xref: /freebsd/sys/netinet/igmp.c (revision 79b015331632b33f22bc8d8fef136c2672f2241f)
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 <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/module.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/socket.h>
59 #include <sys/protosw.h>
60 #include <sys/kernel.h>
61 #include <sys/lock.h>
62 #include <sys/rmlock.h>
63 #include <sys/sysctl.h>
64 #include <sys/ktr.h>
65 #include <sys/condvar.h>
66 
67 #include <net/if.h>
68 #include <net/if_var.h>
69 #include <net/netisr.h>
70 #include <net/vnet.h>
71 
72 #include <netinet/in.h>
73 #include <netinet/in_var.h>
74 #include <netinet/in_systm.h>
75 #include <netinet/ip.h>
76 #include <netinet/ip_var.h>
77 #include <netinet/ip_options.h>
78 #include <netinet/igmp.h>
79 #include <netinet/igmp_var.h>
80 
81 #include <machine/in_cksum.h>
82 
83 #include <security/mac/mac_framework.h>
84 
85 #ifndef KTR_IGMPV3
86 #define KTR_IGMPV3 KTR_INET
87 #endif
88 
89 static struct igmp_ifsoftc *
90 		igi_alloc_locked(struct ifnet *);
91 static void	igi_delete_locked(const struct ifnet *);
92 static void	igmp_dispatch_queue(struct mbufq *, int, const int);
93 static void	igmp_fasttimo_vnet(void);
94 static void	igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
95 static int	igmp_handle_state_change(struct in_multi *,
96 		    struct igmp_ifsoftc *);
97 static int	igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
98 static int	igmp_input_v1_query(struct ifnet *, const struct ip *,
99 		    const struct igmp *);
100 static int	igmp_input_v2_query(struct ifnet *, const struct ip *,
101 		    const struct igmp *);
102 static int	igmp_input_v3_query(struct ifnet *, const struct ip *,
103 		    /*const*/ struct igmpv3 *);
104 static int	igmp_input_v3_group_query(struct in_multi *,
105 		    struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
106 static int	igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
107 		    /*const*/ struct igmp *);
108 static int	igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
109 		    /*const*/ struct igmp *);
110 static void	igmp_intr(struct mbuf *);
111 static int	igmp_isgroupreported(const struct in_addr);
112 static struct mbuf *
113 		igmp_ra_alloc(void);
114 #ifdef KTR
115 static char *	igmp_rec_type_to_str(const int);
116 #endif
117 static void	igmp_set_version(struct igmp_ifsoftc *, const int);
118 static void	igmp_slowtimo_vnet(void);
119 static int	igmp_v1v2_queue_report(struct in_multi *, const int);
120 static void	igmp_v1v2_process_group_timer(struct in_multi *, const int);
121 static void	igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
122 static void	igmp_v2_update_group(struct in_multi *, const int);
123 static void	igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
124 static void	igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
125 static struct mbuf *
126 		igmp_v3_encap_report(struct ifnet *, struct mbuf *);
127 static int	igmp_v3_enqueue_group_record(struct mbufq *,
128 		    struct in_multi *, const int, const int, const int);
129 static int	igmp_v3_enqueue_filter_change(struct mbufq *,
130 		    struct in_multi *);
131 static void	igmp_v3_process_group_timers(struct igmp_ifsoftc *,
132 		    struct mbufq *, struct mbufq *, struct in_multi *,
133 		    const int);
134 static int	igmp_v3_merge_state_changes(struct in_multi *,
135 		    struct mbufq *);
136 static void	igmp_v3_suppress_group_record(struct in_multi *);
137 static int	sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
138 static int	sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
139 static int	sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
140 
141 static const struct netisr_handler igmp_nh = {
142 	.nh_name = "igmp",
143 	.nh_handler = igmp_intr,
144 	.nh_proto = NETISR_IGMP,
145 	.nh_policy = NETISR_POLICY_SOURCE,
146 };
147 
148 /*
149  * System-wide globals.
150  *
151  * Unlocked access to these is OK, except for the global IGMP output
152  * queue. The IGMP subsystem lock ends up being system-wide for the moment,
153  * because all VIMAGEs have to share a global output queue, as netisrs
154  * themselves are not virtualized.
155  *
156  * Locking:
157  *  * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
158  *    Any may be taken independently; if any are held at the same
159  *    time, the above lock order must be followed.
160  *  * All output is delegated to the netisr.
161  *    Now that Giant has been eliminated, the netisr may be inlined.
162  *  * IN_MULTI_LOCK covers in_multi.
163  *  * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
164  *    including the output queue.
165  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
166  *    per-link state iterators.
167  *  * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
168  *    therefore it is not refcounted.
169  *    We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
170  *
171  * Reference counting
172  *  * IGMP acquires its own reference every time an in_multi is passed to
173  *    it and the group is being joined for the first time.
174  *  * IGMP releases its reference(s) on in_multi in a deferred way,
175  *    because the operations which process the release run as part of
176  *    a loop whose control variables are directly affected by the release
177  *    (that, and not recursing on the IF_ADDR_LOCK).
178  *
179  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
180  * to a vnet in ifp->if_vnet.
181  *
182  * SMPng: XXX We may potentially race operations on ifma_protospec.
183  * The problem is that we currently lack a clean way of taking the
184  * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
185  * as anything which modifies ifma needs to be covered by that lock.
186  * So check for ifma_protospec being NULL before proceeding.
187  */
188 struct mtx		 igmp_mtx;
189 
190 struct mbuf		*m_raopt;		 /* Router Alert option */
191 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
192 
193 /*
194  * VIMAGE-wide globals.
195  *
196  * The IGMPv3 timers themselves need to run per-image, however,
197  * protosw timers run globally (see tcp).
198  * An ifnet can only be in one vimage at a time, and the loopback
199  * ifnet, loif, is itself virtualized.
200  * It would otherwise be possible to seriously hose IGMP state,
201  * and create inconsistencies in upstream multicast routing, if you have
202  * multiple VIMAGEs running on the same link joining different multicast
203  * groups, UNLESS the "primary IP address" is different. This is because
204  * IGMP for IPv4 does not force link-local addresses to be used for each
205  * node, unlike MLD for IPv6.
206  * Obviously the IGMPv3 per-interface state has per-vimage granularity
207  * also as a result.
208  *
209  * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
210  * policy to control the address used by IGMP on the link.
211  */
212 static VNET_DEFINE(int, interface_timers_running);	/* IGMPv3 general
213 							 * query response */
214 static VNET_DEFINE(int, state_change_timers_running);	/* IGMPv3 state-change
215 							 * retransmit */
216 static VNET_DEFINE(int, current_state_timers_running);	/* IGMPv1/v2 host
217 							 * report; IGMPv3 g/sg
218 							 * query response */
219 
220 #define	V_interface_timers_running	VNET(interface_timers_running)
221 #define	V_state_change_timers_running	VNET(state_change_timers_running)
222 #define	V_current_state_timers_running	VNET(current_state_timers_running)
223 
224 static VNET_DEFINE(LIST_HEAD(, igmp_ifsoftc), igi_head);
225 static VNET_DEFINE(struct igmpstat, igmpstat) = {
226 	.igps_version = IGPS_VERSION_3,
227 	.igps_len = sizeof(struct igmpstat),
228 };
229 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
230 
231 #define	V_igi_head			VNET(igi_head)
232 #define	V_igmpstat			VNET(igmpstat)
233 #define	V_igmp_gsrdelay			VNET(igmp_gsrdelay)
234 
235 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
236 static VNET_DEFINE(int, igmp_sendra) = 1;
237 static VNET_DEFINE(int, igmp_sendlocal) = 1;
238 static VNET_DEFINE(int, igmp_v1enable) = 1;
239 static VNET_DEFINE(int, igmp_v2enable) = 1;
240 static VNET_DEFINE(int, igmp_legacysupp);
241 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
242 
243 #define	V_igmp_recvifkludge		VNET(igmp_recvifkludge)
244 #define	V_igmp_sendra			VNET(igmp_sendra)
245 #define	V_igmp_sendlocal		VNET(igmp_sendlocal)
246 #define	V_igmp_v1enable			VNET(igmp_v1enable)
247 #define	V_igmp_v2enable			VNET(igmp_v2enable)
248 #define	V_igmp_legacysupp		VNET(igmp_legacysupp)
249 #define	V_igmp_default_version		VNET(igmp_default_version)
250 
251 /*
252  * Virtualized sysctls.
253  */
254 SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW,
255     &VNET_NAME(igmpstat), igmpstat, "");
256 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
257     &VNET_NAME(igmp_recvifkludge), 0,
258     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
259 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
260     &VNET_NAME(igmp_sendra), 0,
261     "Send IP Router Alert option in IGMPv2/v3 messages");
262 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
263     &VNET_NAME(igmp_sendlocal), 0,
264     "Send IGMP membership reports for 224.0.0.0/24 groups");
265 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
266     &VNET_NAME(igmp_v1enable), 0,
267     "Enable backwards compatibility with IGMPv1");
268 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
269     &VNET_NAME(igmp_v2enable), 0,
270     "Enable backwards compatibility with IGMPv2");
271 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
272     &VNET_NAME(igmp_legacysupp), 0,
273     "Allow v1/v2 reports to suppress v3 group responses");
274 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
275     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
276     &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
277     "Default version of IGMP to run on each interface");
278 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
279     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
280     &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
281     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
282 
283 /*
284  * Non-virtualized sysctls.
285  */
286 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
287     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
288     "Per-interface IGMPv3 state");
289 
290 static __inline void
291 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
292 {
293 
294 #ifdef VIMAGE
295 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
296 #endif /* VIMAGE */
297 	m->m_pkthdr.flowid = ifp->if_index;
298 }
299 
300 static __inline void
301 igmp_scrub_context(struct mbuf *m)
302 {
303 
304 	m->m_pkthdr.PH_loc.ptr = NULL;
305 	m->m_pkthdr.flowid = 0;
306 }
307 
308 #ifdef KTR
309 static __inline char *
310 inet_ntoa_haddr(in_addr_t haddr)
311 {
312 	struct in_addr ia;
313 
314 	ia.s_addr = htonl(haddr);
315 	return (inet_ntoa(ia));
316 }
317 #endif
318 
319 /*
320  * Restore context from a queued IGMP output chain.
321  * Return saved ifindex.
322  *
323  * VIMAGE: The assertion is there to make sure that we
324  * actually called CURVNET_SET() with what's in the mbuf chain.
325  */
326 static __inline uint32_t
327 igmp_restore_context(struct mbuf *m)
328 {
329 
330 #ifdef notyet
331 #if defined(VIMAGE) && defined(INVARIANTS)
332 	KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
333 	    ("%s: called when curvnet was not restored", __func__));
334 #endif
335 #endif
336 	return (m->m_pkthdr.flowid);
337 }
338 
339 /*
340  * Retrieve or set default IGMP version.
341  *
342  * VIMAGE: Assume curvnet set by caller.
343  * SMPng: NOTE: Serialized by IGMP lock.
344  */
345 static int
346 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
347 {
348 	int	 error;
349 	int	 new;
350 
351 	error = sysctl_wire_old_buffer(req, sizeof(int));
352 	if (error)
353 		return (error);
354 
355 	IGMP_LOCK();
356 
357 	new = V_igmp_default_version;
358 
359 	error = sysctl_handle_int(oidp, &new, 0, req);
360 	if (error || !req->newptr)
361 		goto out_locked;
362 
363 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
364 		error = EINVAL;
365 		goto out_locked;
366 	}
367 
368 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
369 	     V_igmp_default_version, new);
370 
371 	V_igmp_default_version = new;
372 
373 out_locked:
374 	IGMP_UNLOCK();
375 	return (error);
376 }
377 
378 /*
379  * Retrieve or set threshold between group-source queries in seconds.
380  *
381  * VIMAGE: Assume curvnet set by caller.
382  * SMPng: NOTE: Serialized by IGMP lock.
383  */
384 static int
385 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
386 {
387 	int error;
388 	int i;
389 
390 	error = sysctl_wire_old_buffer(req, sizeof(int));
391 	if (error)
392 		return (error);
393 
394 	IGMP_LOCK();
395 
396 	i = V_igmp_gsrdelay.tv_sec;
397 
398 	error = sysctl_handle_int(oidp, &i, 0, req);
399 	if (error || !req->newptr)
400 		goto out_locked;
401 
402 	if (i < -1 || i >= 60) {
403 		error = EINVAL;
404 		goto out_locked;
405 	}
406 
407 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
408 	     V_igmp_gsrdelay.tv_sec, i);
409 	V_igmp_gsrdelay.tv_sec = i;
410 
411 out_locked:
412 	IGMP_UNLOCK();
413 	return (error);
414 }
415 
416 /*
417  * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
418  * For use by ifmcstat(8).
419  *
420  * SMPng: NOTE: Does an unlocked ifindex space read.
421  * VIMAGE: Assume curvnet set by caller. The node handler itself
422  * is not directly virtualized.
423  */
424 static int
425 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
426 {
427 	int			*name;
428 	int			 error;
429 	u_int			 namelen;
430 	struct ifnet		*ifp;
431 	struct igmp_ifsoftc	*igi;
432 
433 	name = (int *)arg1;
434 	namelen = arg2;
435 
436 	if (req->newptr != NULL)
437 		return (EPERM);
438 
439 	if (namelen != 1)
440 		return (EINVAL);
441 
442 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
443 	if (error)
444 		return (error);
445 
446 	IN_MULTI_LOCK();
447 	IGMP_LOCK();
448 
449 	if (name[0] <= 0 || name[0] > V_if_index) {
450 		error = ENOENT;
451 		goto out_locked;
452 	}
453 
454 	error = ENOENT;
455 
456 	ifp = ifnet_byindex(name[0]);
457 	if (ifp == NULL)
458 		goto out_locked;
459 
460 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
461 		if (ifp == igi->igi_ifp) {
462 			struct igmp_ifinfo info;
463 
464 			info.igi_version = igi->igi_version;
465 			info.igi_v1_timer = igi->igi_v1_timer;
466 			info.igi_v2_timer = igi->igi_v2_timer;
467 			info.igi_v3_timer = igi->igi_v3_timer;
468 			info.igi_flags = igi->igi_flags;
469 			info.igi_rv = igi->igi_rv;
470 			info.igi_qi = igi->igi_qi;
471 			info.igi_qri = igi->igi_qri;
472 			info.igi_uri = igi->igi_uri;
473 			error = SYSCTL_OUT(req, &info, sizeof(info));
474 			break;
475 		}
476 	}
477 
478 out_locked:
479 	IGMP_UNLOCK();
480 	IN_MULTI_UNLOCK();
481 	return (error);
482 }
483 
484 /*
485  * Dispatch an entire queue of pending packet chains
486  * using the netisr.
487  * VIMAGE: Assumes the vnet pointer has been set.
488  */
489 static void
490 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
491 {
492 	struct mbuf *m;
493 
494 	while ((m = mbufq_dequeue(mq)) != NULL) {
495 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
496 		if (loop)
497 			m->m_flags |= M_IGMP_LOOP;
498 		netisr_dispatch(NETISR_IGMP, m);
499 		if (--limit == 0)
500 			break;
501 	}
502 }
503 
504 /*
505  * Filter outgoing IGMP report state by group.
506  *
507  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
508  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
509  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
510  * this may break certain IGMP snooping switches which rely on the old
511  * report behaviour.
512  *
513  * Return zero if the given group is one for which IGMP reports
514  * should be suppressed, or non-zero if reports should be issued.
515  */
516 static __inline int
517 igmp_isgroupreported(const struct in_addr addr)
518 {
519 
520 	if (in_allhosts(addr) ||
521 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
522 		return (0);
523 
524 	return (1);
525 }
526 
527 /*
528  * Construct a Router Alert option to use in outgoing packets.
529  */
530 static struct mbuf *
531 igmp_ra_alloc(void)
532 {
533 	struct mbuf	*m;
534 	struct ipoption	*p;
535 
536 	m = m_get(M_WAITOK, MT_DATA);
537 	p = mtod(m, struct ipoption *);
538 	p->ipopt_dst.s_addr = INADDR_ANY;
539 	p->ipopt_list[0] = IPOPT_RA;	/* Router Alert Option */
540 	p->ipopt_list[1] = 0x04;	/* 4 bytes long */
541 	p->ipopt_list[2] = IPOPT_EOL;	/* End of IP option list */
542 	p->ipopt_list[3] = 0x00;	/* pad byte */
543 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
544 
545 	return (m);
546 }
547 
548 /*
549  * Attach IGMP when PF_INET is attached to an interface.
550  */
551 struct igmp_ifsoftc *
552 igmp_domifattach(struct ifnet *ifp)
553 {
554 	struct igmp_ifsoftc *igi;
555 
556 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
557 	    __func__, ifp, ifp->if_xname);
558 
559 	IGMP_LOCK();
560 
561 	igi = igi_alloc_locked(ifp);
562 	if (!(ifp->if_flags & IFF_MULTICAST))
563 		igi->igi_flags |= IGIF_SILENT;
564 
565 	IGMP_UNLOCK();
566 
567 	return (igi);
568 }
569 
570 /*
571  * VIMAGE: assume curvnet set by caller.
572  */
573 static struct igmp_ifsoftc *
574 igi_alloc_locked(/*const*/ struct ifnet *ifp)
575 {
576 	struct igmp_ifsoftc *igi;
577 
578 	IGMP_LOCK_ASSERT();
579 
580 	igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
581 	if (igi == NULL)
582 		goto out;
583 
584 	igi->igi_ifp = ifp;
585 	igi->igi_version = V_igmp_default_version;
586 	igi->igi_flags = 0;
587 	igi->igi_rv = IGMP_RV_INIT;
588 	igi->igi_qi = IGMP_QI_INIT;
589 	igi->igi_qri = IGMP_QRI_INIT;
590 	igi->igi_uri = IGMP_URI_INIT;
591 	SLIST_INIT(&igi->igi_relinmhead);
592 	mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
593 
594 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
595 
596 	CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
597 	     ifp, ifp->if_xname);
598 
599 out:
600 	return (igi);
601 }
602 
603 /*
604  * Hook for ifdetach.
605  *
606  * NOTE: Some finalization tasks need to run before the protocol domain
607  * is detached, but also before the link layer does its cleanup.
608  *
609  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
610  * XXX This is also bitten by unlocked ifma_protospec access.
611  */
612 void
613 igmp_ifdetach(struct ifnet *ifp)
614 {
615 	struct igmp_ifsoftc	*igi;
616 	struct ifmultiaddr	*ifma;
617 	struct in_multi		*inm, *tinm;
618 
619 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
620 	    ifp->if_xname);
621 
622 	IGMP_LOCK();
623 
624 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
625 	if (igi->igi_version == IGMP_VERSION_3) {
626 		IF_ADDR_RLOCK(ifp);
627 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
628 			if (ifma->ifma_addr->sa_family != AF_INET ||
629 			    ifma->ifma_protospec == NULL)
630 				continue;
631 #if 0
632 			KASSERT(ifma->ifma_protospec != NULL,
633 			    ("%s: ifma_protospec is NULL", __func__));
634 #endif
635 			inm = (struct in_multi *)ifma->ifma_protospec;
636 			if (inm->inm_state == IGMP_LEAVING_MEMBER) {
637 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
638 				    inm, inm_nrele);
639 			}
640 			inm_clear_recorded(inm);
641 		}
642 		IF_ADDR_RUNLOCK(ifp);
643 		/*
644 		 * Free the in_multi reference(s) for this IGMP lifecycle.
645 		 */
646 		SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
647 		    tinm) {
648 			SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
649 			inm_release_locked(inm);
650 		}
651 	}
652 
653 	IGMP_UNLOCK();
654 }
655 
656 /*
657  * Hook for domifdetach.
658  */
659 void
660 igmp_domifdetach(struct ifnet *ifp)
661 {
662 	struct igmp_ifsoftc *igi;
663 
664 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
665 	    __func__, ifp, ifp->if_xname);
666 
667 	IGMP_LOCK();
668 
669 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
670 	igi_delete_locked(ifp);
671 
672 	IGMP_UNLOCK();
673 }
674 
675 static void
676 igi_delete_locked(const struct ifnet *ifp)
677 {
678 	struct igmp_ifsoftc *igi, *tigi;
679 
680 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
681 	    __func__, ifp, ifp->if_xname);
682 
683 	IGMP_LOCK_ASSERT();
684 
685 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
686 		if (igi->igi_ifp == ifp) {
687 			/*
688 			 * Free deferred General Query responses.
689 			 */
690 			mbufq_drain(&igi->igi_gq);
691 
692 			LIST_REMOVE(igi, igi_link);
693 
694 			KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
695 			    ("%s: there are dangling in_multi references",
696 			    __func__));
697 
698 			free(igi, M_IGMP);
699 			return;
700 		}
701 	}
702 
703 #ifdef INVARIANTS
704 	panic("%s: igmp_ifsoftc not found for ifp %p\n", __func__,  ifp);
705 #endif
706 }
707 
708 /*
709  * Process a received IGMPv1 query.
710  * Return non-zero if the message should be dropped.
711  *
712  * VIMAGE: The curvnet pointer is derived from the input ifp.
713  */
714 static int
715 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
716     const struct igmp *igmp)
717 {
718 	struct ifmultiaddr	*ifma;
719 	struct igmp_ifsoftc	*igi;
720 	struct in_multi		*inm;
721 
722 	/*
723 	 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
724 	 * 224.0.0.1. They are always treated as General Queries.
725 	 * igmp_group is always ignored. Do not drop it as a userland
726 	 * daemon may wish to see it.
727 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
728 	 */
729 	if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
730 		IGMPSTAT_INC(igps_rcv_badqueries);
731 		return (0);
732 	}
733 	IGMPSTAT_INC(igps_rcv_gen_queries);
734 
735 	IN_MULTI_LOCK();
736 	IGMP_LOCK();
737 
738 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
739 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
740 
741 	if (igi->igi_flags & IGIF_LOOPBACK) {
742 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
743 		    ifp, ifp->if_xname);
744 		goto out_locked;
745 	}
746 
747 	/*
748 	 * Switch to IGMPv1 host compatibility mode.
749 	 */
750 	igmp_set_version(igi, IGMP_VERSION_1);
751 
752 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
753 
754 	/*
755 	 * Start the timers in all of our group records
756 	 * for the interface on which the query arrived,
757 	 * except those which are already running.
758 	 */
759 	IF_ADDR_RLOCK(ifp);
760 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
761 		if (ifma->ifma_addr->sa_family != AF_INET ||
762 		    ifma->ifma_protospec == NULL)
763 			continue;
764 		inm = (struct in_multi *)ifma->ifma_protospec;
765 		if (inm->inm_timer != 0)
766 			continue;
767 		switch (inm->inm_state) {
768 		case IGMP_NOT_MEMBER:
769 		case IGMP_SILENT_MEMBER:
770 			break;
771 		case IGMP_G_QUERY_PENDING_MEMBER:
772 		case IGMP_SG_QUERY_PENDING_MEMBER:
773 		case IGMP_REPORTING_MEMBER:
774 		case IGMP_IDLE_MEMBER:
775 		case IGMP_LAZY_MEMBER:
776 		case IGMP_SLEEPING_MEMBER:
777 		case IGMP_AWAKENING_MEMBER:
778 			inm->inm_state = IGMP_REPORTING_MEMBER;
779 			inm->inm_timer = IGMP_RANDOM_DELAY(
780 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
781 			V_current_state_timers_running = 1;
782 			break;
783 		case IGMP_LEAVING_MEMBER:
784 			break;
785 		}
786 	}
787 	IF_ADDR_RUNLOCK(ifp);
788 
789 out_locked:
790 	IGMP_UNLOCK();
791 	IN_MULTI_UNLOCK();
792 
793 	return (0);
794 }
795 
796 /*
797  * Process a received IGMPv2 general or group-specific query.
798  */
799 static int
800 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
801     const struct igmp *igmp)
802 {
803 	struct ifmultiaddr	*ifma;
804 	struct igmp_ifsoftc	*igi;
805 	struct in_multi		*inm;
806 	int			 is_general_query;
807 	uint16_t		 timer;
808 
809 	is_general_query = 0;
810 
811 	/*
812 	 * Validate address fields upfront.
813 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
814 	 */
815 	if (in_nullhost(igmp->igmp_group)) {
816 		/*
817 		 * IGMPv2 General Query.
818 		 * If this was not sent to the all-hosts group, ignore it.
819 		 */
820 		if (!in_allhosts(ip->ip_dst))
821 			return (0);
822 		IGMPSTAT_INC(igps_rcv_gen_queries);
823 		is_general_query = 1;
824 	} else {
825 		/* IGMPv2 Group-Specific Query. */
826 		IGMPSTAT_INC(igps_rcv_group_queries);
827 	}
828 
829 	IN_MULTI_LOCK();
830 	IGMP_LOCK();
831 
832 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
833 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
834 
835 	if (igi->igi_flags & IGIF_LOOPBACK) {
836 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
837 		    ifp, ifp->if_xname);
838 		goto out_locked;
839 	}
840 
841 	/*
842 	 * Ignore v2 query if in v1 Compatibility Mode.
843 	 */
844 	if (igi->igi_version == IGMP_VERSION_1)
845 		goto out_locked;
846 
847 	igmp_set_version(igi, IGMP_VERSION_2);
848 
849 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
850 	if (timer == 0)
851 		timer = 1;
852 
853 	if (is_general_query) {
854 		/*
855 		 * For each reporting group joined on this
856 		 * interface, kick the report timer.
857 		 */
858 		CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
859 		    ifp, ifp->if_xname);
860 		IF_ADDR_RLOCK(ifp);
861 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
862 			if (ifma->ifma_addr->sa_family != AF_INET ||
863 			    ifma->ifma_protospec == NULL)
864 				continue;
865 			inm = (struct in_multi *)ifma->ifma_protospec;
866 			igmp_v2_update_group(inm, timer);
867 		}
868 		IF_ADDR_RUNLOCK(ifp);
869 	} else {
870 		/*
871 		 * Group-specific IGMPv2 query, we need only
872 		 * look up the single group to process it.
873 		 */
874 		inm = inm_lookup(ifp, igmp->igmp_group);
875 		if (inm != NULL) {
876 			CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
877 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
878 			igmp_v2_update_group(inm, timer);
879 		}
880 	}
881 
882 out_locked:
883 	IGMP_UNLOCK();
884 	IN_MULTI_UNLOCK();
885 
886 	return (0);
887 }
888 
889 /*
890  * Update the report timer on a group in response to an IGMPv2 query.
891  *
892  * If we are becoming the reporting member for this group, start the timer.
893  * If we already are the reporting member for this group, and timer is
894  * below the threshold, reset it.
895  *
896  * We may be updating the group for the first time since we switched
897  * to IGMPv3. If we are, then we must clear any recorded source lists,
898  * and transition to REPORTING state; the group timer is overloaded
899  * for group and group-source query responses.
900  *
901  * Unlike IGMPv3, the delay per group should be jittered
902  * to avoid bursts of IGMPv2 reports.
903  */
904 static void
905 igmp_v2_update_group(struct in_multi *inm, const int timer)
906 {
907 
908 	CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
909 	    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
910 
911 	IN_MULTI_LOCK_ASSERT();
912 
913 	switch (inm->inm_state) {
914 	case IGMP_NOT_MEMBER:
915 	case IGMP_SILENT_MEMBER:
916 		break;
917 	case IGMP_REPORTING_MEMBER:
918 		if (inm->inm_timer != 0 &&
919 		    inm->inm_timer <= timer) {
920 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
921 			    "skipping.", __func__);
922 			break;
923 		}
924 		/* FALLTHROUGH */
925 	case IGMP_SG_QUERY_PENDING_MEMBER:
926 	case IGMP_G_QUERY_PENDING_MEMBER:
927 	case IGMP_IDLE_MEMBER:
928 	case IGMP_LAZY_MEMBER:
929 	case IGMP_AWAKENING_MEMBER:
930 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
931 		inm->inm_state = IGMP_REPORTING_MEMBER;
932 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
933 		V_current_state_timers_running = 1;
934 		break;
935 	case IGMP_SLEEPING_MEMBER:
936 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
937 		inm->inm_state = IGMP_AWAKENING_MEMBER;
938 		break;
939 	case IGMP_LEAVING_MEMBER:
940 		break;
941 	}
942 }
943 
944 /*
945  * Process a received IGMPv3 general, group-specific or
946  * group-and-source-specific query.
947  * Assumes m has already been pulled up to the full IGMP message length.
948  * Return 0 if successful, otherwise an appropriate error code is returned.
949  */
950 static int
951 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
952     /*const*/ struct igmpv3 *igmpv3)
953 {
954 	struct igmp_ifsoftc	*igi;
955 	struct in_multi		*inm;
956 	int			 is_general_query;
957 	uint32_t		 maxresp, nsrc, qqi;
958 	uint16_t		 timer;
959 	uint8_t			 qrv;
960 
961 	is_general_query = 0;
962 
963 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
964 
965 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
966 	if (maxresp >= 128) {
967 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
968 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
969 	}
970 
971 	/*
972 	 * Robustness must never be less than 2 for on-wire IGMPv3.
973 	 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
974 	 * an exception for interfaces whose IGMPv3 state changes
975 	 * are redirected to loopback (e.g. MANET).
976 	 */
977 	qrv = IGMP_QRV(igmpv3->igmp_misc);
978 	if (qrv < 2) {
979 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
980 		    qrv, IGMP_RV_INIT);
981 		qrv = IGMP_RV_INIT;
982 	}
983 
984 	qqi = igmpv3->igmp_qqi;
985 	if (qqi >= 128) {
986 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
987 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
988 	}
989 
990 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
991 	if (timer == 0)
992 		timer = 1;
993 
994 	nsrc = ntohs(igmpv3->igmp_numsrc);
995 
996 	/*
997 	 * Validate address fields and versions upfront before
998 	 * accepting v3 query.
999 	 * XXX SMPng: Unlocked access to igmpstat counters here.
1000 	 */
1001 	if (in_nullhost(igmpv3->igmp_group)) {
1002 		/*
1003 		 * IGMPv3 General Query.
1004 		 *
1005 		 * General Queries SHOULD be directed to 224.0.0.1.
1006 		 * A general query with a source list has undefined
1007 		 * behaviour; discard it.
1008 		 */
1009 		IGMPSTAT_INC(igps_rcv_gen_queries);
1010 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1011 			IGMPSTAT_INC(igps_rcv_badqueries);
1012 			return (0);
1013 		}
1014 		is_general_query = 1;
1015 	} else {
1016 		/* Group or group-source specific query. */
1017 		if (nsrc == 0)
1018 			IGMPSTAT_INC(igps_rcv_group_queries);
1019 		else
1020 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1021 	}
1022 
1023 	IN_MULTI_LOCK();
1024 	IGMP_LOCK();
1025 
1026 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1027 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1028 
1029 	if (igi->igi_flags & IGIF_LOOPBACK) {
1030 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1031 		    ifp, ifp->if_xname);
1032 		goto out_locked;
1033 	}
1034 
1035 	/*
1036 	 * Discard the v3 query if we're in Compatibility Mode.
1037 	 * The RFC is not obviously worded that hosts need to stay in
1038 	 * compatibility mode until the Old Version Querier Present
1039 	 * timer expires.
1040 	 */
1041 	if (igi->igi_version != IGMP_VERSION_3) {
1042 		CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1043 		    igi->igi_version, ifp, ifp->if_xname);
1044 		goto out_locked;
1045 	}
1046 
1047 	igmp_set_version(igi, IGMP_VERSION_3);
1048 	igi->igi_rv = qrv;
1049 	igi->igi_qi = qqi;
1050 	igi->igi_qri = maxresp;
1051 
1052 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1053 	    maxresp);
1054 
1055 	if (is_general_query) {
1056 		/*
1057 		 * Schedule a current-state report on this ifp for
1058 		 * all groups, possibly containing source lists.
1059 		 * If there is a pending General Query response
1060 		 * scheduled earlier than the selected delay, do
1061 		 * not schedule any other reports.
1062 		 * Otherwise, reset the interface timer.
1063 		 */
1064 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1065 		    ifp, ifp->if_xname);
1066 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1067 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1068 			V_interface_timers_running = 1;
1069 		}
1070 	} else {
1071 		/*
1072 		 * Group-source-specific queries are throttled on
1073 		 * a per-group basis to defeat denial-of-service attempts.
1074 		 * Queries for groups we are not a member of on this
1075 		 * link are simply ignored.
1076 		 */
1077 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1078 		if (inm == NULL)
1079 			goto out_locked;
1080 		if (nsrc > 0) {
1081 			if (!ratecheck(&inm->inm_lastgsrtv,
1082 			    &V_igmp_gsrdelay)) {
1083 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1084 				    __func__);
1085 				IGMPSTAT_INC(igps_drop_gsr_queries);
1086 				goto out_locked;
1087 			}
1088 		}
1089 		CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1090 		     inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1091 		/*
1092 		 * If there is a pending General Query response
1093 		 * scheduled sooner than the selected delay, no
1094 		 * further report need be scheduled.
1095 		 * Otherwise, prepare to respond to the
1096 		 * group-specific or group-and-source query.
1097 		 */
1098 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1099 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1100 	}
1101 
1102 out_locked:
1103 	IGMP_UNLOCK();
1104 	IN_MULTI_UNLOCK();
1105 
1106 	return (0);
1107 }
1108 
1109 /*
1110  * Process a recieved IGMPv3 group-specific or group-and-source-specific
1111  * query.
1112  * Return <0 if any error occured. Currently this is ignored.
1113  */
1114 static int
1115 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1116     int timer, /*const*/ struct igmpv3 *igmpv3)
1117 {
1118 	int			 retval;
1119 	uint16_t		 nsrc;
1120 
1121 	IN_MULTI_LOCK_ASSERT();
1122 	IGMP_LOCK_ASSERT();
1123 
1124 	retval = 0;
1125 
1126 	switch (inm->inm_state) {
1127 	case IGMP_NOT_MEMBER:
1128 	case IGMP_SILENT_MEMBER:
1129 	case IGMP_SLEEPING_MEMBER:
1130 	case IGMP_LAZY_MEMBER:
1131 	case IGMP_AWAKENING_MEMBER:
1132 	case IGMP_IDLE_MEMBER:
1133 	case IGMP_LEAVING_MEMBER:
1134 		return (retval);
1135 		break;
1136 	case IGMP_REPORTING_MEMBER:
1137 	case IGMP_G_QUERY_PENDING_MEMBER:
1138 	case IGMP_SG_QUERY_PENDING_MEMBER:
1139 		break;
1140 	}
1141 
1142 	nsrc = ntohs(igmpv3->igmp_numsrc);
1143 
1144 	/*
1145 	 * Deal with group-specific queries upfront.
1146 	 * If any group query is already pending, purge any recorded
1147 	 * source-list state if it exists, and schedule a query response
1148 	 * for this group-specific query.
1149 	 */
1150 	if (nsrc == 0) {
1151 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1152 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1153 			inm_clear_recorded(inm);
1154 			timer = min(inm->inm_timer, timer);
1155 		}
1156 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1157 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1158 		V_current_state_timers_running = 1;
1159 		return (retval);
1160 	}
1161 
1162 	/*
1163 	 * Deal with the case where a group-and-source-specific query has
1164 	 * been received but a group-specific query is already pending.
1165 	 */
1166 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1167 		timer = min(inm->inm_timer, timer);
1168 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1169 		V_current_state_timers_running = 1;
1170 		return (retval);
1171 	}
1172 
1173 	/*
1174 	 * Finally, deal with the case where a group-and-source-specific
1175 	 * query has been received, where a response to a previous g-s-r
1176 	 * query exists, or none exists.
1177 	 * In this case, we need to parse the source-list which the Querier
1178 	 * has provided us with and check if we have any source list filter
1179 	 * entries at T1 for these sources. If we do not, there is no need
1180 	 * schedule a report and the query may be dropped.
1181 	 * If we do, we must record them and schedule a current-state
1182 	 * report for those sources.
1183 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1184 	 * we pass the mbuf chain pointer down to this function, and use
1185 	 * m_getptr() to walk the chain.
1186 	 */
1187 	if (inm->inm_nsrc > 0) {
1188 		const struct in_addr	*ap;
1189 		int			 i, nrecorded;
1190 
1191 		ap = (const struct in_addr *)(igmpv3 + 1);
1192 		nrecorded = 0;
1193 		for (i = 0; i < nsrc; i++, ap++) {
1194 			retval = inm_record_source(inm, ap->s_addr);
1195 			if (retval < 0)
1196 				break;
1197 			nrecorded += retval;
1198 		}
1199 		if (nrecorded > 0) {
1200 			CTR1(KTR_IGMPV3,
1201 			    "%s: schedule response to SG query", __func__);
1202 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1203 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1204 			V_current_state_timers_running = 1;
1205 		}
1206 	}
1207 
1208 	return (retval);
1209 }
1210 
1211 /*
1212  * Process a received IGMPv1 host membership report.
1213  *
1214  * NOTE: 0.0.0.0 workaround breaks const correctness.
1215  */
1216 static int
1217 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1218     /*const*/ struct igmp *igmp)
1219 {
1220 	struct rm_priotracker in_ifa_tracker;
1221 	struct in_ifaddr *ia;
1222 	struct in_multi *inm;
1223 
1224 	IGMPSTAT_INC(igps_rcv_reports);
1225 
1226 	if (ifp->if_flags & IFF_LOOPBACK)
1227 		return (0);
1228 
1229 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1230 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1231 		IGMPSTAT_INC(igps_rcv_badreports);
1232 		return (EINVAL);
1233 	}
1234 
1235 	/*
1236 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1237 	 * Booting clients may use the source address 0.0.0.0. Some
1238 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1239 	 * the interface upon which this message was received.
1240 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1241 	 */
1242 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1243 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1244 		if (ia != NULL) {
1245 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1246 			ifa_free(&ia->ia_ifa);
1247 		}
1248 	}
1249 
1250 	CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1251 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1252 
1253 	/*
1254 	 * IGMPv1 report suppression.
1255 	 * If we are a member of this group, and our membership should be
1256 	 * reported, stop our group timer and transition to the 'lazy' state.
1257 	 */
1258 	IN_MULTI_LOCK();
1259 	inm = inm_lookup(ifp, igmp->igmp_group);
1260 	if (inm != NULL) {
1261 		struct igmp_ifsoftc *igi;
1262 
1263 		igi = inm->inm_igi;
1264 		if (igi == NULL) {
1265 			KASSERT(igi != NULL,
1266 			    ("%s: no igi for ifp %p", __func__, ifp));
1267 			goto out_locked;
1268 		}
1269 
1270 		IGMPSTAT_INC(igps_rcv_ourreports);
1271 
1272 		/*
1273 		 * If we are in IGMPv3 host mode, do not allow the
1274 		 * other host's IGMPv1 report to suppress our reports
1275 		 * unless explicitly configured to do so.
1276 		 */
1277 		if (igi->igi_version == IGMP_VERSION_3) {
1278 			if (V_igmp_legacysupp)
1279 				igmp_v3_suppress_group_record(inm);
1280 			goto out_locked;
1281 		}
1282 
1283 		inm->inm_timer = 0;
1284 
1285 		switch (inm->inm_state) {
1286 		case IGMP_NOT_MEMBER:
1287 		case IGMP_SILENT_MEMBER:
1288 			break;
1289 		case IGMP_IDLE_MEMBER:
1290 		case IGMP_LAZY_MEMBER:
1291 		case IGMP_AWAKENING_MEMBER:
1292 			CTR3(KTR_IGMPV3,
1293 			    "report suppressed for %s on ifp %p(%s)",
1294 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1295 		case IGMP_SLEEPING_MEMBER:
1296 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1297 			break;
1298 		case IGMP_REPORTING_MEMBER:
1299 			CTR3(KTR_IGMPV3,
1300 			    "report suppressed for %s on ifp %p(%s)",
1301 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1302 			if (igi->igi_version == IGMP_VERSION_1)
1303 				inm->inm_state = IGMP_LAZY_MEMBER;
1304 			else if (igi->igi_version == IGMP_VERSION_2)
1305 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1306 			break;
1307 		case IGMP_G_QUERY_PENDING_MEMBER:
1308 		case IGMP_SG_QUERY_PENDING_MEMBER:
1309 		case IGMP_LEAVING_MEMBER:
1310 			break;
1311 		}
1312 	}
1313 
1314 out_locked:
1315 	IN_MULTI_UNLOCK();
1316 
1317 	return (0);
1318 }
1319 
1320 /*
1321  * Process a received IGMPv2 host membership report.
1322  *
1323  * NOTE: 0.0.0.0 workaround breaks const correctness.
1324  */
1325 static int
1326 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1327     /*const*/ struct igmp *igmp)
1328 {
1329 	struct rm_priotracker in_ifa_tracker;
1330 	struct in_ifaddr *ia;
1331 	struct in_multi *inm;
1332 
1333 	/*
1334 	 * Make sure we don't hear our own membership report.  Fast
1335 	 * leave requires knowing that we are the only member of a
1336 	 * group.
1337 	 */
1338 	IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1339 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1340 		ifa_free(&ia->ia_ifa);
1341 		return (0);
1342 	}
1343 
1344 	IGMPSTAT_INC(igps_rcv_reports);
1345 
1346 	if (ifp->if_flags & IFF_LOOPBACK) {
1347 		if (ia != NULL)
1348 			ifa_free(&ia->ia_ifa);
1349 		return (0);
1350 	}
1351 
1352 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1353 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1354 		if (ia != NULL)
1355 			ifa_free(&ia->ia_ifa);
1356 		IGMPSTAT_INC(igps_rcv_badreports);
1357 		return (EINVAL);
1358 	}
1359 
1360 	/*
1361 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1362 	 * Booting clients may use the source address 0.0.0.0. Some
1363 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1364 	 * the interface upon which this message was received.
1365 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1366 	 */
1367 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1368 		if (ia != NULL)
1369 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1370 	}
1371 	if (ia != NULL)
1372 		ifa_free(&ia->ia_ifa);
1373 
1374 	CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1375 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1376 
1377 	/*
1378 	 * IGMPv2 report suppression.
1379 	 * If we are a member of this group, and our membership should be
1380 	 * reported, and our group timer is pending or about to be reset,
1381 	 * stop our group timer by transitioning to the 'lazy' state.
1382 	 */
1383 	IN_MULTI_LOCK();
1384 	inm = inm_lookup(ifp, igmp->igmp_group);
1385 	if (inm != NULL) {
1386 		struct igmp_ifsoftc *igi;
1387 
1388 		igi = inm->inm_igi;
1389 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1390 
1391 		IGMPSTAT_INC(igps_rcv_ourreports);
1392 
1393 		/*
1394 		 * If we are in IGMPv3 host mode, do not allow the
1395 		 * other host's IGMPv1 report to suppress our reports
1396 		 * unless explicitly configured to do so.
1397 		 */
1398 		if (igi->igi_version == IGMP_VERSION_3) {
1399 			if (V_igmp_legacysupp)
1400 				igmp_v3_suppress_group_record(inm);
1401 			goto out_locked;
1402 		}
1403 
1404 		inm->inm_timer = 0;
1405 
1406 		switch (inm->inm_state) {
1407 		case IGMP_NOT_MEMBER:
1408 		case IGMP_SILENT_MEMBER:
1409 		case IGMP_SLEEPING_MEMBER:
1410 			break;
1411 		case IGMP_REPORTING_MEMBER:
1412 		case IGMP_IDLE_MEMBER:
1413 		case IGMP_AWAKENING_MEMBER:
1414 			CTR3(KTR_IGMPV3,
1415 			    "report suppressed for %s on ifp %p(%s)",
1416 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1417 		case IGMP_LAZY_MEMBER:
1418 			inm->inm_state = IGMP_LAZY_MEMBER;
1419 			break;
1420 		case IGMP_G_QUERY_PENDING_MEMBER:
1421 		case IGMP_SG_QUERY_PENDING_MEMBER:
1422 		case IGMP_LEAVING_MEMBER:
1423 			break;
1424 		}
1425 	}
1426 
1427 out_locked:
1428 	IN_MULTI_UNLOCK();
1429 
1430 	return (0);
1431 }
1432 
1433 int
1434 igmp_input(struct mbuf **mp, int *offp, int proto)
1435 {
1436 	int iphlen;
1437 	struct ifnet *ifp;
1438 	struct igmp *igmp;
1439 	struct ip *ip;
1440 	struct mbuf *m;
1441 	int igmplen;
1442 	int minlen;
1443 	int queryver;
1444 
1445 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1446 
1447 	m = *mp;
1448 	ifp = m->m_pkthdr.rcvif;
1449 	*mp = NULL;
1450 
1451 	IGMPSTAT_INC(igps_rcv_total);
1452 
1453 	ip = mtod(m, struct ip *);
1454 	iphlen = *offp;
1455 	igmplen = ntohs(ip->ip_len) - iphlen;
1456 
1457 	/*
1458 	 * Validate lengths.
1459 	 */
1460 	if (igmplen < IGMP_MINLEN) {
1461 		IGMPSTAT_INC(igps_rcv_tooshort);
1462 		m_freem(m);
1463 		return (IPPROTO_DONE);
1464 	}
1465 
1466 	/*
1467 	 * Always pullup to the minimum size for v1/v2 or v3
1468 	 * to amortize calls to m_pullup().
1469 	 */
1470 	minlen = iphlen;
1471 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1472 		minlen += IGMP_V3_QUERY_MINLEN;
1473 	else
1474 		minlen += IGMP_MINLEN;
1475 	if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1476 	    (m = m_pullup(m, minlen)) == 0) {
1477 		IGMPSTAT_INC(igps_rcv_tooshort);
1478 		return (IPPROTO_DONE);
1479 	}
1480 	ip = mtod(m, struct ip *);
1481 
1482 	/*
1483 	 * Validate checksum.
1484 	 */
1485 	m->m_data += iphlen;
1486 	m->m_len -= iphlen;
1487 	igmp = mtod(m, struct igmp *);
1488 	if (in_cksum(m, igmplen)) {
1489 		IGMPSTAT_INC(igps_rcv_badsum);
1490 		m_freem(m);
1491 		return (IPPROTO_DONE);
1492 	}
1493 	m->m_data -= iphlen;
1494 	m->m_len += iphlen;
1495 
1496 	/*
1497 	 * IGMP control traffic is link-scope, and must have a TTL of 1.
1498 	 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1499 	 * probe packets may come from beyond the LAN.
1500 	 */
1501 	if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1502 		IGMPSTAT_INC(igps_rcv_badttl);
1503 		m_freem(m);
1504 		return (IPPROTO_DONE);
1505 	}
1506 
1507 	switch (igmp->igmp_type) {
1508 	case IGMP_HOST_MEMBERSHIP_QUERY:
1509 		if (igmplen == IGMP_MINLEN) {
1510 			if (igmp->igmp_code == 0)
1511 				queryver = IGMP_VERSION_1;
1512 			else
1513 				queryver = IGMP_VERSION_2;
1514 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1515 			queryver = IGMP_VERSION_3;
1516 		} else {
1517 			IGMPSTAT_INC(igps_rcv_tooshort);
1518 			m_freem(m);
1519 			return (IPPROTO_DONE);
1520 		}
1521 
1522 		switch (queryver) {
1523 		case IGMP_VERSION_1:
1524 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1525 			if (!V_igmp_v1enable)
1526 				break;
1527 			if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1528 				m_freem(m);
1529 				return (IPPROTO_DONE);
1530 			}
1531 			break;
1532 
1533 		case IGMP_VERSION_2:
1534 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1535 			if (!V_igmp_v2enable)
1536 				break;
1537 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1538 				m_freem(m);
1539 				return (IPPROTO_DONE);
1540 			}
1541 			break;
1542 
1543 		case IGMP_VERSION_3: {
1544 				struct igmpv3 *igmpv3;
1545 				uint16_t igmpv3len;
1546 				uint16_t nsrc;
1547 
1548 				IGMPSTAT_INC(igps_rcv_v3_queries);
1549 				igmpv3 = (struct igmpv3 *)igmp;
1550 				/*
1551 				 * Validate length based on source count.
1552 				 */
1553 				nsrc = ntohs(igmpv3->igmp_numsrc);
1554 				if (nsrc * sizeof(in_addr_t) >
1555 				    UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1556 					IGMPSTAT_INC(igps_rcv_tooshort);
1557 					return (IPPROTO_DONE);
1558 				}
1559 				/*
1560 				 * m_pullup() may modify m, so pullup in
1561 				 * this scope.
1562 				 */
1563 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1564 				   sizeof(struct in_addr) * nsrc;
1565 				if ((!M_WRITABLE(m) ||
1566 				     m->m_len < igmpv3len) &&
1567 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1568 					IGMPSTAT_INC(igps_rcv_tooshort);
1569 					return (IPPROTO_DONE);
1570 				}
1571 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1572 				    + iphlen);
1573 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1574 					m_freem(m);
1575 					return (IPPROTO_DONE);
1576 				}
1577 			}
1578 			break;
1579 		}
1580 		break;
1581 
1582 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1583 		if (!V_igmp_v1enable)
1584 			break;
1585 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1586 			m_freem(m);
1587 			return (IPPROTO_DONE);
1588 		}
1589 		break;
1590 
1591 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1592 		if (!V_igmp_v2enable)
1593 			break;
1594 		if (!ip_checkrouteralert(m))
1595 			IGMPSTAT_INC(igps_rcv_nora);
1596 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1597 			m_freem(m);
1598 			return (IPPROTO_DONE);
1599 		}
1600 		break;
1601 
1602 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1603 		/*
1604 		 * Hosts do not need to process IGMPv3 membership reports,
1605 		 * as report suppression is no longer required.
1606 		 */
1607 		if (!ip_checkrouteralert(m))
1608 			IGMPSTAT_INC(igps_rcv_nora);
1609 		break;
1610 
1611 	default:
1612 		break;
1613 	}
1614 
1615 	/*
1616 	 * Pass all valid IGMP packets up to any process(es) listening on a
1617 	 * raw IGMP socket.
1618 	 */
1619 	*mp = m;
1620 	return (rip_input(mp, offp, proto));
1621 }
1622 
1623 
1624 /*
1625  * Fast timeout handler (global).
1626  * VIMAGE: Timeout handlers are expected to service all vimages.
1627  */
1628 void
1629 igmp_fasttimo(void)
1630 {
1631 	VNET_ITERATOR_DECL(vnet_iter);
1632 
1633 	VNET_LIST_RLOCK_NOSLEEP();
1634 	VNET_FOREACH(vnet_iter) {
1635 		CURVNET_SET(vnet_iter);
1636 		igmp_fasttimo_vnet();
1637 		CURVNET_RESTORE();
1638 	}
1639 	VNET_LIST_RUNLOCK_NOSLEEP();
1640 }
1641 
1642 /*
1643  * Fast timeout handler (per-vnet).
1644  * Sends are shuffled off to a netisr to deal with Giant.
1645  *
1646  * VIMAGE: Assume caller has set up our curvnet.
1647  */
1648 static void
1649 igmp_fasttimo_vnet(void)
1650 {
1651 	struct mbufq		 scq;	/* State-change packets */
1652 	struct mbufq		 qrq;	/* Query response packets */
1653 	struct ifnet		*ifp;
1654 	struct igmp_ifsoftc	*igi;
1655 	struct ifmultiaddr	*ifma;
1656 	struct in_multi		*inm;
1657 	int			 loop, uri_fasthz;
1658 
1659 	loop = 0;
1660 	uri_fasthz = 0;
1661 
1662 	/*
1663 	 * Quick check to see if any work needs to be done, in order to
1664 	 * minimize the overhead of fasttimo processing.
1665 	 * SMPng: XXX Unlocked reads.
1666 	 */
1667 	if (!V_current_state_timers_running &&
1668 	    !V_interface_timers_running &&
1669 	    !V_state_change_timers_running)
1670 		return;
1671 
1672 	IN_MULTI_LOCK();
1673 	IGMP_LOCK();
1674 
1675 	/*
1676 	 * IGMPv3 General Query response timer processing.
1677 	 */
1678 	if (V_interface_timers_running) {
1679 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1680 
1681 		V_interface_timers_running = 0;
1682 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1683 			if (igi->igi_v3_timer == 0) {
1684 				/* Do nothing. */
1685 			} else if (--igi->igi_v3_timer == 0) {
1686 				igmp_v3_dispatch_general_query(igi);
1687 			} else {
1688 				V_interface_timers_running = 1;
1689 			}
1690 		}
1691 	}
1692 
1693 	if (!V_current_state_timers_running &&
1694 	    !V_state_change_timers_running)
1695 		goto out_locked;
1696 
1697 	V_current_state_timers_running = 0;
1698 	V_state_change_timers_running = 0;
1699 
1700 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1701 
1702 	/*
1703 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1704 	 * Note: Processing a v3 group timer may remove a node.
1705 	 */
1706 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1707 		ifp = igi->igi_ifp;
1708 
1709 		if (igi->igi_version == IGMP_VERSION_3) {
1710 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1711 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1712 			    PR_FASTHZ);
1713 			mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1714 			mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1715 		}
1716 
1717 		IF_ADDR_RLOCK(ifp);
1718 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1719 			if (ifma->ifma_addr->sa_family != AF_INET ||
1720 			    ifma->ifma_protospec == NULL)
1721 				continue;
1722 			inm = (struct in_multi *)ifma->ifma_protospec;
1723 			switch (igi->igi_version) {
1724 			case IGMP_VERSION_1:
1725 			case IGMP_VERSION_2:
1726 				igmp_v1v2_process_group_timer(inm,
1727 				    igi->igi_version);
1728 				break;
1729 			case IGMP_VERSION_3:
1730 				igmp_v3_process_group_timers(igi, &qrq,
1731 				    &scq, inm, uri_fasthz);
1732 				break;
1733 			}
1734 		}
1735 		IF_ADDR_RUNLOCK(ifp);
1736 
1737 		if (igi->igi_version == IGMP_VERSION_3) {
1738 			struct in_multi		*tinm;
1739 
1740 			igmp_dispatch_queue(&qrq, 0, loop);
1741 			igmp_dispatch_queue(&scq, 0, loop);
1742 
1743 			/*
1744 			 * Free the in_multi reference(s) for this
1745 			 * IGMP lifecycle.
1746 			 */
1747 			SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1748 			    inm_nrele, tinm) {
1749 				SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1750 				    inm_nrele);
1751 				inm_release_locked(inm);
1752 			}
1753 		}
1754 	}
1755 
1756 out_locked:
1757 	IGMP_UNLOCK();
1758 	IN_MULTI_UNLOCK();
1759 }
1760 
1761 /*
1762  * Update host report group timer for IGMPv1/v2.
1763  * Will update the global pending timer flags.
1764  */
1765 static void
1766 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1767 {
1768 	int report_timer_expired;
1769 
1770 	IN_MULTI_LOCK_ASSERT();
1771 	IGMP_LOCK_ASSERT();
1772 
1773 	if (inm->inm_timer == 0) {
1774 		report_timer_expired = 0;
1775 	} else if (--inm->inm_timer == 0) {
1776 		report_timer_expired = 1;
1777 	} else {
1778 		V_current_state_timers_running = 1;
1779 		return;
1780 	}
1781 
1782 	switch (inm->inm_state) {
1783 	case IGMP_NOT_MEMBER:
1784 	case IGMP_SILENT_MEMBER:
1785 	case IGMP_IDLE_MEMBER:
1786 	case IGMP_LAZY_MEMBER:
1787 	case IGMP_SLEEPING_MEMBER:
1788 	case IGMP_AWAKENING_MEMBER:
1789 		break;
1790 	case IGMP_REPORTING_MEMBER:
1791 		if (report_timer_expired) {
1792 			inm->inm_state = IGMP_IDLE_MEMBER;
1793 			(void)igmp_v1v2_queue_report(inm,
1794 			    (version == IGMP_VERSION_2) ?
1795 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1796 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1797 		}
1798 		break;
1799 	case IGMP_G_QUERY_PENDING_MEMBER:
1800 	case IGMP_SG_QUERY_PENDING_MEMBER:
1801 	case IGMP_LEAVING_MEMBER:
1802 		break;
1803 	}
1804 }
1805 
1806 /*
1807  * Update a group's timers for IGMPv3.
1808  * Will update the global pending timer flags.
1809  * Note: Unlocked read from igi.
1810  */
1811 static void
1812 igmp_v3_process_group_timers(struct igmp_ifsoftc *igi,
1813     struct mbufq *qrq, struct mbufq *scq,
1814     struct in_multi *inm, const int uri_fasthz)
1815 {
1816 	int query_response_timer_expired;
1817 	int state_change_retransmit_timer_expired;
1818 
1819 	IN_MULTI_LOCK_ASSERT();
1820 	IGMP_LOCK_ASSERT();
1821 
1822 	query_response_timer_expired = 0;
1823 	state_change_retransmit_timer_expired = 0;
1824 
1825 	/*
1826 	 * During a transition from v1/v2 compatibility mode back to v3,
1827 	 * a group record in REPORTING state may still have its group
1828 	 * timer active. This is a no-op in this function; it is easier
1829 	 * to deal with it here than to complicate the slow-timeout path.
1830 	 */
1831 	if (inm->inm_timer == 0) {
1832 		query_response_timer_expired = 0;
1833 	} else if (--inm->inm_timer == 0) {
1834 		query_response_timer_expired = 1;
1835 	} else {
1836 		V_current_state_timers_running = 1;
1837 	}
1838 
1839 	if (inm->inm_sctimer == 0) {
1840 		state_change_retransmit_timer_expired = 0;
1841 	} else if (--inm->inm_sctimer == 0) {
1842 		state_change_retransmit_timer_expired = 1;
1843 	} else {
1844 		V_state_change_timers_running = 1;
1845 	}
1846 
1847 	/* We are in fasttimo, so be quick about it. */
1848 	if (!state_change_retransmit_timer_expired &&
1849 	    !query_response_timer_expired)
1850 		return;
1851 
1852 	switch (inm->inm_state) {
1853 	case IGMP_NOT_MEMBER:
1854 	case IGMP_SILENT_MEMBER:
1855 	case IGMP_SLEEPING_MEMBER:
1856 	case IGMP_LAZY_MEMBER:
1857 	case IGMP_AWAKENING_MEMBER:
1858 	case IGMP_IDLE_MEMBER:
1859 		break;
1860 	case IGMP_G_QUERY_PENDING_MEMBER:
1861 	case IGMP_SG_QUERY_PENDING_MEMBER:
1862 		/*
1863 		 * Respond to a previously pending Group-Specific
1864 		 * or Group-and-Source-Specific query by enqueueing
1865 		 * the appropriate Current-State report for
1866 		 * immediate transmission.
1867 		 */
1868 		if (query_response_timer_expired) {
1869 			int retval;
1870 
1871 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1872 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1873 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1874 			    __func__, retval);
1875 			inm->inm_state = IGMP_REPORTING_MEMBER;
1876 			/* XXX Clear recorded sources for next time. */
1877 			inm_clear_recorded(inm);
1878 		}
1879 		/* FALLTHROUGH */
1880 	case IGMP_REPORTING_MEMBER:
1881 	case IGMP_LEAVING_MEMBER:
1882 		if (state_change_retransmit_timer_expired) {
1883 			/*
1884 			 * State-change retransmission timer fired.
1885 			 * If there are any further pending retransmissions,
1886 			 * set the global pending state-change flag, and
1887 			 * reset the timer.
1888 			 */
1889 			if (--inm->inm_scrv > 0) {
1890 				inm->inm_sctimer = uri_fasthz;
1891 				V_state_change_timers_running = 1;
1892 			}
1893 			/*
1894 			 * Retransmit the previously computed state-change
1895 			 * report. If there are no further pending
1896 			 * retransmissions, the mbuf queue will be consumed.
1897 			 * Update T0 state to T1 as we have now sent
1898 			 * a state-change.
1899 			 */
1900 			(void)igmp_v3_merge_state_changes(inm, scq);
1901 
1902 			inm_commit(inm);
1903 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1904 			    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1905 
1906 			/*
1907 			 * If we are leaving the group for good, make sure
1908 			 * we release IGMP's reference to it.
1909 			 * This release must be deferred using a SLIST,
1910 			 * as we are called from a loop which traverses
1911 			 * the in_ifmultiaddr TAILQ.
1912 			 */
1913 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1914 			    inm->inm_scrv == 0) {
1915 				inm->inm_state = IGMP_NOT_MEMBER;
1916 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1917 				    inm, inm_nrele);
1918 			}
1919 		}
1920 		break;
1921 	}
1922 }
1923 
1924 
1925 /*
1926  * Suppress a group's pending response to a group or source/group query.
1927  *
1928  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1929  * Do NOT update ST1/ST0 as this operation merely suppresses
1930  * the currently pending group record.
1931  * Do NOT suppress the response to a general query. It is possible but
1932  * it would require adding another state or flag.
1933  */
1934 static void
1935 igmp_v3_suppress_group_record(struct in_multi *inm)
1936 {
1937 
1938 	IN_MULTI_LOCK_ASSERT();
1939 
1940 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1941 		("%s: not IGMPv3 mode on link", __func__));
1942 
1943 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1944 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1945 		return;
1946 
1947 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1948 		inm_clear_recorded(inm);
1949 
1950 	inm->inm_timer = 0;
1951 	inm->inm_state = IGMP_REPORTING_MEMBER;
1952 }
1953 
1954 /*
1955  * Switch to a different IGMP version on the given interface,
1956  * as per Section 7.2.1.
1957  */
1958 static void
1959 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1960 {
1961 	int old_version_timer;
1962 
1963 	IGMP_LOCK_ASSERT();
1964 
1965 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1966 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1967 
1968 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1969 		/*
1970 		 * Compute the "Older Version Querier Present" timer as per
1971 		 * Section 8.12.
1972 		 */
1973 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1974 		old_version_timer *= PR_SLOWHZ;
1975 
1976 		if (version == IGMP_VERSION_1) {
1977 			igi->igi_v1_timer = old_version_timer;
1978 			igi->igi_v2_timer = 0;
1979 		} else if (version == IGMP_VERSION_2) {
1980 			igi->igi_v1_timer = 0;
1981 			igi->igi_v2_timer = old_version_timer;
1982 		}
1983 	}
1984 
1985 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1986 		if (igi->igi_version != IGMP_VERSION_2) {
1987 			igi->igi_version = IGMP_VERSION_2;
1988 			igmp_v3_cancel_link_timers(igi);
1989 		}
1990 	} else if (igi->igi_v1_timer > 0) {
1991 		if (igi->igi_version != IGMP_VERSION_1) {
1992 			igi->igi_version = IGMP_VERSION_1;
1993 			igmp_v3_cancel_link_timers(igi);
1994 		}
1995 	}
1996 }
1997 
1998 /*
1999  * Cancel pending IGMPv3 timers for the given link and all groups
2000  * joined on it; state-change, general-query, and group-query timers.
2001  *
2002  * Only ever called on a transition from v3 to Compatibility mode. Kill
2003  * the timers stone dead (this may be expensive for large N groups), they
2004  * will be restarted if Compatibility Mode deems that they must be due to
2005  * query processing.
2006  */
2007 static void
2008 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2009 {
2010 	struct ifmultiaddr	*ifma;
2011 	struct ifnet		*ifp;
2012 	struct in_multi		*inm, *tinm;
2013 
2014 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2015 	    igi->igi_ifp, igi->igi_ifp->if_xname);
2016 
2017 	IN_MULTI_LOCK_ASSERT();
2018 	IGMP_LOCK_ASSERT();
2019 
2020 	/*
2021 	 * Stop the v3 General Query Response on this link stone dead.
2022 	 * If fasttimo is woken up due to V_interface_timers_running,
2023 	 * the flag will be cleared if there are no pending link timers.
2024 	 */
2025 	igi->igi_v3_timer = 0;
2026 
2027 	/*
2028 	 * Now clear the current-state and state-change report timers
2029 	 * for all memberships scoped to this link.
2030 	 */
2031 	ifp = igi->igi_ifp;
2032 	IF_ADDR_RLOCK(ifp);
2033 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2034 		if (ifma->ifma_addr->sa_family != AF_INET ||
2035 		    ifma->ifma_protospec == NULL)
2036 			continue;
2037 		inm = (struct in_multi *)ifma->ifma_protospec;
2038 		switch (inm->inm_state) {
2039 		case IGMP_NOT_MEMBER:
2040 		case IGMP_SILENT_MEMBER:
2041 		case IGMP_IDLE_MEMBER:
2042 		case IGMP_LAZY_MEMBER:
2043 		case IGMP_SLEEPING_MEMBER:
2044 		case IGMP_AWAKENING_MEMBER:
2045 			/*
2046 			 * These states are either not relevant in v3 mode,
2047 			 * or are unreported. Do nothing.
2048 			 */
2049 			break;
2050 		case IGMP_LEAVING_MEMBER:
2051 			/*
2052 			 * If we are leaving the group and switching to
2053 			 * compatibility mode, we need to release the final
2054 			 * reference held for issuing the INCLUDE {}, and
2055 			 * transition to REPORTING to ensure the host leave
2056 			 * message is sent upstream to the old querier --
2057 			 * transition to NOT would lose the leave and race.
2058 			 */
2059 			SLIST_INSERT_HEAD(&igi->igi_relinmhead, inm, inm_nrele);
2060 			/* FALLTHROUGH */
2061 		case IGMP_G_QUERY_PENDING_MEMBER:
2062 		case IGMP_SG_QUERY_PENDING_MEMBER:
2063 			inm_clear_recorded(inm);
2064 			/* FALLTHROUGH */
2065 		case IGMP_REPORTING_MEMBER:
2066 			inm->inm_state = IGMP_REPORTING_MEMBER;
2067 			break;
2068 		}
2069 		/*
2070 		 * Always clear state-change and group report timers.
2071 		 * Free any pending IGMPv3 state-change records.
2072 		 */
2073 		inm->inm_sctimer = 0;
2074 		inm->inm_timer = 0;
2075 		mbufq_drain(&inm->inm_scq);
2076 	}
2077 	IF_ADDR_RUNLOCK(ifp);
2078 	SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, tinm) {
2079 		SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
2080 		inm_release_locked(inm);
2081 	}
2082 }
2083 
2084 /*
2085  * Update the Older Version Querier Present timers for a link.
2086  * See Section 7.2.1 of RFC 3376.
2087  */
2088 static void
2089 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2090 {
2091 
2092 	IGMP_LOCK_ASSERT();
2093 
2094 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2095 		/*
2096 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2097 		 *
2098 		 * Revert to IGMPv3.
2099 		 */
2100 		if (igi->igi_version != IGMP_VERSION_3) {
2101 			CTR5(KTR_IGMPV3,
2102 			    "%s: transition from v%d -> v%d on %p(%s)",
2103 			    __func__, igi->igi_version, IGMP_VERSION_3,
2104 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2105 			igi->igi_version = IGMP_VERSION_3;
2106 		}
2107 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2108 		/*
2109 		 * IGMPv1 Querier Present timer expired,
2110 		 * IGMPv2 Querier Present timer running.
2111 		 * If IGMPv2 was disabled since last timeout,
2112 		 * revert to IGMPv3.
2113 		 * If IGMPv2 is enabled, revert to IGMPv2.
2114 		 */
2115 		if (!V_igmp_v2enable) {
2116 			CTR5(KTR_IGMPV3,
2117 			    "%s: transition from v%d -> v%d on %p(%s)",
2118 			    __func__, igi->igi_version, IGMP_VERSION_3,
2119 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2120 			igi->igi_v2_timer = 0;
2121 			igi->igi_version = IGMP_VERSION_3;
2122 		} else {
2123 			--igi->igi_v2_timer;
2124 			if (igi->igi_version != IGMP_VERSION_2) {
2125 				CTR5(KTR_IGMPV3,
2126 				    "%s: transition from v%d -> v%d on %p(%s)",
2127 				    __func__, igi->igi_version, IGMP_VERSION_2,
2128 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2129 				igi->igi_version = IGMP_VERSION_2;
2130 				igmp_v3_cancel_link_timers(igi);
2131 			}
2132 		}
2133 	} else if (igi->igi_v1_timer > 0) {
2134 		/*
2135 		 * IGMPv1 Querier Present timer running.
2136 		 * Stop IGMPv2 timer if running.
2137 		 *
2138 		 * If IGMPv1 was disabled since last timeout,
2139 		 * revert to IGMPv3.
2140 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2141 		 */
2142 		if (!V_igmp_v1enable) {
2143 			CTR5(KTR_IGMPV3,
2144 			    "%s: transition from v%d -> v%d on %p(%s)",
2145 			    __func__, igi->igi_version, IGMP_VERSION_3,
2146 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2147 			igi->igi_v1_timer = 0;
2148 			igi->igi_version = IGMP_VERSION_3;
2149 		} else {
2150 			--igi->igi_v1_timer;
2151 		}
2152 		if (igi->igi_v2_timer > 0) {
2153 			CTR3(KTR_IGMPV3,
2154 			    "%s: cancel v2 timer on %p(%s)",
2155 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2156 			igi->igi_v2_timer = 0;
2157 		}
2158 	}
2159 }
2160 
2161 /*
2162  * Global slowtimo handler.
2163  * VIMAGE: Timeout handlers are expected to service all vimages.
2164  */
2165 void
2166 igmp_slowtimo(void)
2167 {
2168 	VNET_ITERATOR_DECL(vnet_iter);
2169 
2170 	VNET_LIST_RLOCK_NOSLEEP();
2171 	VNET_FOREACH(vnet_iter) {
2172 		CURVNET_SET(vnet_iter);
2173 		igmp_slowtimo_vnet();
2174 		CURVNET_RESTORE();
2175 	}
2176 	VNET_LIST_RUNLOCK_NOSLEEP();
2177 }
2178 
2179 /*
2180  * Per-vnet slowtimo handler.
2181  */
2182 static void
2183 igmp_slowtimo_vnet(void)
2184 {
2185 	struct igmp_ifsoftc *igi;
2186 
2187 	IGMP_LOCK();
2188 
2189 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2190 		igmp_v1v2_process_querier_timers(igi);
2191 	}
2192 
2193 	IGMP_UNLOCK();
2194 }
2195 
2196 /*
2197  * Dispatch an IGMPv1/v2 host report or leave message.
2198  * These are always small enough to fit inside a single mbuf.
2199  */
2200 static int
2201 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2202 {
2203 	struct ifnet		*ifp;
2204 	struct igmp		*igmp;
2205 	struct ip		*ip;
2206 	struct mbuf		*m;
2207 
2208 	IN_MULTI_LOCK_ASSERT();
2209 	IGMP_LOCK_ASSERT();
2210 
2211 	ifp = inm->inm_ifp;
2212 
2213 	m = m_gethdr(M_NOWAIT, MT_DATA);
2214 	if (m == NULL)
2215 		return (ENOMEM);
2216 	M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2217 
2218 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2219 
2220 	m->m_data += sizeof(struct ip);
2221 	m->m_len = sizeof(struct igmp);
2222 
2223 	igmp = mtod(m, struct igmp *);
2224 	igmp->igmp_type = type;
2225 	igmp->igmp_code = 0;
2226 	igmp->igmp_group = inm->inm_addr;
2227 	igmp->igmp_cksum = 0;
2228 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2229 
2230 	m->m_data -= sizeof(struct ip);
2231 	m->m_len += sizeof(struct ip);
2232 
2233 	ip = mtod(m, struct ip *);
2234 	ip->ip_tos = 0;
2235 	ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2236 	ip->ip_off = 0;
2237 	ip->ip_p = IPPROTO_IGMP;
2238 	ip->ip_src.s_addr = INADDR_ANY;
2239 
2240 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2241 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2242 	else
2243 		ip->ip_dst = inm->inm_addr;
2244 
2245 	igmp_save_context(m, ifp);
2246 
2247 	m->m_flags |= M_IGMPV2;
2248 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2249 		m->m_flags |= M_IGMP_LOOP;
2250 
2251 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2252 	netisr_dispatch(NETISR_IGMP, m);
2253 
2254 	return (0);
2255 }
2256 
2257 /*
2258  * Process a state change from the upper layer for the given IPv4 group.
2259  *
2260  * Each socket holds a reference on the in_multi in its own ip_moptions.
2261  * The socket layer will have made the necessary updates to.the group
2262  * state, it is now up to IGMP to issue a state change report if there
2263  * has been any change between T0 (when the last state-change was issued)
2264  * and T1 (now).
2265  *
2266  * We use the IGMPv3 state machine at group level. The IGMP module
2267  * however makes the decision as to which IGMP protocol version to speak.
2268  * A state change *from* INCLUDE {} always means an initial join.
2269  * A state change *to* INCLUDE {} always means a final leave.
2270  *
2271  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2272  * save ourselves a bunch of work; any exclusive mode groups need not
2273  * compute source filter lists.
2274  *
2275  * VIMAGE: curvnet should have been set by caller, as this routine
2276  * is called from the socket option handlers.
2277  */
2278 int
2279 igmp_change_state(struct in_multi *inm)
2280 {
2281 	struct igmp_ifsoftc *igi;
2282 	struct ifnet *ifp;
2283 	int error;
2284 
2285 	IN_MULTI_LOCK_ASSERT();
2286 
2287 	error = 0;
2288 
2289 	/*
2290 	 * Try to detect if the upper layer just asked us to change state
2291 	 * for an interface which has now gone away.
2292 	 */
2293 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2294 	ifp = inm->inm_ifma->ifma_ifp;
2295 	/*
2296 	 * Sanity check that netinet's notion of ifp is the
2297 	 * same as net's.
2298 	 */
2299 	KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2300 
2301 	IGMP_LOCK();
2302 
2303 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2304 	KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2305 
2306 	/*
2307 	 * If we detect a state transition to or from MCAST_UNDEFINED
2308 	 * for this group, then we are starting or finishing an IGMP
2309 	 * life cycle for this group.
2310 	 */
2311 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2312 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2313 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2314 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2315 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2316 			error = igmp_initial_join(inm, igi);
2317 			goto out_locked;
2318 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2319 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2320 			igmp_final_leave(inm, igi);
2321 			goto out_locked;
2322 		}
2323 	} else {
2324 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2325 	}
2326 
2327 	error = igmp_handle_state_change(inm, igi);
2328 
2329 out_locked:
2330 	IGMP_UNLOCK();
2331 	return (error);
2332 }
2333 
2334 /*
2335  * Perform the initial join for an IGMP group.
2336  *
2337  * When joining a group:
2338  *  If the group should have its IGMP traffic suppressed, do nothing.
2339  *  IGMPv1 starts sending IGMPv1 host membership reports.
2340  *  IGMPv2 starts sending IGMPv2 host membership reports.
2341  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2342  *  initial state of the membership.
2343  */
2344 static int
2345 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2346 {
2347 	struct ifnet		*ifp;
2348 	struct mbufq		*mq;
2349 	int			 error, retval, syncstates;
2350 
2351 	CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2352 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2353 	    inm->inm_ifp->if_xname);
2354 
2355 	error = 0;
2356 	syncstates = 1;
2357 
2358 	ifp = inm->inm_ifp;
2359 
2360 	IN_MULTI_LOCK_ASSERT();
2361 	IGMP_LOCK_ASSERT();
2362 
2363 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2364 
2365 	/*
2366 	 * Groups joined on loopback or marked as 'not reported',
2367 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2368 	 * are never reported in any IGMP protocol exchanges.
2369 	 * All other groups enter the appropriate IGMP state machine
2370 	 * for the version in use on this link.
2371 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2372 	 * disabled for the link.
2373 	 */
2374 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2375 	    (igi->igi_flags & IGIF_SILENT) ||
2376 	    !igmp_isgroupreported(inm->inm_addr)) {
2377 		CTR1(KTR_IGMPV3,
2378 "%s: not kicking state machine for silent group", __func__);
2379 		inm->inm_state = IGMP_SILENT_MEMBER;
2380 		inm->inm_timer = 0;
2381 	} else {
2382 		/*
2383 		 * Deal with overlapping in_multi lifecycle.
2384 		 * If this group was LEAVING, then make sure
2385 		 * we drop the reference we picked up to keep the
2386 		 * group around for the final INCLUDE {} enqueue.
2387 		 */
2388 		if (igi->igi_version == IGMP_VERSION_3 &&
2389 		    inm->inm_state == IGMP_LEAVING_MEMBER)
2390 			inm_release_locked(inm);
2391 
2392 		inm->inm_state = IGMP_REPORTING_MEMBER;
2393 
2394 		switch (igi->igi_version) {
2395 		case IGMP_VERSION_1:
2396 		case IGMP_VERSION_2:
2397 			inm->inm_state = IGMP_IDLE_MEMBER;
2398 			error = igmp_v1v2_queue_report(inm,
2399 			    (igi->igi_version == IGMP_VERSION_2) ?
2400 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2401 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2402 			if (error == 0) {
2403 				inm->inm_timer = IGMP_RANDOM_DELAY(
2404 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2405 				V_current_state_timers_running = 1;
2406 			}
2407 			break;
2408 
2409 		case IGMP_VERSION_3:
2410 			/*
2411 			 * Defer update of T0 to T1, until the first copy
2412 			 * of the state change has been transmitted.
2413 			 */
2414 			syncstates = 0;
2415 
2416 			/*
2417 			 * Immediately enqueue a State-Change Report for
2418 			 * this interface, freeing any previous reports.
2419 			 * Don't kick the timers if there is nothing to do,
2420 			 * or if an error occurred.
2421 			 */
2422 			mq = &inm->inm_scq;
2423 			mbufq_drain(mq);
2424 			retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2425 			    0, 0);
2426 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2427 			    __func__, retval);
2428 			if (retval <= 0) {
2429 				error = retval * -1;
2430 				break;
2431 			}
2432 
2433 			/*
2434 			 * Schedule transmission of pending state-change
2435 			 * report up to RV times for this link. The timer
2436 			 * will fire at the next igmp_fasttimo (~200ms),
2437 			 * giving us an opportunity to merge the reports.
2438 			 */
2439 			if (igi->igi_flags & IGIF_LOOPBACK) {
2440 				inm->inm_scrv = 1;
2441 			} else {
2442 				KASSERT(igi->igi_rv > 1,
2443 				   ("%s: invalid robustness %d", __func__,
2444 				    igi->igi_rv));
2445 				inm->inm_scrv = igi->igi_rv;
2446 			}
2447 			inm->inm_sctimer = 1;
2448 			V_state_change_timers_running = 1;
2449 
2450 			error = 0;
2451 			break;
2452 		}
2453 	}
2454 
2455 	/*
2456 	 * Only update the T0 state if state change is atomic,
2457 	 * i.e. we don't need to wait for a timer to fire before we
2458 	 * can consider the state change to have been communicated.
2459 	 */
2460 	if (syncstates) {
2461 		inm_commit(inm);
2462 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2463 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2464 	}
2465 
2466 	return (error);
2467 }
2468 
2469 /*
2470  * Issue an intermediate state change during the IGMP life-cycle.
2471  */
2472 static int
2473 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2474 {
2475 	struct ifnet		*ifp;
2476 	int			 retval;
2477 
2478 	CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2479 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2480 	    inm->inm_ifp->if_xname);
2481 
2482 	ifp = inm->inm_ifp;
2483 
2484 	IN_MULTI_LOCK_ASSERT();
2485 	IGMP_LOCK_ASSERT();
2486 
2487 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2488 
2489 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2490 	    (igi->igi_flags & IGIF_SILENT) ||
2491 	    !igmp_isgroupreported(inm->inm_addr) ||
2492 	    (igi->igi_version != IGMP_VERSION_3)) {
2493 		if (!igmp_isgroupreported(inm->inm_addr)) {
2494 			CTR1(KTR_IGMPV3,
2495 "%s: not kicking state machine for silent group", __func__);
2496 		}
2497 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2498 		inm_commit(inm);
2499 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2500 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2501 		return (0);
2502 	}
2503 
2504 	mbufq_drain(&inm->inm_scq);
2505 
2506 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2507 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2508 	if (retval <= 0)
2509 		return (-retval);
2510 
2511 	/*
2512 	 * If record(s) were enqueued, start the state-change
2513 	 * report timer for this group.
2514 	 */
2515 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2516 	inm->inm_sctimer = 1;
2517 	V_state_change_timers_running = 1;
2518 
2519 	return (0);
2520 }
2521 
2522 /*
2523  * Perform the final leave for an IGMP group.
2524  *
2525  * When leaving a group:
2526  *  IGMPv1 does nothing.
2527  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2528  *  IGMPv3 enqueues a state-change report containing a transition
2529  *  to INCLUDE {} for immediate transmission.
2530  */
2531 static void
2532 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2533 {
2534 	int syncstates;
2535 
2536 	syncstates = 1;
2537 
2538 	CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2539 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2540 	    inm->inm_ifp->if_xname);
2541 
2542 	IN_MULTI_LOCK_ASSERT();
2543 	IGMP_LOCK_ASSERT();
2544 
2545 	switch (inm->inm_state) {
2546 	case IGMP_NOT_MEMBER:
2547 	case IGMP_SILENT_MEMBER:
2548 	case IGMP_LEAVING_MEMBER:
2549 		/* Already leaving or left; do nothing. */
2550 		CTR1(KTR_IGMPV3,
2551 "%s: not kicking state machine for silent group", __func__);
2552 		break;
2553 	case IGMP_REPORTING_MEMBER:
2554 	case IGMP_IDLE_MEMBER:
2555 	case IGMP_G_QUERY_PENDING_MEMBER:
2556 	case IGMP_SG_QUERY_PENDING_MEMBER:
2557 		if (igi->igi_version == IGMP_VERSION_2) {
2558 #ifdef INVARIANTS
2559 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2560 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2561 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2562 			     __func__);
2563 #endif
2564 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2565 			inm->inm_state = IGMP_NOT_MEMBER;
2566 		} else if (igi->igi_version == IGMP_VERSION_3) {
2567 			/*
2568 			 * Stop group timer and all pending reports.
2569 			 * Immediately enqueue a state-change report
2570 			 * TO_IN {} to be sent on the next fast timeout,
2571 			 * giving us an opportunity to merge reports.
2572 			 */
2573 			mbufq_drain(&inm->inm_scq);
2574 			inm->inm_timer = 0;
2575 			if (igi->igi_flags & IGIF_LOOPBACK) {
2576 				inm->inm_scrv = 1;
2577 			} else {
2578 				inm->inm_scrv = igi->igi_rv;
2579 			}
2580 			CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2581 			    "pending retransmissions.", __func__,
2582 			    inet_ntoa(inm->inm_addr),
2583 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2584 			if (inm->inm_scrv == 0) {
2585 				inm->inm_state = IGMP_NOT_MEMBER;
2586 				inm->inm_sctimer = 0;
2587 			} else {
2588 				int retval;
2589 
2590 				inm_acquire_locked(inm);
2591 
2592 				retval = igmp_v3_enqueue_group_record(
2593 				    &inm->inm_scq, inm, 1, 0, 0);
2594 				KASSERT(retval != 0,
2595 				    ("%s: enqueue record = %d", __func__,
2596 				     retval));
2597 
2598 				inm->inm_state = IGMP_LEAVING_MEMBER;
2599 				inm->inm_sctimer = 1;
2600 				V_state_change_timers_running = 1;
2601 				syncstates = 0;
2602 			}
2603 			break;
2604 		}
2605 		break;
2606 	case IGMP_LAZY_MEMBER:
2607 	case IGMP_SLEEPING_MEMBER:
2608 	case IGMP_AWAKENING_MEMBER:
2609 		/* Our reports are suppressed; do nothing. */
2610 		break;
2611 	}
2612 
2613 	if (syncstates) {
2614 		inm_commit(inm);
2615 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2616 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2617 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2618 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2619 		    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2620 	}
2621 }
2622 
2623 /*
2624  * Enqueue an IGMPv3 group record to the given output queue.
2625  *
2626  * XXX This function could do with having the allocation code
2627  * split out, and the multiple-tree-walks coalesced into a single
2628  * routine as has been done in igmp_v3_enqueue_filter_change().
2629  *
2630  * If is_state_change is zero, a current-state record is appended.
2631  * If is_state_change is non-zero, a state-change report is appended.
2632  *
2633  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2634  * If is_group_query is zero, and if there is a packet with free space
2635  * at the tail of the queue, it will be appended to providing there
2636  * is enough free space.
2637  * Otherwise a new mbuf packet chain is allocated.
2638  *
2639  * If is_source_query is non-zero, each source is checked to see if
2640  * it was recorded for a Group-Source query, and will be omitted if
2641  * it is not both in-mode and recorded.
2642  *
2643  * The function will attempt to allocate leading space in the packet
2644  * for the IP/IGMP header to be prepended without fragmenting the chain.
2645  *
2646  * If successful the size of all data appended to the queue is returned,
2647  * otherwise an error code less than zero is returned, or zero if
2648  * no record(s) were appended.
2649  */
2650 static int
2651 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2652     const int is_state_change, const int is_group_query,
2653     const int is_source_query)
2654 {
2655 	struct igmp_grouprec	 ig;
2656 	struct igmp_grouprec	*pig;
2657 	struct ifnet		*ifp;
2658 	struct ip_msource	*ims, *nims;
2659 	struct mbuf		*m0, *m, *md;
2660 	int			 error, is_filter_list_change;
2661 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2662 	int			 record_has_sources;
2663 	int			 now;
2664 	int			 type;
2665 	in_addr_t		 naddr;
2666 	uint8_t			 mode;
2667 
2668 	IN_MULTI_LOCK_ASSERT();
2669 
2670 	error = 0;
2671 	ifp = inm->inm_ifp;
2672 	is_filter_list_change = 0;
2673 	m = NULL;
2674 	m0 = NULL;
2675 	m0srcs = 0;
2676 	msrcs = 0;
2677 	nbytes = 0;
2678 	nims = NULL;
2679 	record_has_sources = 1;
2680 	pig = NULL;
2681 	type = IGMP_DO_NOTHING;
2682 	mode = inm->inm_st[1].iss_fmode;
2683 
2684 	/*
2685 	 * If we did not transition out of ASM mode during t0->t1,
2686 	 * and there are no source nodes to process, we can skip
2687 	 * the generation of source records.
2688 	 */
2689 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2690 	    inm->inm_nsrc == 0)
2691 		record_has_sources = 0;
2692 
2693 	if (is_state_change) {
2694 		/*
2695 		 * Queue a state change record.
2696 		 * If the mode did not change, and there are non-ASM
2697 		 * listeners or source filters present,
2698 		 * we potentially need to issue two records for the group.
2699 		 * If we are transitioning to MCAST_UNDEFINED, we need
2700 		 * not send any sources.
2701 		 * If there are ASM listeners, and there was no filter
2702 		 * mode transition of any kind, do nothing.
2703 		 */
2704 		if (mode != inm->inm_st[0].iss_fmode) {
2705 			if (mode == MCAST_EXCLUDE) {
2706 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2707 				    __func__);
2708 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2709 			} else {
2710 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2711 				    __func__);
2712 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2713 				if (mode == MCAST_UNDEFINED)
2714 					record_has_sources = 0;
2715 			}
2716 		} else {
2717 			if (record_has_sources) {
2718 				is_filter_list_change = 1;
2719 			} else {
2720 				type = IGMP_DO_NOTHING;
2721 			}
2722 		}
2723 	} else {
2724 		/*
2725 		 * Queue a current state record.
2726 		 */
2727 		if (mode == MCAST_EXCLUDE) {
2728 			type = IGMP_MODE_IS_EXCLUDE;
2729 		} else if (mode == MCAST_INCLUDE) {
2730 			type = IGMP_MODE_IS_INCLUDE;
2731 			KASSERT(inm->inm_st[1].iss_asm == 0,
2732 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2733 			     __func__, inm, inm->inm_st[1].iss_asm));
2734 		}
2735 	}
2736 
2737 	/*
2738 	 * Generate the filter list changes using a separate function.
2739 	 */
2740 	if (is_filter_list_change)
2741 		return (igmp_v3_enqueue_filter_change(mq, inm));
2742 
2743 	if (type == IGMP_DO_NOTHING) {
2744 		CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2745 		    __func__, inet_ntoa(inm->inm_addr),
2746 		    inm->inm_ifp->if_xname);
2747 		return (0);
2748 	}
2749 
2750 	/*
2751 	 * If any sources are present, we must be able to fit at least
2752 	 * one in the trailing space of the tail packet's mbuf,
2753 	 * ideally more.
2754 	 */
2755 	minrec0len = sizeof(struct igmp_grouprec);
2756 	if (record_has_sources)
2757 		minrec0len += sizeof(in_addr_t);
2758 
2759 	CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2760 	    igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2761 	    inm->inm_ifp->if_xname);
2762 
2763 	/*
2764 	 * Check if we have a packet in the tail of the queue for this
2765 	 * group into which the first group record for this group will fit.
2766 	 * Otherwise allocate a new packet.
2767 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2768 	 * Note: Group records for G/GSR query responses MUST be sent
2769 	 * in their own packet.
2770 	 */
2771 	m0 = mbufq_last(mq);
2772 	if (!is_group_query &&
2773 	    m0 != NULL &&
2774 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2775 	    (m0->m_pkthdr.len + minrec0len) <
2776 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2777 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2778 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2779 		m = m0;
2780 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2781 	} else {
2782 		if (mbufq_full(mq)) {
2783 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2784 			return (-ENOMEM);
2785 		}
2786 		m = NULL;
2787 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2788 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2789 		if (!is_state_change && !is_group_query) {
2790 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2791 			if (m)
2792 				m->m_data += IGMP_LEADINGSPACE;
2793 		}
2794 		if (m == NULL) {
2795 			m = m_gethdr(M_NOWAIT, MT_DATA);
2796 			if (m)
2797 				M_ALIGN(m, IGMP_LEADINGSPACE);
2798 		}
2799 		if (m == NULL)
2800 			return (-ENOMEM);
2801 
2802 		igmp_save_context(m, ifp);
2803 
2804 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2805 	}
2806 
2807 	/*
2808 	 * Append group record.
2809 	 * If we have sources, we don't know how many yet.
2810 	 */
2811 	ig.ig_type = type;
2812 	ig.ig_datalen = 0;
2813 	ig.ig_numsrc = 0;
2814 	ig.ig_group = inm->inm_addr;
2815 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2816 		if (m != m0)
2817 			m_freem(m);
2818 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2819 		return (-ENOMEM);
2820 	}
2821 	nbytes += sizeof(struct igmp_grouprec);
2822 
2823 	/*
2824 	 * Append as many sources as will fit in the first packet.
2825 	 * If we are appending to a new packet, the chain allocation
2826 	 * may potentially use clusters; use m_getptr() in this case.
2827 	 * If we are appending to an existing packet, we need to obtain
2828 	 * a pointer to the group record after m_append(), in case a new
2829 	 * mbuf was allocated.
2830 	 * Only append sources which are in-mode at t1. If we are
2831 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2832 	 * include source entries.
2833 	 * Only report recorded sources in our filter set when responding
2834 	 * to a group-source query.
2835 	 */
2836 	if (record_has_sources) {
2837 		if (m == m0) {
2838 			md = m_last(m);
2839 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2840 			    md->m_len - nbytes);
2841 		} else {
2842 			md = m_getptr(m, 0, &off);
2843 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2844 			    off);
2845 		}
2846 		msrcs = 0;
2847 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2848 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2849 			    inet_ntoa_haddr(ims->ims_haddr));
2850 			now = ims_get_mode(inm, ims, 1);
2851 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2852 			if ((now != mode) ||
2853 			    (now == mode && mode == MCAST_UNDEFINED)) {
2854 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2855 				continue;
2856 			}
2857 			if (is_source_query && ims->ims_stp == 0) {
2858 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2859 				    __func__);
2860 				continue;
2861 			}
2862 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2863 			naddr = htonl(ims->ims_haddr);
2864 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2865 				if (m != m0)
2866 					m_freem(m);
2867 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2868 				    __func__);
2869 				return (-ENOMEM);
2870 			}
2871 			nbytes += sizeof(in_addr_t);
2872 			++msrcs;
2873 			if (msrcs == m0srcs)
2874 				break;
2875 		}
2876 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2877 		    msrcs);
2878 		pig->ig_numsrc = htons(msrcs);
2879 		nbytes += (msrcs * sizeof(in_addr_t));
2880 	}
2881 
2882 	if (is_source_query && msrcs == 0) {
2883 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2884 		if (m != m0)
2885 			m_freem(m);
2886 		return (0);
2887 	}
2888 
2889 	/*
2890 	 * We are good to go with first packet.
2891 	 */
2892 	if (m != m0) {
2893 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2894 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2895 		mbufq_enqueue(mq, m);
2896 	} else
2897 		m->m_pkthdr.PH_vt.vt_nrecs++;
2898 
2899 	/*
2900 	 * No further work needed if no source list in packet(s).
2901 	 */
2902 	if (!record_has_sources)
2903 		return (nbytes);
2904 
2905 	/*
2906 	 * Whilst sources remain to be announced, we need to allocate
2907 	 * a new packet and fill out as many sources as will fit.
2908 	 * Always try for a cluster first.
2909 	 */
2910 	while (nims != NULL) {
2911 		if (mbufq_full(mq)) {
2912 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2913 			return (-ENOMEM);
2914 		}
2915 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2916 		if (m)
2917 			m->m_data += IGMP_LEADINGSPACE;
2918 		if (m == NULL) {
2919 			m = m_gethdr(M_NOWAIT, MT_DATA);
2920 			if (m)
2921 				M_ALIGN(m, IGMP_LEADINGSPACE);
2922 		}
2923 		if (m == NULL)
2924 			return (-ENOMEM);
2925 		igmp_save_context(m, ifp);
2926 		md = m_getptr(m, 0, &off);
2927 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2928 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2929 
2930 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2931 			if (m != m0)
2932 				m_freem(m);
2933 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2934 			return (-ENOMEM);
2935 		}
2936 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2937 		nbytes += sizeof(struct igmp_grouprec);
2938 
2939 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2940 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2941 
2942 		msrcs = 0;
2943 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2944 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2945 			    inet_ntoa_haddr(ims->ims_haddr));
2946 			now = ims_get_mode(inm, ims, 1);
2947 			if ((now != mode) ||
2948 			    (now == mode && mode == MCAST_UNDEFINED)) {
2949 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2950 				continue;
2951 			}
2952 			if (is_source_query && ims->ims_stp == 0) {
2953 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2954 				    __func__);
2955 				continue;
2956 			}
2957 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2958 			naddr = htonl(ims->ims_haddr);
2959 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2960 				if (m != m0)
2961 					m_freem(m);
2962 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2963 				    __func__);
2964 				return (-ENOMEM);
2965 			}
2966 			++msrcs;
2967 			if (msrcs == m0srcs)
2968 				break;
2969 		}
2970 		pig->ig_numsrc = htons(msrcs);
2971 		nbytes += (msrcs * sizeof(in_addr_t));
2972 
2973 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2974 		mbufq_enqueue(mq, m);
2975 	}
2976 
2977 	return (nbytes);
2978 }
2979 
2980 /*
2981  * Type used to mark record pass completion.
2982  * We exploit the fact we can cast to this easily from the
2983  * current filter modes on each ip_msource node.
2984  */
2985 typedef enum {
2986 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2987 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2988 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2989 	REC_FULL = REC_ALLOW | REC_BLOCK
2990 } rectype_t;
2991 
2992 /*
2993  * Enqueue an IGMPv3 filter list change to the given output queue.
2994  *
2995  * Source list filter state is held in an RB-tree. When the filter list
2996  * for a group is changed without changing its mode, we need to compute
2997  * the deltas between T0 and T1 for each source in the filter set,
2998  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2999  *
3000  * As we may potentially queue two record types, and the entire R-B tree
3001  * needs to be walked at once, we break this out into its own function
3002  * so we can generate a tightly packed queue of packets.
3003  *
3004  * XXX This could be written to only use one tree walk, although that makes
3005  * serializing into the mbuf chains a bit harder. For now we do two walks
3006  * which makes things easier on us, and it may or may not be harder on
3007  * the L2 cache.
3008  *
3009  * If successful the size of all data appended to the queue is returned,
3010  * otherwise an error code less than zero is returned, or zero if
3011  * no record(s) were appended.
3012  */
3013 static int
3014 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3015 {
3016 	static const int MINRECLEN =
3017 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3018 	struct ifnet		*ifp;
3019 	struct igmp_grouprec	 ig;
3020 	struct igmp_grouprec	*pig;
3021 	struct ip_msource	*ims, *nims;
3022 	struct mbuf		*m, *m0, *md;
3023 	in_addr_t		 naddr;
3024 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3025 	int			 nallow, nblock;
3026 	uint8_t			 mode, now, then;
3027 	rectype_t		 crt, drt, nrt;
3028 
3029 	IN_MULTI_LOCK_ASSERT();
3030 
3031 	if (inm->inm_nsrc == 0 ||
3032 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3033 		return (0);
3034 
3035 	ifp = inm->inm_ifp;			/* interface */
3036 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
3037 	crt = REC_NONE;	/* current group record type */
3038 	drt = REC_NONE;	/* mask of completed group record types */
3039 	nrt = REC_NONE;	/* record type for current node */
3040 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3041 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3042 	npbytes = 0;	/* # of bytes appended this packet */
3043 	rsrcs = 0;	/* # sources encoded in current record */
3044 	schanged = 0;	/* # nodes encoded in overall filter change */
3045 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3046 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3047 	nims = NULL;	/* next tree node pointer */
3048 
3049 	/*
3050 	 * For each possible filter record mode.
3051 	 * The first kind of source we encounter tells us which
3052 	 * is the first kind of record we start appending.
3053 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3054 	 * as the inverse of the group's filter mode.
3055 	 */
3056 	while (drt != REC_FULL) {
3057 		do {
3058 			m0 = mbufq_last(mq);
3059 			if (m0 != NULL &&
3060 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3061 			     IGMP_V3_REPORT_MAXRECS) &&
3062 			    (m0->m_pkthdr.len + MINRECLEN) <
3063 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3064 				m = m0;
3065 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3066 					    sizeof(struct igmp_grouprec)) /
3067 				    sizeof(in_addr_t);
3068 				CTR1(KTR_IGMPV3,
3069 				    "%s: use previous packet", __func__);
3070 			} else {
3071 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3072 				if (m)
3073 					m->m_data += IGMP_LEADINGSPACE;
3074 				if (m == NULL) {
3075 					m = m_gethdr(M_NOWAIT, MT_DATA);
3076 					if (m)
3077 						M_ALIGN(m, IGMP_LEADINGSPACE);
3078 				}
3079 				if (m == NULL) {
3080 					CTR1(KTR_IGMPV3,
3081 					    "%s: m_get*() failed", __func__);
3082 					return (-ENOMEM);
3083 				}
3084 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3085 				igmp_save_context(m, ifp);
3086 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3087 				    sizeof(struct igmp_grouprec)) /
3088 				    sizeof(in_addr_t);
3089 				npbytes = 0;
3090 				CTR1(KTR_IGMPV3,
3091 				    "%s: allocated new packet", __func__);
3092 			}
3093 			/*
3094 			 * Append the IGMP group record header to the
3095 			 * current packet's data area.
3096 			 * Recalculate pointer to free space for next
3097 			 * group record, in case m_append() allocated
3098 			 * a new mbuf or cluster.
3099 			 */
3100 			memset(&ig, 0, sizeof(ig));
3101 			ig.ig_group = inm->inm_addr;
3102 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3103 				if (m != m0)
3104 					m_freem(m);
3105 				CTR1(KTR_IGMPV3,
3106 				    "%s: m_append() failed", __func__);
3107 				return (-ENOMEM);
3108 			}
3109 			npbytes += sizeof(struct igmp_grouprec);
3110 			if (m != m0) {
3111 				/* new packet; offset in c hain */
3112 				md = m_getptr(m, npbytes -
3113 				    sizeof(struct igmp_grouprec), &off);
3114 				pig = (struct igmp_grouprec *)(mtod(md,
3115 				    uint8_t *) + off);
3116 			} else {
3117 				/* current packet; offset from last append */
3118 				md = m_last(m);
3119 				pig = (struct igmp_grouprec *)(mtod(md,
3120 				    uint8_t *) + md->m_len -
3121 				    sizeof(struct igmp_grouprec));
3122 			}
3123 			/*
3124 			 * Begin walking the tree for this record type
3125 			 * pass, or continue from where we left off
3126 			 * previously if we had to allocate a new packet.
3127 			 * Only report deltas in-mode at t1.
3128 			 * We need not report included sources as allowed
3129 			 * if we are in inclusive mode on the group,
3130 			 * however the converse is not true.
3131 			 */
3132 			rsrcs = 0;
3133 			if (nims == NULL)
3134 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3135 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3136 				CTR2(KTR_IGMPV3, "%s: visit node %s",
3137 				    __func__, inet_ntoa_haddr(ims->ims_haddr));
3138 				now = ims_get_mode(inm, ims, 1);
3139 				then = ims_get_mode(inm, ims, 0);
3140 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3141 				    __func__, then, now);
3142 				if (now == then) {
3143 					CTR1(KTR_IGMPV3,
3144 					    "%s: skip unchanged", __func__);
3145 					continue;
3146 				}
3147 				if (mode == MCAST_EXCLUDE &&
3148 				    now == MCAST_INCLUDE) {
3149 					CTR1(KTR_IGMPV3,
3150 					    "%s: skip IN src on EX group",
3151 					    __func__);
3152 					continue;
3153 				}
3154 				nrt = (rectype_t)now;
3155 				if (nrt == REC_NONE)
3156 					nrt = (rectype_t)(~mode & REC_FULL);
3157 				if (schanged++ == 0) {
3158 					crt = nrt;
3159 				} else if (crt != nrt)
3160 					continue;
3161 				naddr = htonl(ims->ims_haddr);
3162 				if (!m_append(m, sizeof(in_addr_t),
3163 				    (void *)&naddr)) {
3164 					if (m != m0)
3165 						m_freem(m);
3166 					CTR1(KTR_IGMPV3,
3167 					    "%s: m_append() failed", __func__);
3168 					return (-ENOMEM);
3169 				}
3170 				nallow += !!(crt == REC_ALLOW);
3171 				nblock += !!(crt == REC_BLOCK);
3172 				if (++rsrcs == m0srcs)
3173 					break;
3174 			}
3175 			/*
3176 			 * If we did not append any tree nodes on this
3177 			 * pass, back out of allocations.
3178 			 */
3179 			if (rsrcs == 0) {
3180 				npbytes -= sizeof(struct igmp_grouprec);
3181 				if (m != m0) {
3182 					CTR1(KTR_IGMPV3,
3183 					    "%s: m_free(m)", __func__);
3184 					m_freem(m);
3185 				} else {
3186 					CTR1(KTR_IGMPV3,
3187 					    "%s: m_adj(m, -ig)", __func__);
3188 					m_adj(m, -((int)sizeof(
3189 					    struct igmp_grouprec)));
3190 				}
3191 				continue;
3192 			}
3193 			npbytes += (rsrcs * sizeof(in_addr_t));
3194 			if (crt == REC_ALLOW)
3195 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3196 			else if (crt == REC_BLOCK)
3197 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3198 			pig->ig_numsrc = htons(rsrcs);
3199 			/*
3200 			 * Count the new group record, and enqueue this
3201 			 * packet if it wasn't already queued.
3202 			 */
3203 			m->m_pkthdr.PH_vt.vt_nrecs++;
3204 			if (m != m0)
3205 				mbufq_enqueue(mq, m);
3206 			nbytes += npbytes;
3207 		} while (nims != NULL);
3208 		drt |= crt;
3209 		crt = (~crt & REC_FULL);
3210 	}
3211 
3212 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3213 	    nallow, nblock);
3214 
3215 	return (nbytes);
3216 }
3217 
3218 static int
3219 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3220 {
3221 	struct mbufq	*gq;
3222 	struct mbuf	*m;		/* pending state-change */
3223 	struct mbuf	*m0;		/* copy of pending state-change */
3224 	struct mbuf	*mt;		/* last state-change in packet */
3225 	int		 docopy, domerge;
3226 	u_int		 recslen;
3227 
3228 	docopy = 0;
3229 	domerge = 0;
3230 	recslen = 0;
3231 
3232 	IN_MULTI_LOCK_ASSERT();
3233 	IGMP_LOCK_ASSERT();
3234 
3235 	/*
3236 	 * If there are further pending retransmissions, make a writable
3237 	 * copy of each queued state-change message before merging.
3238 	 */
3239 	if (inm->inm_scrv > 0)
3240 		docopy = 1;
3241 
3242 	gq = &inm->inm_scq;
3243 #ifdef KTR
3244 	if (mbufq_first(gq) == NULL) {
3245 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3246 		    __func__, inm);
3247 	}
3248 #endif
3249 
3250 	m = mbufq_first(gq);
3251 	while (m != NULL) {
3252 		/*
3253 		 * Only merge the report into the current packet if
3254 		 * there is sufficient space to do so; an IGMPv3 report
3255 		 * packet may only contain 65,535 group records.
3256 		 * Always use a simple mbuf chain concatentation to do this,
3257 		 * as large state changes for single groups may have
3258 		 * allocated clusters.
3259 		 */
3260 		domerge = 0;
3261 		mt = mbufq_last(scq);
3262 		if (mt != NULL) {
3263 			recslen = m_length(m, NULL);
3264 
3265 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3266 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3267 			    IGMP_V3_REPORT_MAXRECS) &&
3268 			    (mt->m_pkthdr.len + recslen <=
3269 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3270 				domerge = 1;
3271 		}
3272 
3273 		if (!domerge && mbufq_full(gq)) {
3274 			CTR2(KTR_IGMPV3,
3275 			    "%s: outbound queue full, skipping whole packet %p",
3276 			    __func__, m);
3277 			mt = m->m_nextpkt;
3278 			if (!docopy)
3279 				m_freem(m);
3280 			m = mt;
3281 			continue;
3282 		}
3283 
3284 		if (!docopy) {
3285 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3286 			m0 = mbufq_dequeue(gq);
3287 			m = m0->m_nextpkt;
3288 		} else {
3289 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3290 			m0 = m_dup(m, M_NOWAIT);
3291 			if (m0 == NULL)
3292 				return (ENOMEM);
3293 			m0->m_nextpkt = NULL;
3294 			m = m->m_nextpkt;
3295 		}
3296 
3297 		if (!domerge) {
3298 			CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3299 			    __func__, m0, scq);
3300 			mbufq_enqueue(scq, m0);
3301 		} else {
3302 			struct mbuf *mtl;	/* last mbuf of packet mt */
3303 
3304 			CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3305 			    __func__, m0, mt);
3306 
3307 			mtl = m_last(mt);
3308 			m0->m_flags &= ~M_PKTHDR;
3309 			mt->m_pkthdr.len += recslen;
3310 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3311 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3312 
3313 			mtl->m_next = m0;
3314 		}
3315 	}
3316 
3317 	return (0);
3318 }
3319 
3320 /*
3321  * Respond to a pending IGMPv3 General Query.
3322  */
3323 static void
3324 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3325 {
3326 	struct ifmultiaddr	*ifma;
3327 	struct ifnet		*ifp;
3328 	struct in_multi		*inm;
3329 	int			 retval, loop;
3330 
3331 	IN_MULTI_LOCK_ASSERT();
3332 	IGMP_LOCK_ASSERT();
3333 
3334 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3335 	    ("%s: called when version %d", __func__, igi->igi_version));
3336 
3337 	ifp = igi->igi_ifp;
3338 
3339 	IF_ADDR_RLOCK(ifp);
3340 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3341 		if (ifma->ifma_addr->sa_family != AF_INET ||
3342 		    ifma->ifma_protospec == NULL)
3343 			continue;
3344 
3345 		inm = (struct in_multi *)ifma->ifma_protospec;
3346 		KASSERT(ifp == inm->inm_ifp,
3347 		    ("%s: inconsistent ifp", __func__));
3348 
3349 		switch (inm->inm_state) {
3350 		case IGMP_NOT_MEMBER:
3351 		case IGMP_SILENT_MEMBER:
3352 			break;
3353 		case IGMP_REPORTING_MEMBER:
3354 		case IGMP_IDLE_MEMBER:
3355 		case IGMP_LAZY_MEMBER:
3356 		case IGMP_SLEEPING_MEMBER:
3357 		case IGMP_AWAKENING_MEMBER:
3358 			inm->inm_state = IGMP_REPORTING_MEMBER;
3359 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3360 			    inm, 0, 0, 0);
3361 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3362 			    __func__, retval);
3363 			break;
3364 		case IGMP_G_QUERY_PENDING_MEMBER:
3365 		case IGMP_SG_QUERY_PENDING_MEMBER:
3366 		case IGMP_LEAVING_MEMBER:
3367 			break;
3368 		}
3369 	}
3370 	IF_ADDR_RUNLOCK(ifp);
3371 
3372 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3373 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3374 
3375 	/*
3376 	 * Slew transmission of bursts over 500ms intervals.
3377 	 */
3378 	if (mbufq_first(&igi->igi_gq) != NULL) {
3379 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3380 		    IGMP_RESPONSE_BURST_INTERVAL);
3381 		V_interface_timers_running = 1;
3382 	}
3383 }
3384 
3385 /*
3386  * Transmit the next pending IGMP message in the output queue.
3387  *
3388  * We get called from netisr_processqueue(). A mutex private to igmpoq
3389  * will be acquired and released around this routine.
3390  *
3391  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3392  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3393  * a link and uses a link-scope multicast address.
3394  */
3395 static void
3396 igmp_intr(struct mbuf *m)
3397 {
3398 	struct ip_moptions	 imo;
3399 	struct ifnet		*ifp;
3400 	struct mbuf		*ipopts, *m0;
3401 	int			 error;
3402 	uint32_t		 ifindex;
3403 
3404 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3405 
3406 	/*
3407 	 * Set VNET image pointer from enqueued mbuf chain
3408 	 * before doing anything else. Whilst we use interface
3409 	 * indexes to guard against interface detach, they are
3410 	 * unique to each VIMAGE and must be retrieved.
3411 	 */
3412 	CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3413 	ifindex = igmp_restore_context(m);
3414 
3415 	/*
3416 	 * Check if the ifnet still exists. This limits the scope of
3417 	 * any race in the absence of a global ifp lock for low cost
3418 	 * (an array lookup).
3419 	 */
3420 	ifp = ifnet_byindex(ifindex);
3421 	if (ifp == NULL) {
3422 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3423 		    __func__, m, ifindex);
3424 		m_freem(m);
3425 		IPSTAT_INC(ips_noroute);
3426 		goto out;
3427 	}
3428 
3429 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3430 
3431 	imo.imo_multicast_ttl  = 1;
3432 	imo.imo_multicast_vif  = -1;
3433 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3434 
3435 	/*
3436 	 * If the user requested that IGMP traffic be explicitly
3437 	 * redirected to the loopback interface (e.g. they are running a
3438 	 * MANET interface and the routing protocol needs to see the
3439 	 * updates), handle this now.
3440 	 */
3441 	if (m->m_flags & M_IGMP_LOOP)
3442 		imo.imo_multicast_ifp = V_loif;
3443 	else
3444 		imo.imo_multicast_ifp = ifp;
3445 
3446 	if (m->m_flags & M_IGMPV2) {
3447 		m0 = m;
3448 	} else {
3449 		m0 = igmp_v3_encap_report(ifp, m);
3450 		if (m0 == NULL) {
3451 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3452 			m_freem(m);
3453 			IPSTAT_INC(ips_odropped);
3454 			goto out;
3455 		}
3456 	}
3457 
3458 	igmp_scrub_context(m0);
3459 	m_clrprotoflags(m);
3460 	m0->m_pkthdr.rcvif = V_loif;
3461 #ifdef MAC
3462 	mac_netinet_igmp_send(ifp, m0);
3463 #endif
3464 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3465 	if (error) {
3466 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3467 		goto out;
3468 	}
3469 
3470 	IGMPSTAT_INC(igps_snd_reports);
3471 
3472 out:
3473 	/*
3474 	 * We must restore the existing vnet pointer before
3475 	 * continuing as we are run from netisr context.
3476 	 */
3477 	CURVNET_RESTORE();
3478 }
3479 
3480 /*
3481  * Encapsulate an IGMPv3 report.
3482  *
3483  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3484  * chain has already had its IP/IGMPv3 header prepended. In this case
3485  * the function will not attempt to prepend; the lengths and checksums
3486  * will however be re-computed.
3487  *
3488  * Returns a pointer to the new mbuf chain head, or NULL if the
3489  * allocation failed.
3490  */
3491 static struct mbuf *
3492 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3493 {
3494 	struct rm_priotracker	in_ifa_tracker;
3495 	struct igmp_report	*igmp;
3496 	struct ip		*ip;
3497 	int			 hdrlen, igmpreclen;
3498 
3499 	KASSERT((m->m_flags & M_PKTHDR),
3500 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3501 
3502 	igmpreclen = m_length(m, NULL);
3503 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3504 
3505 	if (m->m_flags & M_IGMPV3_HDR) {
3506 		igmpreclen -= hdrlen;
3507 	} else {
3508 		M_PREPEND(m, hdrlen, M_NOWAIT);
3509 		if (m == NULL)
3510 			return (NULL);
3511 		m->m_flags |= M_IGMPV3_HDR;
3512 	}
3513 
3514 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3515 
3516 	m->m_data += sizeof(struct ip);
3517 	m->m_len -= sizeof(struct ip);
3518 
3519 	igmp = mtod(m, struct igmp_report *);
3520 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3521 	igmp->ir_rsv1 = 0;
3522 	igmp->ir_rsv2 = 0;
3523 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3524 	igmp->ir_cksum = 0;
3525 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3526 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3527 
3528 	m->m_data -= sizeof(struct ip);
3529 	m->m_len += sizeof(struct ip);
3530 
3531 	ip = mtod(m, struct ip *);
3532 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3533 	ip->ip_len = htons(hdrlen + igmpreclen);
3534 	ip->ip_off = htons(IP_DF);
3535 	ip->ip_p = IPPROTO_IGMP;
3536 	ip->ip_sum = 0;
3537 
3538 	ip->ip_src.s_addr = INADDR_ANY;
3539 
3540 	if (m->m_flags & M_IGMP_LOOP) {
3541 		struct in_ifaddr *ia;
3542 
3543 		IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3544 		if (ia != NULL) {
3545 			ip->ip_src = ia->ia_addr.sin_addr;
3546 			ifa_free(&ia->ia_ifa);
3547 		}
3548 	}
3549 
3550 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3551 
3552 	return (m);
3553 }
3554 
3555 #ifdef KTR
3556 static char *
3557 igmp_rec_type_to_str(const int type)
3558 {
3559 
3560 	switch (type) {
3561 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3562 			return "TO_EX";
3563 			break;
3564 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3565 			return "TO_IN";
3566 			break;
3567 		case IGMP_MODE_IS_EXCLUDE:
3568 			return "MODE_EX";
3569 			break;
3570 		case IGMP_MODE_IS_INCLUDE:
3571 			return "MODE_IN";
3572 			break;
3573 		case IGMP_ALLOW_NEW_SOURCES:
3574 			return "ALLOW_NEW";
3575 			break;
3576 		case IGMP_BLOCK_OLD_SOURCES:
3577 			return "BLOCK_OLD";
3578 			break;
3579 		default:
3580 			break;
3581 	}
3582 	return "unknown";
3583 }
3584 #endif
3585 
3586 static void
3587 igmp_init(void *unused __unused)
3588 {
3589 
3590 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3591 
3592 	IGMP_LOCK_INIT();
3593 
3594 	m_raopt = igmp_ra_alloc();
3595 
3596 	netisr_register(&igmp_nh);
3597 }
3598 SYSINIT(igmp_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_init, NULL);
3599 
3600 static void
3601 igmp_uninit(void *unused __unused)
3602 {
3603 
3604 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3605 
3606 	netisr_unregister(&igmp_nh);
3607 
3608 	m_free(m_raopt);
3609 	m_raopt = NULL;
3610 
3611 	IGMP_LOCK_DESTROY();
3612 }
3613 SYSUNINIT(igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_uninit, NULL);
3614 
3615 static void
3616 vnet_igmp_init(const void *unused __unused)
3617 {
3618 
3619 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3620 
3621 	LIST_INIT(&V_igi_head);
3622 }
3623 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_igmp_init,
3624     NULL);
3625 
3626 static void
3627 vnet_igmp_uninit(const void *unused __unused)
3628 {
3629 
3630 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3631 
3632 	KASSERT(LIST_EMPTY(&V_igi_head),
3633 	    ("%s: igi list not empty; ifnets not detached?", __func__));
3634 }
3635 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY,
3636     vnet_igmp_uninit, NULL);
3637 
3638 static int
3639 igmp_modevent(module_t mod, int type, void *unused __unused)
3640 {
3641 
3642     switch (type) {
3643     case MOD_LOAD:
3644     case MOD_UNLOAD:
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