igmp_default_version, "I",
261     "Default version of IGMP to run on each interface");
262 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, gsrdelay,
263     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, igmp_gsrdelay.tv_sec, 0,
264     sysctl_igmp_gsr, "I",
265     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
266 
267 /*
268  * Non-virtualized sysctls.
269  */
270 SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
271     sysctl_igmp_ifinfo, "Per-interface IGMPv3 state");
272 
273 static __inline void
274 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
275 {
276 
277 #ifdef VIMAGE
278 	m->m_pkthdr.header = ifp->if_vnet;
279 #endif /* VIMAGE */
280 	m->m_pkthdr.flowid = ifp->if_index;
281 }
282 
283 static __inline void
284 igmp_scrub_context(struct mbuf *m)
285 {
286 
287 	m->m_pkthdr.header = NULL;
288 	m->m_pkthdr.flowid = 0;
289 }
290 
291 #ifdef KTR
292 static __inline char *
293 inet_ntoa_haddr(in_addr_t haddr)
294 {
295 	struct in_addr ia;
296 
297 	ia.s_addr = htonl(haddr);
298 	return (inet_ntoa(ia));
299 }
300 #endif
301 
302 /*
303  * Restore context from a queued IGMP output chain.
304  * Return saved ifindex.
305  *
306  * VIMAGE: The assertion is there to make sure that we
307  * actually called CURVNET_SET() with what's in the mbuf chain.
308  */
309 static __inline uint32_t
310 igmp_restore_context(struct mbuf *m)
311 {
312 
313 #ifdef notyet
314 #if defined(VIMAGE) && defined(INVARIANTS)
315 	KASSERT(curvnet == (m->m_pkthdr.header),
316 	    ("%s: called when curvnet was not restored", __func__));
317 #endif
318 #endif
319 	return (m->m_pkthdr.flowid);
320 }
321 
322 /*
323  * Retrieve or set default IGMP version.
324  *
325  * VIMAGE: Assume curvnet set by caller.
326  * SMPng: NOTE: Serialized by IGMP lock.
327  */
328 static int
329 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
330 {
331 	INIT_VNET_INET(curvnet);
332 	int	 error;
333 	int	 new;
334 
335 	error = sysctl_wire_old_buffer(req, sizeof(int));
336 	if (error)
337 		return (error);
338 
339 	IGMP_LOCK();
340 
341 	new = V_igmp_default_version;
342 
343 	error = sysctl_handle_int(oidp, &new, 0, req);
344 	if (error || !req->newptr)
345 		goto out_locked;
346 
347 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
348 		error = EINVAL;
349 		goto out_locked;
350 	}
351 
352 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
353 	     V_igmp_default_version, new);
354 
355 	V_igmp_default_version = new;
356 
357 out_locked:
358 	IGMP_UNLOCK();
359 	return (error);
360 }
361 
362 /*
363  * Retrieve or set threshold between group-source queries in seconds.
364  *
365  * VIMAGE: Assume curvnet set by caller.
366  * SMPng: NOTE: Serialized by IGMP lock.
367  */
368 static int
369 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
370 {
371 	INIT_VNET_INET(curvnet);
372 	int error;
373 	int i;
374 
375 	error = sysctl_wire_old_buffer(req, sizeof(int));
376 	if (error)
377 		return (error);
378 
379 	IGMP_LOCK();
380 
381 	i = V_igmp_gsrdelay.tv_sec;
382 
383 	error = sysctl_handle_int(oidp, &i, 0, req);
384 	if (error || !req->newptr)
385 		goto out_locked;
386 
387 	if (i < -1 || i >= 60) {
388 		error = EINVAL;
389 		goto out_locked;
390 	}
391 
392 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
393 	     V_igmp_gsrdelay.tv_sec, i);
394 	V_igmp_gsrdelay.tv_sec = i;
395 
396 out_locked:
397 	IGMP_UNLOCK();
398 	return (error);
399 }
400 
401 /*
402  * Expose struct igmp_ifinfo to userland, keyed by ifindex.
403  * For use by ifmcstat(8).
404  *
405  * SMPng: NOTE: Does an unlocked ifindex space read.
406  * VIMAGE: Assume curvnet set by caller. The node handler itself
407  * is not directly virtualized.
408  */
409 static int
410 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
411 {
412 	INIT_VNET_NET(curvnet);
413 	INIT_VNET_INET(curvnet);
414 	int			*name;
415 	int			 error;
416 	u_int			 namelen;
417 	struct ifnet		*ifp;
418 	struct igmp_ifinfo	*igi;
419 
420 	name = (int *)arg1;
421 	namelen = arg2;
422 
423 	if (req->newptr != NULL)
424 		return (EPERM);
425 
426 	if (namelen != 1)
427 		return (EINVAL);
428 
429 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
430 	if (error)
431 		return (error);
432 
433 	IN_MULTI_LOCK();
434 	IGMP_LOCK();
435 
436 	if (name[0] <= 0 || name[0] > V_if_index) {
437 		error = ENOENT;
438 		goto out_locked;
439 	}
440 
441 	error = ENOENT;
442 
443 	ifp = ifnet_byindex(name[0]);
444 	if (ifp == NULL)
445 		goto out_locked;
446 
447 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
448 		if (ifp == igi->igi_ifp) {
449 			error = SYSCTL_OUT(req, igi,
450 			    sizeof(struct igmp_ifinfo));
451 			break;
452 		}
453 	}
454 
455 out_locked:
456 	IGMP_UNLOCK();
457 	IN_MULTI_UNLOCK();
458 	return (error);
459 }
460 
461 /*
462  * Dispatch an entire queue of pending packet chains
463  * using the netisr.
464  * VIMAGE: Assumes the vnet pointer has been set.
465  */
466 static void
467 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop)
468 {
469 	struct mbuf *m;
470 
471 	for (;;) {
472 		_IF_DEQUEUE(ifq, m);
473 		if (m == NULL)
474 			break;
475 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m);
476 		if (loop)
477 			m->m_flags |= M_IGMP_LOOP;
478 		netisr_dispatch(NETISR_IGMP, m);
479 		if (--limit == 0)
480 			break;
481 	}
482 }
483 
484 /*
485  * Filter outgoing IGMP report state by group.
486  *
487  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
488  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
489  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
490  * this may break certain IGMP snooping switches which rely on the old
491  * report behaviour.
492  *
493  * Return zero if the given group is one for which IGMP reports
494  * should be suppressed, or non-zero if reports should be issued.
495  */
496 static __inline int
497 igmp_isgroupreported(const struct in_addr addr)
498 {
499 	INIT_VNET_INET(curvnet);
500 
501 	if (in_allhosts(addr) ||
502 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
503 		return (0);
504 
505 	return (1);
506 }
507 
508 /*
509  * Construct a Router Alert option to use in outgoing packets.
510  */
511 static struct mbuf *
512 igmp_ra_alloc(void)
513 {
514 	struct mbuf	*m;
515 	struct ipoption	*p;
516 
517 	MGET(m, M_DONTWAIT, MT_DATA);
518 	p = mtod(m, struct ipoption *);
519 	p->ipopt_dst.s_addr = INADDR_ANY;
520 	p->ipopt_list[0] = IPOPT_RA;	/* Router Alert Option */
521 	p->ipopt_list[1] = 0x04;	/* 4 bytes long */
522 	p->ipopt_list[2] = IPOPT_EOL;	/* End of IP option list */
523 	p->ipopt_list[3] = 0x00;	/* pad byte */
524 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
525 
526 	return (m);
527 }
528 
529 /*
530  * Attach IGMP when PF_INET is attached to an interface.
531  */
532 struct igmp_ifinfo *
533 igmp_domifattach(struct ifnet *ifp)
534 {
535 	struct igmp_ifinfo *igi;
536 
537 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
538 	    __func__, ifp, ifp->if_xname);
539 
540 	IGMP_LOCK();
541 
542 	igi = igi_alloc_locked(ifp);
543 	if (!(ifp->if_flags & IFF_MULTICAST))
544 		igi->igi_flags |= IGIF_SILENT;
545 
546 	IGMP_UNLOCK();
547 
548 	return (igi);
549 }
550 
551 /*
552  * VIMAGE: assume curvnet set by caller.
553  */
554 static struct igmp_ifinfo *
555 igi_alloc_locked(/*const*/ struct ifnet *ifp)
556 {
557 	INIT_VNET_INET(ifp->if_vnet);
558 	struct igmp_ifinfo *igi;
559 
560 	IGMP_LOCK_ASSERT();
561 
562 	igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO);
563 	if (igi == NULL)
564 		goto out;
565 
566 	igi->igi_ifp = ifp;
567 	igi->igi_version = V_igmp_default_version;
568 	igi->igi_flags = 0;
569 	igi->igi_rv = IGMP_RV_INIT;
570 	igi->igi_qi = IGMP_QI_INIT;
571 	igi->igi_qri = IGMP_QRI_INIT;
572 	igi->igi_uri = IGMP_URI_INIT;
573 
574 	SLIST_INIT(&igi->igi_relinmhead);
575 
576 	/*
577 	 * Responses to general queries are subject to bounds.
578 	 */
579 	IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
580 
581 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
582 
583 	CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)",
584 	     ifp, ifp->if_xname);
585 
586 out:
587 	return (igi);
588 }
589 
590 /*
591  * Hook for ifdetach.
592  *
593  * NOTE: Some finalization tasks need to run before the protocol domain
594  * is detached, but also before the link layer does its cleanup.
595  *
596  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
597  * XXX This is also bitten by unlocked ifma_protospec access.
598  */
599 void
600 igmp_ifdetach(struct ifnet *ifp)
601 {
602 	struct igmp_ifinfo	*igi;
603 	struct ifmultiaddr	*ifma;
604 	struct in_multi		*inm, *tinm;
605 
606 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
607 	    ifp->if_xname);
608 
609 	IGMP_LOCK();
610 
611 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
612 	if (igi->igi_version == IGMP_VERSION_3) {
613 		IF_ADDR_LOCK(ifp);
614 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
615 			if (ifma->ifma_addr->sa_family != AF_INET ||
616 			    ifma->ifma_protospec == NULL)
617 				continue;
618 #if 0
619 			KASSERT(ifma->ifma_protospec != NULL,
620 			    ("%s: ifma_protospec is NULL", __func__));
621 #endif
622 			inm = (struct in_multi *)ifma->ifma_protospec;
623 			if (inm->inm_state == IGMP_LEAVING_MEMBER) {
624 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
625 				    inm, inm_nrele);
626 			}
627 			inm_clear_recorded(inm);
628 		}
629 		IF_ADDR_UNLOCK(ifp);
630 		/*
631 		 * Free the in_multi reference(s) for this IGMP lifecycle.
632 		 */
633 		SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
634 		    tinm) {
635 			SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
636 			inm_release_locked(inm);
637 		}
638 	}
639 
640 	IGMP_UNLOCK();
641 }
642 
643 /*
644  * Hook for domifdetach.
645  */
646 void
647 igmp_domifdetach(struct ifnet *ifp)
648 {
649 	struct igmp_ifinfo *igi;
650 
651 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
652 	    __func__, ifp, ifp->if_xname);
653 
654 	IGMP_LOCK();
655 
656 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
657 	igi_delete_locked(ifp);
658 
659 	IGMP_UNLOCK();
660 }
661 
662 static void
663 igi_delete_locked(const struct ifnet *ifp)
664 {
665 	INIT_VNET_INET(ifp->if_vnet);
666 	struct igmp_ifinfo *igi, *tigi;
667 
668 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
669 	    __func__, ifp, ifp->if_xname);
670 
671 	IGMP_LOCK_ASSERT();
672 
673 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
674 		if (igi->igi_ifp == ifp) {
675 			/*
676 			 * Free deferred General Query responses.
677 			 */
678 			_IF_DRAIN(&igi->igi_gq);
679 
680 			LIST_REMOVE(igi, igi_link);
681 
682 			KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
683 			    ("%s: there are dangling in_multi references",
684 			    __func__));
685 
686 			free(igi, M_IGMP);
687 			return;
688 		}
689 	}
690 
691 #ifdef INVARIANTS
692 	panic("%s: igmp_ifinfo not found for ifp %p\n", __func__,  ifp);
693 #endif
694 }
695 
696 /*
697  * Process a received IGMPv1 query.
698  * Return non-zero if the message should be dropped.
699  *
700  * VIMAGE: The curvnet pointer is derived from the input ifp.
701  */
702 static int
703 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip)
704 {
705 	INIT_VNET_INET(ifp->if_vnet);
706 	struct ifmultiaddr	*ifma;
707 	struct igmp_ifinfo	*igi;
708 	struct in_multi		*inm;
709 
710 	/*
711 	 * IGMPv1 General Queries SHOULD always addressed to 224.0.0.1.
712 	 * igmp_group is always ignored. Do not drop it as a userland
713 	 * daemon may wish to see it.
714 	 */
715 	if (!in_allhosts(ip->ip_dst)) {
716 		IGMPSTAT_INC(igps_rcv_badqueries);
717 		return (0);
718 	}
719 
720 	IGMPSTAT_INC(igps_rcv_gen_queries);
721 
722 	/*
723 	 * Switch to IGMPv1 host compatibility mode.
724 	 */
725 	IN_MULTI_LOCK();
726 	IGMP_LOCK();
727 
728 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
729 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
730 
731 	if (igi->igi_flags & IGIF_LOOPBACK) {
732 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
733 		    ifp, ifp->if_xname);
734 		goto out_locked;
735 	}
736 
737 	igmp_set_version(igi, IGMP_VERSION_1);
738 
739 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
740 
741 	/*
742 	 * Start the timers in all of our group records
743 	 * for the interface on which the query arrived,
744 	 * except those which are already running.
745 	 */
746 	IF_ADDR_LOCK(ifp);
747 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
748 		if (ifma->ifma_addr->sa_family != AF_INET ||
749 		    ifma->ifma_protospec == NULL)
750 			continue;
751 		inm = (struct in_multi *)ifma->ifma_protospec;
752 		if (inm->inm_timer != 0)
753 			continue;
754 		switch (inm->inm_state) {
755 		case IGMP_NOT_MEMBER:
756 		case IGMP_SILENT_MEMBER:
757 			break;
758 		case IGMP_G_QUERY_PENDING_MEMBER:
759 		case IGMP_SG_QUERY_PENDING_MEMBER:
760 		case IGMP_REPORTING_MEMBER:
761 		case IGMP_IDLE_MEMBER:
762 		case IGMP_LAZY_MEMBER:
763 		case IGMP_SLEEPING_MEMBER:
764 		case IGMP_AWAKENING_MEMBER:
765 			inm->inm_state = IGMP_REPORTING_MEMBER;
766 			inm->inm_timer = IGMP_RANDOM_DELAY(
767 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
768 			V_current_state_timers_running = 1;
769 			break;
770 		case IGMP_LEAVING_MEMBER:
771 			break;
772 		}
773 	}
774 	IF_ADDR_UNLOCK(ifp);
775 
776 out_locked:
777 	IGMP_UNLOCK();
778 	IN_MULTI_UNLOCK();
779 
780 	return (0);
781 }
782 
783 /*
784  * Process a received IGMPv2 general or group-specific query.
785  */
786 static int
787 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
788     const struct igmp *igmp)
789 {
790 	INIT_VNET_INET(ifp->if_vnet);
791 	struct ifmultiaddr	*ifma;
792 	struct igmp_ifinfo	*igi;
793 	struct in_multi		*inm;
794 	uint16_t		 timer;
795 
796 	/*
797 	 * Perform lazy allocation of IGMP link info if required,
798 	 * and switch to IGMPv2 host compatibility mode.
799 	 */
800 	IN_MULTI_LOCK();
801 	IGMP_LOCK();
802 
803 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
804 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
805 
806 	if (igi->igi_flags & IGIF_LOOPBACK) {
807 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
808 		    ifp, ifp->if_xname);
809 		goto out_locked;
810 	}
811 
812 	igmp_set_version(igi, IGMP_VERSION_2);
813 
814 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
815 	if (timer == 0)
816 		timer = 1;
817 
818 	if (!in_nullhost(igmp->igmp_group)) {
819 		/*
820 		 * IGMPv2 Group-Specific Query.
821 		 * If this is a group-specific IGMPv2 query, we need only
822 		 * look up the single group to process it.
823 		 */
824 		inm = inm_lookup(ifp, igmp->igmp_group);
825 		if (inm != NULL) {
826 			CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
827 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
828 			igmp_v2_update_group(inm, timer);
829 		}
830 		IGMPSTAT_INC(igps_rcv_group_queries);
831 	} else {
832 		/*
833 		 * IGMPv2 General Query.
834 		 * If this was not sent to the all-hosts group, ignore it.
835 		 */
836 		if (in_allhosts(ip->ip_dst)) {
837 			/*
838 			 * For each reporting group joined on this
839 			 * interface, kick the report timer.
840 			 */
841 			CTR2(KTR_IGMPV3,
842 			    "process v2 general query on ifp %p(%s)",
843 			    ifp, ifp->if_xname);
844 
845 			IF_ADDR_LOCK(ifp);
846 			TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
847 				if (ifma->ifma_addr->sa_family != AF_INET ||
848 				    ifma->ifma_protospec == NULL)
849 					continue;
850 				inm = (struct in_multi *)ifma->ifma_protospec;
851 				igmp_v2_update_group(inm, timer);
852 			}
853 			IF_ADDR_UNLOCK(ifp);
854 		}
855 		IGMPSTAT_INC(igps_rcv_gen_queries);
856 	}
857 
858 out_locked:
859 	IGMP_UNLOCK();
860 	IN_MULTI_UNLOCK();
861 
862 	return (0);
863 }
864 
865 /*
866  * Update the report timer on a group in response to an IGMPv2 query.
867  *
868  * If we are becoming the reporting member for this group, start the timer.
869  * If we already are the reporting member for this group, and timer is
870  * below the threshold, reset it.
871  *
872  * We may be updating the group for the first time since we switched
873  * to IGMPv3. If we are, then we must clear any recorded source lists,
874  * and transition to REPORTING state; the group timer is overloaded
875  * for group and group-source query responses.
876  *
877  * Unlike IGMPv3, the delay per group should be jittered
878  * to avoid bursts of IGMPv2 reports.
879  */
880 static void
881 igmp_v2_update_group(struct in_multi *inm, const int timer)
882 {
883 	INIT_VNET_INET(curvnet);
884 
885 	CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
886 	    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
887 
888 	IN_MULTI_LOCK_ASSERT();
889 
890 	switch (inm->inm_state) {
891 	case IGMP_NOT_MEMBER:
892 	case IGMP_SILENT_MEMBER:
893 		break;
894 	case IGMP_REPORTING_MEMBER:
895 		if (inm->inm_timer != 0 &&
896 		    inm->inm_timer <= timer) {
897 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
898 			    "skipping.", __func__);
899 			break;
900 		}
901 		/* FALLTHROUGH */
902 	case IGMP_SG_QUERY_PENDING_MEMBER:
903 	case IGMP_G_QUERY_PENDING_MEMBER:
904 	case IGMP_IDLE_MEMBER:
905 	case IGMP_LAZY_MEMBER:
906 	case IGMP_AWAKENING_MEMBER:
907 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
908 		inm->inm_state = IGMP_REPORTING_MEMBER;
909 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
910 		V_current_state_timers_running = 1;
911 		break;
912 	case IGMP_SLEEPING_MEMBER:
913 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
914 		inm->inm_state = IGMP_AWAKENING_MEMBER;
915 		break;
916 	case IGMP_LEAVING_MEMBER:
917 		break;
918 	}
919 }
920 
921 /*
922  * Process a received IGMPv3 general, group-specific or
923  * group-and-source-specific query.
924  * Assumes m has already been pulled up to the full IGMP message length.
925  * Return 0 if successful, otherwise an appropriate error code is returned.
926  */
927 static int
928 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
929     /*const*/ struct igmpv3 *igmpv3)
930 {
931 	INIT_VNET_INET(ifp->if_vnet);
932 	struct igmp_ifinfo	*igi;
933 	struct in_multi		*inm;
934 	uint32_t		 maxresp, nsrc, qqi;
935 	uint16_t		 timer;
936 	uint8_t			 qrv;
937 
938 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
939 
940 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
941 	if (maxresp >= 128) {
942 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
943 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
944 	}
945 
946 	/*
947 	 * Robustness must never be less than 2 for on-wire IGMPv3.
948 	 * FIXME: Check if ifp has IGIF_LOOPBACK set, as we make
949 	 * an exception for interfaces whose IGMPv3 state changes
950 	 * are redirected to loopback (e.g. MANET).
951 	 */
952 	qrv = IGMP_QRV(igmpv3->igmp_misc);
953 	if (qrv < 2) {
954 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
955 		    qrv, IGMP_RV_INIT);
956 		qrv = IGMP_RV_INIT;
957 	}
958 
959 	qqi = igmpv3->igmp_qqi;
960 	if (qqi >= 128) {
961 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
962 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
963 	}
964 
965 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
966 	if (timer == 0)
967 		timer = 1;
968 
969 	nsrc = ntohs(igmpv3->igmp_numsrc);
970 
971 	IN_MULTI_LOCK();
972 	IGMP_LOCK();
973 
974 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
975 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
976 
977 	if (igi->igi_flags & IGIF_LOOPBACK) {
978 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
979 		    ifp, ifp->if_xname);
980 		goto out_locked;
981 	}
982 
983 	igmp_set_version(igi, IGMP_VERSION_3);
984 
985 	igi->igi_rv = qrv;
986 	igi->igi_qi = qqi;
987 	igi->igi_qri = maxresp;
988 
989 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
990 	    maxresp);
991 
992 	if (in_nullhost(igmpv3->igmp_group)) {
993 		/*
994 		 * IGMPv3 General Query.
995 		 * Schedule a current-state report on this ifp for
996 		 * all groups, possibly containing source lists.
997 		 */
998 		IGMPSTAT_INC(igps_rcv_gen_queries);
999 
1000 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1001 			/*
1002 			 * General Queries SHOULD be directed to 224.0.0.1.
1003 			 * A general query with a source list has undefined
1004 			 * behaviour; discard it.
1005 			 */
1006 			IGMPSTAT_INC(igps_rcv_badqueries);
1007 			goto out_locked;
1008 		}
1009 
1010 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1011 		    ifp, ifp->if_xname);
1012 
1013 		/*
1014 		 * If there is a pending General Query response
1015 		 * scheduled earlier than the selected delay, do
1016 		 * not schedule any other reports.
1017 		 * Otherwise, reset the interface timer.
1018 		 */
1019 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1020 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1021 			V_interface_timers_running = 1;
1022 		}
1023 	} else {
1024 		/*
1025 		 * IGMPv3 Group-specific or Group-and-source-specific Query.
1026 		 *
1027 		 * Group-source-specific queries are throttled on
1028 		 * a per-group basis to defeat denial-of-service attempts.
1029 		 * Queries for groups we are not a member of on this
1030 		 * link are simply ignored.
1031 		 */
1032 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1033 		if (inm == NULL)
1034 			goto out_locked;
1035 		if (nsrc > 0) {
1036 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1037 			if (!ratecheck(&inm->inm_lastgsrtv,
1038 			    &V_igmp_gsrdelay)) {
1039 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1040 				    __func__);
1041 				IGMPSTAT_INC(igps_drop_gsr_queries);
1042 				goto out_locked;
1043 			}
1044 		} else {
1045 			IGMPSTAT_INC(igps_rcv_group_queries);
1046 		}
1047 		CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1048 		     inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1049 		/*
1050 		 * If there is a pending General Query response
1051 		 * scheduled sooner than the selected delay, no
1052 		 * further report need be scheduled.
1053 		 * Otherwise, prepare to respond to the
1054 		 * group-specific or group-and-source query.
1055 		 */
1056 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1057 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1058 	}
1059 
1060 out_locked:
1061 	IGMP_UNLOCK();
1062 	IN_MULTI_UNLOCK();
1063 
1064 	return (0);
1065 }
1066 
1067 /*
1068  * Process a recieved IGMPv3 group-specific or group-and-source-specific
1069  * query.
1070  * Return <0 if any error occured. Currently this is ignored.
1071  */
1072 static int
1073 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1074     int timer, /*const*/ struct igmpv3 *igmpv3)
1075 {
1076 	INIT_VNET_INET(curvnet);
1077 	int			 retval;
1078 	uint16_t		 nsrc;
1079 
1080 	IN_MULTI_LOCK_ASSERT();
1081 	IGMP_LOCK_ASSERT();
1082 
1083 	retval = 0;
1084 
1085 	switch (inm->inm_state) {
1086 	case IGMP_NOT_MEMBER:
1087 	case IGMP_SILENT_MEMBER:
1088 	case IGMP_SLEEPING_MEMBER:
1089 	case IGMP_LAZY_MEMBER:
1090 	case IGMP_AWAKENING_MEMBER:
1091 	case IGMP_IDLE_MEMBER:
1092 	case IGMP_LEAVING_MEMBER:
1093 		return (retval);
1094 		break;
1095 	case IGMP_REPORTING_MEMBER:
1096 	case IGMP_G_QUERY_PENDING_MEMBER:
1097 	case IGMP_SG_QUERY_PENDING_MEMBER:
1098 		break;
1099 	}
1100 
1101 	nsrc = ntohs(igmpv3->igmp_numsrc);
1102 
1103 	/*
1104 	 * Deal with group-specific queries upfront.
1105 	 * If any group query is already pending, purge any recorded
1106 	 * source-list state if it exists, and schedule a query response
1107 	 * for this group-specific query.
1108 	 */
1109 	if (nsrc == 0) {
1110 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1111 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1112 			inm_clear_recorded(inm);
1113 			timer = min(inm->inm_timer, timer);
1114 		}
1115 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1116 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1117 		V_current_state_timers_running = 1;
1118 		return (retval);
1119 	}
1120 
1121 	/*
1122 	 * Deal with the case where a group-and-source-specific query has
1123 	 * been received but a group-specific query is already pending.
1124 	 */
1125 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1126 		timer = min(inm->inm_timer, timer);
1127 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1128 		V_current_state_timers_running = 1;
1129 		return (retval);
1130 	}
1131 
1132 	/*
1133 	 * Finally, deal with the case where a group-and-source-specific
1134 	 * query has been received, where a response to a previous g-s-r
1135 	 * query exists, or none exists.
1136 	 * In this case, we need to parse the source-list which the Querier
1137 	 * has provided us with and check if we have any source list filter
1138 	 * entries at T1 for these sources. If we do not, there is no need
1139 	 * schedule a report and the query may be dropped.
1140 	 * If we do, we must record them and schedule a current-state
1141 	 * report for those sources.
1142 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1143 	 * we pass the mbuf chain pointer down to this function, and use
1144 	 * m_getptr() to walk the chain.
1145 	 */
1146 	if (inm->inm_nsrc > 0) {
1147 		const struct in_addr	*ap;
1148 		int			 i, nrecorded;
1149 
1150 		ap = (const struct in_addr *)(igmpv3 + 1);
1151 		nrecorded = 0;
1152 		for (i = 0; i < nsrc; i++, ap++) {
1153 			retval = inm_record_source(inm, ap->s_addr);
1154 			if (retval < 0)
1155 				break;
1156 			nrecorded += retval;
1157 		}
1158 		if (nrecorded > 0) {
1159 			CTR1(KTR_IGMPV3,
1160 			    "%s: schedule response to SG query", __func__);
1161 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1162 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1163 			V_current_state_timers_running = 1;
1164 		}
1165 	}
1166 
1167 	return (retval);
1168 }
1169 
1170 /*
1171  * Process a received IGMPv1 host membership report.
1172  *
1173  * NOTE: 0.0.0.0 workaround breaks const correctness.
1174  */
1175 static int
1176 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1177     /*const*/ struct igmp *igmp)
1178 {
1179 	INIT_VNET_INET(ifp->if_vnet);
1180 	struct in_ifaddr *ia;
1181 	struct in_multi *inm;
1182 
1183 	IGMPSTAT_INC(igps_rcv_reports);
1184 
1185 	if (ifp->if_flags & IFF_LOOPBACK)
1186 		return (0);
1187 
1188 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr) ||
1189 	    !in_hosteq(igmp->igmp_group, ip->ip_dst))) {
1190 		IGMPSTAT_INC(igps_rcv_badreports);
1191 		return (EINVAL);
1192 	}
1193 
1194 	/*
1195 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1196 	 * Booting clients may use the source address 0.0.0.0. Some
1197 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1198 	 * the interface upon which this message was received.
1199 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1200 	 */
1201 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1202 		IFP_TO_IA(ifp, ia);
1203 		if (ia != NULL)
1204 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1205 	}
1206 
1207 	CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1208 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1209 
1210 	/*
1211 	 * IGMPv1 report suppression.
1212 	 * If we are a member of this group, and our membership should be
1213 	 * reported, stop our group timer and transition to the 'lazy' state.
1214 	 */
1215 	IN_MULTI_LOCK();
1216 	inm = inm_lookup(ifp, igmp->igmp_group);
1217 	if (inm != NULL) {
1218 		struct igmp_ifinfo *igi;
1219 
1220 		igi = inm->inm_igi;
1221 		if (igi == NULL) {
1222 			KASSERT(igi != NULL,
1223 			    ("%s: no igi for ifp %p", __func__, ifp));
1224 			goto out_locked;
1225 		}
1226 
1227 		IGMPSTAT_INC(igps_rcv_ourreports);
1228 
1229 		/*
1230 		 * If we are in IGMPv3 host mode, do not allow the
1231 		 * other host's IGMPv1 report to suppress our reports
1232 		 * unless explicitly configured to do so.
1233 		 */
1234 		if (igi->igi_version == IGMP_VERSION_3) {
1235 			if (V_igmp_legacysupp)
1236 				igmp_v3_suppress_group_record(inm);
1237 			goto out_locked;
1238 		}
1239 
1240 		inm->inm_timer = 0;
1241 
1242 		switch (inm->inm_state) {
1243 		case IGMP_NOT_MEMBER:
1244 		case IGMP_SILENT_MEMBER:
1245 			break;
1246 		case IGMP_IDLE_MEMBER:
1247 		case IGMP_LAZY_MEMBER:
1248 		case IGMP_AWAKENING_MEMBER:
1249 			CTR3(KTR_IGMPV3,
1250 			    "report suppressed for %s on ifp %p(%s)",
1251 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1252 		case IGMP_SLEEPING_MEMBER:
1253 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1254 			break;
1255 		case IGMP_REPORTING_MEMBER:
1256 			CTR3(KTR_IGMPV3,
1257 			    "report suppressed for %s on ifp %p(%s)",
1258 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1259 			if (igi->igi_version == IGMP_VERSION_1)
1260 				inm->inm_state = IGMP_LAZY_MEMBER;
1261 			else if (igi->igi_version == IGMP_VERSION_2)
1262 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1263 			break;
1264 		case IGMP_G_QUERY_PENDING_MEMBER:
1265 		case IGMP_SG_QUERY_PENDING_MEMBER:
1266 		case IGMP_LEAVING_MEMBER:
1267 			break;
1268 		}
1269 	}
1270 
1271 out_locked:
1272 	IN_MULTI_UNLOCK();
1273 
1274 	return (0);
1275 }
1276 
1277 /*
1278  * Process a received IGMPv2 host membership report.
1279  *
1280  * NOTE: 0.0.0.0 workaround breaks const correctness.
1281  */
1282 static int
1283 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1284     /*const*/ struct igmp *igmp)
1285 {
1286 	INIT_VNET_INET(ifp->if_vnet);
1287 	struct in_ifaddr *ia;
1288 	struct in_multi *inm;
1289 
1290 	/*
1291 	 * Make sure we don't hear our own membership report.  Fast
1292 	 * leave requires knowing that we are the only member of a
1293 	 * group.
1294 	 */
1295 	IFP_TO_IA(ifp, ia);
1296 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr))
1297 		return (0);
1298 
1299 	IGMPSTAT_INC(igps_rcv_reports);
1300 
1301 	if (ifp->if_flags & IFF_LOOPBACK)
1302 		return (0);
1303 
1304 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1305 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1306 		IGMPSTAT_INC(igps_rcv_badreports);
1307 		return (EINVAL);
1308 	}
1309 
1310 	/*
1311 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1312 	 * Booting clients may use the source address 0.0.0.0. Some
1313 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1314 	 * the interface upon which this message was received.
1315 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1316 	 */
1317 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1318 		if (ia != NULL)
1319 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1320 	}
1321 
1322 	CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1323 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1324 
1325 	/*
1326 	 * IGMPv2 report suppression.
1327 	 * If we are a member of this group, and our membership should be
1328 	 * reported, and our group timer is pending or about to be reset,
1329 	 * stop our group timer by transitioning to the 'lazy' state.
1330 	 */
1331 	IN_MULTI_LOCK();
1332 	inm = inm_lookup(ifp, igmp->igmp_group);
1333 	if (inm != NULL) {
1334 		struct igmp_ifinfo *igi;
1335 
1336 		igi = inm->inm_igi;
1337 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1338 
1339 		IGMPSTAT_INC(igps_rcv_ourreports);
1340 
1341 		/*
1342 		 * If we are in IGMPv3 host mode, do not allow the
1343 		 * other host's IGMPv1 report to suppress our reports
1344 		 * unless explicitly configured to do so.
1345 		 */
1346 		if (igi->igi_version == IGMP_VERSION_3) {
1347 			if (V_igmp_legacysupp)
1348 				igmp_v3_suppress_group_record(inm);
1349 			goto out_locked;
1350 		}
1351 
1352 		inm->inm_timer = 0;
1353 
1354 		switch (inm->inm_state) {
1355 		case IGMP_NOT_MEMBER:
1356 		case IGMP_SILENT_MEMBER:
1357 		case IGMP_SLEEPING_MEMBER:
1358 			break;
1359 		case IGMP_REPORTING_MEMBER:
1360 		case IGMP_IDLE_MEMBER:
1361 		case IGMP_AWAKENING_MEMBER:
1362 			CTR3(KTR_IGMPV3,
1363 			    "report suppressed for %s on ifp %p(%s)",
1364 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1365 		case IGMP_LAZY_MEMBER:
1366 			inm->inm_state = IGMP_LAZY_MEMBER;
1367 			break;
1368 		case IGMP_G_QUERY_PENDING_MEMBER:
1369 		case IGMP_SG_QUERY_PENDING_MEMBER:
1370 		case IGMP_LEAVING_MEMBER:
1371 			break;
1372 		}
1373 	}
1374 
1375 out_locked:
1376 	IN_MULTI_UNLOCK();
1377 
1378 	return (0);
1379 }
1380 
1381 void
1382 igmp_input(struct mbuf *m, int off)
1383 {
1384 	int iphlen;
1385 	struct ifnet *ifp;
1386 	struct igmp *igmp;
1387 	struct ip *ip;
1388 	int igmplen;
1389 	int minlen;
1390 	int queryver;
1391 
1392 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1393 
1394 	ifp = m->m_pkthdr.rcvif;
1395 	INIT_VNET_INET(ifp->if_vnet);
1396 
1397 	IGMPSTAT_INC(igps_rcv_total);
1398 
1399 	ip = mtod(m, struct ip *);
1400 	iphlen = off;
1401 	igmplen = ip->ip_len;
1402 
1403 	/*
1404 	 * Validate lengths.
1405 	 */
1406 	if (igmplen < IGMP_MINLEN) {
1407 		IGMPSTAT_INC(igps_rcv_tooshort);
1408 		m_freem(m);
1409 		return;
1410 	}
1411 
1412 	/*
1413 	 * Always pullup to the minimum size for v1/v2 or v3
1414 	 * to amortize calls to m_pullup().
1415 	 */
1416 	minlen = iphlen;
1417 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1418 		minlen += IGMP_V3_QUERY_MINLEN;
1419 	else
1420 		minlen += IGMP_MINLEN;
1421 	if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1422 	    (m = m_pullup(m, minlen)) == 0) {
1423 		IGMPSTAT_INC(igps_rcv_tooshort);
1424 		return;
1425 	}
1426 	ip = mtod(m, struct ip *);
1427 
1428 	if (ip->ip_ttl != 1) {
1429 		IGMPSTAT_INC(igps_rcv_badttl);
1430 		m_freem(m);
1431 		return;
1432 	}
1433 
1434 	/*
1435 	 * Validate checksum.
1436 	 */
1437 	m->m_data += iphlen;
1438 	m->m_len -= iphlen;
1439 	igmp = mtod(m, struct igmp *);
1440 	if (in_cksum(m, igmplen)) {
1441 		IGMPSTAT_INC(igps_rcv_badsum);
1442 		m_freem(m);
1443 		return;
1444 	}
1445 	m->m_data -= iphlen;
1446 	m->m_len += iphlen;
1447 
1448 	switch (igmp->igmp_type) {
1449 	case IGMP_HOST_MEMBERSHIP_QUERY:
1450 		if (igmplen == IGMP_MINLEN) {
1451 			if (igmp->igmp_code == 0)
1452 				queryver = IGMP_VERSION_1;
1453 			else
1454 				queryver = IGMP_VERSION_2;
1455 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1456 			queryver = IGMP_VERSION_3;
1457 		} else {
1458 			IGMPSTAT_INC(igps_rcv_tooshort);
1459 			m_freem(m);
1460 			return;
1461 		}
1462 
1463 		switch (queryver) {
1464 		case IGMP_VERSION_1:
1465 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1466 			if (!V_igmp_v1enable)
1467 				break;
1468 			if (igmp_input_v1_query(ifp, ip) != 0) {
1469 				m_freem(m);
1470 				return;
1471 			}
1472 			break;
1473 
1474 		case IGMP_VERSION_2:
1475 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1476 			if (!V_igmp_v2enable)
1477 				break;
1478 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1479 				m_freem(m);
1480 				return;
1481 			}
1482 			break;
1483 
1484 		case IGMP_VERSION_3: {
1485 				struct igmpv3 *igmpv3;
1486 				uint16_t igmpv3len;
1487 				uint16_t srclen;
1488 				int nsrc;
1489 
1490 				IGMPSTAT_INC(igps_rcv_v3_queries);
1491 				igmpv3 = (struct igmpv3 *)igmp;
1492 				/*
1493 				 * Validate length based on source count.
1494 				 */
1495 				nsrc = ntohs(igmpv3->igmp_numsrc);
1496 				srclen = sizeof(struct in_addr) * nsrc;
1497 				if (nsrc * sizeof(in_addr_t) > srclen) {
1498 					IGMPSTAT_INC(igps_rcv_tooshort);
1499 					return;
1500 				}
1501 				/*
1502 				 * m_pullup() may modify m, so pullup in
1503 				 * this scope.
1504 				 */
1505 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1506 				    srclen;
1507 				if ((m->m_flags & M_EXT ||
1508 				     m->m_len < igmpv3len) &&
1509 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1510 					IGMPSTAT_INC(igps_rcv_tooshort);
1511 					return;
1512 				}
1513 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1514 				    + iphlen);
1515 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1516 					m_freem(m);
1517 					return;
1518 				}
1519 			}
1520 			break;
1521 		}
1522 		break;
1523 
1524 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1525 		if (!V_igmp_v1enable)
1526 			break;
1527 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1528 			m_freem(m);
1529 			return;
1530 		}
1531 		break;
1532 
1533 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1534 		if (!V_igmp_v2enable)
1535 			break;
1536 		if (!ip_checkrouteralert(m))
1537 			IGMPSTAT_INC(igps_rcv_nora);
1538 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1539 			m_freem(m);
1540 			return;
1541 		}
1542 		break;
1543 
1544 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1545 		/*
1546 		 * Hosts do not need to process IGMPv3 membership reports,
1547 		 * as report suppression is no longer required.
1548 		 */
1549 		if (!ip_checkrouteralert(m))
1550 			IGMPSTAT_INC(igps_rcv_nora);
1551 		break;
1552 
1553 	default:
1554 		break;
1555 	}
1556 
1557 	/*
1558 	 * Pass all valid IGMP packets up to any process(es) listening on a
1559 	 * raw IGMP socket.
1560 	 */
1561 	rip_input(m, off);
1562 }
1563 
1564 
1565 /*
1566  * Fast timeout handler (global).
1567  * VIMAGE: Timeout handlers are expected to service all vimages.
1568  */
1569 void
1570 igmp_fasttimo(void)
1571 {
1572 	VNET_ITERATOR_DECL(vnet_iter);
1573 
1574 	VNET_LIST_RLOCK();
1575 	VNET_FOREACH(vnet_iter) {
1576 		CURVNET_SET(vnet_iter);
1577 		igmp_fasttimo_vnet();
1578 		CURVNET_RESTORE();
1579 	}
1580 	VNET_LIST_RUNLOCK();
1581 }
1582 
1583 /*
1584  * Fast timeout handler (per-vnet).
1585  * Sends are shuffled off to a netisr to deal with Giant.
1586  *
1587  * VIMAGE: Assume caller has set up our curvnet.
1588  */
1589 static void
1590 igmp_fasttimo_vnet(void)
1591 {
1592 	INIT_VNET_INET(curvnet);
1593 	struct ifqueue		 scq;	/* State-change packets */
1594 	struct ifqueue		 qrq;	/* Query response packets */
1595 	struct ifnet		*ifp;
1596 	struct igmp_ifinfo	*igi;
1597 	struct ifmultiaddr	*ifma, *tifma;
1598 	struct in_multi		*inm;
1599 	int			 loop, uri_fasthz;
1600 
1601 	loop = 0;
1602 	uri_fasthz = 0;
1603 
1604 	/*
1605 	 * Quick check to see if any work needs to be done, in order to
1606 	 * minimize the overhead of fasttimo processing.
1607 	 * SMPng: XXX Unlocked reads.
1608 	 */
1609 	if (!V_current_state_timers_running &&
1610 	    !V_interface_timers_running &&
1611 	    !V_state_change_timers_running)
1612 		return;
1613 
1614 	IN_MULTI_LOCK();
1615 	IGMP_LOCK();
1616 
1617 	/*
1618 	 * IGMPv3 General Query response timer processing.
1619 	 */
1620 	if (V_interface_timers_running) {
1621 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1622 
1623 		V_interface_timers_running = 0;
1624 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1625 			if (igi->igi_v3_timer == 0) {
1626 				/* Do nothing. */
1627 			} else if (--igi->igi_v3_timer == 0) {
1628 				igmp_v3_dispatch_general_query(igi);
1629 			} else {
1630 				V_interface_timers_running = 1;
1631 			}
1632 		}
1633 	}
1634 
1635 	if (!V_current_state_timers_running &&
1636 	    !V_state_change_timers_running)
1637 		goto out_locked;
1638 
1639 	V_current_state_timers_running = 0;
1640 	V_state_change_timers_running = 0;
1641 
1642 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1643 
1644 	/*
1645 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1646 	 * Note: Processing a v3 group timer may remove a node.
1647 	 */
1648 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1649 		ifp = igi->igi_ifp;
1650 
1651 		if (igi->igi_version == IGMP_VERSION_3) {
1652 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1653 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1654 			    PR_FASTHZ);
1655 
1656 			memset(&qrq, 0, sizeof(struct ifqueue));
1657 			IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1658 
1659 			memset(&scq, 0, sizeof(struct ifqueue));
1660 			IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1661 		}
1662 
1663 		IF_ADDR_LOCK(ifp);
1664 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1665 		    tifma) {
1666 			if (ifma->ifma_addr->sa_family != AF_INET ||
1667 			    ifma->ifma_protospec == NULL)
1668 				continue;
1669 			inm = (struct in_multi *)ifma->ifma_protospec;
1670 			switch (igi->igi_version) {
1671 			case IGMP_VERSION_1:
1672 			case IGMP_VERSION_2:
1673 				igmp_v1v2_process_group_timer(inm,
1674 				    igi->igi_version);
1675 				break;
1676 			case IGMP_VERSION_3:
1677 				igmp_v3_process_group_timers(igi, &qrq,
1678 				    &scq, inm, uri_fasthz);
1679 				break;
1680 			}
1681 		}
1682 		IF_ADDR_UNLOCK(ifp);
1683 
1684 		if (igi->igi_version == IGMP_VERSION_3) {
1685 			struct in_multi		*tinm;
1686 
1687 			igmp_dispatch_queue(&qrq, 0, loop);
1688 			igmp_dispatch_queue(&scq, 0, loop);
1689 
1690 			/*
1691 			 * Free the in_multi reference(s) for this
1692 			 * IGMP lifecycle.
1693 			 */
1694 			SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1695 			    inm_nrele, tinm) {
1696 				SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1697 				    inm_nrele);
1698 				inm_release_locked(inm);
1699 			}
1700 		}
1701 	}
1702 
1703 out_locked:
1704 	IGMP_UNLOCK();
1705 	IN_MULTI_UNLOCK();
1706 }
1707 
1708 /*
1709  * Update host report group timer for IGMPv1/v2.
1710  * Will update the global pending timer flags.
1711  */
1712 static void
1713 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1714 {
1715 	INIT_VNET_INET(curvnet);
1716 	int report_timer_expired;
1717 
1718 	IN_MULTI_LOCK_ASSERT();
1719 	IGMP_LOCK_ASSERT();
1720 
1721 	if (inm->inm_timer == 0) {
1722 		report_timer_expired = 0;
1723 	} else if (--inm->inm_timer == 0) {
1724 		report_timer_expired = 1;
1725 	} else {
1726 		V_current_state_timers_running = 1;
1727 		return;
1728 	}
1729 
1730 	switch (inm->inm_state) {
1731 	case IGMP_NOT_MEMBER:
1732 	case IGMP_SILENT_MEMBER:
1733 	case IGMP_IDLE_MEMBER:
1734 	case IGMP_LAZY_MEMBER:
1735 	case IGMP_SLEEPING_MEMBER:
1736 	case IGMP_AWAKENING_MEMBER:
1737 		break;
1738 	case IGMP_REPORTING_MEMBER:
1739 		if (report_timer_expired) {
1740 			inm->inm_state = IGMP_IDLE_MEMBER;
1741 			(void)igmp_v1v2_queue_report(inm,
1742 			    (version == IGMP_VERSION_2) ?
1743 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1744 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1745 		}
1746 		break;
1747 	case IGMP_G_QUERY_PENDING_MEMBER:
1748 	case IGMP_SG_QUERY_PENDING_MEMBER:
1749 	case IGMP_LEAVING_MEMBER:
1750 		break;
1751 	}
1752 }
1753 
1754 /*
1755  * Update a group's timers for IGMPv3.
1756  * Will update the global pending timer flags.
1757  * Note: Unlocked read from igi.
1758  */
1759 static void
1760 igmp_v3_process_group_timers(struct igmp_ifinfo *igi,
1761     struct ifqueue *qrq, struct ifqueue *scq,
1762     struct in_multi *inm, const int uri_fasthz)
1763 {
1764 	INIT_VNET_INET(curvnet);
1765 	int query_response_timer_expired;
1766 	int state_change_retransmit_timer_expired;
1767 
1768 	IN_MULTI_LOCK_ASSERT();
1769 	IGMP_LOCK_ASSERT();
1770 
1771 	query_response_timer_expired = 0;
1772 	state_change_retransmit_timer_expired = 0;
1773 
1774 	/*
1775 	 * During a transition from v1/v2 compatibility mode back to v3,
1776 	 * a group record in REPORTING state may still have its group
1777 	 * timer active. This is a no-op in this function; it is easier
1778 	 * to deal with it here than to complicate the slow-timeout path.
1779 	 */
1780 	if (inm->inm_timer == 0) {
1781 		query_response_timer_expired = 0;
1782 	} else if (--inm->inm_timer == 0) {
1783 		query_response_timer_expired = 1;
1784 	} else {
1785 		V_current_state_timers_running = 1;
1786 	}
1787 
1788 	if (inm->inm_sctimer == 0) {
1789 		state_change_retransmit_timer_expired = 0;
1790 	} else if (--inm->inm_sctimer == 0) {
1791 		state_change_retransmit_timer_expired = 1;
1792 	} else {
1793 		V_state_change_timers_running = 1;
1794 	}
1795 
1796 	/* We are in fasttimo, so be quick about it. */
1797 	if (!state_change_retransmit_timer_expired &&
1798 	    !query_response_timer_expired)
1799 		return;
1800 
1801 	switch (inm->inm_state) {
1802 	case IGMP_NOT_MEMBER:
1803 	case IGMP_SILENT_MEMBER:
1804 	case IGMP_SLEEPING_MEMBER:
1805 	case IGMP_LAZY_MEMBER:
1806 	case IGMP_AWAKENING_MEMBER:
1807 	case IGMP_IDLE_MEMBER:
1808 		break;
1809 	case IGMP_G_QUERY_PENDING_MEMBER:
1810 	case IGMP_SG_QUERY_PENDING_MEMBER:
1811 		/*
1812 		 * Respond to a previously pending Group-Specific
1813 		 * or Group-and-Source-Specific query by enqueueing
1814 		 * the appropriate Current-State report for
1815 		 * immediate transmission.
1816 		 */
1817 		if (query_response_timer_expired) {
1818 			int retval;
1819 
1820 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1821 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1822 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1823 			    __func__, retval);
1824 			inm->inm_state = IGMP_REPORTING_MEMBER;
1825 			/* XXX Clear recorded sources for next time. */
1826 			inm_clear_recorded(inm);
1827 		}
1828 		/* FALLTHROUGH */
1829 	case IGMP_REPORTING_MEMBER:
1830 	case IGMP_LEAVING_MEMBER:
1831 		if (state_change_retransmit_timer_expired) {
1832 			/*
1833 			 * State-change retransmission timer fired.
1834 			 * If there are any further pending retransmissions,
1835 			 * set the global pending state-change flag, and
1836 			 * reset the timer.
1837 			 */
1838 			if (--inm->inm_scrv > 0) {
1839 				inm->inm_sctimer = uri_fasthz;
1840 				V_state_change_timers_running = 1;
1841 			}
1842 			/*
1843 			 * Retransmit the previously computed state-change
1844 			 * report. If there are no further pending
1845 			 * retransmissions, the mbuf queue will be consumed.
1846 			 * Update T0 state to T1 as we have now sent
1847 			 * a state-change.
1848 			 */
1849 			(void)igmp_v3_merge_state_changes(inm, scq);
1850 
1851 			inm_commit(inm);
1852 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1853 			    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1854 
1855 			/*
1856 			 * If we are leaving the group for good, make sure
1857 			 * we release IGMP's reference to it.
1858 			 * This release must be deferred using a SLIST,
1859 			 * as we are called from a loop which traverses
1860 			 * the in_ifmultiaddr TAILQ.
1861 			 */
1862 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1863 			    inm->inm_scrv == 0) {
1864 				inm->inm_state = IGMP_NOT_MEMBER;
1865 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1866 				    inm, inm_nrele);
1867 			}
1868 		}
1869 		break;
1870 	}
1871 }
1872 
1873 
1874 /*
1875  * Suppress a group's pending response to a group or source/group query.
1876  *
1877  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1878  * Do NOT update ST1/ST0 as this operation merely suppresses
1879  * the currently pending group record.
1880  * Do NOT suppress the response to a general query. It is possible but
1881  * it would require adding another state or flag.
1882  */
1883 static void
1884 igmp_v3_suppress_group_record(struct in_multi *inm)
1885 {
1886 
1887 	IN_MULTI_LOCK_ASSERT();
1888 
1889 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1890 		("%s: not IGMPv3 mode on link", __func__));
1891 
1892 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1893 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1894 		return;
1895 
1896 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1897 		inm_clear_recorded(inm);
1898 
1899 	inm->inm_timer = 0;
1900 	inm->inm_state = IGMP_REPORTING_MEMBER;
1901 }
1902 
1903 /*
1904  * Switch to a different IGMP version on the given interface,
1905  * as per Section 7.2.1.
1906  */
1907 static void
1908 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1909 {
1910 
1911 	IGMP_LOCK_ASSERT();
1912 
1913 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1914 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1915 
1916 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1917 		int old_version_timer;
1918 		/*
1919 		 * Compute the "Older Version Querier Present" timer as per
1920 		 * Section 8.12.
1921 		 */
1922 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1923 		old_version_timer *= PR_SLOWHZ;
1924 
1925 		if (version == IGMP_VERSION_1) {
1926 			igi->igi_v1_timer = old_version_timer;
1927 			igi->igi_v2_timer = 0;
1928 		} else if (version == IGMP_VERSION_2) {
1929 			igi->igi_v1_timer = 0;
1930 			igi->igi_v2_timer = old_version_timer;
1931 		}
1932 	}
1933 
1934 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1935 		if (igi->igi_version != IGMP_VERSION_2) {
1936 			igi->igi_version = IGMP_VERSION_2;
1937 			igmp_v3_cancel_link_timers(igi);
1938 		}
1939 	} else if (igi->igi_v1_timer > 0) {
1940 		if (igi->igi_version != IGMP_VERSION_1) {
1941 			igi->igi_version = IGMP_VERSION_1;
1942 			igmp_v3_cancel_link_timers(igi);
1943 		}
1944 	}
1945 }
1946 
1947 /*
1948  * Cancel pending IGMPv3 timers for the given link and all groups
1949  * joined on it; state-change, general-query, and group-query timers.
1950  */
1951 static void
1952 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
1953 {
1954 	INIT_VNET_INET(curvnet);
1955 	struct ifmultiaddr	*ifma;
1956 	struct ifnet		*ifp;
1957 	struct in_multi		*inm;
1958 
1959 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
1960 	    igi->igi_ifp, igi->igi_ifp->if_xname);
1961 
1962 	IN_MULTI_LOCK_ASSERT();
1963 	IGMP_LOCK_ASSERT();
1964 
1965 	/*
1966 	 * Fast-track this potentially expensive operation
1967 	 * by checking all the global 'timer pending' flags.
1968 	 */
1969 	if (!V_interface_timers_running &&
1970 	    !V_state_change_timers_running &&
1971 	    !V_current_state_timers_running)
1972 		return;
1973 
1974 	igi->igi_v3_timer = 0;
1975 
1976 	ifp = igi->igi_ifp;
1977 
1978 	IF_ADDR_LOCK(ifp);
1979 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1980 		if (ifma->ifma_addr->sa_family != AF_INET)
1981 			continue;
1982 		inm = (struct in_multi *)ifma->ifma_protospec;
1983 		switch (inm->inm_state) {
1984 		case IGMP_NOT_MEMBER:
1985 		case IGMP_SILENT_MEMBER:
1986 		case IGMP_IDLE_MEMBER:
1987 		case IGMP_LAZY_MEMBER:
1988 		case IGMP_SLEEPING_MEMBER:
1989 		case IGMP_AWAKENING_MEMBER:
1990 			break;
1991 		case IGMP_LEAVING_MEMBER:
1992 			/*
1993 			 * If we are leaving the group and switching
1994 			 * IGMP version, we need to release the final
1995 			 * reference held for issuing the INCLUDE {}.
1996 			 *
1997 			 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
1998 			 * around inm_release_locked(), as it is not
1999 			 * a recursive mutex.
2000 			 */
2001 			IF_ADDR_UNLOCK(ifp);
2002 			inm_release_locked(inm);
2003 			IF_ADDR_LOCK(ifp);
2004 			/* FALLTHROUGH */
2005 		case IGMP_G_QUERY_PENDING_MEMBER:
2006 		case IGMP_SG_QUERY_PENDING_MEMBER:
2007 			inm_clear_recorded(inm);
2008 			/* FALLTHROUGH */
2009 		case IGMP_REPORTING_MEMBER:
2010 			inm->inm_sctimer = 0;
2011 			inm->inm_timer = 0;
2012 			inm->inm_state = IGMP_REPORTING_MEMBER;
2013 			/*
2014 			 * Free any pending IGMPv3 state-change records.
2015 			 */
2016 			_IF_DRAIN(&inm->inm_scq);
2017 			break;
2018 		}
2019 	}
2020 	IF_ADDR_UNLOCK(ifp);
2021 }
2022 
2023 /*
2024  * Update the Older Version Querier Present timers for a link.
2025  * See Section 7.2.1 of RFC 3376.
2026  */
2027 static void
2028 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2029 {
2030 	INIT_VNET_INET(curvnet);
2031 
2032 	IGMP_LOCK_ASSERT();
2033 
2034 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2035 		/*
2036 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2037 		 *
2038 		 * Revert to IGMPv3.
2039 		 */
2040 		if (igi->igi_version != IGMP_VERSION_3) {
2041 			CTR5(KTR_IGMPV3,
2042 			    "%s: transition from v%d -> v%d on %p(%s)",
2043 			    __func__, igi->igi_version, IGMP_VERSION_3,
2044 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2045 			igi->igi_version = IGMP_VERSION_3;
2046 		}
2047 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2048 		/*
2049 		 * IGMPv1 Querier Present timer expired,
2050 		 * IGMPv2 Querier Present timer running.
2051 		 * If IGMPv2 was disabled since last timeout,
2052 		 * revert to IGMPv3.
2053 		 * If IGMPv2 is enabled, revert to IGMPv2.
2054 		 */
2055 		if (!V_igmp_v2enable) {
2056 			CTR5(KTR_IGMPV3,
2057 			    "%s: transition from v%d -> v%d on %p(%s)",
2058 			    __func__, igi->igi_version, IGMP_VERSION_3,
2059 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2060 			igi->igi_v2_timer = 0;
2061 			igi->igi_version = IGMP_VERSION_3;
2062 		} else {
2063 			--igi->igi_v2_timer;
2064 			if (igi->igi_version != IGMP_VERSION_2) {
2065 				CTR5(KTR_IGMPV3,
2066 				    "%s: transition from v%d -> v%d on %p(%s)",
2067 				    __func__, igi->igi_version, IGMP_VERSION_2,
2068 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2069 				igi->igi_version = IGMP_VERSION_2;
2070 			}
2071 		}
2072 	} else if (igi->igi_v1_timer > 0) {
2073 		/*
2074 		 * IGMPv1 Querier Present timer running.
2075 		 * Stop IGMPv2 timer if running.
2076 		 *
2077 		 * If IGMPv1 was disabled since last timeout,
2078 		 * revert to IGMPv3.
2079 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2080 		 */
2081 		if (!V_igmp_v1enable) {
2082 			CTR5(KTR_IGMPV3,
2083 			    "%s: transition from v%d -> v%d on %p(%s)",
2084 			    __func__, igi->igi_version, IGMP_VERSION_3,
2085 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2086 			igi->igi_v1_timer = 0;
2087 			igi->igi_version = IGMP_VERSION_3;
2088 		} else {
2089 			--igi->igi_v1_timer;
2090 		}
2091 		if (igi->igi_v2_timer > 0) {
2092 			CTR3(KTR_IGMPV3,
2093 			    "%s: cancel v2 timer on %p(%s)",
2094 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2095 			igi->igi_v2_timer = 0;
2096 		}
2097 	}
2098 }
2099 
2100 /*
2101  * Global slowtimo handler.
2102  * VIMAGE: Timeout handlers are expected to service all vimages.
2103  */
2104 void
2105 igmp_slowtimo(void)
2106 {
2107 	VNET_ITERATOR_DECL(vnet_iter);
2108 
2109 	VNET_LIST_RLOCK();
2110 	VNET_FOREACH(vnet_iter) {
2111 		CURVNET_SET(vnet_iter);
2112 		igmp_slowtimo_vnet();
2113 		CURVNET_RESTORE();
2114 	}
2115 	VNET_LIST_RUNLOCK();
2116 }
2117 
2118 /*
2119  * Per-vnet slowtimo handler.
2120  */
2121 static void
2122 igmp_slowtimo_vnet(void)
2123 {
2124 	INIT_VNET_INET(curvnet);
2125 	struct igmp_ifinfo *igi;
2126 
2127 	IGMP_LOCK();
2128 
2129 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2130 		igmp_v1v2_process_querier_timers(igi);
2131 	}
2132 
2133 	IGMP_UNLOCK();
2134 }
2135 
2136 /*
2137  * Dispatch an IGMPv1/v2 host report or leave message.
2138  * These are always small enough to fit inside a single mbuf.
2139  */
2140 static int
2141 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2142 {
2143 	struct ifnet		*ifp;
2144 	struct igmp		*igmp;
2145 	struct ip		*ip;
2146 	struct mbuf		*m;
2147 
2148 	IN_MULTI_LOCK_ASSERT();
2149 	IGMP_LOCK_ASSERT();
2150 
2151 	ifp = inm->inm_ifp;
2152 
2153 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2154 	if (m == NULL)
2155 		return (ENOMEM);
2156 	MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2157 
2158 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2159 
2160 	m->m_data += sizeof(struct ip);
2161 	m->m_len = sizeof(struct igmp);
2162 
2163 	igmp = mtod(m, struct igmp *);
2164 	igmp->igmp_type = type;
2165 	igmp->igmp_code = 0;
2166 	igmp->igmp_group = inm->inm_addr;
2167 	igmp->igmp_cksum = 0;
2168 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2169 
2170 	m->m_data -= sizeof(struct ip);
2171 	m->m_len += sizeof(struct ip);
2172 
2173 	ip = mtod(m, struct ip *);
2174 	ip->ip_tos = 0;
2175 	ip->ip_len = sizeof(struct ip) + sizeof(struct igmp);
2176 	ip->ip_off = 0;
2177 	ip->ip_p = IPPROTO_IGMP;
2178 	ip->ip_src.s_addr = INADDR_ANY;
2179 
2180 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2181 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2182 	else
2183 		ip->ip_dst = inm->inm_addr;
2184 
2185 	igmp_save_context(m, ifp);
2186 
2187 	m->m_flags |= M_IGMPV2;
2188 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2189 		m->m_flags |= M_IGMP_LOOP;
2190 
2191 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2192 	netisr_dispatch(NETISR_IGMP, m);
2193 
2194 	return (0);
2195 }
2196 
2197 /*
2198  * Process a state change from the upper layer for the given IPv4 group.
2199  *
2200  * Each socket holds a reference on the in_multi in its own ip_moptions.
2201  * The socket layer will have made the necessary updates to.the group
2202  * state, it is now up to IGMP to issue a state change report if there
2203  * has been any change between T0 (when the last state-change was issued)
2204  * and T1 (now).
2205  *
2206  * We use the IGMPv3 state machine at group level. The IGMP module
2207  * however makes the decision as to which IGMP protocol version to speak.
2208  * A state change *from* INCLUDE {} always means an initial join.
2209  * A state change *to* INCLUDE {} always means a final leave.
2210  *
2211  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2212  * save ourselves a bunch of work; any exclusive mode groups need not
2213  * compute source filter lists.
2214  *
2215  * VIMAGE: curvnet should have been set by caller, as this routine
2216  * is called from the socket option handlers.
2217  */
2218 int
2219 igmp_change_state(struct in_multi *inm)
2220 {
2221 	struct igmp_ifinfo *igi;
2222 	struct ifnet *ifp;
2223 	int error;
2224 
2225 	IN_MULTI_LOCK_ASSERT();
2226 
2227 	error = 0;
2228 
2229 	/*
2230 	 * Try to detect if the upper layer just asked us to change state
2231 	 * for an interface which has now gone away.
2232 	 */
2233 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2234 	ifp = inm->inm_ifma->ifma_ifp;
2235 	if (ifp != NULL) {
2236 		/*
2237 		 * Sanity check that netinet's notion of ifp is the
2238 		 * same as net's.
2239 		 */
2240 		KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2241 	}
2242 
2243 	IGMP_LOCK();
2244 
2245 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2246 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2247 
2248 	/*
2249 	 * If we detect a state transition to or from MCAST_UNDEFINED
2250 	 * for this group, then we are starting or finishing an IGMP
2251 	 * life cycle for this group.
2252 	 */
2253 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2254 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2255 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2256 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2257 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2258 			error = igmp_initial_join(inm, igi);
2259 			goto out_locked;
2260 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2261 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2262 			igmp_final_leave(inm, igi);
2263 			goto out_locked;
2264 		}
2265 	} else {
2266 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2267 	}
2268 
2269 	error = igmp_handle_state_change(inm, igi);
2270 
2271 out_locked:
2272 	IGMP_UNLOCK();
2273 	return (error);
2274 }
2275 
2276 /*
2277  * Perform the initial join for an IGMP group.
2278  *
2279  * When joining a group:
2280  *  If the group should have its IGMP traffic suppressed, do nothing.
2281  *  IGMPv1 starts sending IGMPv1 host membership reports.
2282  *  IGMPv2 starts sending IGMPv2 host membership reports.
2283  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2284  *  initial state of the membership.
2285  */
2286 static int
2287 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2288 {
2289 	INIT_VNET_INET(curvnet);
2290 	struct ifnet		*ifp;
2291 	struct ifqueue		*ifq;
2292 	int			 error, retval, syncstates;
2293 
2294 	CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2295 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2296 	    inm->inm_ifp->if_xname);
2297 
2298 	error = 0;
2299 	syncstates = 1;
2300 
2301 	ifp = inm->inm_ifp;
2302 
2303 	IN_MULTI_LOCK_ASSERT();
2304 	IGMP_LOCK_ASSERT();
2305 
2306 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2307 
2308 	/*
2309 	 * Groups joined on loopback or marked as 'not reported',
2310 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2311 	 * are never reported in any IGMP protocol exchanges.
2312 	 * All other groups enter the appropriate IGMP state machine
2313 	 * for the version in use on this link.
2314 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2315 	 * disabled for the link.
2316 	 */
2317 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2318 	    (igi->igi_flags & IGIF_SILENT) ||
2319 	    !igmp_isgroupreported(inm->inm_addr)) {
2320 		CTR1(KTR_IGMPV3,
2321 "%s: not kicking state machine for silent group", __func__);
2322 		inm->inm_state = IGMP_SILENT_MEMBER;
2323 		inm->inm_timer = 0;
2324 	} else {
2325 		/*
2326 		 * Deal with overlapping in_multi lifecycle.
2327 		 * If this group was LEAVING, then make sure
2328 		 * we drop the reference we picked up to keep the
2329 		 * group around for the final INCLUDE {} enqueue.
2330 		 */
2331 		if (igi->igi_version == IGMP_VERSION_3 &&
2332 		    inm->inm_state == IGMP_LEAVING_MEMBER)
2333 			inm_release_locked(inm);
2334 
2335 		inm->inm_state = IGMP_REPORTING_MEMBER;
2336 
2337 		switch (igi->igi_version) {
2338 		case IGMP_VERSION_1:
2339 		case IGMP_VERSION_2:
2340 			inm->inm_state = IGMP_IDLE_MEMBER;
2341 			error = igmp_v1v2_queue_report(inm,
2342 			    (igi->igi_version == IGMP_VERSION_2) ?
2343 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2344 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2345 			if (error == 0) {
2346 				inm->inm_timer = IGMP_RANDOM_DELAY(
2347 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2348 				V_current_state_timers_running = 1;
2349 			}
2350 			break;
2351 
2352 		case IGMP_VERSION_3:
2353 			/*
2354 			 * Defer update of T0 to T1, until the first copy
2355 			 * of the state change has been transmitted.
2356 			 */
2357 			syncstates = 0;
2358 
2359 			/*
2360 			 * Immediately enqueue a State-Change Report for
2361 			 * this interface, freeing any previous reports.
2362 			 * Don't kick the timers if there is nothing to do,
2363 			 * or if an error occurred.
2364 			 */
2365 			ifq = &inm->inm_scq;
2366 			_IF_DRAIN(ifq);
2367 			retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2368 			    0, 0);
2369 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2370 			    __func__, retval);
2371 			if (retval <= 0) {
2372 				error = retval * -1;
2373 				break;
2374 			}
2375 
2376 			/*
2377 			 * Schedule transmission of pending state-change
2378 			 * report up to RV times for this link. The timer
2379 			 * will fire at the next igmp_fasttimo (~200ms),
2380 			 * giving us an opportunity to merge the reports.
2381 			 */
2382 			if (igi->igi_flags & IGIF_LOOPBACK) {
2383 				inm->inm_scrv = 1;
2384 			} else {
2385 				KASSERT(igi->igi_rv > 1,
2386 				   ("%s: invalid robustness %d", __func__,
2387 				    igi->igi_rv));
2388 				inm->inm_scrv = igi->igi_rv;
2389 			}
2390 			inm->inm_sctimer = 1;
2391 			V_state_change_timers_running = 1;
2392 
2393 			error = 0;
2394 			break;
2395 		}
2396 	}
2397 
2398 	/*
2399 	 * Only update the T0 state if state change is atomic,
2400 	 * i.e. we don't need to wait for a timer to fire before we
2401 	 * can consider the state change to have been communicated.
2402 	 */
2403 	if (syncstates) {
2404 		inm_commit(inm);
2405 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2406 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2407 	}
2408 
2409 	return (error);
2410 }
2411 
2412 /*
2413  * Issue an intermediate state change during the IGMP life-cycle.
2414  */
2415 static int
2416 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2417 {
2418 	INIT_VNET_INET(curvnet);
2419 	struct ifnet		*ifp;
2420 	int			 retval;
2421 
2422 	CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2423 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2424 	    inm->inm_ifp->if_xname);
2425 
2426 	ifp = inm->inm_ifp;
2427 
2428 	IN_MULTI_LOCK_ASSERT();
2429 	IGMP_LOCK_ASSERT();
2430 
2431 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2432 
2433 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2434 	    (igi->igi_flags & IGIF_SILENT) ||
2435 	    !igmp_isgroupreported(inm->inm_addr) ||
2436 	    (igi->igi_version != IGMP_VERSION_3)) {
2437 		if (!igmp_isgroupreported(inm->inm_addr)) {
2438 			CTR1(KTR_IGMPV3,
2439 "%s: not kicking state machine for silent group", __func__);
2440 		}
2441 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2442 		inm_commit(inm);
2443 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2444 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2445 		return (0);
2446 	}
2447 
2448 	_IF_DRAIN(&inm->inm_scq);
2449 
2450 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2451 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2452 	if (retval <= 0)
2453 		return (-retval);
2454 
2455 	/*
2456 	 * If record(s) were enqueued, start the state-change
2457 	 * report timer for this group.
2458 	 */
2459 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2460 	inm->inm_sctimer = 1;
2461 	V_state_change_timers_running = 1;
2462 
2463 	return (0);
2464 }
2465 
2466 /*
2467  * Perform the final leave for an IGMP group.
2468  *
2469  * When leaving a group:
2470  *  IGMPv1 does nothing.
2471  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2472  *  IGMPv3 enqueues a state-change report containing a transition
2473  *  to INCLUDE {} for immediate transmission.
2474  */
2475 static void
2476 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2477 {
2478 	INIT_VNET_INET(curvnet);
2479 	int syncstates;
2480 
2481 	syncstates = 1;
2482 
2483 	CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2484 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2485 	    inm->inm_ifp->if_xname);
2486 
2487 	IN_MULTI_LOCK_ASSERT();
2488 	IGMP_LOCK_ASSERT();
2489 
2490 	switch (inm->inm_state) {
2491 	case IGMP_NOT_MEMBER:
2492 	case IGMP_SILENT_MEMBER:
2493 	case IGMP_LEAVING_MEMBER:
2494 		/* Already leaving or left; do nothing. */
2495 		CTR1(KTR_IGMPV3,
2496 "%s: not kicking state machine for silent group", __func__);
2497 		break;
2498 	case IGMP_REPORTING_MEMBER:
2499 	case IGMP_IDLE_MEMBER:
2500 	case IGMP_G_QUERY_PENDING_MEMBER:
2501 	case IGMP_SG_QUERY_PENDING_MEMBER:
2502 		if (igi->igi_version == IGMP_VERSION_2) {
2503 #ifdef INVARIANTS
2504 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2505 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2506 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2507 			     __func__);
2508 #endif
2509 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2510 			inm->inm_state = IGMP_NOT_MEMBER;
2511 		} else if (igi->igi_version == IGMP_VERSION_3) {
2512 			/*
2513 			 * Stop group timer and all pending reports.
2514 			 * Immediately enqueue a state-change report
2515 			 * TO_IN {} to be sent on the next fast timeout,
2516 			 * giving us an opportunity to merge reports.
2517 			 */
2518 			_IF_DRAIN(&inm->inm_scq);
2519 			inm->inm_timer = 0;
2520 			if (igi->igi_flags & IGIF_LOOPBACK) {
2521 				inm->inm_scrv = 1;
2522 			} else {
2523 				inm->inm_scrv = igi->igi_rv;
2524 			}
2525 			CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2526 			    "pending retransmissions.", __func__,
2527 			    inet_ntoa(inm->inm_addr),
2528 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2529 			if (inm->inm_scrv == 0) {
2530 				inm->inm_state = IGMP_NOT_MEMBER;
2531 				inm->inm_sctimer = 0;
2532 			} else {
2533 				int retval;
2534 
2535 				inm_acquire_locked(inm);
2536 
2537 				retval = igmp_v3_enqueue_group_record(
2538 				    &inm->inm_scq, inm, 1, 0, 0);
2539 				KASSERT(retval != 0,
2540 				    ("%s: enqueue record = %d", __func__,
2541 				     retval));
2542 
2543 				inm->inm_state = IGMP_LEAVING_MEMBER;
2544 				inm->inm_sctimer = 1;
2545 				V_state_change_timers_running = 1;
2546 				syncstates = 0;
2547 			}
2548 			break;
2549 		}
2550 		break;
2551 	case IGMP_LAZY_MEMBER:
2552 	case IGMP_SLEEPING_MEMBER:
2553 	case IGMP_AWAKENING_MEMBER:
2554 		/* Our reports are suppressed; do nothing. */
2555 		break;
2556 	}
2557 
2558 	if (syncstates) {
2559 		inm_commit(inm);
2560 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2561 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2562 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2563 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2564 		    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2565 	}
2566 }
2567 
2568 /*
2569  * Enqueue an IGMPv3 group record to the given output queue.
2570  *
2571  * XXX This function could do with having the allocation code
2572  * split out, and the multiple-tree-walks coalesced into a single
2573  * routine as has been done in igmp_v3_enqueue_filter_change().
2574  *
2575  * If is_state_change is zero, a current-state record is appended.
2576  * If is_state_change is non-zero, a state-change report is appended.
2577  *
2578  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2579  * If is_group_query is zero, and if there is a packet with free space
2580  * at the tail of the queue, it will be appended to providing there
2581  * is enough free space.
2582  * Otherwise a new mbuf packet chain is allocated.
2583  *
2584  * If is_source_query is non-zero, each source is checked to see if
2585  * it was recorded for a Group-Source query, and will be omitted if
2586  * it is not both in-mode and recorded.
2587  *
2588  * The function will attempt to allocate leading space in the packet
2589  * for the IP/IGMP header to be prepended without fragmenting the chain.
2590  *
2591  * If successful the size of all data appended to the queue is returned,
2592  * otherwise an error code less than zero is returned, or zero if
2593  * no record(s) were appended.
2594  */
2595 static int
2596 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2597     const int is_state_change, const int is_group_query,
2598     const int is_source_query)
2599 {
2600 	struct igmp_grouprec	 ig;
2601 	struct igmp_grouprec	*pig;
2602 	struct ifnet		*ifp;
2603 	struct ip_msource	*ims, *nims;
2604 	struct mbuf		*m0, *m, *md;
2605 	int			 error, is_filter_list_change;
2606 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2607 	int			 record_has_sources;
2608 	int			 now;
2609 	int			 type;
2610 	in_addr_t		 naddr;
2611 	uint8_t			 mode;
2612 
2613 	IN_MULTI_LOCK_ASSERT();
2614 
2615 	error = 0;
2616 	ifp = inm->inm_ifp;
2617 	is_filter_list_change = 0;
2618 	m = NULL;
2619 	m0 = NULL;
2620 	m0srcs = 0;
2621 	msrcs = 0;
2622 	nbytes = 0;
2623 	nims = NULL;
2624 	record_has_sources = 1;
2625 	pig = NULL;
2626 	type = IGMP_DO_NOTHING;
2627 	mode = inm->inm_st[1].iss_fmode;
2628 
2629 	/*
2630 	 * If we did not transition out of ASM mode during t0->t1,
2631 	 * and there are no source nodes to process, we can skip
2632 	 * the generation of source records.
2633 	 */
2634 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2635 	    inm->inm_nsrc == 0)
2636 		record_has_sources = 0;
2637 
2638 	if (is_state_change) {
2639 		/*
2640 		 * Queue a state change record.
2641 		 * If the mode did not change, and there are non-ASM
2642 		 * listeners or source filters present,
2643 		 * we potentially need to issue two records for the group.
2644 		 * If we are transitioning to MCAST_UNDEFINED, we need
2645 		 * not send any sources.
2646 		 * If there are ASM listeners, and there was no filter
2647 		 * mode transition of any kind, do nothing.
2648 		 */
2649 		if (mode != inm->inm_st[0].iss_fmode) {
2650 			if (mode == MCAST_EXCLUDE) {
2651 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2652 				    __func__);
2653 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2654 			} else {
2655 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2656 				    __func__);
2657 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2658 				if (mode == MCAST_UNDEFINED)
2659 					record_has_sources = 0;
2660 			}
2661 		} else {
2662 			if (record_has_sources) {
2663 				is_filter_list_change = 1;
2664 			} else {
2665 				type = IGMP_DO_NOTHING;
2666 			}
2667 		}
2668 	} else {
2669 		/*
2670 		 * Queue a current state record.
2671 		 */
2672 		if (mode == MCAST_EXCLUDE) {
2673 			type = IGMP_MODE_IS_EXCLUDE;
2674 		} else if (mode == MCAST_INCLUDE) {
2675 			type = IGMP_MODE_IS_INCLUDE;
2676 			KASSERT(inm->inm_st[1].iss_asm == 0,
2677 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2678 			     __func__, inm, inm->inm_st[1].iss_asm));
2679 		}
2680 	}
2681 
2682 	/*
2683 	 * Generate the filter list changes using a separate function.
2684 	 */
2685 	if (is_filter_list_change)
2686 		return (igmp_v3_enqueue_filter_change(ifq, inm));
2687 
2688 	if (type == IGMP_DO_NOTHING) {
2689 		CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2690 		    __func__, inet_ntoa(inm->inm_addr),
2691 		    inm->inm_ifp->if_xname);
2692 		return (0);
2693 	}
2694 
2695 	/*
2696 	 * If any sources are present, we must be able to fit at least
2697 	 * one in the trailing space of the tail packet's mbuf,
2698 	 * ideally more.
2699 	 */
2700 	minrec0len = sizeof(struct igmp_grouprec);
2701 	if (record_has_sources)
2702 		minrec0len += sizeof(in_addr_t);
2703 
2704 	CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2705 	    igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2706 	    inm->inm_ifp->if_xname);
2707 
2708 	/*
2709 	 * Check if we have a packet in the tail of the queue for this
2710 	 * group into which the first group record for this group will fit.
2711 	 * Otherwise allocate a new packet.
2712 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2713 	 * Note: Group records for G/GSR query responses MUST be sent
2714 	 * in their own packet.
2715 	 */
2716 	m0 = ifq->ifq_tail;
2717 	if (!is_group_query &&
2718 	    m0 != NULL &&
2719 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2720 	    (m0->m_pkthdr.len + minrec0len) <
2721 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2722 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2723 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2724 		m = m0;
2725 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2726 	} else {
2727 		if (_IF_QFULL(ifq)) {
2728 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2729 			return (-ENOMEM);
2730 		}
2731 		m = NULL;
2732 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2733 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2734 		if (!is_state_change && !is_group_query) {
2735 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2736 			if (m)
2737 				m->m_data += IGMP_LEADINGSPACE;
2738 		}
2739 		if (m == NULL) {
2740 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2741 			if (m)
2742 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2743 		}
2744 		if (m == NULL)
2745 			return (-ENOMEM);
2746 
2747 		igmp_save_context(m, ifp);
2748 
2749 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2750 	}
2751 
2752 	/*
2753 	 * Append group record.
2754 	 * If we have sources, we don't know how many yet.
2755 	 */
2756 	ig.ig_type = type;
2757 	ig.ig_datalen = 0;
2758 	ig.ig_numsrc = 0;
2759 	ig.ig_group = inm->inm_addr;
2760 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2761 		if (m != m0)
2762 			m_freem(m);
2763 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2764 		return (-ENOMEM);
2765 	}
2766 	nbytes += sizeof(struct igmp_grouprec);
2767 
2768 	/*
2769 	 * Append as many sources as will fit in the first packet.
2770 	 * If we are appending to a new packet, the chain allocation
2771 	 * may potentially use clusters; use m_getptr() in this case.
2772 	 * If we are appending to an existing packet, we need to obtain
2773 	 * a pointer to the group record after m_append(), in case a new
2774 	 * mbuf was allocated.
2775 	 * Only append sources which are in-mode at t1. If we are
2776 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2777 	 * include source entries.
2778 	 * Only report recorded sources in our filter set when responding
2779 	 * to a group-source query.
2780 	 */
2781 	if (record_has_sources) {
2782 		if (m == m0) {
2783 			md = m_last(m);
2784 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2785 			    md->m_len - nbytes);
2786 		} else {
2787 			md = m_getptr(m, 0, &off);
2788 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2789 			    off);
2790 		}
2791 		msrcs = 0;
2792 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2793 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2794 			    inet_ntoa_haddr(ims->ims_haddr));
2795 			now = ims_get_mode(inm, ims, 1);
2796 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2797 			if ((now != mode) ||
2798 			    (now == mode && mode == MCAST_UNDEFINED)) {
2799 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2800 				continue;
2801 			}
2802 			if (is_source_query && ims->ims_stp == 0) {
2803 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2804 				    __func__);
2805 				continue;
2806 			}
2807 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2808 			naddr = htonl(ims->ims_haddr);
2809 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2810 				if (m != m0)
2811 					m_freem(m);
2812 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2813 				    __func__);
2814 				return (-ENOMEM);
2815 			}
2816 			nbytes += sizeof(in_addr_t);
2817 			++msrcs;
2818 			if (msrcs == m0srcs)
2819 				break;
2820 		}
2821 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2822 		    msrcs);
2823 		pig->ig_numsrc = htons(msrcs);
2824 		nbytes += (msrcs * sizeof(in_addr_t));
2825 	}
2826 
2827 	if (is_source_query && msrcs == 0) {
2828 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2829 		if (m != m0)
2830 			m_freem(m);
2831 		return (0);
2832 	}
2833 
2834 	/*
2835 	 * We are good to go with first packet.
2836 	 */
2837 	if (m != m0) {
2838 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2839 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2840 		_IF_ENQUEUE(ifq, m);
2841 	} else
2842 		m->m_pkthdr.PH_vt.vt_nrecs++;
2843 
2844 	/*
2845 	 * No further work needed if no source list in packet(s).
2846 	 */
2847 	if (!record_has_sources)
2848 		return (nbytes);
2849 
2850 	/*
2851 	 * Whilst sources remain to be announced, we need to allocate
2852 	 * a new packet and fill out as many sources as will fit.
2853 	 * Always try for a cluster first.
2854 	 */
2855 	while (nims != NULL) {
2856 		if (_IF_QFULL(ifq)) {
2857 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2858 			return (-ENOMEM);
2859 		}
2860 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2861 		if (m)
2862 			m->m_data += IGMP_LEADINGSPACE;
2863 		if (m == NULL) {
2864 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2865 			if (m)
2866 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2867 		}
2868 		if (m == NULL)
2869 			return (-ENOMEM);
2870 		igmp_save_context(m, ifp);
2871 		md = m_getptr(m, 0, &off);
2872 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2873 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2874 
2875 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2876 			if (m != m0)
2877 				m_freem(m);
2878 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2879 			return (-ENOMEM);
2880 		}
2881 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2882 		nbytes += sizeof(struct igmp_grouprec);
2883 
2884 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2885 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2886 
2887 		msrcs = 0;
2888 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2889 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2890 			    inet_ntoa_haddr(ims->ims_haddr));
2891 			now = ims_get_mode(inm, ims, 1);
2892 			if ((now != mode) ||
2893 			    (now == mode && mode == MCAST_UNDEFINED)) {
2894 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2895 				continue;
2896 			}
2897 			if (is_source_query && ims->ims_stp == 0) {
2898 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2899 				    __func__);
2900 				continue;
2901 			}
2902 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2903 			naddr = htonl(ims->ims_haddr);
2904 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2905 				if (m != m0)
2906 					m_freem(m);
2907 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2908 				    __func__);
2909 				return (-ENOMEM);
2910 			}
2911 			++msrcs;
2912 			if (msrcs == m0srcs)
2913 				break;
2914 		}
2915 		pig->ig_numsrc = htons(msrcs);
2916 		nbytes += (msrcs * sizeof(in_addr_t));
2917 
2918 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2919 		_IF_ENQUEUE(ifq, m);
2920 	}
2921 
2922 	return (nbytes);
2923 }
2924 
2925 /*
2926  * Type used to mark record pass completion.
2927  * We exploit the fact we can cast to this easily from the
2928  * current filter modes on each ip_msource node.
2929  */
2930 typedef enum {
2931 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2932 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2933 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2934 	REC_FULL = REC_ALLOW | REC_BLOCK
2935 } rectype_t;
2936 
2937 /*
2938  * Enqueue an IGMPv3 filter list change to the given output queue.
2939  *
2940  * Source list filter state is held in an RB-tree. When the filter list
2941  * for a group is changed without changing its mode, we need to compute
2942  * the deltas between T0 and T1 for each source in the filter set,
2943  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2944  *
2945  * As we may potentially queue two record types, and the entire R-B tree
2946  * needs to be walked at once, we break this out into its own function
2947  * so we can generate a tightly packed queue of packets.
2948  *
2949  * XXX This could be written to only use one tree walk, although that makes
2950  * serializing into the mbuf chains a bit harder. For now we do two walks
2951  * which makes things easier on us, and it may or may not be harder on
2952  * the L2 cache.
2953  *
2954  * If successful the size of all data appended to the queue is returned,
2955  * otherwise an error code less than zero is returned, or zero if
2956  * no record(s) were appended.
2957  */
2958 static int
2959 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
2960 {
2961 	static const int MINRECLEN =
2962 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
2963 	struct ifnet		*ifp;
2964 	struct igmp_grouprec	 ig;
2965 	struct igmp_grouprec	*pig;
2966 	struct ip_msource	*ims, *nims;
2967 	struct mbuf		*m, *m0, *md;
2968 	in_addr_t		 naddr;
2969 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2970 	int			 nallow, nblock;
2971 	uint8_t			 mode, now, then;
2972 	rectype_t		 crt, drt, nrt;
2973 
2974 	IN_MULTI_LOCK_ASSERT();
2975 
2976 	if (inm->inm_nsrc == 0 ||
2977 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
2978 		return (0);
2979 
2980 	ifp = inm->inm_ifp;			/* interface */
2981 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
2982 	crt = REC_NONE;	/* current group record type */
2983 	drt = REC_NONE;	/* mask of completed group record types */
2984 	nrt = REC_NONE;	/* record type for current node */
2985 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2986 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2987 	npbytes = 0;	/* # of bytes appended this packet */
2988 	rsrcs = 0;	/* # sources encoded in current record */
2989 	schanged = 0;	/* # nodes encoded in overall filter change */
2990 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2991 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2992 	nims = NULL;	/* next tree node pointer */
2993 
2994 	/*
2995 	 * For each possible filter record mode.
2996 	 * The first kind of source we encounter tells us which
2997 	 * is the first kind of record we start appending.
2998 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2999 	 * as the inverse of the group's filter mode.
3000 	 */
3001 	while (drt != REC_FULL) {
3002 		do {
3003 			m0 = ifq->ifq_tail;
3004 			if (m0 != NULL &&
3005 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3006 			     IGMP_V3_REPORT_MAXRECS) &&
3007 			    (m0->m_pkthdr.len + MINRECLEN) <
3008 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3009 				m = m0;
3010 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3011 					    sizeof(struct igmp_grouprec)) /
3012 				    sizeof(in_addr_t);
3013 				CTR1(KTR_IGMPV3,
3014 				    "%s: use previous packet", __func__);
3015 			} else {
3016 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3017 				if (m)
3018 					m->m_data += IGMP_LEADINGSPACE;
3019 				if (m == NULL) {
3020 					m = m_gethdr(M_DONTWAIT, MT_DATA);
3021 					if (m)
3022 						MH_ALIGN(m, IGMP_LEADINGSPACE);
3023 				}
3024 				if (m == NULL) {
3025 					CTR1(KTR_IGMPV3,
3026 					    "%s: m_get*() failed", __func__);
3027 					return (-ENOMEM);
3028 				}
3029 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3030 				igmp_save_context(m, ifp);
3031 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3032 				    sizeof(struct igmp_grouprec)) /
3033 				    sizeof(in_addr_t);
3034 				npbytes = 0;
3035 				CTR1(KTR_IGMPV3,
3036 				    "%s: allocated new packet", __func__);
3037 			}
3038 			/*
3039 			 * Append the IGMP group record header to the
3040 			 * current packet's data area.
3041 			 * Recalculate pointer to free space for next
3042 			 * group record, in case m_append() allocated
3043 			 * a new mbuf or cluster.
3044 			 */
3045 			memset(&ig, 0, sizeof(ig));
3046 			ig.ig_group = inm->inm_addr;
3047 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3048 				if (m != m0)
3049 					m_freem(m);
3050 				CTR1(KTR_IGMPV3,
3051 				    "%s: m_append() failed", __func__);
3052 				return (-ENOMEM);
3053 			}
3054 			npbytes += sizeof(struct igmp_grouprec);
3055 			if (m != m0) {
3056 				/* new packet; offset in c hain */
3057 				md = m_getptr(m, npbytes -
3058 				    sizeof(struct igmp_grouprec), &off);
3059 				pig = (struct igmp_grouprec *)(mtod(md,
3060 				    uint8_t *) + off);
3061 			} else {
3062 				/* current packet; offset from last append */
3063 				md = m_last(m);
3064 				pig = (struct igmp_grouprec *)(mtod(md,
3065 				    uint8_t *) + md->m_len -
3066 				    sizeof(struct igmp_grouprec));
3067 			}
3068 			/*
3069 			 * Begin walking the tree for this record type
3070 			 * pass, or continue from where we left off
3071 			 * previously if we had to allocate a new packet.
3072 			 * Only report deltas in-mode at t1.
3073 			 * We need not report included sources as allowed
3074 			 * if we are in inclusive mode on the group,
3075 			 * however the converse is not true.
3076 			 */
3077 			rsrcs = 0;
3078 			if (nims == NULL)
3079 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3080 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3081 				CTR2(KTR_IGMPV3, "%s: visit node %s",
3082 				    __func__, inet_ntoa_haddr(ims->ims_haddr));
3083 				now = ims_get_mode(inm, ims, 1);
3084 				then = ims_get_mode(inm, ims, 0);
3085 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3086 				    __func__, then, now);
3087 				if (now == then) {
3088 					CTR1(KTR_IGMPV3,
3089 					    "%s: skip unchanged", __func__);
3090 					continue;
3091 				}
3092 				if (mode == MCAST_EXCLUDE &&
3093 				    now == MCAST_INCLUDE) {
3094 					CTR1(KTR_IGMPV3,
3095 					    "%s: skip IN src on EX group",
3096 					    __func__);
3097 					continue;
3098 				}
3099 				nrt = (rectype_t)now;
3100 				if (nrt == REC_NONE)
3101 					nrt = (rectype_t)(~mode & REC_FULL);
3102 				if (schanged++ == 0) {
3103 					crt = nrt;
3104 				} else if (crt != nrt)
3105 					continue;
3106 				naddr = htonl(ims->ims_haddr);
3107 				if (!m_append(m, sizeof(in_addr_t),
3108 				    (void *)&naddr)) {
3109 					if (m != m0)
3110 						m_freem(m);
3111 					CTR1(KTR_IGMPV3,
3112 					    "%s: m_append() failed", __func__);
3113 					return (-ENOMEM);
3114 				}
3115 				nallow += !!(crt == REC_ALLOW);
3116 				nblock += !!(crt == REC_BLOCK);
3117 				if (++rsrcs == m0srcs)
3118 					break;
3119 			}
3120 			/*
3121 			 * If we did not append any tree nodes on this
3122 			 * pass, back out of allocations.
3123 			 */
3124 			if (rsrcs == 0) {
3125 				npbytes -= sizeof(struct igmp_grouprec);
3126 				if (m != m0) {
3127 					CTR1(KTR_IGMPV3,
3128 					    "%s: m_free(m)", __func__);
3129 					m_freem(m);
3130 				} else {
3131 					CTR1(KTR_IGMPV3,
3132 					    "%s: m_adj(m, -ig)", __func__);
3133 					m_adj(m, -((int)sizeof(
3134 					    struct igmp_grouprec)));
3135 				}
3136 				continue;
3137 			}
3138 			npbytes += (rsrcs * sizeof(in_addr_t));
3139 			if (crt == REC_ALLOW)
3140 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3141 			else if (crt == REC_BLOCK)
3142 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3143 			pig->ig_numsrc = htons(rsrcs);
3144 			/*
3145 			 * Count the new group record, and enqueue this
3146 			 * packet if it wasn't already queued.
3147 			 */
3148 			m->m_pkthdr.PH_vt.vt_nrecs++;
3149 			if (m != m0)
3150 				_IF_ENQUEUE(ifq, m);
3151 			nbytes += npbytes;
3152 		} while (nims != NULL);
3153 		drt |= crt;
3154 		crt = (~crt & REC_FULL);
3155 	}
3156 
3157 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3158 	    nallow, nblock);
3159 
3160 	return (nbytes);
3161 }
3162 
3163 static int
3164 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3165 {
3166 	struct ifqueue	*gq;
3167 	struct mbuf	*m;		/* pending state-change */
3168 	struct mbuf	*m0;		/* copy of pending state-change */
3169 	struct mbuf	*mt;		/* last state-change in packet */
3170 	int		 docopy, domerge;
3171 	u_int		 recslen;
3172 
3173 	docopy = 0;
3174 	domerge = 0;
3175 	recslen = 0;
3176 
3177 	IN_MULTI_LOCK_ASSERT();
3178 	IGMP_LOCK_ASSERT();
3179 
3180 	/*
3181 	 * If there are further pending retransmissions, make a writable
3182 	 * copy of each queued state-change message before merging.
3183 	 */
3184 	if (inm->inm_scrv > 0)
3185 		docopy = 1;
3186 
3187 	gq = &inm->inm_scq;
3188 #ifdef KTR
3189 	if (gq->ifq_head == NULL) {
3190 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3191 		    __func__, inm);
3192 	}
3193 #endif
3194 
3195 	m = gq->ifq_head;
3196 	while (m != NULL) {
3197 		/*
3198 		 * Only merge the report into the current packet if
3199 		 * there is sufficient space to do so; an IGMPv3 report
3200 		 * packet may only contain 65,535 group records.
3201 		 * Always use a simple mbuf chain concatentation to do this,
3202 		 * as large state changes for single groups may have
3203 		 * allocated clusters.
3204 		 */
3205 		domerge = 0;
3206 		mt = ifscq->ifq_tail;
3207 		if (mt != NULL) {
3208 			recslen = m_length(m, NULL);
3209 
3210 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3211 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3212 			    IGMP_V3_REPORT_MAXRECS) &&
3213 			    (mt->m_pkthdr.len + recslen <=
3214 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3215 				domerge = 1;
3216 		}
3217 
3218 		if (!domerge && _IF_QFULL(gq)) {
3219 			CTR2(KTR_IGMPV3,
3220 			    "%s: outbound queue full, skipping whole packet %p",
3221 			    __func__, m);
3222 			mt = m->m_nextpkt;
3223 			if (!docopy)
3224 				m_freem(m);
3225 			m = mt;
3226 			continue;
3227 		}
3228 
3229 		if (!docopy) {
3230 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3231 			_IF_DEQUEUE(gq, m0);
3232 			m = m0->m_nextpkt;
3233 		} else {
3234 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3235 			m0 = m_dup(m, M_NOWAIT);
3236 			if (m0 == NULL)
3237 				return (ENOMEM);
3238 			m0->m_nextpkt = NULL;
3239 			m = m->m_nextpkt;
3240 		}
3241 
3242 		if (!domerge) {
3243 			CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3244 			    __func__, m0, ifscq);
3245 			_IF_ENQUEUE(ifscq, m0);
3246 		} else {
3247 			struct mbuf *mtl;	/* last mbuf of packet mt */
3248 
3249 			CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3250 			    __func__, m0, mt);
3251 
3252 			mtl = m_last(mt);
3253 			m0->m_flags &= ~M_PKTHDR;
3254 			mt->m_pkthdr.len += recslen;
3255 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3256 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3257 
3258 			mtl->m_next = m0;
3259 		}
3260 	}
3261 
3262 	return (0);
3263 }
3264 
3265 /*
3266  * Respond to a pending IGMPv3 General Query.
3267  */
3268 static void
3269 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3270 {
3271 	INIT_VNET_INET(curvnet);
3272 	struct ifmultiaddr	*ifma, *tifma;
3273 	struct ifnet		*ifp;
3274 	struct in_multi		*inm;
3275 	int			 retval, loop;
3276 
3277 	IN_MULTI_LOCK_ASSERT();
3278 	IGMP_LOCK_ASSERT();
3279 
3280 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3281 	    ("%s: called when version %d", __func__, igi->igi_version));
3282 
3283 	ifp = igi->igi_ifp;
3284 
3285 	IF_ADDR_LOCK(ifp);
3286 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
3287 		if (ifma->ifma_addr->sa_family != AF_INET ||
3288 		    ifma->ifma_protospec == NULL)
3289 			continue;
3290 
3291 		inm = (struct in_multi *)ifma->ifma_protospec;
3292 		KASSERT(ifp == inm->inm_ifp,
3293 		    ("%s: inconsistent ifp", __func__));
3294 
3295 		switch (inm->inm_state) {
3296 		case IGMP_NOT_MEMBER:
3297 		case IGMP_SILENT_MEMBER:
3298 			break;
3299 		case IGMP_REPORTING_MEMBER:
3300 		case IGMP_IDLE_MEMBER:
3301 		case IGMP_LAZY_MEMBER:
3302 		case IGMP_SLEEPING_MEMBER:
3303 		case IGMP_AWAKENING_MEMBER:
3304 			inm->inm_state = IGMP_REPORTING_MEMBER;
3305 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3306 			    inm, 0, 0, 0);
3307 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3308 			    __func__, retval);
3309 			break;
3310 		case IGMP_G_QUERY_PENDING_MEMBER:
3311 		case IGMP_SG_QUERY_PENDING_MEMBER:
3312 		case IGMP_LEAVING_MEMBER:
3313 			break;
3314 		}
3315 	}
3316 	IF_ADDR_UNLOCK(ifp);
3317 
3318 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3319 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3320 
3321 	/*
3322 	 * Slew transmission of bursts over 500ms intervals.
3323 	 */
3324 	if (igi->igi_gq.ifq_head != NULL) {
3325 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3326 		    IGMP_RESPONSE_BURST_INTERVAL);
3327 		V_interface_timers_running = 1;
3328 	}
3329 }
3330 
3331 /*
3332  * Transmit the next pending IGMP message in the output queue.
3333  *
3334  * We get called from netisr_processqueue(). A mutex private to igmpoq
3335  * will be acquired and released around this routine.
3336  *
3337  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3338  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3339  * a link and uses a link-scope multicast address.
3340  */
3341 static void
3342 igmp_intr(struct mbuf *m)
3343 {
3344 	struct ip_moptions	 imo;
3345 	struct ifnet		*ifp;
3346 	struct mbuf		*ipopts, *m0;
3347 	int			 error;
3348 	uint32_t		 ifindex;
3349 
3350 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3351 
3352 	/*
3353 	 * Set VNET image pointer from enqueued mbuf chain
3354 	 * before doing anything else. Whilst we use interface
3355 	 * indexes to guard against interface detach, they are
3356 	 * unique to each VIMAGE and must be retrieved.
3357 	 */
3358 	CURVNET_SET((struct vnet *)(m->m_pkthdr.header));
3359 	INIT_VNET_NET(curvnet);
3360 	INIT_VNET_INET(curvnet);
3361 	ifindex = igmp_restore_context(m);
3362 
3363 	/*
3364 	 * Check if the ifnet still exists. This limits the scope of
3365 	 * any race in the absence of a global ifp lock for low cost
3366 	 * (an array lookup).
3367 	 */
3368 	ifp = ifnet_byindex(ifindex);
3369 	if (ifp == NULL) {
3370 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3371 		    __func__, m, ifindex);
3372 		m_freem(m);
3373 		IPSTAT_INC(ips_noroute);
3374 		goto out;
3375 	}
3376 
3377 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3378 
3379 	imo.imo_multicast_ttl  = 1;
3380 	imo.imo_multicast_vif  = -1;
3381 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3382 
3383 	/*
3384 	 * If the user requested that IGMP traffic be explicitly
3385 	 * redirected to the loopback interface (e.g. they are running a
3386 	 * MANET interface and the routing protocol needs to see the
3387 	 * updates), handle this now.
3388 	 */
3389 	if (m->m_flags & M_IGMP_LOOP)
3390 		imo.imo_multicast_ifp = V_loif;
3391 	else
3392 		imo.imo_multicast_ifp = ifp;
3393 
3394 	if (m->m_flags & M_IGMPV2) {
3395 		m0 = m;
3396 	} else {
3397 		m0 = igmp_v3_encap_report(ifp, m);
3398 		if (m0 == NULL) {
3399 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3400 			m_freem(m);
3401 			IPSTAT_INC(ips_odropped);
3402 			goto out;
3403 		}
3404 	}
3405 
3406 	igmp_scrub_context(m0);
3407 	m->m_flags &= ~(M_PROTOFLAGS);
3408 	m0->m_pkthdr.rcvif = V_loif;
3409 #ifdef MAC
3410 	mac_netinet_igmp_send(ifp, m0);
3411 #endif
3412 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3413 	if (error) {
3414 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3415 		goto out;
3416 	}
3417 
3418 	IGMPSTAT_INC(igps_snd_reports);
3419 
3420 out:
3421 	/*
3422 	 * We must restore the existing vnet pointer before
3423 	 * continuing as we are run from netisr context.
3424 	 */
3425 	CURVNET_RESTORE();
3426 }
3427 
3428 /*
3429  * Encapsulate an IGMPv3 report.
3430  *
3431  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3432  * chain has already had its IP/IGMPv3 header prepended. In this case
3433  * the function will not attempt to prepend; the lengths and checksums
3434  * will however be re-computed.
3435  *
3436  * Returns a pointer to the new mbuf chain head, or NULL if the
3437  * allocation failed.
3438  */
3439 static struct mbuf *
3440 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3441 {
3442 	INIT_VNET_INET(curvnet);
3443 	struct igmp_report	*igmp;
3444 	struct ip		*ip;
3445 	int			 hdrlen, igmpreclen;
3446 
3447 	KASSERT((m->m_flags & M_PKTHDR),
3448 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3449 
3450 	igmpreclen = m_length(m, NULL);
3451 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3452 
3453 	if (m->m_flags & M_IGMPV3_HDR) {
3454 		igmpreclen -= hdrlen;
3455 	} else {
3456 		M_PREPEND(m, hdrlen, M_DONTWAIT);
3457 		if (m == NULL)
3458 			return (NULL);
3459 		m->m_flags |= M_IGMPV3_HDR;
3460 	}
3461 
3462 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3463 
3464 	m->m_data += sizeof(struct ip);
3465 	m->m_len -= sizeof(struct ip);
3466 
3467 	igmp = mtod(m, struct igmp_report *);
3468 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3469 	igmp->ir_rsv1 = 0;
3470 	igmp->ir_rsv2 = 0;
3471 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3472 	igmp->ir_cksum = 0;
3473 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3474 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3475 
3476 	m->m_data -= sizeof(struct ip);
3477 	m->m_len += sizeof(struct ip);
3478 
3479 	ip = mtod(m, struct ip *);
3480 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3481 	ip->ip_len = hdrlen + igmpreclen;
3482 	ip->ip_off = IP_DF;
3483 	ip->ip_p = IPPROTO_IGMP;
3484 	ip->ip_sum = 0;
3485 
3486 	ip->ip_src.s_addr = INADDR_ANY;
3487 
3488 	if (m->m_flags & M_IGMP_LOOP) {
3489 		struct in_ifaddr *ia;
3490 
3491 		IFP_TO_IA(ifp, ia);
3492 		if (ia != NULL)
3493 			ip->ip_src = ia->ia_addr.sin_addr;
3494 	}
3495 
3496 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3497 
3498 	return (m);
3499 }
3500 
3501 #ifdef KTR
3502 static char *
3503 igmp_rec_type_to_str(const int type)
3504 {
3505 
3506 	switch (type) {
3507 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3508 			return "TO_EX";
3509 			break;
3510 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3511 			return "TO_IN";
3512 			break;
3513 		case IGMP_MODE_IS_EXCLUDE:
3514 			return "MODE_EX";
3515 			break;
3516 		case IGMP_MODE_IS_INCLUDE:
3517 			return "MODE_IN";
3518 			break;
3519 		case IGMP_ALLOW_NEW_SOURCES:
3520 			return "ALLOW_NEW";
3521 			break;
3522 		case IGMP_BLOCK_OLD_SOURCES:
3523 			return "BLOCK_OLD";
3524 			break;
3525 		default:
3526 			break;
3527 	}
3528 	return "unknown";
3529 }
3530 #endif
3531 
3532 static void
3533 igmp_sysinit(void)
3534 {
3535 
3536 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3537 
3538 	IGMP_LOCK_INIT();
3539 
3540 	mtx_init(&igmpoq.ifq_mtx, "igmpoq_mtx", NULL, MTX_DEF);
3541 	IFQ_SET_MAXLEN(&igmpoq, IFQ_MAXLEN);
3542 
3543 	m_raopt = igmp_ra_alloc();
3544 
3545 	netisr_register(NETISR_IGMP, igmp_intr, &igmpoq, 0);
3546 }
3547 
3548 static void
3549 igmp_sysuninit(void)
3550 {
3551 
3552 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3553 
3554 	netisr_unregister(NETISR_IGMP);
3555 	mtx_destroy(&igmpoq.ifq_mtx);
3556 
3557 	m_free(m_raopt);
3558 	m_raopt = NULL;
3559 
3560 	IGMP_LOCK_DESTROY();
3561 }
3562 
3563 /*
3564  * Initialize an IGMPv3 instance.
3565  * VIMAGE: Assumes curvnet set by caller and called per vimage.
3566  */
3567 static int
3568 vnet_igmp_iattach(const void *unused __unused)
3569 {
3570 	INIT_VNET_INET(curvnet);
3571 
3572 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3573 
3574 	LIST_INIT(&V_igi_head);
3575 
3576 	V_current_state_timers_running = 0;
3577 	V_state_change_timers_running = 0;
3578 	V_interface_timers_running = 0;
3579 
3580 	/*
3581 	 * Initialize sysctls to default values.
3582 	 */
3583 	V_igmp_recvifkludge = 1;
3584 	V_igmp_sendra = 1;
3585 	V_igmp_sendlocal = 1;
3586 	V_igmp_v1enable = 1;
3587 	V_igmp_v2enable = 1;
3588 	V_igmp_legacysupp = 0;
3589 	V_igmp_default_version = IGMP_VERSION_3;
3590 	V_igmp_gsrdelay.tv_sec = 10;
3591 	V_igmp_gsrdelay.tv_usec = 0;
3592 
3593 	memset(&V_igmpstat, 0, sizeof(struct igmpstat));
3594 	V_igmpstat.igps_version = IGPS_VERSION_3;
3595 	V_igmpstat.igps_len = sizeof(struct igmpstat);
3596 
3597 	return (0);
3598 }
3599 
3600 static int
3601 vnet_igmp_idetach(const void *unused __unused)
3602 {
3603 #ifdef INVARIANTS
3604 	INIT_VNET_INET(curvnet);
3605 #endif
3606 
3607 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3608 
3609 	KASSERT(LIST_EMPTY(&V_igi_head),
3610 	    ("%s: igi list not empty; ifnets not detached?", __func__));
3611 
3612 	return (0);
3613 }
3614 
3615 #ifndef VIMAGE_GLOBALS
3616 static vnet_modinfo_t vnet_igmp_modinfo = {
3617 	.vmi_id		= VNET_MOD_IGMP,
3618 	.vmi_name	= "igmp",
3619 	.vmi_dependson	= VNET_MOD_INET,
3620 	.vmi_iattach	= vnet_igmp_iattach,
3621 	.vmi_idetach	= vnet_igmp_idetach
3622 };
3623 #endif
3624 
3625 static int
3626 igmp_modevent(module_t mod, int type, void *unused __unused)
3627 {
3628 
3629     switch (type) {
3630     case MOD_LOAD:
3631 	igmp_sysinit();
3632 #ifndef VIMAGE_GLOBALS
3633 	vnet_mod_register(&vnet_igmp_modinfo);
3634 #else
3635 	vnet_igmp_iattach(NULL);
3636 #endif
3637 	break;
3638     case MOD_UNLOAD:
3639 #ifndef VIMAGE_GLOBALS
3640 	vnet_mod_deregister(&vnet_igmp_modinfo);
3641 #else
3642 	vnet_igmp_idetach(NULL);
3643 #endif
3644 	igmp_sysuninit();
3645 	break;
3646     default:
3647 	return (EOPNOTSUPP);
3648     }
3649     return (0);
3650 }
3651 
3652 static moduledata_t igmp_mod = {
3653     "igmp",
3654     igmp_modevent,
3655     0
3656 };
3657 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3658