FreeBSD/Linux Kernel Cross Reference
sys/netinet/igmp.c
1 /*-
2 * Copyright (c) 2007-2009 Bruce Simpson.
3 * Copyright (c) 1988 Stephen Deering.
4 * Copyright (c) 1992, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Stephen Deering of Stanford University.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)igmp.c 8.1 (Berkeley) 7/19/93
35 */
36
37 /*
38 * Internet Group Management Protocol (IGMP) routines.
39 * [RFC1112, RFC2236, RFC3376]
40 *
41 * Written by Steve Deering, Stanford, May 1988.
42 * Modified by Rosen Sharma, Stanford, Aug 1994.
43 * Modified by Bill Fenner, Xerox PARC, Feb 1995.
44 * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995.
45 * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson.
46 *
47 * MULTICAST Revision: 3.5.1.4
48 */
49
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
52
53 #include "opt_ddb.h"
54
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/module.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/socket.h>
61 #include <sys/protosw.h>
62 #include <sys/kernel.h>
63 #include <sys/lock.h>
64 #include <sys/rmlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/ktr.h>
67 #include <sys/condvar.h>
68
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif
72
73 #include <net/if.h>
74 #include <net/if_var.h>
75 #include <net/netisr.h>
76 #include <net/vnet.h>
77
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_var.h>
83 #include <netinet/ip_options.h>
84 #include <netinet/igmp.h>
85 #include <netinet/igmp_var.h>
86
87 #include <machine/in_cksum.h>
88
89 #include <security/mac/mac_framework.h>
90
91 #ifndef KTR_IGMPV3
92 #define KTR_IGMPV3 KTR_INET
93 #endif
94
95 static struct igmp_ifsoftc *
96 igi_alloc_locked(struct ifnet *);
97 static void igi_delete_locked(const struct ifnet *);
98 static void igmp_dispatch_queue(struct mbufq *, int, const int);
99 static void igmp_fasttimo_vnet(void);
100 static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
101 static int igmp_handle_state_change(struct in_multi *,
102 struct igmp_ifsoftc *);
103 static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
104 static int igmp_input_v1_query(struct ifnet *, const struct ip *,
105 const struct igmp *);
106 static int igmp_input_v2_query(struct ifnet *, const struct ip *,
107 const struct igmp *);
108 static int igmp_input_v3_query(struct ifnet *, const struct ip *,
109 /*const*/ struct igmpv3 *);
110 static int igmp_input_v3_group_query(struct in_multi *,
111 struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
112 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
113 /*const*/ struct igmp *);
114 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
115 /*const*/ struct igmp *);
116 static void igmp_intr(struct mbuf *);
117 static int igmp_isgroupreported(const struct in_addr);
118 static struct mbuf *
119 igmp_ra_alloc(void);
120 #ifdef KTR
121 static char * igmp_rec_type_to_str(const int);
122 #endif
123 static void igmp_set_version(struct igmp_ifsoftc *, const int);
124 static void igmp_slowtimo_vnet(void);
125 static int igmp_v1v2_queue_report(struct in_multi *, const int);
126 static void igmp_v1v2_process_group_timer(struct in_multi *, const int);
127 static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
128 static void igmp_v2_update_group(struct in_multi *, const int);
129 static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
130 static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
131 static struct mbuf *
132 igmp_v3_encap_report(struct ifnet *, struct mbuf *);
133 static int igmp_v3_enqueue_group_record(struct mbufq *,
134 struct in_multi *, const int, const int, const int);
135 static int igmp_v3_enqueue_filter_change(struct mbufq *,
136 struct in_multi *);
137 static void igmp_v3_process_group_timers(struct igmp_ifsoftc *,
138 struct mbufq *, struct mbufq *, struct in_multi *,
139 const int);
140 static int igmp_v3_merge_state_changes(struct in_multi *,
141 struct mbufq *);
142 static void igmp_v3_suppress_group_record(struct in_multi *);
143 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
144 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
145 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
146
147 static const struct netisr_handler igmp_nh = {
148 .nh_name = "igmp",
149 .nh_handler = igmp_intr,
150 .nh_proto = NETISR_IGMP,
151 .nh_policy = NETISR_POLICY_SOURCE,
152 };
153
154 /*
155 * System-wide globals.
156 *
157 * Unlocked access to these is OK, except for the global IGMP output
158 * queue. The IGMP subsystem lock ends up being system-wide for the moment,
159 * because all VIMAGEs have to share a global output queue, as netisrs
160 * themselves are not virtualized.
161 *
162 * Locking:
163 * * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
164 * Any may be taken independently; if any are held at the same
165 * time, the above lock order must be followed.
166 * * All output is delegated to the netisr.
167 * Now that Giant has been eliminated, the netisr may be inlined.
168 * * IN_MULTI_LOCK covers in_multi.
169 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
170 * including the output queue.
171 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
172 * per-link state iterators.
173 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
174 * therefore it is not refcounted.
175 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
176 *
177 * Reference counting
178 * * IGMP acquires its own reference every time an in_multi is passed to
179 * it and the group is being joined for the first time.
180 * * IGMP releases its reference(s) on in_multi in a deferred way,
181 * because the operations which process the release run as part of
182 * a loop whose control variables are directly affected by the release
183 * (that, and not recursing on the IF_ADDR_LOCK).
184 *
185 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
186 * to a vnet in ifp->if_vnet.
187 *
188 * SMPng: XXX We may potentially race operations on ifma_protospec.
189 * The problem is that we currently lack a clean way of taking the
190 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
191 * as anything which modifies ifma needs to be covered by that lock.
192 * So check for ifma_protospec being NULL before proceeding.
193 */
194 struct mtx igmp_mtx;
195
196 struct mbuf *m_raopt; /* Router Alert option */
197 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
198
199 /*
200 * VIMAGE-wide globals.
201 *
202 * The IGMPv3 timers themselves need to run per-image, however,
203 * protosw timers run globally (see tcp).
204 * An ifnet can only be in one vimage at a time, and the loopback
205 * ifnet, loif, is itself virtualized.
206 * It would otherwise be possible to seriously hose IGMP state,
207 * and create inconsistencies in upstream multicast routing, if you have
208 * multiple VIMAGEs running on the same link joining different multicast
209 * groups, UNLESS the "primary IP address" is different. This is because
210 * IGMP for IPv4 does not force link-local addresses to be used for each
211 * node, unlike MLD for IPv6.
212 * Obviously the IGMPv3 per-interface state has per-vimage granularity
213 * also as a result.
214 *
215 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
216 * policy to control the address used by IGMP on the link.
217 */
218 static VNET_DEFINE(int, interface_timers_running); /* IGMPv3 general
219 * query response */
220 static VNET_DEFINE(int, state_change_timers_running); /* IGMPv3 state-change
221 * retransmit */
222 static VNET_DEFINE(int, current_state_timers_running); /* IGMPv1/v2 host
223 * report; IGMPv3 g/sg
224 * query response */
225
226 #define V_interface_timers_running VNET(interface_timers_running)
227 #define V_state_change_timers_running VNET(state_change_timers_running)
228 #define V_current_state_timers_running VNET(current_state_timers_running)
229
230 static VNET_DEFINE(LIST_HEAD(, igmp_ifsoftc), igi_head) =
231 LIST_HEAD_INITIALIZER(igi_head);
232 static VNET_DEFINE(struct igmpstat, igmpstat) = {
233 .igps_version = IGPS_VERSION_3,
234 .igps_len = sizeof(struct igmpstat),
235 };
236 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
237
238 #define V_igi_head VNET(igi_head)
239 #define V_igmpstat VNET(igmpstat)
240 #define V_igmp_gsrdelay VNET(igmp_gsrdelay)
241
242 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
243 static VNET_DEFINE(int, igmp_sendra) = 1;
244 static VNET_DEFINE(int, igmp_sendlocal) = 1;
245 static VNET_DEFINE(int, igmp_v1enable) = 1;
246 static VNET_DEFINE(int, igmp_v2enable) = 1;
247 static VNET_DEFINE(int, igmp_legacysupp);
248 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
249
250 #define V_igmp_recvifkludge VNET(igmp_recvifkludge)
251 #define V_igmp_sendra VNET(igmp_sendra)
252 #define V_igmp_sendlocal VNET(igmp_sendlocal)
253 #define V_igmp_v1enable VNET(igmp_v1enable)
254 #define V_igmp_v2enable VNET(igmp_v2enable)
255 #define V_igmp_legacysupp VNET(igmp_legacysupp)
256 #define V_igmp_default_version VNET(igmp_default_version)
257
258 /*
259 * Virtualized sysctls.
260 */
261 SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW,
262 &VNET_NAME(igmpstat), igmpstat, "");
263 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
264 &VNET_NAME(igmp_recvifkludge), 0,
265 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
266 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
267 &VNET_NAME(igmp_sendra), 0,
268 "Send IP Router Alert option in IGMPv2/v3 messages");
269 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
270 &VNET_NAME(igmp_sendlocal), 0,
271 "Send IGMP membership reports for 224.0.0.0/24 groups");
272 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
273 &VNET_NAME(igmp_v1enable), 0,
274 "Enable backwards compatibility with IGMPv1");
275 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
276 &VNET_NAME(igmp_v2enable), 0,
277 "Enable backwards compatibility with IGMPv2");
278 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
279 &VNET_NAME(igmp_legacysupp), 0,
280 "Allow v1/v2 reports to suppress v3 group responses");
281 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
282 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
283 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
284 "Default version of IGMP to run on each interface");
285 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
286 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
287 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
288 "Rate limit for IGMPv3 Group-and-Source queries in seconds");
289
290 /*
291 * Non-virtualized sysctls.
292 */
293 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
294 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
295 "Per-interface IGMPv3 state");
296
297 static __inline void
298 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
299 {
300
301 #ifdef VIMAGE
302 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
303 #endif /* VIMAGE */
304 m->m_pkthdr.rcvif = ifp;
305 m->m_pkthdr.flowid = ifp->if_index;
306 }
307
308 static __inline void
309 igmp_scrub_context(struct mbuf *m)
310 {
311
312 m->m_pkthdr.PH_loc.ptr = NULL;
313 m->m_pkthdr.flowid = 0;
314 }
315
316 /*
317 * Restore context from a queued IGMP output chain.
318 * Return saved ifindex.
319 *
320 * VIMAGE: The assertion is there to make sure that we
321 * actually called CURVNET_SET() with what's in the mbuf chain.
322 */
323 static __inline uint32_t
324 igmp_restore_context(struct mbuf *m)
325 {
326
327 #ifdef notyet
328 #if defined(VIMAGE) && defined(INVARIANTS)
329 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
330 ("%s: called when curvnet was not restored", __func__));
331 #endif
332 #endif
333 return (m->m_pkthdr.flowid);
334 }
335
336 /*
337 * Retrieve or set default IGMP version.
338 *
339 * VIMAGE: Assume curvnet set by caller.
340 * SMPng: NOTE: Serialized by IGMP lock.
341 */
342 static int
343 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
344 {
345 int error;
346 int new;
347
348 error = sysctl_wire_old_buffer(req, sizeof(int));
349 if (error)
350 return (error);
351
352 IGMP_LOCK();
353
354 new = V_igmp_default_version;
355
356 error = sysctl_handle_int(oidp, &new, 0, req);
357 if (error || !req->newptr)
358 goto out_locked;
359
360 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
361 error = EINVAL;
362 goto out_locked;
363 }
364
365 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
366 V_igmp_default_version, new);
367
368 V_igmp_default_version = new;
369
370 out_locked:
371 IGMP_UNLOCK();
372 return (error);
373 }
374
375 /*
376 * Retrieve or set threshold between group-source queries in seconds.
377 *
378 * VIMAGE: Assume curvnet set by caller.
379 * SMPng: NOTE: Serialized by IGMP lock.
380 */
381 static int
382 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
383 {
384 int error;
385 int i;
386
387 error = sysctl_wire_old_buffer(req, sizeof(int));
388 if (error)
389 return (error);
390
391 IGMP_LOCK();
392
393 i = V_igmp_gsrdelay.tv_sec;
394
395 error = sysctl_handle_int(oidp, &i, 0, req);
396 if (error || !req->newptr)
397 goto out_locked;
398
399 if (i < -1 || i >= 60) {
400 error = EINVAL;
401 goto out_locked;
402 }
403
404 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
405 V_igmp_gsrdelay.tv_sec, i);
406 V_igmp_gsrdelay.tv_sec = i;
407
408 out_locked:
409 IGMP_UNLOCK();
410 return (error);
411 }
412
413 /*
414 * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
415 * For use by ifmcstat(8).
416 *
417 * SMPng: NOTE: Does an unlocked ifindex space read.
418 * VIMAGE: Assume curvnet set by caller. The node handler itself
419 * is not directly virtualized.
420 */
421 static int
422 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
423 {
424 int *name;
425 int error;
426 u_int namelen;
427 struct ifnet *ifp;
428 struct igmp_ifsoftc *igi;
429
430 name = (int *)arg1;
431 namelen = arg2;
432
433 if (req->newptr != NULL)
434 return (EPERM);
435
436 if (namelen != 1)
437 return (EINVAL);
438
439 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
440 if (error)
441 return (error);
442
443 IN_MULTI_LOCK();
444 IGMP_LOCK();
445
446 if (name[0] <= 0 || name[0] > V_if_index) {
447 error = ENOENT;
448 goto out_locked;
449 }
450
451 error = ENOENT;
452
453 ifp = ifnet_byindex(name[0]);
454 if (ifp == NULL)
455 goto out_locked;
456
457 LIST_FOREACH(igi, &V_igi_head, igi_link) {
458 if (ifp == igi->igi_ifp) {
459 struct igmp_ifinfo info;
460
461 info.igi_version = igi->igi_version;
462 info.igi_v1_timer = igi->igi_v1_timer;
463 info.igi_v2_timer = igi->igi_v2_timer;
464 info.igi_v3_timer = igi->igi_v3_timer;
465 info.igi_flags = igi->igi_flags;
466 info.igi_rv = igi->igi_rv;
467 info.igi_qi = igi->igi_qi;
468 info.igi_qri = igi->igi_qri;
469 info.igi_uri = igi->igi_uri;
470 error = SYSCTL_OUT(req, &info, sizeof(info));
471 break;
472 }
473 }
474
475 out_locked:
476 IGMP_UNLOCK();
477 IN_MULTI_UNLOCK();
478 return (error);
479 }
480
481 /*
482 * Dispatch an entire queue of pending packet chains
483 * using the netisr.
484 * VIMAGE: Assumes the vnet pointer has been set.
485 */
486 static void
487 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
488 {
489 struct mbuf *m;
490
491 while ((m = mbufq_dequeue(mq)) != NULL) {
492 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
493 if (loop)
494 m->m_flags |= M_IGMP_LOOP;
495 netisr_dispatch(NETISR_IGMP, m);
496 if (--limit == 0)
497 break;
498 }
499 }
500
501 /*
502 * Filter outgoing IGMP report state by group.
503 *
504 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
505 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
506 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
507 * this may break certain IGMP snooping switches which rely on the old
508 * report behaviour.
509 *
510 * Return zero if the given group is one for which IGMP reports
511 * should be suppressed, or non-zero if reports should be issued.
512 */
513 static __inline int
514 igmp_isgroupreported(const struct in_addr addr)
515 {
516
517 if (in_allhosts(addr) ||
518 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
519 return (0);
520
521 return (1);
522 }
523
524 /*
525 * Construct a Router Alert option to use in outgoing packets.
526 */
527 static struct mbuf *
528 igmp_ra_alloc(void)
529 {
530 struct mbuf *m;
531 struct ipoption *p;
532
533 m = m_get(M_WAITOK, MT_DATA);
534 p = mtod(m, struct ipoption *);
535 p->ipopt_dst.s_addr = INADDR_ANY;
536 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */
537 p->ipopt_list[1] = 0x04; /* 4 bytes long */
538 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
539 p->ipopt_list[3] = 0x00; /* pad byte */
540 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
541
542 return (m);
543 }
544
545 /*
546 * Attach IGMP when PF_INET is attached to an interface.
547 */
548 struct igmp_ifsoftc *
549 igmp_domifattach(struct ifnet *ifp)
550 {
551 struct igmp_ifsoftc *igi;
552
553 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
554 __func__, ifp, ifp->if_xname);
555
556 IGMP_LOCK();
557
558 igi = igi_alloc_locked(ifp);
559 if (!(ifp->if_flags & IFF_MULTICAST))
560 igi->igi_flags |= IGIF_SILENT;
561
562 IGMP_UNLOCK();
563
564 return (igi);
565 }
566
567 /*
568 * VIMAGE: assume curvnet set by caller.
569 */
570 static struct igmp_ifsoftc *
571 igi_alloc_locked(/*const*/ struct ifnet *ifp)
572 {
573 struct igmp_ifsoftc *igi;
574
575 IGMP_LOCK_ASSERT();
576
577 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
578 if (igi == NULL)
579 goto out;
580
581 igi->igi_ifp = ifp;
582 igi->igi_version = V_igmp_default_version;
583 igi->igi_flags = 0;
584 igi->igi_rv = IGMP_RV_INIT;
585 igi->igi_qi = IGMP_QI_INIT;
586 igi->igi_qri = IGMP_QRI_INIT;
587 igi->igi_uri = IGMP_URI_INIT;
588 SLIST_INIT(&igi->igi_relinmhead);
589 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
590
591 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
592
593 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
594 ifp, ifp->if_xname);
595
596 out:
597 return (igi);
598 }
599
600 /*
601 * Hook for ifdetach.
602 *
603 * NOTE: Some finalization tasks need to run before the protocol domain
604 * is detached, but also before the link layer does its cleanup.
605 *
606 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
607 * XXX This is also bitten by unlocked ifma_protospec access.
608 */
609 void
610 igmp_ifdetach(struct ifnet *ifp)
611 {
612 struct igmp_ifsoftc *igi;
613 struct ifmultiaddr *ifma;
614 struct in_multi *inm, *tinm;
615
616 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
617 ifp->if_xname);
618
619 IGMP_LOCK();
620
621 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
622 if (igi->igi_version == IGMP_VERSION_3) {
623 IF_ADDR_RLOCK(ifp);
624 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
625 if (ifma->ifma_addr->sa_family != AF_INET ||
626 ifma->ifma_protospec == NULL)
627 continue;
628 #if 0
629 KASSERT(ifma->ifma_protospec != NULL,
630 ("%s: ifma_protospec is NULL", __func__));
631 #endif
632 inm = (struct in_multi *)ifma->ifma_protospec;
633 if (inm->inm_state == IGMP_LEAVING_MEMBER) {
634 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
635 inm, inm_nrele);
636 }
637 inm_clear_recorded(inm);
638 }
639 IF_ADDR_RUNLOCK(ifp);
640 /*
641 * Free the in_multi reference(s) for this IGMP lifecycle.
642 */
643 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
644 tinm) {
645 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
646 inm_release_locked(inm);
647 }
648 }
649
650 IGMP_UNLOCK();
651 }
652
653 /*
654 * Hook for domifdetach.
655 */
656 void
657 igmp_domifdetach(struct ifnet *ifp)
658 {
659
660 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
661 __func__, ifp, ifp->if_xname);
662
663 IGMP_LOCK();
664 igi_delete_locked(ifp);
665 IGMP_UNLOCK();
666 }
667
668 static void
669 igi_delete_locked(const struct ifnet *ifp)
670 {
671 struct igmp_ifsoftc *igi, *tigi;
672
673 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
674 __func__, ifp, ifp->if_xname);
675
676 IGMP_LOCK_ASSERT();
677
678 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
679 if (igi->igi_ifp == ifp) {
680 /*
681 * Free deferred General Query responses.
682 */
683 mbufq_drain(&igi->igi_gq);
684
685 LIST_REMOVE(igi, igi_link);
686
687 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
688 ("%s: there are dangling in_multi references",
689 __func__));
690
691 free(igi, M_IGMP);
692 return;
693 }
694 }
695 }
696
697 /*
698 * Process a received IGMPv1 query.
699 * Return non-zero if the message should be dropped.
700 *
701 * VIMAGE: The curvnet pointer is derived from the input ifp.
702 */
703 static int
704 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
705 const struct igmp *igmp)
706 {
707 struct ifmultiaddr *ifma;
708 struct igmp_ifsoftc *igi;
709 struct in_multi *inm;
710
711 /*
712 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
713 * 224.0.0.1. They are always treated as General Queries.
714 * igmp_group is always ignored. Do not drop it as a userland
715 * daemon may wish to see it.
716 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
717 */
718 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
719 IGMPSTAT_INC(igps_rcv_badqueries);
720 return (0);
721 }
722 IGMPSTAT_INC(igps_rcv_gen_queries);
723
724 IN_MULTI_LOCK();
725 IGMP_LOCK();
726
727 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
728 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
729
730 if (igi->igi_flags & IGIF_LOOPBACK) {
731 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
732 ifp, ifp->if_xname);
733 goto out_locked;
734 }
735
736 /*
737 * Switch to IGMPv1 host compatibility mode.
738 */
739 igmp_set_version(igi, IGMP_VERSION_1);
740
741 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
742
743 /*
744 * Start the timers in all of our group records
745 * for the interface on which the query arrived,
746 * except those which are already running.
747 */
748 IF_ADDR_RLOCK(ifp);
749 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
750 if (ifma->ifma_addr->sa_family != AF_INET ||
751 ifma->ifma_protospec == NULL)
752 continue;
753 inm = (struct in_multi *)ifma->ifma_protospec;
754 if (inm->inm_timer != 0)
755 continue;
756 switch (inm->inm_state) {
757 case IGMP_NOT_MEMBER:
758 case IGMP_SILENT_MEMBER:
759 break;
760 case IGMP_G_QUERY_PENDING_MEMBER:
761 case IGMP_SG_QUERY_PENDING_MEMBER:
762 case IGMP_REPORTING_MEMBER:
763 case IGMP_IDLE_MEMBER:
764 case IGMP_LAZY_MEMBER:
765 case IGMP_SLEEPING_MEMBER:
766 case IGMP_AWAKENING_MEMBER:
767 inm->inm_state = IGMP_REPORTING_MEMBER;
768 inm->inm_timer = IGMP_RANDOM_DELAY(
769 IGMP_V1V2_MAX_RI * PR_FASTHZ);
770 V_current_state_timers_running = 1;
771 break;
772 case IGMP_LEAVING_MEMBER:
773 break;
774 }
775 }
776 IF_ADDR_RUNLOCK(ifp);
777
778 out_locked:
779 IGMP_UNLOCK();
780 IN_MULTI_UNLOCK();
781
782 return (0);
783 }
784
785 /*
786 * Process a received IGMPv2 general or group-specific query.
787 */
788 static int
789 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
790 const struct igmp *igmp)
791 {
792 struct ifmultiaddr *ifma;
793 struct igmp_ifsoftc *igi;
794 struct in_multi *inm;
795 int is_general_query;
796 uint16_t timer;
797
798 is_general_query = 0;
799
800 /*
801 * Validate address fields upfront.
802 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
803 */
804 if (in_nullhost(igmp->igmp_group)) {
805 /*
806 * IGMPv2 General Query.
807 * If this was not sent to the all-hosts group, ignore it.
808 */
809 if (!in_allhosts(ip->ip_dst))
810 return (0);
811 IGMPSTAT_INC(igps_rcv_gen_queries);
812 is_general_query = 1;
813 } else {
814 /* IGMPv2 Group-Specific Query. */
815 IGMPSTAT_INC(igps_rcv_group_queries);
816 }
817
818 IN_MULTI_LOCK();
819 IGMP_LOCK();
820
821 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
822 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
823
824 if (igi->igi_flags & IGIF_LOOPBACK) {
825 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
826 ifp, ifp->if_xname);
827 goto out_locked;
828 }
829
830 /*
831 * Ignore v2 query if in v1 Compatibility Mode.
832 */
833 if (igi->igi_version == IGMP_VERSION_1)
834 goto out_locked;
835
836 igmp_set_version(igi, IGMP_VERSION_2);
837
838 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
839 if (timer == 0)
840 timer = 1;
841
842 if (is_general_query) {
843 /*
844 * For each reporting group joined on this
845 * interface, kick the report timer.
846 */
847 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
848 ifp, ifp->if_xname);
849 IF_ADDR_RLOCK(ifp);
850 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
851 if (ifma->ifma_addr->sa_family != AF_INET ||
852 ifma->ifma_protospec == NULL)
853 continue;
854 inm = (struct in_multi *)ifma->ifma_protospec;
855 igmp_v2_update_group(inm, timer);
856 }
857 IF_ADDR_RUNLOCK(ifp);
858 } else {
859 /*
860 * Group-specific IGMPv2 query, we need only
861 * look up the single group to process it.
862 */
863 inm = inm_lookup(ifp, igmp->igmp_group);
864 if (inm != NULL) {
865 CTR3(KTR_IGMPV3,
866 "process v2 query 0x%08x on ifp %p(%s)",
867 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
868 igmp_v2_update_group(inm, timer);
869 }
870 }
871
872 out_locked:
873 IGMP_UNLOCK();
874 IN_MULTI_UNLOCK();
875
876 return (0);
877 }
878
879 /*
880 * Update the report timer on a group in response to an IGMPv2 query.
881 *
882 * If we are becoming the reporting member for this group, start the timer.
883 * If we already are the reporting member for this group, and timer is
884 * below the threshold, reset it.
885 *
886 * We may be updating the group for the first time since we switched
887 * to IGMPv3. If we are, then we must clear any recorded source lists,
888 * and transition to REPORTING state; the group timer is overloaded
889 * for group and group-source query responses.
890 *
891 * Unlike IGMPv3, the delay per group should be jittered
892 * to avoid bursts of IGMPv2 reports.
893 */
894 static void
895 igmp_v2_update_group(struct in_multi *inm, const int timer)
896 {
897
898 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
899 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
900
901 IN_MULTI_LOCK_ASSERT();
902
903 switch (inm->inm_state) {
904 case IGMP_NOT_MEMBER:
905 case IGMP_SILENT_MEMBER:
906 break;
907 case IGMP_REPORTING_MEMBER:
908 if (inm->inm_timer != 0 &&
909 inm->inm_timer <= timer) {
910 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
911 "skipping.", __func__);
912 break;
913 }
914 /* FALLTHROUGH */
915 case IGMP_SG_QUERY_PENDING_MEMBER:
916 case IGMP_G_QUERY_PENDING_MEMBER:
917 case IGMP_IDLE_MEMBER:
918 case IGMP_LAZY_MEMBER:
919 case IGMP_AWAKENING_MEMBER:
920 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
921 inm->inm_state = IGMP_REPORTING_MEMBER;
922 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
923 V_current_state_timers_running = 1;
924 break;
925 case IGMP_SLEEPING_MEMBER:
926 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
927 inm->inm_state = IGMP_AWAKENING_MEMBER;
928 break;
929 case IGMP_LEAVING_MEMBER:
930 break;
931 }
932 }
933
934 /*
935 * Process a received IGMPv3 general, group-specific or
936 * group-and-source-specific query.
937 * Assumes m has already been pulled up to the full IGMP message length.
938 * Return 0 if successful, otherwise an appropriate error code is returned.
939 */
940 static int
941 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
942 /*const*/ struct igmpv3 *igmpv3)
943 {
944 struct igmp_ifsoftc *igi;
945 struct in_multi *inm;
946 int is_general_query;
947 uint32_t maxresp, nsrc, qqi;
948 uint16_t timer;
949 uint8_t qrv;
950
951 is_general_query = 0;
952
953 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
954
955 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
956 if (maxresp >= 128) {
957 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
958 (IGMP_EXP(igmpv3->igmp_code) + 3);
959 }
960
961 /*
962 * Robustness must never be less than 2 for on-wire IGMPv3.
963 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
964 * an exception for interfaces whose IGMPv3 state changes
965 * are redirected to loopback (e.g. MANET).
966 */
967 qrv = IGMP_QRV(igmpv3->igmp_misc);
968 if (qrv < 2) {
969 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
970 qrv, IGMP_RV_INIT);
971 qrv = IGMP_RV_INIT;
972 }
973
974 qqi = igmpv3->igmp_qqi;
975 if (qqi >= 128) {
976 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
977 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
978 }
979
980 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
981 if (timer == 0)
982 timer = 1;
983
984 nsrc = ntohs(igmpv3->igmp_numsrc);
985
986 /*
987 * Validate address fields and versions upfront before
988 * accepting v3 query.
989 * XXX SMPng: Unlocked access to igmpstat counters here.
990 */
991 if (in_nullhost(igmpv3->igmp_group)) {
992 /*
993 * IGMPv3 General Query.
994 *
995 * General Queries SHOULD be directed to 224.0.0.1.
996 * A general query with a source list has undefined
997 * behaviour; discard it.
998 */
999 IGMPSTAT_INC(igps_rcv_gen_queries);
1000 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1001 IGMPSTAT_INC(igps_rcv_badqueries);
1002 return (0);
1003 }
1004 is_general_query = 1;
1005 } else {
1006 /* Group or group-source specific query. */
1007 if (nsrc == 0)
1008 IGMPSTAT_INC(igps_rcv_group_queries);
1009 else
1010 IGMPSTAT_INC(igps_rcv_gsr_queries);
1011 }
1012
1013 IN_MULTI_LOCK();
1014 IGMP_LOCK();
1015
1016 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1017 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1018
1019 if (igi->igi_flags & IGIF_LOOPBACK) {
1020 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1021 ifp, ifp->if_xname);
1022 goto out_locked;
1023 }
1024
1025 /*
1026 * Discard the v3 query if we're in Compatibility Mode.
1027 * The RFC is not obviously worded that hosts need to stay in
1028 * compatibility mode until the Old Version Querier Present
1029 * timer expires.
1030 */
1031 if (igi->igi_version != IGMP_VERSION_3) {
1032 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1033 igi->igi_version, ifp, ifp->if_xname);
1034 goto out_locked;
1035 }
1036
1037 igmp_set_version(igi, IGMP_VERSION_3);
1038 igi->igi_rv = qrv;
1039 igi->igi_qi = qqi;
1040 igi->igi_qri = maxresp;
1041
1042 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1043 maxresp);
1044
1045 if (is_general_query) {
1046 /*
1047 * Schedule a current-state report on this ifp for
1048 * all groups, possibly containing source lists.
1049 * If there is a pending General Query response
1050 * scheduled earlier than the selected delay, do
1051 * not schedule any other reports.
1052 * Otherwise, reset the interface timer.
1053 */
1054 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1055 ifp, ifp->if_xname);
1056 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1057 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1058 V_interface_timers_running = 1;
1059 }
1060 } else {
1061 /*
1062 * Group-source-specific queries are throttled on
1063 * a per-group basis to defeat denial-of-service attempts.
1064 * Queries for groups we are not a member of on this
1065 * link are simply ignored.
1066 */
1067 inm = inm_lookup(ifp, igmpv3->igmp_group);
1068 if (inm == NULL)
1069 goto out_locked;
1070 if (nsrc > 0) {
1071 if (!ratecheck(&inm->inm_lastgsrtv,
1072 &V_igmp_gsrdelay)) {
1073 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1074 __func__);
1075 IGMPSTAT_INC(igps_drop_gsr_queries);
1076 goto out_locked;
1077 }
1078 }
1079 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1080 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1081 /*
1082 * If there is a pending General Query response
1083 * scheduled sooner than the selected delay, no
1084 * further report need be scheduled.
1085 * Otherwise, prepare to respond to the
1086 * group-specific or group-and-source query.
1087 */
1088 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1089 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1090 }
1091
1092 out_locked:
1093 IGMP_UNLOCK();
1094 IN_MULTI_UNLOCK();
1095
1096 return (0);
1097 }
1098
1099 /*
1100 * Process a received IGMPv3 group-specific or group-and-source-specific
1101 * query.
1102 * Return <0 if any error occurred. Currently this is ignored.
1103 */
1104 static int
1105 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1106 int timer, /*const*/ struct igmpv3 *igmpv3)
1107 {
1108 int retval;
1109 uint16_t nsrc;
1110
1111 IN_MULTI_LOCK_ASSERT();
1112 IGMP_LOCK_ASSERT();
1113
1114 retval = 0;
1115
1116 switch (inm->inm_state) {
1117 case IGMP_NOT_MEMBER:
1118 case IGMP_SILENT_MEMBER:
1119 case IGMP_SLEEPING_MEMBER:
1120 case IGMP_LAZY_MEMBER:
1121 case IGMP_AWAKENING_MEMBER:
1122 case IGMP_IDLE_MEMBER:
1123 case IGMP_LEAVING_MEMBER:
1124 return (retval);
1125 break;
1126 case IGMP_REPORTING_MEMBER:
1127 case IGMP_G_QUERY_PENDING_MEMBER:
1128 case IGMP_SG_QUERY_PENDING_MEMBER:
1129 break;
1130 }
1131
1132 nsrc = ntohs(igmpv3->igmp_numsrc);
1133
1134 /*
1135 * Deal with group-specific queries upfront.
1136 * If any group query is already pending, purge any recorded
1137 * source-list state if it exists, and schedule a query response
1138 * for this group-specific query.
1139 */
1140 if (nsrc == 0) {
1141 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1142 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1143 inm_clear_recorded(inm);
1144 timer = min(inm->inm_timer, timer);
1145 }
1146 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1147 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1148 V_current_state_timers_running = 1;
1149 return (retval);
1150 }
1151
1152 /*
1153 * Deal with the case where a group-and-source-specific query has
1154 * been received but a group-specific query is already pending.
1155 */
1156 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1157 timer = min(inm->inm_timer, timer);
1158 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1159 V_current_state_timers_running = 1;
1160 return (retval);
1161 }
1162
1163 /*
1164 * Finally, deal with the case where a group-and-source-specific
1165 * query has been received, where a response to a previous g-s-r
1166 * query exists, or none exists.
1167 * In this case, we need to parse the source-list which the Querier
1168 * has provided us with and check if we have any source list filter
1169 * entries at T1 for these sources. If we do not, there is no need
1170 * schedule a report and the query may be dropped.
1171 * If we do, we must record them and schedule a current-state
1172 * report for those sources.
1173 * FIXME: Handling source lists larger than 1 mbuf requires that
1174 * we pass the mbuf chain pointer down to this function, and use
1175 * m_getptr() to walk the chain.
1176 */
1177 if (inm->inm_nsrc > 0) {
1178 const struct in_addr *ap;
1179 int i, nrecorded;
1180
1181 ap = (const struct in_addr *)(igmpv3 + 1);
1182 nrecorded = 0;
1183 for (i = 0; i < nsrc; i++, ap++) {
1184 retval = inm_record_source(inm, ap->s_addr);
1185 if (retval < 0)
1186 break;
1187 nrecorded += retval;
1188 }
1189 if (nrecorded > 0) {
1190 CTR1(KTR_IGMPV3,
1191 "%s: schedule response to SG query", __func__);
1192 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1193 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1194 V_current_state_timers_running = 1;
1195 }
1196 }
1197
1198 return (retval);
1199 }
1200
1201 /*
1202 * Process a received IGMPv1 host membership report.
1203 *
1204 * NOTE: 0.0.0.0 workaround breaks const correctness.
1205 */
1206 static int
1207 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1208 /*const*/ struct igmp *igmp)
1209 {
1210 struct rm_priotracker in_ifa_tracker;
1211 struct in_ifaddr *ia;
1212 struct in_multi *inm;
1213
1214 IGMPSTAT_INC(igps_rcv_reports);
1215
1216 if (ifp->if_flags & IFF_LOOPBACK)
1217 return (0);
1218
1219 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1220 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1221 IGMPSTAT_INC(igps_rcv_badreports);
1222 return (EINVAL);
1223 }
1224
1225 /*
1226 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1227 * Booting clients may use the source address 0.0.0.0. Some
1228 * IGMP daemons may not know how to use IP_RECVIF to determine
1229 * the interface upon which this message was received.
1230 * Replace 0.0.0.0 with the subnet address if told to do so.
1231 */
1232 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1233 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1234 if (ia != NULL) {
1235 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1236 ifa_free(&ia->ia_ifa);
1237 }
1238 }
1239
1240 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1241 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1242
1243 /*
1244 * IGMPv1 report suppression.
1245 * If we are a member of this group, and our membership should be
1246 * reported, stop our group timer and transition to the 'lazy' state.
1247 */
1248 IN_MULTI_LOCK();
1249 inm = inm_lookup(ifp, igmp->igmp_group);
1250 if (inm != NULL) {
1251 struct igmp_ifsoftc *igi;
1252
1253 igi = inm->inm_igi;
1254 if (igi == NULL) {
1255 KASSERT(igi != NULL,
1256 ("%s: no igi for ifp %p", __func__, ifp));
1257 goto out_locked;
1258 }
1259
1260 IGMPSTAT_INC(igps_rcv_ourreports);
1261
1262 /*
1263 * If we are in IGMPv3 host mode, do not allow the
1264 * other host's IGMPv1 report to suppress our reports
1265 * unless explicitly configured to do so.
1266 */
1267 if (igi->igi_version == IGMP_VERSION_3) {
1268 if (V_igmp_legacysupp)
1269 igmp_v3_suppress_group_record(inm);
1270 goto out_locked;
1271 }
1272
1273 inm->inm_timer = 0;
1274
1275 switch (inm->inm_state) {
1276 case IGMP_NOT_MEMBER:
1277 case IGMP_SILENT_MEMBER:
1278 break;
1279 case IGMP_IDLE_MEMBER:
1280 case IGMP_LAZY_MEMBER:
1281 case IGMP_AWAKENING_MEMBER:
1282 CTR3(KTR_IGMPV3,
1283 "report suppressed for 0x%08x on ifp %p(%s)",
1284 ntohl(igmp->igmp_group.s_addr), ifp,
1285 ifp->if_xname);
1286 case IGMP_SLEEPING_MEMBER:
1287 inm->inm_state = IGMP_SLEEPING_MEMBER;
1288 break;
1289 case IGMP_REPORTING_MEMBER:
1290 CTR3(KTR_IGMPV3,
1291 "report suppressed for 0x%08x on ifp %p(%s)",
1292 ntohl(igmp->igmp_group.s_addr), ifp,
1293 ifp->if_xname);
1294 if (igi->igi_version == IGMP_VERSION_1)
1295 inm->inm_state = IGMP_LAZY_MEMBER;
1296 else if (igi->igi_version == IGMP_VERSION_2)
1297 inm->inm_state = IGMP_SLEEPING_MEMBER;
1298 break;
1299 case IGMP_G_QUERY_PENDING_MEMBER:
1300 case IGMP_SG_QUERY_PENDING_MEMBER:
1301 case IGMP_LEAVING_MEMBER:
1302 break;
1303 }
1304 }
1305
1306 out_locked:
1307 IN_MULTI_UNLOCK();
1308
1309 return (0);
1310 }
1311
1312 /*
1313 * Process a received IGMPv2 host membership report.
1314 *
1315 * NOTE: 0.0.0.0 workaround breaks const correctness.
1316 */
1317 static int
1318 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1319 /*const*/ struct igmp *igmp)
1320 {
1321 struct rm_priotracker in_ifa_tracker;
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, &in_ifa_tracker);
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 0x%08x on ifp %p(%s)",
1367 ntohl(igmp->igmp_group.s_addr), 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_ifsoftc *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 0x%08x on ifp %p(%s)",
1408 ntohl(igmp->igmp_group.s_addr), 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 int
1426 igmp_input(struct mbuf **mp, int *offp, int proto)
1427 {
1428 int iphlen;
1429 struct ifnet *ifp;
1430 struct igmp *igmp;
1431 struct ip *ip;
1432 struct mbuf *m;
1433 int igmplen;
1434 int minlen;
1435 int queryver;
1436
1437 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1438
1439 m = *mp;
1440 ifp = m->m_pkthdr.rcvif;
1441 *mp = NULL;
1442
1443 IGMPSTAT_INC(igps_rcv_total);
1444
1445 ip = mtod(m, struct ip *);
1446 iphlen = *offp;
1447 igmplen = ntohs(ip->ip_len) - iphlen;
1448
1449 /*
1450 * Validate lengths.
1451 */
1452 if (igmplen < IGMP_MINLEN) {
1453 IGMPSTAT_INC(igps_rcv_tooshort);
1454 m_freem(m);
1455 return (IPPROTO_DONE);
1456 }
1457
1458 /*
1459 * Always pullup to the minimum size for v1/v2 or v3
1460 * to amortize calls to m_pullup().
1461 */
1462 minlen = iphlen;
1463 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1464 minlen += IGMP_V3_QUERY_MINLEN;
1465 else
1466 minlen += IGMP_MINLEN;
1467 if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1468 (m = m_pullup(m, minlen)) == NULL) {
1469 IGMPSTAT_INC(igps_rcv_tooshort);
1470 return (IPPROTO_DONE);
1471 }
1472 ip = mtod(m, struct ip *);
1473
1474 /*
1475 * Validate checksum.
1476 */
1477 m->m_data += iphlen;
1478 m->m_len -= iphlen;
1479 igmp = mtod(m, struct igmp *);
1480 if (in_cksum(m, igmplen)) {
1481 IGMPSTAT_INC(igps_rcv_badsum);
1482 m_freem(m);
1483 return (IPPROTO_DONE);
1484 }
1485 m->m_data -= iphlen;
1486 m->m_len += iphlen;
1487
1488 /*
1489 * IGMP control traffic is link-scope, and must have a TTL of 1.
1490 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1491 * probe packets may come from beyond the LAN.
1492 */
1493 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1494 IGMPSTAT_INC(igps_rcv_badttl);
1495 m_freem(m);
1496 return (IPPROTO_DONE);
1497 }
1498
1499 switch (igmp->igmp_type) {
1500 case IGMP_HOST_MEMBERSHIP_QUERY:
1501 if (igmplen == IGMP_MINLEN) {
1502 if (igmp->igmp_code == 0)
1503 queryver = IGMP_VERSION_1;
1504 else
1505 queryver = IGMP_VERSION_2;
1506 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1507 queryver = IGMP_VERSION_3;
1508 } else {
1509 IGMPSTAT_INC(igps_rcv_tooshort);
1510 m_freem(m);
1511 return (IPPROTO_DONE);
1512 }
1513
1514 switch (queryver) {
1515 case IGMP_VERSION_1:
1516 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1517 if (!V_igmp_v1enable)
1518 break;
1519 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1520 m_freem(m);
1521 return (IPPROTO_DONE);
1522 }
1523 break;
1524
1525 case IGMP_VERSION_2:
1526 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1527 if (!V_igmp_v2enable)
1528 break;
1529 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1530 m_freem(m);
1531 return (IPPROTO_DONE);
1532 }
1533 break;
1534
1535 case IGMP_VERSION_3: {
1536 struct igmpv3 *igmpv3;
1537 uint16_t igmpv3len;
1538 uint16_t nsrc;
1539
1540 IGMPSTAT_INC(igps_rcv_v3_queries);
1541 igmpv3 = (struct igmpv3 *)igmp;
1542 /*
1543 * Validate length based on source count.
1544 */
1545 nsrc = ntohs(igmpv3->igmp_numsrc);
1546 if (nsrc * sizeof(in_addr_t) >
1547 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1548 IGMPSTAT_INC(igps_rcv_tooshort);
1549 return (IPPROTO_DONE);
1550 }
1551 /*
1552 * m_pullup() may modify m, so pullup in
1553 * this scope.
1554 */
1555 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1556 sizeof(struct in_addr) * nsrc;
1557 if ((!M_WRITABLE(m) ||
1558 m->m_len < igmpv3len) &&
1559 (m = m_pullup(m, igmpv3len)) == NULL) {
1560 IGMPSTAT_INC(igps_rcv_tooshort);
1561 return (IPPROTO_DONE);
1562 }
1563 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1564 + iphlen);
1565 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1566 m_freem(m);
1567 return (IPPROTO_DONE);
1568 }
1569 }
1570 break;
1571 }
1572 break;
1573
1574 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1575 if (!V_igmp_v1enable)
1576 break;
1577 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1578 m_freem(m);
1579 return (IPPROTO_DONE);
1580 }
1581 break;
1582
1583 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1584 if (!V_igmp_v2enable)
1585 break;
1586 if (!ip_checkrouteralert(m))
1587 IGMPSTAT_INC(igps_rcv_nora);
1588 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1589 m_freem(m);
1590 return (IPPROTO_DONE);
1591 }
1592 break;
1593
1594 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1595 /*
1596 * Hosts do not need to process IGMPv3 membership reports,
1597 * as report suppression is no longer required.
1598 */
1599 if (!ip_checkrouteralert(m))
1600 IGMPSTAT_INC(igps_rcv_nora);
1601 break;
1602
1603 default:
1604 break;
1605 }
1606
1607 /*
1608 * Pass all valid IGMP packets up to any process(es) listening on a
1609 * raw IGMP socket.
1610 */
1611 *mp = m;
1612 return (rip_input(mp, offp, proto));
1613 }
1614
1615
1616 /*
1617 * Fast timeout handler (global).
1618 * VIMAGE: Timeout handlers are expected to service all vimages.
1619 */
1620 void
1621 igmp_fasttimo(void)
1622 {
1623 VNET_ITERATOR_DECL(vnet_iter);
1624
1625 VNET_LIST_RLOCK_NOSLEEP();
1626 VNET_FOREACH(vnet_iter) {
1627 CURVNET_SET(vnet_iter);
1628 igmp_fasttimo_vnet();
1629 CURVNET_RESTORE();
1630 }
1631 VNET_LIST_RUNLOCK_NOSLEEP();
1632 }
1633
1634 /*
1635 * Fast timeout handler (per-vnet).
1636 * Sends are shuffled off to a netisr to deal with Giant.
1637 *
1638 * VIMAGE: Assume caller has set up our curvnet.
1639 */
1640 static void
1641 igmp_fasttimo_vnet(void)
1642 {
1643 struct mbufq scq; /* State-change packets */
1644 struct mbufq qrq; /* Query response packets */
1645 struct ifnet *ifp;
1646 struct igmp_ifsoftc *igi;
1647 struct ifmultiaddr *ifma;
1648 struct in_multi *inm;
1649 int loop, uri_fasthz;
1650
1651 loop = 0;
1652 uri_fasthz = 0;
1653
1654 /*
1655 * Quick check to see if any work needs to be done, in order to
1656 * minimize the overhead of fasttimo processing.
1657 * SMPng: XXX Unlocked reads.
1658 */
1659 if (!V_current_state_timers_running &&
1660 !V_interface_timers_running &&
1661 !V_state_change_timers_running)
1662 return;
1663
1664 IN_MULTI_LOCK();
1665 IGMP_LOCK();
1666
1667 /*
1668 * IGMPv3 General Query response timer processing.
1669 */
1670 if (V_interface_timers_running) {
1671 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1672
1673 V_interface_timers_running = 0;
1674 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1675 if (igi->igi_v3_timer == 0) {
1676 /* Do nothing. */
1677 } else if (--igi->igi_v3_timer == 0) {
1678 igmp_v3_dispatch_general_query(igi);
1679 } else {
1680 V_interface_timers_running = 1;
1681 }
1682 }
1683 }
1684
1685 if (!V_current_state_timers_running &&
1686 !V_state_change_timers_running)
1687 goto out_locked;
1688
1689 V_current_state_timers_running = 0;
1690 V_state_change_timers_running = 0;
1691
1692 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1693
1694 /*
1695 * IGMPv1/v2/v3 host report and state-change timer processing.
1696 * Note: Processing a v3 group timer may remove a node.
1697 */
1698 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1699 ifp = igi->igi_ifp;
1700
1701 if (igi->igi_version == IGMP_VERSION_3) {
1702 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1703 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1704 PR_FASTHZ);
1705 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1706 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1707 }
1708
1709 IF_ADDR_RLOCK(ifp);
1710 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1711 if (ifma->ifma_addr->sa_family != AF_INET ||
1712 ifma->ifma_protospec == NULL)
1713 continue;
1714 inm = (struct in_multi *)ifma->ifma_protospec;
1715 switch (igi->igi_version) {
1716 case IGMP_VERSION_1:
1717 case IGMP_VERSION_2:
1718 igmp_v1v2_process_group_timer(inm,
1719 igi->igi_version);
1720 break;
1721 case IGMP_VERSION_3:
1722 igmp_v3_process_group_timers(igi, &qrq,
1723 &scq, inm, uri_fasthz);
1724 break;
1725 }
1726 }
1727 IF_ADDR_RUNLOCK(ifp);
1728
1729 if (igi->igi_version == IGMP_VERSION_3) {
1730 struct in_multi *tinm;
1731
1732 igmp_dispatch_queue(&qrq, 0, loop);
1733 igmp_dispatch_queue(&scq, 0, loop);
1734
1735 /*
1736 * Free the in_multi reference(s) for this
1737 * IGMP lifecycle.
1738 */
1739 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1740 inm_nrele, tinm) {
1741 SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1742 inm_nrele);
1743 inm_release_locked(inm);
1744 }
1745 }
1746 }
1747
1748 out_locked:
1749 IGMP_UNLOCK();
1750 IN_MULTI_UNLOCK();
1751 }
1752
1753 /*
1754 * Update host report group timer for IGMPv1/v2.
1755 * Will update the global pending timer flags.
1756 */
1757 static void
1758 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1759 {
1760 int report_timer_expired;
1761
1762 IN_MULTI_LOCK_ASSERT();
1763 IGMP_LOCK_ASSERT();
1764
1765 if (inm->inm_timer == 0) {
1766 report_timer_expired = 0;
1767 } else if (--inm->inm_timer == 0) {
1768 report_timer_expired = 1;
1769 } else {
1770 V_current_state_timers_running = 1;
1771 return;
1772 }
1773
1774 switch (inm->inm_state) {
1775 case IGMP_NOT_MEMBER:
1776 case IGMP_SILENT_MEMBER:
1777 case IGMP_IDLE_MEMBER:
1778 case IGMP_LAZY_MEMBER:
1779 case IGMP_SLEEPING_MEMBER:
1780 case IGMP_AWAKENING_MEMBER:
1781 break;
1782 case IGMP_REPORTING_MEMBER:
1783 if (report_timer_expired) {
1784 inm->inm_state = IGMP_IDLE_MEMBER;
1785 (void)igmp_v1v2_queue_report(inm,
1786 (version == IGMP_VERSION_2) ?
1787 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1788 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1789 }
1790 break;
1791 case IGMP_G_QUERY_PENDING_MEMBER:
1792 case IGMP_SG_QUERY_PENDING_MEMBER:
1793 case IGMP_LEAVING_MEMBER:
1794 break;
1795 }
1796 }
1797
1798 /*
1799 * Update a group's timers for IGMPv3.
1800 * Will update the global pending timer flags.
1801 * Note: Unlocked read from igi.
1802 */
1803 static void
1804 igmp_v3_process_group_timers(struct igmp_ifsoftc *igi,
1805 struct mbufq *qrq, struct mbufq *scq,
1806 struct in_multi *inm, const int uri_fasthz)
1807 {
1808 int query_response_timer_expired;
1809 int state_change_retransmit_timer_expired;
1810
1811 IN_MULTI_LOCK_ASSERT();
1812 IGMP_LOCK_ASSERT();
1813
1814 query_response_timer_expired = 0;
1815 state_change_retransmit_timer_expired = 0;
1816
1817 /*
1818 * During a transition from v1/v2 compatibility mode back to v3,
1819 * a group record in REPORTING state may still have its group
1820 * timer active. This is a no-op in this function; it is easier
1821 * to deal with it here than to complicate the slow-timeout path.
1822 */
1823 if (inm->inm_timer == 0) {
1824 query_response_timer_expired = 0;
1825 } else if (--inm->inm_timer == 0) {
1826 query_response_timer_expired = 1;
1827 } else {
1828 V_current_state_timers_running = 1;
1829 }
1830
1831 if (inm->inm_sctimer == 0) {
1832 state_change_retransmit_timer_expired = 0;
1833 } else if (--inm->inm_sctimer == 0) {
1834 state_change_retransmit_timer_expired = 1;
1835 } else {
1836 V_state_change_timers_running = 1;
1837 }
1838
1839 /* We are in fasttimo, so be quick about it. */
1840 if (!state_change_retransmit_timer_expired &&
1841 !query_response_timer_expired)
1842 return;
1843
1844 switch (inm->inm_state) {
1845 case IGMP_NOT_MEMBER:
1846 case IGMP_SILENT_MEMBER:
1847 case IGMP_SLEEPING_MEMBER:
1848 case IGMP_LAZY_MEMBER:
1849 case IGMP_AWAKENING_MEMBER:
1850 case IGMP_IDLE_MEMBER:
1851 break;
1852 case IGMP_G_QUERY_PENDING_MEMBER:
1853 case IGMP_SG_QUERY_PENDING_MEMBER:
1854 /*
1855 * Respond to a previously pending Group-Specific
1856 * or Group-and-Source-Specific query by enqueueing
1857 * the appropriate Current-State report for
1858 * immediate transmission.
1859 */
1860 if (query_response_timer_expired) {
1861 int retval;
1862
1863 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1864 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1865 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1866 __func__, retval);
1867 inm->inm_state = IGMP_REPORTING_MEMBER;
1868 /* XXX Clear recorded sources for next time. */
1869 inm_clear_recorded(inm);
1870 }
1871 /* FALLTHROUGH */
1872 case IGMP_REPORTING_MEMBER:
1873 case IGMP_LEAVING_MEMBER:
1874 if (state_change_retransmit_timer_expired) {
1875 /*
1876 * State-change retransmission timer fired.
1877 * If there are any further pending retransmissions,
1878 * set the global pending state-change flag, and
1879 * reset the timer.
1880 */
1881 if (--inm->inm_scrv > 0) {
1882 inm->inm_sctimer = uri_fasthz;
1883 V_state_change_timers_running = 1;
1884 }
1885 /*
1886 * Retransmit the previously computed state-change
1887 * report. If there are no further pending
1888 * retransmissions, the mbuf queue will be consumed.
1889 * Update T0 state to T1 as we have now sent
1890 * a state-change.
1891 */
1892 (void)igmp_v3_merge_state_changes(inm, scq);
1893
1894 inm_commit(inm);
1895 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1896 ntohl(inm->inm_addr.s_addr),
1897 inm->inm_ifp->if_xname);
1898
1899 /*
1900 * If we are leaving the group for good, make sure
1901 * we release IGMP's reference to it.
1902 * This release must be deferred using a SLIST,
1903 * as we are called from a loop which traverses
1904 * the in_ifmultiaddr TAILQ.
1905 */
1906 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1907 inm->inm_scrv == 0) {
1908 inm->inm_state = IGMP_NOT_MEMBER;
1909 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1910 inm, inm_nrele);
1911 }
1912 }
1913 break;
1914 }
1915 }
1916
1917
1918 /*
1919 * Suppress a group's pending response to a group or source/group query.
1920 *
1921 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1922 * Do NOT update ST1/ST0 as this operation merely suppresses
1923 * the currently pending group record.
1924 * Do NOT suppress the response to a general query. It is possible but
1925 * it would require adding another state or flag.
1926 */
1927 static void
1928 igmp_v3_suppress_group_record(struct in_multi *inm)
1929 {
1930
1931 IN_MULTI_LOCK_ASSERT();
1932
1933 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1934 ("%s: not IGMPv3 mode on link", __func__));
1935
1936 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1937 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1938 return;
1939
1940 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1941 inm_clear_recorded(inm);
1942
1943 inm->inm_timer = 0;
1944 inm->inm_state = IGMP_REPORTING_MEMBER;
1945 }
1946
1947 /*
1948 * Switch to a different IGMP version on the given interface,
1949 * as per Section 7.2.1.
1950 */
1951 static void
1952 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1953 {
1954 int old_version_timer;
1955
1956 IGMP_LOCK_ASSERT();
1957
1958 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1959 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1960
1961 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1962 /*
1963 * Compute the "Older Version Querier Present" timer as per
1964 * Section 8.12.
1965 */
1966 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1967 old_version_timer *= PR_SLOWHZ;
1968
1969 if (version == IGMP_VERSION_1) {
1970 igi->igi_v1_timer = old_version_timer;
1971 igi->igi_v2_timer = 0;
1972 } else if (version == IGMP_VERSION_2) {
1973 igi->igi_v1_timer = 0;
1974 igi->igi_v2_timer = old_version_timer;
1975 }
1976 }
1977
1978 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1979 if (igi->igi_version != IGMP_VERSION_2) {
1980 igi->igi_version = IGMP_VERSION_2;
1981 igmp_v3_cancel_link_timers(igi);
1982 }
1983 } else if (igi->igi_v1_timer > 0) {
1984 if (igi->igi_version != IGMP_VERSION_1) {
1985 igi->igi_version = IGMP_VERSION_1;
1986 igmp_v3_cancel_link_timers(igi);
1987 }
1988 }
1989 }
1990
1991 /*
1992 * Cancel pending IGMPv3 timers for the given link and all groups
1993 * joined on it; state-change, general-query, and group-query timers.
1994 *
1995 * Only ever called on a transition from v3 to Compatibility mode. Kill
1996 * the timers stone dead (this may be expensive for large N groups), they
1997 * will be restarted if Compatibility Mode deems that they must be due to
1998 * query processing.
1999 */
2000 static void
2001 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2002 {
2003 struct ifmultiaddr *ifma;
2004 struct ifnet *ifp;
2005 struct in_multi *inm, *tinm;
2006
2007 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2008 igi->igi_ifp, igi->igi_ifp->if_xname);
2009
2010 IN_MULTI_LOCK_ASSERT();
2011 IGMP_LOCK_ASSERT();
2012
2013 /*
2014 * Stop the v3 General Query Response on this link stone dead.
2015 * If fasttimo is woken up due to V_interface_timers_running,
2016 * the flag will be cleared if there are no pending link timers.
2017 */
2018 igi->igi_v3_timer = 0;
2019
2020 /*
2021 * Now clear the current-state and state-change report timers
2022 * for all memberships scoped to this link.
2023 */
2024 ifp = igi->igi_ifp;
2025 IF_ADDR_RLOCK(ifp);
2026 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2027 if (ifma->ifma_addr->sa_family != AF_INET ||
2028 ifma->ifma_protospec == NULL)
2029 continue;
2030 inm = (struct in_multi *)ifma->ifma_protospec;
2031 switch (inm->inm_state) {
2032 case IGMP_NOT_MEMBER:
2033 case IGMP_SILENT_MEMBER:
2034 case IGMP_IDLE_MEMBER:
2035 case IGMP_LAZY_MEMBER:
2036 case IGMP_SLEEPING_MEMBER:
2037 case IGMP_AWAKENING_MEMBER:
2038 /*
2039 * These states are either not relevant in v3 mode,
2040 * or are unreported. Do nothing.
2041 */
2042 break;
2043 case IGMP_LEAVING_MEMBER:
2044 /*
2045 * If we are leaving the group and switching to
2046 * compatibility mode, we need to release the final
2047 * reference held for issuing the INCLUDE {}, and
2048 * transition to REPORTING to ensure the host leave
2049 * message is sent upstream to the old querier --
2050 * transition to NOT would lose the leave and race.
2051 */
2052 SLIST_INSERT_HEAD(&igi->igi_relinmhead, inm, inm_nrele);
2053 /* FALLTHROUGH */
2054 case IGMP_G_QUERY_PENDING_MEMBER:
2055 case IGMP_SG_QUERY_PENDING_MEMBER:
2056 inm_clear_recorded(inm);
2057 /* FALLTHROUGH */
2058 case IGMP_REPORTING_MEMBER:
2059 inm->inm_state = IGMP_REPORTING_MEMBER;
2060 break;
2061 }
2062 /*
2063 * Always clear state-change and group report timers.
2064 * Free any pending IGMPv3 state-change records.
2065 */
2066 inm->inm_sctimer = 0;
2067 inm->inm_timer = 0;
2068 mbufq_drain(&inm->inm_scq);
2069 }
2070 IF_ADDR_RUNLOCK(ifp);
2071 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, tinm) {
2072 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
2073 inm_release_locked(inm);
2074 }
2075 }
2076
2077 /*
2078 * Update the Older Version Querier Present timers for a link.
2079 * See Section 7.2.1 of RFC 3376.
2080 */
2081 static void
2082 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2083 {
2084
2085 IGMP_LOCK_ASSERT();
2086
2087 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2088 /*
2089 * IGMPv1 and IGMPv2 Querier Present timers expired.
2090 *
2091 * Revert to IGMPv3.
2092 */
2093 if (igi->igi_version != IGMP_VERSION_3) {
2094 CTR5(KTR_IGMPV3,
2095 "%s: transition from v%d -> v%d on %p(%s)",
2096 __func__, igi->igi_version, IGMP_VERSION_3,
2097 igi->igi_ifp, igi->igi_ifp->if_xname);
2098 igi->igi_version = IGMP_VERSION_3;
2099 }
2100 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2101 /*
2102 * IGMPv1 Querier Present timer expired,
2103 * IGMPv2 Querier Present timer running.
2104 * If IGMPv2 was disabled since last timeout,
2105 * revert to IGMPv3.
2106 * If IGMPv2 is enabled, revert to IGMPv2.
2107 */
2108 if (!V_igmp_v2enable) {
2109 CTR5(KTR_IGMPV3,
2110 "%s: transition from v%d -> v%d on %p(%s)",
2111 __func__, igi->igi_version, IGMP_VERSION_3,
2112 igi->igi_ifp, igi->igi_ifp->if_xname);
2113 igi->igi_v2_timer = 0;
2114 igi->igi_version = IGMP_VERSION_3;
2115 } else {
2116 --igi->igi_v2_timer;
2117 if (igi->igi_version != IGMP_VERSION_2) {
2118 CTR5(KTR_IGMPV3,
2119 "%s: transition from v%d -> v%d on %p(%s)",
2120 __func__, igi->igi_version, IGMP_VERSION_2,
2121 igi->igi_ifp, igi->igi_ifp->if_xname);
2122 igi->igi_version = IGMP_VERSION_2;
2123 igmp_v3_cancel_link_timers(igi);
2124 }
2125 }
2126 } else if (igi->igi_v1_timer > 0) {
2127 /*
2128 * IGMPv1 Querier Present timer running.
2129 * Stop IGMPv2 timer if running.
2130 *
2131 * If IGMPv1 was disabled since last timeout,
2132 * revert to IGMPv3.
2133 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2134 */
2135 if (!V_igmp_v1enable) {
2136 CTR5(KTR_IGMPV3,
2137 "%s: transition from v%d -> v%d on %p(%s)",
2138 __func__, igi->igi_version, IGMP_VERSION_3,
2139 igi->igi_ifp, igi->igi_ifp->if_xname);
2140 igi->igi_v1_timer = 0;
2141 igi->igi_version = IGMP_VERSION_3;
2142 } else {
2143 --igi->igi_v1_timer;
2144 }
2145 if (igi->igi_v2_timer > 0) {
2146 CTR3(KTR_IGMPV3,
2147 "%s: cancel v2 timer on %p(%s)",
2148 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2149 igi->igi_v2_timer = 0;
2150 }
2151 }
2152 }
2153
2154 /*
2155 * Global slowtimo handler.
2156 * VIMAGE: Timeout handlers are expected to service all vimages.
2157 */
2158 void
2159 igmp_slowtimo(void)
2160 {
2161 VNET_ITERATOR_DECL(vnet_iter);
2162
2163 VNET_LIST_RLOCK_NOSLEEP();
2164 VNET_FOREACH(vnet_iter) {
2165 CURVNET_SET(vnet_iter);
2166 igmp_slowtimo_vnet();
2167 CURVNET_RESTORE();
2168 }
2169 VNET_LIST_RUNLOCK_NOSLEEP();
2170 }
2171
2172 /*
2173 * Per-vnet slowtimo handler.
2174 */
2175 static void
2176 igmp_slowtimo_vnet(void)
2177 {
2178 struct igmp_ifsoftc *igi;
2179
2180 IGMP_LOCK();
2181
2182 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2183 igmp_v1v2_process_querier_timers(igi);
2184 }
2185
2186 IGMP_UNLOCK();
2187 }
2188
2189 /*
2190 * Dispatch an IGMPv1/v2 host report or leave message.
2191 * These are always small enough to fit inside a single mbuf.
2192 */
2193 static int
2194 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2195 {
2196 struct ifnet *ifp;
2197 struct igmp *igmp;
2198 struct ip *ip;
2199 struct mbuf *m;
2200
2201 IN_MULTI_LOCK_ASSERT();
2202 IGMP_LOCK_ASSERT();
2203
2204 ifp = inm->inm_ifp;
2205
2206 m = m_gethdr(M_NOWAIT, MT_DATA);
2207 if (m == NULL)
2208 return (ENOMEM);
2209 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2210
2211 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2212
2213 m->m_data += sizeof(struct ip);
2214 m->m_len = sizeof(struct igmp);
2215
2216 igmp = mtod(m, struct igmp *);
2217 igmp->igmp_type = type;
2218 igmp->igmp_code = 0;
2219 igmp->igmp_group = inm->inm_addr;
2220 igmp->igmp_cksum = 0;
2221 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2222
2223 m->m_data -= sizeof(struct ip);
2224 m->m_len += sizeof(struct ip);
2225
2226 ip = mtod(m, struct ip *);
2227 ip->ip_tos = 0;
2228 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2229 ip->ip_off = 0;
2230 ip->ip_p = IPPROTO_IGMP;
2231 ip->ip_src.s_addr = INADDR_ANY;
2232
2233 if (type == IGMP_HOST_LEAVE_MESSAGE)
2234 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2235 else
2236 ip->ip_dst = inm->inm_addr;
2237
2238 igmp_save_context(m, ifp);
2239
2240 m->m_flags |= M_IGMPV2;
2241 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2242 m->m_flags |= M_IGMP_LOOP;
2243
2244 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2245 netisr_dispatch(NETISR_IGMP, m);
2246
2247 return (0);
2248 }
2249
2250 /*
2251 * Process a state change from the upper layer for the given IPv4 group.
2252 *
2253 * Each socket holds a reference on the in_multi in its own ip_moptions.
2254 * The socket layer will have made the necessary updates to.the group
2255 * state, it is now up to IGMP to issue a state change report if there
2256 * has been any change between T0 (when the last state-change was issued)
2257 * and T1 (now).
2258 *
2259 * We use the IGMPv3 state machine at group level. The IGMP module
2260 * however makes the decision as to which IGMP protocol version to speak.
2261 * A state change *from* INCLUDE {} always means an initial join.
2262 * A state change *to* INCLUDE {} always means a final leave.
2263 *
2264 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2265 * save ourselves a bunch of work; any exclusive mode groups need not
2266 * compute source filter lists.
2267 *
2268 * VIMAGE: curvnet should have been set by caller, as this routine
2269 * is called from the socket option handlers.
2270 */
2271 int
2272 igmp_change_state(struct in_multi *inm)
2273 {
2274 struct igmp_ifsoftc *igi;
2275 struct ifnet *ifp;
2276 int error;
2277
2278 IN_MULTI_LOCK_ASSERT();
2279
2280 error = 0;
2281
2282 /*
2283 * Try to detect if the upper layer just asked us to change state
2284 * for an interface which has now gone away.
2285 */
2286 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2287 ifp = inm->inm_ifma->ifma_ifp;
2288 /*
2289 * Sanity check that netinet's notion of ifp is the
2290 * same as net's.
2291 */
2292 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2293
2294 IGMP_LOCK();
2295
2296 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2297 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2298
2299 /*
2300 * If we detect a state transition to or from MCAST_UNDEFINED
2301 * for this group, then we are starting or finishing an IGMP
2302 * life cycle for this group.
2303 */
2304 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2305 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2306 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2307 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2308 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2309 error = igmp_initial_join(inm, igi);
2310 goto out_locked;
2311 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2312 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2313 igmp_final_leave(inm, igi);
2314 goto out_locked;
2315 }
2316 } else {
2317 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2318 }
2319
2320 error = igmp_handle_state_change(inm, igi);
2321
2322 out_locked:
2323 IGMP_UNLOCK();
2324 return (error);
2325 }
2326
2327 /*
2328 * Perform the initial join for an IGMP group.
2329 *
2330 * When joining a group:
2331 * If the group should have its IGMP traffic suppressed, do nothing.
2332 * IGMPv1 starts sending IGMPv1 host membership reports.
2333 * IGMPv2 starts sending IGMPv2 host membership reports.
2334 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2335 * initial state of the membership.
2336 */
2337 static int
2338 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2339 {
2340 struct ifnet *ifp;
2341 struct mbufq *mq;
2342 int error, retval, syncstates;
2343
2344 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2345 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, 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 mq = &inm->inm_scq;
2415 mbufq_drain(mq);
2416 retval = igmp_v3_enqueue_group_record(mq, 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 0x%08x/%s", __func__,
2455 ntohl(inm->inm_addr.s_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_ifsoftc *igi)
2466 {
2467 struct ifnet *ifp;
2468 int retval;
2469
2470 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2471 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2472
2473 ifp = inm->inm_ifp;
2474
2475 IN_MULTI_LOCK_ASSERT();
2476 IGMP_LOCK_ASSERT();
2477
2478 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2479
2480 if ((ifp->if_flags & IFF_LOOPBACK) ||
2481 (igi->igi_flags & IGIF_SILENT) ||
2482 !igmp_isgroupreported(inm->inm_addr) ||
2483 (igi->igi_version != IGMP_VERSION_3)) {
2484 if (!igmp_isgroupreported(inm->inm_addr)) {
2485 CTR1(KTR_IGMPV3,
2486 "%s: not kicking state machine for silent group", __func__);
2487 }
2488 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2489 inm_commit(inm);
2490 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2491 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2492 return (0);
2493 }
2494
2495 mbufq_drain(&inm->inm_scq);
2496
2497 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2498 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2499 if (retval <= 0)
2500 return (-retval);
2501
2502 /*
2503 * If record(s) were enqueued, start the state-change
2504 * report timer for this group.
2505 */
2506 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2507 inm->inm_sctimer = 1;
2508 V_state_change_timers_running = 1;
2509
2510 return (0);
2511 }
2512
2513 /*
2514 * Perform the final leave for an IGMP group.
2515 *
2516 * When leaving a group:
2517 * IGMPv1 does nothing.
2518 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2519 * IGMPv3 enqueues a state-change report containing a transition
2520 * to INCLUDE {} for immediate transmission.
2521 */
2522 static void
2523 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2524 {
2525 int syncstates;
2526
2527 syncstates = 1;
2528
2529 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2530 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2531 inm->inm_ifp->if_xname);
2532
2533 IN_MULTI_LOCK_ASSERT();
2534 IGMP_LOCK_ASSERT();
2535
2536 switch (inm->inm_state) {
2537 case IGMP_NOT_MEMBER:
2538 case IGMP_SILENT_MEMBER:
2539 case IGMP_LEAVING_MEMBER:
2540 /* Already leaving or left; do nothing. */
2541 CTR1(KTR_IGMPV3,
2542 "%s: not kicking state machine for silent group", __func__);
2543 break;
2544 case IGMP_REPORTING_MEMBER:
2545 case IGMP_IDLE_MEMBER:
2546 case IGMP_G_QUERY_PENDING_MEMBER:
2547 case IGMP_SG_QUERY_PENDING_MEMBER:
2548 if (igi->igi_version == IGMP_VERSION_2) {
2549 #ifdef INVARIANTS
2550 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2551 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2552 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2553 __func__);
2554 #endif
2555 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2556 inm->inm_state = IGMP_NOT_MEMBER;
2557 } else if (igi->igi_version == IGMP_VERSION_3) {
2558 /*
2559 * Stop group timer and all pending reports.
2560 * Immediately enqueue a state-change report
2561 * TO_IN {} to be sent on the next fast timeout,
2562 * giving us an opportunity to merge reports.
2563 */
2564 mbufq_drain(&inm->inm_scq);
2565 inm->inm_timer = 0;
2566 if (igi->igi_flags & IGIF_LOOPBACK) {
2567 inm->inm_scrv = 1;
2568 } else {
2569 inm->inm_scrv = igi->igi_rv;
2570 }
2571 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2572 "pending retransmissions.", __func__,
2573 ntohl(inm->inm_addr.s_addr),
2574 inm->inm_ifp->if_xname, inm->inm_scrv);
2575 if (inm->inm_scrv == 0) {
2576 inm->inm_state = IGMP_NOT_MEMBER;
2577 inm->inm_sctimer = 0;
2578 } else {
2579 int retval;
2580
2581 inm_acquire_locked(inm);
2582
2583 retval = igmp_v3_enqueue_group_record(
2584 &inm->inm_scq, inm, 1, 0, 0);
2585 KASSERT(retval != 0,
2586 ("%s: enqueue record = %d", __func__,
2587 retval));
2588
2589 inm->inm_state = IGMP_LEAVING_MEMBER;
2590 inm->inm_sctimer = 1;
2591 V_state_change_timers_running = 1;
2592 syncstates = 0;
2593 }
2594 break;
2595 }
2596 break;
2597 case IGMP_LAZY_MEMBER:
2598 case IGMP_SLEEPING_MEMBER:
2599 case IGMP_AWAKENING_MEMBER:
2600 /* Our reports are suppressed; do nothing. */
2601 break;
2602 }
2603
2604 if (syncstates) {
2605 inm_commit(inm);
2606 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2607 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2608 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2609 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2610 __func__, ntohl(inm->inm_addr.s_addr),
2611 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 mbufq *mq, 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(mq, inm));
2734
2735 if (type == IGMP_DO_NOTHING) {
2736 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2737 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2738 return (0);
2739 }
2740
2741 /*
2742 * If any sources are present, we must be able to fit at least
2743 * one in the trailing space of the tail packet's mbuf,
2744 * ideally more.
2745 */
2746 minrec0len = sizeof(struct igmp_grouprec);
2747 if (record_has_sources)
2748 minrec0len += sizeof(in_addr_t);
2749
2750 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2751 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2752 inm->inm_ifp->if_xname);
2753
2754 /*
2755 * Check if we have a packet in the tail of the queue for this
2756 * group into which the first group record for this group will fit.
2757 * Otherwise allocate a new packet.
2758 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2759 * Note: Group records for G/GSR query responses MUST be sent
2760 * in their own packet.
2761 */
2762 m0 = mbufq_last(mq);
2763 if (!is_group_query &&
2764 m0 != NULL &&
2765 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2766 (m0->m_pkthdr.len + minrec0len) <
2767 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2768 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2769 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2770 m = m0;
2771 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2772 } else {
2773 if (mbufq_full(mq)) {
2774 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2775 return (-ENOMEM);
2776 }
2777 m = NULL;
2778 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2779 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2780 if (!is_state_change && !is_group_query) {
2781 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2782 if (m)
2783 m->m_data += IGMP_LEADINGSPACE;
2784 }
2785 if (m == NULL) {
2786 m = m_gethdr(M_NOWAIT, MT_DATA);
2787 if (m)
2788 M_ALIGN(m, IGMP_LEADINGSPACE);
2789 }
2790 if (m == NULL)
2791 return (-ENOMEM);
2792
2793 igmp_save_context(m, ifp);
2794
2795 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2796 }
2797
2798 /*
2799 * Append group record.
2800 * If we have sources, we don't know how many yet.
2801 */
2802 ig.ig_type = type;
2803 ig.ig_datalen = 0;
2804 ig.ig_numsrc = 0;
2805 ig.ig_group = inm->inm_addr;
2806 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2807 if (m != m0)
2808 m_freem(m);
2809 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2810 return (-ENOMEM);
2811 }
2812 nbytes += sizeof(struct igmp_grouprec);
2813
2814 /*
2815 * Append as many sources as will fit in the first packet.
2816 * If we are appending to a new packet, the chain allocation
2817 * may potentially use clusters; use m_getptr() in this case.
2818 * If we are appending to an existing packet, we need to obtain
2819 * a pointer to the group record after m_append(), in case a new
2820 * mbuf was allocated.
2821 * Only append sources which are in-mode at t1. If we are
2822 * transitioning to MCAST_UNDEFINED state on the group, do not
2823 * include source entries.
2824 * Only report recorded sources in our filter set when responding
2825 * to a group-source query.
2826 */
2827 if (record_has_sources) {
2828 if (m == m0) {
2829 md = m_last(m);
2830 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2831 md->m_len - nbytes);
2832 } else {
2833 md = m_getptr(m, 0, &off);
2834 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2835 off);
2836 }
2837 msrcs = 0;
2838 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2839 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2840 ims->ims_haddr);
2841 now = ims_get_mode(inm, ims, 1);
2842 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2843 if ((now != mode) ||
2844 (now == mode && mode == MCAST_UNDEFINED)) {
2845 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2846 continue;
2847 }
2848 if (is_source_query && ims->ims_stp == 0) {
2849 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2850 __func__);
2851 continue;
2852 }
2853 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2854 naddr = htonl(ims->ims_haddr);
2855 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2856 if (m != m0)
2857 m_freem(m);
2858 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2859 __func__);
2860 return (-ENOMEM);
2861 }
2862 nbytes += sizeof(in_addr_t);
2863 ++msrcs;
2864 if (msrcs == m0srcs)
2865 break;
2866 }
2867 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2868 msrcs);
2869 pig->ig_numsrc = htons(msrcs);
2870 nbytes += (msrcs * sizeof(in_addr_t));
2871 }
2872
2873 if (is_source_query && msrcs == 0) {
2874 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2875 if (m != m0)
2876 m_freem(m);
2877 return (0);
2878 }
2879
2880 /*
2881 * We are good to go with first packet.
2882 */
2883 if (m != m0) {
2884 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2885 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2886 mbufq_enqueue(mq, m);
2887 } else
2888 m->m_pkthdr.PH_vt.vt_nrecs++;
2889
2890 /*
2891 * No further work needed if no source list in packet(s).
2892 */
2893 if (!record_has_sources)
2894 return (nbytes);
2895
2896 /*
2897 * Whilst sources remain to be announced, we need to allocate
2898 * a new packet and fill out as many sources as will fit.
2899 * Always try for a cluster first.
2900 */
2901 while (nims != NULL) {
2902 if (mbufq_full(mq)) {
2903 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2904 return (-ENOMEM);
2905 }
2906 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2907 if (m)
2908 m->m_data += IGMP_LEADINGSPACE;
2909 if (m == NULL) {
2910 m = m_gethdr(M_NOWAIT, MT_DATA);
2911 if (m)
2912 M_ALIGN(m, IGMP_LEADINGSPACE);
2913 }
2914 if (m == NULL)
2915 return (-ENOMEM);
2916 igmp_save_context(m, ifp);
2917 md = m_getptr(m, 0, &off);
2918 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2919 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2920
2921 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2922 if (m != m0)
2923 m_freem(m);
2924 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2925 return (-ENOMEM);
2926 }
2927 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2928 nbytes += sizeof(struct igmp_grouprec);
2929
2930 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2931 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2932
2933 msrcs = 0;
2934 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2935 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2936 ims->ims_haddr);
2937 now = ims_get_mode(inm, ims, 1);
2938 if ((now != mode) ||
2939 (now == mode && mode == MCAST_UNDEFINED)) {
2940 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2941 continue;
2942 }
2943 if (is_source_query && ims->ims_stp == 0) {
2944 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2945 __func__);
2946 continue;
2947 }
2948 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2949 naddr = htonl(ims->ims_haddr);
2950 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2951 if (m != m0)
2952 m_freem(m);
2953 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2954 __func__);
2955 return (-ENOMEM);
2956 }
2957 ++msrcs;
2958 if (msrcs == m0srcs)
2959 break;
2960 }
2961 pig->ig_numsrc = htons(msrcs);
2962 nbytes += (msrcs * sizeof(in_addr_t));
2963
2964 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2965 mbufq_enqueue(mq, m);
2966 }
2967
2968 return (nbytes);
2969 }
2970
2971 /*
2972 * Type used to mark record pass completion.
2973 * We exploit the fact we can cast to this easily from the
2974 * current filter modes on each ip_msource node.
2975 */
2976 typedef enum {
2977 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2978 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2979 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2980 REC_FULL = REC_ALLOW | REC_BLOCK
2981 } rectype_t;
2982
2983 /*
2984 * Enqueue an IGMPv3 filter list change to the given output queue.
2985 *
2986 * Source list filter state is held in an RB-tree. When the filter list
2987 * for a group is changed without changing its mode, we need to compute
2988 * the deltas between T0 and T1 for each source in the filter set,
2989 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2990 *
2991 * As we may potentially queue two record types, and the entire R-B tree
2992 * needs to be walked at once, we break this out into its own function
2993 * so we can generate a tightly packed queue of packets.
2994 *
2995 * XXX This could be written to only use one tree walk, although that makes
2996 * serializing into the mbuf chains a bit harder. For now we do two walks
2997 * which makes things easier on us, and it may or may not be harder on
2998 * the L2 cache.
2999 *
3000 * If successful the size of all data appended to the queue is returned,
3001 * otherwise an error code less than zero is returned, or zero if
3002 * no record(s) were appended.
3003 */
3004 static int
3005 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3006 {
3007 static const int MINRECLEN =
3008 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3009 struct ifnet *ifp;
3010 struct igmp_grouprec ig;
3011 struct igmp_grouprec *pig;
3012 struct ip_msource *ims, *nims;
3013 struct mbuf *m, *m0, *md;
3014 in_addr_t naddr;
3015 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3016 int nallow, nblock;
3017 uint8_t mode, now, then;
3018 rectype_t crt, drt, nrt;
3019
3020 IN_MULTI_LOCK_ASSERT();
3021
3022 if (inm->inm_nsrc == 0 ||
3023 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3024 return (0);
3025
3026 ifp = inm->inm_ifp; /* interface */
3027 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3028 crt = REC_NONE; /* current group record type */
3029 drt = REC_NONE; /* mask of completed group record types */
3030 nrt = REC_NONE; /* record type for current node */
3031 m0srcs = 0; /* # source which will fit in current mbuf chain */
3032 nbytes = 0; /* # of bytes appended to group's state-change queue */
3033 npbytes = 0; /* # of bytes appended this packet */
3034 rsrcs = 0; /* # sources encoded in current record */
3035 schanged = 0; /* # nodes encoded in overall filter change */
3036 nallow = 0; /* # of source entries in ALLOW_NEW */
3037 nblock = 0; /* # of source entries in BLOCK_OLD */
3038 nims = NULL; /* next tree node pointer */
3039
3040 /*
3041 * For each possible filter record mode.
3042 * The first kind of source we encounter tells us which
3043 * is the first kind of record we start appending.
3044 * If a node transitioned to UNDEFINED at t1, its mode is treated
3045 * as the inverse of the group's filter mode.
3046 */
3047 while (drt != REC_FULL) {
3048 do {
3049 m0 = mbufq_last(mq);
3050 if (m0 != NULL &&
3051 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3052 IGMP_V3_REPORT_MAXRECS) &&
3053 (m0->m_pkthdr.len + MINRECLEN) <
3054 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3055 m = m0;
3056 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3057 sizeof(struct igmp_grouprec)) /
3058 sizeof(in_addr_t);
3059 CTR1(KTR_IGMPV3,
3060 "%s: use previous packet", __func__);
3061 } else {
3062 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3063 if (m)
3064 m->m_data += IGMP_LEADINGSPACE;
3065 if (m == NULL) {
3066 m = m_gethdr(M_NOWAIT, MT_DATA);
3067 if (m)
3068 M_ALIGN(m, IGMP_LEADINGSPACE);
3069 }
3070 if (m == NULL) {
3071 CTR1(KTR_IGMPV3,
3072 "%s: m_get*() failed", __func__);
3073 return (-ENOMEM);
3074 }
3075 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3076 igmp_save_context(m, ifp);
3077 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3078 sizeof(struct igmp_grouprec)) /
3079 sizeof(in_addr_t);
3080 npbytes = 0;
3081 CTR1(KTR_IGMPV3,
3082 "%s: allocated new packet", __func__);
3083 }
3084 /*
3085 * Append the IGMP group record header to the
3086 * current packet's data area.
3087 * Recalculate pointer to free space for next
3088 * group record, in case m_append() allocated
3089 * a new mbuf or cluster.
3090 */
3091 memset(&ig, 0, sizeof(ig));
3092 ig.ig_group = inm->inm_addr;
3093 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3094 if (m != m0)
3095 m_freem(m);
3096 CTR1(KTR_IGMPV3,
3097 "%s: m_append() failed", __func__);
3098 return (-ENOMEM);
3099 }
3100 npbytes += sizeof(struct igmp_grouprec);
3101 if (m != m0) {
3102 /* new packet; offset in c hain */
3103 md = m_getptr(m, npbytes -
3104 sizeof(struct igmp_grouprec), &off);
3105 pig = (struct igmp_grouprec *)(mtod(md,
3106 uint8_t *) + off);
3107 } else {
3108 /* current packet; offset from last append */
3109 md = m_last(m);
3110 pig = (struct igmp_grouprec *)(mtod(md,
3111 uint8_t *) + md->m_len -
3112 sizeof(struct igmp_grouprec));
3113 }
3114 /*
3115 * Begin walking the tree for this record type
3116 * pass, or continue from where we left off
3117 * previously if we had to allocate a new packet.
3118 * Only report deltas in-mode at t1.
3119 * We need not report included sources as allowed
3120 * if we are in inclusive mode on the group,
3121 * however the converse is not true.
3122 */
3123 rsrcs = 0;
3124 if (nims == NULL)
3125 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3126 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3127 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3128 __func__, ims->ims_haddr);
3129 now = ims_get_mode(inm, ims, 1);
3130 then = ims_get_mode(inm, ims, 0);
3131 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3132 __func__, then, now);
3133 if (now == then) {
3134 CTR1(KTR_IGMPV3,
3135 "%s: skip unchanged", __func__);
3136 continue;
3137 }
3138 if (mode == MCAST_EXCLUDE &&
3139 now == MCAST_INCLUDE) {
3140 CTR1(KTR_IGMPV3,
3141 "%s: skip IN src on EX group",
3142 __func__);
3143 continue;
3144 }
3145 nrt = (rectype_t)now;
3146 if (nrt == REC_NONE)
3147 nrt = (rectype_t)(~mode & REC_FULL);
3148 if (schanged++ == 0) {
3149 crt = nrt;
3150 } else if (crt != nrt)
3151 continue;
3152 naddr = htonl(ims->ims_haddr);
3153 if (!m_append(m, sizeof(in_addr_t),
3154 (void *)&naddr)) {
3155 if (m != m0)
3156 m_freem(m);
3157 CTR1(KTR_IGMPV3,
3158 "%s: m_append() failed", __func__);
3159 return (-ENOMEM);
3160 }
3161 nallow += !!(crt == REC_ALLOW);
3162 nblock += !!(crt == REC_BLOCK);
3163 if (++rsrcs == m0srcs)
3164 break;
3165 }
3166 /*
3167 * If we did not append any tree nodes on this
3168 * pass, back out of allocations.
3169 */
3170 if (rsrcs == 0) {
3171 npbytes -= sizeof(struct igmp_grouprec);
3172 if (m != m0) {
3173 CTR1(KTR_IGMPV3,
3174 "%s: m_free(m)", __func__);
3175 m_freem(m);
3176 } else {
3177 CTR1(KTR_IGMPV3,
3178 "%s: m_adj(m, -ig)", __func__);
3179 m_adj(m, -((int)sizeof(
3180 struct igmp_grouprec)));
3181 }
3182 continue;
3183 }
3184 npbytes += (rsrcs * sizeof(in_addr_t));
3185 if (crt == REC_ALLOW)
3186 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3187 else if (crt == REC_BLOCK)
3188 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3189 pig->ig_numsrc = htons(rsrcs);
3190 /*
3191 * Count the new group record, and enqueue this
3192 * packet if it wasn't already queued.
3193 */
3194 m->m_pkthdr.PH_vt.vt_nrecs++;
3195 if (m != m0)
3196 mbufq_enqueue(mq, m);
3197 nbytes += npbytes;
3198 } while (nims != NULL);
3199 drt |= crt;
3200 crt = (~crt & REC_FULL);
3201 }
3202
3203 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3204 nallow, nblock);
3205
3206 return (nbytes);
3207 }
3208
3209 static int
3210 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3211 {
3212 struct mbufq *gq;
3213 struct mbuf *m; /* pending state-change */
3214 struct mbuf *m0; /* copy of pending state-change */
3215 struct mbuf *mt; /* last state-change in packet */
3216 int docopy, domerge;
3217 u_int recslen;
3218
3219 docopy = 0;
3220 domerge = 0;
3221 recslen = 0;
3222
3223 IN_MULTI_LOCK_ASSERT();
3224 IGMP_LOCK_ASSERT();
3225
3226 /*
3227 * If there are further pending retransmissions, make a writable
3228 * copy of each queued state-change message before merging.
3229 */
3230 if (inm->inm_scrv > 0)
3231 docopy = 1;
3232
3233 gq = &inm->inm_scq;
3234 #ifdef KTR
3235 if (mbufq_first(gq) == NULL) {
3236 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3237 __func__, inm);
3238 }
3239 #endif
3240
3241 m = mbufq_first(gq);
3242 while (m != NULL) {
3243 /*
3244 * Only merge the report into the current packet if
3245 * there is sufficient space to do so; an IGMPv3 report
3246 * packet may only contain 65,535 group records.
3247 * Always use a simple mbuf chain concatentation to do this,
3248 * as large state changes for single groups may have
3249 * allocated clusters.
3250 */
3251 domerge = 0;
3252 mt = mbufq_last(scq);
3253 if (mt != NULL) {
3254 recslen = m_length(m, NULL);
3255
3256 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3257 m->m_pkthdr.PH_vt.vt_nrecs <=
3258 IGMP_V3_REPORT_MAXRECS) &&
3259 (mt->m_pkthdr.len + recslen <=
3260 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3261 domerge = 1;
3262 }
3263
3264 if (!domerge && mbufq_full(gq)) {
3265 CTR2(KTR_IGMPV3,
3266 "%s: outbound queue full, skipping whole packet %p",
3267 __func__, m);
3268 mt = m->m_nextpkt;
3269 if (!docopy)
3270 m_freem(m);
3271 m = mt;
3272 continue;
3273 }
3274
3275 if (!docopy) {
3276 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3277 m0 = mbufq_dequeue(gq);
3278 m = m0->m_nextpkt;
3279 } else {
3280 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3281 m0 = m_dup(m, M_NOWAIT);
3282 if (m0 == NULL)
3283 return (ENOMEM);
3284 m0->m_nextpkt = NULL;
3285 m = m->m_nextpkt;
3286 }
3287
3288 if (!domerge) {
3289 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3290 __func__, m0, scq);
3291 mbufq_enqueue(scq, m0);
3292 } else {
3293 struct mbuf *mtl; /* last mbuf of packet mt */
3294
3295 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3296 __func__, m0, mt);
3297
3298 mtl = m_last(mt);
3299 m0->m_flags &= ~M_PKTHDR;
3300 mt->m_pkthdr.len += recslen;
3301 mt->m_pkthdr.PH_vt.vt_nrecs +=
3302 m0->m_pkthdr.PH_vt.vt_nrecs;
3303
3304 mtl->m_next = m0;
3305 }
3306 }
3307
3308 return (0);
3309 }
3310
3311 /*
3312 * Respond to a pending IGMPv3 General Query.
3313 */
3314 static void
3315 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3316 {
3317 struct ifmultiaddr *ifma;
3318 struct ifnet *ifp;
3319 struct in_multi *inm;
3320 int retval, loop;
3321
3322 IN_MULTI_LOCK_ASSERT();
3323 IGMP_LOCK_ASSERT();
3324
3325 KASSERT(igi->igi_version == IGMP_VERSION_3,
3326 ("%s: called when version %d", __func__, igi->igi_version));
3327
3328 /*
3329 * Check that there are some packets queued. If so, send them first.
3330 * For large number of groups the reply to general query can take
3331 * many packets, we should finish sending them before starting of
3332 * queuing the new reply.
3333 */
3334 if (mbufq_len(&igi->igi_gq) != 0)
3335 goto send;
3336
3337 ifp = igi->igi_ifp;
3338
3339 IF_ADDR_RLOCK(ifp);
3340 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3341 if (ifma->ifma_addr->sa_family != AF_INET ||
3342 ifma->ifma_protospec == NULL)
3343 continue;
3344
3345 inm = (struct in_multi *)ifma->ifma_protospec;
3346 KASSERT(ifp == inm->inm_ifp,
3347 ("%s: inconsistent ifp", __func__));
3348
3349 switch (inm->inm_state) {
3350 case IGMP_NOT_MEMBER:
3351 case IGMP_SILENT_MEMBER:
3352 break;
3353 case IGMP_REPORTING_MEMBER:
3354 case IGMP_IDLE_MEMBER:
3355 case IGMP_LAZY_MEMBER:
3356 case IGMP_SLEEPING_MEMBER:
3357 case IGMP_AWAKENING_MEMBER:
3358 inm->inm_state = IGMP_REPORTING_MEMBER;
3359 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3360 inm, 0, 0, 0);
3361 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3362 __func__, retval);
3363 break;
3364 case IGMP_G_QUERY_PENDING_MEMBER:
3365 case IGMP_SG_QUERY_PENDING_MEMBER:
3366 case IGMP_LEAVING_MEMBER:
3367 break;
3368 }
3369 }
3370 IF_ADDR_RUNLOCK(ifp);
3371
3372 send:
3373 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3374 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3375
3376 /*
3377 * Slew transmission of bursts over 500ms intervals.
3378 */
3379 if (mbufq_first(&igi->igi_gq) != NULL) {
3380 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3381 IGMP_RESPONSE_BURST_INTERVAL);
3382 V_interface_timers_running = 1;
3383 }
3384 }
3385
3386 /*
3387 * Transmit the next pending IGMP message in the output queue.
3388 *
3389 * We get called from netisr_processqueue(). A mutex private to igmpoq
3390 * will be acquired and released around this routine.
3391 *
3392 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3393 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3394 * a link and uses a link-scope multicast address.
3395 */
3396 static void
3397 igmp_intr(struct mbuf *m)
3398 {
3399 struct ip_moptions imo;
3400 struct ifnet *ifp;
3401 struct mbuf *ipopts, *m0;
3402 int error;
3403 uint32_t ifindex;
3404
3405 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3406
3407 /*
3408 * Set VNET image pointer from enqueued mbuf chain
3409 * before doing anything else. Whilst we use interface
3410 * indexes to guard against interface detach, they are
3411 * unique to each VIMAGE and must be retrieved.
3412 */
3413 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3414 ifindex = igmp_restore_context(m);
3415
3416 /*
3417 * Check if the ifnet still exists. This limits the scope of
3418 * any race in the absence of a global ifp lock for low cost
3419 * (an array lookup).
3420 */
3421 ifp = ifnet_byindex(ifindex);
3422 if (ifp == NULL) {
3423 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3424 __func__, m, ifindex);
3425 m_freem(m);
3426 IPSTAT_INC(ips_noroute);
3427 goto out;
3428 }
3429
3430 ipopts = V_igmp_sendra ? m_raopt : NULL;
3431
3432 imo.imo_multicast_ttl = 1;
3433 imo.imo_multicast_vif = -1;
3434 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3435
3436 /*
3437 * If the user requested that IGMP traffic be explicitly
3438 * redirected to the loopback interface (e.g. they are running a
3439 * MANET interface and the routing protocol needs to see the
3440 * updates), handle this now.
3441 */
3442 if (m->m_flags & M_IGMP_LOOP)
3443 imo.imo_multicast_ifp = V_loif;
3444 else
3445 imo.imo_multicast_ifp = ifp;
3446
3447 if (m->m_flags & M_IGMPV2) {
3448 m0 = m;
3449 } else {
3450 m0 = igmp_v3_encap_report(ifp, m);
3451 if (m0 == NULL) {
3452 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3453 m_freem(m);
3454 IPSTAT_INC(ips_odropped);
3455 goto out;
3456 }
3457 }
3458
3459 igmp_scrub_context(m0);
3460 m_clrprotoflags(m);
3461 m0->m_pkthdr.rcvif = V_loif;
3462 #ifdef MAC
3463 mac_netinet_igmp_send(ifp, m0);
3464 #endif
3465 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3466 if (error) {
3467 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3468 goto out;
3469 }
3470
3471 IGMPSTAT_INC(igps_snd_reports);
3472
3473 out:
3474 /*
3475 * We must restore the existing vnet pointer before
3476 * continuing as we are run from netisr context.
3477 */
3478 CURVNET_RESTORE();
3479 }
3480
3481 /*
3482 * Encapsulate an IGMPv3 report.
3483 *
3484 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3485 * chain has already had its IP/IGMPv3 header prepended. In this case
3486 * the function will not attempt to prepend; the lengths and checksums
3487 * will however be re-computed.
3488 *
3489 * Returns a pointer to the new mbuf chain head, or NULL if the
3490 * allocation failed.
3491 */
3492 static struct mbuf *
3493 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3494 {
3495 struct rm_priotracker in_ifa_tracker;
3496 struct igmp_report *igmp;
3497 struct ip *ip;
3498 int hdrlen, igmpreclen;
3499
3500 KASSERT((m->m_flags & M_PKTHDR),
3501 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3502
3503 igmpreclen = m_length(m, NULL);
3504 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3505
3506 if (m->m_flags & M_IGMPV3_HDR) {
3507 igmpreclen -= hdrlen;
3508 } else {
3509 M_PREPEND(m, hdrlen, M_NOWAIT);
3510 if (m == NULL)
3511 return (NULL);
3512 m->m_flags |= M_IGMPV3_HDR;
3513 }
3514
3515 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3516
3517 m->m_data += sizeof(struct ip);
3518 m->m_len -= sizeof(struct ip);
3519
3520 igmp = mtod(m, struct igmp_report *);
3521 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3522 igmp->ir_rsv1 = 0;
3523 igmp->ir_rsv2 = 0;
3524 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3525 igmp->ir_cksum = 0;
3526 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3527 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3528
3529 m->m_data -= sizeof(struct ip);
3530 m->m_len += sizeof(struct ip);
3531
3532 ip = mtod(m, struct ip *);
3533 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3534 ip->ip_len = htons(hdrlen + igmpreclen);
3535 ip->ip_off = htons(IP_DF);
3536 ip->ip_p = IPPROTO_IGMP;
3537 ip->ip_sum = 0;
3538
3539 ip->ip_src.s_addr = INADDR_ANY;
3540
3541 if (m->m_flags & M_IGMP_LOOP) {
3542 struct in_ifaddr *ia;
3543
3544 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3545 if (ia != NULL) {
3546 ip->ip_src = ia->ia_addr.sin_addr;
3547 ifa_free(&ia->ia_ifa);
3548 }
3549 }
3550
3551 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3552
3553 return (m);
3554 }
3555
3556 #ifdef KTR
3557 static char *
3558 igmp_rec_type_to_str(const int type)
3559 {
3560
3561 switch (type) {
3562 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3563 return "TO_EX";
3564 break;
3565 case IGMP_CHANGE_TO_INCLUDE_MODE:
3566 return "TO_IN";
3567 break;
3568 case IGMP_MODE_IS_EXCLUDE:
3569 return "MODE_EX";
3570 break;
3571 case IGMP_MODE_IS_INCLUDE:
3572 return "MODE_IN";
3573 break;
3574 case IGMP_ALLOW_NEW_SOURCES:
3575 return "ALLOW_NEW";
3576 break;
3577 case IGMP_BLOCK_OLD_SOURCES:
3578 return "BLOCK_OLD";
3579 break;
3580 default:
3581 break;
3582 }
3583 return "unknown";
3584 }
3585 #endif
3586
3587 #ifdef VIMAGE
3588 static void
3589 vnet_igmp_init(const void *unused __unused)
3590 {
3591
3592 netisr_register_vnet(&igmp_nh);
3593 }
3594 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3595 vnet_igmp_init, NULL);
3596
3597 static void
3598 vnet_igmp_uninit(const void *unused __unused)
3599 {
3600
3601 /* This can happen when we shutdown the entire network stack. */
3602 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3603
3604 netisr_unregister_vnet(&igmp_nh);
3605 }
3606 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3607 vnet_igmp_uninit, NULL);
3608 #endif
3609
3610 #ifdef DDB
3611 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3612 {
3613 struct igmp_ifsoftc *igi, *tigi;
3614 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3615
3616 if (!have_addr) {
3617 db_printf("usage: show igi_list <addr>\n");
3618 return;
3619 }
3620 igi_head = (struct _igi_list *)addr;
3621
3622 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3623 db_printf("igmp_ifsoftc %p:\n", igi);
3624 db_printf(" ifp %p\n", igi->igi_ifp);
3625 db_printf(" version %u\n", igi->igi_version);
3626 db_printf(" v1_timer %u\n", igi->igi_v1_timer);
3627 db_printf(" v2_timer %u\n", igi->igi_v2_timer);
3628 db_printf(" v3_timer %u\n", igi->igi_v3_timer);
3629 db_printf(" flags %#x\n", igi->igi_flags);
3630 db_printf(" rv %u\n", igi->igi_rv);
3631 db_printf(" qi %u\n", igi->igi_qi);
3632 db_printf(" qri %u\n", igi->igi_qri);
3633 db_printf(" uri %u\n", igi->igi_uri);
3634 /* SLIST_HEAD(,in_multi) igi_relinmhead */
3635 /* struct mbufq igi_gq; */
3636 db_printf("\n");
3637 }
3638 }
3639 #endif
3640
3641 static int
3642 igmp_modevent(module_t mod, int type, void *unused __unused)
3643 {
3644
3645 switch (type) {
3646 case MOD_LOAD:
3647 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3648 IGMP_LOCK_INIT();
3649 m_raopt = igmp_ra_alloc();
3650 netisr_register(&igmp_nh);
3651 break;
3652 case MOD_UNLOAD:
3653 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3654 netisr_unregister(&igmp_nh);
3655 m_free(m_raopt);
3656 m_raopt = NULL;
3657 IGMP_LOCK_DESTROY();
3658 break;
3659 default:
3660 return (EOPNOTSUPP);
3661 }
3662 return (0);
3663 }
3664
3665 static moduledata_t igmp_mod = {
3666 "igmp",
3667 igmp_modevent,
3668 0
3669 };
3670 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);
Cache object: a0c9245449baa51cfba6337fa3aa1ad9
|