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: releng/8.1/sys/netinet/igmp.c 207558 2010-05-03 09:31:51Z bms $");
52
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/module.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/socket.h>
59 #include <sys/protosw.h>
60 #include <sys/kernel.h>
61 #include <sys/sysctl.h>
62 #include <sys/ktr.h>
63 #include <sys/condvar.h>
64
65 #include <net/if.h>
66 #include <net/netisr.h>
67 #include <net/vnet.h>
68
69 #include <netinet/in.h>
70 #include <netinet/in_var.h>
71 #include <netinet/in_systm.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_var.h>
74 #include <netinet/ip_options.h>
75 #include <netinet/igmp.h>
76 #include <netinet/igmp_var.h>
77
78 #include <machine/in_cksum.h>
79
80 #include <security/mac/mac_framework.h>
81
82 #ifndef KTR_IGMPV3
83 #define KTR_IGMPV3 KTR_INET
84 #endif
85
86 static struct igmp_ifinfo *
87 igi_alloc_locked(struct ifnet *);
88 static void igi_delete_locked(const struct ifnet *);
89 static void igmp_dispatch_queue(struct ifqueue *, int, const int);
90 static void igmp_fasttimo_vnet(void);
91 static void igmp_final_leave(struct in_multi *, struct igmp_ifinfo *);
92 static int igmp_handle_state_change(struct in_multi *,
93 struct igmp_ifinfo *);
94 static int igmp_initial_join(struct in_multi *, struct igmp_ifinfo *);
95 static int igmp_input_v1_query(struct ifnet *, const struct ip *,
96 const struct igmp *);
97 static int igmp_input_v2_query(struct ifnet *, const struct ip *,
98 const struct igmp *);
99 static int igmp_input_v3_query(struct ifnet *, const struct ip *,
100 /*const*/ struct igmpv3 *);
101 static int igmp_input_v3_group_query(struct in_multi *,
102 struct igmp_ifinfo *, int, /*const*/ struct igmpv3 *);
103 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
104 /*const*/ struct igmp *);
105 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
106 /*const*/ struct igmp *);
107 static void igmp_intr(struct mbuf *);
108 static int igmp_isgroupreported(const struct in_addr);
109 static struct mbuf *
110 igmp_ra_alloc(void);
111 #ifdef KTR
112 static char * igmp_rec_type_to_str(const int);
113 #endif
114 static void igmp_set_version(struct igmp_ifinfo *, const int);
115 static void igmp_slowtimo_vnet(void);
116 static int igmp_v1v2_queue_report(struct in_multi *, const int);
117 static void igmp_v1v2_process_group_timer(struct in_multi *, const int);
118 static void igmp_v1v2_process_querier_timers(struct igmp_ifinfo *);
119 static void igmp_v2_update_group(struct in_multi *, const int);
120 static void igmp_v3_cancel_link_timers(struct igmp_ifinfo *);
121 static void igmp_v3_dispatch_general_query(struct igmp_ifinfo *);
122 static struct mbuf *
123 igmp_v3_encap_report(struct ifnet *, struct mbuf *);
124 static int igmp_v3_enqueue_group_record(struct ifqueue *,
125 struct in_multi *, const int, const int, const int);
126 static int igmp_v3_enqueue_filter_change(struct ifqueue *,
127 struct in_multi *);
128 static void igmp_v3_process_group_timers(struct igmp_ifinfo *,
129 struct ifqueue *, struct ifqueue *, struct in_multi *,
130 const int);
131 static int igmp_v3_merge_state_changes(struct in_multi *,
132 struct ifqueue *);
133 static void igmp_v3_suppress_group_record(struct in_multi *);
134 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
135 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
136 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
137
138 static const struct netisr_handler igmp_nh = {
139 .nh_name = "igmp",
140 .nh_handler = igmp_intr,
141 .nh_proto = NETISR_IGMP,
142 .nh_policy = NETISR_POLICY_SOURCE,
143 };
144
145 /*
146 * System-wide globals.
147 *
148 * Unlocked access to these is OK, except for the global IGMP output
149 * queue. The IGMP subsystem lock ends up being system-wide for the moment,
150 * because all VIMAGEs have to share a global output queue, as netisrs
151 * themselves are not virtualized.
152 *
153 * Locking:
154 * * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
155 * Any may be taken independently; if any are held at the same
156 * time, the above lock order must be followed.
157 * * All output is delegated to the netisr.
158 * Now that Giant has been eliminated, the netisr may be inlined.
159 * * IN_MULTI_LOCK covers in_multi.
160 * * IGMP_LOCK covers igmp_ifinfo and any global variables in this file,
161 * including the output queue.
162 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
163 * per-link state iterators.
164 * * igmp_ifinfo is valid as long as PF_INET is attached to the interface,
165 * therefore it is not refcounted.
166 * We allow unlocked reads of igmp_ifinfo when accessed via in_multi.
167 *
168 * Reference counting
169 * * IGMP acquires its own reference every time an in_multi is passed to
170 * it and the group is being joined for the first time.
171 * * IGMP releases its reference(s) on in_multi in a deferred way,
172 * because the operations which process the release run as part of
173 * a loop whose control variables are directly affected by the release
174 * (that, and not recursing on the IF_ADDR_LOCK).
175 *
176 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
177 * to a vnet in ifp->if_vnet.
178 *
179 * SMPng: XXX We may potentially race operations on ifma_protospec.
180 * The problem is that we currently lack a clean way of taking the
181 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
182 * as anything which modifies ifma needs to be covered by that lock.
183 * So check for ifma_protospec being NULL before proceeding.
184 */
185 struct mtx igmp_mtx;
186
187 struct mbuf *m_raopt; /* Router Alert option */
188 MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
189
190 /*
191 * VIMAGE-wide globals.
192 *
193 * The IGMPv3 timers themselves need to run per-image, however,
194 * protosw timers run globally (see tcp).
195 * An ifnet can only be in one vimage at a time, and the loopback
196 * ifnet, loif, is itself virtualized.
197 * It would otherwise be possible to seriously hose IGMP state,
198 * and create inconsistencies in upstream multicast routing, if you have
199 * multiple VIMAGEs running on the same link joining different multicast
200 * groups, UNLESS the "primary IP address" is different. This is because
201 * IGMP for IPv4 does not force link-local addresses to be used for each
202 * node, unlike MLD for IPv6.
203 * Obviously the IGMPv3 per-interface state has per-vimage granularity
204 * also as a result.
205 *
206 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
207 * policy to control the address used by IGMP on the link.
208 */
209 static VNET_DEFINE(int, interface_timers_running); /* IGMPv3 general
210 * query response */
211 static VNET_DEFINE(int, state_change_timers_running); /* IGMPv3 state-change
212 * retransmit */
213 static VNET_DEFINE(int, current_state_timers_running); /* IGMPv1/v2 host
214 * report; IGMPv3 g/sg
215 * query response */
216
217 #define V_interface_timers_running VNET(interface_timers_running)
218 #define V_state_change_timers_running VNET(state_change_timers_running)
219 #define V_current_state_timers_running VNET(current_state_timers_running)
220
221 static VNET_DEFINE(LIST_HEAD(, igmp_ifinfo), igi_head);
222 static VNET_DEFINE(struct igmpstat, igmpstat) = {
223 .igps_version = IGPS_VERSION_3,
224 .igps_len = sizeof(struct igmpstat),
225 };
226 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
227
228 #define V_igi_head VNET(igi_head)
229 #define V_igmpstat VNET(igmpstat)
230 #define V_igmp_gsrdelay VNET(igmp_gsrdelay)
231
232 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
233 static VNET_DEFINE(int, igmp_sendra) = 1;
234 static VNET_DEFINE(int, igmp_sendlocal) = 1;
235 static VNET_DEFINE(int, igmp_v1enable) = 1;
236 static VNET_DEFINE(int, igmp_v2enable) = 1;
237 static VNET_DEFINE(int, igmp_legacysupp);
238 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
239
240 #define V_igmp_recvifkludge VNET(igmp_recvifkludge)
241 #define V_igmp_sendra VNET(igmp_sendra)
242 #define V_igmp_sendlocal VNET(igmp_sendlocal)
243 #define V_igmp_v1enable VNET(igmp_v1enable)
244 #define V_igmp_v2enable VNET(igmp_v2enable)
245 #define V_igmp_legacysupp VNET(igmp_legacysupp)
246 #define V_igmp_default_version VNET(igmp_default_version)
247
248 /*
249 * Virtualized sysctls.
250 */
251 SYSCTL_VNET_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_RW,
252 &VNET_NAME(igmpstat), igmpstat, "");
253 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_RW,
254 &VNET_NAME(igmp_recvifkludge), 0,
255 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
256 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_RW,
257 &VNET_NAME(igmp_sendra), 0,
258 "Send IP Router Alert option in IGMPv2/v3 messages");
259 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_RW,
260 &VNET_NAME(igmp_sendlocal), 0,
261 "Send IGMP membership reports for 224.0.0.0/24 groups");
262 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_RW,
263 &VNET_NAME(igmp_v1enable), 0,
264 "Enable backwards compatibility with IGMPv1");
265 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_RW,
266 &VNET_NAME(igmp_v2enable), 0,
267 "Enable backwards compatibility with IGMPv2");
268 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_RW,
269 &VNET_NAME(igmp_legacysupp), 0,
270 "Allow v1/v2 reports to suppress v3 group responses");
271 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, default_version,
272 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
273 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
274 "Default version of IGMP to run on each interface");
275 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
276 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
277 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
278 "Rate limit for IGMPv3 Group-and-Source queries in seconds");
279
280 /*
281 * Non-virtualized sysctls.
282 */
283 SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
284 sysctl_igmp_ifinfo, "Per-interface IGMPv3 state");
285
286 static __inline void
287 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
288 {
289
290 #ifdef VIMAGE
291 m->m_pkthdr.header = ifp->if_vnet;
292 #endif /* VIMAGE */
293 m->m_pkthdr.flowid = ifp->if_index;
294 }
295
296 static __inline void
297 igmp_scrub_context(struct mbuf *m)
298 {
299
300 m->m_pkthdr.header = NULL;
301 m->m_pkthdr.flowid = 0;
302 }
303
304 #ifdef KTR
305 static __inline char *
306 inet_ntoa_haddr(in_addr_t haddr)
307 {
308 struct in_addr ia;
309
310 ia.s_addr = htonl(haddr);
311 return (inet_ntoa(ia));
312 }
313 #endif
314
315 /*
316 * Restore context from a queued IGMP output chain.
317 * Return saved ifindex.
318 *
319 * VIMAGE: The assertion is there to make sure that we
320 * actually called CURVNET_SET() with what's in the mbuf chain.
321 */
322 static __inline uint32_t
323 igmp_restore_context(struct mbuf *m)
324 {
325
326 #ifdef notyet
327 #if defined(VIMAGE) && defined(INVARIANTS)
328 KASSERT(curvnet == (m->m_pkthdr.header),
329 ("%s: called when curvnet was not restored", __func__));
330 #endif
331 #endif
332 return (m->m_pkthdr.flowid);
333 }
334
335 /*
336 * Retrieve or set default IGMP version.
337 *
338 * VIMAGE: Assume curvnet set by caller.
339 * SMPng: NOTE: Serialized by IGMP lock.
340 */
341 static int
342 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
343 {
344 int error;
345 int new;
346
347 error = sysctl_wire_old_buffer(req, sizeof(int));
348 if (error)
349 return (error);
350
351 IGMP_LOCK();
352
353 new = V_igmp_default_version;
354
355 error = sysctl_handle_int(oidp, &new, 0, req);
356 if (error || !req->newptr)
357 goto out_locked;
358
359 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
360 error = EINVAL;
361 goto out_locked;
362 }
363
364 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
365 V_igmp_default_version, new);
366
367 V_igmp_default_version = new;
368
369 out_locked:
370 IGMP_UNLOCK();
371 return (error);
372 }
373
374 /*
375 * Retrieve or set threshold between group-source queries in seconds.
376 *
377 * VIMAGE: Assume curvnet set by caller.
378 * SMPng: NOTE: Serialized by IGMP lock.
379 */
380 static int
381 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
382 {
383 int error;
384 int i;
385
386 error = sysctl_wire_old_buffer(req, sizeof(int));
387 if (error)
388 return (error);
389
390 IGMP_LOCK();
391
392 i = V_igmp_gsrdelay.tv_sec;
393
394 error = sysctl_handle_int(oidp, &i, 0, req);
395 if (error || !req->newptr)
396 goto out_locked;
397
398 if (i < -1 || i >= 60) {
399 error = EINVAL;
400 goto out_locked;
401 }
402
403 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
404 V_igmp_gsrdelay.tv_sec, i);
405 V_igmp_gsrdelay.tv_sec = i;
406
407 out_locked:
408 IGMP_UNLOCK();
409 return (error);
410 }
411
412 /*
413 * Expose struct igmp_ifinfo to userland, keyed by ifindex.
414 * For use by ifmcstat(8).
415 *
416 * SMPng: NOTE: Does an unlocked ifindex space read.
417 * VIMAGE: Assume curvnet set by caller. The node handler itself
418 * is not directly virtualized.
419 */
420 static int
421 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
422 {
423 int *name;
424 int error;
425 u_int namelen;
426 struct ifnet *ifp;
427 struct igmp_ifinfo *igi;
428
429 name = (int *)arg1;
430 namelen = arg2;
431
432 if (req->newptr != NULL)
433 return (EPERM);
434
435 if (namelen != 1)
436 return (EINVAL);
437
438 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
439 if (error)
440 return (error);
441
442 IN_MULTI_LOCK();
443 IGMP_LOCK();
444
445 if (name[0] <= 0 || name[0] > V_if_index) {
446 error = ENOENT;
447 goto out_locked;
448 }
449
450 error = ENOENT;
451
452 ifp = ifnet_byindex(name[0]);
453 if (ifp == NULL)
454 goto out_locked;
455
456 LIST_FOREACH(igi, &V_igi_head, igi_link) {
457 if (ifp == igi->igi_ifp) {
458 error = SYSCTL_OUT(req, igi,
459 sizeof(struct igmp_ifinfo));
460 break;
461 }
462 }
463
464 out_locked:
465 IGMP_UNLOCK();
466 IN_MULTI_UNLOCK();
467 return (error);
468 }
469
470 /*
471 * Dispatch an entire queue of pending packet chains
472 * using the netisr.
473 * VIMAGE: Assumes the vnet pointer has been set.
474 */
475 static void
476 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop)
477 {
478 struct mbuf *m;
479
480 for (;;) {
481 _IF_DEQUEUE(ifq, m);
482 if (m == NULL)
483 break;
484 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m);
485 if (loop)
486 m->m_flags |= M_IGMP_LOOP;
487 netisr_dispatch(NETISR_IGMP, m);
488 if (--limit == 0)
489 break;
490 }
491 }
492
493 /*
494 * Filter outgoing IGMP report state by group.
495 *
496 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
497 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
498 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
499 * this may break certain IGMP snooping switches which rely on the old
500 * report behaviour.
501 *
502 * Return zero if the given group is one for which IGMP reports
503 * should be suppressed, or non-zero if reports should be issued.
504 */
505 static __inline int
506 igmp_isgroupreported(const struct in_addr addr)
507 {
508
509 if (in_allhosts(addr) ||
510 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
511 return (0);
512
513 return (1);
514 }
515
516 /*
517 * Construct a Router Alert option to use in outgoing packets.
518 */
519 static struct mbuf *
520 igmp_ra_alloc(void)
521 {
522 struct mbuf *m;
523 struct ipoption *p;
524
525 MGET(m, M_DONTWAIT, MT_DATA);
526 p = mtod(m, struct ipoption *);
527 p->ipopt_dst.s_addr = INADDR_ANY;
528 p->ipopt_list[0] = IPOPT_RA; /* Router Alert Option */
529 p->ipopt_list[1] = 0x04; /* 4 bytes long */
530 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
531 p->ipopt_list[3] = 0x00; /* pad byte */
532 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
533
534 return (m);
535 }
536
537 /*
538 * Attach IGMP when PF_INET is attached to an interface.
539 */
540 struct igmp_ifinfo *
541 igmp_domifattach(struct ifnet *ifp)
542 {
543 struct igmp_ifinfo *igi;
544
545 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
546 __func__, ifp, ifp->if_xname);
547
548 IGMP_LOCK();
549
550 igi = igi_alloc_locked(ifp);
551 if (!(ifp->if_flags & IFF_MULTICAST))
552 igi->igi_flags |= IGIF_SILENT;
553
554 IGMP_UNLOCK();
555
556 return (igi);
557 }
558
559 /*
560 * VIMAGE: assume curvnet set by caller.
561 */
562 static struct igmp_ifinfo *
563 igi_alloc_locked(/*const*/ struct ifnet *ifp)
564 {
565 struct igmp_ifinfo *igi;
566
567 IGMP_LOCK_ASSERT();
568
569 igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO);
570 if (igi == NULL)
571 goto out;
572
573 igi->igi_ifp = ifp;
574 igi->igi_version = V_igmp_default_version;
575 igi->igi_flags = 0;
576 igi->igi_rv = IGMP_RV_INIT;
577 igi->igi_qi = IGMP_QI_INIT;
578 igi->igi_qri = IGMP_QRI_INIT;
579 igi->igi_uri = IGMP_URI_INIT;
580
581 SLIST_INIT(&igi->igi_relinmhead);
582
583 /*
584 * Responses to general queries are subject to bounds.
585 */
586 IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
587
588 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
589
590 CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)",
591 ifp, ifp->if_xname);
592
593 out:
594 return (igi);
595 }
596
597 /*
598 * Hook for ifdetach.
599 *
600 * NOTE: Some finalization tasks need to run before the protocol domain
601 * is detached, but also before the link layer does its cleanup.
602 *
603 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
604 * XXX This is also bitten by unlocked ifma_protospec access.
605 */
606 void
607 igmp_ifdetach(struct ifnet *ifp)
608 {
609 struct igmp_ifinfo *igi;
610 struct ifmultiaddr *ifma;
611 struct in_multi *inm, *tinm;
612
613 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
614 ifp->if_xname);
615
616 IGMP_LOCK();
617
618 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
619 if (igi->igi_version == IGMP_VERSION_3) {
620 IF_ADDR_LOCK(ifp);
621 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
622 if (ifma->ifma_addr->sa_family != AF_INET ||
623 ifma->ifma_protospec == NULL)
624 continue;
625 #if 0
626 KASSERT(ifma->ifma_protospec != NULL,
627 ("%s: ifma_protospec is NULL", __func__));
628 #endif
629 inm = (struct in_multi *)ifma->ifma_protospec;
630 if (inm->inm_state == IGMP_LEAVING_MEMBER) {
631 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
632 inm, inm_nrele);
633 }
634 inm_clear_recorded(inm);
635 }
636 IF_ADDR_UNLOCK(ifp);
637 /*
638 * Free the in_multi reference(s) for this IGMP lifecycle.
639 */
640 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
641 tinm) {
642 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
643 inm_release_locked(inm);
644 }
645 }
646
647 IGMP_UNLOCK();
648 }
649
650 /*
651 * Hook for domifdetach.
652 */
653 void
654 igmp_domifdetach(struct ifnet *ifp)
655 {
656 struct igmp_ifinfo *igi;
657
658 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
659 __func__, ifp, ifp->if_xname);
660
661 IGMP_LOCK();
662
663 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
664 igi_delete_locked(ifp);
665
666 IGMP_UNLOCK();
667 }
668
669 static void
670 igi_delete_locked(const struct ifnet *ifp)
671 {
672 struct igmp_ifinfo *igi, *tigi;
673
674 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
675 __func__, ifp, ifp->if_xname);
676
677 IGMP_LOCK_ASSERT();
678
679 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
680 if (igi->igi_ifp == ifp) {
681 /*
682 * Free deferred General Query responses.
683 */
684 _IF_DRAIN(&igi->igi_gq);
685
686 LIST_REMOVE(igi, igi_link);
687
688 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
689 ("%s: there are dangling in_multi references",
690 __func__));
691
692 free(igi, M_IGMP);
693 return;
694 }
695 }
696
697 #ifdef INVARIANTS
698 panic("%s: igmp_ifinfo not found for ifp %p\n", __func__, ifp);
699 #endif
700 }
701
702 /*
703 * Process a received IGMPv1 query.
704 * Return non-zero if the message should be dropped.
705 *
706 * VIMAGE: The curvnet pointer is derived from the input ifp.
707 */
708 static int
709 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
710 const struct igmp *igmp)
711 {
712 struct ifmultiaddr *ifma;
713 struct igmp_ifinfo *igi;
714 struct in_multi *inm;
715
716 /*
717 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
718 * 224.0.0.1. They are always treated as General Queries.
719 * igmp_group is always ignored. Do not drop it as a userland
720 * daemon may wish to see it.
721 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
722 */
723 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
724 IGMPSTAT_INC(igps_rcv_badqueries);
725 return (0);
726 }
727 IGMPSTAT_INC(igps_rcv_gen_queries);
728
729 IN_MULTI_LOCK();
730 IGMP_LOCK();
731
732 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
733 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
734
735 if (igi->igi_flags & IGIF_LOOPBACK) {
736 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
737 ifp, ifp->if_xname);
738 goto out_locked;
739 }
740
741 /*
742 * Switch to IGMPv1 host compatibility mode.
743 */
744 igmp_set_version(igi, IGMP_VERSION_1);
745
746 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
747
748 /*
749 * Start the timers in all of our group records
750 * for the interface on which the query arrived,
751 * except those which are already running.
752 */
753 IF_ADDR_LOCK(ifp);
754 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
755 if (ifma->ifma_addr->sa_family != AF_INET ||
756 ifma->ifma_protospec == NULL)
757 continue;
758 inm = (struct in_multi *)ifma->ifma_protospec;
759 if (inm->inm_timer != 0)
760 continue;
761 switch (inm->inm_state) {
762 case IGMP_NOT_MEMBER:
763 case IGMP_SILENT_MEMBER:
764 break;
765 case IGMP_G_QUERY_PENDING_MEMBER:
766 case IGMP_SG_QUERY_PENDING_MEMBER:
767 case IGMP_REPORTING_MEMBER:
768 case IGMP_IDLE_MEMBER:
769 case IGMP_LAZY_MEMBER:
770 case IGMP_SLEEPING_MEMBER:
771 case IGMP_AWAKENING_MEMBER:
772 inm->inm_state = IGMP_REPORTING_MEMBER;
773 inm->inm_timer = IGMP_RANDOM_DELAY(
774 IGMP_V1V2_MAX_RI * PR_FASTHZ);
775 V_current_state_timers_running = 1;
776 break;
777 case IGMP_LEAVING_MEMBER:
778 break;
779 }
780 }
781 IF_ADDR_UNLOCK(ifp);
782
783 out_locked:
784 IGMP_UNLOCK();
785 IN_MULTI_UNLOCK();
786
787 return (0);
788 }
789
790 /*
791 * Process a received IGMPv2 general or group-specific query.
792 */
793 static int
794 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
795 const struct igmp *igmp)
796 {
797 struct ifmultiaddr *ifma;
798 struct igmp_ifinfo *igi;
799 struct in_multi *inm;
800 int is_general_query;
801 uint16_t timer;
802
803 is_general_query = 0;
804
805 /*
806 * Validate address fields upfront.
807 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
808 */
809 if (in_nullhost(igmp->igmp_group)) {
810 /*
811 * IGMPv2 General Query.
812 * If this was not sent to the all-hosts group, ignore it.
813 */
814 if (!in_allhosts(ip->ip_dst))
815 return (0);
816 IGMPSTAT_INC(igps_rcv_gen_queries);
817 is_general_query = 1;
818 } else {
819 /* IGMPv2 Group-Specific Query. */
820 IGMPSTAT_INC(igps_rcv_group_queries);
821 }
822
823 IN_MULTI_LOCK();
824 IGMP_LOCK();
825
826 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
827 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
828
829 if (igi->igi_flags & IGIF_LOOPBACK) {
830 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
831 ifp, ifp->if_xname);
832 goto out_locked;
833 }
834
835 /*
836 * Ignore v2 query if in v1 Compatibility Mode.
837 */
838 if (igi->igi_version == IGMP_VERSION_1)
839 goto out_locked;
840
841 igmp_set_version(igi, IGMP_VERSION_2);
842
843 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
844 if (timer == 0)
845 timer = 1;
846
847 if (is_general_query) {
848 /*
849 * For each reporting group joined on this
850 * interface, kick the report timer.
851 */
852 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
853 ifp, ifp->if_xname);
854 IF_ADDR_LOCK(ifp);
855 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
856 if (ifma->ifma_addr->sa_family != AF_INET ||
857 ifma->ifma_protospec == NULL)
858 continue;
859 inm = (struct in_multi *)ifma->ifma_protospec;
860 igmp_v2_update_group(inm, timer);
861 }
862 IF_ADDR_UNLOCK(ifp);
863 } else {
864 /*
865 * Group-specific IGMPv2 query, we need only
866 * look up the single group to process it.
867 */
868 inm = inm_lookup(ifp, igmp->igmp_group);
869 if (inm != NULL) {
870 CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
871 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
872 igmp_v2_update_group(inm, timer);
873 }
874 }
875
876 out_locked:
877 IGMP_UNLOCK();
878 IN_MULTI_UNLOCK();
879
880 return (0);
881 }
882
883 /*
884 * Update the report timer on a group in response to an IGMPv2 query.
885 *
886 * If we are becoming the reporting member for this group, start the timer.
887 * If we already are the reporting member for this group, and timer is
888 * below the threshold, reset it.
889 *
890 * We may be updating the group for the first time since we switched
891 * to IGMPv3. If we are, then we must clear any recorded source lists,
892 * and transition to REPORTING state; the group timer is overloaded
893 * for group and group-source query responses.
894 *
895 * Unlike IGMPv3, the delay per group should be jittered
896 * to avoid bursts of IGMPv2 reports.
897 */
898 static void
899 igmp_v2_update_group(struct in_multi *inm, const int timer)
900 {
901
902 CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
903 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
904
905 IN_MULTI_LOCK_ASSERT();
906
907 switch (inm->inm_state) {
908 case IGMP_NOT_MEMBER:
909 case IGMP_SILENT_MEMBER:
910 break;
911 case IGMP_REPORTING_MEMBER:
912 if (inm->inm_timer != 0 &&
913 inm->inm_timer <= timer) {
914 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
915 "skipping.", __func__);
916 break;
917 }
918 /* FALLTHROUGH */
919 case IGMP_SG_QUERY_PENDING_MEMBER:
920 case IGMP_G_QUERY_PENDING_MEMBER:
921 case IGMP_IDLE_MEMBER:
922 case IGMP_LAZY_MEMBER:
923 case IGMP_AWAKENING_MEMBER:
924 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
925 inm->inm_state = IGMP_REPORTING_MEMBER;
926 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
927 V_current_state_timers_running = 1;
928 break;
929 case IGMP_SLEEPING_MEMBER:
930 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
931 inm->inm_state = IGMP_AWAKENING_MEMBER;
932 break;
933 case IGMP_LEAVING_MEMBER:
934 break;
935 }
936 }
937
938 /*
939 * Process a received IGMPv3 general, group-specific or
940 * group-and-source-specific query.
941 * Assumes m has already been pulled up to the full IGMP message length.
942 * Return 0 if successful, otherwise an appropriate error code is returned.
943 */
944 static int
945 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
946 /*const*/ struct igmpv3 *igmpv3)
947 {
948 struct igmp_ifinfo *igi;
949 struct in_multi *inm;
950 int is_general_query;
951 uint32_t maxresp, nsrc, qqi;
952 uint16_t timer;
953 uint8_t qrv;
954
955 is_general_query = 0;
956
957 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
958
959 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
960 if (maxresp >= 128) {
961 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
962 (IGMP_EXP(igmpv3->igmp_code) + 3);
963 }
964
965 /*
966 * Robustness must never be less than 2 for on-wire IGMPv3.
967 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
968 * an exception for interfaces whose IGMPv3 state changes
969 * are redirected to loopback (e.g. MANET).
970 */
971 qrv = IGMP_QRV(igmpv3->igmp_misc);
972 if (qrv < 2) {
973 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
974 qrv, IGMP_RV_INIT);
975 qrv = IGMP_RV_INIT;
976 }
977
978 qqi = igmpv3->igmp_qqi;
979 if (qqi >= 128) {
980 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
981 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
982 }
983
984 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
985 if (timer == 0)
986 timer = 1;
987
988 nsrc = ntohs(igmpv3->igmp_numsrc);
989
990 /*
991 * Validate address fields and versions upfront before
992 * accepting v3 query.
993 * XXX SMPng: Unlocked access to igmpstat counters here.
994 */
995 if (in_nullhost(igmpv3->igmp_group)) {
996 /*
997 * IGMPv3 General Query.
998 *
999 * General Queries SHOULD be directed to 224.0.0.1.
1000 * A general query with a source list has undefined
1001 * behaviour; discard it.
1002 */
1003 IGMPSTAT_INC(igps_rcv_gen_queries);
1004 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1005 IGMPSTAT_INC(igps_rcv_badqueries);
1006 return (0);
1007 }
1008 is_general_query = 1;
1009 } else {
1010 /* Group or group-source specific query. */
1011 if (nsrc == 0)
1012 IGMPSTAT_INC(igps_rcv_group_queries);
1013 else
1014 IGMPSTAT_INC(igps_rcv_gsr_queries);
1015 }
1016
1017 IN_MULTI_LOCK();
1018 IGMP_LOCK();
1019
1020 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1021 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
1022
1023 if (igi->igi_flags & IGIF_LOOPBACK) {
1024 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1025 ifp, ifp->if_xname);
1026 goto out_locked;
1027 }
1028
1029 /*
1030 * Discard the v3 query if we're in Compatibility Mode.
1031 * The RFC is not obviously worded that hosts need to stay in
1032 * compatibility mode until the Old Version Querier Present
1033 * timer expires.
1034 */
1035 if (igi->igi_version != IGMP_VERSION_3) {
1036 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1037 igi->igi_version, ifp, ifp->if_xname);
1038 goto out_locked;
1039 }
1040
1041 igmp_set_version(igi, IGMP_VERSION_3);
1042 igi->igi_rv = qrv;
1043 igi->igi_qi = qqi;
1044 igi->igi_qri = maxresp;
1045
1046 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1047 maxresp);
1048
1049 if (is_general_query) {
1050 /*
1051 * Schedule a current-state report on this ifp for
1052 * all groups, possibly containing source lists.
1053 * If there is a pending General Query response
1054 * scheduled earlier than the selected delay, do
1055 * not schedule any other reports.
1056 * Otherwise, reset the interface timer.
1057 */
1058 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1059 ifp, ifp->if_xname);
1060 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1061 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1062 V_interface_timers_running = 1;
1063 }
1064 } else {
1065 /*
1066 * Group-source-specific queries are throttled on
1067 * a per-group basis to defeat denial-of-service attempts.
1068 * Queries for groups we are not a member of on this
1069 * link are simply ignored.
1070 */
1071 inm = inm_lookup(ifp, igmpv3->igmp_group);
1072 if (inm == NULL)
1073 goto out_locked;
1074 if (nsrc > 0) {
1075 if (!ratecheck(&inm->inm_lastgsrtv,
1076 &V_igmp_gsrdelay)) {
1077 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1078 __func__);
1079 IGMPSTAT_INC(igps_drop_gsr_queries);
1080 goto out_locked;
1081 }
1082 }
1083 CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1084 inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1085 /*
1086 * If there is a pending General Query response
1087 * scheduled sooner than the selected delay, no
1088 * further report need be scheduled.
1089 * Otherwise, prepare to respond to the
1090 * group-specific or group-and-source query.
1091 */
1092 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1093 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1094 }
1095
1096 out_locked:
1097 IGMP_UNLOCK();
1098 IN_MULTI_UNLOCK();
1099
1100 return (0);
1101 }
1102
1103 /*
1104 * Process a recieved IGMPv3 group-specific or group-and-source-specific
1105 * query.
1106 * Return <0 if any error occured. Currently this is ignored.
1107 */
1108 static int
1109 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1110 int timer, /*const*/ struct igmpv3 *igmpv3)
1111 {
1112 int retval;
1113 uint16_t nsrc;
1114
1115 IN_MULTI_LOCK_ASSERT();
1116 IGMP_LOCK_ASSERT();
1117
1118 retval = 0;
1119
1120 switch (inm->inm_state) {
1121 case IGMP_NOT_MEMBER:
1122 case IGMP_SILENT_MEMBER:
1123 case IGMP_SLEEPING_MEMBER:
1124 case IGMP_LAZY_MEMBER:
1125 case IGMP_AWAKENING_MEMBER:
1126 case IGMP_IDLE_MEMBER:
1127 case IGMP_LEAVING_MEMBER:
1128 return (retval);
1129 break;
1130 case IGMP_REPORTING_MEMBER:
1131 case IGMP_G_QUERY_PENDING_MEMBER:
1132 case IGMP_SG_QUERY_PENDING_MEMBER:
1133 break;
1134 }
1135
1136 nsrc = ntohs(igmpv3->igmp_numsrc);
1137
1138 /*
1139 * Deal with group-specific queries upfront.
1140 * If any group query is already pending, purge any recorded
1141 * source-list state if it exists, and schedule a query response
1142 * for this group-specific query.
1143 */
1144 if (nsrc == 0) {
1145 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1146 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1147 inm_clear_recorded(inm);
1148 timer = min(inm->inm_timer, timer);
1149 }
1150 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1151 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1152 V_current_state_timers_running = 1;
1153 return (retval);
1154 }
1155
1156 /*
1157 * Deal with the case where a group-and-source-specific query has
1158 * been received but a group-specific query is already pending.
1159 */
1160 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1161 timer = min(inm->inm_timer, timer);
1162 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1163 V_current_state_timers_running = 1;
1164 return (retval);
1165 }
1166
1167 /*
1168 * Finally, deal with the case where a group-and-source-specific
1169 * query has been received, where a response to a previous g-s-r
1170 * query exists, or none exists.
1171 * In this case, we need to parse the source-list which the Querier
1172 * has provided us with and check if we have any source list filter
1173 * entries at T1 for these sources. If we do not, there is no need
1174 * schedule a report and the query may be dropped.
1175 * If we do, we must record them and schedule a current-state
1176 * report for those sources.
1177 * FIXME: Handling source lists larger than 1 mbuf requires that
1178 * we pass the mbuf chain pointer down to this function, and use
1179 * m_getptr() to walk the chain.
1180 */
1181 if (inm->inm_nsrc > 0) {
1182 const struct in_addr *ap;
1183 int i, nrecorded;
1184
1185 ap = (const struct in_addr *)(igmpv3 + 1);
1186 nrecorded = 0;
1187 for (i = 0; i < nsrc; i++, ap++) {
1188 retval = inm_record_source(inm, ap->s_addr);
1189 if (retval < 0)
1190 break;
1191 nrecorded += retval;
1192 }
1193 if (nrecorded > 0) {
1194 CTR1(KTR_IGMPV3,
1195 "%s: schedule response to SG query", __func__);
1196 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1197 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1198 V_current_state_timers_running = 1;
1199 }
1200 }
1201
1202 return (retval);
1203 }
1204
1205 /*
1206 * Process a received IGMPv1 host membership report.
1207 *
1208 * NOTE: 0.0.0.0 workaround breaks const correctness.
1209 */
1210 static int
1211 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1212 /*const*/ struct igmp *igmp)
1213 {
1214 struct in_ifaddr *ia;
1215 struct in_multi *inm;
1216
1217 IGMPSTAT_INC(igps_rcv_reports);
1218
1219 if (ifp->if_flags & IFF_LOOPBACK)
1220 return (0);
1221
1222 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr) ||
1223 !in_hosteq(igmp->igmp_group, ip->ip_dst))) {
1224 IGMPSTAT_INC(igps_rcv_badreports);
1225 return (EINVAL);
1226 }
1227
1228 /*
1229 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1230 * Booting clients may use the source address 0.0.0.0. Some
1231 * IGMP daemons may not know how to use IP_RECVIF to determine
1232 * the interface upon which this message was received.
1233 * Replace 0.0.0.0 with the subnet address if told to do so.
1234 */
1235 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1236 IFP_TO_IA(ifp, ia);
1237 if (ia != NULL) {
1238 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1239 ifa_free(&ia->ia_ifa);
1240 }
1241 }
1242
1243 CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1244 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1245
1246 /*
1247 * IGMPv1 report suppression.
1248 * If we are a member of this group, and our membership should be
1249 * reported, stop our group timer and transition to the 'lazy' state.
1250 */
1251 IN_MULTI_LOCK();
1252 inm = inm_lookup(ifp, igmp->igmp_group);
1253 if (inm != NULL) {
1254 struct igmp_ifinfo *igi;
1255
1256 igi = inm->inm_igi;
1257 if (igi == NULL) {
1258 KASSERT(igi != NULL,
1259 ("%s: no igi for ifp %p", __func__, ifp));
1260 goto out_locked;
1261 }
1262
1263 IGMPSTAT_INC(igps_rcv_ourreports);
1264
1265 /*
1266 * If we are in IGMPv3 host mode, do not allow the
1267 * other host's IGMPv1 report to suppress our reports
1268 * unless explicitly configured to do so.
1269 */
1270 if (igi->igi_version == IGMP_VERSION_3) {
1271 if (V_igmp_legacysupp)
1272 igmp_v3_suppress_group_record(inm);
1273 goto out_locked;
1274 }
1275
1276 inm->inm_timer = 0;
1277
1278 switch (inm->inm_state) {
1279 case IGMP_NOT_MEMBER:
1280 case IGMP_SILENT_MEMBER:
1281 break;
1282 case IGMP_IDLE_MEMBER:
1283 case IGMP_LAZY_MEMBER:
1284 case IGMP_AWAKENING_MEMBER:
1285 CTR3(KTR_IGMPV3,
1286 "report suppressed for %s on ifp %p(%s)",
1287 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1288 case IGMP_SLEEPING_MEMBER:
1289 inm->inm_state = IGMP_SLEEPING_MEMBER;
1290 break;
1291 case IGMP_REPORTING_MEMBER:
1292 CTR3(KTR_IGMPV3,
1293 "report suppressed for %s on ifp %p(%s)",
1294 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1295 if (igi->igi_version == IGMP_VERSION_1)
1296 inm->inm_state = IGMP_LAZY_MEMBER;
1297 else if (igi->igi_version == IGMP_VERSION_2)
1298 inm->inm_state = IGMP_SLEEPING_MEMBER;
1299 break;
1300 case IGMP_G_QUERY_PENDING_MEMBER:
1301 case IGMP_SG_QUERY_PENDING_MEMBER:
1302 case IGMP_LEAVING_MEMBER:
1303 break;
1304 }
1305 }
1306
1307 out_locked:
1308 IN_MULTI_UNLOCK();
1309
1310 return (0);
1311 }
1312
1313 /*
1314 * Process a received IGMPv2 host membership report.
1315 *
1316 * NOTE: 0.0.0.0 workaround breaks const correctness.
1317 */
1318 static int
1319 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1320 /*const*/ struct igmp *igmp)
1321 {
1322 struct in_ifaddr *ia;
1323 struct in_multi *inm;
1324
1325 /*
1326 * Make sure we don't hear our own membership report. Fast
1327 * leave requires knowing that we are the only member of a
1328 * group.
1329 */
1330 IFP_TO_IA(ifp, ia);
1331 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1332 ifa_free(&ia->ia_ifa);
1333 return (0);
1334 }
1335
1336 IGMPSTAT_INC(igps_rcv_reports);
1337
1338 if (ifp->if_flags & IFF_LOOPBACK) {
1339 if (ia != NULL)
1340 ifa_free(&ia->ia_ifa);
1341 return (0);
1342 }
1343
1344 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1345 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1346 if (ia != NULL)
1347 ifa_free(&ia->ia_ifa);
1348 IGMPSTAT_INC(igps_rcv_badreports);
1349 return (EINVAL);
1350 }
1351
1352 /*
1353 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1354 * Booting clients may use the source address 0.0.0.0. Some
1355 * IGMP daemons may not know how to use IP_RECVIF to determine
1356 * the interface upon which this message was received.
1357 * Replace 0.0.0.0 with the subnet address if told to do so.
1358 */
1359 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1360 if (ia != NULL)
1361 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1362 }
1363 if (ia != NULL)
1364 ifa_free(&ia->ia_ifa);
1365
1366 CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1367 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1368
1369 /*
1370 * IGMPv2 report suppression.
1371 * If we are a member of this group, and our membership should be
1372 * reported, and our group timer is pending or about to be reset,
1373 * stop our group timer by transitioning to the 'lazy' state.
1374 */
1375 IN_MULTI_LOCK();
1376 inm = inm_lookup(ifp, igmp->igmp_group);
1377 if (inm != NULL) {
1378 struct igmp_ifinfo *igi;
1379
1380 igi = inm->inm_igi;
1381 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1382
1383 IGMPSTAT_INC(igps_rcv_ourreports);
1384
1385 /*
1386 * If we are in IGMPv3 host mode, do not allow the
1387 * other host's IGMPv1 report to suppress our reports
1388 * unless explicitly configured to do so.
1389 */
1390 if (igi->igi_version == IGMP_VERSION_3) {
1391 if (V_igmp_legacysupp)
1392 igmp_v3_suppress_group_record(inm);
1393 goto out_locked;
1394 }
1395
1396 inm->inm_timer = 0;
1397
1398 switch (inm->inm_state) {
1399 case IGMP_NOT_MEMBER:
1400 case IGMP_SILENT_MEMBER:
1401 case IGMP_SLEEPING_MEMBER:
1402 break;
1403 case IGMP_REPORTING_MEMBER:
1404 case IGMP_IDLE_MEMBER:
1405 case IGMP_AWAKENING_MEMBER:
1406 CTR3(KTR_IGMPV3,
1407 "report suppressed for %s on ifp %p(%s)",
1408 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1409 case IGMP_LAZY_MEMBER:
1410 inm->inm_state = IGMP_LAZY_MEMBER;
1411 break;
1412 case IGMP_G_QUERY_PENDING_MEMBER:
1413 case IGMP_SG_QUERY_PENDING_MEMBER:
1414 case IGMP_LEAVING_MEMBER:
1415 break;
1416 }
1417 }
1418
1419 out_locked:
1420 IN_MULTI_UNLOCK();
1421
1422 return (0);
1423 }
1424
1425 void
1426 igmp_input(struct mbuf *m, int off)
1427 {
1428 int iphlen;
1429 struct ifnet *ifp;
1430 struct igmp *igmp;
1431 struct ip *ip;
1432 int igmplen;
1433 int minlen;
1434 int queryver;
1435
1436 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1437
1438 ifp = m->m_pkthdr.rcvif;
1439
1440 IGMPSTAT_INC(igps_rcv_total);
1441
1442 ip = mtod(m, struct ip *);
1443 iphlen = off;
1444 igmplen = ip->ip_len;
1445
1446 /*
1447 * Validate lengths.
1448 */
1449 if (igmplen < IGMP_MINLEN) {
1450 IGMPSTAT_INC(igps_rcv_tooshort);
1451 m_freem(m);
1452 return;
1453 }
1454
1455 /*
1456 * Always pullup to the minimum size for v1/v2 or v3
1457 * to amortize calls to m_pullup().
1458 */
1459 minlen = iphlen;
1460 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1461 minlen += IGMP_V3_QUERY_MINLEN;
1462 else
1463 minlen += IGMP_MINLEN;
1464 if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1465 (m = m_pullup(m, minlen)) == 0) {
1466 IGMPSTAT_INC(igps_rcv_tooshort);
1467 return;
1468 }
1469 ip = mtod(m, struct ip *);
1470
1471 /*
1472 * Validate checksum.
1473 */
1474 m->m_data += iphlen;
1475 m->m_len -= iphlen;
1476 igmp = mtod(m, struct igmp *);
1477 if (in_cksum(m, igmplen)) {
1478 IGMPSTAT_INC(igps_rcv_badsum);
1479 m_freem(m);
1480 return;
1481 }
1482 m->m_data -= iphlen;
1483 m->m_len += iphlen;
1484
1485 /*
1486 * IGMP control traffic is link-scope, and must have a TTL of 1.
1487 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1488 * probe packets may come from beyond the LAN.
1489 */
1490 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1491 IGMPSTAT_INC(igps_rcv_badttl);
1492 m_freem(m);
1493 return;
1494 }
1495
1496 switch (igmp->igmp_type) {
1497 case IGMP_HOST_MEMBERSHIP_QUERY:
1498 if (igmplen == IGMP_MINLEN) {
1499 if (igmp->igmp_code == 0)
1500 queryver = IGMP_VERSION_1;
1501 else
1502 queryver = IGMP_VERSION_2;
1503 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1504 queryver = IGMP_VERSION_3;
1505 } else {
1506 IGMPSTAT_INC(igps_rcv_tooshort);
1507 m_freem(m);
1508 return;
1509 }
1510
1511 switch (queryver) {
1512 case IGMP_VERSION_1:
1513 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1514 if (!V_igmp_v1enable)
1515 break;
1516 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1517 m_freem(m);
1518 return;
1519 }
1520 break;
1521
1522 case IGMP_VERSION_2:
1523 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1524 if (!V_igmp_v2enable)
1525 break;
1526 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1527 m_freem(m);
1528 return;
1529 }
1530 break;
1531
1532 case IGMP_VERSION_3: {
1533 struct igmpv3 *igmpv3;
1534 uint16_t igmpv3len;
1535 uint16_t srclen;
1536 int nsrc;
1537
1538 IGMPSTAT_INC(igps_rcv_v3_queries);
1539 igmpv3 = (struct igmpv3 *)igmp;
1540 /*
1541 * Validate length based on source count.
1542 */
1543 nsrc = ntohs(igmpv3->igmp_numsrc);
1544 srclen = sizeof(struct in_addr) * nsrc;
1545 if (nsrc * sizeof(in_addr_t) > srclen) {
1546 IGMPSTAT_INC(igps_rcv_tooshort);
1547 return;
1548 }
1549 /*
1550 * m_pullup() may modify m, so pullup in
1551 * this scope.
1552 */
1553 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1554 srclen;
1555 if ((m->m_flags & M_EXT ||
1556 m->m_len < igmpv3len) &&
1557 (m = m_pullup(m, igmpv3len)) == NULL) {
1558 IGMPSTAT_INC(igps_rcv_tooshort);
1559 return;
1560 }
1561 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1562 + iphlen);
1563 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1564 m_freem(m);
1565 return;
1566 }
1567 }
1568 break;
1569 }
1570 break;
1571
1572 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1573 if (!V_igmp_v1enable)
1574 break;
1575 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1576 m_freem(m);
1577 return;
1578 }
1579 break;
1580
1581 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1582 if (!V_igmp_v2enable)
1583 break;
1584 if (!ip_checkrouteralert(m))
1585 IGMPSTAT_INC(igps_rcv_nora);
1586 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1587 m_freem(m);
1588 return;
1589 }
1590 break;
1591
1592 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1593 /*
1594 * Hosts do not need to process IGMPv3 membership reports,
1595 * as report suppression is no longer required.
1596 */
1597 if (!ip_checkrouteralert(m))
1598 IGMPSTAT_INC(igps_rcv_nora);
1599 break;
1600
1601 default:
1602 break;
1603 }
1604
1605 /*
1606 * Pass all valid IGMP packets up to any process(es) listening on a
1607 * raw IGMP socket.
1608 */
1609 rip_input(m, off);
1610 }
1611
1612
1613 /*
1614 * Fast timeout handler (global).
1615 * VIMAGE: Timeout handlers are expected to service all vimages.
1616 */
1617 void
1618 igmp_fasttimo(void)
1619 {
1620 VNET_ITERATOR_DECL(vnet_iter);
1621
1622 VNET_LIST_RLOCK_NOSLEEP();
1623 VNET_FOREACH(vnet_iter) {
1624 CURVNET_SET(vnet_iter);
1625 igmp_fasttimo_vnet();
1626 CURVNET_RESTORE();
1627 }
1628 VNET_LIST_RUNLOCK_NOSLEEP();
1629 }
1630
1631 /*
1632 * Fast timeout handler (per-vnet).
1633 * Sends are shuffled off to a netisr to deal with Giant.
1634 *
1635 * VIMAGE: Assume caller has set up our curvnet.
1636 */
1637 static void
1638 igmp_fasttimo_vnet(void)
1639 {
1640 struct ifqueue scq; /* State-change packets */
1641 struct ifqueue qrq; /* Query response packets */
1642 struct ifnet *ifp;
1643 struct igmp_ifinfo *igi;
1644 struct ifmultiaddr *ifma, *tifma;
1645 struct in_multi *inm;
1646 int loop, uri_fasthz;
1647
1648 loop = 0;
1649 uri_fasthz = 0;
1650
1651 /*
1652 * Quick check to see if any work needs to be done, in order to
1653 * minimize the overhead of fasttimo processing.
1654 * SMPng: XXX Unlocked reads.
1655 */
1656 if (!V_current_state_timers_running &&
1657 !V_interface_timers_running &&
1658 !V_state_change_timers_running)
1659 return;
1660
1661 IN_MULTI_LOCK();
1662 IGMP_LOCK();
1663
1664 /*
1665 * IGMPv3 General Query response timer processing.
1666 */
1667 if (V_interface_timers_running) {
1668 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1669
1670 V_interface_timers_running = 0;
1671 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1672 if (igi->igi_v3_timer == 0) {
1673 /* Do nothing. */
1674 } else if (--igi->igi_v3_timer == 0) {
1675 igmp_v3_dispatch_general_query(igi);
1676 } else {
1677 V_interface_timers_running = 1;
1678 }
1679 }
1680 }
1681
1682 if (!V_current_state_timers_running &&
1683 !V_state_change_timers_running)
1684 goto out_locked;
1685
1686 V_current_state_timers_running = 0;
1687 V_state_change_timers_running = 0;
1688
1689 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1690
1691 /*
1692 * IGMPv1/v2/v3 host report and state-change timer processing.
1693 * Note: Processing a v3 group timer may remove a node.
1694 */
1695 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1696 ifp = igi->igi_ifp;
1697
1698 if (igi->igi_version == IGMP_VERSION_3) {
1699 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1700 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1701 PR_FASTHZ);
1702
1703 memset(&qrq, 0, sizeof(struct ifqueue));
1704 IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1705
1706 memset(&scq, 0, sizeof(struct ifqueue));
1707 IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1708 }
1709
1710 IF_ADDR_LOCK(ifp);
1711 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1712 tifma) {
1713 if (ifma->ifma_addr->sa_family != AF_INET ||
1714 ifma->ifma_protospec == NULL)
1715 continue;
1716 inm = (struct in_multi *)ifma->ifma_protospec;
1717 switch (igi->igi_version) {
1718 case IGMP_VERSION_1:
1719 case IGMP_VERSION_2:
1720 igmp_v1v2_process_group_timer(inm,
1721 igi->igi_version);
1722 break;
1723 case IGMP_VERSION_3:
1724 igmp_v3_process_group_timers(igi, &qrq,
1725 &scq, inm, uri_fasthz);
1726 break;
1727 }
1728 }
1729 IF_ADDR_UNLOCK(ifp);
1730
1731 if (igi->igi_version == IGMP_VERSION_3) {
1732 struct in_multi *tinm;
1733
1734 igmp_dispatch_queue(&qrq, 0, loop);
1735 igmp_dispatch_queue(&scq, 0, loop);
1736
1737 /*
1738 * Free the in_multi reference(s) for this
1739 * IGMP lifecycle.
1740 */
1741 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1742 inm_nrele, tinm) {
1743 SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1744 inm_nrele);
1745 inm_release_locked(inm);
1746 }
1747 }
1748 }
1749
1750 out_locked:
1751 IGMP_UNLOCK();
1752 IN_MULTI_UNLOCK();
1753 }
1754
1755 /*
1756 * Update host report group timer for IGMPv1/v2.
1757 * Will update the global pending timer flags.
1758 */
1759 static void
1760 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1761 {
1762 int report_timer_expired;
1763
1764 IN_MULTI_LOCK_ASSERT();
1765 IGMP_LOCK_ASSERT();
1766
1767 if (inm->inm_timer == 0) {
1768 report_timer_expired = 0;
1769 } else if (--inm->inm_timer == 0) {
1770 report_timer_expired = 1;
1771 } else {
1772 V_current_state_timers_running = 1;
1773 return;
1774 }
1775
1776 switch (inm->inm_state) {
1777 case IGMP_NOT_MEMBER:
1778 case IGMP_SILENT_MEMBER:
1779 case IGMP_IDLE_MEMBER:
1780 case IGMP_LAZY_MEMBER:
1781 case IGMP_SLEEPING_MEMBER:
1782 case IGMP_AWAKENING_MEMBER:
1783 break;
1784 case IGMP_REPORTING_MEMBER:
1785 if (report_timer_expired) {
1786 inm->inm_state = IGMP_IDLE_MEMBER;
1787 (void)igmp_v1v2_queue_report(inm,
1788 (version == IGMP_VERSION_2) ?
1789 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1790 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1791 }
1792 break;
1793 case IGMP_G_QUERY_PENDING_MEMBER:
1794 case IGMP_SG_QUERY_PENDING_MEMBER:
1795 case IGMP_LEAVING_MEMBER:
1796 break;
1797 }
1798 }
1799
1800 /*
1801 * Update a group's timers for IGMPv3.
1802 * Will update the global pending timer flags.
1803 * Note: Unlocked read from igi.
1804 */
1805 static void
1806 igmp_v3_process_group_timers(struct igmp_ifinfo *igi,
1807 struct ifqueue *qrq, struct ifqueue *scq,
1808 struct in_multi *inm, const int uri_fasthz)
1809 {
1810 int query_response_timer_expired;
1811 int state_change_retransmit_timer_expired;
1812
1813 IN_MULTI_LOCK_ASSERT();
1814 IGMP_LOCK_ASSERT();
1815
1816 query_response_timer_expired = 0;
1817 state_change_retransmit_timer_expired = 0;
1818
1819 /*
1820 * During a transition from v1/v2 compatibility mode back to v3,
1821 * a group record in REPORTING state may still have its group
1822 * timer active. This is a no-op in this function; it is easier
1823 * to deal with it here than to complicate the slow-timeout path.
1824 */
1825 if (inm->inm_timer == 0) {
1826 query_response_timer_expired = 0;
1827 } else if (--inm->inm_timer == 0) {
1828 query_response_timer_expired = 1;
1829 } else {
1830 V_current_state_timers_running = 1;
1831 }
1832
1833 if (inm->inm_sctimer == 0) {
1834 state_change_retransmit_timer_expired = 0;
1835 } else if (--inm->inm_sctimer == 0) {
1836 state_change_retransmit_timer_expired = 1;
1837 } else {
1838 V_state_change_timers_running = 1;
1839 }
1840
1841 /* We are in fasttimo, so be quick about it. */
1842 if (!state_change_retransmit_timer_expired &&
1843 !query_response_timer_expired)
1844 return;
1845
1846 switch (inm->inm_state) {
1847 case IGMP_NOT_MEMBER:
1848 case IGMP_SILENT_MEMBER:
1849 case IGMP_SLEEPING_MEMBER:
1850 case IGMP_LAZY_MEMBER:
1851 case IGMP_AWAKENING_MEMBER:
1852 case IGMP_IDLE_MEMBER:
1853 break;
1854 case IGMP_G_QUERY_PENDING_MEMBER:
1855 case IGMP_SG_QUERY_PENDING_MEMBER:
1856 /*
1857 * Respond to a previously pending Group-Specific
1858 * or Group-and-Source-Specific query by enqueueing
1859 * the appropriate Current-State report for
1860 * immediate transmission.
1861 */
1862 if (query_response_timer_expired) {
1863 int retval;
1864
1865 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1866 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1867 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1868 __func__, retval);
1869 inm->inm_state = IGMP_REPORTING_MEMBER;
1870 /* XXX Clear recorded sources for next time. */
1871 inm_clear_recorded(inm);
1872 }
1873 /* FALLTHROUGH */
1874 case IGMP_REPORTING_MEMBER:
1875 case IGMP_LEAVING_MEMBER:
1876 if (state_change_retransmit_timer_expired) {
1877 /*
1878 * State-change retransmission timer fired.
1879 * If there are any further pending retransmissions,
1880 * set the global pending state-change flag, and
1881 * reset the timer.
1882 */
1883 if (--inm->inm_scrv > 0) {
1884 inm->inm_sctimer = uri_fasthz;
1885 V_state_change_timers_running = 1;
1886 }
1887 /*
1888 * Retransmit the previously computed state-change
1889 * report. If there are no further pending
1890 * retransmissions, the mbuf queue will be consumed.
1891 * Update T0 state to T1 as we have now sent
1892 * a state-change.
1893 */
1894 (void)igmp_v3_merge_state_changes(inm, scq);
1895
1896 inm_commit(inm);
1897 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1898 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1899
1900 /*
1901 * If we are leaving the group for good, make sure
1902 * we release IGMP's reference to it.
1903 * This release must be deferred using a SLIST,
1904 * as we are called from a loop which traverses
1905 * the in_ifmultiaddr TAILQ.
1906 */
1907 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1908 inm->inm_scrv == 0) {
1909 inm->inm_state = IGMP_NOT_MEMBER;
1910 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1911 inm, inm_nrele);
1912 }
1913 }
1914 break;
1915 }
1916 }
1917
1918
1919 /*
1920 * Suppress a group's pending response to a group or source/group query.
1921 *
1922 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1923 * Do NOT update ST1/ST0 as this operation merely suppresses
1924 * the currently pending group record.
1925 * Do NOT suppress the response to a general query. It is possible but
1926 * it would require adding another state or flag.
1927 */
1928 static void
1929 igmp_v3_suppress_group_record(struct in_multi *inm)
1930 {
1931
1932 IN_MULTI_LOCK_ASSERT();
1933
1934 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1935 ("%s: not IGMPv3 mode on link", __func__));
1936
1937 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1938 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1939 return;
1940
1941 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1942 inm_clear_recorded(inm);
1943
1944 inm->inm_timer = 0;
1945 inm->inm_state = IGMP_REPORTING_MEMBER;
1946 }
1947
1948 /*
1949 * Switch to a different IGMP version on the given interface,
1950 * as per Section 7.2.1.
1951 */
1952 static void
1953 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1954 {
1955 int old_version_timer;
1956
1957 IGMP_LOCK_ASSERT();
1958
1959 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1960 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1961
1962 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1963 /*
1964 * Compute the "Older Version Querier Present" timer as per
1965 * Section 8.12.
1966 */
1967 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1968 old_version_timer *= PR_SLOWHZ;
1969
1970 if (version == IGMP_VERSION_1) {
1971 igi->igi_v1_timer = old_version_timer;
1972 igi->igi_v2_timer = 0;
1973 } else if (version == IGMP_VERSION_2) {
1974 igi->igi_v1_timer = 0;
1975 igi->igi_v2_timer = old_version_timer;
1976 }
1977 }
1978
1979 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1980 if (igi->igi_version != IGMP_VERSION_2) {
1981 igi->igi_version = IGMP_VERSION_2;
1982 igmp_v3_cancel_link_timers(igi);
1983 }
1984 } else if (igi->igi_v1_timer > 0) {
1985 if (igi->igi_version != IGMP_VERSION_1) {
1986 igi->igi_version = IGMP_VERSION_1;
1987 igmp_v3_cancel_link_timers(igi);
1988 }
1989 }
1990 }
1991
1992 /*
1993 * Cancel pending IGMPv3 timers for the given link and all groups
1994 * joined on it; state-change, general-query, and group-query timers.
1995 *
1996 * Only ever called on a transition from v3 to Compatibility mode. Kill
1997 * the timers stone dead (this may be expensive for large N groups), they
1998 * will be restarted if Compatibility Mode deems that they must be due to
1999 * query processing.
2000 */
2001 static void
2002 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
2003 {
2004 struct ifmultiaddr *ifma;
2005 struct ifnet *ifp;
2006 struct in_multi *inm;
2007
2008 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2009 igi->igi_ifp, igi->igi_ifp->if_xname);
2010
2011 IN_MULTI_LOCK_ASSERT();
2012 IGMP_LOCK_ASSERT();
2013
2014 /*
2015 * Stop the v3 General Query Response on this link stone dead.
2016 * If fasttimo is woken up due to V_interface_timers_running,
2017 * the flag will be cleared if there are no pending link timers.
2018 */
2019 igi->igi_v3_timer = 0;
2020
2021 /*
2022 * Now clear the current-state and state-change report timers
2023 * for all memberships scoped to this link.
2024 */
2025 ifp = igi->igi_ifp;
2026 IF_ADDR_LOCK(ifp);
2027 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2028 if (ifma->ifma_addr->sa_family != AF_INET ||
2029 ifma->ifma_protospec == NULL)
2030 continue;
2031 inm = (struct in_multi *)ifma->ifma_protospec;
2032 switch (inm->inm_state) {
2033 case IGMP_NOT_MEMBER:
2034 case IGMP_SILENT_MEMBER:
2035 case IGMP_IDLE_MEMBER:
2036 case IGMP_LAZY_MEMBER:
2037 case IGMP_SLEEPING_MEMBER:
2038 case IGMP_AWAKENING_MEMBER:
2039 /*
2040 * These states are either not relevant in v3 mode,
2041 * or are unreported. Do nothing.
2042 */
2043 break;
2044 case IGMP_LEAVING_MEMBER:
2045 /*
2046 * If we are leaving the group and switching to
2047 * compatibility mode, we need to release the final
2048 * reference held for issuing the INCLUDE {}, and
2049 * transition to REPORTING to ensure the host leave
2050 * message is sent upstream to the old querier --
2051 * transition to NOT would lose the leave and race.
2052 *
2053 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
2054 * around inm_release_locked(), as it is not
2055 * a recursive mutex.
2056 */
2057 IF_ADDR_UNLOCK(ifp);
2058 inm_release_locked(inm);
2059 IF_ADDR_LOCK(ifp);
2060 /* FALLTHROUGH */
2061 case IGMP_G_QUERY_PENDING_MEMBER:
2062 case IGMP_SG_QUERY_PENDING_MEMBER:
2063 inm_clear_recorded(inm);
2064 /* FALLTHROUGH */
2065 case IGMP_REPORTING_MEMBER:
2066 inm->inm_state = IGMP_REPORTING_MEMBER;
2067 break;
2068 }
2069 /*
2070 * Always clear state-change and group report timers.
2071 * Free any pending IGMPv3 state-change records.
2072 */
2073 inm->inm_sctimer = 0;
2074 inm->inm_timer = 0;
2075 _IF_DRAIN(&inm->inm_scq);
2076 }
2077 IF_ADDR_UNLOCK(ifp);
2078 }
2079
2080 /*
2081 * Update the Older Version Querier Present timers for a link.
2082 * See Section 7.2.1 of RFC 3376.
2083 */
2084 static void
2085 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2086 {
2087
2088 IGMP_LOCK_ASSERT();
2089
2090 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2091 /*
2092 * IGMPv1 and IGMPv2 Querier Present timers expired.
2093 *
2094 * Revert to IGMPv3.
2095 */
2096 if (igi->igi_version != IGMP_VERSION_3) {
2097 CTR5(KTR_IGMPV3,
2098 "%s: transition from v%d -> v%d on %p(%s)",
2099 __func__, igi->igi_version, IGMP_VERSION_3,
2100 igi->igi_ifp, igi->igi_ifp->if_xname);
2101 igi->igi_version = IGMP_VERSION_3;
2102 }
2103 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2104 /*
2105 * IGMPv1 Querier Present timer expired,
2106 * IGMPv2 Querier Present timer running.
2107 * If IGMPv2 was disabled since last timeout,
2108 * revert to IGMPv3.
2109 * If IGMPv2 is enabled, revert to IGMPv2.
2110 */
2111 if (!V_igmp_v2enable) {
2112 CTR5(KTR_IGMPV3,
2113 "%s: transition from v%d -> v%d on %p(%s)",
2114 __func__, igi->igi_version, IGMP_VERSION_3,
2115 igi->igi_ifp, igi->igi_ifp->if_xname);
2116 igi->igi_v2_timer = 0;
2117 igi->igi_version = IGMP_VERSION_3;
2118 } else {
2119 --igi->igi_v2_timer;
2120 if (igi->igi_version != IGMP_VERSION_2) {
2121 CTR5(KTR_IGMPV3,
2122 "%s: transition from v%d -> v%d on %p(%s)",
2123 __func__, igi->igi_version, IGMP_VERSION_2,
2124 igi->igi_ifp, igi->igi_ifp->if_xname);
2125 igi->igi_version = IGMP_VERSION_2;
2126 }
2127 }
2128 } else if (igi->igi_v1_timer > 0) {
2129 /*
2130 * IGMPv1 Querier Present timer running.
2131 * Stop IGMPv2 timer if running.
2132 *
2133 * If IGMPv1 was disabled since last timeout,
2134 * revert to IGMPv3.
2135 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2136 */
2137 if (!V_igmp_v1enable) {
2138 CTR5(KTR_IGMPV3,
2139 "%s: transition from v%d -> v%d on %p(%s)",
2140 __func__, igi->igi_version, IGMP_VERSION_3,
2141 igi->igi_ifp, igi->igi_ifp->if_xname);
2142 igi->igi_v1_timer = 0;
2143 igi->igi_version = IGMP_VERSION_3;
2144 } else {
2145 --igi->igi_v1_timer;
2146 }
2147 if (igi->igi_v2_timer > 0) {
2148 CTR3(KTR_IGMPV3,
2149 "%s: cancel v2 timer on %p(%s)",
2150 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2151 igi->igi_v2_timer = 0;
2152 }
2153 }
2154 }
2155
2156 /*
2157 * Global slowtimo handler.
2158 * VIMAGE: Timeout handlers are expected to service all vimages.
2159 */
2160 void
2161 igmp_slowtimo(void)
2162 {
2163 VNET_ITERATOR_DECL(vnet_iter);
2164
2165 VNET_LIST_RLOCK_NOSLEEP();
2166 VNET_FOREACH(vnet_iter) {
2167 CURVNET_SET(vnet_iter);
2168 igmp_slowtimo_vnet();
2169 CURVNET_RESTORE();
2170 }
2171 VNET_LIST_RUNLOCK_NOSLEEP();
2172 }
2173
2174 /*
2175 * Per-vnet slowtimo handler.
2176 */
2177 static void
2178 igmp_slowtimo_vnet(void)
2179 {
2180 struct igmp_ifinfo *igi;
2181
2182 IGMP_LOCK();
2183
2184 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2185 igmp_v1v2_process_querier_timers(igi);
2186 }
2187
2188 IGMP_UNLOCK();
2189 }
2190
2191 /*
2192 * Dispatch an IGMPv1/v2 host report or leave message.
2193 * These are always small enough to fit inside a single mbuf.
2194 */
2195 static int
2196 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2197 {
2198 struct ifnet *ifp;
2199 struct igmp *igmp;
2200 struct ip *ip;
2201 struct mbuf *m;
2202
2203 IN_MULTI_LOCK_ASSERT();
2204 IGMP_LOCK_ASSERT();
2205
2206 ifp = inm->inm_ifp;
2207
2208 MGETHDR(m, M_DONTWAIT, MT_DATA);
2209 if (m == NULL)
2210 return (ENOMEM);
2211 MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2212
2213 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2214
2215 m->m_data += sizeof(struct ip);
2216 m->m_len = sizeof(struct igmp);
2217
2218 igmp = mtod(m, struct igmp *);
2219 igmp->igmp_type = type;
2220 igmp->igmp_code = 0;
2221 igmp->igmp_group = inm->inm_addr;
2222 igmp->igmp_cksum = 0;
2223 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2224
2225 m->m_data -= sizeof(struct ip);
2226 m->m_len += sizeof(struct ip);
2227
2228 ip = mtod(m, struct ip *);
2229 ip->ip_tos = 0;
2230 ip->ip_len = sizeof(struct ip) + sizeof(struct igmp);
2231 ip->ip_off = 0;
2232 ip->ip_p = IPPROTO_IGMP;
2233 ip->ip_src.s_addr = INADDR_ANY;
2234
2235 if (type == IGMP_HOST_LEAVE_MESSAGE)
2236 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2237 else
2238 ip->ip_dst = inm->inm_addr;
2239
2240 igmp_save_context(m, ifp);
2241
2242 m->m_flags |= M_IGMPV2;
2243 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2244 m->m_flags |= M_IGMP_LOOP;
2245
2246 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2247 netisr_dispatch(NETISR_IGMP, m);
2248
2249 return (0);
2250 }
2251
2252 /*
2253 * Process a state change from the upper layer for the given IPv4 group.
2254 *
2255 * Each socket holds a reference on the in_multi in its own ip_moptions.
2256 * The socket layer will have made the necessary updates to.the group
2257 * state, it is now up to IGMP to issue a state change report if there
2258 * has been any change between T0 (when the last state-change was issued)
2259 * and T1 (now).
2260 *
2261 * We use the IGMPv3 state machine at group level. The IGMP module
2262 * however makes the decision as to which IGMP protocol version to speak.
2263 * A state change *from* INCLUDE {} always means an initial join.
2264 * A state change *to* INCLUDE {} always means a final leave.
2265 *
2266 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2267 * save ourselves a bunch of work; any exclusive mode groups need not
2268 * compute source filter lists.
2269 *
2270 * VIMAGE: curvnet should have been set by caller, as this routine
2271 * is called from the socket option handlers.
2272 */
2273 int
2274 igmp_change_state(struct in_multi *inm)
2275 {
2276 struct igmp_ifinfo *igi;
2277 struct ifnet *ifp;
2278 int error;
2279
2280 IN_MULTI_LOCK_ASSERT();
2281
2282 error = 0;
2283
2284 /*
2285 * Try to detect if the upper layer just asked us to change state
2286 * for an interface which has now gone away.
2287 */
2288 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2289 ifp = inm->inm_ifma->ifma_ifp;
2290 if (ifp != NULL) {
2291 /*
2292 * Sanity check that netinet's notion of ifp is the
2293 * same as net's.
2294 */
2295 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2296 }
2297
2298 IGMP_LOCK();
2299
2300 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2301 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2302
2303 /*
2304 * If we detect a state transition to or from MCAST_UNDEFINED
2305 * for this group, then we are starting or finishing an IGMP
2306 * life cycle for this group.
2307 */
2308 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2309 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2310 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2311 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2312 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2313 error = igmp_initial_join(inm, igi);
2314 goto out_locked;
2315 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2316 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2317 igmp_final_leave(inm, igi);
2318 goto out_locked;
2319 }
2320 } else {
2321 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2322 }
2323
2324 error = igmp_handle_state_change(inm, igi);
2325
2326 out_locked:
2327 IGMP_UNLOCK();
2328 return (error);
2329 }
2330
2331 /*
2332 * Perform the initial join for an IGMP group.
2333 *
2334 * When joining a group:
2335 * If the group should have its IGMP traffic suppressed, do nothing.
2336 * IGMPv1 starts sending IGMPv1 host membership reports.
2337 * IGMPv2 starts sending IGMPv2 host membership reports.
2338 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2339 * initial state of the membership.
2340 */
2341 static int
2342 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2343 {
2344 struct ifnet *ifp;
2345 struct ifqueue *ifq;
2346 int error, retval, syncstates;
2347
2348 CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2349 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2350 inm->inm_ifp->if_xname);
2351
2352 error = 0;
2353 syncstates = 1;
2354
2355 ifp = inm->inm_ifp;
2356
2357 IN_MULTI_LOCK_ASSERT();
2358 IGMP_LOCK_ASSERT();
2359
2360 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2361
2362 /*
2363 * Groups joined on loopback or marked as 'not reported',
2364 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2365 * are never reported in any IGMP protocol exchanges.
2366 * All other groups enter the appropriate IGMP state machine
2367 * for the version in use on this link.
2368 * A link marked as IGIF_SILENT causes IGMP to be completely
2369 * disabled for the link.
2370 */
2371 if ((ifp->if_flags & IFF_LOOPBACK) ||
2372 (igi->igi_flags & IGIF_SILENT) ||
2373 !igmp_isgroupreported(inm->inm_addr)) {
2374 CTR1(KTR_IGMPV3,
2375 "%s: not kicking state machine for silent group", __func__);
2376 inm->inm_state = IGMP_SILENT_MEMBER;
2377 inm->inm_timer = 0;
2378 } else {
2379 /*
2380 * Deal with overlapping in_multi lifecycle.
2381 * If this group was LEAVING, then make sure
2382 * we drop the reference we picked up to keep the
2383 * group around for the final INCLUDE {} enqueue.
2384 */
2385 if (igi->igi_version == IGMP_VERSION_3 &&
2386 inm->inm_state == IGMP_LEAVING_MEMBER)
2387 inm_release_locked(inm);
2388
2389 inm->inm_state = IGMP_REPORTING_MEMBER;
2390
2391 switch (igi->igi_version) {
2392 case IGMP_VERSION_1:
2393 case IGMP_VERSION_2:
2394 inm->inm_state = IGMP_IDLE_MEMBER;
2395 error = igmp_v1v2_queue_report(inm,
2396 (igi->igi_version == IGMP_VERSION_2) ?
2397 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2398 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2399 if (error == 0) {
2400 inm->inm_timer = IGMP_RANDOM_DELAY(
2401 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2402 V_current_state_timers_running = 1;
2403 }
2404 break;
2405
2406 case IGMP_VERSION_3:
2407 /*
2408 * Defer update of T0 to T1, until the first copy
2409 * of the state change has been transmitted.
2410 */
2411 syncstates = 0;
2412
2413 /*
2414 * Immediately enqueue a State-Change Report for
2415 * this interface, freeing any previous reports.
2416 * Don't kick the timers if there is nothing to do,
2417 * or if an error occurred.
2418 */
2419 ifq = &inm->inm_scq;
2420 _IF_DRAIN(ifq);
2421 retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2422 0, 0);
2423 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2424 __func__, retval);
2425 if (retval <= 0) {
2426 error = retval * -1;
2427 break;
2428 }
2429
2430 /*
2431 * Schedule transmission of pending state-change
2432 * report up to RV times for this link. The timer
2433 * will fire at the next igmp_fasttimo (~200ms),
2434 * giving us an opportunity to merge the reports.
2435 */
2436 if (igi->igi_flags & IGIF_LOOPBACK) {
2437 inm->inm_scrv = 1;
2438 } else {
2439 KASSERT(igi->igi_rv > 1,
2440 ("%s: invalid robustness %d", __func__,
2441 igi->igi_rv));
2442 inm->inm_scrv = igi->igi_rv;
2443 }
2444 inm->inm_sctimer = 1;
2445 V_state_change_timers_running = 1;
2446
2447 error = 0;
2448 break;
2449 }
2450 }
2451
2452 /*
2453 * Only update the T0 state if state change is atomic,
2454 * i.e. we don't need to wait for a timer to fire before we
2455 * can consider the state change to have been communicated.
2456 */
2457 if (syncstates) {
2458 inm_commit(inm);
2459 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2460 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2461 }
2462
2463 return (error);
2464 }
2465
2466 /*
2467 * Issue an intermediate state change during the IGMP life-cycle.
2468 */
2469 static int
2470 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2471 {
2472 struct ifnet *ifp;
2473 int retval;
2474
2475 CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2476 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2477 inm->inm_ifp->if_xname);
2478
2479 ifp = inm->inm_ifp;
2480
2481 IN_MULTI_LOCK_ASSERT();
2482 IGMP_LOCK_ASSERT();
2483
2484 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2485
2486 if ((ifp->if_flags & IFF_LOOPBACK) ||
2487 (igi->igi_flags & IGIF_SILENT) ||
2488 !igmp_isgroupreported(inm->inm_addr) ||
2489 (igi->igi_version != IGMP_VERSION_3)) {
2490 if (!igmp_isgroupreported(inm->inm_addr)) {
2491 CTR1(KTR_IGMPV3,
2492 "%s: not kicking state machine for silent group", __func__);
2493 }
2494 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2495 inm_commit(inm);
2496 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2497 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2498 return (0);
2499 }
2500
2501 _IF_DRAIN(&inm->inm_scq);
2502
2503 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2504 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2505 if (retval <= 0)
2506 return (-retval);
2507
2508 /*
2509 * If record(s) were enqueued, start the state-change
2510 * report timer for this group.
2511 */
2512 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2513 inm->inm_sctimer = 1;
2514 V_state_change_timers_running = 1;
2515
2516 return (0);
2517 }
2518
2519 /*
2520 * Perform the final leave for an IGMP group.
2521 *
2522 * When leaving a group:
2523 * IGMPv1 does nothing.
2524 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2525 * IGMPv3 enqueues a state-change report containing a transition
2526 * to INCLUDE {} for immediate transmission.
2527 */
2528 static void
2529 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2530 {
2531 int syncstates;
2532
2533 syncstates = 1;
2534
2535 CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2536 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2537 inm->inm_ifp->if_xname);
2538
2539 IN_MULTI_LOCK_ASSERT();
2540 IGMP_LOCK_ASSERT();
2541
2542 switch (inm->inm_state) {
2543 case IGMP_NOT_MEMBER:
2544 case IGMP_SILENT_MEMBER:
2545 case IGMP_LEAVING_MEMBER:
2546 /* Already leaving or left; do nothing. */
2547 CTR1(KTR_IGMPV3,
2548 "%s: not kicking state machine for silent group", __func__);
2549 break;
2550 case IGMP_REPORTING_MEMBER:
2551 case IGMP_IDLE_MEMBER:
2552 case IGMP_G_QUERY_PENDING_MEMBER:
2553 case IGMP_SG_QUERY_PENDING_MEMBER:
2554 if (igi->igi_version == IGMP_VERSION_2) {
2555 #ifdef INVARIANTS
2556 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2557 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2558 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2559 __func__);
2560 #endif
2561 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2562 inm->inm_state = IGMP_NOT_MEMBER;
2563 } else if (igi->igi_version == IGMP_VERSION_3) {
2564 /*
2565 * Stop group timer and all pending reports.
2566 * Immediately enqueue a state-change report
2567 * TO_IN {} to be sent on the next fast timeout,
2568 * giving us an opportunity to merge reports.
2569 */
2570 _IF_DRAIN(&inm->inm_scq);
2571 inm->inm_timer = 0;
2572 if (igi->igi_flags & IGIF_LOOPBACK) {
2573 inm->inm_scrv = 1;
2574 } else {
2575 inm->inm_scrv = igi->igi_rv;
2576 }
2577 CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2578 "pending retransmissions.", __func__,
2579 inet_ntoa(inm->inm_addr),
2580 inm->inm_ifp->if_xname, inm->inm_scrv);
2581 if (inm->inm_scrv == 0) {
2582 inm->inm_state = IGMP_NOT_MEMBER;
2583 inm->inm_sctimer = 0;
2584 } else {
2585 int retval;
2586
2587 inm_acquire_locked(inm);
2588
2589 retval = igmp_v3_enqueue_group_record(
2590 &inm->inm_scq, inm, 1, 0, 0);
2591 KASSERT(retval != 0,
2592 ("%s: enqueue record = %d", __func__,
2593 retval));
2594
2595 inm->inm_state = IGMP_LEAVING_MEMBER;
2596 inm->inm_sctimer = 1;
2597 V_state_change_timers_running = 1;
2598 syncstates = 0;
2599 }
2600 break;
2601 }
2602 break;
2603 case IGMP_LAZY_MEMBER:
2604 case IGMP_SLEEPING_MEMBER:
2605 case IGMP_AWAKENING_MEMBER:
2606 /* Our reports are suppressed; do nothing. */
2607 break;
2608 }
2609
2610 if (syncstates) {
2611 inm_commit(inm);
2612 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2613 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2614 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2615 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2616 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2617 }
2618 }
2619
2620 /*
2621 * Enqueue an IGMPv3 group record to the given output queue.
2622 *
2623 * XXX This function could do with having the allocation code
2624 * split out, and the multiple-tree-walks coalesced into a single
2625 * routine as has been done in igmp_v3_enqueue_filter_change().
2626 *
2627 * If is_state_change is zero, a current-state record is appended.
2628 * If is_state_change is non-zero, a state-change report is appended.
2629 *
2630 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2631 * If is_group_query is zero, and if there is a packet with free space
2632 * at the tail of the queue, it will be appended to providing there
2633 * is enough free space.
2634 * Otherwise a new mbuf packet chain is allocated.
2635 *
2636 * If is_source_query is non-zero, each source is checked to see if
2637 * it was recorded for a Group-Source query, and will be omitted if
2638 * it is not both in-mode and recorded.
2639 *
2640 * The function will attempt to allocate leading space in the packet
2641 * for the IP/IGMP header to be prepended without fragmenting the chain.
2642 *
2643 * If successful the size of all data appended to the queue is returned,
2644 * otherwise an error code less than zero is returned, or zero if
2645 * no record(s) were appended.
2646 */
2647 static int
2648 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2649 const int is_state_change, const int is_group_query,
2650 const int is_source_query)
2651 {
2652 struct igmp_grouprec ig;
2653 struct igmp_grouprec *pig;
2654 struct ifnet *ifp;
2655 struct ip_msource *ims, *nims;
2656 struct mbuf *m0, *m, *md;
2657 int error, is_filter_list_change;
2658 int minrec0len, m0srcs, msrcs, nbytes, off;
2659 int record_has_sources;
2660 int now;
2661 int type;
2662 in_addr_t naddr;
2663 uint8_t mode;
2664
2665 IN_MULTI_LOCK_ASSERT();
2666
2667 error = 0;
2668 ifp = inm->inm_ifp;
2669 is_filter_list_change = 0;
2670 m = NULL;
2671 m0 = NULL;
2672 m0srcs = 0;
2673 msrcs = 0;
2674 nbytes = 0;
2675 nims = NULL;
2676 record_has_sources = 1;
2677 pig = NULL;
2678 type = IGMP_DO_NOTHING;
2679 mode = inm->inm_st[1].iss_fmode;
2680
2681 /*
2682 * If we did not transition out of ASM mode during t0->t1,
2683 * and there are no source nodes to process, we can skip
2684 * the generation of source records.
2685 */
2686 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2687 inm->inm_nsrc == 0)
2688 record_has_sources = 0;
2689
2690 if (is_state_change) {
2691 /*
2692 * Queue a state change record.
2693 * If the mode did not change, and there are non-ASM
2694 * listeners or source filters present,
2695 * we potentially need to issue two records for the group.
2696 * If we are transitioning to MCAST_UNDEFINED, we need
2697 * not send any sources.
2698 * If there are ASM listeners, and there was no filter
2699 * mode transition of any kind, do nothing.
2700 */
2701 if (mode != inm->inm_st[0].iss_fmode) {
2702 if (mode == MCAST_EXCLUDE) {
2703 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2704 __func__);
2705 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2706 } else {
2707 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2708 __func__);
2709 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2710 if (mode == MCAST_UNDEFINED)
2711 record_has_sources = 0;
2712 }
2713 } else {
2714 if (record_has_sources) {
2715 is_filter_list_change = 1;
2716 } else {
2717 type = IGMP_DO_NOTHING;
2718 }
2719 }
2720 } else {
2721 /*
2722 * Queue a current state record.
2723 */
2724 if (mode == MCAST_EXCLUDE) {
2725 type = IGMP_MODE_IS_EXCLUDE;
2726 } else if (mode == MCAST_INCLUDE) {
2727 type = IGMP_MODE_IS_INCLUDE;
2728 KASSERT(inm->inm_st[1].iss_asm == 0,
2729 ("%s: inm %p is INCLUDE but ASM count is %d",
2730 __func__, inm, inm->inm_st[1].iss_asm));
2731 }
2732 }
2733
2734 /*
2735 * Generate the filter list changes using a separate function.
2736 */
2737 if (is_filter_list_change)
2738 return (igmp_v3_enqueue_filter_change(ifq, inm));
2739
2740 if (type == IGMP_DO_NOTHING) {
2741 CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2742 __func__, inet_ntoa(inm->inm_addr),
2743 inm->inm_ifp->if_xname);
2744 return (0);
2745 }
2746
2747 /*
2748 * If any sources are present, we must be able to fit at least
2749 * one in the trailing space of the tail packet's mbuf,
2750 * ideally more.
2751 */
2752 minrec0len = sizeof(struct igmp_grouprec);
2753 if (record_has_sources)
2754 minrec0len += sizeof(in_addr_t);
2755
2756 CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2757 igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2758 inm->inm_ifp->if_xname);
2759
2760 /*
2761 * Check if we have a packet in the tail of the queue for this
2762 * group into which the first group record for this group will fit.
2763 * Otherwise allocate a new packet.
2764 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2765 * Note: Group records for G/GSR query responses MUST be sent
2766 * in their own packet.
2767 */
2768 m0 = ifq->ifq_tail;
2769 if (!is_group_query &&
2770 m0 != NULL &&
2771 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2772 (m0->m_pkthdr.len + minrec0len) <
2773 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2774 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2775 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2776 m = m0;
2777 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2778 } else {
2779 if (_IF_QFULL(ifq)) {
2780 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2781 return (-ENOMEM);
2782 }
2783 m = NULL;
2784 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2785 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2786 if (!is_state_change && !is_group_query) {
2787 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2788 if (m)
2789 m->m_data += IGMP_LEADINGSPACE;
2790 }
2791 if (m == NULL) {
2792 m = m_gethdr(M_DONTWAIT, MT_DATA);
2793 if (m)
2794 MH_ALIGN(m, IGMP_LEADINGSPACE);
2795 }
2796 if (m == NULL)
2797 return (-ENOMEM);
2798
2799 igmp_save_context(m, ifp);
2800
2801 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2802 }
2803
2804 /*
2805 * Append group record.
2806 * If we have sources, we don't know how many yet.
2807 */
2808 ig.ig_type = type;
2809 ig.ig_datalen = 0;
2810 ig.ig_numsrc = 0;
2811 ig.ig_group = inm->inm_addr;
2812 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2813 if (m != m0)
2814 m_freem(m);
2815 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2816 return (-ENOMEM);
2817 }
2818 nbytes += sizeof(struct igmp_grouprec);
2819
2820 /*
2821 * Append as many sources as will fit in the first packet.
2822 * If we are appending to a new packet, the chain allocation
2823 * may potentially use clusters; use m_getptr() in this case.
2824 * If we are appending to an existing packet, we need to obtain
2825 * a pointer to the group record after m_append(), in case a new
2826 * mbuf was allocated.
2827 * Only append sources which are in-mode at t1. If we are
2828 * transitioning to MCAST_UNDEFINED state on the group, do not
2829 * include source entries.
2830 * Only report recorded sources in our filter set when responding
2831 * to a group-source query.
2832 */
2833 if (record_has_sources) {
2834 if (m == m0) {
2835 md = m_last(m);
2836 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2837 md->m_len - nbytes);
2838 } else {
2839 md = m_getptr(m, 0, &off);
2840 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2841 off);
2842 }
2843 msrcs = 0;
2844 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2845 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2846 inet_ntoa_haddr(ims->ims_haddr));
2847 now = ims_get_mode(inm, ims, 1);
2848 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2849 if ((now != mode) ||
2850 (now == mode && mode == MCAST_UNDEFINED)) {
2851 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2852 continue;
2853 }
2854 if (is_source_query && ims->ims_stp == 0) {
2855 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2856 __func__);
2857 continue;
2858 }
2859 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2860 naddr = htonl(ims->ims_haddr);
2861 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2862 if (m != m0)
2863 m_freem(m);
2864 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2865 __func__);
2866 return (-ENOMEM);
2867 }
2868 nbytes += sizeof(in_addr_t);
2869 ++msrcs;
2870 if (msrcs == m0srcs)
2871 break;
2872 }
2873 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2874 msrcs);
2875 pig->ig_numsrc = htons(msrcs);
2876 nbytes += (msrcs * sizeof(in_addr_t));
2877 }
2878
2879 if (is_source_query && msrcs == 0) {
2880 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2881 if (m != m0)
2882 m_freem(m);
2883 return (0);
2884 }
2885
2886 /*
2887 * We are good to go with first packet.
2888 */
2889 if (m != m0) {
2890 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2891 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2892 _IF_ENQUEUE(ifq, m);
2893 } else
2894 m->m_pkthdr.PH_vt.vt_nrecs++;
2895
2896 /*
2897 * No further work needed if no source list in packet(s).
2898 */
2899 if (!record_has_sources)
2900 return (nbytes);
2901
2902 /*
2903 * Whilst sources remain to be announced, we need to allocate
2904 * a new packet and fill out as many sources as will fit.
2905 * Always try for a cluster first.
2906 */
2907 while (nims != NULL) {
2908 if (_IF_QFULL(ifq)) {
2909 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2910 return (-ENOMEM);
2911 }
2912 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2913 if (m)
2914 m->m_data += IGMP_LEADINGSPACE;
2915 if (m == NULL) {
2916 m = m_gethdr(M_DONTWAIT, MT_DATA);
2917 if (m)
2918 MH_ALIGN(m, IGMP_LEADINGSPACE);
2919 }
2920 if (m == NULL)
2921 return (-ENOMEM);
2922 igmp_save_context(m, ifp);
2923 md = m_getptr(m, 0, &off);
2924 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2925 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2926
2927 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2928 if (m != m0)
2929 m_freem(m);
2930 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2931 return (-ENOMEM);
2932 }
2933 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2934 nbytes += sizeof(struct igmp_grouprec);
2935
2936 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2937 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2938
2939 msrcs = 0;
2940 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2941 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2942 inet_ntoa_haddr(ims->ims_haddr));
2943 now = ims_get_mode(inm, ims, 1);
2944 if ((now != mode) ||
2945 (now == mode && mode == MCAST_UNDEFINED)) {
2946 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2947 continue;
2948 }
2949 if (is_source_query && ims->ims_stp == 0) {
2950 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2951 __func__);
2952 continue;
2953 }
2954 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2955 naddr = htonl(ims->ims_haddr);
2956 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2957 if (m != m0)
2958 m_freem(m);
2959 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2960 __func__);
2961 return (-ENOMEM);
2962 }
2963 ++msrcs;
2964 if (msrcs == m0srcs)
2965 break;
2966 }
2967 pig->ig_numsrc = htons(msrcs);
2968 nbytes += (msrcs * sizeof(in_addr_t));
2969
2970 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2971 _IF_ENQUEUE(ifq, m);
2972 }
2973
2974 return (nbytes);
2975 }
2976
2977 /*
2978 * Type used to mark record pass completion.
2979 * We exploit the fact we can cast to this easily from the
2980 * current filter modes on each ip_msource node.
2981 */
2982 typedef enum {
2983 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2984 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2985 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2986 REC_FULL = REC_ALLOW | REC_BLOCK
2987 } rectype_t;
2988
2989 /*
2990 * Enqueue an IGMPv3 filter list change to the given output queue.
2991 *
2992 * Source list filter state is held in an RB-tree. When the filter list
2993 * for a group is changed without changing its mode, we need to compute
2994 * the deltas between T0 and T1 for each source in the filter set,
2995 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2996 *
2997 * As we may potentially queue two record types, and the entire R-B tree
2998 * needs to be walked at once, we break this out into its own function
2999 * so we can generate a tightly packed queue of packets.
3000 *
3001 * XXX This could be written to only use one tree walk, although that makes
3002 * serializing into the mbuf chains a bit harder. For now we do two walks
3003 * which makes things easier on us, and it may or may not be harder on
3004 * the L2 cache.
3005 *
3006 * If successful the size of all data appended to the queue is returned,
3007 * otherwise an error code less than zero is returned, or zero if
3008 * no record(s) were appended.
3009 */
3010 static int
3011 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
3012 {
3013 static const int MINRECLEN =
3014 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3015 struct ifnet *ifp;
3016 struct igmp_grouprec ig;
3017 struct igmp_grouprec *pig;
3018 struct ip_msource *ims, *nims;
3019 struct mbuf *m, *m0, *md;
3020 in_addr_t naddr;
3021 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3022 int nallow, nblock;
3023 uint8_t mode, now, then;
3024 rectype_t crt, drt, nrt;
3025
3026 IN_MULTI_LOCK_ASSERT();
3027
3028 if (inm->inm_nsrc == 0 ||
3029 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3030 return (0);
3031
3032 ifp = inm->inm_ifp; /* interface */
3033 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3034 crt = REC_NONE; /* current group record type */
3035 drt = REC_NONE; /* mask of completed group record types */
3036 nrt = REC_NONE; /* record type for current node */
3037 m0srcs = 0; /* # source which will fit in current mbuf chain */
3038 nbytes = 0; /* # of bytes appended to group's state-change queue */
3039 npbytes = 0; /* # of bytes appended this packet */
3040 rsrcs = 0; /* # sources encoded in current record */
3041 schanged = 0; /* # nodes encoded in overall filter change */
3042 nallow = 0; /* # of source entries in ALLOW_NEW */
3043 nblock = 0; /* # of source entries in BLOCK_OLD */
3044 nims = NULL; /* next tree node pointer */
3045
3046 /*
3047 * For each possible filter record mode.
3048 * The first kind of source we encounter tells us which
3049 * is the first kind of record we start appending.
3050 * If a node transitioned to UNDEFINED at t1, its mode is treated
3051 * as the inverse of the group's filter mode.
3052 */
3053 while (drt != REC_FULL) {
3054 do {
3055 m0 = ifq->ifq_tail;
3056 if (m0 != NULL &&
3057 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3058 IGMP_V3_REPORT_MAXRECS) &&
3059 (m0->m_pkthdr.len + MINRECLEN) <
3060 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3061 m = m0;
3062 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3063 sizeof(struct igmp_grouprec)) /
3064 sizeof(in_addr_t);
3065 CTR1(KTR_IGMPV3,
3066 "%s: use previous packet", __func__);
3067 } else {
3068 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3069 if (m)
3070 m->m_data += IGMP_LEADINGSPACE;
3071 if (m == NULL) {
3072 m = m_gethdr(M_DONTWAIT, MT_DATA);
3073 if (m)
3074 MH_ALIGN(m, IGMP_LEADINGSPACE);
3075 }
3076 if (m == NULL) {
3077 CTR1(KTR_IGMPV3,
3078 "%s: m_get*() failed", __func__);
3079 return (-ENOMEM);
3080 }
3081 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3082 igmp_save_context(m, ifp);
3083 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3084 sizeof(struct igmp_grouprec)) /
3085 sizeof(in_addr_t);
3086 npbytes = 0;
3087 CTR1(KTR_IGMPV3,
3088 "%s: allocated new packet", __func__);
3089 }
3090 /*
3091 * Append the IGMP group record header to the
3092 * current packet's data area.
3093 * Recalculate pointer to free space for next
3094 * group record, in case m_append() allocated
3095 * a new mbuf or cluster.
3096 */
3097 memset(&ig, 0, sizeof(ig));
3098 ig.ig_group = inm->inm_addr;
3099 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3100 if (m != m0)
3101 m_freem(m);
3102 CTR1(KTR_IGMPV3,
3103 "%s: m_append() failed", __func__);
3104 return (-ENOMEM);
3105 }
3106 npbytes += sizeof(struct igmp_grouprec);
3107 if (m != m0) {
3108 /* new packet; offset in c hain */
3109 md = m_getptr(m, npbytes -
3110 sizeof(struct igmp_grouprec), &off);
3111 pig = (struct igmp_grouprec *)(mtod(md,
3112 uint8_t *) + off);
3113 } else {
3114 /* current packet; offset from last append */
3115 md = m_last(m);
3116 pig = (struct igmp_grouprec *)(mtod(md,
3117 uint8_t *) + md->m_len -
3118 sizeof(struct igmp_grouprec));
3119 }
3120 /*
3121 * Begin walking the tree for this record type
3122 * pass, or continue from where we left off
3123 * previously if we had to allocate a new packet.
3124 * Only report deltas in-mode at t1.
3125 * We need not report included sources as allowed
3126 * if we are in inclusive mode on the group,
3127 * however the converse is not true.
3128 */
3129 rsrcs = 0;
3130 if (nims == NULL)
3131 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3132 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3133 CTR2(KTR_IGMPV3, "%s: visit node %s",
3134 __func__, inet_ntoa_haddr(ims->ims_haddr));
3135 now = ims_get_mode(inm, ims, 1);
3136 then = ims_get_mode(inm, ims, 0);
3137 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3138 __func__, then, now);
3139 if (now == then) {
3140 CTR1(KTR_IGMPV3,
3141 "%s: skip unchanged", __func__);
3142 continue;
3143 }
3144 if (mode == MCAST_EXCLUDE &&
3145 now == MCAST_INCLUDE) {
3146 CTR1(KTR_IGMPV3,
3147 "%s: skip IN src on EX group",
3148 __func__);
3149 continue;
3150 }
3151 nrt = (rectype_t)now;
3152 if (nrt == REC_NONE)
3153 nrt = (rectype_t)(~mode & REC_FULL);
3154 if (schanged++ == 0) {
3155 crt = nrt;
3156 } else if (crt != nrt)
3157 continue;
3158 naddr = htonl(ims->ims_haddr);
3159 if (!m_append(m, sizeof(in_addr_t),
3160 (void *)&naddr)) {
3161 if (m != m0)
3162 m_freem(m);
3163 CTR1(KTR_IGMPV3,
3164 "%s: m_append() failed", __func__);
3165 return (-ENOMEM);
3166 }
3167 nallow += !!(crt == REC_ALLOW);
3168 nblock += !!(crt == REC_BLOCK);
3169 if (++rsrcs == m0srcs)
3170 break;
3171 }
3172 /*
3173 * If we did not append any tree nodes on this
3174 * pass, back out of allocations.
3175 */
3176 if (rsrcs == 0) {
3177 npbytes -= sizeof(struct igmp_grouprec);
3178 if (m != m0) {
3179 CTR1(KTR_IGMPV3,
3180 "%s: m_free(m)", __func__);
3181 m_freem(m);
3182 } else {
3183 CTR1(KTR_IGMPV3,
3184 "%s: m_adj(m, -ig)", __func__);
3185 m_adj(m, -((int)sizeof(
3186 struct igmp_grouprec)));
3187 }
3188 continue;
3189 }
3190 npbytes += (rsrcs * sizeof(in_addr_t));
3191 if (crt == REC_ALLOW)
3192 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3193 else if (crt == REC_BLOCK)
3194 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3195 pig->ig_numsrc = htons(rsrcs);
3196 /*
3197 * Count the new group record, and enqueue this
3198 * packet if it wasn't already queued.
3199 */
3200 m->m_pkthdr.PH_vt.vt_nrecs++;
3201 if (m != m0)
3202 _IF_ENQUEUE(ifq, m);
3203 nbytes += npbytes;
3204 } while (nims != NULL);
3205 drt |= crt;
3206 crt = (~crt & REC_FULL);
3207 }
3208
3209 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3210 nallow, nblock);
3211
3212 return (nbytes);
3213 }
3214
3215 static int
3216 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3217 {
3218 struct ifqueue *gq;
3219 struct mbuf *m; /* pending state-change */
3220 struct mbuf *m0; /* copy of pending state-change */
3221 struct mbuf *mt; /* last state-change in packet */
3222 int docopy, domerge;
3223 u_int recslen;
3224
3225 docopy = 0;
3226 domerge = 0;
3227 recslen = 0;
3228
3229 IN_MULTI_LOCK_ASSERT();
3230 IGMP_LOCK_ASSERT();
3231
3232 /*
3233 * If there are further pending retransmissions, make a writable
3234 * copy of each queued state-change message before merging.
3235 */
3236 if (inm->inm_scrv > 0)
3237 docopy = 1;
3238
3239 gq = &inm->inm_scq;
3240 #ifdef KTR
3241 if (gq->ifq_head == NULL) {
3242 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3243 __func__, inm);
3244 }
3245 #endif
3246
3247 m = gq->ifq_head;
3248 while (m != NULL) {
3249 /*
3250 * Only merge the report into the current packet if
3251 * there is sufficient space to do so; an IGMPv3 report
3252 * packet may only contain 65,535 group records.
3253 * Always use a simple mbuf chain concatentation to do this,
3254 * as large state changes for single groups may have
3255 * allocated clusters.
3256 */
3257 domerge = 0;
3258 mt = ifscq->ifq_tail;
3259 if (mt != NULL) {
3260 recslen = m_length(m, NULL);
3261
3262 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3263 m->m_pkthdr.PH_vt.vt_nrecs <=
3264 IGMP_V3_REPORT_MAXRECS) &&
3265 (mt->m_pkthdr.len + recslen <=
3266 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3267 domerge = 1;
3268 }
3269
3270 if (!domerge && _IF_QFULL(gq)) {
3271 CTR2(KTR_IGMPV3,
3272 "%s: outbound queue full, skipping whole packet %p",
3273 __func__, m);
3274 mt = m->m_nextpkt;
3275 if (!docopy)
3276 m_freem(m);
3277 m = mt;
3278 continue;
3279 }
3280
3281 if (!docopy) {
3282 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3283 _IF_DEQUEUE(gq, m0);
3284 m = m0->m_nextpkt;
3285 } else {
3286 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3287 m0 = m_dup(m, M_NOWAIT);
3288 if (m0 == NULL)
3289 return (ENOMEM);
3290 m0->m_nextpkt = NULL;
3291 m = m->m_nextpkt;
3292 }
3293
3294 if (!domerge) {
3295 CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3296 __func__, m0, ifscq);
3297 _IF_ENQUEUE(ifscq, m0);
3298 } else {
3299 struct mbuf *mtl; /* last mbuf of packet mt */
3300
3301 CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3302 __func__, m0, mt);
3303
3304 mtl = m_last(mt);
3305 m0->m_flags &= ~M_PKTHDR;
3306 mt->m_pkthdr.len += recslen;
3307 mt->m_pkthdr.PH_vt.vt_nrecs +=
3308 m0->m_pkthdr.PH_vt.vt_nrecs;
3309
3310 mtl->m_next = m0;
3311 }
3312 }
3313
3314 return (0);
3315 }
3316
3317 /*
3318 * Respond to a pending IGMPv3 General Query.
3319 */
3320 static void
3321 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3322 {
3323 struct ifmultiaddr *ifma, *tifma;
3324 struct ifnet *ifp;
3325 struct in_multi *inm;
3326 int retval, loop;
3327
3328 IN_MULTI_LOCK_ASSERT();
3329 IGMP_LOCK_ASSERT();
3330
3331 KASSERT(igi->igi_version == IGMP_VERSION_3,
3332 ("%s: called when version %d", __func__, igi->igi_version));
3333
3334 ifp = igi->igi_ifp;
3335
3336 IF_ADDR_LOCK(ifp);
3337 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
3338 if (ifma->ifma_addr->sa_family != AF_INET ||
3339 ifma->ifma_protospec == NULL)
3340 continue;
3341
3342 inm = (struct in_multi *)ifma->ifma_protospec;
3343 KASSERT(ifp == inm->inm_ifp,
3344 ("%s: inconsistent ifp", __func__));
3345
3346 switch (inm->inm_state) {
3347 case IGMP_NOT_MEMBER:
3348 case IGMP_SILENT_MEMBER:
3349 break;
3350 case IGMP_REPORTING_MEMBER:
3351 case IGMP_IDLE_MEMBER:
3352 case IGMP_LAZY_MEMBER:
3353 case IGMP_SLEEPING_MEMBER:
3354 case IGMP_AWAKENING_MEMBER:
3355 inm->inm_state = IGMP_REPORTING_MEMBER;
3356 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3357 inm, 0, 0, 0);
3358 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3359 __func__, retval);
3360 break;
3361 case IGMP_G_QUERY_PENDING_MEMBER:
3362 case IGMP_SG_QUERY_PENDING_MEMBER:
3363 case IGMP_LEAVING_MEMBER:
3364 break;
3365 }
3366 }
3367 IF_ADDR_UNLOCK(ifp);
3368
3369 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3370 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3371
3372 /*
3373 * Slew transmission of bursts over 500ms intervals.
3374 */
3375 if (igi->igi_gq.ifq_head != NULL) {
3376 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3377 IGMP_RESPONSE_BURST_INTERVAL);
3378 V_interface_timers_running = 1;
3379 }
3380 }
3381
3382 /*
3383 * Transmit the next pending IGMP message in the output queue.
3384 *
3385 * We get called from netisr_processqueue(). A mutex private to igmpoq
3386 * will be acquired and released around this routine.
3387 *
3388 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3389 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3390 * a link and uses a link-scope multicast address.
3391 */
3392 static void
3393 igmp_intr(struct mbuf *m)
3394 {
3395 struct ip_moptions imo;
3396 struct ifnet *ifp;
3397 struct mbuf *ipopts, *m0;
3398 int error;
3399 uint32_t ifindex;
3400
3401 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3402
3403 /*
3404 * Set VNET image pointer from enqueued mbuf chain
3405 * before doing anything else. Whilst we use interface
3406 * indexes to guard against interface detach, they are
3407 * unique to each VIMAGE and must be retrieved.
3408 */
3409 CURVNET_SET((struct vnet *)(m->m_pkthdr.header));
3410 ifindex = igmp_restore_context(m);
3411
3412 /*
3413 * Check if the ifnet still exists. This limits the scope of
3414 * any race in the absence of a global ifp lock for low cost
3415 * (an array lookup).
3416 */
3417 ifp = ifnet_byindex(ifindex);
3418 if (ifp == NULL) {
3419 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3420 __func__, m, ifindex);
3421 m_freem(m);
3422 IPSTAT_INC(ips_noroute);
3423 goto out;
3424 }
3425
3426 ipopts = V_igmp_sendra ? m_raopt : NULL;
3427
3428 imo.imo_multicast_ttl = 1;
3429 imo.imo_multicast_vif = -1;
3430 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3431
3432 /*
3433 * If the user requested that IGMP traffic be explicitly
3434 * redirected to the loopback interface (e.g. they are running a
3435 * MANET interface and the routing protocol needs to see the
3436 * updates), handle this now.
3437 */
3438 if (m->m_flags & M_IGMP_LOOP)
3439 imo.imo_multicast_ifp = V_loif;
3440 else
3441 imo.imo_multicast_ifp = ifp;
3442
3443 if (m->m_flags & M_IGMPV2) {
3444 m0 = m;
3445 } else {
3446 m0 = igmp_v3_encap_report(ifp, m);
3447 if (m0 == NULL) {
3448 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3449 m_freem(m);
3450 IPSTAT_INC(ips_odropped);
3451 goto out;
3452 }
3453 }
3454
3455 igmp_scrub_context(m0);
3456 m->m_flags &= ~(M_PROTOFLAGS);
3457 m0->m_pkthdr.rcvif = V_loif;
3458 #ifdef MAC
3459 mac_netinet_igmp_send(ifp, m0);
3460 #endif
3461 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3462 if (error) {
3463 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3464 goto out;
3465 }
3466
3467 IGMPSTAT_INC(igps_snd_reports);
3468
3469 out:
3470 /*
3471 * We must restore the existing vnet pointer before
3472 * continuing as we are run from netisr context.
3473 */
3474 CURVNET_RESTORE();
3475 }
3476
3477 /*
3478 * Encapsulate an IGMPv3 report.
3479 *
3480 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3481 * chain has already had its IP/IGMPv3 header prepended. In this case
3482 * the function will not attempt to prepend; the lengths and checksums
3483 * will however be re-computed.
3484 *
3485 * Returns a pointer to the new mbuf chain head, or NULL if the
3486 * allocation failed.
3487 */
3488 static struct mbuf *
3489 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3490 {
3491 struct igmp_report *igmp;
3492 struct ip *ip;
3493 int hdrlen, igmpreclen;
3494
3495 KASSERT((m->m_flags & M_PKTHDR),
3496 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3497
3498 igmpreclen = m_length(m, NULL);
3499 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3500
3501 if (m->m_flags & M_IGMPV3_HDR) {
3502 igmpreclen -= hdrlen;
3503 } else {
3504 M_PREPEND(m, hdrlen, M_DONTWAIT);
3505 if (m == NULL)
3506 return (NULL);
3507 m->m_flags |= M_IGMPV3_HDR;
3508 }
3509
3510 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3511
3512 m->m_data += sizeof(struct ip);
3513 m->m_len -= sizeof(struct ip);
3514
3515 igmp = mtod(m, struct igmp_report *);
3516 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3517 igmp->ir_rsv1 = 0;
3518 igmp->ir_rsv2 = 0;
3519 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3520 igmp->ir_cksum = 0;
3521 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3522 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3523
3524 m->m_data -= sizeof(struct ip);
3525 m->m_len += sizeof(struct ip);
3526
3527 ip = mtod(m, struct ip *);
3528 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3529 ip->ip_len = hdrlen + igmpreclen;
3530 ip->ip_off = IP_DF;
3531 ip->ip_p = IPPROTO_IGMP;
3532 ip->ip_sum = 0;
3533
3534 ip->ip_src.s_addr = INADDR_ANY;
3535
3536 if (m->m_flags & M_IGMP_LOOP) {
3537 struct in_ifaddr *ia;
3538
3539 IFP_TO_IA(ifp, ia);
3540 if (ia != NULL) {
3541 ip->ip_src = ia->ia_addr.sin_addr;
3542 ifa_free(&ia->ia_ifa);
3543 }
3544 }
3545
3546 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3547
3548 return (m);
3549 }
3550
3551 #ifdef KTR
3552 static char *
3553 igmp_rec_type_to_str(const int type)
3554 {
3555
3556 switch (type) {
3557 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3558 return "TO_EX";
3559 break;
3560 case IGMP_CHANGE_TO_INCLUDE_MODE:
3561 return "TO_IN";
3562 break;
3563 case IGMP_MODE_IS_EXCLUDE:
3564 return "MODE_EX";
3565 break;
3566 case IGMP_MODE_IS_INCLUDE:
3567 return "MODE_IN";
3568 break;
3569 case IGMP_ALLOW_NEW_SOURCES:
3570 return "ALLOW_NEW";
3571 break;
3572 case IGMP_BLOCK_OLD_SOURCES:
3573 return "BLOCK_OLD";
3574 break;
3575 default:
3576 break;
3577 }
3578 return "unknown";
3579 }
3580 #endif
3581
3582 static void
3583 igmp_init(void *unused __unused)
3584 {
3585
3586 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3587
3588 IGMP_LOCK_INIT();
3589
3590 m_raopt = igmp_ra_alloc();
3591
3592 netisr_register(&igmp_nh);
3593 }
3594 SYSINIT(igmp_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_init, NULL);
3595
3596 static void
3597 igmp_uninit(void *unused __unused)
3598 {
3599
3600 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3601
3602 netisr_unregister(&igmp_nh);
3603
3604 m_free(m_raopt);
3605 m_raopt = NULL;
3606
3607 IGMP_LOCK_DESTROY();
3608 }
3609 SYSUNINIT(igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_uninit, NULL);
3610
3611 static void
3612 vnet_igmp_init(const void *unused __unused)
3613 {
3614
3615 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3616
3617 LIST_INIT(&V_igi_head);
3618 }
3619 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_igmp_init,
3620 NULL);
3621
3622 static void
3623 vnet_igmp_uninit(const void *unused __unused)
3624 {
3625
3626 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3627
3628 KASSERT(LIST_EMPTY(&V_igi_head),
3629 ("%s: igi list not empty; ifnets not detached?", __func__));
3630 }
3631 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY,
3632 vnet_igmp_uninit, NULL);
3633
3634 static int
3635 igmp_modevent(module_t mod, int type, void *unused __unused)
3636 {
3637
3638 switch (type) {
3639 case MOD_LOAD:
3640 case MOD_UNLOAD:
3641 break;
3642 default:
3643 return (EOPNOTSUPP);
3644 }
3645 return (0);
3646 }
3647
3648 static moduledata_t igmp_mod = {
3649 "igmp",
3650 igmp_modevent,
3651 0
3652 };
3653 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
Cache object: 3d5d465d83a3d9c79640c6b840415375
|