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.PH_loc.ptr = ifp->if_vnet;
294 m->m_pkthdr.flowid = ifp->if_index;
298 igmp_scrub_context(struct mbuf *m)
301 m->m_pkthdr.PH_loc.ptr = 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.PH_loc.ptr),
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 m = m_get(M_WAITOK, 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)
658 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
659 __func__, ifp, ifp->if_xname);
662 igi_delete_locked(ifp);
667 igi_delete_locked(const struct ifnet *ifp)
669 struct igmp_ifinfo *igi, *tigi;
671 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
672 __func__, ifp, ifp->if_xname);
676 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
677 if (igi->igi_ifp == ifp) {
679 * Free deferred General Query responses.
681 _IF_DRAIN(&igi->igi_gq);
683 LIST_REMOVE(igi, igi_link);
685 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
686 ("%s: there are dangling in_multi references",
695 panic("%s: igmp_ifinfo not found for ifp %p\n", __func__, ifp);
700 * Process a received IGMPv1 query.
701 * Return non-zero if the message should be dropped.
703 * VIMAGE: The curvnet pointer is derived from the input ifp.
706 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
707 const struct igmp *igmp)
709 struct ifmultiaddr *ifma;
710 struct igmp_ifinfo *igi;
711 struct in_multi *inm;
714 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
715 * 224.0.0.1. They are always treated as General Queries.
716 * igmp_group is always ignored. Do not drop it as a userland
717 * daemon may wish to see it.
718 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
720 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
721 IGMPSTAT_INC(igps_rcv_badqueries);
724 IGMPSTAT_INC(igps_rcv_gen_queries);
729 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
730 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
732 if (igi->igi_flags & IGIF_LOOPBACK) {
733 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
739 * Switch to IGMPv1 host compatibility mode.
741 igmp_set_version(igi, IGMP_VERSION_1);
743 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
746 * Start the timers in all of our group records
747 * for the interface on which the query arrived,
748 * except those which are already running.
751 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
752 if (ifma->ifma_addr->sa_family != AF_INET ||
753 ifma->ifma_protospec == NULL)
755 inm = (struct in_multi *)ifma->ifma_protospec;
756 if (inm->inm_timer != 0)
758 switch (inm->inm_state) {
759 case IGMP_NOT_MEMBER:
760 case IGMP_SILENT_MEMBER:
762 case IGMP_G_QUERY_PENDING_MEMBER:
763 case IGMP_SG_QUERY_PENDING_MEMBER:
764 case IGMP_REPORTING_MEMBER:
765 case IGMP_IDLE_MEMBER:
766 case IGMP_LAZY_MEMBER:
767 case IGMP_SLEEPING_MEMBER:
768 case IGMP_AWAKENING_MEMBER:
769 inm->inm_state = IGMP_REPORTING_MEMBER;
770 inm->inm_timer = IGMP_RANDOM_DELAY(
771 IGMP_V1V2_MAX_RI * PR_FASTHZ);
772 V_current_state_timers_running = 1;
774 case IGMP_LEAVING_MEMBER:
778 IF_ADDR_RUNLOCK(ifp);
788 * Process a received IGMPv2 general or group-specific query.
791 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
792 const struct igmp *igmp)
794 struct ifmultiaddr *ifma;
795 struct igmp_ifinfo *igi;
796 struct in_multi *inm;
797 int is_general_query;
800 is_general_query = 0;
803 * Validate address fields upfront.
804 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
806 if (in_nullhost(igmp->igmp_group)) {
808 * IGMPv2 General Query.
809 * If this was not sent to the all-hosts group, ignore it.
811 if (!in_allhosts(ip->ip_dst))
813 IGMPSTAT_INC(igps_rcv_gen_queries);
814 is_general_query = 1;
816 /* IGMPv2 Group-Specific Query. */
817 IGMPSTAT_INC(igps_rcv_group_queries);
823 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
824 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
826 if (igi->igi_flags & IGIF_LOOPBACK) {
827 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
833 * Ignore v2 query if in v1 Compatibility Mode.
835 if (igi->igi_version == IGMP_VERSION_1)
838 igmp_set_version(igi, IGMP_VERSION_2);
840 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
844 if (is_general_query) {
846 * For each reporting group joined on this
847 * interface, kick the report timer.
849 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
852 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
853 if (ifma->ifma_addr->sa_family != AF_INET ||
854 ifma->ifma_protospec == NULL)
856 inm = (struct in_multi *)ifma->ifma_protospec;
857 igmp_v2_update_group(inm, timer);
859 IF_ADDR_RUNLOCK(ifp);
862 * Group-specific IGMPv2 query, we need only
863 * look up the single group to process it.
865 inm = inm_lookup(ifp, igmp->igmp_group);
867 CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
868 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
869 igmp_v2_update_group(inm, timer);
881 * Update the report timer on a group in response to an IGMPv2 query.
883 * If we are becoming the reporting member for this group, start the timer.
884 * If we already are the reporting member for this group, and timer is
885 * below the threshold, reset it.
887 * We may be updating the group for the first time since we switched
888 * to IGMPv3. If we are, then we must clear any recorded source lists,
889 * and transition to REPORTING state; the group timer is overloaded
890 * for group and group-source query responses.
892 * Unlike IGMPv3, the delay per group should be jittered
893 * to avoid bursts of IGMPv2 reports.
896 igmp_v2_update_group(struct in_multi *inm, const int timer)
899 CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
900 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
902 IN_MULTI_LOCK_ASSERT();
904 switch (inm->inm_state) {
905 case IGMP_NOT_MEMBER:
906 case IGMP_SILENT_MEMBER:
908 case IGMP_REPORTING_MEMBER:
909 if (inm->inm_timer != 0 &&
910 inm->inm_timer <= timer) {
911 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
912 "skipping.", __func__);
916 case IGMP_SG_QUERY_PENDING_MEMBER:
917 case IGMP_G_QUERY_PENDING_MEMBER:
918 case IGMP_IDLE_MEMBER:
919 case IGMP_LAZY_MEMBER:
920 case IGMP_AWAKENING_MEMBER:
921 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
922 inm->inm_state = IGMP_REPORTING_MEMBER;
923 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
924 V_current_state_timers_running = 1;
926 case IGMP_SLEEPING_MEMBER:
927 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
928 inm->inm_state = IGMP_AWAKENING_MEMBER;
930 case IGMP_LEAVING_MEMBER:
936 * Process a received IGMPv3 general, group-specific or
937 * group-and-source-specific query.
938 * Assumes m has already been pulled up to the full IGMP message length.
939 * Return 0 if successful, otherwise an appropriate error code is returned.
942 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
943 /*const*/ struct igmpv3 *igmpv3)
945 struct igmp_ifinfo *igi;
946 struct in_multi *inm;
947 int is_general_query;
948 uint32_t maxresp, nsrc, qqi;
952 is_general_query = 0;
954 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
956 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
957 if (maxresp >= 128) {
958 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
959 (IGMP_EXP(igmpv3->igmp_code) + 3);
963 * Robustness must never be less than 2 for on-wire IGMPv3.
964 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
965 * an exception for interfaces whose IGMPv3 state changes
966 * are redirected to loopback (e.g. MANET).
968 qrv = IGMP_QRV(igmpv3->igmp_misc);
970 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
975 qqi = igmpv3->igmp_qqi;
977 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
978 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
981 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
985 nsrc = ntohs(igmpv3->igmp_numsrc);
988 * Validate address fields and versions upfront before
989 * accepting v3 query.
990 * XXX SMPng: Unlocked access to igmpstat counters here.
992 if (in_nullhost(igmpv3->igmp_group)) {
994 * IGMPv3 General Query.
996 * General Queries SHOULD be directed to 224.0.0.1.
997 * A general query with a source list has undefined
998 * behaviour; discard it.
1000 IGMPSTAT_INC(igps_rcv_gen_queries);
1001 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1002 IGMPSTAT_INC(igps_rcv_badqueries);
1005 is_general_query = 1;
1007 /* Group or group-source specific query. */
1009 IGMPSTAT_INC(igps_rcv_group_queries);
1011 IGMPSTAT_INC(igps_rcv_gsr_queries);
1017 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1018 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
1020 if (igi->igi_flags & IGIF_LOOPBACK) {
1021 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1022 ifp, ifp->if_xname);
1027 * Discard the v3 query if we're in Compatibility Mode.
1028 * The RFC is not obviously worded that hosts need to stay in
1029 * compatibility mode until the Old Version Querier Present
1032 if (igi->igi_version != IGMP_VERSION_3) {
1033 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1034 igi->igi_version, ifp, ifp->if_xname);
1038 igmp_set_version(igi, IGMP_VERSION_3);
1041 igi->igi_qri = maxresp;
1043 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1046 if (is_general_query) {
1048 * Schedule a current-state report on this ifp for
1049 * all groups, possibly containing source lists.
1050 * If there is a pending General Query response
1051 * scheduled earlier than the selected delay, do
1052 * not schedule any other reports.
1053 * Otherwise, reset the interface timer.
1055 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1056 ifp, ifp->if_xname);
1057 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1058 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1059 V_interface_timers_running = 1;
1063 * Group-source-specific queries are throttled on
1064 * a per-group basis to defeat denial-of-service attempts.
1065 * Queries for groups we are not a member of on this
1066 * link are simply ignored.
1068 inm = inm_lookup(ifp, igmpv3->igmp_group);
1072 if (!ratecheck(&inm->inm_lastgsrtv,
1073 &V_igmp_gsrdelay)) {
1074 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1076 IGMPSTAT_INC(igps_drop_gsr_queries);
1080 CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1081 inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1083 * If there is a pending General Query response
1084 * scheduled sooner than the selected delay, no
1085 * further report need be scheduled.
1086 * Otherwise, prepare to respond to the
1087 * group-specific or group-and-source query.
1089 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1090 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1101 * Process a recieved IGMPv3 group-specific or group-and-source-specific
1103 * Return <0 if any error occured. Currently this is ignored.
1106 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1107 int timer, /*const*/ struct igmpv3 *igmpv3)
1112 IN_MULTI_LOCK_ASSERT();
1117 switch (inm->inm_state) {
1118 case IGMP_NOT_MEMBER:
1119 case IGMP_SILENT_MEMBER:
1120 case IGMP_SLEEPING_MEMBER:
1121 case IGMP_LAZY_MEMBER:
1122 case IGMP_AWAKENING_MEMBER:
1123 case IGMP_IDLE_MEMBER:
1124 case IGMP_LEAVING_MEMBER:
1127 case IGMP_REPORTING_MEMBER:
1128 case IGMP_G_QUERY_PENDING_MEMBER:
1129 case IGMP_SG_QUERY_PENDING_MEMBER:
1133 nsrc = ntohs(igmpv3->igmp_numsrc);
1136 * Deal with group-specific queries upfront.
1137 * If any group query is already pending, purge any recorded
1138 * source-list state if it exists, and schedule a query response
1139 * for this group-specific query.
1142 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1143 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1144 inm_clear_recorded(inm);
1145 timer = min(inm->inm_timer, timer);
1147 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1148 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1149 V_current_state_timers_running = 1;
1154 * Deal with the case where a group-and-source-specific query has
1155 * been received but a group-specific query is already pending.
1157 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1158 timer = min(inm->inm_timer, timer);
1159 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1160 V_current_state_timers_running = 1;
1165 * Finally, deal with the case where a group-and-source-specific
1166 * query has been received, where a response to a previous g-s-r
1167 * query exists, or none exists.
1168 * In this case, we need to parse the source-list which the Querier
1169 * has provided us with and check if we have any source list filter
1170 * entries at T1 for these sources. If we do not, there is no need
1171 * schedule a report and the query may be dropped.
1172 * If we do, we must record them and schedule a current-state
1173 * report for those sources.
1174 * FIXME: Handling source lists larger than 1 mbuf requires that
1175 * we pass the mbuf chain pointer down to this function, and use
1176 * m_getptr() to walk the chain.
1178 if (inm->inm_nsrc > 0) {
1179 const struct in_addr *ap;
1182 ap = (const struct in_addr *)(igmpv3 + 1);
1184 for (i = 0; i < nsrc; i++, ap++) {
1185 retval = inm_record_source(inm, ap->s_addr);
1188 nrecorded += retval;
1190 if (nrecorded > 0) {
1192 "%s: schedule response to SG query", __func__);
1193 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1194 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1195 V_current_state_timers_running = 1;
1203 * Process a received IGMPv1 host membership report.
1205 * NOTE: 0.0.0.0 workaround breaks const correctness.
1208 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1209 /*const*/ struct igmp *igmp)
1211 struct in_ifaddr *ia;
1212 struct in_multi *inm;
1214 IGMPSTAT_INC(igps_rcv_reports);
1216 if (ifp->if_flags & IFF_LOOPBACK)
1219 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1220 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1221 IGMPSTAT_INC(igps_rcv_badreports);
1226 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1227 * Booting clients may use the source address 0.0.0.0. Some
1228 * IGMP daemons may not know how to use IP_RECVIF to determine
1229 * the interface upon which this message was received.
1230 * Replace 0.0.0.0 with the subnet address if told to do so.
1232 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1235 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1236 ifa_free(&ia->ia_ifa);
1240 CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1241 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1244 * IGMPv1 report suppression.
1245 * If we are a member of this group, and our membership should be
1246 * reported, stop our group timer and transition to the 'lazy' state.
1249 inm = inm_lookup(ifp, igmp->igmp_group);
1251 struct igmp_ifinfo *igi;
1255 KASSERT(igi != NULL,
1256 ("%s: no igi for ifp %p", __func__, ifp));
1260 IGMPSTAT_INC(igps_rcv_ourreports);
1263 * If we are in IGMPv3 host mode, do not allow the
1264 * other host's IGMPv1 report to suppress our reports
1265 * unless explicitly configured to do so.
1267 if (igi->igi_version == IGMP_VERSION_3) {
1268 if (V_igmp_legacysupp)
1269 igmp_v3_suppress_group_record(inm);
1275 switch (inm->inm_state) {
1276 case IGMP_NOT_MEMBER:
1277 case IGMP_SILENT_MEMBER:
1279 case IGMP_IDLE_MEMBER:
1280 case IGMP_LAZY_MEMBER:
1281 case IGMP_AWAKENING_MEMBER:
1283 "report suppressed for %s on ifp %p(%s)",
1284 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1285 case IGMP_SLEEPING_MEMBER:
1286 inm->inm_state = IGMP_SLEEPING_MEMBER;
1288 case IGMP_REPORTING_MEMBER:
1290 "report suppressed for %s on ifp %p(%s)",
1291 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1292 if (igi->igi_version == IGMP_VERSION_1)
1293 inm->inm_state = IGMP_LAZY_MEMBER;
1294 else if (igi->igi_version == IGMP_VERSION_2)
1295 inm->inm_state = IGMP_SLEEPING_MEMBER;
1297 case IGMP_G_QUERY_PENDING_MEMBER:
1298 case IGMP_SG_QUERY_PENDING_MEMBER:
1299 case IGMP_LEAVING_MEMBER:
1311 * Process a received IGMPv2 host membership report.
1313 * NOTE: 0.0.0.0 workaround breaks const correctness.
1316 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1317 /*const*/ struct igmp *igmp)
1319 struct in_ifaddr *ia;
1320 struct in_multi *inm;
1323 * Make sure we don't hear our own membership report. Fast
1324 * leave requires knowing that we are the only member of a
1328 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1329 ifa_free(&ia->ia_ifa);
1333 IGMPSTAT_INC(igps_rcv_reports);
1335 if (ifp->if_flags & IFF_LOOPBACK) {
1337 ifa_free(&ia->ia_ifa);
1341 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1342 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1344 ifa_free(&ia->ia_ifa);
1345 IGMPSTAT_INC(igps_rcv_badreports);
1350 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1351 * Booting clients may use the source address 0.0.0.0. Some
1352 * IGMP daemons may not know how to use IP_RECVIF to determine
1353 * the interface upon which this message was received.
1354 * Replace 0.0.0.0 with the subnet address if told to do so.
1356 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1358 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1361 ifa_free(&ia->ia_ifa);
1363 CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1364 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1367 * IGMPv2 report suppression.
1368 * If we are a member of this group, and our membership should be
1369 * reported, and our group timer is pending or about to be reset,
1370 * stop our group timer by transitioning to the 'lazy' state.
1373 inm = inm_lookup(ifp, igmp->igmp_group);
1375 struct igmp_ifinfo *igi;
1378 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1380 IGMPSTAT_INC(igps_rcv_ourreports);
1383 * If we are in IGMPv3 host mode, do not allow the
1384 * other host's IGMPv1 report to suppress our reports
1385 * unless explicitly configured to do so.
1387 if (igi->igi_version == IGMP_VERSION_3) {
1388 if (V_igmp_legacysupp)
1389 igmp_v3_suppress_group_record(inm);
1395 switch (inm->inm_state) {
1396 case IGMP_NOT_MEMBER:
1397 case IGMP_SILENT_MEMBER:
1398 case IGMP_SLEEPING_MEMBER:
1400 case IGMP_REPORTING_MEMBER:
1401 case IGMP_IDLE_MEMBER:
1402 case IGMP_AWAKENING_MEMBER:
1404 "report suppressed for %s on ifp %p(%s)",
1405 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1406 case IGMP_LAZY_MEMBER:
1407 inm->inm_state = IGMP_LAZY_MEMBER;
1409 case IGMP_G_QUERY_PENDING_MEMBER:
1410 case IGMP_SG_QUERY_PENDING_MEMBER:
1411 case IGMP_LEAVING_MEMBER:
1423 igmp_input(struct mbuf *m, int off)
1433 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1435 ifp = m->m_pkthdr.rcvif;
1437 IGMPSTAT_INC(igps_rcv_total);
1439 ip = mtod(m, struct ip *);
1441 igmplen = ntohs(ip->ip_len) - off;
1446 if (igmplen < IGMP_MINLEN) {
1447 IGMPSTAT_INC(igps_rcv_tooshort);
1453 * Always pullup to the minimum size for v1/v2 or v3
1454 * to amortize calls to m_pullup().
1457 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1458 minlen += IGMP_V3_QUERY_MINLEN;
1460 minlen += IGMP_MINLEN;
1461 if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1462 (m = m_pullup(m, minlen)) == 0) {
1463 IGMPSTAT_INC(igps_rcv_tooshort);
1466 ip = mtod(m, struct ip *);
1469 * Validate checksum.
1471 m->m_data += iphlen;
1473 igmp = mtod(m, struct igmp *);
1474 if (in_cksum(m, igmplen)) {
1475 IGMPSTAT_INC(igps_rcv_badsum);
1479 m->m_data -= iphlen;
1483 * IGMP control traffic is link-scope, and must have a TTL of 1.
1484 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1485 * probe packets may come from beyond the LAN.
1487 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1488 IGMPSTAT_INC(igps_rcv_badttl);
1493 switch (igmp->igmp_type) {
1494 case IGMP_HOST_MEMBERSHIP_QUERY:
1495 if (igmplen == IGMP_MINLEN) {
1496 if (igmp->igmp_code == 0)
1497 queryver = IGMP_VERSION_1;
1499 queryver = IGMP_VERSION_2;
1500 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1501 queryver = IGMP_VERSION_3;
1503 IGMPSTAT_INC(igps_rcv_tooshort);
1509 case IGMP_VERSION_1:
1510 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1511 if (!V_igmp_v1enable)
1513 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1519 case IGMP_VERSION_2:
1520 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1521 if (!V_igmp_v2enable)
1523 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1529 case IGMP_VERSION_3: {
1530 struct igmpv3 *igmpv3;
1534 IGMPSTAT_INC(igps_rcv_v3_queries);
1535 igmpv3 = (struct igmpv3 *)igmp;
1537 * Validate length based on source count.
1539 nsrc = ntohs(igmpv3->igmp_numsrc);
1540 if (nsrc * sizeof(in_addr_t) >
1541 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1542 IGMPSTAT_INC(igps_rcv_tooshort);
1546 * m_pullup() may modify m, so pullup in
1549 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1550 sizeof(struct in_addr) * nsrc;
1551 if ((m->m_flags & M_EXT ||
1552 m->m_len < igmpv3len) &&
1553 (m = m_pullup(m, igmpv3len)) == NULL) {
1554 IGMPSTAT_INC(igps_rcv_tooshort);
1557 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1559 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1568 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1569 if (!V_igmp_v1enable)
1571 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1577 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1578 if (!V_igmp_v2enable)
1580 if (!ip_checkrouteralert(m))
1581 IGMPSTAT_INC(igps_rcv_nora);
1582 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1588 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1590 * Hosts do not need to process IGMPv3 membership reports,
1591 * as report suppression is no longer required.
1593 if (!ip_checkrouteralert(m))
1594 IGMPSTAT_INC(igps_rcv_nora);
1602 * Pass all valid IGMP packets up to any process(es) listening on a
1610 * Fast timeout handler (global).
1611 * VIMAGE: Timeout handlers are expected to service all vimages.
1616 VNET_ITERATOR_DECL(vnet_iter);
1618 VNET_LIST_RLOCK_NOSLEEP();
1619 VNET_FOREACH(vnet_iter) {
1620 CURVNET_SET(vnet_iter);
1621 igmp_fasttimo_vnet();
1624 VNET_LIST_RUNLOCK_NOSLEEP();
1628 * Fast timeout handler (per-vnet).
1629 * Sends are shuffled off to a netisr to deal with Giant.
1631 * VIMAGE: Assume caller has set up our curvnet.
1634 igmp_fasttimo_vnet(void)
1636 struct ifqueue scq; /* State-change packets */
1637 struct ifqueue qrq; /* Query response packets */
1639 struct igmp_ifinfo *igi;
1640 struct ifmultiaddr *ifma;
1641 struct in_multi *inm;
1642 int loop, uri_fasthz;
1648 * Quick check to see if any work needs to be done, in order to
1649 * minimize the overhead of fasttimo processing.
1650 * SMPng: XXX Unlocked reads.
1652 if (!V_current_state_timers_running &&
1653 !V_interface_timers_running &&
1654 !V_state_change_timers_running)
1661 * IGMPv3 General Query response timer processing.
1663 if (V_interface_timers_running) {
1664 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1666 V_interface_timers_running = 0;
1667 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1668 if (igi->igi_v3_timer == 0) {
1670 } else if (--igi->igi_v3_timer == 0) {
1671 igmp_v3_dispatch_general_query(igi);
1673 V_interface_timers_running = 1;
1678 if (!V_current_state_timers_running &&
1679 !V_state_change_timers_running)
1682 V_current_state_timers_running = 0;
1683 V_state_change_timers_running = 0;
1685 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1688 * IGMPv1/v2/v3 host report and state-change timer processing.
1689 * Note: Processing a v3 group timer may remove a node.
1691 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1694 if (igi->igi_version == IGMP_VERSION_3) {
1695 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1696 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1699 memset(&qrq, 0, sizeof(struct ifqueue));
1700 IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1702 memset(&scq, 0, sizeof(struct ifqueue));
1703 IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1707 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1708 if (ifma->ifma_addr->sa_family != AF_INET ||
1709 ifma->ifma_protospec == NULL)
1711 inm = (struct in_multi *)ifma->ifma_protospec;
1712 switch (igi->igi_version) {
1713 case IGMP_VERSION_1:
1714 case IGMP_VERSION_2:
1715 igmp_v1v2_process_group_timer(inm,
1718 case IGMP_VERSION_3:
1719 igmp_v3_process_group_timers(igi, &qrq,
1720 &scq, inm, uri_fasthz);
1724 IF_ADDR_RUNLOCK(ifp);
1726 if (igi->igi_version == IGMP_VERSION_3) {
1727 struct in_multi *tinm;
1729 igmp_dispatch_queue(&qrq, 0, loop);
1730 igmp_dispatch_queue(&scq, 0, loop);
1733 * Free the in_multi reference(s) for this
1736 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1738 SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1740 inm_release_locked(inm);
1751 * Update host report group timer for IGMPv1/v2.
1752 * Will update the global pending timer flags.
1755 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1757 int report_timer_expired;
1759 IN_MULTI_LOCK_ASSERT();
1762 if (inm->inm_timer == 0) {
1763 report_timer_expired = 0;
1764 } else if (--inm->inm_timer == 0) {
1765 report_timer_expired = 1;
1767 V_current_state_timers_running = 1;
1771 switch (inm->inm_state) {
1772 case IGMP_NOT_MEMBER:
1773 case IGMP_SILENT_MEMBER:
1774 case IGMP_IDLE_MEMBER:
1775 case IGMP_LAZY_MEMBER:
1776 case IGMP_SLEEPING_MEMBER:
1777 case IGMP_AWAKENING_MEMBER:
1779 case IGMP_REPORTING_MEMBER:
1780 if (report_timer_expired) {
1781 inm->inm_state = IGMP_IDLE_MEMBER;
1782 (void)igmp_v1v2_queue_report(inm,
1783 (version == IGMP_VERSION_2) ?
1784 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1785 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1788 case IGMP_G_QUERY_PENDING_MEMBER:
1789 case IGMP_SG_QUERY_PENDING_MEMBER:
1790 case IGMP_LEAVING_MEMBER:
1796 * Update a group's timers for IGMPv3.
1797 * Will update the global pending timer flags.
1798 * Note: Unlocked read from igi.
1801 igmp_v3_process_group_timers(struct igmp_ifinfo *igi,
1802 struct ifqueue *qrq, struct ifqueue *scq,
1803 struct in_multi *inm, const int uri_fasthz)
1805 int query_response_timer_expired;
1806 int state_change_retransmit_timer_expired;
1808 IN_MULTI_LOCK_ASSERT();
1811 query_response_timer_expired = 0;
1812 state_change_retransmit_timer_expired = 0;
1815 * During a transition from v1/v2 compatibility mode back to v3,
1816 * a group record in REPORTING state may still have its group
1817 * timer active. This is a no-op in this function; it is easier
1818 * to deal with it here than to complicate the slow-timeout path.
1820 if (inm->inm_timer == 0) {
1821 query_response_timer_expired = 0;
1822 } else if (--inm->inm_timer == 0) {
1823 query_response_timer_expired = 1;
1825 V_current_state_timers_running = 1;
1828 if (inm->inm_sctimer == 0) {
1829 state_change_retransmit_timer_expired = 0;
1830 } else if (--inm->inm_sctimer == 0) {
1831 state_change_retransmit_timer_expired = 1;
1833 V_state_change_timers_running = 1;
1836 /* We are in fasttimo, so be quick about it. */
1837 if (!state_change_retransmit_timer_expired &&
1838 !query_response_timer_expired)
1841 switch (inm->inm_state) {
1842 case IGMP_NOT_MEMBER:
1843 case IGMP_SILENT_MEMBER:
1844 case IGMP_SLEEPING_MEMBER:
1845 case IGMP_LAZY_MEMBER:
1846 case IGMP_AWAKENING_MEMBER:
1847 case IGMP_IDLE_MEMBER:
1849 case IGMP_G_QUERY_PENDING_MEMBER:
1850 case IGMP_SG_QUERY_PENDING_MEMBER:
1852 * Respond to a previously pending Group-Specific
1853 * or Group-and-Source-Specific query by enqueueing
1854 * the appropriate Current-State report for
1855 * immediate transmission.
1857 if (query_response_timer_expired) {
1860 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1861 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1862 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1864 inm->inm_state = IGMP_REPORTING_MEMBER;
1865 /* XXX Clear recorded sources for next time. */
1866 inm_clear_recorded(inm);
1869 case IGMP_REPORTING_MEMBER:
1870 case IGMP_LEAVING_MEMBER:
1871 if (state_change_retransmit_timer_expired) {
1873 * State-change retransmission timer fired.
1874 * If there are any further pending retransmissions,
1875 * set the global pending state-change flag, and
1878 if (--inm->inm_scrv > 0) {
1879 inm->inm_sctimer = uri_fasthz;
1880 V_state_change_timers_running = 1;
1883 * Retransmit the previously computed state-change
1884 * report. If there are no further pending
1885 * retransmissions, the mbuf queue will be consumed.
1886 * Update T0 state to T1 as we have now sent
1889 (void)igmp_v3_merge_state_changes(inm, scq);
1892 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1893 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1896 * If we are leaving the group for good, make sure
1897 * we release IGMP's reference to it.
1898 * This release must be deferred using a SLIST,
1899 * as we are called from a loop which traverses
1900 * the in_ifmultiaddr TAILQ.
1902 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1903 inm->inm_scrv == 0) {
1904 inm->inm_state = IGMP_NOT_MEMBER;
1905 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1915 * Suppress a group's pending response to a group or source/group query.
1917 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1918 * Do NOT update ST1/ST0 as this operation merely suppresses
1919 * the currently pending group record.
1920 * Do NOT suppress the response to a general query. It is possible but
1921 * it would require adding another state or flag.
1924 igmp_v3_suppress_group_record(struct in_multi *inm)
1927 IN_MULTI_LOCK_ASSERT();
1929 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1930 ("%s: not IGMPv3 mode on link", __func__));
1932 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1933 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1936 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1937 inm_clear_recorded(inm);
1940 inm->inm_state = IGMP_REPORTING_MEMBER;
1944 * Switch to a different IGMP version on the given interface,
1945 * as per Section 7.2.1.
1948 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1950 int old_version_timer;
1954 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1955 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1957 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1959 * Compute the "Older Version Querier Present" timer as per
1962 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1963 old_version_timer *= PR_SLOWHZ;
1965 if (version == IGMP_VERSION_1) {
1966 igi->igi_v1_timer = old_version_timer;
1967 igi->igi_v2_timer = 0;
1968 } else if (version == IGMP_VERSION_2) {
1969 igi->igi_v1_timer = 0;
1970 igi->igi_v2_timer = old_version_timer;
1974 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1975 if (igi->igi_version != IGMP_VERSION_2) {
1976 igi->igi_version = IGMP_VERSION_2;
1977 igmp_v3_cancel_link_timers(igi);
1979 } else if (igi->igi_v1_timer > 0) {
1980 if (igi->igi_version != IGMP_VERSION_1) {
1981 igi->igi_version = IGMP_VERSION_1;
1982 igmp_v3_cancel_link_timers(igi);
1988 * Cancel pending IGMPv3 timers for the given link and all groups
1989 * joined on it; state-change, general-query, and group-query timers.
1991 * Only ever called on a transition from v3 to Compatibility mode. Kill
1992 * the timers stone dead (this may be expensive for large N groups), they
1993 * will be restarted if Compatibility Mode deems that they must be due to
1997 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
1999 struct ifmultiaddr *ifma;
2001 struct in_multi *inm, *tinm;
2003 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2004 igi->igi_ifp, igi->igi_ifp->if_xname);
2006 IN_MULTI_LOCK_ASSERT();
2010 * Stop the v3 General Query Response on this link stone dead.
2011 * If fasttimo is woken up due to V_interface_timers_running,
2012 * the flag will be cleared if there are no pending link timers.
2014 igi->igi_v3_timer = 0;
2017 * Now clear the current-state and state-change report timers
2018 * for all memberships scoped to this link.
2022 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2023 if (ifma->ifma_addr->sa_family != AF_INET ||
2024 ifma->ifma_protospec == NULL)
2026 inm = (struct in_multi *)ifma->ifma_protospec;
2027 switch (inm->inm_state) {
2028 case IGMP_NOT_MEMBER:
2029 case IGMP_SILENT_MEMBER:
2030 case IGMP_IDLE_MEMBER:
2031 case IGMP_LAZY_MEMBER:
2032 case IGMP_SLEEPING_MEMBER:
2033 case IGMP_AWAKENING_MEMBER:
2035 * These states are either not relevant in v3 mode,
2036 * or are unreported. Do nothing.
2039 case IGMP_LEAVING_MEMBER:
2041 * If we are leaving the group and switching to
2042 * compatibility mode, we need to release the final
2043 * reference held for issuing the INCLUDE {}, and
2044 * transition to REPORTING to ensure the host leave
2045 * message is sent upstream to the old querier --
2046 * transition to NOT would lose the leave and race.
2048 SLIST_INSERT_HEAD(&igi->igi_relinmhead, inm, inm_nrele);
2050 case IGMP_G_QUERY_PENDING_MEMBER:
2051 case IGMP_SG_QUERY_PENDING_MEMBER:
2052 inm_clear_recorded(inm);
2054 case IGMP_REPORTING_MEMBER:
2055 inm->inm_state = IGMP_REPORTING_MEMBER;
2059 * Always clear state-change and group report timers.
2060 * Free any pending IGMPv3 state-change records.
2062 inm->inm_sctimer = 0;
2064 _IF_DRAIN(&inm->inm_scq);
2066 IF_ADDR_RUNLOCK(ifp);
2067 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, tinm) {
2068 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
2069 inm_release_locked(inm);
2074 * Update the Older Version Querier Present timers for a link.
2075 * See Section 7.2.1 of RFC 3376.
2078 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2083 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2085 * IGMPv1 and IGMPv2 Querier Present timers expired.
2089 if (igi->igi_version != IGMP_VERSION_3) {
2091 "%s: transition from v%d -> v%d on %p(%s)",
2092 __func__, igi->igi_version, IGMP_VERSION_3,
2093 igi->igi_ifp, igi->igi_ifp->if_xname);
2094 igi->igi_version = IGMP_VERSION_3;
2096 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2098 * IGMPv1 Querier Present timer expired,
2099 * IGMPv2 Querier Present timer running.
2100 * If IGMPv2 was disabled since last timeout,
2102 * If IGMPv2 is enabled, revert to IGMPv2.
2104 if (!V_igmp_v2enable) {
2106 "%s: transition from v%d -> v%d on %p(%s)",
2107 __func__, igi->igi_version, IGMP_VERSION_3,
2108 igi->igi_ifp, igi->igi_ifp->if_xname);
2109 igi->igi_v2_timer = 0;
2110 igi->igi_version = IGMP_VERSION_3;
2112 --igi->igi_v2_timer;
2113 if (igi->igi_version != IGMP_VERSION_2) {
2115 "%s: transition from v%d -> v%d on %p(%s)",
2116 __func__, igi->igi_version, IGMP_VERSION_2,
2117 igi->igi_ifp, igi->igi_ifp->if_xname);
2118 igi->igi_version = IGMP_VERSION_2;
2119 igmp_v3_cancel_link_timers(igi);
2122 } else if (igi->igi_v1_timer > 0) {
2124 * IGMPv1 Querier Present timer running.
2125 * Stop IGMPv2 timer if running.
2127 * If IGMPv1 was disabled since last timeout,
2129 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2131 if (!V_igmp_v1enable) {
2133 "%s: transition from v%d -> v%d on %p(%s)",
2134 __func__, igi->igi_version, IGMP_VERSION_3,
2135 igi->igi_ifp, igi->igi_ifp->if_xname);
2136 igi->igi_v1_timer = 0;
2137 igi->igi_version = IGMP_VERSION_3;
2139 --igi->igi_v1_timer;
2141 if (igi->igi_v2_timer > 0) {
2143 "%s: cancel v2 timer on %p(%s)",
2144 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2145 igi->igi_v2_timer = 0;
2151 * Global slowtimo handler.
2152 * VIMAGE: Timeout handlers are expected to service all vimages.
2157 VNET_ITERATOR_DECL(vnet_iter);
2159 VNET_LIST_RLOCK_NOSLEEP();
2160 VNET_FOREACH(vnet_iter) {
2161 CURVNET_SET(vnet_iter);
2162 igmp_slowtimo_vnet();
2165 VNET_LIST_RUNLOCK_NOSLEEP();
2169 * Per-vnet slowtimo handler.
2172 igmp_slowtimo_vnet(void)
2174 struct igmp_ifinfo *igi;
2178 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2179 igmp_v1v2_process_querier_timers(igi);
2186 * Dispatch an IGMPv1/v2 host report or leave message.
2187 * These are always small enough to fit inside a single mbuf.
2190 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2197 IN_MULTI_LOCK_ASSERT();
2202 m = m_gethdr(M_NOWAIT, MT_DATA);
2205 MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2207 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2209 m->m_data += sizeof(struct ip);
2210 m->m_len = sizeof(struct igmp);
2212 igmp = mtod(m, struct igmp *);
2213 igmp->igmp_type = type;
2214 igmp->igmp_code = 0;
2215 igmp->igmp_group = inm->inm_addr;
2216 igmp->igmp_cksum = 0;
2217 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2219 m->m_data -= sizeof(struct ip);
2220 m->m_len += sizeof(struct ip);
2222 ip = mtod(m, struct ip *);
2224 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2226 ip->ip_p = IPPROTO_IGMP;
2227 ip->ip_src.s_addr = INADDR_ANY;
2229 if (type == IGMP_HOST_LEAVE_MESSAGE)
2230 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2232 ip->ip_dst = inm->inm_addr;
2234 igmp_save_context(m, ifp);
2236 m->m_flags |= M_IGMPV2;
2237 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2238 m->m_flags |= M_IGMP_LOOP;
2240 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2241 netisr_dispatch(NETISR_IGMP, m);
2247 * Process a state change from the upper layer for the given IPv4 group.
2249 * Each socket holds a reference on the in_multi in its own ip_moptions.
2250 * The socket layer will have made the necessary updates to.the group
2251 * state, it is now up to IGMP to issue a state change report if there
2252 * has been any change between T0 (when the last state-change was issued)
2255 * We use the IGMPv3 state machine at group level. The IGMP module
2256 * however makes the decision as to which IGMP protocol version to speak.
2257 * A state change *from* INCLUDE {} always means an initial join.
2258 * A state change *to* INCLUDE {} always means a final leave.
2260 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2261 * save ourselves a bunch of work; any exclusive mode groups need not
2262 * compute source filter lists.
2264 * VIMAGE: curvnet should have been set by caller, as this routine
2265 * is called from the socket option handlers.
2268 igmp_change_state(struct in_multi *inm)
2270 struct igmp_ifinfo *igi;
2274 IN_MULTI_LOCK_ASSERT();
2279 * Try to detect if the upper layer just asked us to change state
2280 * for an interface which has now gone away.
2282 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2283 ifp = inm->inm_ifma->ifma_ifp;
2285 * Sanity check that netinet's notion of ifp is the
2288 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2292 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2293 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2296 * If we detect a state transition to or from MCAST_UNDEFINED
2297 * for this group, then we are starting or finishing an IGMP
2298 * life cycle for this group.
2300 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2301 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2302 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2303 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2304 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2305 error = igmp_initial_join(inm, igi);
2307 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2308 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2309 igmp_final_leave(inm, igi);
2313 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2316 error = igmp_handle_state_change(inm, igi);
2324 * Perform the initial join for an IGMP group.
2326 * When joining a group:
2327 * If the group should have its IGMP traffic suppressed, do nothing.
2328 * IGMPv1 starts sending IGMPv1 host membership reports.
2329 * IGMPv2 starts sending IGMPv2 host membership reports.
2330 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2331 * initial state of the membership.
2334 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2337 struct ifqueue *ifq;
2338 int error, retval, syncstates;
2340 CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2341 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2342 inm->inm_ifp->if_xname);
2349 IN_MULTI_LOCK_ASSERT();
2352 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2355 * Groups joined on loopback or marked as 'not reported',
2356 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2357 * are never reported in any IGMP protocol exchanges.
2358 * All other groups enter the appropriate IGMP state machine
2359 * for the version in use on this link.
2360 * A link marked as IGIF_SILENT causes IGMP to be completely
2361 * disabled for the link.
2363 if ((ifp->if_flags & IFF_LOOPBACK) ||
2364 (igi->igi_flags & IGIF_SILENT) ||
2365 !igmp_isgroupreported(inm->inm_addr)) {
2367 "%s: not kicking state machine for silent group", __func__);
2368 inm->inm_state = IGMP_SILENT_MEMBER;
2372 * Deal with overlapping in_multi lifecycle.
2373 * If this group was LEAVING, then make sure
2374 * we drop the reference we picked up to keep the
2375 * group around for the final INCLUDE {} enqueue.
2377 if (igi->igi_version == IGMP_VERSION_3 &&
2378 inm->inm_state == IGMP_LEAVING_MEMBER)
2379 inm_release_locked(inm);
2381 inm->inm_state = IGMP_REPORTING_MEMBER;
2383 switch (igi->igi_version) {
2384 case IGMP_VERSION_1:
2385 case IGMP_VERSION_2:
2386 inm->inm_state = IGMP_IDLE_MEMBER;
2387 error = igmp_v1v2_queue_report(inm,
2388 (igi->igi_version == IGMP_VERSION_2) ?
2389 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2390 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2392 inm->inm_timer = IGMP_RANDOM_DELAY(
2393 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2394 V_current_state_timers_running = 1;
2398 case IGMP_VERSION_3:
2400 * Defer update of T0 to T1, until the first copy
2401 * of the state change has been transmitted.
2406 * Immediately enqueue a State-Change Report for
2407 * this interface, freeing any previous reports.
2408 * Don't kick the timers if there is nothing to do,
2409 * or if an error occurred.
2411 ifq = &inm->inm_scq;
2413 retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2415 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2418 error = retval * -1;
2423 * Schedule transmission of pending state-change
2424 * report up to RV times for this link. The timer
2425 * will fire at the next igmp_fasttimo (~200ms),
2426 * giving us an opportunity to merge the reports.
2428 if (igi->igi_flags & IGIF_LOOPBACK) {
2431 KASSERT(igi->igi_rv > 1,
2432 ("%s: invalid robustness %d", __func__,
2434 inm->inm_scrv = igi->igi_rv;
2436 inm->inm_sctimer = 1;
2437 V_state_change_timers_running = 1;
2445 * Only update the T0 state if state change is atomic,
2446 * i.e. we don't need to wait for a timer to fire before we
2447 * can consider the state change to have been communicated.
2451 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2452 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2459 * Issue an intermediate state change during the IGMP life-cycle.
2462 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2467 CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2468 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2469 inm->inm_ifp->if_xname);
2473 IN_MULTI_LOCK_ASSERT();
2476 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2478 if ((ifp->if_flags & IFF_LOOPBACK) ||
2479 (igi->igi_flags & IGIF_SILENT) ||
2480 !igmp_isgroupreported(inm->inm_addr) ||
2481 (igi->igi_version != IGMP_VERSION_3)) {
2482 if (!igmp_isgroupreported(inm->inm_addr)) {
2484 "%s: not kicking state machine for silent group", __func__);
2486 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2488 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2489 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2493 _IF_DRAIN(&inm->inm_scq);
2495 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2496 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2501 * If record(s) were enqueued, start the state-change
2502 * report timer for this group.
2504 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2505 inm->inm_sctimer = 1;
2506 V_state_change_timers_running = 1;
2512 * Perform the final leave for an IGMP group.
2514 * When leaving a group:
2515 * IGMPv1 does nothing.
2516 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2517 * IGMPv3 enqueues a state-change report containing a transition
2518 * to INCLUDE {} for immediate transmission.
2521 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2527 CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2528 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2529 inm->inm_ifp->if_xname);
2531 IN_MULTI_LOCK_ASSERT();
2534 switch (inm->inm_state) {
2535 case IGMP_NOT_MEMBER:
2536 case IGMP_SILENT_MEMBER:
2537 case IGMP_LEAVING_MEMBER:
2538 /* Already leaving or left; do nothing. */
2540 "%s: not kicking state machine for silent group", __func__);
2542 case IGMP_REPORTING_MEMBER:
2543 case IGMP_IDLE_MEMBER:
2544 case IGMP_G_QUERY_PENDING_MEMBER:
2545 case IGMP_SG_QUERY_PENDING_MEMBER:
2546 if (igi->igi_version == IGMP_VERSION_2) {
2548 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2549 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2550 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2553 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2554 inm->inm_state = IGMP_NOT_MEMBER;
2555 } else if (igi->igi_version == IGMP_VERSION_3) {
2557 * Stop group timer and all pending reports.
2558 * Immediately enqueue a state-change report
2559 * TO_IN {} to be sent on the next fast timeout,
2560 * giving us an opportunity to merge reports.
2562 _IF_DRAIN(&inm->inm_scq);
2564 if (igi->igi_flags & IGIF_LOOPBACK) {
2567 inm->inm_scrv = igi->igi_rv;
2569 CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2570 "pending retransmissions.", __func__,
2571 inet_ntoa(inm->inm_addr),
2572 inm->inm_ifp->if_xname, inm->inm_scrv);
2573 if (inm->inm_scrv == 0) {
2574 inm->inm_state = IGMP_NOT_MEMBER;
2575 inm->inm_sctimer = 0;
2579 inm_acquire_locked(inm);
2581 retval = igmp_v3_enqueue_group_record(
2582 &inm->inm_scq, inm, 1, 0, 0);
2583 KASSERT(retval != 0,
2584 ("%s: enqueue record = %d", __func__,
2587 inm->inm_state = IGMP_LEAVING_MEMBER;
2588 inm->inm_sctimer = 1;
2589 V_state_change_timers_running = 1;
2595 case IGMP_LAZY_MEMBER:
2596 case IGMP_SLEEPING_MEMBER:
2597 case IGMP_AWAKENING_MEMBER:
2598 /* Our reports are suppressed; do nothing. */
2604 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2605 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2606 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2607 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2608 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2613 * Enqueue an IGMPv3 group record to the given output queue.
2615 * XXX This function could do with having the allocation code
2616 * split out, and the multiple-tree-walks coalesced into a single
2617 * routine as has been done in igmp_v3_enqueue_filter_change().
2619 * If is_state_change is zero, a current-state record is appended.
2620 * If is_state_change is non-zero, a state-change report is appended.
2622 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2623 * If is_group_query is zero, and if there is a packet with free space
2624 * at the tail of the queue, it will be appended to providing there
2625 * is enough free space.
2626 * Otherwise a new mbuf packet chain is allocated.
2628 * If is_source_query is non-zero, each source is checked to see if
2629 * it was recorded for a Group-Source query, and will be omitted if
2630 * it is not both in-mode and recorded.
2632 * The function will attempt to allocate leading space in the packet
2633 * for the IP/IGMP header to be prepended without fragmenting the chain.
2635 * If successful the size of all data appended to the queue is returned,
2636 * otherwise an error code less than zero is returned, or zero if
2637 * no record(s) were appended.
2640 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2641 const int is_state_change, const int is_group_query,
2642 const int is_source_query)
2644 struct igmp_grouprec ig;
2645 struct igmp_grouprec *pig;
2647 struct ip_msource *ims, *nims;
2648 struct mbuf *m0, *m, *md;
2649 int error, is_filter_list_change;
2650 int minrec0len, m0srcs, msrcs, nbytes, off;
2651 int record_has_sources;
2657 IN_MULTI_LOCK_ASSERT();
2661 is_filter_list_change = 0;
2668 record_has_sources = 1;
2670 type = IGMP_DO_NOTHING;
2671 mode = inm->inm_st[1].iss_fmode;
2674 * If we did not transition out of ASM mode during t0->t1,
2675 * and there are no source nodes to process, we can skip
2676 * the generation of source records.
2678 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2680 record_has_sources = 0;
2682 if (is_state_change) {
2684 * Queue a state change record.
2685 * If the mode did not change, and there are non-ASM
2686 * listeners or source filters present,
2687 * we potentially need to issue two records for the group.
2688 * If we are transitioning to MCAST_UNDEFINED, we need
2689 * not send any sources.
2690 * If there are ASM listeners, and there was no filter
2691 * mode transition of any kind, do nothing.
2693 if (mode != inm->inm_st[0].iss_fmode) {
2694 if (mode == MCAST_EXCLUDE) {
2695 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2697 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2699 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2701 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2702 if (mode == MCAST_UNDEFINED)
2703 record_has_sources = 0;
2706 if (record_has_sources) {
2707 is_filter_list_change = 1;
2709 type = IGMP_DO_NOTHING;
2714 * Queue a current state record.
2716 if (mode == MCAST_EXCLUDE) {
2717 type = IGMP_MODE_IS_EXCLUDE;
2718 } else if (mode == MCAST_INCLUDE) {
2719 type = IGMP_MODE_IS_INCLUDE;
2720 KASSERT(inm->inm_st[1].iss_asm == 0,
2721 ("%s: inm %p is INCLUDE but ASM count is %d",
2722 __func__, inm, inm->inm_st[1].iss_asm));
2727 * Generate the filter list changes using a separate function.
2729 if (is_filter_list_change)
2730 return (igmp_v3_enqueue_filter_change(ifq, inm));
2732 if (type == IGMP_DO_NOTHING) {
2733 CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2734 __func__, inet_ntoa(inm->inm_addr),
2735 inm->inm_ifp->if_xname);
2740 * If any sources are present, we must be able to fit at least
2741 * one in the trailing space of the tail packet's mbuf,
2744 minrec0len = sizeof(struct igmp_grouprec);
2745 if (record_has_sources)
2746 minrec0len += sizeof(in_addr_t);
2748 CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2749 igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2750 inm->inm_ifp->if_xname);
2753 * Check if we have a packet in the tail of the queue for this
2754 * group into which the first group record for this group will fit.
2755 * Otherwise allocate a new packet.
2756 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2757 * Note: Group records for G/GSR query responses MUST be sent
2758 * in their own packet.
2761 if (!is_group_query &&
2763 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2764 (m0->m_pkthdr.len + minrec0len) <
2765 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2766 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2767 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2769 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2771 if (_IF_QFULL(ifq)) {
2772 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2776 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2777 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2778 if (!is_state_change && !is_group_query) {
2779 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2781 m->m_data += IGMP_LEADINGSPACE;
2784 m = m_gethdr(M_NOWAIT, MT_DATA);
2786 MH_ALIGN(m, IGMP_LEADINGSPACE);
2791 igmp_save_context(m, ifp);
2793 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2797 * Append group record.
2798 * If we have sources, we don't know how many yet.
2803 ig.ig_group = inm->inm_addr;
2804 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2807 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2810 nbytes += sizeof(struct igmp_grouprec);
2813 * Append as many sources as will fit in the first packet.
2814 * If we are appending to a new packet, the chain allocation
2815 * may potentially use clusters; use m_getptr() in this case.
2816 * If we are appending to an existing packet, we need to obtain
2817 * a pointer to the group record after m_append(), in case a new
2818 * mbuf was allocated.
2819 * Only append sources which are in-mode at t1. If we are
2820 * transitioning to MCAST_UNDEFINED state on the group, do not
2821 * include source entries.
2822 * Only report recorded sources in our filter set when responding
2823 * to a group-source query.
2825 if (record_has_sources) {
2828 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2829 md->m_len - nbytes);
2831 md = m_getptr(m, 0, &off);
2832 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2836 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2837 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2838 inet_ntoa_haddr(ims->ims_haddr));
2839 now = ims_get_mode(inm, ims, 1);
2840 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2841 if ((now != mode) ||
2842 (now == mode && mode == MCAST_UNDEFINED)) {
2843 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2846 if (is_source_query && ims->ims_stp == 0) {
2847 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2851 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2852 naddr = htonl(ims->ims_haddr);
2853 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2856 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2860 nbytes += sizeof(in_addr_t);
2862 if (msrcs == m0srcs)
2865 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2867 pig->ig_numsrc = htons(msrcs);
2868 nbytes += (msrcs * sizeof(in_addr_t));
2871 if (is_source_query && msrcs == 0) {
2872 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2879 * We are good to go with first packet.
2882 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2883 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2884 _IF_ENQUEUE(ifq, m);
2886 m->m_pkthdr.PH_vt.vt_nrecs++;
2889 * No further work needed if no source list in packet(s).
2891 if (!record_has_sources)
2895 * Whilst sources remain to be announced, we need to allocate
2896 * a new packet and fill out as many sources as will fit.
2897 * Always try for a cluster first.
2899 while (nims != NULL) {
2900 if (_IF_QFULL(ifq)) {
2901 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2904 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2906 m->m_data += IGMP_LEADINGSPACE;
2908 m = m_gethdr(M_NOWAIT, MT_DATA);
2910 MH_ALIGN(m, IGMP_LEADINGSPACE);
2914 igmp_save_context(m, ifp);
2915 md = m_getptr(m, 0, &off);
2916 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2917 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2919 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2922 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2925 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2926 nbytes += sizeof(struct igmp_grouprec);
2928 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2929 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2932 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2933 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2934 inet_ntoa_haddr(ims->ims_haddr));
2935 now = ims_get_mode(inm, ims, 1);
2936 if ((now != mode) ||
2937 (now == mode && mode == MCAST_UNDEFINED)) {
2938 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2941 if (is_source_query && ims->ims_stp == 0) {
2942 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2946 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2947 naddr = htonl(ims->ims_haddr);
2948 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2951 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2956 if (msrcs == m0srcs)
2959 pig->ig_numsrc = htons(msrcs);
2960 nbytes += (msrcs * sizeof(in_addr_t));
2962 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2963 _IF_ENQUEUE(ifq, m);
2970 * Type used to mark record pass completion.
2971 * We exploit the fact we can cast to this easily from the
2972 * current filter modes on each ip_msource node.
2975 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2976 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2977 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2978 REC_FULL = REC_ALLOW | REC_BLOCK
2982 * Enqueue an IGMPv3 filter list change to the given output queue.
2984 * Source list filter state is held in an RB-tree. When the filter list
2985 * for a group is changed without changing its mode, we need to compute
2986 * the deltas between T0 and T1 for each source in the filter set,
2987 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2989 * As we may potentially queue two record types, and the entire R-B tree
2990 * needs to be walked at once, we break this out into its own function
2991 * so we can generate a tightly packed queue of packets.
2993 * XXX This could be written to only use one tree walk, although that makes
2994 * serializing into the mbuf chains a bit harder. For now we do two walks
2995 * which makes things easier on us, and it may or may not be harder on
2998 * If successful the size of all data appended to the queue is returned,
2999 * otherwise an error code less than zero is returned, or zero if
3000 * no record(s) were appended.
3003 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
3005 static const int MINRECLEN =
3006 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3008 struct igmp_grouprec ig;
3009 struct igmp_grouprec *pig;
3010 struct ip_msource *ims, *nims;
3011 struct mbuf *m, *m0, *md;
3013 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3015 uint8_t mode, now, then;
3016 rectype_t crt, drt, nrt;
3018 IN_MULTI_LOCK_ASSERT();
3020 if (inm->inm_nsrc == 0 ||
3021 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3024 ifp = inm->inm_ifp; /* interface */
3025 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3026 crt = REC_NONE; /* current group record type */
3027 drt = REC_NONE; /* mask of completed group record types */
3028 nrt = REC_NONE; /* record type for current node */
3029 m0srcs = 0; /* # source which will fit in current mbuf chain */
3030 nbytes = 0; /* # of bytes appended to group's state-change queue */
3031 npbytes = 0; /* # of bytes appended this packet */
3032 rsrcs = 0; /* # sources encoded in current record */
3033 schanged = 0; /* # nodes encoded in overall filter change */
3034 nallow = 0; /* # of source entries in ALLOW_NEW */
3035 nblock = 0; /* # of source entries in BLOCK_OLD */
3036 nims = NULL; /* next tree node pointer */
3039 * For each possible filter record mode.
3040 * The first kind of source we encounter tells us which
3041 * is the first kind of record we start appending.
3042 * If a node transitioned to UNDEFINED at t1, its mode is treated
3043 * as the inverse of the group's filter mode.
3045 while (drt != REC_FULL) {
3049 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3050 IGMP_V3_REPORT_MAXRECS) &&
3051 (m0->m_pkthdr.len + MINRECLEN) <
3052 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3054 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3055 sizeof(struct igmp_grouprec)) /
3058 "%s: use previous packet", __func__);
3060 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3062 m->m_data += IGMP_LEADINGSPACE;
3064 m = m_gethdr(M_NOWAIT, MT_DATA);
3066 MH_ALIGN(m, IGMP_LEADINGSPACE);
3070 "%s: m_get*() failed", __func__);
3073 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3074 igmp_save_context(m, ifp);
3075 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3076 sizeof(struct igmp_grouprec)) /
3080 "%s: allocated new packet", __func__);
3083 * Append the IGMP group record header to the
3084 * current packet's data area.
3085 * Recalculate pointer to free space for next
3086 * group record, in case m_append() allocated
3087 * a new mbuf or cluster.
3089 memset(&ig, 0, sizeof(ig));
3090 ig.ig_group = inm->inm_addr;
3091 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3095 "%s: m_append() failed", __func__);
3098 npbytes += sizeof(struct igmp_grouprec);
3100 /* new packet; offset in c hain */
3101 md = m_getptr(m, npbytes -
3102 sizeof(struct igmp_grouprec), &off);
3103 pig = (struct igmp_grouprec *)(mtod(md,
3106 /* current packet; offset from last append */
3108 pig = (struct igmp_grouprec *)(mtod(md,
3109 uint8_t *) + md->m_len -
3110 sizeof(struct igmp_grouprec));
3113 * Begin walking the tree for this record type
3114 * pass, or continue from where we left off
3115 * previously if we had to allocate a new packet.
3116 * Only report deltas in-mode at t1.
3117 * We need not report included sources as allowed
3118 * if we are in inclusive mode on the group,
3119 * however the converse is not true.
3123 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3124 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3125 CTR2(KTR_IGMPV3, "%s: visit node %s",
3126 __func__, inet_ntoa_haddr(ims->ims_haddr));
3127 now = ims_get_mode(inm, ims, 1);
3128 then = ims_get_mode(inm, ims, 0);
3129 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3130 __func__, then, now);
3133 "%s: skip unchanged", __func__);
3136 if (mode == MCAST_EXCLUDE &&
3137 now == MCAST_INCLUDE) {
3139 "%s: skip IN src on EX group",
3143 nrt = (rectype_t)now;
3144 if (nrt == REC_NONE)
3145 nrt = (rectype_t)(~mode & REC_FULL);
3146 if (schanged++ == 0) {
3148 } else if (crt != nrt)
3150 naddr = htonl(ims->ims_haddr);
3151 if (!m_append(m, sizeof(in_addr_t),
3156 "%s: m_append() failed", __func__);
3159 nallow += !!(crt == REC_ALLOW);
3160 nblock += !!(crt == REC_BLOCK);
3161 if (++rsrcs == m0srcs)
3165 * If we did not append any tree nodes on this
3166 * pass, back out of allocations.
3169 npbytes -= sizeof(struct igmp_grouprec);
3172 "%s: m_free(m)", __func__);
3176 "%s: m_adj(m, -ig)", __func__);
3177 m_adj(m, -((int)sizeof(
3178 struct igmp_grouprec)));
3182 npbytes += (rsrcs * sizeof(in_addr_t));
3183 if (crt == REC_ALLOW)
3184 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3185 else if (crt == REC_BLOCK)
3186 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3187 pig->ig_numsrc = htons(rsrcs);
3189 * Count the new group record, and enqueue this
3190 * packet if it wasn't already queued.
3192 m->m_pkthdr.PH_vt.vt_nrecs++;
3194 _IF_ENQUEUE(ifq, m);
3196 } while (nims != NULL);
3198 crt = (~crt & REC_FULL);
3201 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3208 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3211 struct mbuf *m; /* pending state-change */
3212 struct mbuf *m0; /* copy of pending state-change */
3213 struct mbuf *mt; /* last state-change in packet */
3214 int docopy, domerge;
3221 IN_MULTI_LOCK_ASSERT();
3225 * If there are further pending retransmissions, make a writable
3226 * copy of each queued state-change message before merging.
3228 if (inm->inm_scrv > 0)
3233 if (gq->ifq_head == NULL) {
3234 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3242 * Only merge the report into the current packet if
3243 * there is sufficient space to do so; an IGMPv3 report
3244 * packet may only contain 65,535 group records.
3245 * Always use a simple mbuf chain concatentation to do this,
3246 * as large state changes for single groups may have
3247 * allocated clusters.
3250 mt = ifscq->ifq_tail;
3252 recslen = m_length(m, NULL);
3254 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3255 m->m_pkthdr.PH_vt.vt_nrecs <=
3256 IGMP_V3_REPORT_MAXRECS) &&
3257 (mt->m_pkthdr.len + recslen <=
3258 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3262 if (!domerge && _IF_QFULL(gq)) {
3264 "%s: outbound queue full, skipping whole packet %p",
3274 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3275 _IF_DEQUEUE(gq, m0);
3278 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3279 m0 = m_dup(m, M_NOWAIT);
3282 m0->m_nextpkt = NULL;
3287 CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3288 __func__, m0, ifscq);
3289 _IF_ENQUEUE(ifscq, m0);
3291 struct mbuf *mtl; /* last mbuf of packet mt */
3293 CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3297 m0->m_flags &= ~M_PKTHDR;
3298 mt->m_pkthdr.len += recslen;
3299 mt->m_pkthdr.PH_vt.vt_nrecs +=
3300 m0->m_pkthdr.PH_vt.vt_nrecs;
3310 * Respond to a pending IGMPv3 General Query.
3313 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3315 struct ifmultiaddr *ifma;
3317 struct in_multi *inm;
3320 IN_MULTI_LOCK_ASSERT();
3323 KASSERT(igi->igi_version == IGMP_VERSION_3,
3324 ("%s: called when version %d", __func__, igi->igi_version));
3327 * Check that there are some packets queued. If so, send them first.
3328 * For large number of groups the reply to general query can take
3329 * many packets, we should finish sending them before starting of
3330 * queuing the new reply.
3332 if (igi->igi_gq.ifq_head != NULL)
3338 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3339 if (ifma->ifma_addr->sa_family != AF_INET ||
3340 ifma->ifma_protospec == NULL)
3343 inm = (struct in_multi *)ifma->ifma_protospec;
3344 KASSERT(ifp == inm->inm_ifp,
3345 ("%s: inconsistent ifp", __func__));
3347 switch (inm->inm_state) {
3348 case IGMP_NOT_MEMBER:
3349 case IGMP_SILENT_MEMBER:
3351 case IGMP_REPORTING_MEMBER:
3352 case IGMP_IDLE_MEMBER:
3353 case IGMP_LAZY_MEMBER:
3354 case IGMP_SLEEPING_MEMBER:
3355 case IGMP_AWAKENING_MEMBER:
3356 inm->inm_state = IGMP_REPORTING_MEMBER;
3357 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3359 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3362 case IGMP_G_QUERY_PENDING_MEMBER:
3363 case IGMP_SG_QUERY_PENDING_MEMBER:
3364 case IGMP_LEAVING_MEMBER:
3368 IF_ADDR_RUNLOCK(ifp);
3371 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3372 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3375 * Slew transmission of bursts over 500ms intervals.
3377 if (igi->igi_gq.ifq_head != NULL) {
3378 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3379 IGMP_RESPONSE_BURST_INTERVAL);
3380 V_interface_timers_running = 1;
3385 * Transmit the next pending IGMP message in the output queue.
3387 * We get called from netisr_processqueue(). A mutex private to igmpoq
3388 * will be acquired and released around this routine.
3390 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3391 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3392 * a link and uses a link-scope multicast address.
3395 igmp_intr(struct mbuf *m)
3397 struct ip_moptions imo;
3399 struct mbuf *ipopts, *m0;
3403 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3406 * Set VNET image pointer from enqueued mbuf chain
3407 * before doing anything else. Whilst we use interface
3408 * indexes to guard against interface detach, they are
3409 * unique to each VIMAGE and must be retrieved.
3411 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3412 ifindex = igmp_restore_context(m);
3415 * Check if the ifnet still exists. This limits the scope of
3416 * any race in the absence of a global ifp lock for low cost
3417 * (an array lookup).
3419 ifp = ifnet_byindex(ifindex);
3421 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3422 __func__, m, ifindex);
3424 IPSTAT_INC(ips_noroute);
3428 ipopts = V_igmp_sendra ? m_raopt : NULL;
3430 imo.imo_multicast_ttl = 1;
3431 imo.imo_multicast_vif = -1;
3432 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3435 * If the user requested that IGMP traffic be explicitly
3436 * redirected to the loopback interface (e.g. they are running a
3437 * MANET interface and the routing protocol needs to see the
3438 * updates), handle this now.
3440 if (m->m_flags & M_IGMP_LOOP)
3441 imo.imo_multicast_ifp = V_loif;
3443 imo.imo_multicast_ifp = ifp;
3445 if (m->m_flags & M_IGMPV2) {
3448 m0 = igmp_v3_encap_report(ifp, m);
3450 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3452 IPSTAT_INC(ips_odropped);
3457 igmp_scrub_context(m0);
3459 m0->m_pkthdr.rcvif = V_loif;
3461 mac_netinet_igmp_send(ifp, m0);
3463 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3465 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3469 IGMPSTAT_INC(igps_snd_reports);
3473 * We must restore the existing vnet pointer before
3474 * continuing as we are run from netisr context.
3480 * Encapsulate an IGMPv3 report.
3482 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3483 * chain has already had its IP/IGMPv3 header prepended. In this case
3484 * the function will not attempt to prepend; the lengths and checksums
3485 * will however be re-computed.
3487 * Returns a pointer to the new mbuf chain head, or NULL if the
3488 * allocation failed.
3490 static struct mbuf *
3491 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3493 struct igmp_report *igmp;
3495 int hdrlen, igmpreclen;
3497 KASSERT((m->m_flags & M_PKTHDR),
3498 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3500 igmpreclen = m_length(m, NULL);
3501 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3503 if (m->m_flags & M_IGMPV3_HDR) {
3504 igmpreclen -= hdrlen;
3506 M_PREPEND(m, hdrlen, M_NOWAIT);
3509 m->m_flags |= M_IGMPV3_HDR;
3512 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3514 m->m_data += sizeof(struct ip);
3515 m->m_len -= sizeof(struct ip);
3517 igmp = mtod(m, struct igmp_report *);
3518 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3521 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3523 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3524 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3526 m->m_data -= sizeof(struct ip);
3527 m->m_len += sizeof(struct ip);
3529 ip = mtod(m, struct ip *);
3530 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3531 ip->ip_len = htons(hdrlen + igmpreclen);
3532 ip->ip_off = htons(IP_DF);
3533 ip->ip_p = IPPROTO_IGMP;
3536 ip->ip_src.s_addr = INADDR_ANY;
3538 if (m->m_flags & M_IGMP_LOOP) {
3539 struct in_ifaddr *ia;
3543 ip->ip_src = ia->ia_addr.sin_addr;
3544 ifa_free(&ia->ia_ifa);
3548 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3555 igmp_rec_type_to_str(const int type)
3559 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3562 case IGMP_CHANGE_TO_INCLUDE_MODE:
3565 case IGMP_MODE_IS_EXCLUDE:
3568 case IGMP_MODE_IS_INCLUDE:
3571 case IGMP_ALLOW_NEW_SOURCES:
3574 case IGMP_BLOCK_OLD_SOURCES:
3585 igmp_init(void *unused __unused)
3588 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3592 m_raopt = igmp_ra_alloc();
3594 netisr_register(&igmp_nh);
3596 SYSINIT(igmp_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_init, NULL);
3599 igmp_uninit(void *unused __unused)
3602 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3604 netisr_unregister(&igmp_nh);
3609 IGMP_LOCK_DESTROY();
3611 SYSUNINIT(igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_uninit, NULL);
3614 vnet_igmp_init(const void *unused __unused)
3617 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3619 LIST_INIT(&V_igi_head);
3621 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_igmp_init,
3625 vnet_igmp_uninit(const void *unused __unused)
3628 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3630 KASSERT(LIST_EMPTY(&V_igi_head),
3631 ("%s: igi list not empty; ifnets not detached?", __func__));
3633 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY,
3634 vnet_igmp_uninit, NULL);
3637 igmp_modevent(module_t mod, int type, void *unused __unused)
3645 return (EOPNOTSUPP);
3650 static moduledata_t igmp_mod = {
3655 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);