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.
7 * This code is derived from software contributed to Berkeley by
8 * Stephen Deering of Stanford University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
34 * @(#)igmp.c 8.1 (Berkeley) 7/19/93
38 * Internet Group Management Protocol (IGMP) routines.
39 * [RFC1112, RFC2236, RFC3376]
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.
47 * MULTICAST Revision: 3.5.1.4
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/module.h>
56 #include <sys/malloc.h>
58 #include <sys/socket.h>
59 #include <sys/protosw.h>
60 #include <sys/kernel.h>
61 #include <sys/sysctl.h>
63 #include <sys/condvar.h>
66 #include <net/netisr.h>
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>
78 #include <machine/in_cksum.h>
80 #include <security/mac/mac_framework.h>
83 #define KTR_IGMPV3 KTR_INET
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 *,
97 static int igmp_input_v2_query(struct ifnet *, const struct ip *,
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);
112 static char * igmp_rec_type_to_str(const int);
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 *);
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 *,
128 static void igmp_v3_process_group_timers(struct igmp_ifinfo *,
129 struct ifqueue *, struct ifqueue *, struct in_multi *,
131 static int igmp_v3_merge_state_changes(struct in_multi *,
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);
138 static const struct netisr_handler igmp_nh = {
140 .nh_handler = igmp_intr,
141 .nh_proto = NETISR_IGMP,
142 .nh_policy = NETISR_POLICY_SOURCE,
146 * System-wide globals.
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.
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.
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).
176 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
177 * to a vnet in ifp->if_vnet.
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.
187 struct mbuf *m_raopt; /* Router Alert option */
188 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
191 * VIMAGE-wide globals.
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
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.
209 static VNET_DEFINE(int, interface_timers_running); /* IGMPv3 general
211 static VNET_DEFINE(int, state_change_timers_running); /* IGMPv3 state-change
213 static VNET_DEFINE(int, current_state_timers_running); /* IGMPv1/v2 host
214 * report; IGMPv3 g/sg
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)
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),
226 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
228 #define V_igi_head VNET(igi_head)
229 #define V_igmpstat VNET(igmpstat)
230 #define V_igmp_gsrdelay VNET(igmp_gsrdelay)
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;
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)
249 * Virtualized sysctls.
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");
281 * Non-virtualized sysctls.
283 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
284 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
285 "Per-interface IGMPv3 state");
288 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
292 m->m_pkthdr.header = ifp->if_vnet;
294 m->m_pkthdr.flowid = ifp->if_index;
298 igmp_scrub_context(struct mbuf *m)
301 m->m_pkthdr.header = NULL;
302 m->m_pkthdr.flowid = 0;
306 static __inline char *
307 inet_ntoa_haddr(in_addr_t haddr)
311 ia.s_addr = htonl(haddr);
312 return (inet_ntoa(ia));
317 * Restore context from a queued IGMP output chain.
318 * Return saved ifindex.
320 * VIMAGE: The assertion is there to make sure that we
321 * actually called CURVNET_SET() with what's in the mbuf chain.
323 static __inline uint32_t
324 igmp_restore_context(struct mbuf *m)
328 #if defined(VIMAGE) && defined(INVARIANTS)
329 KASSERT(curvnet == (m->m_pkthdr.header),
330 ("%s: called when curvnet was not restored", __func__));
333 return (m->m_pkthdr.flowid);
337 * Retrieve or set default IGMP version.
339 * VIMAGE: Assume curvnet set by caller.
340 * SMPng: NOTE: Serialized by IGMP lock.
343 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
348 error = sysctl_wire_old_buffer(req, sizeof(int));
354 new = V_igmp_default_version;
356 error = sysctl_handle_int(oidp, &new, 0, req);
357 if (error || !req->newptr)
360 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
365 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
366 V_igmp_default_version, new);
368 V_igmp_default_version = new;
376 * Retrieve or set threshold between group-source queries in seconds.
378 * VIMAGE: Assume curvnet set by caller.
379 * SMPng: NOTE: Serialized by IGMP lock.
382 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
387 error = sysctl_wire_old_buffer(req, sizeof(int));
393 i = V_igmp_gsrdelay.tv_sec;
395 error = sysctl_handle_int(oidp, &i, 0, req);
396 if (error || !req->newptr)
399 if (i < -1 || i >= 60) {
404 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
405 V_igmp_gsrdelay.tv_sec, i);
406 V_igmp_gsrdelay.tv_sec = i;
414 * Expose struct igmp_ifinfo to userland, keyed by ifindex.
415 * For use by ifmcstat(8).
417 * SMPng: NOTE: Does an unlocked ifindex space read.
418 * VIMAGE: Assume curvnet set by caller. The node handler itself
419 * is not directly virtualized.
422 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
428 struct igmp_ifinfo *igi;
433 if (req->newptr != NULL)
439 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
446 if (name[0] <= 0 || name[0] > V_if_index) {
453 ifp = ifnet_byindex(name[0]);
457 LIST_FOREACH(igi, &V_igi_head, igi_link) {
458 if (ifp == igi->igi_ifp) {
459 error = SYSCTL_OUT(req, igi,
460 sizeof(struct igmp_ifinfo));
472 * Dispatch an entire queue of pending packet chains
474 * VIMAGE: Assumes the vnet pointer has been set.
477 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop)
485 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m);
487 m->m_flags |= M_IGMP_LOOP;
488 netisr_dispatch(NETISR_IGMP, m);
495 * Filter outgoing IGMP report state by group.
497 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
498 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
499 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
500 * this may break certain IGMP snooping switches which rely on the old
503 * Return zero if the given group is one for which IGMP reports
504 * should be suppressed, or non-zero if reports should be issued.
507 igmp_isgroupreported(const struct in_addr addr)
510 if (in_allhosts(addr) ||
511 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
518 * Construct a Router Alert option to use in outgoing packets.
526 MGET(m, M_DONTWAIT, MT_DATA);
527 p = mtod(m, struct ipoption *);
528 p->ipopt_dst.s_addr = INADDR_ANY;
529 p->ipopt_list[0] = IPOPT_RA; /* Router Alert Option */
530 p->ipopt_list[1] = 0x04; /* 4 bytes long */
531 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
532 p->ipopt_list[3] = 0x00; /* pad byte */
533 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
539 * Attach IGMP when PF_INET is attached to an interface.
542 igmp_domifattach(struct ifnet *ifp)
544 struct igmp_ifinfo *igi;
546 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
547 __func__, ifp, ifp->if_xname);
551 igi = igi_alloc_locked(ifp);
552 if (!(ifp->if_flags & IFF_MULTICAST))
553 igi->igi_flags |= IGIF_SILENT;
561 * VIMAGE: assume curvnet set by caller.
563 static struct igmp_ifinfo *
564 igi_alloc_locked(/*const*/ struct ifnet *ifp)
566 struct igmp_ifinfo *igi;
570 igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO);
575 igi->igi_version = V_igmp_default_version;
577 igi->igi_rv = IGMP_RV_INIT;
578 igi->igi_qi = IGMP_QI_INIT;
579 igi->igi_qri = IGMP_QRI_INIT;
580 igi->igi_uri = IGMP_URI_INIT;
582 SLIST_INIT(&igi->igi_relinmhead);
585 * Responses to general queries are subject to bounds.
587 IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
589 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
591 CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)",
601 * NOTE: Some finalization tasks need to run before the protocol domain
602 * is detached, but also before the link layer does its cleanup.
604 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
605 * XXX This is also bitten by unlocked ifma_protospec access.
608 igmp_ifdetach(struct ifnet *ifp)
610 struct igmp_ifinfo *igi;
611 struct ifmultiaddr *ifma;
612 struct in_multi *inm, *tinm;
614 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
619 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
620 if (igi->igi_version == IGMP_VERSION_3) {
622 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
623 if (ifma->ifma_addr->sa_family != AF_INET ||
624 ifma->ifma_protospec == NULL)
627 KASSERT(ifma->ifma_protospec != NULL,
628 ("%s: ifma_protospec is NULL", __func__));
630 inm = (struct in_multi *)ifma->ifma_protospec;
631 if (inm->inm_state == IGMP_LEAVING_MEMBER) {
632 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
635 inm_clear_recorded(inm);
637 IF_ADDR_RUNLOCK(ifp);
639 * Free the in_multi reference(s) for this IGMP lifecycle.
641 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
643 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
644 inm_release_locked(inm);
652 * Hook for domifdetach.
655 igmp_domifdetach(struct ifnet *ifp)
657 struct igmp_ifinfo *igi;
659 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
660 __func__, ifp, ifp->if_xname);
664 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
665 igi_delete_locked(ifp);
671 igi_delete_locked(const struct ifnet *ifp)
673 struct igmp_ifinfo *igi, *tigi;
675 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
676 __func__, ifp, ifp->if_xname);
680 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
681 if (igi->igi_ifp == ifp) {
683 * Free deferred General Query responses.
685 _IF_DRAIN(&igi->igi_gq);
687 LIST_REMOVE(igi, igi_link);
689 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
690 ("%s: there are dangling in_multi references",
699 panic("%s: igmp_ifinfo not found for ifp %p\n", __func__, ifp);
704 * Process a received IGMPv1 query.
705 * Return non-zero if the message should be dropped.
707 * VIMAGE: The curvnet pointer is derived from the input ifp.
710 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
711 const struct igmp *igmp)
713 struct ifmultiaddr *ifma;
714 struct igmp_ifinfo *igi;
715 struct in_multi *inm;
718 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
719 * 224.0.0.1. They are always treated as General Queries.
720 * igmp_group is always ignored. Do not drop it as a userland
721 * daemon may wish to see it.
722 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
724 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
725 IGMPSTAT_INC(igps_rcv_badqueries);
728 IGMPSTAT_INC(igps_rcv_gen_queries);
733 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
734 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
736 if (igi->igi_flags & IGIF_LOOPBACK) {
737 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
743 * Switch to IGMPv1 host compatibility mode.
745 igmp_set_version(igi, IGMP_VERSION_1);
747 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
750 * Start the timers in all of our group records
751 * for the interface on which the query arrived,
752 * except those which are already running.
755 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
756 if (ifma->ifma_addr->sa_family != AF_INET ||
757 ifma->ifma_protospec == NULL)
759 inm = (struct in_multi *)ifma->ifma_protospec;
760 if (inm->inm_timer != 0)
762 switch (inm->inm_state) {
763 case IGMP_NOT_MEMBER:
764 case IGMP_SILENT_MEMBER:
766 case IGMP_G_QUERY_PENDING_MEMBER:
767 case IGMP_SG_QUERY_PENDING_MEMBER:
768 case IGMP_REPORTING_MEMBER:
769 case IGMP_IDLE_MEMBER:
770 case IGMP_LAZY_MEMBER:
771 case IGMP_SLEEPING_MEMBER:
772 case IGMP_AWAKENING_MEMBER:
773 inm->inm_state = IGMP_REPORTING_MEMBER;
774 inm->inm_timer = IGMP_RANDOM_DELAY(
775 IGMP_V1V2_MAX_RI * PR_FASTHZ);
776 V_current_state_timers_running = 1;
778 case IGMP_LEAVING_MEMBER:
782 IF_ADDR_RUNLOCK(ifp);
792 * Process a received IGMPv2 general or group-specific query.
795 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
796 const struct igmp *igmp)
798 struct ifmultiaddr *ifma;
799 struct igmp_ifinfo *igi;
800 struct in_multi *inm;
801 int is_general_query;
804 is_general_query = 0;
807 * Validate address fields upfront.
808 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
810 if (in_nullhost(igmp->igmp_group)) {
812 * IGMPv2 General Query.
813 * If this was not sent to the all-hosts group, ignore it.
815 if (!in_allhosts(ip->ip_dst))
817 IGMPSTAT_INC(igps_rcv_gen_queries);
818 is_general_query = 1;
820 /* IGMPv2 Group-Specific Query. */
821 IGMPSTAT_INC(igps_rcv_group_queries);
827 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
828 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
830 if (igi->igi_flags & IGIF_LOOPBACK) {
831 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
837 * Ignore v2 query if in v1 Compatibility Mode.
839 if (igi->igi_version == IGMP_VERSION_1)
842 igmp_set_version(igi, IGMP_VERSION_2);
844 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
848 if (is_general_query) {
850 * For each reporting group joined on this
851 * interface, kick the report timer.
853 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
856 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
857 if (ifma->ifma_addr->sa_family != AF_INET ||
858 ifma->ifma_protospec == NULL)
860 inm = (struct in_multi *)ifma->ifma_protospec;
861 igmp_v2_update_group(inm, timer);
863 IF_ADDR_RUNLOCK(ifp);
866 * Group-specific IGMPv2 query, we need only
867 * look up the single group to process it.
869 inm = inm_lookup(ifp, igmp->igmp_group);
871 CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
872 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
873 igmp_v2_update_group(inm, timer);
885 * Update the report timer on a group in response to an IGMPv2 query.
887 * If we are becoming the reporting member for this group, start the timer.
888 * If we already are the reporting member for this group, and timer is
889 * below the threshold, reset it.
891 * We may be updating the group for the first time since we switched
892 * to IGMPv3. If we are, then we must clear any recorded source lists,
893 * and transition to REPORTING state; the group timer is overloaded
894 * for group and group-source query responses.
896 * Unlike IGMPv3, the delay per group should be jittered
897 * to avoid bursts of IGMPv2 reports.
900 igmp_v2_update_group(struct in_multi *inm, const int timer)
903 CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
904 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
906 IN_MULTI_LOCK_ASSERT();
908 switch (inm->inm_state) {
909 case IGMP_NOT_MEMBER:
910 case IGMP_SILENT_MEMBER:
912 case IGMP_REPORTING_MEMBER:
913 if (inm->inm_timer != 0 &&
914 inm->inm_timer <= timer) {
915 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
916 "skipping.", __func__);
920 case IGMP_SG_QUERY_PENDING_MEMBER:
921 case IGMP_G_QUERY_PENDING_MEMBER:
922 case IGMP_IDLE_MEMBER:
923 case IGMP_LAZY_MEMBER:
924 case IGMP_AWAKENING_MEMBER:
925 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
926 inm->inm_state = IGMP_REPORTING_MEMBER;
927 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
928 V_current_state_timers_running = 1;
930 case IGMP_SLEEPING_MEMBER:
931 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
932 inm->inm_state = IGMP_AWAKENING_MEMBER;
934 case IGMP_LEAVING_MEMBER:
940 * Process a received IGMPv3 general, group-specific or
941 * group-and-source-specific query.
942 * Assumes m has already been pulled up to the full IGMP message length.
943 * Return 0 if successful, otherwise an appropriate error code is returned.
946 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
947 /*const*/ struct igmpv3 *igmpv3)
949 struct igmp_ifinfo *igi;
950 struct in_multi *inm;
951 int is_general_query;
952 uint32_t maxresp, nsrc, qqi;
956 is_general_query = 0;
958 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
960 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
961 if (maxresp >= 128) {
962 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
963 (IGMP_EXP(igmpv3->igmp_code) + 3);
967 * Robustness must never be less than 2 for on-wire IGMPv3.
968 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
969 * an exception for interfaces whose IGMPv3 state changes
970 * are redirected to loopback (e.g. MANET).
972 qrv = IGMP_QRV(igmpv3->igmp_misc);
974 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
979 qqi = igmpv3->igmp_qqi;
981 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
982 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
985 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
989 nsrc = ntohs(igmpv3->igmp_numsrc);
992 * Validate address fields and versions upfront before
993 * accepting v3 query.
994 * XXX SMPng: Unlocked access to igmpstat counters here.
996 if (in_nullhost(igmpv3->igmp_group)) {
998 * IGMPv3 General Query.
1000 * General Queries SHOULD be directed to 224.0.0.1.
1001 * A general query with a source list has undefined
1002 * behaviour; discard it.
1004 IGMPSTAT_INC(igps_rcv_gen_queries);
1005 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1006 IGMPSTAT_INC(igps_rcv_badqueries);
1009 is_general_query = 1;
1011 /* Group or group-source specific query. */
1013 IGMPSTAT_INC(igps_rcv_group_queries);
1015 IGMPSTAT_INC(igps_rcv_gsr_queries);
1021 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1022 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
1024 if (igi->igi_flags & IGIF_LOOPBACK) {
1025 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1026 ifp, ifp->if_xname);
1031 * Discard the v3 query if we're in Compatibility Mode.
1032 * The RFC is not obviously worded that hosts need to stay in
1033 * compatibility mode until the Old Version Querier Present
1036 if (igi->igi_version != IGMP_VERSION_3) {
1037 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1038 igi->igi_version, ifp, ifp->if_xname);
1042 igmp_set_version(igi, IGMP_VERSION_3);
1045 igi->igi_qri = maxresp;
1047 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1050 if (is_general_query) {
1052 * Schedule a current-state report on this ifp for
1053 * all groups, possibly containing source lists.
1054 * If there is a pending General Query response
1055 * scheduled earlier than the selected delay, do
1056 * not schedule any other reports.
1057 * Otherwise, reset the interface timer.
1059 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1060 ifp, ifp->if_xname);
1061 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1062 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1063 V_interface_timers_running = 1;
1067 * Group-source-specific queries are throttled on
1068 * a per-group basis to defeat denial-of-service attempts.
1069 * Queries for groups we are not a member of on this
1070 * link are simply ignored.
1072 inm = inm_lookup(ifp, igmpv3->igmp_group);
1076 if (!ratecheck(&inm->inm_lastgsrtv,
1077 &V_igmp_gsrdelay)) {
1078 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1080 IGMPSTAT_INC(igps_drop_gsr_queries);
1084 CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1085 inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1087 * If there is a pending General Query response
1088 * scheduled sooner than the selected delay, no
1089 * further report need be scheduled.
1090 * Otherwise, prepare to respond to the
1091 * group-specific or group-and-source query.
1093 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1094 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1105 * Process a recieved IGMPv3 group-specific or group-and-source-specific
1107 * Return <0 if any error occured. Currently this is ignored.
1110 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1111 int timer, /*const*/ struct igmpv3 *igmpv3)
1116 IN_MULTI_LOCK_ASSERT();
1121 switch (inm->inm_state) {
1122 case IGMP_NOT_MEMBER:
1123 case IGMP_SILENT_MEMBER:
1124 case IGMP_SLEEPING_MEMBER:
1125 case IGMP_LAZY_MEMBER:
1126 case IGMP_AWAKENING_MEMBER:
1127 case IGMP_IDLE_MEMBER:
1128 case IGMP_LEAVING_MEMBER:
1131 case IGMP_REPORTING_MEMBER:
1132 case IGMP_G_QUERY_PENDING_MEMBER:
1133 case IGMP_SG_QUERY_PENDING_MEMBER:
1137 nsrc = ntohs(igmpv3->igmp_numsrc);
1140 * Deal with group-specific queries upfront.
1141 * If any group query is already pending, purge any recorded
1142 * source-list state if it exists, and schedule a query response
1143 * for this group-specific query.
1146 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1147 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1148 inm_clear_recorded(inm);
1149 timer = min(inm->inm_timer, timer);
1151 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1152 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1153 V_current_state_timers_running = 1;
1158 * Deal with the case where a group-and-source-specific query has
1159 * been received but a group-specific query is already pending.
1161 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1162 timer = min(inm->inm_timer, timer);
1163 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1164 V_current_state_timers_running = 1;
1169 * Finally, deal with the case where a group-and-source-specific
1170 * query has been received, where a response to a previous g-s-r
1171 * query exists, or none exists.
1172 * In this case, we need to parse the source-list which the Querier
1173 * has provided us with and check if we have any source list filter
1174 * entries at T1 for these sources. If we do not, there is no need
1175 * schedule a report and the query may be dropped.
1176 * If we do, we must record them and schedule a current-state
1177 * report for those sources.
1178 * FIXME: Handling source lists larger than 1 mbuf requires that
1179 * we pass the mbuf chain pointer down to this function, and use
1180 * m_getptr() to walk the chain.
1182 if (inm->inm_nsrc > 0) {
1183 const struct in_addr *ap;
1186 ap = (const struct in_addr *)(igmpv3 + 1);
1188 for (i = 0; i < nsrc; i++, ap++) {
1189 retval = inm_record_source(inm, ap->s_addr);
1192 nrecorded += retval;
1194 if (nrecorded > 0) {
1196 "%s: schedule response to SG query", __func__);
1197 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1198 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1199 V_current_state_timers_running = 1;
1207 * Process a received IGMPv1 host membership report.
1209 * NOTE: 0.0.0.0 workaround breaks const correctness.
1212 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1213 /*const*/ struct igmp *igmp)
1215 struct in_ifaddr *ia;
1216 struct in_multi *inm;
1218 IGMPSTAT_INC(igps_rcv_reports);
1220 if (ifp->if_flags & IFF_LOOPBACK)
1223 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1224 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1225 IGMPSTAT_INC(igps_rcv_badreports);
1230 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1231 * Booting clients may use the source address 0.0.0.0. Some
1232 * IGMP daemons may not know how to use IP_RECVIF to determine
1233 * the interface upon which this message was received.
1234 * Replace 0.0.0.0 with the subnet address if told to do so.
1236 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1239 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1240 ifa_free(&ia->ia_ifa);
1244 CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1245 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1248 * IGMPv1 report suppression.
1249 * If we are a member of this group, and our membership should be
1250 * reported, stop our group timer and transition to the 'lazy' state.
1253 inm = inm_lookup(ifp, igmp->igmp_group);
1255 struct igmp_ifinfo *igi;
1259 KASSERT(igi != NULL,
1260 ("%s: no igi for ifp %p", __func__, ifp));
1264 IGMPSTAT_INC(igps_rcv_ourreports);
1267 * If we are in IGMPv3 host mode, do not allow the
1268 * other host's IGMPv1 report to suppress our reports
1269 * unless explicitly configured to do so.
1271 if (igi->igi_version == IGMP_VERSION_3) {
1272 if (V_igmp_legacysupp)
1273 igmp_v3_suppress_group_record(inm);
1279 switch (inm->inm_state) {
1280 case IGMP_NOT_MEMBER:
1281 case IGMP_SILENT_MEMBER:
1283 case IGMP_IDLE_MEMBER:
1284 case IGMP_LAZY_MEMBER:
1285 case IGMP_AWAKENING_MEMBER:
1287 "report suppressed for %s on ifp %p(%s)",
1288 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1289 case IGMP_SLEEPING_MEMBER:
1290 inm->inm_state = IGMP_SLEEPING_MEMBER;
1292 case IGMP_REPORTING_MEMBER:
1294 "report suppressed for %s on ifp %p(%s)",
1295 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1296 if (igi->igi_version == IGMP_VERSION_1)
1297 inm->inm_state = IGMP_LAZY_MEMBER;
1298 else if (igi->igi_version == IGMP_VERSION_2)
1299 inm->inm_state = IGMP_SLEEPING_MEMBER;
1301 case IGMP_G_QUERY_PENDING_MEMBER:
1302 case IGMP_SG_QUERY_PENDING_MEMBER:
1303 case IGMP_LEAVING_MEMBER:
1315 * Process a received IGMPv2 host membership report.
1317 * NOTE: 0.0.0.0 workaround breaks const correctness.
1320 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1321 /*const*/ struct igmp *igmp)
1323 struct in_ifaddr *ia;
1324 struct in_multi *inm;
1327 * Make sure we don't hear our own membership report. Fast
1328 * leave requires knowing that we are the only member of a
1332 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1333 ifa_free(&ia->ia_ifa);
1337 IGMPSTAT_INC(igps_rcv_reports);
1339 if (ifp->if_flags & IFF_LOOPBACK) {
1341 ifa_free(&ia->ia_ifa);
1345 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1346 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1348 ifa_free(&ia->ia_ifa);
1349 IGMPSTAT_INC(igps_rcv_badreports);
1354 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1355 * Booting clients may use the source address 0.0.0.0. Some
1356 * IGMP daemons may not know how to use IP_RECVIF to determine
1357 * the interface upon which this message was received.
1358 * Replace 0.0.0.0 with the subnet address if told to do so.
1360 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1362 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1365 ifa_free(&ia->ia_ifa);
1367 CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1368 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1371 * IGMPv2 report suppression.
1372 * If we are a member of this group, and our membership should be
1373 * reported, and our group timer is pending or about to be reset,
1374 * stop our group timer by transitioning to the 'lazy' state.
1377 inm = inm_lookup(ifp, igmp->igmp_group);
1379 struct igmp_ifinfo *igi;
1382 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1384 IGMPSTAT_INC(igps_rcv_ourreports);
1387 * If we are in IGMPv3 host mode, do not allow the
1388 * other host's IGMPv1 report to suppress our reports
1389 * unless explicitly configured to do so.
1391 if (igi->igi_version == IGMP_VERSION_3) {
1392 if (V_igmp_legacysupp)
1393 igmp_v3_suppress_group_record(inm);
1399 switch (inm->inm_state) {
1400 case IGMP_NOT_MEMBER:
1401 case IGMP_SILENT_MEMBER:
1402 case IGMP_SLEEPING_MEMBER:
1404 case IGMP_REPORTING_MEMBER:
1405 case IGMP_IDLE_MEMBER:
1406 case IGMP_AWAKENING_MEMBER:
1408 "report suppressed for %s on ifp %p(%s)",
1409 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1410 case IGMP_LAZY_MEMBER:
1411 inm->inm_state = IGMP_LAZY_MEMBER;
1413 case IGMP_G_QUERY_PENDING_MEMBER:
1414 case IGMP_SG_QUERY_PENDING_MEMBER:
1415 case IGMP_LEAVING_MEMBER:
1427 igmp_input(struct mbuf *m, int off)
1437 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1439 ifp = m->m_pkthdr.rcvif;
1441 IGMPSTAT_INC(igps_rcv_total);
1443 ip = mtod(m, struct ip *);
1445 igmplen = ip->ip_len;
1450 if (igmplen < IGMP_MINLEN) {
1451 IGMPSTAT_INC(igps_rcv_tooshort);
1457 * Always pullup to the minimum size for v1/v2 or v3
1458 * to amortize calls to m_pullup().
1461 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1462 minlen += IGMP_V3_QUERY_MINLEN;
1464 minlen += IGMP_MINLEN;
1465 if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1466 (m = m_pullup(m, minlen)) == 0) {
1467 IGMPSTAT_INC(igps_rcv_tooshort);
1470 ip = mtod(m, struct ip *);
1473 * Validate checksum.
1475 m->m_data += iphlen;
1477 igmp = mtod(m, struct igmp *);
1478 if (in_cksum(m, igmplen)) {
1479 IGMPSTAT_INC(igps_rcv_badsum);
1483 m->m_data -= iphlen;
1487 * IGMP control traffic is link-scope, and must have a TTL of 1.
1488 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1489 * probe packets may come from beyond the LAN.
1491 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1492 IGMPSTAT_INC(igps_rcv_badttl);
1497 switch (igmp->igmp_type) {
1498 case IGMP_HOST_MEMBERSHIP_QUERY:
1499 if (igmplen == IGMP_MINLEN) {
1500 if (igmp->igmp_code == 0)
1501 queryver = IGMP_VERSION_1;
1503 queryver = IGMP_VERSION_2;
1504 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1505 queryver = IGMP_VERSION_3;
1507 IGMPSTAT_INC(igps_rcv_tooshort);
1513 case IGMP_VERSION_1:
1514 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1515 if (!V_igmp_v1enable)
1517 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1523 case IGMP_VERSION_2:
1524 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1525 if (!V_igmp_v2enable)
1527 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1533 case IGMP_VERSION_3: {
1534 struct igmpv3 *igmpv3;
1539 IGMPSTAT_INC(igps_rcv_v3_queries);
1540 igmpv3 = (struct igmpv3 *)igmp;
1542 * Validate length based on source count.
1544 nsrc = ntohs(igmpv3->igmp_numsrc);
1545 srclen = sizeof(struct in_addr) * nsrc;
1546 if (nsrc * sizeof(in_addr_t) > srclen) {
1547 IGMPSTAT_INC(igps_rcv_tooshort);
1551 * m_pullup() may modify m, so pullup in
1554 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1556 if ((m->m_flags & M_EXT ||
1557 m->m_len < igmpv3len) &&
1558 (m = m_pullup(m, igmpv3len)) == NULL) {
1559 IGMPSTAT_INC(igps_rcv_tooshort);
1562 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1564 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1573 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1574 if (!V_igmp_v1enable)
1576 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1582 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1583 if (!V_igmp_v2enable)
1585 if (!ip_checkrouteralert(m))
1586 IGMPSTAT_INC(igps_rcv_nora);
1587 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1593 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1595 * Hosts do not need to process IGMPv3 membership reports,
1596 * as report suppression is no longer required.
1598 if (!ip_checkrouteralert(m))
1599 IGMPSTAT_INC(igps_rcv_nora);
1607 * Pass all valid IGMP packets up to any process(es) listening on a
1615 * Fast timeout handler (global).
1616 * VIMAGE: Timeout handlers are expected to service all vimages.
1621 VNET_ITERATOR_DECL(vnet_iter);
1623 VNET_LIST_RLOCK_NOSLEEP();
1624 VNET_FOREACH(vnet_iter) {
1625 CURVNET_SET(vnet_iter);
1626 igmp_fasttimo_vnet();
1629 VNET_LIST_RUNLOCK_NOSLEEP();
1633 * Fast timeout handler (per-vnet).
1634 * Sends are shuffled off to a netisr to deal with Giant.
1636 * VIMAGE: Assume caller has set up our curvnet.
1639 igmp_fasttimo_vnet(void)
1641 struct ifqueue scq; /* State-change packets */
1642 struct ifqueue qrq; /* Query response packets */
1644 struct igmp_ifinfo *igi;
1645 struct ifmultiaddr *ifma;
1646 struct in_multi *inm;
1647 int loop, uri_fasthz;
1653 * Quick check to see if any work needs to be done, in order to
1654 * minimize the overhead of fasttimo processing.
1655 * SMPng: XXX Unlocked reads.
1657 if (!V_current_state_timers_running &&
1658 !V_interface_timers_running &&
1659 !V_state_change_timers_running)
1666 * IGMPv3 General Query response timer processing.
1668 if (V_interface_timers_running) {
1669 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1671 V_interface_timers_running = 0;
1672 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1673 if (igi->igi_v3_timer == 0) {
1675 } else if (--igi->igi_v3_timer == 0) {
1676 igmp_v3_dispatch_general_query(igi);
1678 V_interface_timers_running = 1;
1683 if (!V_current_state_timers_running &&
1684 !V_state_change_timers_running)
1687 V_current_state_timers_running = 0;
1688 V_state_change_timers_running = 0;
1690 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1693 * IGMPv1/v2/v3 host report and state-change timer processing.
1694 * Note: Processing a v3 group timer may remove a node.
1696 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1699 if (igi->igi_version == IGMP_VERSION_3) {
1700 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1701 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1704 memset(&qrq, 0, sizeof(struct ifqueue));
1705 IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1707 memset(&scq, 0, sizeof(struct ifqueue));
1708 IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1712 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1713 if (ifma->ifma_addr->sa_family != AF_INET ||
1714 ifma->ifma_protospec == NULL)
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,
1723 case IGMP_VERSION_3:
1724 igmp_v3_process_group_timers(igi, &qrq,
1725 &scq, inm, uri_fasthz);
1729 IF_ADDR_RUNLOCK(ifp);
1731 if (igi->igi_version == IGMP_VERSION_3) {
1732 struct in_multi *tinm;
1734 igmp_dispatch_queue(&qrq, 0, loop);
1735 igmp_dispatch_queue(&scq, 0, loop);
1738 * Free the in_multi reference(s) for this
1741 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1743 SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1745 inm_release_locked(inm);
1756 * Update host report group timer for IGMPv1/v2.
1757 * Will update the global pending timer flags.
1760 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1762 int report_timer_expired;
1764 IN_MULTI_LOCK_ASSERT();
1767 if (inm->inm_timer == 0) {
1768 report_timer_expired = 0;
1769 } else if (--inm->inm_timer == 0) {
1770 report_timer_expired = 1;
1772 V_current_state_timers_running = 1;
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:
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);
1793 case IGMP_G_QUERY_PENDING_MEMBER:
1794 case IGMP_SG_QUERY_PENDING_MEMBER:
1795 case IGMP_LEAVING_MEMBER:
1801 * Update a group's timers for IGMPv3.
1802 * Will update the global pending timer flags.
1803 * Note: Unlocked read from igi.
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)
1810 int query_response_timer_expired;
1811 int state_change_retransmit_timer_expired;
1813 IN_MULTI_LOCK_ASSERT();
1816 query_response_timer_expired = 0;
1817 state_change_retransmit_timer_expired = 0;
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.
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;
1830 V_current_state_timers_running = 1;
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;
1838 V_state_change_timers_running = 1;
1841 /* We are in fasttimo, so be quick about it. */
1842 if (!state_change_retransmit_timer_expired &&
1843 !query_response_timer_expired)
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:
1854 case IGMP_G_QUERY_PENDING_MEMBER:
1855 case IGMP_SG_QUERY_PENDING_MEMBER:
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.
1862 if (query_response_timer_expired) {
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",
1869 inm->inm_state = IGMP_REPORTING_MEMBER;
1870 /* XXX Clear recorded sources for next time. */
1871 inm_clear_recorded(inm);
1874 case IGMP_REPORTING_MEMBER:
1875 case IGMP_LEAVING_MEMBER:
1876 if (state_change_retransmit_timer_expired) {
1878 * State-change retransmission timer fired.
1879 * If there are any further pending retransmissions,
1880 * set the global pending state-change flag, and
1883 if (--inm->inm_scrv > 0) {
1884 inm->inm_sctimer = uri_fasthz;
1885 V_state_change_timers_running = 1;
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
1894 (void)igmp_v3_merge_state_changes(inm, scq);
1897 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1898 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
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.
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,
1920 * Suppress a group's pending response to a group or source/group query.
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.
1929 igmp_v3_suppress_group_record(struct in_multi *inm)
1932 IN_MULTI_LOCK_ASSERT();
1934 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1935 ("%s: not IGMPv3 mode on link", __func__));
1937 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1938 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1941 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1942 inm_clear_recorded(inm);
1945 inm->inm_state = IGMP_REPORTING_MEMBER;
1949 * Switch to a different IGMP version on the given interface,
1950 * as per Section 7.2.1.
1953 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1955 int old_version_timer;
1959 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1960 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1962 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1964 * Compute the "Older Version Querier Present" timer as per
1967 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1968 old_version_timer *= PR_SLOWHZ;
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;
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);
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);
1993 * Cancel pending IGMPv3 timers for the given link and all groups
1994 * joined on it; state-change, general-query, and group-query timers.
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
2002 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
2004 struct ifmultiaddr *ifma;
2006 struct in_multi *inm, *tinm;
2008 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2009 igi->igi_ifp, igi->igi_ifp->if_xname);
2011 IN_MULTI_LOCK_ASSERT();
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.
2019 igi->igi_v3_timer = 0;
2022 * Now clear the current-state and state-change report timers
2023 * for all memberships scoped to this link.
2027 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2028 if (ifma->ifma_addr->sa_family != AF_INET ||
2029 ifma->ifma_protospec == NULL)
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:
2040 * These states are either not relevant in v3 mode,
2041 * or are unreported. Do nothing.
2044 case IGMP_LEAVING_MEMBER:
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.
2053 SLIST_INSERT_HEAD(&igi->igi_relinmhead, inm, inm_nrele);
2055 case IGMP_G_QUERY_PENDING_MEMBER:
2056 case IGMP_SG_QUERY_PENDING_MEMBER:
2057 inm_clear_recorded(inm);
2059 case IGMP_REPORTING_MEMBER:
2060 inm->inm_state = IGMP_REPORTING_MEMBER;
2064 * Always clear state-change and group report timers.
2065 * Free any pending IGMPv3 state-change records.
2067 inm->inm_sctimer = 0;
2069 _IF_DRAIN(&inm->inm_scq);
2071 IF_ADDR_RUNLOCK(ifp);
2072 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, tinm) {
2073 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
2074 inm_release_locked(inm);
2079 * Update the Older Version Querier Present timers for a link.
2080 * See Section 7.2.1 of RFC 3376.
2083 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2088 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2090 * IGMPv1 and IGMPv2 Querier Present timers expired.
2094 if (igi->igi_version != IGMP_VERSION_3) {
2096 "%s: transition from v%d -> v%d on %p(%s)",
2097 __func__, igi->igi_version, IGMP_VERSION_3,
2098 igi->igi_ifp, igi->igi_ifp->if_xname);
2099 igi->igi_version = IGMP_VERSION_3;
2101 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2103 * IGMPv1 Querier Present timer expired,
2104 * IGMPv2 Querier Present timer running.
2105 * If IGMPv2 was disabled since last timeout,
2107 * If IGMPv2 is enabled, revert to IGMPv2.
2109 if (!V_igmp_v2enable) {
2111 "%s: transition from v%d -> v%d on %p(%s)",
2112 __func__, igi->igi_version, IGMP_VERSION_3,
2113 igi->igi_ifp, igi->igi_ifp->if_xname);
2114 igi->igi_v2_timer = 0;
2115 igi->igi_version = IGMP_VERSION_3;
2117 --igi->igi_v2_timer;
2118 if (igi->igi_version != IGMP_VERSION_2) {
2120 "%s: transition from v%d -> v%d on %p(%s)",
2121 __func__, igi->igi_version, IGMP_VERSION_2,
2122 igi->igi_ifp, igi->igi_ifp->if_xname);
2123 igi->igi_version = IGMP_VERSION_2;
2126 } else if (igi->igi_v1_timer > 0) {
2128 * IGMPv1 Querier Present timer running.
2129 * Stop IGMPv2 timer if running.
2131 * If IGMPv1 was disabled since last timeout,
2133 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2135 if (!V_igmp_v1enable) {
2137 "%s: transition from v%d -> v%d on %p(%s)",
2138 __func__, igi->igi_version, IGMP_VERSION_3,
2139 igi->igi_ifp, igi->igi_ifp->if_xname);
2140 igi->igi_v1_timer = 0;
2141 igi->igi_version = IGMP_VERSION_3;
2143 --igi->igi_v1_timer;
2145 if (igi->igi_v2_timer > 0) {
2147 "%s: cancel v2 timer on %p(%s)",
2148 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2149 igi->igi_v2_timer = 0;
2155 * Global slowtimo handler.
2156 * VIMAGE: Timeout handlers are expected to service all vimages.
2161 VNET_ITERATOR_DECL(vnet_iter);
2163 VNET_LIST_RLOCK_NOSLEEP();
2164 VNET_FOREACH(vnet_iter) {
2165 CURVNET_SET(vnet_iter);
2166 igmp_slowtimo_vnet();
2169 VNET_LIST_RUNLOCK_NOSLEEP();
2173 * Per-vnet slowtimo handler.
2176 igmp_slowtimo_vnet(void)
2178 struct igmp_ifinfo *igi;
2182 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2183 igmp_v1v2_process_querier_timers(igi);
2190 * Dispatch an IGMPv1/v2 host report or leave message.
2191 * These are always small enough to fit inside a single mbuf.
2194 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2201 IN_MULTI_LOCK_ASSERT();
2206 MGETHDR(m, M_DONTWAIT, MT_DATA);
2209 MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2211 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2213 m->m_data += sizeof(struct ip);
2214 m->m_len = sizeof(struct igmp);
2216 igmp = mtod(m, struct igmp *);
2217 igmp->igmp_type = type;
2218 igmp->igmp_code = 0;
2219 igmp->igmp_group = inm->inm_addr;
2220 igmp->igmp_cksum = 0;
2221 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2223 m->m_data -= sizeof(struct ip);
2224 m->m_len += sizeof(struct ip);
2226 ip = mtod(m, struct ip *);
2228 ip->ip_len = sizeof(struct ip) + sizeof(struct igmp);
2230 ip->ip_p = IPPROTO_IGMP;
2231 ip->ip_src.s_addr = INADDR_ANY;
2233 if (type == IGMP_HOST_LEAVE_MESSAGE)
2234 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2236 ip->ip_dst = inm->inm_addr;
2238 igmp_save_context(m, ifp);
2240 m->m_flags |= M_IGMPV2;
2241 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2242 m->m_flags |= M_IGMP_LOOP;
2244 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2245 netisr_dispatch(NETISR_IGMP, m);
2251 * Process a state change from the upper layer for the given IPv4 group.
2253 * Each socket holds a reference on the in_multi in its own ip_moptions.
2254 * The socket layer will have made the necessary updates to.the group
2255 * state, it is now up to IGMP to issue a state change report if there
2256 * has been any change between T0 (when the last state-change was issued)
2259 * We use the IGMPv3 state machine at group level. The IGMP module
2260 * however makes the decision as to which IGMP protocol version to speak.
2261 * A state change *from* INCLUDE {} always means an initial join.
2262 * A state change *to* INCLUDE {} always means a final leave.
2264 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2265 * save ourselves a bunch of work; any exclusive mode groups need not
2266 * compute source filter lists.
2268 * VIMAGE: curvnet should have been set by caller, as this routine
2269 * is called from the socket option handlers.
2272 igmp_change_state(struct in_multi *inm)
2274 struct igmp_ifinfo *igi;
2278 IN_MULTI_LOCK_ASSERT();
2283 * Try to detect if the upper layer just asked us to change state
2284 * for an interface which has now gone away.
2286 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2287 ifp = inm->inm_ifma->ifma_ifp;
2289 * Sanity check that netinet's notion of ifp is the
2292 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2296 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2297 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2300 * If we detect a state transition to or from MCAST_UNDEFINED
2301 * for this group, then we are starting or finishing an IGMP
2302 * life cycle for this group.
2304 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2305 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2306 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2307 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2308 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2309 error = igmp_initial_join(inm, igi);
2311 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2312 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2313 igmp_final_leave(inm, igi);
2317 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2320 error = igmp_handle_state_change(inm, igi);
2328 * Perform the initial join for an IGMP group.
2330 * When joining a group:
2331 * If the group should have its IGMP traffic suppressed, do nothing.
2332 * IGMPv1 starts sending IGMPv1 host membership reports.
2333 * IGMPv2 starts sending IGMPv2 host membership reports.
2334 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2335 * initial state of the membership.
2338 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2341 struct ifqueue *ifq;
2342 int error, retval, syncstates;
2344 CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2345 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2346 inm->inm_ifp->if_xname);
2353 IN_MULTI_LOCK_ASSERT();
2356 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2359 * Groups joined on loopback or marked as 'not reported',
2360 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2361 * are never reported in any IGMP protocol exchanges.
2362 * All other groups enter the appropriate IGMP state machine
2363 * for the version in use on this link.
2364 * A link marked as IGIF_SILENT causes IGMP to be completely
2365 * disabled for the link.
2367 if ((ifp->if_flags & IFF_LOOPBACK) ||
2368 (igi->igi_flags & IGIF_SILENT) ||
2369 !igmp_isgroupreported(inm->inm_addr)) {
2371 "%s: not kicking state machine for silent group", __func__);
2372 inm->inm_state = IGMP_SILENT_MEMBER;
2376 * Deal with overlapping in_multi lifecycle.
2377 * If this group was LEAVING, then make sure
2378 * we drop the reference we picked up to keep the
2379 * group around for the final INCLUDE {} enqueue.
2381 if (igi->igi_version == IGMP_VERSION_3 &&
2382 inm->inm_state == IGMP_LEAVING_MEMBER)
2383 inm_release_locked(inm);
2385 inm->inm_state = IGMP_REPORTING_MEMBER;
2387 switch (igi->igi_version) {
2388 case IGMP_VERSION_1:
2389 case IGMP_VERSION_2:
2390 inm->inm_state = IGMP_IDLE_MEMBER;
2391 error = igmp_v1v2_queue_report(inm,
2392 (igi->igi_version == IGMP_VERSION_2) ?
2393 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2394 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2396 inm->inm_timer = IGMP_RANDOM_DELAY(
2397 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2398 V_current_state_timers_running = 1;
2402 case IGMP_VERSION_3:
2404 * Defer update of T0 to T1, until the first copy
2405 * of the state change has been transmitted.
2410 * Immediately enqueue a State-Change Report for
2411 * this interface, freeing any previous reports.
2412 * Don't kick the timers if there is nothing to do,
2413 * or if an error occurred.
2415 ifq = &inm->inm_scq;
2417 retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2419 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2422 error = retval * -1;
2427 * Schedule transmission of pending state-change
2428 * report up to RV times for this link. The timer
2429 * will fire at the next igmp_fasttimo (~200ms),
2430 * giving us an opportunity to merge the reports.
2432 if (igi->igi_flags & IGIF_LOOPBACK) {
2435 KASSERT(igi->igi_rv > 1,
2436 ("%s: invalid robustness %d", __func__,
2438 inm->inm_scrv = igi->igi_rv;
2440 inm->inm_sctimer = 1;
2441 V_state_change_timers_running = 1;
2449 * Only update the T0 state if state change is atomic,
2450 * i.e. we don't need to wait for a timer to fire before we
2451 * can consider the state change to have been communicated.
2455 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2456 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2463 * Issue an intermediate state change during the IGMP life-cycle.
2466 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2471 CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2472 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2473 inm->inm_ifp->if_xname);
2477 IN_MULTI_LOCK_ASSERT();
2480 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2482 if ((ifp->if_flags & IFF_LOOPBACK) ||
2483 (igi->igi_flags & IGIF_SILENT) ||
2484 !igmp_isgroupreported(inm->inm_addr) ||
2485 (igi->igi_version != IGMP_VERSION_3)) {
2486 if (!igmp_isgroupreported(inm->inm_addr)) {
2488 "%s: not kicking state machine for silent group", __func__);
2490 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2492 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2493 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2497 _IF_DRAIN(&inm->inm_scq);
2499 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2500 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2505 * If record(s) were enqueued, start the state-change
2506 * report timer for this group.
2508 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2509 inm->inm_sctimer = 1;
2510 V_state_change_timers_running = 1;
2516 * Perform the final leave for an IGMP group.
2518 * When leaving a group:
2519 * IGMPv1 does nothing.
2520 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2521 * IGMPv3 enqueues a state-change report containing a transition
2522 * to INCLUDE {} for immediate transmission.
2525 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2531 CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2532 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2533 inm->inm_ifp->if_xname);
2535 IN_MULTI_LOCK_ASSERT();
2538 switch (inm->inm_state) {
2539 case IGMP_NOT_MEMBER:
2540 case IGMP_SILENT_MEMBER:
2541 case IGMP_LEAVING_MEMBER:
2542 /* Already leaving or left; do nothing. */
2544 "%s: not kicking state machine for silent group", __func__);
2546 case IGMP_REPORTING_MEMBER:
2547 case IGMP_IDLE_MEMBER:
2548 case IGMP_G_QUERY_PENDING_MEMBER:
2549 case IGMP_SG_QUERY_PENDING_MEMBER:
2550 if (igi->igi_version == IGMP_VERSION_2) {
2552 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2553 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2554 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2557 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2558 inm->inm_state = IGMP_NOT_MEMBER;
2559 } else if (igi->igi_version == IGMP_VERSION_3) {
2561 * Stop group timer and all pending reports.
2562 * Immediately enqueue a state-change report
2563 * TO_IN {} to be sent on the next fast timeout,
2564 * giving us an opportunity to merge reports.
2566 _IF_DRAIN(&inm->inm_scq);
2568 if (igi->igi_flags & IGIF_LOOPBACK) {
2571 inm->inm_scrv = igi->igi_rv;
2573 CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2574 "pending retransmissions.", __func__,
2575 inet_ntoa(inm->inm_addr),
2576 inm->inm_ifp->if_xname, inm->inm_scrv);
2577 if (inm->inm_scrv == 0) {
2578 inm->inm_state = IGMP_NOT_MEMBER;
2579 inm->inm_sctimer = 0;
2583 inm_acquire_locked(inm);
2585 retval = igmp_v3_enqueue_group_record(
2586 &inm->inm_scq, inm, 1, 0, 0);
2587 KASSERT(retval != 0,
2588 ("%s: enqueue record = %d", __func__,
2591 inm->inm_state = IGMP_LEAVING_MEMBER;
2592 inm->inm_sctimer = 1;
2593 V_state_change_timers_running = 1;
2599 case IGMP_LAZY_MEMBER:
2600 case IGMP_SLEEPING_MEMBER:
2601 case IGMP_AWAKENING_MEMBER:
2602 /* Our reports are suppressed; do nothing. */
2608 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2609 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2610 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2611 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2612 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2617 * Enqueue an IGMPv3 group record to the given output queue.
2619 * XXX This function could do with having the allocation code
2620 * split out, and the multiple-tree-walks coalesced into a single
2621 * routine as has been done in igmp_v3_enqueue_filter_change().
2623 * If is_state_change is zero, a current-state record is appended.
2624 * If is_state_change is non-zero, a state-change report is appended.
2626 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2627 * If is_group_query is zero, and if there is a packet with free space
2628 * at the tail of the queue, it will be appended to providing there
2629 * is enough free space.
2630 * Otherwise a new mbuf packet chain is allocated.
2632 * If is_source_query is non-zero, each source is checked to see if
2633 * it was recorded for a Group-Source query, and will be omitted if
2634 * it is not both in-mode and recorded.
2636 * The function will attempt to allocate leading space in the packet
2637 * for the IP/IGMP header to be prepended without fragmenting the chain.
2639 * If successful the size of all data appended to the queue is returned,
2640 * otherwise an error code less than zero is returned, or zero if
2641 * no record(s) were appended.
2644 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2645 const int is_state_change, const int is_group_query,
2646 const int is_source_query)
2648 struct igmp_grouprec ig;
2649 struct igmp_grouprec *pig;
2651 struct ip_msource *ims, *nims;
2652 struct mbuf *m0, *m, *md;
2653 int error, is_filter_list_change;
2654 int minrec0len, m0srcs, msrcs, nbytes, off;
2655 int record_has_sources;
2661 IN_MULTI_LOCK_ASSERT();
2665 is_filter_list_change = 0;
2672 record_has_sources = 1;
2674 type = IGMP_DO_NOTHING;
2675 mode = inm->inm_st[1].iss_fmode;
2678 * If we did not transition out of ASM mode during t0->t1,
2679 * and there are no source nodes to process, we can skip
2680 * the generation of source records.
2682 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2684 record_has_sources = 0;
2686 if (is_state_change) {
2688 * Queue a state change record.
2689 * If the mode did not change, and there are non-ASM
2690 * listeners or source filters present,
2691 * we potentially need to issue two records for the group.
2692 * If we are transitioning to MCAST_UNDEFINED, we need
2693 * not send any sources.
2694 * If there are ASM listeners, and there was no filter
2695 * mode transition of any kind, do nothing.
2697 if (mode != inm->inm_st[0].iss_fmode) {
2698 if (mode == MCAST_EXCLUDE) {
2699 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2701 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2703 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2705 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2706 if (mode == MCAST_UNDEFINED)
2707 record_has_sources = 0;
2710 if (record_has_sources) {
2711 is_filter_list_change = 1;
2713 type = IGMP_DO_NOTHING;
2718 * Queue a current state record.
2720 if (mode == MCAST_EXCLUDE) {
2721 type = IGMP_MODE_IS_EXCLUDE;
2722 } else if (mode == MCAST_INCLUDE) {
2723 type = IGMP_MODE_IS_INCLUDE;
2724 KASSERT(inm->inm_st[1].iss_asm == 0,
2725 ("%s: inm %p is INCLUDE but ASM count is %d",
2726 __func__, inm, inm->inm_st[1].iss_asm));
2731 * Generate the filter list changes using a separate function.
2733 if (is_filter_list_change)
2734 return (igmp_v3_enqueue_filter_change(ifq, inm));
2736 if (type == IGMP_DO_NOTHING) {
2737 CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2738 __func__, inet_ntoa(inm->inm_addr),
2739 inm->inm_ifp->if_xname);
2744 * If any sources are present, we must be able to fit at least
2745 * one in the trailing space of the tail packet's mbuf,
2748 minrec0len = sizeof(struct igmp_grouprec);
2749 if (record_has_sources)
2750 minrec0len += sizeof(in_addr_t);
2752 CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2753 igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2754 inm->inm_ifp->if_xname);
2757 * Check if we have a packet in the tail of the queue for this
2758 * group into which the first group record for this group will fit.
2759 * Otherwise allocate a new packet.
2760 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2761 * Note: Group records for G/GSR query responses MUST be sent
2762 * in their own packet.
2765 if (!is_group_query &&
2767 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2768 (m0->m_pkthdr.len + minrec0len) <
2769 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2770 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2771 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2773 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2775 if (_IF_QFULL(ifq)) {
2776 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2780 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2781 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2782 if (!is_state_change && !is_group_query) {
2783 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2785 m->m_data += IGMP_LEADINGSPACE;
2788 m = m_gethdr(M_DONTWAIT, MT_DATA);
2790 MH_ALIGN(m, IGMP_LEADINGSPACE);
2795 igmp_save_context(m, ifp);
2797 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2801 * Append group record.
2802 * If we have sources, we don't know how many yet.
2807 ig.ig_group = inm->inm_addr;
2808 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2811 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2814 nbytes += sizeof(struct igmp_grouprec);
2817 * Append as many sources as will fit in the first packet.
2818 * If we are appending to a new packet, the chain allocation
2819 * may potentially use clusters; use m_getptr() in this case.
2820 * If we are appending to an existing packet, we need to obtain
2821 * a pointer to the group record after m_append(), in case a new
2822 * mbuf was allocated.
2823 * Only append sources which are in-mode at t1. If we are
2824 * transitioning to MCAST_UNDEFINED state on the group, do not
2825 * include source entries.
2826 * Only report recorded sources in our filter set when responding
2827 * to a group-source query.
2829 if (record_has_sources) {
2832 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2833 md->m_len - nbytes);
2835 md = m_getptr(m, 0, &off);
2836 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2840 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2841 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2842 inet_ntoa_haddr(ims->ims_haddr));
2843 now = ims_get_mode(inm, ims, 1);
2844 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2845 if ((now != mode) ||
2846 (now == mode && mode == MCAST_UNDEFINED)) {
2847 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2850 if (is_source_query && ims->ims_stp == 0) {
2851 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2855 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2856 naddr = htonl(ims->ims_haddr);
2857 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2860 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2864 nbytes += sizeof(in_addr_t);
2866 if (msrcs == m0srcs)
2869 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2871 pig->ig_numsrc = htons(msrcs);
2872 nbytes += (msrcs * sizeof(in_addr_t));
2875 if (is_source_query && msrcs == 0) {
2876 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2883 * We are good to go with first packet.
2886 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2887 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2888 _IF_ENQUEUE(ifq, m);
2890 m->m_pkthdr.PH_vt.vt_nrecs++;
2893 * No further work needed if no source list in packet(s).
2895 if (!record_has_sources)
2899 * Whilst sources remain to be announced, we need to allocate
2900 * a new packet and fill out as many sources as will fit.
2901 * Always try for a cluster first.
2903 while (nims != NULL) {
2904 if (_IF_QFULL(ifq)) {
2905 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2908 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2910 m->m_data += IGMP_LEADINGSPACE;
2912 m = m_gethdr(M_DONTWAIT, MT_DATA);
2914 MH_ALIGN(m, IGMP_LEADINGSPACE);
2918 igmp_save_context(m, ifp);
2919 md = m_getptr(m, 0, &off);
2920 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2921 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2923 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2926 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2929 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2930 nbytes += sizeof(struct igmp_grouprec);
2932 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2933 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2936 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2937 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2938 inet_ntoa_haddr(ims->ims_haddr));
2939 now = ims_get_mode(inm, ims, 1);
2940 if ((now != mode) ||
2941 (now == mode && mode == MCAST_UNDEFINED)) {
2942 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2945 if (is_source_query && ims->ims_stp == 0) {
2946 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2950 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2951 naddr = htonl(ims->ims_haddr);
2952 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2955 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2960 if (msrcs == m0srcs)
2963 pig->ig_numsrc = htons(msrcs);
2964 nbytes += (msrcs * sizeof(in_addr_t));
2966 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2967 _IF_ENQUEUE(ifq, m);
2974 * Type used to mark record pass completion.
2975 * We exploit the fact we can cast to this easily from the
2976 * current filter modes on each ip_msource node.
2979 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2980 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2981 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2982 REC_FULL = REC_ALLOW | REC_BLOCK
2986 * Enqueue an IGMPv3 filter list change to the given output queue.
2988 * Source list filter state is held in an RB-tree. When the filter list
2989 * for a group is changed without changing its mode, we need to compute
2990 * the deltas between T0 and T1 for each source in the filter set,
2991 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2993 * As we may potentially queue two record types, and the entire R-B tree
2994 * needs to be walked at once, we break this out into its own function
2995 * so we can generate a tightly packed queue of packets.
2997 * XXX This could be written to only use one tree walk, although that makes
2998 * serializing into the mbuf chains a bit harder. For now we do two walks
2999 * which makes things easier on us, and it may or may not be harder on
3002 * If successful the size of all data appended to the queue is returned,
3003 * otherwise an error code less than zero is returned, or zero if
3004 * no record(s) were appended.
3007 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
3009 static const int MINRECLEN =
3010 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3012 struct igmp_grouprec ig;
3013 struct igmp_grouprec *pig;
3014 struct ip_msource *ims, *nims;
3015 struct mbuf *m, *m0, *md;
3017 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3019 uint8_t mode, now, then;
3020 rectype_t crt, drt, nrt;
3022 IN_MULTI_LOCK_ASSERT();
3024 if (inm->inm_nsrc == 0 ||
3025 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3028 ifp = inm->inm_ifp; /* interface */
3029 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3030 crt = REC_NONE; /* current group record type */
3031 drt = REC_NONE; /* mask of completed group record types */
3032 nrt = REC_NONE; /* record type for current node */
3033 m0srcs = 0; /* # source which will fit in current mbuf chain */
3034 nbytes = 0; /* # of bytes appended to group's state-change queue */
3035 npbytes = 0; /* # of bytes appended this packet */
3036 rsrcs = 0; /* # sources encoded in current record */
3037 schanged = 0; /* # nodes encoded in overall filter change */
3038 nallow = 0; /* # of source entries in ALLOW_NEW */
3039 nblock = 0; /* # of source entries in BLOCK_OLD */
3040 nims = NULL; /* next tree node pointer */
3043 * For each possible filter record mode.
3044 * The first kind of source we encounter tells us which
3045 * is the first kind of record we start appending.
3046 * If a node transitioned to UNDEFINED at t1, its mode is treated
3047 * as the inverse of the group's filter mode.
3049 while (drt != REC_FULL) {
3053 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3054 IGMP_V3_REPORT_MAXRECS) &&
3055 (m0->m_pkthdr.len + MINRECLEN) <
3056 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3058 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3059 sizeof(struct igmp_grouprec)) /
3062 "%s: use previous packet", __func__);
3064 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3066 m->m_data += IGMP_LEADINGSPACE;
3068 m = m_gethdr(M_DONTWAIT, MT_DATA);
3070 MH_ALIGN(m, IGMP_LEADINGSPACE);
3074 "%s: m_get*() failed", __func__);
3077 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3078 igmp_save_context(m, ifp);
3079 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3080 sizeof(struct igmp_grouprec)) /
3084 "%s: allocated new packet", __func__);
3087 * Append the IGMP group record header to the
3088 * current packet's data area.
3089 * Recalculate pointer to free space for next
3090 * group record, in case m_append() allocated
3091 * a new mbuf or cluster.
3093 memset(&ig, 0, sizeof(ig));
3094 ig.ig_group = inm->inm_addr;
3095 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3099 "%s: m_append() failed", __func__);
3102 npbytes += sizeof(struct igmp_grouprec);
3104 /* new packet; offset in c hain */
3105 md = m_getptr(m, npbytes -
3106 sizeof(struct igmp_grouprec), &off);
3107 pig = (struct igmp_grouprec *)(mtod(md,
3110 /* current packet; offset from last append */
3112 pig = (struct igmp_grouprec *)(mtod(md,
3113 uint8_t *) + md->m_len -
3114 sizeof(struct igmp_grouprec));
3117 * Begin walking the tree for this record type
3118 * pass, or continue from where we left off
3119 * previously if we had to allocate a new packet.
3120 * Only report deltas in-mode at t1.
3121 * We need not report included sources as allowed
3122 * if we are in inclusive mode on the group,
3123 * however the converse is not true.
3127 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3128 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3129 CTR2(KTR_IGMPV3, "%s: visit node %s",
3130 __func__, inet_ntoa_haddr(ims->ims_haddr));
3131 now = ims_get_mode(inm, ims, 1);
3132 then = ims_get_mode(inm, ims, 0);
3133 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3134 __func__, then, now);
3137 "%s: skip unchanged", __func__);
3140 if (mode == MCAST_EXCLUDE &&
3141 now == MCAST_INCLUDE) {
3143 "%s: skip IN src on EX group",
3147 nrt = (rectype_t)now;
3148 if (nrt == REC_NONE)
3149 nrt = (rectype_t)(~mode & REC_FULL);
3150 if (schanged++ == 0) {
3152 } else if (crt != nrt)
3154 naddr = htonl(ims->ims_haddr);
3155 if (!m_append(m, sizeof(in_addr_t),
3160 "%s: m_append() failed", __func__);
3163 nallow += !!(crt == REC_ALLOW);
3164 nblock += !!(crt == REC_BLOCK);
3165 if (++rsrcs == m0srcs)
3169 * If we did not append any tree nodes on this
3170 * pass, back out of allocations.
3173 npbytes -= sizeof(struct igmp_grouprec);
3176 "%s: m_free(m)", __func__);
3180 "%s: m_adj(m, -ig)", __func__);
3181 m_adj(m, -((int)sizeof(
3182 struct igmp_grouprec)));
3186 npbytes += (rsrcs * sizeof(in_addr_t));
3187 if (crt == REC_ALLOW)
3188 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3189 else if (crt == REC_BLOCK)
3190 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3191 pig->ig_numsrc = htons(rsrcs);
3193 * Count the new group record, and enqueue this
3194 * packet if it wasn't already queued.
3196 m->m_pkthdr.PH_vt.vt_nrecs++;
3198 _IF_ENQUEUE(ifq, m);
3200 } while (nims != NULL);
3202 crt = (~crt & REC_FULL);
3205 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3212 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3215 struct mbuf *m; /* pending state-change */
3216 struct mbuf *m0; /* copy of pending state-change */
3217 struct mbuf *mt; /* last state-change in packet */
3218 int docopy, domerge;
3225 IN_MULTI_LOCK_ASSERT();
3229 * If there are further pending retransmissions, make a writable
3230 * copy of each queued state-change message before merging.
3232 if (inm->inm_scrv > 0)
3237 if (gq->ifq_head == NULL) {
3238 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3246 * Only merge the report into the current packet if
3247 * there is sufficient space to do so; an IGMPv3 report
3248 * packet may only contain 65,535 group records.
3249 * Always use a simple mbuf chain concatentation to do this,
3250 * as large state changes for single groups may have
3251 * allocated clusters.
3254 mt = ifscq->ifq_tail;
3256 recslen = m_length(m, NULL);
3258 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3259 m->m_pkthdr.PH_vt.vt_nrecs <=
3260 IGMP_V3_REPORT_MAXRECS) &&
3261 (mt->m_pkthdr.len + recslen <=
3262 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3266 if (!domerge && _IF_QFULL(gq)) {
3268 "%s: outbound queue full, skipping whole packet %p",
3278 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3279 _IF_DEQUEUE(gq, m0);
3282 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3283 m0 = m_dup(m, M_NOWAIT);
3286 m0->m_nextpkt = NULL;
3291 CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3292 __func__, m0, ifscq);
3293 _IF_ENQUEUE(ifscq, m0);
3295 struct mbuf *mtl; /* last mbuf of packet mt */
3297 CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3301 m0->m_flags &= ~M_PKTHDR;
3302 mt->m_pkthdr.len += recslen;
3303 mt->m_pkthdr.PH_vt.vt_nrecs +=
3304 m0->m_pkthdr.PH_vt.vt_nrecs;
3314 * Respond to a pending IGMPv3 General Query.
3317 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3319 struct ifmultiaddr *ifma;
3321 struct in_multi *inm;
3324 IN_MULTI_LOCK_ASSERT();
3327 KASSERT(igi->igi_version == IGMP_VERSION_3,
3328 ("%s: called when version %d", __func__, igi->igi_version));
3333 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3334 if (ifma->ifma_addr->sa_family != AF_INET ||
3335 ifma->ifma_protospec == NULL)
3338 inm = (struct in_multi *)ifma->ifma_protospec;
3339 KASSERT(ifp == inm->inm_ifp,
3340 ("%s: inconsistent ifp", __func__));
3342 switch (inm->inm_state) {
3343 case IGMP_NOT_MEMBER:
3344 case IGMP_SILENT_MEMBER:
3346 case IGMP_REPORTING_MEMBER:
3347 case IGMP_IDLE_MEMBER:
3348 case IGMP_LAZY_MEMBER:
3349 case IGMP_SLEEPING_MEMBER:
3350 case IGMP_AWAKENING_MEMBER:
3351 inm->inm_state = IGMP_REPORTING_MEMBER;
3352 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3354 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3357 case IGMP_G_QUERY_PENDING_MEMBER:
3358 case IGMP_SG_QUERY_PENDING_MEMBER:
3359 case IGMP_LEAVING_MEMBER:
3363 IF_ADDR_RUNLOCK(ifp);
3365 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3366 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3369 * Slew transmission of bursts over 500ms intervals.
3371 if (igi->igi_gq.ifq_head != NULL) {
3372 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3373 IGMP_RESPONSE_BURST_INTERVAL);
3374 V_interface_timers_running = 1;
3379 * Transmit the next pending IGMP message in the output queue.
3381 * We get called from netisr_processqueue(). A mutex private to igmpoq
3382 * will be acquired and released around this routine.
3384 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3385 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3386 * a link and uses a link-scope multicast address.
3389 igmp_intr(struct mbuf *m)
3391 struct ip_moptions imo;
3393 struct mbuf *ipopts, *m0;
3397 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3400 * Set VNET image pointer from enqueued mbuf chain
3401 * before doing anything else. Whilst we use interface
3402 * indexes to guard against interface detach, they are
3403 * unique to each VIMAGE and must be retrieved.
3405 CURVNET_SET((struct vnet *)(m->m_pkthdr.header));
3406 ifindex = igmp_restore_context(m);
3409 * Check if the ifnet still exists. This limits the scope of
3410 * any race in the absence of a global ifp lock for low cost
3411 * (an array lookup).
3413 ifp = ifnet_byindex(ifindex);
3415 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3416 __func__, m, ifindex);
3418 IPSTAT_INC(ips_noroute);
3422 ipopts = V_igmp_sendra ? m_raopt : NULL;
3424 imo.imo_multicast_ttl = 1;
3425 imo.imo_multicast_vif = -1;
3426 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3429 * If the user requested that IGMP traffic be explicitly
3430 * redirected to the loopback interface (e.g. they are running a
3431 * MANET interface and the routing protocol needs to see the
3432 * updates), handle this now.
3434 if (m->m_flags & M_IGMP_LOOP)
3435 imo.imo_multicast_ifp = V_loif;
3437 imo.imo_multicast_ifp = ifp;
3439 if (m->m_flags & M_IGMPV2) {
3442 m0 = igmp_v3_encap_report(ifp, m);
3444 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3446 IPSTAT_INC(ips_odropped);
3451 igmp_scrub_context(m0);
3452 m->m_flags &= ~(M_PROTOFLAGS);
3453 m0->m_pkthdr.rcvif = V_loif;
3455 mac_netinet_igmp_send(ifp, m0);
3457 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3459 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3463 IGMPSTAT_INC(igps_snd_reports);
3467 * We must restore the existing vnet pointer before
3468 * continuing as we are run from netisr context.
3474 * Encapsulate an IGMPv3 report.
3476 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3477 * chain has already had its IP/IGMPv3 header prepended. In this case
3478 * the function will not attempt to prepend; the lengths and checksums
3479 * will however be re-computed.
3481 * Returns a pointer to the new mbuf chain head, or NULL if the
3482 * allocation failed.
3484 static struct mbuf *
3485 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3487 struct igmp_report *igmp;
3489 int hdrlen, igmpreclen;
3491 KASSERT((m->m_flags & M_PKTHDR),
3492 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3494 igmpreclen = m_length(m, NULL);
3495 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3497 if (m->m_flags & M_IGMPV3_HDR) {
3498 igmpreclen -= hdrlen;
3500 M_PREPEND(m, hdrlen, M_DONTWAIT);
3503 m->m_flags |= M_IGMPV3_HDR;
3506 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3508 m->m_data += sizeof(struct ip);
3509 m->m_len -= sizeof(struct ip);
3511 igmp = mtod(m, struct igmp_report *);
3512 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3515 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3517 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3518 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3520 m->m_data -= sizeof(struct ip);
3521 m->m_len += sizeof(struct ip);
3523 ip = mtod(m, struct ip *);
3524 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3525 ip->ip_len = hdrlen + igmpreclen;
3527 ip->ip_p = IPPROTO_IGMP;
3530 ip->ip_src.s_addr = INADDR_ANY;
3532 if (m->m_flags & M_IGMP_LOOP) {
3533 struct in_ifaddr *ia;
3537 ip->ip_src = ia->ia_addr.sin_addr;
3538 ifa_free(&ia->ia_ifa);
3542 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3549 igmp_rec_type_to_str(const int type)
3553 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3556 case IGMP_CHANGE_TO_INCLUDE_MODE:
3559 case IGMP_MODE_IS_EXCLUDE:
3562 case IGMP_MODE_IS_INCLUDE:
3565 case IGMP_ALLOW_NEW_SOURCES:
3568 case IGMP_BLOCK_OLD_SOURCES:
3579 igmp_init(void *unused __unused)
3582 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3586 m_raopt = igmp_ra_alloc();
3588 netisr_register(&igmp_nh);
3590 SYSINIT(igmp_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_init, NULL);
3593 igmp_uninit(void *unused __unused)
3596 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3598 netisr_unregister(&igmp_nh);
3603 IGMP_LOCK_DESTROY();
3605 SYSUNINIT(igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_uninit, NULL);
3608 vnet_igmp_init(const void *unused __unused)
3611 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3613 LIST_INIT(&V_igi_head);
3615 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_igmp_init,
3619 vnet_igmp_uninit(const void *unused __unused)
3622 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3624 KASSERT(LIST_EMPTY(&V_igi_head),
3625 ("%s: igi list not empty; ifnets not detached?", __func__));
3627 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY,
3628 vnet_igmp_uninit, NULL);
3631 igmp_modevent(module_t mod, int type, void *unused __unused)
3639 return (EOPNOTSUPP);
3644 static moduledata_t igmp_mod = {
3649 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);