xref: /freebsd/sys/netinet/igmp.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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/sysctl.h>
62 #include <sys/ktr.h>
63 #include <sys/condvar.h>
64 
65 #include <net/if.h>
66 #include <net/netisr.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_var.h>
71 #include <netinet/in_systm.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_options.h>
75 #include <netinet/igmp.h>
76 #include <netinet/igmp_var.h>
77 
78 #include <machine/in_cksum.h>
79 
80 #include <security/mac/mac_framework.h>
81 
82 #ifndef KTR_IGMPV3
83 #define KTR_IGMPV3 KTR_INET
84 #endif
85 
86 static struct igmp_ifinfo *
87 		igi_alloc_locked(struct ifnet *);
88 static void	igi_delete_locked(const struct ifnet *);
89 static void	igmp_dispatch_queue(struct ifqueue *, int, const int);
90 static void	igmp_fasttimo_vnet(void);
91 static void	igmp_final_leave(struct in_multi *, struct igmp_ifinfo *);
92 static int	igmp_handle_state_change(struct in_multi *,
93 		    struct igmp_ifinfo *);
94 static int	igmp_initial_join(struct in_multi *, struct igmp_ifinfo *);
95 static int	igmp_input_v1_query(struct ifnet *, const struct ip *,
96 		    const struct igmp *);
97 static int	igmp_input_v2_query(struct ifnet *, const struct ip *,
98 		    const struct igmp *);
99 static int	igmp_input_v3_query(struct ifnet *, const struct ip *,
100 		    /*const*/ struct igmpv3 *);
101 static int	igmp_input_v3_group_query(struct in_multi *,
102 		    struct igmp_ifinfo *, int, /*const*/ struct igmpv3 *);
103 static int	igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
104 		    /*const*/ struct igmp *);
105 static int	igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
106 		    /*const*/ struct igmp *);
107 static void	igmp_intr(struct mbuf *);
108 static int	igmp_isgroupreported(const struct in_addr);
109 static struct mbuf *
110 		igmp_ra_alloc(void);
111 #ifdef KTR
112 static char *	igmp_rec_type_to_str(const int);
113 #endif
114 static void	igmp_set_version(struct igmp_ifinfo *, const int);
115 static void	igmp_slowtimo_vnet(void);
116 static int	igmp_v1v2_queue_report(struct in_multi *, const int);
117 static void	igmp_v1v2_process_group_timer(struct in_multi *, const int);
118 static void	igmp_v1v2_process_querier_timers(struct igmp_ifinfo *);
119 static void	igmp_v2_update_group(struct in_multi *, const int);
120 static void	igmp_v3_cancel_link_timers(struct igmp_ifinfo *);
121 static void	igmp_v3_dispatch_general_query(struct igmp_ifinfo *);
122 static struct mbuf *
123 		igmp_v3_encap_report(struct ifnet *, struct mbuf *);
124 static int	igmp_v3_enqueue_group_record(struct ifqueue *,
125 		    struct in_multi *, const int, const int, const int);
126 static int	igmp_v3_enqueue_filter_change(struct ifqueue *,
127 		    struct in_multi *);
128 static void	igmp_v3_process_group_timers(struct igmp_ifinfo *,
129 		    struct ifqueue *, struct ifqueue *, struct in_multi *,
130 		    const int);
131 static int	igmp_v3_merge_state_changes(struct in_multi *,
132 		    struct ifqueue *);
133 static void	igmp_v3_suppress_group_record(struct in_multi *);
134 static int	sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
135 static int	sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
136 static int	sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
137 
138 static const struct netisr_handler igmp_nh = {
139 	.nh_name = "igmp",
140 	.nh_handler = igmp_intr,
141 	.nh_proto = NETISR_IGMP,
142 	.nh_policy = NETISR_POLICY_SOURCE,
143 };
144 
145 /*
146  * System-wide globals.
147  *
148  * Unlocked access to these is OK, except for the global IGMP output
149  * queue. The IGMP subsystem lock ends up being system-wide for the moment,
150  * because all VIMAGEs have to share a global output queue, as netisrs
151  * themselves are not virtualized.
152  *
153  * Locking:
154  *  * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
155  *    Any may be taken independently; if any are held at the same
156  *    time, the above lock order must be followed.
157  *  * All output is delegated to the netisr.
158  *    Now that Giant has been eliminated, the netisr may be inlined.
159  *  * IN_MULTI_LOCK covers in_multi.
160  *  * IGMP_LOCK covers igmp_ifinfo and any global variables in this file,
161  *    including the output queue.
162  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
163  *    per-link state iterators.
164  *  * igmp_ifinfo is valid as long as PF_INET is attached to the interface,
165  *    therefore it is not refcounted.
166  *    We allow unlocked reads of igmp_ifinfo when accessed via in_multi.
167  *
168  * Reference counting
169  *  * IGMP acquires its own reference every time an in_multi is passed to
170  *    it and the group is being joined for the first time.
171  *  * IGMP releases its reference(s) on in_multi in a deferred way,
172  *    because the operations which process the release run as part of
173  *    a loop whose control variables are directly affected by the release
174  *    (that, and not recursing on the IF_ADDR_LOCK).
175  *
176  * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
177  * to a vnet in ifp->if_vnet.
178  *
179  * SMPng: XXX We may potentially race operations on ifma_protospec.
180  * The problem is that we currently lack a clean way of taking the
181  * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
182  * as anything which modifies ifma needs to be covered by that lock.
183  * So check for ifma_protospec being NULL before proceeding.
184  */
185 struct mtx		 igmp_mtx;
186 
187 struct mbuf		*m_raopt;		 /* Router Alert option */
188 MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
189 
190 /*
191  * VIMAGE-wide globals.
192  *
193  * The IGMPv3 timers themselves need to run per-image, however,
194  * protosw timers run globally (see tcp).
195  * An ifnet can only be in one vimage at a time, and the loopback
196  * ifnet, loif, is itself virtualized.
197  * It would otherwise be possible to seriously hose IGMP state,
198  * and create inconsistencies in upstream multicast routing, if you have
199  * multiple VIMAGEs running on the same link joining different multicast
200  * groups, UNLESS the "primary IP address" is different. This is because
201  * IGMP for IPv4 does not force link-local addresses to be used for each
202  * node, unlike MLD for IPv6.
203  * Obviously the IGMPv3 per-interface state has per-vimage granularity
204  * also as a result.
205  *
206  * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
207  * policy to control the address used by IGMP on the link.
208  */
209 static VNET_DEFINE(int, interface_timers_running);	/* IGMPv3 general
210 							 * query response */
211 static VNET_DEFINE(int, state_change_timers_running);	/* IGMPv3 state-change
212 							 * retransmit */
213 static VNET_DEFINE(int, current_state_timers_running);	/* IGMPv1/v2 host
214 							 * report; IGMPv3 g/sg
215 							 * query response */
216 
217 #define	V_interface_timers_running	VNET(interface_timers_running)
218 #define	V_state_change_timers_running	VNET(state_change_timers_running)
219 #define	V_current_state_timers_running	VNET(current_state_timers_running)
220 
221 static VNET_DEFINE(LIST_HEAD(, igmp_ifinfo), igi_head);
222 static VNET_DEFINE(struct igmpstat, igmpstat) = {
223 	.igps_version = IGPS_VERSION_3,
224 	.igps_len = sizeof(struct igmpstat),
225 };
226 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
227 
228 #define	V_igi_head			VNET(igi_head)
229 #define	V_igmpstat			VNET(igmpstat)
230 #define	V_igmp_gsrdelay			VNET(igmp_gsrdelay)
231 
232 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
233 static VNET_DEFINE(int, igmp_sendra) = 1;
234 static VNET_DEFINE(int, igmp_sendlocal) = 1;
235 static VNET_DEFINE(int, igmp_v1enable) = 1;
236 static VNET_DEFINE(int, igmp_v2enable) = 1;
237 static VNET_DEFINE(int, igmp_legacysupp);
238 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
239 
240 #define	V_igmp_recvifkludge		VNET(igmp_recvifkludge)
241 #define	V_igmp_sendra			VNET(igmp_sendra)
242 #define	V_igmp_sendlocal		VNET(igmp_sendlocal)
243 #define	V_igmp_v1enable			VNET(igmp_v1enable)
244 #define	V_igmp_v2enable			VNET(igmp_v2enable)
245 #define	V_igmp_legacysupp		VNET(igmp_legacysupp)
246 #define	V_igmp_default_version		VNET(igmp_default_version)
247 
248 /*
249  * Virtualized sysctls.
250  */
251 SYSCTL_VNET_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_RW,
252     &VNET_NAME(igmpstat), igmpstat, "");
253 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_RW,
254     &VNET_NAME(igmp_recvifkludge), 0,
255     "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
256 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_RW,
257     &VNET_NAME(igmp_sendra), 0,
258     "Send IP Router Alert option in IGMPv2/v3 messages");
259 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_RW,
260     &VNET_NAME(igmp_sendlocal), 0,
261     "Send IGMP membership reports for 224.0.0.0/24 groups");
262 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_RW,
263     &VNET_NAME(igmp_v1enable), 0,
264     "Enable backwards compatibility with IGMPv1");
265 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_RW,
266     &VNET_NAME(igmp_v2enable), 0,
267     "Enable backwards compatibility with IGMPv2");
268 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_RW,
269     &VNET_NAME(igmp_legacysupp), 0,
270     "Allow v1/v2 reports to suppress v3 group responses");
271 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, default_version,
272     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
273     &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
274     "Default version of IGMP to run on each interface");
275 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
276     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
277     &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
278     "Rate limit for IGMPv3 Group-and-Source queries in seconds");
279 
280 /*
281  * Non-virtualized sysctls.
282  */
283 SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
284     sysctl_igmp_ifinfo, "Per-interface IGMPv3 state");
285 
286 static __inline void
287 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
288 {
289 
290 #ifdef VIMAGE
291 	m->m_pkthdr.header = ifp->if_vnet;
292 #endif /* VIMAGE */
293 	m->m_pkthdr.flowid = ifp->if_index;
294 }
295 
296 static __inline void
297 igmp_scrub_context(struct mbuf *m)
298 {
299 
300 	m->m_pkthdr.header = NULL;
301 	m->m_pkthdr.flowid = 0;
302 }
303 
304 #ifdef KTR
305 static __inline char *
306 inet_ntoa_haddr(in_addr_t haddr)
307 {
308 	struct in_addr ia;
309 
310 	ia.s_addr = htonl(haddr);
311 	return (inet_ntoa(ia));
312 }
313 #endif
314 
315 /*
316  * Restore context from a queued IGMP output chain.
317  * Return saved ifindex.
318  *
319  * VIMAGE: The assertion is there to make sure that we
320  * actually called CURVNET_SET() with what's in the mbuf chain.
321  */
322 static __inline uint32_t
323 igmp_restore_context(struct mbuf *m)
324 {
325 
326 #ifdef notyet
327 #if defined(VIMAGE) && defined(INVARIANTS)
328 	KASSERT(curvnet == (m->m_pkthdr.header),
329 	    ("%s: called when curvnet was not restored", __func__));
330 #endif
331 #endif
332 	return (m->m_pkthdr.flowid);
333 }
334 
335 /*
336  * Retrieve or set default IGMP version.
337  *
338  * VIMAGE: Assume curvnet set by caller.
339  * SMPng: NOTE: Serialized by IGMP lock.
340  */
341 static int
342 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
343 {
344 	int	 error;
345 	int	 new;
346 
347 	error = sysctl_wire_old_buffer(req, sizeof(int));
348 	if (error)
349 		return (error);
350 
351 	IGMP_LOCK();
352 
353 	new = V_igmp_default_version;
354 
355 	error = sysctl_handle_int(oidp, &new, 0, req);
356 	if (error || !req->newptr)
357 		goto out_locked;
358 
359 	if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
360 		error = EINVAL;
361 		goto out_locked;
362 	}
363 
364 	CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
365 	     V_igmp_default_version, new);
366 
367 	V_igmp_default_version = new;
368 
369 out_locked:
370 	IGMP_UNLOCK();
371 	return (error);
372 }
373 
374 /*
375  * Retrieve or set threshold between group-source queries in seconds.
376  *
377  * VIMAGE: Assume curvnet set by caller.
378  * SMPng: NOTE: Serialized by IGMP lock.
379  */
380 static int
381 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
382 {
383 	int error;
384 	int i;
385 
386 	error = sysctl_wire_old_buffer(req, sizeof(int));
387 	if (error)
388 		return (error);
389 
390 	IGMP_LOCK();
391 
392 	i = V_igmp_gsrdelay.tv_sec;
393 
394 	error = sysctl_handle_int(oidp, &i, 0, req);
395 	if (error || !req->newptr)
396 		goto out_locked;
397 
398 	if (i < -1 || i >= 60) {
399 		error = EINVAL;
400 		goto out_locked;
401 	}
402 
403 	CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
404 	     V_igmp_gsrdelay.tv_sec, i);
405 	V_igmp_gsrdelay.tv_sec = i;
406 
407 out_locked:
408 	IGMP_UNLOCK();
409 	return (error);
410 }
411 
412 /*
413  * Expose struct igmp_ifinfo to userland, keyed by ifindex.
414  * For use by ifmcstat(8).
415  *
416  * SMPng: NOTE: Does an unlocked ifindex space read.
417  * VIMAGE: Assume curvnet set by caller. The node handler itself
418  * is not directly virtualized.
419  */
420 static int
421 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
422 {
423 	int			*name;
424 	int			 error;
425 	u_int			 namelen;
426 	struct ifnet		*ifp;
427 	struct igmp_ifinfo	*igi;
428 
429 	name = (int *)arg1;
430 	namelen = arg2;
431 
432 	if (req->newptr != NULL)
433 		return (EPERM);
434 
435 	if (namelen != 1)
436 		return (EINVAL);
437 
438 	error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
439 	if (error)
440 		return (error);
441 
442 	IN_MULTI_LOCK();
443 	IGMP_LOCK();
444 
445 	if (name[0] <= 0 || name[0] > V_if_index) {
446 		error = ENOENT;
447 		goto out_locked;
448 	}
449 
450 	error = ENOENT;
451 
452 	ifp = ifnet_byindex(name[0]);
453 	if (ifp == NULL)
454 		goto out_locked;
455 
456 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
457 		if (ifp == igi->igi_ifp) {
458 			error = SYSCTL_OUT(req, igi,
459 			    sizeof(struct igmp_ifinfo));
460 			break;
461 		}
462 	}
463 
464 out_locked:
465 	IGMP_UNLOCK();
466 	IN_MULTI_UNLOCK();
467 	return (error);
468 }
469 
470 /*
471  * Dispatch an entire queue of pending packet chains
472  * using the netisr.
473  * VIMAGE: Assumes the vnet pointer has been set.
474  */
475 static void
476 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop)
477 {
478 	struct mbuf *m;
479 
480 	for (;;) {
481 		_IF_DEQUEUE(ifq, m);
482 		if (m == NULL)
483 			break;
484 		CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m);
485 		if (loop)
486 			m->m_flags |= M_IGMP_LOOP;
487 		netisr_dispatch(NETISR_IGMP, m);
488 		if (--limit == 0)
489 			break;
490 	}
491 }
492 
493 /*
494  * Filter outgoing IGMP report state by group.
495  *
496  * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
497  * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
498  * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
499  * this may break certain IGMP snooping switches which rely on the old
500  * report behaviour.
501  *
502  * Return zero if the given group is one for which IGMP reports
503  * should be suppressed, or non-zero if reports should be issued.
504  */
505 static __inline int
506 igmp_isgroupreported(const struct in_addr addr)
507 {
508 
509 	if (in_allhosts(addr) ||
510 	    ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
511 		return (0);
512 
513 	return (1);
514 }
515 
516 /*
517  * Construct a Router Alert option to use in outgoing packets.
518  */
519 static struct mbuf *
520 igmp_ra_alloc(void)
521 {
522 	struct mbuf	*m;
523 	struct ipoption	*p;
524 
525 	MGET(m, M_DONTWAIT, MT_DATA);
526 	p = mtod(m, struct ipoption *);
527 	p->ipopt_dst.s_addr = INADDR_ANY;
528 	p->ipopt_list[0] = IPOPT_RA;	/* Router Alert Option */
529 	p->ipopt_list[1] = 0x04;	/* 4 bytes long */
530 	p->ipopt_list[2] = IPOPT_EOL;	/* End of IP option list */
531 	p->ipopt_list[3] = 0x00;	/* pad byte */
532 	m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
533 
534 	return (m);
535 }
536 
537 /*
538  * Attach IGMP when PF_INET is attached to an interface.
539  */
540 struct igmp_ifinfo *
541 igmp_domifattach(struct ifnet *ifp)
542 {
543 	struct igmp_ifinfo *igi;
544 
545 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
546 	    __func__, ifp, ifp->if_xname);
547 
548 	IGMP_LOCK();
549 
550 	igi = igi_alloc_locked(ifp);
551 	if (!(ifp->if_flags & IFF_MULTICAST))
552 		igi->igi_flags |= IGIF_SILENT;
553 
554 	IGMP_UNLOCK();
555 
556 	return (igi);
557 }
558 
559 /*
560  * VIMAGE: assume curvnet set by caller.
561  */
562 static struct igmp_ifinfo *
563 igi_alloc_locked(/*const*/ struct ifnet *ifp)
564 {
565 	struct igmp_ifinfo *igi;
566 
567 	IGMP_LOCK_ASSERT();
568 
569 	igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO);
570 	if (igi == NULL)
571 		goto out;
572 
573 	igi->igi_ifp = ifp;
574 	igi->igi_version = V_igmp_default_version;
575 	igi->igi_flags = 0;
576 	igi->igi_rv = IGMP_RV_INIT;
577 	igi->igi_qi = IGMP_QI_INIT;
578 	igi->igi_qri = IGMP_QRI_INIT;
579 	igi->igi_uri = IGMP_URI_INIT;
580 
581 	SLIST_INIT(&igi->igi_relinmhead);
582 
583 	/*
584 	 * Responses to general queries are subject to bounds.
585 	 */
586 	IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
587 
588 	LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
589 
590 	CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)",
591 	     ifp, ifp->if_xname);
592 
593 out:
594 	return (igi);
595 }
596 
597 /*
598  * Hook for ifdetach.
599  *
600  * NOTE: Some finalization tasks need to run before the protocol domain
601  * is detached, but also before the link layer does its cleanup.
602  *
603  * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
604  * XXX This is also bitten by unlocked ifma_protospec access.
605  */
606 void
607 igmp_ifdetach(struct ifnet *ifp)
608 {
609 	struct igmp_ifinfo	*igi;
610 	struct ifmultiaddr	*ifma;
611 	struct in_multi		*inm, *tinm;
612 
613 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
614 	    ifp->if_xname);
615 
616 	IGMP_LOCK();
617 
618 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
619 	if (igi->igi_version == IGMP_VERSION_3) {
620 		IF_ADDR_LOCK(ifp);
621 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
622 			if (ifma->ifma_addr->sa_family != AF_INET ||
623 			    ifma->ifma_protospec == NULL)
624 				continue;
625 #if 0
626 			KASSERT(ifma->ifma_protospec != NULL,
627 			    ("%s: ifma_protospec is NULL", __func__));
628 #endif
629 			inm = (struct in_multi *)ifma->ifma_protospec;
630 			if (inm->inm_state == IGMP_LEAVING_MEMBER) {
631 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
632 				    inm, inm_nrele);
633 			}
634 			inm_clear_recorded(inm);
635 		}
636 		IF_ADDR_UNLOCK(ifp);
637 		/*
638 		 * Free the in_multi reference(s) for this IGMP lifecycle.
639 		 */
640 		SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
641 		    tinm) {
642 			SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
643 			inm_release_locked(inm);
644 		}
645 	}
646 
647 	IGMP_UNLOCK();
648 }
649 
650 /*
651  * Hook for domifdetach.
652  */
653 void
654 igmp_domifdetach(struct ifnet *ifp)
655 {
656 	struct igmp_ifinfo *igi;
657 
658 	CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
659 	    __func__, ifp, ifp->if_xname);
660 
661 	IGMP_LOCK();
662 
663 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
664 	igi_delete_locked(ifp);
665 
666 	IGMP_UNLOCK();
667 }
668 
669 static void
670 igi_delete_locked(const struct ifnet *ifp)
671 {
672 	struct igmp_ifinfo *igi, *tigi;
673 
674 	CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
675 	    __func__, ifp, ifp->if_xname);
676 
677 	IGMP_LOCK_ASSERT();
678 
679 	LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
680 		if (igi->igi_ifp == ifp) {
681 			/*
682 			 * Free deferred General Query responses.
683 			 */
684 			_IF_DRAIN(&igi->igi_gq);
685 
686 			LIST_REMOVE(igi, igi_link);
687 
688 			KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
689 			    ("%s: there are dangling in_multi references",
690 			    __func__));
691 
692 			free(igi, M_IGMP);
693 			return;
694 		}
695 	}
696 
697 #ifdef INVARIANTS
698 	panic("%s: igmp_ifinfo not found for ifp %p\n", __func__,  ifp);
699 #endif
700 }
701 
702 /*
703  * Process a received IGMPv1 query.
704  * Return non-zero if the message should be dropped.
705  *
706  * VIMAGE: The curvnet pointer is derived from the input ifp.
707  */
708 static int
709 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
710     const struct igmp *igmp)
711 {
712 	struct ifmultiaddr	*ifma;
713 	struct igmp_ifinfo	*igi;
714 	struct in_multi		*inm;
715 
716 	/*
717 	 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
718 	 * 224.0.0.1. They are always treated as General Queries.
719 	 * igmp_group is always ignored. Do not drop it as a userland
720 	 * daemon may wish to see it.
721 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
722 	 */
723 	if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
724 		IGMPSTAT_INC(igps_rcv_badqueries);
725 		return (0);
726 	}
727 	IGMPSTAT_INC(igps_rcv_gen_queries);
728 
729 	IN_MULTI_LOCK();
730 	IGMP_LOCK();
731 
732 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
733 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
734 
735 	if (igi->igi_flags & IGIF_LOOPBACK) {
736 		CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
737 		    ifp, ifp->if_xname);
738 		goto out_locked;
739 	}
740 
741 	/*
742 	 * Switch to IGMPv1 host compatibility mode.
743 	 */
744 	igmp_set_version(igi, IGMP_VERSION_1);
745 
746 	CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
747 
748 	/*
749 	 * Start the timers in all of our group records
750 	 * for the interface on which the query arrived,
751 	 * except those which are already running.
752 	 */
753 	IF_ADDR_LOCK(ifp);
754 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
755 		if (ifma->ifma_addr->sa_family != AF_INET ||
756 		    ifma->ifma_protospec == NULL)
757 			continue;
758 		inm = (struct in_multi *)ifma->ifma_protospec;
759 		if (inm->inm_timer != 0)
760 			continue;
761 		switch (inm->inm_state) {
762 		case IGMP_NOT_MEMBER:
763 		case IGMP_SILENT_MEMBER:
764 			break;
765 		case IGMP_G_QUERY_PENDING_MEMBER:
766 		case IGMP_SG_QUERY_PENDING_MEMBER:
767 		case IGMP_REPORTING_MEMBER:
768 		case IGMP_IDLE_MEMBER:
769 		case IGMP_LAZY_MEMBER:
770 		case IGMP_SLEEPING_MEMBER:
771 		case IGMP_AWAKENING_MEMBER:
772 			inm->inm_state = IGMP_REPORTING_MEMBER;
773 			inm->inm_timer = IGMP_RANDOM_DELAY(
774 			    IGMP_V1V2_MAX_RI * PR_FASTHZ);
775 			V_current_state_timers_running = 1;
776 			break;
777 		case IGMP_LEAVING_MEMBER:
778 			break;
779 		}
780 	}
781 	IF_ADDR_UNLOCK(ifp);
782 
783 out_locked:
784 	IGMP_UNLOCK();
785 	IN_MULTI_UNLOCK();
786 
787 	return (0);
788 }
789 
790 /*
791  * Process a received IGMPv2 general or group-specific query.
792  */
793 static int
794 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
795     const struct igmp *igmp)
796 {
797 	struct ifmultiaddr	*ifma;
798 	struct igmp_ifinfo	*igi;
799 	struct in_multi		*inm;
800 	int			 is_general_query;
801 	uint16_t		 timer;
802 
803 	is_general_query = 0;
804 
805 	/*
806 	 * Validate address fields upfront.
807 	 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
808 	 */
809 	if (in_nullhost(igmp->igmp_group)) {
810 		/*
811 		 * IGMPv2 General Query.
812 		 * If this was not sent to the all-hosts group, ignore it.
813 		 */
814 		if (!in_allhosts(ip->ip_dst))
815 			return (0);
816 		IGMPSTAT_INC(igps_rcv_gen_queries);
817 		is_general_query = 1;
818 	} else {
819 		/* IGMPv2 Group-Specific Query. */
820 		IGMPSTAT_INC(igps_rcv_group_queries);
821 	}
822 
823 	IN_MULTI_LOCK();
824 	IGMP_LOCK();
825 
826 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
827 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
828 
829 	if (igi->igi_flags & IGIF_LOOPBACK) {
830 		CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
831 		    ifp, ifp->if_xname);
832 		goto out_locked;
833 	}
834 
835 	/*
836 	 * Ignore v2 query if in v1 Compatibility Mode.
837 	 */
838 	if (igi->igi_version == IGMP_VERSION_1)
839 		goto out_locked;
840 
841 	igmp_set_version(igi, IGMP_VERSION_2);
842 
843 	timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
844 	if (timer == 0)
845 		timer = 1;
846 
847 	if (is_general_query) {
848 		/*
849 		 * For each reporting group joined on this
850 		 * interface, kick the report timer.
851 		 */
852 		CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
853 		    ifp, ifp->if_xname);
854 		IF_ADDR_LOCK(ifp);
855 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
856 			if (ifma->ifma_addr->sa_family != AF_INET ||
857 			    ifma->ifma_protospec == NULL)
858 				continue;
859 			inm = (struct in_multi *)ifma->ifma_protospec;
860 			igmp_v2_update_group(inm, timer);
861 		}
862 		IF_ADDR_UNLOCK(ifp);
863 	} else {
864 		/*
865 		 * Group-specific IGMPv2 query, we need only
866 		 * look up the single group to process it.
867 		 */
868 		inm = inm_lookup(ifp, igmp->igmp_group);
869 		if (inm != NULL) {
870 			CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
871 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
872 			igmp_v2_update_group(inm, timer);
873 		}
874 	}
875 
876 out_locked:
877 	IGMP_UNLOCK();
878 	IN_MULTI_UNLOCK();
879 
880 	return (0);
881 }
882 
883 /*
884  * Update the report timer on a group in response to an IGMPv2 query.
885  *
886  * If we are becoming the reporting member for this group, start the timer.
887  * If we already are the reporting member for this group, and timer is
888  * below the threshold, reset it.
889  *
890  * We may be updating the group for the first time since we switched
891  * to IGMPv3. If we are, then we must clear any recorded source lists,
892  * and transition to REPORTING state; the group timer is overloaded
893  * for group and group-source query responses.
894  *
895  * Unlike IGMPv3, the delay per group should be jittered
896  * to avoid bursts of IGMPv2 reports.
897  */
898 static void
899 igmp_v2_update_group(struct in_multi *inm, const int timer)
900 {
901 
902 	CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
903 	    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
904 
905 	IN_MULTI_LOCK_ASSERT();
906 
907 	switch (inm->inm_state) {
908 	case IGMP_NOT_MEMBER:
909 	case IGMP_SILENT_MEMBER:
910 		break;
911 	case IGMP_REPORTING_MEMBER:
912 		if (inm->inm_timer != 0 &&
913 		    inm->inm_timer <= timer) {
914 			CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
915 			    "skipping.", __func__);
916 			break;
917 		}
918 		/* FALLTHROUGH */
919 	case IGMP_SG_QUERY_PENDING_MEMBER:
920 	case IGMP_G_QUERY_PENDING_MEMBER:
921 	case IGMP_IDLE_MEMBER:
922 	case IGMP_LAZY_MEMBER:
923 	case IGMP_AWAKENING_MEMBER:
924 		CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
925 		inm->inm_state = IGMP_REPORTING_MEMBER;
926 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
927 		V_current_state_timers_running = 1;
928 		break;
929 	case IGMP_SLEEPING_MEMBER:
930 		CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
931 		inm->inm_state = IGMP_AWAKENING_MEMBER;
932 		break;
933 	case IGMP_LEAVING_MEMBER:
934 		break;
935 	}
936 }
937 
938 /*
939  * Process a received IGMPv3 general, group-specific or
940  * group-and-source-specific query.
941  * Assumes m has already been pulled up to the full IGMP message length.
942  * Return 0 if successful, otherwise an appropriate error code is returned.
943  */
944 static int
945 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
946     /*const*/ struct igmpv3 *igmpv3)
947 {
948 	struct igmp_ifinfo	*igi;
949 	struct in_multi		*inm;
950 	int			 is_general_query;
951 	uint32_t		 maxresp, nsrc, qqi;
952 	uint16_t		 timer;
953 	uint8_t			 qrv;
954 
955 	is_general_query = 0;
956 
957 	CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
958 
959 	maxresp = igmpv3->igmp_code;	/* in 1/10ths of a second */
960 	if (maxresp >= 128) {
961 		maxresp = IGMP_MANT(igmpv3->igmp_code) <<
962 			  (IGMP_EXP(igmpv3->igmp_code) + 3);
963 	}
964 
965 	/*
966 	 * Robustness must never be less than 2 for on-wire IGMPv3.
967 	 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
968 	 * an exception for interfaces whose IGMPv3 state changes
969 	 * are redirected to loopback (e.g. MANET).
970 	 */
971 	qrv = IGMP_QRV(igmpv3->igmp_misc);
972 	if (qrv < 2) {
973 		CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
974 		    qrv, IGMP_RV_INIT);
975 		qrv = IGMP_RV_INIT;
976 	}
977 
978 	qqi = igmpv3->igmp_qqi;
979 	if (qqi >= 128) {
980 		qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
981 		     (IGMP_EXP(igmpv3->igmp_qqi) + 3);
982 	}
983 
984 	timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
985 	if (timer == 0)
986 		timer = 1;
987 
988 	nsrc = ntohs(igmpv3->igmp_numsrc);
989 
990 	/*
991 	 * Validate address fields and versions upfront before
992 	 * accepting v3 query.
993 	 * XXX SMPng: Unlocked access to igmpstat counters here.
994 	 */
995 	if (in_nullhost(igmpv3->igmp_group)) {
996 		/*
997 		 * IGMPv3 General Query.
998 		 *
999 		 * General Queries SHOULD be directed to 224.0.0.1.
1000 		 * A general query with a source list has undefined
1001 		 * behaviour; discard it.
1002 		 */
1003 		IGMPSTAT_INC(igps_rcv_gen_queries);
1004 		if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1005 			IGMPSTAT_INC(igps_rcv_badqueries);
1006 			return (0);
1007 		}
1008 		is_general_query = 1;
1009 	} else {
1010 		/* Group or group-source specific query. */
1011 		if (nsrc == 0)
1012 			IGMPSTAT_INC(igps_rcv_group_queries);
1013 		else
1014 			IGMPSTAT_INC(igps_rcv_gsr_queries);
1015 	}
1016 
1017 	IN_MULTI_LOCK();
1018 	IGMP_LOCK();
1019 
1020 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1021 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
1022 
1023 	if (igi->igi_flags & IGIF_LOOPBACK) {
1024 		CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1025 		    ifp, ifp->if_xname);
1026 		goto out_locked;
1027 	}
1028 
1029 	/*
1030 	 * Discard the v3 query if we're in Compatibility Mode.
1031 	 * The RFC is not obviously worded that hosts need to stay in
1032 	 * compatibility mode until the Old Version Querier Present
1033 	 * timer expires.
1034 	 */
1035 	if (igi->igi_version != IGMP_VERSION_3) {
1036 		CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1037 		    igi->igi_version, ifp, ifp->if_xname);
1038 		goto out_locked;
1039 	}
1040 
1041 	igmp_set_version(igi, IGMP_VERSION_3);
1042 	igi->igi_rv = qrv;
1043 	igi->igi_qi = qqi;
1044 	igi->igi_qri = maxresp;
1045 
1046 	CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1047 	    maxresp);
1048 
1049 	if (is_general_query) {
1050 		/*
1051 		 * Schedule a current-state report on this ifp for
1052 		 * all groups, possibly containing source lists.
1053 		 * If there is a pending General Query response
1054 		 * scheduled earlier than the selected delay, do
1055 		 * not schedule any other reports.
1056 		 * Otherwise, reset the interface timer.
1057 		 */
1058 		CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1059 		    ifp, ifp->if_xname);
1060 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1061 			igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1062 			V_interface_timers_running = 1;
1063 		}
1064 	} else {
1065 		/*
1066 		 * Group-source-specific queries are throttled on
1067 		 * a per-group basis to defeat denial-of-service attempts.
1068 		 * Queries for groups we are not a member of on this
1069 		 * link are simply ignored.
1070 		 */
1071 		inm = inm_lookup(ifp, igmpv3->igmp_group);
1072 		if (inm == NULL)
1073 			goto out_locked;
1074 		if (nsrc > 0) {
1075 			if (!ratecheck(&inm->inm_lastgsrtv,
1076 			    &V_igmp_gsrdelay)) {
1077 				CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1078 				    __func__);
1079 				IGMPSTAT_INC(igps_drop_gsr_queries);
1080 				goto out_locked;
1081 			}
1082 		}
1083 		CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1084 		     inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1085 		/*
1086 		 * If there is a pending General Query response
1087 		 * scheduled sooner than the selected delay, no
1088 		 * further report need be scheduled.
1089 		 * Otherwise, prepare to respond to the
1090 		 * group-specific or group-and-source query.
1091 		 */
1092 		if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1093 			igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1094 	}
1095 
1096 out_locked:
1097 	IGMP_UNLOCK();
1098 	IN_MULTI_UNLOCK();
1099 
1100 	return (0);
1101 }
1102 
1103 /*
1104  * Process a recieved IGMPv3 group-specific or group-and-source-specific
1105  * query.
1106  * Return <0 if any error occured. Currently this is ignored.
1107  */
1108 static int
1109 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1110     int timer, /*const*/ struct igmpv3 *igmpv3)
1111 {
1112 	int			 retval;
1113 	uint16_t		 nsrc;
1114 
1115 	IN_MULTI_LOCK_ASSERT();
1116 	IGMP_LOCK_ASSERT();
1117 
1118 	retval = 0;
1119 
1120 	switch (inm->inm_state) {
1121 	case IGMP_NOT_MEMBER:
1122 	case IGMP_SILENT_MEMBER:
1123 	case IGMP_SLEEPING_MEMBER:
1124 	case IGMP_LAZY_MEMBER:
1125 	case IGMP_AWAKENING_MEMBER:
1126 	case IGMP_IDLE_MEMBER:
1127 	case IGMP_LEAVING_MEMBER:
1128 		return (retval);
1129 		break;
1130 	case IGMP_REPORTING_MEMBER:
1131 	case IGMP_G_QUERY_PENDING_MEMBER:
1132 	case IGMP_SG_QUERY_PENDING_MEMBER:
1133 		break;
1134 	}
1135 
1136 	nsrc = ntohs(igmpv3->igmp_numsrc);
1137 
1138 	/*
1139 	 * Deal with group-specific queries upfront.
1140 	 * If any group query is already pending, purge any recorded
1141 	 * source-list state if it exists, and schedule a query response
1142 	 * for this group-specific query.
1143 	 */
1144 	if (nsrc == 0) {
1145 		if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1146 		    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1147 			inm_clear_recorded(inm);
1148 			timer = min(inm->inm_timer, timer);
1149 		}
1150 		inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1151 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1152 		V_current_state_timers_running = 1;
1153 		return (retval);
1154 	}
1155 
1156 	/*
1157 	 * Deal with the case where a group-and-source-specific query has
1158 	 * been received but a group-specific query is already pending.
1159 	 */
1160 	if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1161 		timer = min(inm->inm_timer, timer);
1162 		inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1163 		V_current_state_timers_running = 1;
1164 		return (retval);
1165 	}
1166 
1167 	/*
1168 	 * Finally, deal with the case where a group-and-source-specific
1169 	 * query has been received, where a response to a previous g-s-r
1170 	 * query exists, or none exists.
1171 	 * In this case, we need to parse the source-list which the Querier
1172 	 * has provided us with and check if we have any source list filter
1173 	 * entries at T1 for these sources. If we do not, there is no need
1174 	 * schedule a report and the query may be dropped.
1175 	 * If we do, we must record them and schedule a current-state
1176 	 * report for those sources.
1177 	 * FIXME: Handling source lists larger than 1 mbuf requires that
1178 	 * we pass the mbuf chain pointer down to this function, and use
1179 	 * m_getptr() to walk the chain.
1180 	 */
1181 	if (inm->inm_nsrc > 0) {
1182 		const struct in_addr	*ap;
1183 		int			 i, nrecorded;
1184 
1185 		ap = (const struct in_addr *)(igmpv3 + 1);
1186 		nrecorded = 0;
1187 		for (i = 0; i < nsrc; i++, ap++) {
1188 			retval = inm_record_source(inm, ap->s_addr);
1189 			if (retval < 0)
1190 				break;
1191 			nrecorded += retval;
1192 		}
1193 		if (nrecorded > 0) {
1194 			CTR1(KTR_IGMPV3,
1195 			    "%s: schedule response to SG query", __func__);
1196 			inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1197 			inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1198 			V_current_state_timers_running = 1;
1199 		}
1200 	}
1201 
1202 	return (retval);
1203 }
1204 
1205 /*
1206  * Process a received IGMPv1 host membership report.
1207  *
1208  * NOTE: 0.0.0.0 workaround breaks const correctness.
1209  */
1210 static int
1211 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1212     /*const*/ struct igmp *igmp)
1213 {
1214 	struct in_ifaddr *ia;
1215 	struct in_multi *inm;
1216 
1217 	IGMPSTAT_INC(igps_rcv_reports);
1218 
1219 	if (ifp->if_flags & IFF_LOOPBACK)
1220 		return (0);
1221 
1222 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr) ||
1223 	    !in_hosteq(igmp->igmp_group, ip->ip_dst))) {
1224 		IGMPSTAT_INC(igps_rcv_badreports);
1225 		return (EINVAL);
1226 	}
1227 
1228 	/*
1229 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1230 	 * Booting clients may use the source address 0.0.0.0. Some
1231 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1232 	 * the interface upon which this message was received.
1233 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1234 	 */
1235 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1236 		IFP_TO_IA(ifp, ia);
1237 		if (ia != NULL) {
1238 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1239 			ifa_free(&ia->ia_ifa);
1240 		}
1241 	}
1242 
1243 	CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1244 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1245 
1246 	/*
1247 	 * IGMPv1 report suppression.
1248 	 * If we are a member of this group, and our membership should be
1249 	 * reported, stop our group timer and transition to the 'lazy' state.
1250 	 */
1251 	IN_MULTI_LOCK();
1252 	inm = inm_lookup(ifp, igmp->igmp_group);
1253 	if (inm != NULL) {
1254 		struct igmp_ifinfo *igi;
1255 
1256 		igi = inm->inm_igi;
1257 		if (igi == NULL) {
1258 			KASSERT(igi != NULL,
1259 			    ("%s: no igi for ifp %p", __func__, ifp));
1260 			goto out_locked;
1261 		}
1262 
1263 		IGMPSTAT_INC(igps_rcv_ourreports);
1264 
1265 		/*
1266 		 * If we are in IGMPv3 host mode, do not allow the
1267 		 * other host's IGMPv1 report to suppress our reports
1268 		 * unless explicitly configured to do so.
1269 		 */
1270 		if (igi->igi_version == IGMP_VERSION_3) {
1271 			if (V_igmp_legacysupp)
1272 				igmp_v3_suppress_group_record(inm);
1273 			goto out_locked;
1274 		}
1275 
1276 		inm->inm_timer = 0;
1277 
1278 		switch (inm->inm_state) {
1279 		case IGMP_NOT_MEMBER:
1280 		case IGMP_SILENT_MEMBER:
1281 			break;
1282 		case IGMP_IDLE_MEMBER:
1283 		case IGMP_LAZY_MEMBER:
1284 		case IGMP_AWAKENING_MEMBER:
1285 			CTR3(KTR_IGMPV3,
1286 			    "report suppressed for %s on ifp %p(%s)",
1287 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1288 		case IGMP_SLEEPING_MEMBER:
1289 			inm->inm_state = IGMP_SLEEPING_MEMBER;
1290 			break;
1291 		case IGMP_REPORTING_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 			if (igi->igi_version == IGMP_VERSION_1)
1296 				inm->inm_state = IGMP_LAZY_MEMBER;
1297 			else if (igi->igi_version == IGMP_VERSION_2)
1298 				inm->inm_state = IGMP_SLEEPING_MEMBER;
1299 			break;
1300 		case IGMP_G_QUERY_PENDING_MEMBER:
1301 		case IGMP_SG_QUERY_PENDING_MEMBER:
1302 		case IGMP_LEAVING_MEMBER:
1303 			break;
1304 		}
1305 	}
1306 
1307 out_locked:
1308 	IN_MULTI_UNLOCK();
1309 
1310 	return (0);
1311 }
1312 
1313 /*
1314  * Process a received IGMPv2 host membership report.
1315  *
1316  * NOTE: 0.0.0.0 workaround breaks const correctness.
1317  */
1318 static int
1319 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1320     /*const*/ struct igmp *igmp)
1321 {
1322 	struct in_ifaddr *ia;
1323 	struct in_multi *inm;
1324 
1325 	/*
1326 	 * Make sure we don't hear our own membership report.  Fast
1327 	 * leave requires knowing that we are the only member of a
1328 	 * group.
1329 	 */
1330 	IFP_TO_IA(ifp, ia);
1331 	if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1332 		ifa_free(&ia->ia_ifa);
1333 		return (0);
1334 	}
1335 
1336 	IGMPSTAT_INC(igps_rcv_reports);
1337 
1338 	if (ifp->if_flags & IFF_LOOPBACK) {
1339 		if (ia != NULL)
1340 			ifa_free(&ia->ia_ifa);
1341 		return (0);
1342 	}
1343 
1344 	if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1345 	    !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1346 		if (ia != NULL)
1347 			ifa_free(&ia->ia_ifa);
1348 		IGMPSTAT_INC(igps_rcv_badreports);
1349 		return (EINVAL);
1350 	}
1351 
1352 	/*
1353 	 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1354 	 * Booting clients may use the source address 0.0.0.0. Some
1355 	 * IGMP daemons may not know how to use IP_RECVIF to determine
1356 	 * the interface upon which this message was received.
1357 	 * Replace 0.0.0.0 with the subnet address if told to do so.
1358 	 */
1359 	if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1360 		if (ia != NULL)
1361 			ip->ip_src.s_addr = htonl(ia->ia_subnet);
1362 	}
1363 	if (ia != NULL)
1364 		ifa_free(&ia->ia_ifa);
1365 
1366 	CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1367 	     inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1368 
1369 	/*
1370 	 * IGMPv2 report suppression.
1371 	 * If we are a member of this group, and our membership should be
1372 	 * reported, and our group timer is pending or about to be reset,
1373 	 * stop our group timer by transitioning to the 'lazy' state.
1374 	 */
1375 	IN_MULTI_LOCK();
1376 	inm = inm_lookup(ifp, igmp->igmp_group);
1377 	if (inm != NULL) {
1378 		struct igmp_ifinfo *igi;
1379 
1380 		igi = inm->inm_igi;
1381 		KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1382 
1383 		IGMPSTAT_INC(igps_rcv_ourreports);
1384 
1385 		/*
1386 		 * If we are in IGMPv3 host mode, do not allow the
1387 		 * other host's IGMPv1 report to suppress our reports
1388 		 * unless explicitly configured to do so.
1389 		 */
1390 		if (igi->igi_version == IGMP_VERSION_3) {
1391 			if (V_igmp_legacysupp)
1392 				igmp_v3_suppress_group_record(inm);
1393 			goto out_locked;
1394 		}
1395 
1396 		inm->inm_timer = 0;
1397 
1398 		switch (inm->inm_state) {
1399 		case IGMP_NOT_MEMBER:
1400 		case IGMP_SILENT_MEMBER:
1401 		case IGMP_SLEEPING_MEMBER:
1402 			break;
1403 		case IGMP_REPORTING_MEMBER:
1404 		case IGMP_IDLE_MEMBER:
1405 		case IGMP_AWAKENING_MEMBER:
1406 			CTR3(KTR_IGMPV3,
1407 			    "report suppressed for %s on ifp %p(%s)",
1408 			    inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1409 		case IGMP_LAZY_MEMBER:
1410 			inm->inm_state = IGMP_LAZY_MEMBER;
1411 			break;
1412 		case IGMP_G_QUERY_PENDING_MEMBER:
1413 		case IGMP_SG_QUERY_PENDING_MEMBER:
1414 		case IGMP_LEAVING_MEMBER:
1415 			break;
1416 		}
1417 	}
1418 
1419 out_locked:
1420 	IN_MULTI_UNLOCK();
1421 
1422 	return (0);
1423 }
1424 
1425 void
1426 igmp_input(struct mbuf *m, int off)
1427 {
1428 	int iphlen;
1429 	struct ifnet *ifp;
1430 	struct igmp *igmp;
1431 	struct ip *ip;
1432 	int igmplen;
1433 	int minlen;
1434 	int queryver;
1435 
1436 	CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1437 
1438 	ifp = m->m_pkthdr.rcvif;
1439 
1440 	IGMPSTAT_INC(igps_rcv_total);
1441 
1442 	ip = mtod(m, struct ip *);
1443 	iphlen = off;
1444 	igmplen = ip->ip_len;
1445 
1446 	/*
1447 	 * Validate lengths.
1448 	 */
1449 	if (igmplen < IGMP_MINLEN) {
1450 		IGMPSTAT_INC(igps_rcv_tooshort);
1451 		m_freem(m);
1452 		return;
1453 	}
1454 
1455 	/*
1456 	 * Always pullup to the minimum size for v1/v2 or v3
1457 	 * to amortize calls to m_pullup().
1458 	 */
1459 	minlen = iphlen;
1460 	if (igmplen >= IGMP_V3_QUERY_MINLEN)
1461 		minlen += IGMP_V3_QUERY_MINLEN;
1462 	else
1463 		minlen += IGMP_MINLEN;
1464 	if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1465 	    (m = m_pullup(m, minlen)) == 0) {
1466 		IGMPSTAT_INC(igps_rcv_tooshort);
1467 		return;
1468 	}
1469 	ip = mtod(m, struct ip *);
1470 
1471 	if (ip->ip_ttl != 1) {
1472 		IGMPSTAT_INC(igps_rcv_badttl);
1473 		m_freem(m);
1474 		return;
1475 	}
1476 
1477 	/*
1478 	 * Validate checksum.
1479 	 */
1480 	m->m_data += iphlen;
1481 	m->m_len -= iphlen;
1482 	igmp = mtod(m, struct igmp *);
1483 	if (in_cksum(m, igmplen)) {
1484 		IGMPSTAT_INC(igps_rcv_badsum);
1485 		m_freem(m);
1486 		return;
1487 	}
1488 	m->m_data -= iphlen;
1489 	m->m_len += iphlen;
1490 
1491 	switch (igmp->igmp_type) {
1492 	case IGMP_HOST_MEMBERSHIP_QUERY:
1493 		if (igmplen == IGMP_MINLEN) {
1494 			if (igmp->igmp_code == 0)
1495 				queryver = IGMP_VERSION_1;
1496 			else
1497 				queryver = IGMP_VERSION_2;
1498 		} else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1499 			queryver = IGMP_VERSION_3;
1500 		} else {
1501 			IGMPSTAT_INC(igps_rcv_tooshort);
1502 			m_freem(m);
1503 			return;
1504 		}
1505 
1506 		switch (queryver) {
1507 		case IGMP_VERSION_1:
1508 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1509 			if (!V_igmp_v1enable)
1510 				break;
1511 			if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1512 				m_freem(m);
1513 				return;
1514 			}
1515 			break;
1516 
1517 		case IGMP_VERSION_2:
1518 			IGMPSTAT_INC(igps_rcv_v1v2_queries);
1519 			if (!V_igmp_v2enable)
1520 				break;
1521 			if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1522 				m_freem(m);
1523 				return;
1524 			}
1525 			break;
1526 
1527 		case IGMP_VERSION_3: {
1528 				struct igmpv3 *igmpv3;
1529 				uint16_t igmpv3len;
1530 				uint16_t srclen;
1531 				int nsrc;
1532 
1533 				IGMPSTAT_INC(igps_rcv_v3_queries);
1534 				igmpv3 = (struct igmpv3 *)igmp;
1535 				/*
1536 				 * Validate length based on source count.
1537 				 */
1538 				nsrc = ntohs(igmpv3->igmp_numsrc);
1539 				srclen = sizeof(struct in_addr) * nsrc;
1540 				if (nsrc * sizeof(in_addr_t) > srclen) {
1541 					IGMPSTAT_INC(igps_rcv_tooshort);
1542 					return;
1543 				}
1544 				/*
1545 				 * m_pullup() may modify m, so pullup in
1546 				 * this scope.
1547 				 */
1548 				igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1549 				    srclen;
1550 				if ((m->m_flags & M_EXT ||
1551 				     m->m_len < igmpv3len) &&
1552 				    (m = m_pullup(m, igmpv3len)) == NULL) {
1553 					IGMPSTAT_INC(igps_rcv_tooshort);
1554 					return;
1555 				}
1556 				igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1557 				    + iphlen);
1558 				if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1559 					m_freem(m);
1560 					return;
1561 				}
1562 			}
1563 			break;
1564 		}
1565 		break;
1566 
1567 	case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1568 		if (!V_igmp_v1enable)
1569 			break;
1570 		if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1571 			m_freem(m);
1572 			return;
1573 		}
1574 		break;
1575 
1576 	case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1577 		if (!V_igmp_v2enable)
1578 			break;
1579 		if (!ip_checkrouteralert(m))
1580 			IGMPSTAT_INC(igps_rcv_nora);
1581 		if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1582 			m_freem(m);
1583 			return;
1584 		}
1585 		break;
1586 
1587 	case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1588 		/*
1589 		 * Hosts do not need to process IGMPv3 membership reports,
1590 		 * as report suppression is no longer required.
1591 		 */
1592 		if (!ip_checkrouteralert(m))
1593 			IGMPSTAT_INC(igps_rcv_nora);
1594 		break;
1595 
1596 	default:
1597 		break;
1598 	}
1599 
1600 	/*
1601 	 * Pass all valid IGMP packets up to any process(es) listening on a
1602 	 * raw IGMP socket.
1603 	 */
1604 	rip_input(m, off);
1605 }
1606 
1607 
1608 /*
1609  * Fast timeout handler (global).
1610  * VIMAGE: Timeout handlers are expected to service all vimages.
1611  */
1612 void
1613 igmp_fasttimo(void)
1614 {
1615 	VNET_ITERATOR_DECL(vnet_iter);
1616 
1617 	VNET_LIST_RLOCK_NOSLEEP();
1618 	VNET_FOREACH(vnet_iter) {
1619 		CURVNET_SET(vnet_iter);
1620 		igmp_fasttimo_vnet();
1621 		CURVNET_RESTORE();
1622 	}
1623 	VNET_LIST_RUNLOCK_NOSLEEP();
1624 }
1625 
1626 /*
1627  * Fast timeout handler (per-vnet).
1628  * Sends are shuffled off to a netisr to deal with Giant.
1629  *
1630  * VIMAGE: Assume caller has set up our curvnet.
1631  */
1632 static void
1633 igmp_fasttimo_vnet(void)
1634 {
1635 	struct ifqueue		 scq;	/* State-change packets */
1636 	struct ifqueue		 qrq;	/* Query response packets */
1637 	struct ifnet		*ifp;
1638 	struct igmp_ifinfo	*igi;
1639 	struct ifmultiaddr	*ifma, *tifma;
1640 	struct in_multi		*inm;
1641 	int			 loop, uri_fasthz;
1642 
1643 	loop = 0;
1644 	uri_fasthz = 0;
1645 
1646 	/*
1647 	 * Quick check to see if any work needs to be done, in order to
1648 	 * minimize the overhead of fasttimo processing.
1649 	 * SMPng: XXX Unlocked reads.
1650 	 */
1651 	if (!V_current_state_timers_running &&
1652 	    !V_interface_timers_running &&
1653 	    !V_state_change_timers_running)
1654 		return;
1655 
1656 	IN_MULTI_LOCK();
1657 	IGMP_LOCK();
1658 
1659 	/*
1660 	 * IGMPv3 General Query response timer processing.
1661 	 */
1662 	if (V_interface_timers_running) {
1663 		CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1664 
1665 		V_interface_timers_running = 0;
1666 		LIST_FOREACH(igi, &V_igi_head, igi_link) {
1667 			if (igi->igi_v3_timer == 0) {
1668 				/* Do nothing. */
1669 			} else if (--igi->igi_v3_timer == 0) {
1670 				igmp_v3_dispatch_general_query(igi);
1671 			} else {
1672 				V_interface_timers_running = 1;
1673 			}
1674 		}
1675 	}
1676 
1677 	if (!V_current_state_timers_running &&
1678 	    !V_state_change_timers_running)
1679 		goto out_locked;
1680 
1681 	V_current_state_timers_running = 0;
1682 	V_state_change_timers_running = 0;
1683 
1684 	CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1685 
1686 	/*
1687 	 * IGMPv1/v2/v3 host report and state-change timer processing.
1688 	 * Note: Processing a v3 group timer may remove a node.
1689 	 */
1690 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
1691 		ifp = igi->igi_ifp;
1692 
1693 		if (igi->igi_version == IGMP_VERSION_3) {
1694 			loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1695 			uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1696 			    PR_FASTHZ);
1697 
1698 			memset(&qrq, 0, sizeof(struct ifqueue));
1699 			IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1700 
1701 			memset(&scq, 0, sizeof(struct ifqueue));
1702 			IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1703 		}
1704 
1705 		IF_ADDR_LOCK(ifp);
1706 		TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1707 		    tifma) {
1708 			if (ifma->ifma_addr->sa_family != AF_INET ||
1709 			    ifma->ifma_protospec == NULL)
1710 				continue;
1711 			inm = (struct in_multi *)ifma->ifma_protospec;
1712 			switch (igi->igi_version) {
1713 			case IGMP_VERSION_1:
1714 			case IGMP_VERSION_2:
1715 				igmp_v1v2_process_group_timer(inm,
1716 				    igi->igi_version);
1717 				break;
1718 			case IGMP_VERSION_3:
1719 				igmp_v3_process_group_timers(igi, &qrq,
1720 				    &scq, inm, uri_fasthz);
1721 				break;
1722 			}
1723 		}
1724 		IF_ADDR_UNLOCK(ifp);
1725 
1726 		if (igi->igi_version == IGMP_VERSION_3) {
1727 			struct in_multi		*tinm;
1728 
1729 			igmp_dispatch_queue(&qrq, 0, loop);
1730 			igmp_dispatch_queue(&scq, 0, loop);
1731 
1732 			/*
1733 			 * Free the in_multi reference(s) for this
1734 			 * IGMP lifecycle.
1735 			 */
1736 			SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1737 			    inm_nrele, tinm) {
1738 				SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1739 				    inm_nrele);
1740 				inm_release_locked(inm);
1741 			}
1742 		}
1743 	}
1744 
1745 out_locked:
1746 	IGMP_UNLOCK();
1747 	IN_MULTI_UNLOCK();
1748 }
1749 
1750 /*
1751  * Update host report group timer for IGMPv1/v2.
1752  * Will update the global pending timer flags.
1753  */
1754 static void
1755 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1756 {
1757 	int report_timer_expired;
1758 
1759 	IN_MULTI_LOCK_ASSERT();
1760 	IGMP_LOCK_ASSERT();
1761 
1762 	if (inm->inm_timer == 0) {
1763 		report_timer_expired = 0;
1764 	} else if (--inm->inm_timer == 0) {
1765 		report_timer_expired = 1;
1766 	} else {
1767 		V_current_state_timers_running = 1;
1768 		return;
1769 	}
1770 
1771 	switch (inm->inm_state) {
1772 	case IGMP_NOT_MEMBER:
1773 	case IGMP_SILENT_MEMBER:
1774 	case IGMP_IDLE_MEMBER:
1775 	case IGMP_LAZY_MEMBER:
1776 	case IGMP_SLEEPING_MEMBER:
1777 	case IGMP_AWAKENING_MEMBER:
1778 		break;
1779 	case IGMP_REPORTING_MEMBER:
1780 		if (report_timer_expired) {
1781 			inm->inm_state = IGMP_IDLE_MEMBER;
1782 			(void)igmp_v1v2_queue_report(inm,
1783 			    (version == IGMP_VERSION_2) ?
1784 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
1785 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
1786 		}
1787 		break;
1788 	case IGMP_G_QUERY_PENDING_MEMBER:
1789 	case IGMP_SG_QUERY_PENDING_MEMBER:
1790 	case IGMP_LEAVING_MEMBER:
1791 		break;
1792 	}
1793 }
1794 
1795 /*
1796  * Update a group's timers for IGMPv3.
1797  * Will update the global pending timer flags.
1798  * Note: Unlocked read from igi.
1799  */
1800 static void
1801 igmp_v3_process_group_timers(struct igmp_ifinfo *igi,
1802     struct ifqueue *qrq, struct ifqueue *scq,
1803     struct in_multi *inm, const int uri_fasthz)
1804 {
1805 	int query_response_timer_expired;
1806 	int state_change_retransmit_timer_expired;
1807 
1808 	IN_MULTI_LOCK_ASSERT();
1809 	IGMP_LOCK_ASSERT();
1810 
1811 	query_response_timer_expired = 0;
1812 	state_change_retransmit_timer_expired = 0;
1813 
1814 	/*
1815 	 * During a transition from v1/v2 compatibility mode back to v3,
1816 	 * a group record in REPORTING state may still have its group
1817 	 * timer active. This is a no-op in this function; it is easier
1818 	 * to deal with it here than to complicate the slow-timeout path.
1819 	 */
1820 	if (inm->inm_timer == 0) {
1821 		query_response_timer_expired = 0;
1822 	} else if (--inm->inm_timer == 0) {
1823 		query_response_timer_expired = 1;
1824 	} else {
1825 		V_current_state_timers_running = 1;
1826 	}
1827 
1828 	if (inm->inm_sctimer == 0) {
1829 		state_change_retransmit_timer_expired = 0;
1830 	} else if (--inm->inm_sctimer == 0) {
1831 		state_change_retransmit_timer_expired = 1;
1832 	} else {
1833 		V_state_change_timers_running = 1;
1834 	}
1835 
1836 	/* We are in fasttimo, so be quick about it. */
1837 	if (!state_change_retransmit_timer_expired &&
1838 	    !query_response_timer_expired)
1839 		return;
1840 
1841 	switch (inm->inm_state) {
1842 	case IGMP_NOT_MEMBER:
1843 	case IGMP_SILENT_MEMBER:
1844 	case IGMP_SLEEPING_MEMBER:
1845 	case IGMP_LAZY_MEMBER:
1846 	case IGMP_AWAKENING_MEMBER:
1847 	case IGMP_IDLE_MEMBER:
1848 		break;
1849 	case IGMP_G_QUERY_PENDING_MEMBER:
1850 	case IGMP_SG_QUERY_PENDING_MEMBER:
1851 		/*
1852 		 * Respond to a previously pending Group-Specific
1853 		 * or Group-and-Source-Specific query by enqueueing
1854 		 * the appropriate Current-State report for
1855 		 * immediate transmission.
1856 		 */
1857 		if (query_response_timer_expired) {
1858 			int retval;
1859 
1860 			retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1861 			    (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1862 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1863 			    __func__, retval);
1864 			inm->inm_state = IGMP_REPORTING_MEMBER;
1865 			/* XXX Clear recorded sources for next time. */
1866 			inm_clear_recorded(inm);
1867 		}
1868 		/* FALLTHROUGH */
1869 	case IGMP_REPORTING_MEMBER:
1870 	case IGMP_LEAVING_MEMBER:
1871 		if (state_change_retransmit_timer_expired) {
1872 			/*
1873 			 * State-change retransmission timer fired.
1874 			 * If there are any further pending retransmissions,
1875 			 * set the global pending state-change flag, and
1876 			 * reset the timer.
1877 			 */
1878 			if (--inm->inm_scrv > 0) {
1879 				inm->inm_sctimer = uri_fasthz;
1880 				V_state_change_timers_running = 1;
1881 			}
1882 			/*
1883 			 * Retransmit the previously computed state-change
1884 			 * report. If there are no further pending
1885 			 * retransmissions, the mbuf queue will be consumed.
1886 			 * Update T0 state to T1 as we have now sent
1887 			 * a state-change.
1888 			 */
1889 			(void)igmp_v3_merge_state_changes(inm, scq);
1890 
1891 			inm_commit(inm);
1892 			CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1893 			    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1894 
1895 			/*
1896 			 * If we are leaving the group for good, make sure
1897 			 * we release IGMP's reference to it.
1898 			 * This release must be deferred using a SLIST,
1899 			 * as we are called from a loop which traverses
1900 			 * the in_ifmultiaddr TAILQ.
1901 			 */
1902 			if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1903 			    inm->inm_scrv == 0) {
1904 				inm->inm_state = IGMP_NOT_MEMBER;
1905 				SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1906 				    inm, inm_nrele);
1907 			}
1908 		}
1909 		break;
1910 	}
1911 }
1912 
1913 
1914 /*
1915  * Suppress a group's pending response to a group or source/group query.
1916  *
1917  * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1918  * Do NOT update ST1/ST0 as this operation merely suppresses
1919  * the currently pending group record.
1920  * Do NOT suppress the response to a general query. It is possible but
1921  * it would require adding another state or flag.
1922  */
1923 static void
1924 igmp_v3_suppress_group_record(struct in_multi *inm)
1925 {
1926 
1927 	IN_MULTI_LOCK_ASSERT();
1928 
1929 	KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1930 		("%s: not IGMPv3 mode on link", __func__));
1931 
1932 	if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1933 	    inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1934 		return;
1935 
1936 	if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1937 		inm_clear_recorded(inm);
1938 
1939 	inm->inm_timer = 0;
1940 	inm->inm_state = IGMP_REPORTING_MEMBER;
1941 }
1942 
1943 /*
1944  * Switch to a different IGMP version on the given interface,
1945  * as per Section 7.2.1.
1946  */
1947 static void
1948 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1949 {
1950 	int old_version_timer;
1951 
1952 	IGMP_LOCK_ASSERT();
1953 
1954 	CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1955 	    version, igi->igi_ifp, igi->igi_ifp->if_xname);
1956 
1957 	if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1958 		/*
1959 		 * Compute the "Older Version Querier Present" timer as per
1960 		 * Section 8.12.
1961 		 */
1962 		old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1963 		old_version_timer *= PR_SLOWHZ;
1964 
1965 		if (version == IGMP_VERSION_1) {
1966 			igi->igi_v1_timer = old_version_timer;
1967 			igi->igi_v2_timer = 0;
1968 		} else if (version == IGMP_VERSION_2) {
1969 			igi->igi_v1_timer = 0;
1970 			igi->igi_v2_timer = old_version_timer;
1971 		}
1972 	}
1973 
1974 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1975 		if (igi->igi_version != IGMP_VERSION_2) {
1976 			igi->igi_version = IGMP_VERSION_2;
1977 			igmp_v3_cancel_link_timers(igi);
1978 		}
1979 	} else if (igi->igi_v1_timer > 0) {
1980 		if (igi->igi_version != IGMP_VERSION_1) {
1981 			igi->igi_version = IGMP_VERSION_1;
1982 			igmp_v3_cancel_link_timers(igi);
1983 		}
1984 	}
1985 }
1986 
1987 /*
1988  * Cancel pending IGMPv3 timers for the given link and all groups
1989  * joined on it; state-change, general-query, and group-query timers.
1990  *
1991  * Only ever called on a transition from v3 to Compatibility mode. Kill
1992  * the timers stone dead (this may be expensive for large N groups), they
1993  * will be restarted if Compatibility Mode deems that they must be due to
1994  * query processing.
1995  */
1996 static void
1997 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
1998 {
1999 	struct ifmultiaddr	*ifma;
2000 	struct ifnet		*ifp;
2001 	struct in_multi		*inm;
2002 
2003 	CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2004 	    igi->igi_ifp, igi->igi_ifp->if_xname);
2005 
2006 	IN_MULTI_LOCK_ASSERT();
2007 	IGMP_LOCK_ASSERT();
2008 
2009 	/*
2010 	 * Stop the v3 General Query Response on this link stone dead.
2011 	 * If fasttimo is woken up due to V_interface_timers_running,
2012 	 * the flag will be cleared if there are no pending link timers.
2013 	 */
2014 	igi->igi_v3_timer = 0;
2015 
2016 	/*
2017 	 * Now clear the current-state and state-change report timers
2018 	 * for all memberships scoped to this link.
2019 	 */
2020 	ifp = igi->igi_ifp;
2021 	IF_ADDR_LOCK(ifp);
2022 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2023 		if (ifma->ifma_addr->sa_family != AF_INET ||
2024 		    ifma->ifma_protospec == NULL)
2025 			continue;
2026 		inm = (struct in_multi *)ifma->ifma_protospec;
2027 		switch (inm->inm_state) {
2028 		case IGMP_NOT_MEMBER:
2029 		case IGMP_SILENT_MEMBER:
2030 		case IGMP_IDLE_MEMBER:
2031 		case IGMP_LAZY_MEMBER:
2032 		case IGMP_SLEEPING_MEMBER:
2033 		case IGMP_AWAKENING_MEMBER:
2034 			/*
2035 			 * These states are either not relevant in v3 mode,
2036 			 * or are unreported. Do nothing.
2037 			 */
2038 			break;
2039 		case IGMP_LEAVING_MEMBER:
2040 			/*
2041 			 * If we are leaving the group and switching to
2042 			 * compatibility mode, we need to release the final
2043 			 * reference held for issuing the INCLUDE {}, and
2044 			 * transition to REPORTING to ensure the host leave
2045 			 * message is sent upstream to the old querier --
2046 			 * transition to NOT would lose the leave and race.
2047 			 *
2048 			 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
2049 			 * around inm_release_locked(), as it is not
2050 			 * a recursive mutex.
2051 			 */
2052 			IF_ADDR_UNLOCK(ifp);
2053 			inm_release_locked(inm);
2054 			IF_ADDR_LOCK(ifp);
2055 			/* FALLTHROUGH */
2056 		case IGMP_G_QUERY_PENDING_MEMBER:
2057 		case IGMP_SG_QUERY_PENDING_MEMBER:
2058 			inm_clear_recorded(inm);
2059 			/* FALLTHROUGH */
2060 		case IGMP_REPORTING_MEMBER:
2061 			inm->inm_state = IGMP_REPORTING_MEMBER;
2062 			break;
2063 		}
2064 		/*
2065 		 * Always clear state-change and group report timers.
2066 		 * Free any pending IGMPv3 state-change records.
2067 		 */
2068 		inm->inm_sctimer = 0;
2069 		inm->inm_timer = 0;
2070 		_IF_DRAIN(&inm->inm_scq);
2071 	}
2072 	IF_ADDR_UNLOCK(ifp);
2073 }
2074 
2075 /*
2076  * Update the Older Version Querier Present timers for a link.
2077  * See Section 7.2.1 of RFC 3376.
2078  */
2079 static void
2080 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2081 {
2082 
2083 	IGMP_LOCK_ASSERT();
2084 
2085 	if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2086 		/*
2087 		 * IGMPv1 and IGMPv2 Querier Present timers expired.
2088 		 *
2089 		 * Revert to IGMPv3.
2090 		 */
2091 		if (igi->igi_version != IGMP_VERSION_3) {
2092 			CTR5(KTR_IGMPV3,
2093 			    "%s: transition from v%d -> v%d on %p(%s)",
2094 			    __func__, igi->igi_version, IGMP_VERSION_3,
2095 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2096 			igi->igi_version = IGMP_VERSION_3;
2097 		}
2098 	} else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2099 		/*
2100 		 * IGMPv1 Querier Present timer expired,
2101 		 * IGMPv2 Querier Present timer running.
2102 		 * If IGMPv2 was disabled since last timeout,
2103 		 * revert to IGMPv3.
2104 		 * If IGMPv2 is enabled, revert to IGMPv2.
2105 		 */
2106 		if (!V_igmp_v2enable) {
2107 			CTR5(KTR_IGMPV3,
2108 			    "%s: transition from v%d -> v%d on %p(%s)",
2109 			    __func__, igi->igi_version, IGMP_VERSION_3,
2110 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2111 			igi->igi_v2_timer = 0;
2112 			igi->igi_version = IGMP_VERSION_3;
2113 		} else {
2114 			--igi->igi_v2_timer;
2115 			if (igi->igi_version != IGMP_VERSION_2) {
2116 				CTR5(KTR_IGMPV3,
2117 				    "%s: transition from v%d -> v%d on %p(%s)",
2118 				    __func__, igi->igi_version, IGMP_VERSION_2,
2119 				    igi->igi_ifp, igi->igi_ifp->if_xname);
2120 				igi->igi_version = IGMP_VERSION_2;
2121 			}
2122 		}
2123 	} else if (igi->igi_v1_timer > 0) {
2124 		/*
2125 		 * IGMPv1 Querier Present timer running.
2126 		 * Stop IGMPv2 timer if running.
2127 		 *
2128 		 * If IGMPv1 was disabled since last timeout,
2129 		 * revert to IGMPv3.
2130 		 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2131 		 */
2132 		if (!V_igmp_v1enable) {
2133 			CTR5(KTR_IGMPV3,
2134 			    "%s: transition from v%d -> v%d on %p(%s)",
2135 			    __func__, igi->igi_version, IGMP_VERSION_3,
2136 			    igi->igi_ifp, igi->igi_ifp->if_xname);
2137 			igi->igi_v1_timer = 0;
2138 			igi->igi_version = IGMP_VERSION_3;
2139 		} else {
2140 			--igi->igi_v1_timer;
2141 		}
2142 		if (igi->igi_v2_timer > 0) {
2143 			CTR3(KTR_IGMPV3,
2144 			    "%s: cancel v2 timer on %p(%s)",
2145 			    __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2146 			igi->igi_v2_timer = 0;
2147 		}
2148 	}
2149 }
2150 
2151 /*
2152  * Global slowtimo handler.
2153  * VIMAGE: Timeout handlers are expected to service all vimages.
2154  */
2155 void
2156 igmp_slowtimo(void)
2157 {
2158 	VNET_ITERATOR_DECL(vnet_iter);
2159 
2160 	VNET_LIST_RLOCK_NOSLEEP();
2161 	VNET_FOREACH(vnet_iter) {
2162 		CURVNET_SET(vnet_iter);
2163 		igmp_slowtimo_vnet();
2164 		CURVNET_RESTORE();
2165 	}
2166 	VNET_LIST_RUNLOCK_NOSLEEP();
2167 }
2168 
2169 /*
2170  * Per-vnet slowtimo handler.
2171  */
2172 static void
2173 igmp_slowtimo_vnet(void)
2174 {
2175 	struct igmp_ifinfo *igi;
2176 
2177 	IGMP_LOCK();
2178 
2179 	LIST_FOREACH(igi, &V_igi_head, igi_link) {
2180 		igmp_v1v2_process_querier_timers(igi);
2181 	}
2182 
2183 	IGMP_UNLOCK();
2184 }
2185 
2186 /*
2187  * Dispatch an IGMPv1/v2 host report or leave message.
2188  * These are always small enough to fit inside a single mbuf.
2189  */
2190 static int
2191 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2192 {
2193 	struct ifnet		*ifp;
2194 	struct igmp		*igmp;
2195 	struct ip		*ip;
2196 	struct mbuf		*m;
2197 
2198 	IN_MULTI_LOCK_ASSERT();
2199 	IGMP_LOCK_ASSERT();
2200 
2201 	ifp = inm->inm_ifp;
2202 
2203 	MGETHDR(m, M_DONTWAIT, MT_DATA);
2204 	if (m == NULL)
2205 		return (ENOMEM);
2206 	MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2207 
2208 	m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2209 
2210 	m->m_data += sizeof(struct ip);
2211 	m->m_len = sizeof(struct igmp);
2212 
2213 	igmp = mtod(m, struct igmp *);
2214 	igmp->igmp_type = type;
2215 	igmp->igmp_code = 0;
2216 	igmp->igmp_group = inm->inm_addr;
2217 	igmp->igmp_cksum = 0;
2218 	igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2219 
2220 	m->m_data -= sizeof(struct ip);
2221 	m->m_len += sizeof(struct ip);
2222 
2223 	ip = mtod(m, struct ip *);
2224 	ip->ip_tos = 0;
2225 	ip->ip_len = sizeof(struct ip) + sizeof(struct igmp);
2226 	ip->ip_off = 0;
2227 	ip->ip_p = IPPROTO_IGMP;
2228 	ip->ip_src.s_addr = INADDR_ANY;
2229 
2230 	if (type == IGMP_HOST_LEAVE_MESSAGE)
2231 		ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2232 	else
2233 		ip->ip_dst = inm->inm_addr;
2234 
2235 	igmp_save_context(m, ifp);
2236 
2237 	m->m_flags |= M_IGMPV2;
2238 	if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2239 		m->m_flags |= M_IGMP_LOOP;
2240 
2241 	CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2242 	netisr_dispatch(NETISR_IGMP, m);
2243 
2244 	return (0);
2245 }
2246 
2247 /*
2248  * Process a state change from the upper layer for the given IPv4 group.
2249  *
2250  * Each socket holds a reference on the in_multi in its own ip_moptions.
2251  * The socket layer will have made the necessary updates to.the group
2252  * state, it is now up to IGMP to issue a state change report if there
2253  * has been any change between T0 (when the last state-change was issued)
2254  * and T1 (now).
2255  *
2256  * We use the IGMPv3 state machine at group level. The IGMP module
2257  * however makes the decision as to which IGMP protocol version to speak.
2258  * A state change *from* INCLUDE {} always means an initial join.
2259  * A state change *to* INCLUDE {} always means a final leave.
2260  *
2261  * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2262  * save ourselves a bunch of work; any exclusive mode groups need not
2263  * compute source filter lists.
2264  *
2265  * VIMAGE: curvnet should have been set by caller, as this routine
2266  * is called from the socket option handlers.
2267  */
2268 int
2269 igmp_change_state(struct in_multi *inm)
2270 {
2271 	struct igmp_ifinfo *igi;
2272 	struct ifnet *ifp;
2273 	int error;
2274 
2275 	IN_MULTI_LOCK_ASSERT();
2276 
2277 	error = 0;
2278 
2279 	/*
2280 	 * Try to detect if the upper layer just asked us to change state
2281 	 * for an interface which has now gone away.
2282 	 */
2283 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2284 	ifp = inm->inm_ifma->ifma_ifp;
2285 	if (ifp != NULL) {
2286 		/*
2287 		 * Sanity check that netinet's notion of ifp is the
2288 		 * same as net's.
2289 		 */
2290 		KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2291 	}
2292 
2293 	IGMP_LOCK();
2294 
2295 	igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2296 	KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2297 
2298 	/*
2299 	 * If we detect a state transition to or from MCAST_UNDEFINED
2300 	 * for this group, then we are starting or finishing an IGMP
2301 	 * life cycle for this group.
2302 	 */
2303 	if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2304 		CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2305 		    inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2306 		if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2307 			CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2308 			error = igmp_initial_join(inm, igi);
2309 			goto out_locked;
2310 		} else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2311 			CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2312 			igmp_final_leave(inm, igi);
2313 			goto out_locked;
2314 		}
2315 	} else {
2316 		CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2317 	}
2318 
2319 	error = igmp_handle_state_change(inm, igi);
2320 
2321 out_locked:
2322 	IGMP_UNLOCK();
2323 	return (error);
2324 }
2325 
2326 /*
2327  * Perform the initial join for an IGMP group.
2328  *
2329  * When joining a group:
2330  *  If the group should have its IGMP traffic suppressed, do nothing.
2331  *  IGMPv1 starts sending IGMPv1 host membership reports.
2332  *  IGMPv2 starts sending IGMPv2 host membership reports.
2333  *  IGMPv3 will schedule an IGMPv3 state-change report containing the
2334  *  initial state of the membership.
2335  */
2336 static int
2337 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2338 {
2339 	struct ifnet		*ifp;
2340 	struct ifqueue		*ifq;
2341 	int			 error, retval, syncstates;
2342 
2343 	CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2344 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2345 	    inm->inm_ifp->if_xname);
2346 
2347 	error = 0;
2348 	syncstates = 1;
2349 
2350 	ifp = inm->inm_ifp;
2351 
2352 	IN_MULTI_LOCK_ASSERT();
2353 	IGMP_LOCK_ASSERT();
2354 
2355 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2356 
2357 	/*
2358 	 * Groups joined on loopback or marked as 'not reported',
2359 	 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2360 	 * are never reported in any IGMP protocol exchanges.
2361 	 * All other groups enter the appropriate IGMP state machine
2362 	 * for the version in use on this link.
2363 	 * A link marked as IGIF_SILENT causes IGMP to be completely
2364 	 * disabled for the link.
2365 	 */
2366 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2367 	    (igi->igi_flags & IGIF_SILENT) ||
2368 	    !igmp_isgroupreported(inm->inm_addr)) {
2369 		CTR1(KTR_IGMPV3,
2370 "%s: not kicking state machine for silent group", __func__);
2371 		inm->inm_state = IGMP_SILENT_MEMBER;
2372 		inm->inm_timer = 0;
2373 	} else {
2374 		/*
2375 		 * Deal with overlapping in_multi lifecycle.
2376 		 * If this group was LEAVING, then make sure
2377 		 * we drop the reference we picked up to keep the
2378 		 * group around for the final INCLUDE {} enqueue.
2379 		 */
2380 		if (igi->igi_version == IGMP_VERSION_3 &&
2381 		    inm->inm_state == IGMP_LEAVING_MEMBER)
2382 			inm_release_locked(inm);
2383 
2384 		inm->inm_state = IGMP_REPORTING_MEMBER;
2385 
2386 		switch (igi->igi_version) {
2387 		case IGMP_VERSION_1:
2388 		case IGMP_VERSION_2:
2389 			inm->inm_state = IGMP_IDLE_MEMBER;
2390 			error = igmp_v1v2_queue_report(inm,
2391 			    (igi->igi_version == IGMP_VERSION_2) ?
2392 			     IGMP_v2_HOST_MEMBERSHIP_REPORT :
2393 			     IGMP_v1_HOST_MEMBERSHIP_REPORT);
2394 			if (error == 0) {
2395 				inm->inm_timer = IGMP_RANDOM_DELAY(
2396 				    IGMP_V1V2_MAX_RI * PR_FASTHZ);
2397 				V_current_state_timers_running = 1;
2398 			}
2399 			break;
2400 
2401 		case IGMP_VERSION_3:
2402 			/*
2403 			 * Defer update of T0 to T1, until the first copy
2404 			 * of the state change has been transmitted.
2405 			 */
2406 			syncstates = 0;
2407 
2408 			/*
2409 			 * Immediately enqueue a State-Change Report for
2410 			 * this interface, freeing any previous reports.
2411 			 * Don't kick the timers if there is nothing to do,
2412 			 * or if an error occurred.
2413 			 */
2414 			ifq = &inm->inm_scq;
2415 			_IF_DRAIN(ifq);
2416 			retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2417 			    0, 0);
2418 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2419 			    __func__, retval);
2420 			if (retval <= 0) {
2421 				error = retval * -1;
2422 				break;
2423 			}
2424 
2425 			/*
2426 			 * Schedule transmission of pending state-change
2427 			 * report up to RV times for this link. The timer
2428 			 * will fire at the next igmp_fasttimo (~200ms),
2429 			 * giving us an opportunity to merge the reports.
2430 			 */
2431 			if (igi->igi_flags & IGIF_LOOPBACK) {
2432 				inm->inm_scrv = 1;
2433 			} else {
2434 				KASSERT(igi->igi_rv > 1,
2435 				   ("%s: invalid robustness %d", __func__,
2436 				    igi->igi_rv));
2437 				inm->inm_scrv = igi->igi_rv;
2438 			}
2439 			inm->inm_sctimer = 1;
2440 			V_state_change_timers_running = 1;
2441 
2442 			error = 0;
2443 			break;
2444 		}
2445 	}
2446 
2447 	/*
2448 	 * Only update the T0 state if state change is atomic,
2449 	 * i.e. we don't need to wait for a timer to fire before we
2450 	 * can consider the state change to have been communicated.
2451 	 */
2452 	if (syncstates) {
2453 		inm_commit(inm);
2454 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2455 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2456 	}
2457 
2458 	return (error);
2459 }
2460 
2461 /*
2462  * Issue an intermediate state change during the IGMP life-cycle.
2463  */
2464 static int
2465 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2466 {
2467 	struct ifnet		*ifp;
2468 	int			 retval;
2469 
2470 	CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2471 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2472 	    inm->inm_ifp->if_xname);
2473 
2474 	ifp = inm->inm_ifp;
2475 
2476 	IN_MULTI_LOCK_ASSERT();
2477 	IGMP_LOCK_ASSERT();
2478 
2479 	KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2480 
2481 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2482 	    (igi->igi_flags & IGIF_SILENT) ||
2483 	    !igmp_isgroupreported(inm->inm_addr) ||
2484 	    (igi->igi_version != IGMP_VERSION_3)) {
2485 		if (!igmp_isgroupreported(inm->inm_addr)) {
2486 			CTR1(KTR_IGMPV3,
2487 "%s: not kicking state machine for silent group", __func__);
2488 		}
2489 		CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2490 		inm_commit(inm);
2491 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2492 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2493 		return (0);
2494 	}
2495 
2496 	_IF_DRAIN(&inm->inm_scq);
2497 
2498 	retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2499 	CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2500 	if (retval <= 0)
2501 		return (-retval);
2502 
2503 	/*
2504 	 * If record(s) were enqueued, start the state-change
2505 	 * report timer for this group.
2506 	 */
2507 	inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2508 	inm->inm_sctimer = 1;
2509 	V_state_change_timers_running = 1;
2510 
2511 	return (0);
2512 }
2513 
2514 /*
2515  * Perform the final leave for an IGMP group.
2516  *
2517  * When leaving a group:
2518  *  IGMPv1 does nothing.
2519  *  IGMPv2 sends a host leave message, if and only if we are the reporter.
2520  *  IGMPv3 enqueues a state-change report containing a transition
2521  *  to INCLUDE {} for immediate transmission.
2522  */
2523 static void
2524 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2525 {
2526 	int syncstates;
2527 
2528 	syncstates = 1;
2529 
2530 	CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2531 	    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2532 	    inm->inm_ifp->if_xname);
2533 
2534 	IN_MULTI_LOCK_ASSERT();
2535 	IGMP_LOCK_ASSERT();
2536 
2537 	switch (inm->inm_state) {
2538 	case IGMP_NOT_MEMBER:
2539 	case IGMP_SILENT_MEMBER:
2540 	case IGMP_LEAVING_MEMBER:
2541 		/* Already leaving or left; do nothing. */
2542 		CTR1(KTR_IGMPV3,
2543 "%s: not kicking state machine for silent group", __func__);
2544 		break;
2545 	case IGMP_REPORTING_MEMBER:
2546 	case IGMP_IDLE_MEMBER:
2547 	case IGMP_G_QUERY_PENDING_MEMBER:
2548 	case IGMP_SG_QUERY_PENDING_MEMBER:
2549 		if (igi->igi_version == IGMP_VERSION_2) {
2550 #ifdef INVARIANTS
2551 			if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2552 			    inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2553 			panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2554 			     __func__);
2555 #endif
2556 			igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2557 			inm->inm_state = IGMP_NOT_MEMBER;
2558 		} else if (igi->igi_version == IGMP_VERSION_3) {
2559 			/*
2560 			 * Stop group timer and all pending reports.
2561 			 * Immediately enqueue a state-change report
2562 			 * TO_IN {} to be sent on the next fast timeout,
2563 			 * giving us an opportunity to merge reports.
2564 			 */
2565 			_IF_DRAIN(&inm->inm_scq);
2566 			inm->inm_timer = 0;
2567 			if (igi->igi_flags & IGIF_LOOPBACK) {
2568 				inm->inm_scrv = 1;
2569 			} else {
2570 				inm->inm_scrv = igi->igi_rv;
2571 			}
2572 			CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2573 			    "pending retransmissions.", __func__,
2574 			    inet_ntoa(inm->inm_addr),
2575 			    inm->inm_ifp->if_xname, inm->inm_scrv);
2576 			if (inm->inm_scrv == 0) {
2577 				inm->inm_state = IGMP_NOT_MEMBER;
2578 				inm->inm_sctimer = 0;
2579 			} else {
2580 				int retval;
2581 
2582 				inm_acquire_locked(inm);
2583 
2584 				retval = igmp_v3_enqueue_group_record(
2585 				    &inm->inm_scq, inm, 1, 0, 0);
2586 				KASSERT(retval != 0,
2587 				    ("%s: enqueue record = %d", __func__,
2588 				     retval));
2589 
2590 				inm->inm_state = IGMP_LEAVING_MEMBER;
2591 				inm->inm_sctimer = 1;
2592 				V_state_change_timers_running = 1;
2593 				syncstates = 0;
2594 			}
2595 			break;
2596 		}
2597 		break;
2598 	case IGMP_LAZY_MEMBER:
2599 	case IGMP_SLEEPING_MEMBER:
2600 	case IGMP_AWAKENING_MEMBER:
2601 		/* Our reports are suppressed; do nothing. */
2602 		break;
2603 	}
2604 
2605 	if (syncstates) {
2606 		inm_commit(inm);
2607 		CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2608 		    inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2609 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2610 		CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2611 		    __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2612 	}
2613 }
2614 
2615 /*
2616  * Enqueue an IGMPv3 group record to the given output queue.
2617  *
2618  * XXX This function could do with having the allocation code
2619  * split out, and the multiple-tree-walks coalesced into a single
2620  * routine as has been done in igmp_v3_enqueue_filter_change().
2621  *
2622  * If is_state_change is zero, a current-state record is appended.
2623  * If is_state_change is non-zero, a state-change report is appended.
2624  *
2625  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2626  * If is_group_query is zero, and if there is a packet with free space
2627  * at the tail of the queue, it will be appended to providing there
2628  * is enough free space.
2629  * Otherwise a new mbuf packet chain is allocated.
2630  *
2631  * If is_source_query is non-zero, each source is checked to see if
2632  * it was recorded for a Group-Source query, and will be omitted if
2633  * it is not both in-mode and recorded.
2634  *
2635  * The function will attempt to allocate leading space in the packet
2636  * for the IP/IGMP header to be prepended without fragmenting the chain.
2637  *
2638  * If successful the size of all data appended to the queue is returned,
2639  * otherwise an error code less than zero is returned, or zero if
2640  * no record(s) were appended.
2641  */
2642 static int
2643 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2644     const int is_state_change, const int is_group_query,
2645     const int is_source_query)
2646 {
2647 	struct igmp_grouprec	 ig;
2648 	struct igmp_grouprec	*pig;
2649 	struct ifnet		*ifp;
2650 	struct ip_msource	*ims, *nims;
2651 	struct mbuf		*m0, *m, *md;
2652 	int			 error, is_filter_list_change;
2653 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2654 	int			 record_has_sources;
2655 	int			 now;
2656 	int			 type;
2657 	in_addr_t		 naddr;
2658 	uint8_t			 mode;
2659 
2660 	IN_MULTI_LOCK_ASSERT();
2661 
2662 	error = 0;
2663 	ifp = inm->inm_ifp;
2664 	is_filter_list_change = 0;
2665 	m = NULL;
2666 	m0 = NULL;
2667 	m0srcs = 0;
2668 	msrcs = 0;
2669 	nbytes = 0;
2670 	nims = NULL;
2671 	record_has_sources = 1;
2672 	pig = NULL;
2673 	type = IGMP_DO_NOTHING;
2674 	mode = inm->inm_st[1].iss_fmode;
2675 
2676 	/*
2677 	 * If we did not transition out of ASM mode during t0->t1,
2678 	 * and there are no source nodes to process, we can skip
2679 	 * the generation of source records.
2680 	 */
2681 	if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2682 	    inm->inm_nsrc == 0)
2683 		record_has_sources = 0;
2684 
2685 	if (is_state_change) {
2686 		/*
2687 		 * Queue a state change record.
2688 		 * If the mode did not change, and there are non-ASM
2689 		 * listeners or source filters present,
2690 		 * we potentially need to issue two records for the group.
2691 		 * If we are transitioning to MCAST_UNDEFINED, we need
2692 		 * not send any sources.
2693 		 * If there are ASM listeners, and there was no filter
2694 		 * mode transition of any kind, do nothing.
2695 		 */
2696 		if (mode != inm->inm_st[0].iss_fmode) {
2697 			if (mode == MCAST_EXCLUDE) {
2698 				CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2699 				    __func__);
2700 				type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2701 			} else {
2702 				CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2703 				    __func__);
2704 				type = IGMP_CHANGE_TO_INCLUDE_MODE;
2705 				if (mode == MCAST_UNDEFINED)
2706 					record_has_sources = 0;
2707 			}
2708 		} else {
2709 			if (record_has_sources) {
2710 				is_filter_list_change = 1;
2711 			} else {
2712 				type = IGMP_DO_NOTHING;
2713 			}
2714 		}
2715 	} else {
2716 		/*
2717 		 * Queue a current state record.
2718 		 */
2719 		if (mode == MCAST_EXCLUDE) {
2720 			type = IGMP_MODE_IS_EXCLUDE;
2721 		} else if (mode == MCAST_INCLUDE) {
2722 			type = IGMP_MODE_IS_INCLUDE;
2723 			KASSERT(inm->inm_st[1].iss_asm == 0,
2724 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2725 			     __func__, inm, inm->inm_st[1].iss_asm));
2726 		}
2727 	}
2728 
2729 	/*
2730 	 * Generate the filter list changes using a separate function.
2731 	 */
2732 	if (is_filter_list_change)
2733 		return (igmp_v3_enqueue_filter_change(ifq, inm));
2734 
2735 	if (type == IGMP_DO_NOTHING) {
2736 		CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2737 		    __func__, inet_ntoa(inm->inm_addr),
2738 		    inm->inm_ifp->if_xname);
2739 		return (0);
2740 	}
2741 
2742 	/*
2743 	 * If any sources are present, we must be able to fit at least
2744 	 * one in the trailing space of the tail packet's mbuf,
2745 	 * ideally more.
2746 	 */
2747 	minrec0len = sizeof(struct igmp_grouprec);
2748 	if (record_has_sources)
2749 		minrec0len += sizeof(in_addr_t);
2750 
2751 	CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2752 	    igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2753 	    inm->inm_ifp->if_xname);
2754 
2755 	/*
2756 	 * Check if we have a packet in the tail of the queue for this
2757 	 * group into which the first group record for this group will fit.
2758 	 * Otherwise allocate a new packet.
2759 	 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2760 	 * Note: Group records for G/GSR query responses MUST be sent
2761 	 * in their own packet.
2762 	 */
2763 	m0 = ifq->ifq_tail;
2764 	if (!is_group_query &&
2765 	    m0 != NULL &&
2766 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2767 	    (m0->m_pkthdr.len + minrec0len) <
2768 	     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2769 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2770 			    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2771 		m = m0;
2772 		CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2773 	} else {
2774 		if (_IF_QFULL(ifq)) {
2775 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2776 			return (-ENOMEM);
2777 		}
2778 		m = NULL;
2779 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2780 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2781 		if (!is_state_change && !is_group_query) {
2782 			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2783 			if (m)
2784 				m->m_data += IGMP_LEADINGSPACE;
2785 		}
2786 		if (m == NULL) {
2787 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2788 			if (m)
2789 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2790 		}
2791 		if (m == NULL)
2792 			return (-ENOMEM);
2793 
2794 		igmp_save_context(m, ifp);
2795 
2796 		CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2797 	}
2798 
2799 	/*
2800 	 * Append group record.
2801 	 * If we have sources, we don't know how many yet.
2802 	 */
2803 	ig.ig_type = type;
2804 	ig.ig_datalen = 0;
2805 	ig.ig_numsrc = 0;
2806 	ig.ig_group = inm->inm_addr;
2807 	if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2808 		if (m != m0)
2809 			m_freem(m);
2810 		CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2811 		return (-ENOMEM);
2812 	}
2813 	nbytes += sizeof(struct igmp_grouprec);
2814 
2815 	/*
2816 	 * Append as many sources as will fit in the first packet.
2817 	 * If we are appending to a new packet, the chain allocation
2818 	 * may potentially use clusters; use m_getptr() in this case.
2819 	 * If we are appending to an existing packet, we need to obtain
2820 	 * a pointer to the group record after m_append(), in case a new
2821 	 * mbuf was allocated.
2822 	 * Only append sources which are in-mode at t1. If we are
2823 	 * transitioning to MCAST_UNDEFINED state on the group, do not
2824 	 * include source entries.
2825 	 * Only report recorded sources in our filter set when responding
2826 	 * to a group-source query.
2827 	 */
2828 	if (record_has_sources) {
2829 		if (m == m0) {
2830 			md = m_last(m);
2831 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2832 			    md->m_len - nbytes);
2833 		} else {
2834 			md = m_getptr(m, 0, &off);
2835 			pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2836 			    off);
2837 		}
2838 		msrcs = 0;
2839 		RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2840 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2841 			    inet_ntoa_haddr(ims->ims_haddr));
2842 			now = ims_get_mode(inm, ims, 1);
2843 			CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2844 			if ((now != mode) ||
2845 			    (now == mode && mode == MCAST_UNDEFINED)) {
2846 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2847 				continue;
2848 			}
2849 			if (is_source_query && ims->ims_stp == 0) {
2850 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2851 				    __func__);
2852 				continue;
2853 			}
2854 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2855 			naddr = htonl(ims->ims_haddr);
2856 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2857 				if (m != m0)
2858 					m_freem(m);
2859 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2860 				    __func__);
2861 				return (-ENOMEM);
2862 			}
2863 			nbytes += sizeof(in_addr_t);
2864 			++msrcs;
2865 			if (msrcs == m0srcs)
2866 				break;
2867 		}
2868 		CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2869 		    msrcs);
2870 		pig->ig_numsrc = htons(msrcs);
2871 		nbytes += (msrcs * sizeof(in_addr_t));
2872 	}
2873 
2874 	if (is_source_query && msrcs == 0) {
2875 		CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2876 		if (m != m0)
2877 			m_freem(m);
2878 		return (0);
2879 	}
2880 
2881 	/*
2882 	 * We are good to go with first packet.
2883 	 */
2884 	if (m != m0) {
2885 		CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2886 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2887 		_IF_ENQUEUE(ifq, m);
2888 	} else
2889 		m->m_pkthdr.PH_vt.vt_nrecs++;
2890 
2891 	/*
2892 	 * No further work needed if no source list in packet(s).
2893 	 */
2894 	if (!record_has_sources)
2895 		return (nbytes);
2896 
2897 	/*
2898 	 * Whilst sources remain to be announced, we need to allocate
2899 	 * a new packet and fill out as many sources as will fit.
2900 	 * Always try for a cluster first.
2901 	 */
2902 	while (nims != NULL) {
2903 		if (_IF_QFULL(ifq)) {
2904 			CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2905 			return (-ENOMEM);
2906 		}
2907 		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2908 		if (m)
2909 			m->m_data += IGMP_LEADINGSPACE;
2910 		if (m == NULL) {
2911 			m = m_gethdr(M_DONTWAIT, MT_DATA);
2912 			if (m)
2913 				MH_ALIGN(m, IGMP_LEADINGSPACE);
2914 		}
2915 		if (m == NULL)
2916 			return (-ENOMEM);
2917 		igmp_save_context(m, ifp);
2918 		md = m_getptr(m, 0, &off);
2919 		pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2920 		CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2921 
2922 		if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2923 			if (m != m0)
2924 				m_freem(m);
2925 			CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2926 			return (-ENOMEM);
2927 		}
2928 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2929 		nbytes += sizeof(struct igmp_grouprec);
2930 
2931 		m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2932 		    sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2933 
2934 		msrcs = 0;
2935 		RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2936 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2937 			    inet_ntoa_haddr(ims->ims_haddr));
2938 			now = ims_get_mode(inm, ims, 1);
2939 			if ((now != mode) ||
2940 			    (now == mode && mode == MCAST_UNDEFINED)) {
2941 				CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2942 				continue;
2943 			}
2944 			if (is_source_query && ims->ims_stp == 0) {
2945 				CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2946 				    __func__);
2947 				continue;
2948 			}
2949 			CTR1(KTR_IGMPV3, "%s: append node", __func__);
2950 			naddr = htonl(ims->ims_haddr);
2951 			if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2952 				if (m != m0)
2953 					m_freem(m);
2954 				CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2955 				    __func__);
2956 				return (-ENOMEM);
2957 			}
2958 			++msrcs;
2959 			if (msrcs == m0srcs)
2960 				break;
2961 		}
2962 		pig->ig_numsrc = htons(msrcs);
2963 		nbytes += (msrcs * sizeof(in_addr_t));
2964 
2965 		CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2966 		_IF_ENQUEUE(ifq, m);
2967 	}
2968 
2969 	return (nbytes);
2970 }
2971 
2972 /*
2973  * Type used to mark record pass completion.
2974  * We exploit the fact we can cast to this easily from the
2975  * current filter modes on each ip_msource node.
2976  */
2977 typedef enum {
2978 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2979 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2980 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2981 	REC_FULL = REC_ALLOW | REC_BLOCK
2982 } rectype_t;
2983 
2984 /*
2985  * Enqueue an IGMPv3 filter list change to the given output queue.
2986  *
2987  * Source list filter state is held in an RB-tree. When the filter list
2988  * for a group is changed without changing its mode, we need to compute
2989  * the deltas between T0 and T1 for each source in the filter set,
2990  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2991  *
2992  * As we may potentially queue two record types, and the entire R-B tree
2993  * needs to be walked at once, we break this out into its own function
2994  * so we can generate a tightly packed queue of packets.
2995  *
2996  * XXX This could be written to only use one tree walk, although that makes
2997  * serializing into the mbuf chains a bit harder. For now we do two walks
2998  * which makes things easier on us, and it may or may not be harder on
2999  * the L2 cache.
3000  *
3001  * If successful the size of all data appended to the queue is returned,
3002  * otherwise an error code less than zero is returned, or zero if
3003  * no record(s) were appended.
3004  */
3005 static int
3006 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
3007 {
3008 	static const int MINRECLEN =
3009 	    sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3010 	struct ifnet		*ifp;
3011 	struct igmp_grouprec	 ig;
3012 	struct igmp_grouprec	*pig;
3013 	struct ip_msource	*ims, *nims;
3014 	struct mbuf		*m, *m0, *md;
3015 	in_addr_t		 naddr;
3016 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3017 	int			 nallow, nblock;
3018 	uint8_t			 mode, now, then;
3019 	rectype_t		 crt, drt, nrt;
3020 
3021 	IN_MULTI_LOCK_ASSERT();
3022 
3023 	if (inm->inm_nsrc == 0 ||
3024 	    (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3025 		return (0);
3026 
3027 	ifp = inm->inm_ifp;			/* interface */
3028 	mode = inm->inm_st[1].iss_fmode;	/* filter mode at t1 */
3029 	crt = REC_NONE;	/* current group record type */
3030 	drt = REC_NONE;	/* mask of completed group record types */
3031 	nrt = REC_NONE;	/* record type for current node */
3032 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3033 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3034 	npbytes = 0;	/* # of bytes appended this packet */
3035 	rsrcs = 0;	/* # sources encoded in current record */
3036 	schanged = 0;	/* # nodes encoded in overall filter change */
3037 	nallow = 0;	/* # of source entries in ALLOW_NEW */
3038 	nblock = 0;	/* # of source entries in BLOCK_OLD */
3039 	nims = NULL;	/* next tree node pointer */
3040 
3041 	/*
3042 	 * For each possible filter record mode.
3043 	 * The first kind of source we encounter tells us which
3044 	 * is the first kind of record we start appending.
3045 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3046 	 * as the inverse of the group's filter mode.
3047 	 */
3048 	while (drt != REC_FULL) {
3049 		do {
3050 			m0 = ifq->ifq_tail;
3051 			if (m0 != NULL &&
3052 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3053 			     IGMP_V3_REPORT_MAXRECS) &&
3054 			    (m0->m_pkthdr.len + MINRECLEN) <
3055 			     (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3056 				m = m0;
3057 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3058 					    sizeof(struct igmp_grouprec)) /
3059 				    sizeof(in_addr_t);
3060 				CTR1(KTR_IGMPV3,
3061 				    "%s: use previous packet", __func__);
3062 			} else {
3063 				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3064 				if (m)
3065 					m->m_data += IGMP_LEADINGSPACE;
3066 				if (m == NULL) {
3067 					m = m_gethdr(M_DONTWAIT, MT_DATA);
3068 					if (m)
3069 						MH_ALIGN(m, IGMP_LEADINGSPACE);
3070 				}
3071 				if (m == NULL) {
3072 					CTR1(KTR_IGMPV3,
3073 					    "%s: m_get*() failed", __func__);
3074 					return (-ENOMEM);
3075 				}
3076 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
3077 				igmp_save_context(m, ifp);
3078 				m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3079 				    sizeof(struct igmp_grouprec)) /
3080 				    sizeof(in_addr_t);
3081 				npbytes = 0;
3082 				CTR1(KTR_IGMPV3,
3083 				    "%s: allocated new packet", __func__);
3084 			}
3085 			/*
3086 			 * Append the IGMP group record header to the
3087 			 * current packet's data area.
3088 			 * Recalculate pointer to free space for next
3089 			 * group record, in case m_append() allocated
3090 			 * a new mbuf or cluster.
3091 			 */
3092 			memset(&ig, 0, sizeof(ig));
3093 			ig.ig_group = inm->inm_addr;
3094 			if (!m_append(m, sizeof(ig), (void *)&ig)) {
3095 				if (m != m0)
3096 					m_freem(m);
3097 				CTR1(KTR_IGMPV3,
3098 				    "%s: m_append() failed", __func__);
3099 				return (-ENOMEM);
3100 			}
3101 			npbytes += sizeof(struct igmp_grouprec);
3102 			if (m != m0) {
3103 				/* new packet; offset in c hain */
3104 				md = m_getptr(m, npbytes -
3105 				    sizeof(struct igmp_grouprec), &off);
3106 				pig = (struct igmp_grouprec *)(mtod(md,
3107 				    uint8_t *) + off);
3108 			} else {
3109 				/* current packet; offset from last append */
3110 				md = m_last(m);
3111 				pig = (struct igmp_grouprec *)(mtod(md,
3112 				    uint8_t *) + md->m_len -
3113 				    sizeof(struct igmp_grouprec));
3114 			}
3115 			/*
3116 			 * Begin walking the tree for this record type
3117 			 * pass, or continue from where we left off
3118 			 * previously if we had to allocate a new packet.
3119 			 * Only report deltas in-mode at t1.
3120 			 * We need not report included sources as allowed
3121 			 * if we are in inclusive mode on the group,
3122 			 * however the converse is not true.
3123 			 */
3124 			rsrcs = 0;
3125 			if (nims == NULL)
3126 				nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3127 			RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3128 				CTR2(KTR_IGMPV3, "%s: visit node %s",
3129 				    __func__, inet_ntoa_haddr(ims->ims_haddr));
3130 				now = ims_get_mode(inm, ims, 1);
3131 				then = ims_get_mode(inm, ims, 0);
3132 				CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3133 				    __func__, then, now);
3134 				if (now == then) {
3135 					CTR1(KTR_IGMPV3,
3136 					    "%s: skip unchanged", __func__);
3137 					continue;
3138 				}
3139 				if (mode == MCAST_EXCLUDE &&
3140 				    now == MCAST_INCLUDE) {
3141 					CTR1(KTR_IGMPV3,
3142 					    "%s: skip IN src on EX group",
3143 					    __func__);
3144 					continue;
3145 				}
3146 				nrt = (rectype_t)now;
3147 				if (nrt == REC_NONE)
3148 					nrt = (rectype_t)(~mode & REC_FULL);
3149 				if (schanged++ == 0) {
3150 					crt = nrt;
3151 				} else if (crt != nrt)
3152 					continue;
3153 				naddr = htonl(ims->ims_haddr);
3154 				if (!m_append(m, sizeof(in_addr_t),
3155 				    (void *)&naddr)) {
3156 					if (m != m0)
3157 						m_freem(m);
3158 					CTR1(KTR_IGMPV3,
3159 					    "%s: m_append() failed", __func__);
3160 					return (-ENOMEM);
3161 				}
3162 				nallow += !!(crt == REC_ALLOW);
3163 				nblock += !!(crt == REC_BLOCK);
3164 				if (++rsrcs == m0srcs)
3165 					break;
3166 			}
3167 			/*
3168 			 * If we did not append any tree nodes on this
3169 			 * pass, back out of allocations.
3170 			 */
3171 			if (rsrcs == 0) {
3172 				npbytes -= sizeof(struct igmp_grouprec);
3173 				if (m != m0) {
3174 					CTR1(KTR_IGMPV3,
3175 					    "%s: m_free(m)", __func__);
3176 					m_freem(m);
3177 				} else {
3178 					CTR1(KTR_IGMPV3,
3179 					    "%s: m_adj(m, -ig)", __func__);
3180 					m_adj(m, -((int)sizeof(
3181 					    struct igmp_grouprec)));
3182 				}
3183 				continue;
3184 			}
3185 			npbytes += (rsrcs * sizeof(in_addr_t));
3186 			if (crt == REC_ALLOW)
3187 				pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3188 			else if (crt == REC_BLOCK)
3189 				pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3190 			pig->ig_numsrc = htons(rsrcs);
3191 			/*
3192 			 * Count the new group record, and enqueue this
3193 			 * packet if it wasn't already queued.
3194 			 */
3195 			m->m_pkthdr.PH_vt.vt_nrecs++;
3196 			if (m != m0)
3197 				_IF_ENQUEUE(ifq, m);
3198 			nbytes += npbytes;
3199 		} while (nims != NULL);
3200 		drt |= crt;
3201 		crt = (~crt & REC_FULL);
3202 	}
3203 
3204 	CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3205 	    nallow, nblock);
3206 
3207 	return (nbytes);
3208 }
3209 
3210 static int
3211 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3212 {
3213 	struct ifqueue	*gq;
3214 	struct mbuf	*m;		/* pending state-change */
3215 	struct mbuf	*m0;		/* copy of pending state-change */
3216 	struct mbuf	*mt;		/* last state-change in packet */
3217 	int		 docopy, domerge;
3218 	u_int		 recslen;
3219 
3220 	docopy = 0;
3221 	domerge = 0;
3222 	recslen = 0;
3223 
3224 	IN_MULTI_LOCK_ASSERT();
3225 	IGMP_LOCK_ASSERT();
3226 
3227 	/*
3228 	 * If there are further pending retransmissions, make a writable
3229 	 * copy of each queued state-change message before merging.
3230 	 */
3231 	if (inm->inm_scrv > 0)
3232 		docopy = 1;
3233 
3234 	gq = &inm->inm_scq;
3235 #ifdef KTR
3236 	if (gq->ifq_head == NULL) {
3237 		CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3238 		    __func__, inm);
3239 	}
3240 #endif
3241 
3242 	m = gq->ifq_head;
3243 	while (m != NULL) {
3244 		/*
3245 		 * Only merge the report into the current packet if
3246 		 * there is sufficient space to do so; an IGMPv3 report
3247 		 * packet may only contain 65,535 group records.
3248 		 * Always use a simple mbuf chain concatentation to do this,
3249 		 * as large state changes for single groups may have
3250 		 * allocated clusters.
3251 		 */
3252 		domerge = 0;
3253 		mt = ifscq->ifq_tail;
3254 		if (mt != NULL) {
3255 			recslen = m_length(m, NULL);
3256 
3257 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3258 			    m->m_pkthdr.PH_vt.vt_nrecs <=
3259 			    IGMP_V3_REPORT_MAXRECS) &&
3260 			    (mt->m_pkthdr.len + recslen <=
3261 			    (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3262 				domerge = 1;
3263 		}
3264 
3265 		if (!domerge && _IF_QFULL(gq)) {
3266 			CTR2(KTR_IGMPV3,
3267 			    "%s: outbound queue full, skipping whole packet %p",
3268 			    __func__, m);
3269 			mt = m->m_nextpkt;
3270 			if (!docopy)
3271 				m_freem(m);
3272 			m = mt;
3273 			continue;
3274 		}
3275 
3276 		if (!docopy) {
3277 			CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3278 			_IF_DEQUEUE(gq, m0);
3279 			m = m0->m_nextpkt;
3280 		} else {
3281 			CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3282 			m0 = m_dup(m, M_NOWAIT);
3283 			if (m0 == NULL)
3284 				return (ENOMEM);
3285 			m0->m_nextpkt = NULL;
3286 			m = m->m_nextpkt;
3287 		}
3288 
3289 		if (!domerge) {
3290 			CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3291 			    __func__, m0, ifscq);
3292 			_IF_ENQUEUE(ifscq, m0);
3293 		} else {
3294 			struct mbuf *mtl;	/* last mbuf of packet mt */
3295 
3296 			CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3297 			    __func__, m0, mt);
3298 
3299 			mtl = m_last(mt);
3300 			m0->m_flags &= ~M_PKTHDR;
3301 			mt->m_pkthdr.len += recslen;
3302 			mt->m_pkthdr.PH_vt.vt_nrecs +=
3303 			    m0->m_pkthdr.PH_vt.vt_nrecs;
3304 
3305 			mtl->m_next = m0;
3306 		}
3307 	}
3308 
3309 	return (0);
3310 }
3311 
3312 /*
3313  * Respond to a pending IGMPv3 General Query.
3314  */
3315 static void
3316 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3317 {
3318 	struct ifmultiaddr	*ifma, *tifma;
3319 	struct ifnet		*ifp;
3320 	struct in_multi		*inm;
3321 	int			 retval, loop;
3322 
3323 	IN_MULTI_LOCK_ASSERT();
3324 	IGMP_LOCK_ASSERT();
3325 
3326 	KASSERT(igi->igi_version == IGMP_VERSION_3,
3327 	    ("%s: called when version %d", __func__, igi->igi_version));
3328 
3329 	ifp = igi->igi_ifp;
3330 
3331 	IF_ADDR_LOCK(ifp);
3332 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
3333 		if (ifma->ifma_addr->sa_family != AF_INET ||
3334 		    ifma->ifma_protospec == NULL)
3335 			continue;
3336 
3337 		inm = (struct in_multi *)ifma->ifma_protospec;
3338 		KASSERT(ifp == inm->inm_ifp,
3339 		    ("%s: inconsistent ifp", __func__));
3340 
3341 		switch (inm->inm_state) {
3342 		case IGMP_NOT_MEMBER:
3343 		case IGMP_SILENT_MEMBER:
3344 			break;
3345 		case IGMP_REPORTING_MEMBER:
3346 		case IGMP_IDLE_MEMBER:
3347 		case IGMP_LAZY_MEMBER:
3348 		case IGMP_SLEEPING_MEMBER:
3349 		case IGMP_AWAKENING_MEMBER:
3350 			inm->inm_state = IGMP_REPORTING_MEMBER;
3351 			retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3352 			    inm, 0, 0, 0);
3353 			CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3354 			    __func__, retval);
3355 			break;
3356 		case IGMP_G_QUERY_PENDING_MEMBER:
3357 		case IGMP_SG_QUERY_PENDING_MEMBER:
3358 		case IGMP_LEAVING_MEMBER:
3359 			break;
3360 		}
3361 	}
3362 	IF_ADDR_UNLOCK(ifp);
3363 
3364 	loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3365 	igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3366 
3367 	/*
3368 	 * Slew transmission of bursts over 500ms intervals.
3369 	 */
3370 	if (igi->igi_gq.ifq_head != NULL) {
3371 		igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3372 		    IGMP_RESPONSE_BURST_INTERVAL);
3373 		V_interface_timers_running = 1;
3374 	}
3375 }
3376 
3377 /*
3378  * Transmit the next pending IGMP message in the output queue.
3379  *
3380  * We get called from netisr_processqueue(). A mutex private to igmpoq
3381  * will be acquired and released around this routine.
3382  *
3383  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3384  * MRT: Nothing needs to be done, as IGMP traffic is always local to
3385  * a link and uses a link-scope multicast address.
3386  */
3387 static void
3388 igmp_intr(struct mbuf *m)
3389 {
3390 	struct ip_moptions	 imo;
3391 	struct ifnet		*ifp;
3392 	struct mbuf		*ipopts, *m0;
3393 	int			 error;
3394 	uint32_t		 ifindex;
3395 
3396 	CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3397 
3398 	/*
3399 	 * Set VNET image pointer from enqueued mbuf chain
3400 	 * before doing anything else. Whilst we use interface
3401 	 * indexes to guard against interface detach, they are
3402 	 * unique to each VIMAGE and must be retrieved.
3403 	 */
3404 	CURVNET_SET((struct vnet *)(m->m_pkthdr.header));
3405 	ifindex = igmp_restore_context(m);
3406 
3407 	/*
3408 	 * Check if the ifnet still exists. This limits the scope of
3409 	 * any race in the absence of a global ifp lock for low cost
3410 	 * (an array lookup).
3411 	 */
3412 	ifp = ifnet_byindex(ifindex);
3413 	if (ifp == NULL) {
3414 		CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3415 		    __func__, m, ifindex);
3416 		m_freem(m);
3417 		IPSTAT_INC(ips_noroute);
3418 		goto out;
3419 	}
3420 
3421 	ipopts = V_igmp_sendra ? m_raopt : NULL;
3422 
3423 	imo.imo_multicast_ttl  = 1;
3424 	imo.imo_multicast_vif  = -1;
3425 	imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3426 
3427 	/*
3428 	 * If the user requested that IGMP traffic be explicitly
3429 	 * redirected to the loopback interface (e.g. they are running a
3430 	 * MANET interface and the routing protocol needs to see the
3431 	 * updates), handle this now.
3432 	 */
3433 	if (m->m_flags & M_IGMP_LOOP)
3434 		imo.imo_multicast_ifp = V_loif;
3435 	else
3436 		imo.imo_multicast_ifp = ifp;
3437 
3438 	if (m->m_flags & M_IGMPV2) {
3439 		m0 = m;
3440 	} else {
3441 		m0 = igmp_v3_encap_report(ifp, m);
3442 		if (m0 == NULL) {
3443 			CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3444 			m_freem(m);
3445 			IPSTAT_INC(ips_odropped);
3446 			goto out;
3447 		}
3448 	}
3449 
3450 	igmp_scrub_context(m0);
3451 	m->m_flags &= ~(M_PROTOFLAGS);
3452 	m0->m_pkthdr.rcvif = V_loif;
3453 #ifdef MAC
3454 	mac_netinet_igmp_send(ifp, m0);
3455 #endif
3456 	error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3457 	if (error) {
3458 		CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3459 		goto out;
3460 	}
3461 
3462 	IGMPSTAT_INC(igps_snd_reports);
3463 
3464 out:
3465 	/*
3466 	 * We must restore the existing vnet pointer before
3467 	 * continuing as we are run from netisr context.
3468 	 */
3469 	CURVNET_RESTORE();
3470 }
3471 
3472 /*
3473  * Encapsulate an IGMPv3 report.
3474  *
3475  * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3476  * chain has already had its IP/IGMPv3 header prepended. In this case
3477  * the function will not attempt to prepend; the lengths and checksums
3478  * will however be re-computed.
3479  *
3480  * Returns a pointer to the new mbuf chain head, or NULL if the
3481  * allocation failed.
3482  */
3483 static struct mbuf *
3484 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3485 {
3486 	struct igmp_report	*igmp;
3487 	struct ip		*ip;
3488 	int			 hdrlen, igmpreclen;
3489 
3490 	KASSERT((m->m_flags & M_PKTHDR),
3491 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3492 
3493 	igmpreclen = m_length(m, NULL);
3494 	hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3495 
3496 	if (m->m_flags & M_IGMPV3_HDR) {
3497 		igmpreclen -= hdrlen;
3498 	} else {
3499 		M_PREPEND(m, hdrlen, M_DONTWAIT);
3500 		if (m == NULL)
3501 			return (NULL);
3502 		m->m_flags |= M_IGMPV3_HDR;
3503 	}
3504 
3505 	CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3506 
3507 	m->m_data += sizeof(struct ip);
3508 	m->m_len -= sizeof(struct ip);
3509 
3510 	igmp = mtod(m, struct igmp_report *);
3511 	igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3512 	igmp->ir_rsv1 = 0;
3513 	igmp->ir_rsv2 = 0;
3514 	igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3515 	igmp->ir_cksum = 0;
3516 	igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3517 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3518 
3519 	m->m_data -= sizeof(struct ip);
3520 	m->m_len += sizeof(struct ip);
3521 
3522 	ip = mtod(m, struct ip *);
3523 	ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3524 	ip->ip_len = hdrlen + igmpreclen;
3525 	ip->ip_off = IP_DF;
3526 	ip->ip_p = IPPROTO_IGMP;
3527 	ip->ip_sum = 0;
3528 
3529 	ip->ip_src.s_addr = INADDR_ANY;
3530 
3531 	if (m->m_flags & M_IGMP_LOOP) {
3532 		struct in_ifaddr *ia;
3533 
3534 		IFP_TO_IA(ifp, ia);
3535 		if (ia != NULL) {
3536 			ip->ip_src = ia->ia_addr.sin_addr;
3537 			ifa_free(&ia->ia_ifa);
3538 		}
3539 	}
3540 
3541 	ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3542 
3543 	return (m);
3544 }
3545 
3546 #ifdef KTR
3547 static char *
3548 igmp_rec_type_to_str(const int type)
3549 {
3550 
3551 	switch (type) {
3552 		case IGMP_CHANGE_TO_EXCLUDE_MODE:
3553 			return "TO_EX";
3554 			break;
3555 		case IGMP_CHANGE_TO_INCLUDE_MODE:
3556 			return "TO_IN";
3557 			break;
3558 		case IGMP_MODE_IS_EXCLUDE:
3559 			return "MODE_EX";
3560 			break;
3561 		case IGMP_MODE_IS_INCLUDE:
3562 			return "MODE_IN";
3563 			break;
3564 		case IGMP_ALLOW_NEW_SOURCES:
3565 			return "ALLOW_NEW";
3566 			break;
3567 		case IGMP_BLOCK_OLD_SOURCES:
3568 			return "BLOCK_OLD";
3569 			break;
3570 		default:
3571 			break;
3572 	}
3573 	return "unknown";
3574 }
3575 #endif
3576 
3577 static void
3578 igmp_init(void *unused __unused)
3579 {
3580 
3581 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3582 
3583 	IGMP_LOCK_INIT();
3584 
3585 	m_raopt = igmp_ra_alloc();
3586 
3587 	netisr_register(&igmp_nh);
3588 }
3589 SYSINIT(igmp_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_init, NULL);
3590 
3591 static void
3592 igmp_uninit(void *unused __unused)
3593 {
3594 
3595 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3596 
3597 	netisr_unregister(&igmp_nh);
3598 
3599 	m_free(m_raopt);
3600 	m_raopt = NULL;
3601 
3602 	IGMP_LOCK_DESTROY();
3603 }
3604 SYSUNINIT(igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_uninit, NULL);
3605 
3606 static void
3607 vnet_igmp_init(const void *unused __unused)
3608 {
3609 
3610 	CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3611 
3612 	LIST_INIT(&V_igi_head);
3613 }
3614 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_igmp_init,
3615     NULL);
3616 
3617 static void
3618 vnet_igmp_uninit(const void *unused __unused)
3619 {
3620 
3621 	CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3622 
3623 	KASSERT(LIST_EMPTY(&V_igi_head),
3624 	    ("%s: igi list not empty; ifnets not detached?", __func__));
3625 }
3626 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY,
3627     vnet_igmp_uninit, NULL);
3628 
3629 static int
3630 igmp_modevent(module_t mod, int type, void *unused __unused)
3631 {
3632 
3633     switch (type) {
3634     case MOD_LOAD:
3635     case MOD_UNLOAD:
3636 	break;
3637     default:
3638 	return (EOPNOTSUPP);
3639     }
3640     return (0);
3641 }
3642 
3643 static moduledata_t igmp_mod = {
3644     "igmp",
3645     igmp_modevent,
3646     0
3647 };
3648 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
3649