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("$FreeBSD$");
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/module.h>
58 #include <sys/malloc.h>
60 #include <sys/socket.h>
61 #include <sys/protosw.h>
62 #include <sys/kernel.h>
64 #include <sys/rmlock.h>
65 #include <sys/sysctl.h>
67 #include <sys/condvar.h>
74 #include <net/if_var.h>
75 #include <net/netisr.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_var.h>
83 #include <netinet/ip_options.h>
84 #include <netinet/igmp.h>
85 #include <netinet/igmp_var.h>
87 #include <machine/in_cksum.h>
89 #include <security/mac/mac_framework.h>
92 #define KTR_IGMPV3 KTR_INET
95 static struct igmp_ifsoftc *
96 igi_alloc_locked(struct ifnet *);
97 static void igi_delete_locked(const struct ifnet *);
98 static void igmp_dispatch_queue(struct mbufq *, int, const int);
99 static void igmp_fasttimo_vnet(void);
100 static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
101 static int igmp_handle_state_change(struct in_multi *,
102 struct igmp_ifsoftc *);
103 static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
104 static int igmp_input_v1_query(struct ifnet *, const struct ip *,
105 const struct igmp *);
106 static int igmp_input_v2_query(struct ifnet *, const struct ip *,
107 const struct igmp *);
108 static int igmp_input_v3_query(struct ifnet *, const struct ip *,
109 /*const*/ struct igmpv3 *);
110 static int igmp_input_v3_group_query(struct in_multi *,
111 struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
112 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
113 /*const*/ struct igmp *);
114 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
115 /*const*/ struct igmp *);
116 static void igmp_intr(struct mbuf *);
117 static int igmp_isgroupreported(const struct in_addr);
121 static char * igmp_rec_type_to_str(const int);
123 static void igmp_set_version(struct igmp_ifsoftc *, const int);
124 static void igmp_slowtimo_vnet(void);
125 static int igmp_v1v2_queue_report(struct in_multi *, const int);
126 static void igmp_v1v2_process_group_timer(struct in_multi *, const int);
127 static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
128 static void igmp_v2_update_group(struct in_multi *, const int);
129 static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
130 static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
132 igmp_v3_encap_report(struct ifnet *, struct mbuf *);
133 static int igmp_v3_enqueue_group_record(struct mbufq *,
134 struct in_multi *, const int, const int, const int);
135 static int igmp_v3_enqueue_filter_change(struct mbufq *,
137 static void igmp_v3_process_group_timers(struct igmp_ifsoftc *,
138 struct mbufq *, struct mbufq *, struct in_multi *,
140 static int igmp_v3_merge_state_changes(struct in_multi *,
142 static void igmp_v3_suppress_group_record(struct in_multi *);
143 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
144 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
145 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
147 static const struct netisr_handler igmp_nh = {
149 .nh_handler = igmp_intr,
150 .nh_proto = NETISR_IGMP,
151 .nh_policy = NETISR_POLICY_SOURCE,
155 * System-wide globals.
157 * Unlocked access to these is OK, except for the global IGMP output
158 * queue. The IGMP subsystem lock ends up being system-wide for the moment,
159 * because all VIMAGEs have to share a global output queue, as netisrs
160 * themselves are not virtualized.
163 * * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
164 * Any may be taken independently; if any are held at the same
165 * time, the above lock order must be followed.
166 * * All output is delegated to the netisr.
167 * Now that Giant has been eliminated, the netisr may be inlined.
168 * * IN_MULTI_LOCK covers in_multi.
169 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
170 * including the output queue.
171 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
172 * per-link state iterators.
173 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
174 * therefore it is not refcounted.
175 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
178 * * IGMP acquires its own reference every time an in_multi is passed to
179 * it and the group is being joined for the first time.
180 * * IGMP releases its reference(s) on in_multi in a deferred way,
181 * because the operations which process the release run as part of
182 * a loop whose control variables are directly affected by the release
183 * (that, and not recursing on the IF_ADDR_LOCK).
185 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
186 * to a vnet in ifp->if_vnet.
188 * SMPng: XXX We may potentially race operations on ifma_protospec.
189 * The problem is that we currently lack a clean way of taking the
190 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
191 * as anything which modifies ifma needs to be covered by that lock.
192 * So check for ifma_protospec being NULL before proceeding.
196 struct mbuf *m_raopt; /* Router Alert option */
197 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
200 * VIMAGE-wide globals.
202 * The IGMPv3 timers themselves need to run per-image, however,
203 * protosw timers run globally (see tcp).
204 * An ifnet can only be in one vimage at a time, and the loopback
205 * ifnet, loif, is itself virtualized.
206 * It would otherwise be possible to seriously hose IGMP state,
207 * and create inconsistencies in upstream multicast routing, if you have
208 * multiple VIMAGEs running on the same link joining different multicast
209 * groups, UNLESS the "primary IP address" is different. This is because
210 * IGMP for IPv4 does not force link-local addresses to be used for each
211 * node, unlike MLD for IPv6.
212 * Obviously the IGMPv3 per-interface state has per-vimage granularity
215 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
216 * policy to control the address used by IGMP on the link.
218 static VNET_DEFINE(int, interface_timers_running); /* IGMPv3 general
220 static VNET_DEFINE(int, state_change_timers_running); /* IGMPv3 state-change
222 static VNET_DEFINE(int, current_state_timers_running); /* IGMPv1/v2 host
223 * report; IGMPv3 g/sg
226 #define V_interface_timers_running VNET(interface_timers_running)
227 #define V_state_change_timers_running VNET(state_change_timers_running)
228 #define V_current_state_timers_running VNET(current_state_timers_running)
230 static VNET_DEFINE(LIST_HEAD(, igmp_ifsoftc), igi_head) =
231 LIST_HEAD_INITIALIZER(igi_head);
232 static VNET_DEFINE(struct igmpstat, igmpstat) = {
233 .igps_version = IGPS_VERSION_3,
234 .igps_len = sizeof(struct igmpstat),
236 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0};
238 #define V_igi_head VNET(igi_head)
239 #define V_igmpstat VNET(igmpstat)
240 #define V_igmp_gsrdelay VNET(igmp_gsrdelay)
242 static VNET_DEFINE(int, igmp_recvifkludge) = 1;
243 static VNET_DEFINE(int, igmp_sendra) = 1;
244 static VNET_DEFINE(int, igmp_sendlocal) = 1;
245 static VNET_DEFINE(int, igmp_v1enable) = 1;
246 static VNET_DEFINE(int, igmp_v2enable) = 1;
247 static VNET_DEFINE(int, igmp_legacysupp);
248 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3;
250 #define V_igmp_recvifkludge VNET(igmp_recvifkludge)
251 #define V_igmp_sendra VNET(igmp_sendra)
252 #define V_igmp_sendlocal VNET(igmp_sendlocal)
253 #define V_igmp_v1enable VNET(igmp_v1enable)
254 #define V_igmp_v2enable VNET(igmp_v2enable)
255 #define V_igmp_legacysupp VNET(igmp_legacysupp)
256 #define V_igmp_default_version VNET(igmp_default_version)
259 * Virtualized sysctls.
261 SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW,
262 &VNET_NAME(igmpstat), igmpstat, "");
263 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
264 &VNET_NAME(igmp_recvifkludge), 0,
265 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
266 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
267 &VNET_NAME(igmp_sendra), 0,
268 "Send IP Router Alert option in IGMPv2/v3 messages");
269 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
270 &VNET_NAME(igmp_sendlocal), 0,
271 "Send IGMP membership reports for 224.0.0.0/24 groups");
272 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
273 &VNET_NAME(igmp_v1enable), 0,
274 "Enable backwards compatibility with IGMPv1");
275 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
276 &VNET_NAME(igmp_v2enable), 0,
277 "Enable backwards compatibility with IGMPv2");
278 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
279 &VNET_NAME(igmp_legacysupp), 0,
280 "Allow v1/v2 reports to suppress v3 group responses");
281 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
282 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
283 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
284 "Default version of IGMP to run on each interface");
285 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
286 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
287 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
288 "Rate limit for IGMPv3 Group-and-Source queries in seconds");
291 * Non-virtualized sysctls.
293 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
294 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
295 "Per-interface IGMPv3 state");
298 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
302 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
304 m->m_pkthdr.flowid = ifp->if_index;
308 igmp_scrub_context(struct mbuf *m)
311 m->m_pkthdr.PH_loc.ptr = NULL;
312 m->m_pkthdr.flowid = 0;
316 static __inline char *
317 inet_ntoa_haddr(in_addr_t haddr, char *addrbuf)
321 ia.s_addr = htonl(haddr);
322 return (inet_ntoa_r(ia, addrbuf));
327 * Restore context from a queued IGMP output chain.
328 * Return saved ifindex.
330 * VIMAGE: The assertion is there to make sure that we
331 * actually called CURVNET_SET() with what's in the mbuf chain.
333 static __inline uint32_t
334 igmp_restore_context(struct mbuf *m)
338 #if defined(VIMAGE) && defined(INVARIANTS)
339 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
340 ("%s: called when curvnet was not restored", __func__));
343 return (m->m_pkthdr.flowid);
347 * Retrieve or set default IGMP version.
349 * VIMAGE: Assume curvnet set by caller.
350 * SMPng: NOTE: Serialized by IGMP lock.
353 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
358 error = sysctl_wire_old_buffer(req, sizeof(int));
364 new = V_igmp_default_version;
366 error = sysctl_handle_int(oidp, &new, 0, req);
367 if (error || !req->newptr)
370 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
375 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
376 V_igmp_default_version, new);
378 V_igmp_default_version = new;
386 * Retrieve or set threshold between group-source queries in seconds.
388 * VIMAGE: Assume curvnet set by caller.
389 * SMPng: NOTE: Serialized by IGMP lock.
392 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
397 error = sysctl_wire_old_buffer(req, sizeof(int));
403 i = V_igmp_gsrdelay.tv_sec;
405 error = sysctl_handle_int(oidp, &i, 0, req);
406 if (error || !req->newptr)
409 if (i < -1 || i >= 60) {
414 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
415 V_igmp_gsrdelay.tv_sec, i);
416 V_igmp_gsrdelay.tv_sec = i;
424 * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
425 * For use by ifmcstat(8).
427 * SMPng: NOTE: Does an unlocked ifindex space read.
428 * VIMAGE: Assume curvnet set by caller. The node handler itself
429 * is not directly virtualized.
432 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
438 struct igmp_ifsoftc *igi;
443 if (req->newptr != NULL)
449 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
456 if (name[0] <= 0 || name[0] > V_if_index) {
463 ifp = ifnet_byindex(name[0]);
467 LIST_FOREACH(igi, &V_igi_head, igi_link) {
468 if (ifp == igi->igi_ifp) {
469 struct igmp_ifinfo info;
471 info.igi_version = igi->igi_version;
472 info.igi_v1_timer = igi->igi_v1_timer;
473 info.igi_v2_timer = igi->igi_v2_timer;
474 info.igi_v3_timer = igi->igi_v3_timer;
475 info.igi_flags = igi->igi_flags;
476 info.igi_rv = igi->igi_rv;
477 info.igi_qi = igi->igi_qi;
478 info.igi_qri = igi->igi_qri;
479 info.igi_uri = igi->igi_uri;
480 error = SYSCTL_OUT(req, &info, sizeof(info));
492 * Dispatch an entire queue of pending packet chains
494 * VIMAGE: Assumes the vnet pointer has been set.
497 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
501 while ((m = mbufq_dequeue(mq)) != NULL) {
502 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
504 m->m_flags |= M_IGMP_LOOP;
505 netisr_dispatch(NETISR_IGMP, m);
512 * Filter outgoing IGMP report state by group.
514 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
515 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
516 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
517 * this may break certain IGMP snooping switches which rely on the old
520 * Return zero if the given group is one for which IGMP reports
521 * should be suppressed, or non-zero if reports should be issued.
524 igmp_isgroupreported(const struct in_addr addr)
527 if (in_allhosts(addr) ||
528 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
535 * Construct a Router Alert option to use in outgoing packets.
543 m = m_get(M_WAITOK, MT_DATA);
544 p = mtod(m, struct ipoption *);
545 p->ipopt_dst.s_addr = INADDR_ANY;
546 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */
547 p->ipopt_list[1] = 0x04; /* 4 bytes long */
548 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
549 p->ipopt_list[3] = 0x00; /* pad byte */
550 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
556 * Attach IGMP when PF_INET is attached to an interface.
558 struct igmp_ifsoftc *
559 igmp_domifattach(struct ifnet *ifp)
561 struct igmp_ifsoftc *igi;
563 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
564 __func__, ifp, ifp->if_xname);
568 igi = igi_alloc_locked(ifp);
569 if (!(ifp->if_flags & IFF_MULTICAST))
570 igi->igi_flags |= IGIF_SILENT;
578 * VIMAGE: assume curvnet set by caller.
580 static struct igmp_ifsoftc *
581 igi_alloc_locked(/*const*/ struct ifnet *ifp)
583 struct igmp_ifsoftc *igi;
587 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
592 igi->igi_version = V_igmp_default_version;
594 igi->igi_rv = IGMP_RV_INIT;
595 igi->igi_qi = IGMP_QI_INIT;
596 igi->igi_qri = IGMP_QRI_INIT;
597 igi->igi_uri = IGMP_URI_INIT;
598 SLIST_INIT(&igi->igi_relinmhead);
599 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
601 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
603 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
613 * NOTE: Some finalization tasks need to run before the protocol domain
614 * is detached, but also before the link layer does its cleanup.
616 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
617 * XXX This is also bitten by unlocked ifma_protospec access.
620 igmp_ifdetach(struct ifnet *ifp)
622 struct igmp_ifsoftc *igi;
623 struct ifmultiaddr *ifma;
624 struct in_multi *inm, *tinm;
626 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
631 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
632 if (igi->igi_version == IGMP_VERSION_3) {
634 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
635 if (ifma->ifma_addr->sa_family != AF_INET ||
636 ifma->ifma_protospec == NULL)
639 KASSERT(ifma->ifma_protospec != NULL,
640 ("%s: ifma_protospec is NULL", __func__));
642 inm = (struct in_multi *)ifma->ifma_protospec;
643 if (inm->inm_state == IGMP_LEAVING_MEMBER) {
644 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
647 inm_clear_recorded(inm);
649 IF_ADDR_RUNLOCK(ifp);
651 * Free the in_multi reference(s) for this IGMP lifecycle.
653 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
655 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
656 inm_release_locked(inm);
664 * Hook for domifdetach.
667 igmp_domifdetach(struct ifnet *ifp)
670 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
671 __func__, ifp, ifp->if_xname);
674 igi_delete_locked(ifp);
679 igi_delete_locked(const struct ifnet *ifp)
681 struct igmp_ifsoftc *igi, *tigi;
683 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
684 __func__, ifp, ifp->if_xname);
688 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
689 if (igi->igi_ifp == ifp) {
691 * Free deferred General Query responses.
693 mbufq_drain(&igi->igi_gq);
695 LIST_REMOVE(igi, igi_link);
697 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
698 ("%s: there are dangling in_multi references",
708 * Process a received IGMPv1 query.
709 * Return non-zero if the message should be dropped.
711 * VIMAGE: The curvnet pointer is derived from the input ifp.
714 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
715 const struct igmp *igmp)
717 struct ifmultiaddr *ifma;
718 struct igmp_ifsoftc *igi;
719 struct in_multi *inm;
722 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
723 * 224.0.0.1. They are always treated as General Queries.
724 * igmp_group is always ignored. Do not drop it as a userland
725 * daemon may wish to see it.
726 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
728 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
729 IGMPSTAT_INC(igps_rcv_badqueries);
732 IGMPSTAT_INC(igps_rcv_gen_queries);
737 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
738 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
740 if (igi->igi_flags & IGIF_LOOPBACK) {
741 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
747 * Switch to IGMPv1 host compatibility mode.
749 igmp_set_version(igi, IGMP_VERSION_1);
751 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
754 * Start the timers in all of our group records
755 * for the interface on which the query arrived,
756 * except those which are already running.
759 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
760 if (ifma->ifma_addr->sa_family != AF_INET ||
761 ifma->ifma_protospec == NULL)
763 inm = (struct in_multi *)ifma->ifma_protospec;
764 if (inm->inm_timer != 0)
766 switch (inm->inm_state) {
767 case IGMP_NOT_MEMBER:
768 case IGMP_SILENT_MEMBER:
770 case IGMP_G_QUERY_PENDING_MEMBER:
771 case IGMP_SG_QUERY_PENDING_MEMBER:
772 case IGMP_REPORTING_MEMBER:
773 case IGMP_IDLE_MEMBER:
774 case IGMP_LAZY_MEMBER:
775 case IGMP_SLEEPING_MEMBER:
776 case IGMP_AWAKENING_MEMBER:
777 inm->inm_state = IGMP_REPORTING_MEMBER;
778 inm->inm_timer = IGMP_RANDOM_DELAY(
779 IGMP_V1V2_MAX_RI * PR_FASTHZ);
780 V_current_state_timers_running = 1;
782 case IGMP_LEAVING_MEMBER:
786 IF_ADDR_RUNLOCK(ifp);
796 * Process a received IGMPv2 general or group-specific query.
799 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
800 const struct igmp *igmp)
802 struct ifmultiaddr *ifma;
803 struct igmp_ifsoftc *igi;
804 struct in_multi *inm;
805 int is_general_query;
808 char addrbuf[INET_ADDRSTRLEN];
811 is_general_query = 0;
814 * Validate address fields upfront.
815 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
817 if (in_nullhost(igmp->igmp_group)) {
819 * IGMPv2 General Query.
820 * If this was not sent to the all-hosts group, ignore it.
822 if (!in_allhosts(ip->ip_dst))
824 IGMPSTAT_INC(igps_rcv_gen_queries);
825 is_general_query = 1;
827 /* IGMPv2 Group-Specific Query. */
828 IGMPSTAT_INC(igps_rcv_group_queries);
834 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
835 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
837 if (igi->igi_flags & IGIF_LOOPBACK) {
838 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
844 * Ignore v2 query if in v1 Compatibility Mode.
846 if (igi->igi_version == IGMP_VERSION_1)
849 igmp_set_version(igi, IGMP_VERSION_2);
851 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
855 if (is_general_query) {
857 * For each reporting group joined on this
858 * interface, kick the report timer.
860 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
863 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
864 if (ifma->ifma_addr->sa_family != AF_INET ||
865 ifma->ifma_protospec == NULL)
867 inm = (struct in_multi *)ifma->ifma_protospec;
868 igmp_v2_update_group(inm, timer);
870 IF_ADDR_RUNLOCK(ifp);
873 * Group-specific IGMPv2 query, we need only
874 * look up the single group to process it.
876 inm = inm_lookup(ifp, igmp->igmp_group);
878 CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
879 inet_ntoa_r(igmp->igmp_group, addrbuf), ifp,
881 igmp_v2_update_group(inm, timer);
893 * Update the report timer on a group in response to an IGMPv2 query.
895 * If we are becoming the reporting member for this group, start the timer.
896 * If we already are the reporting member for this group, and timer is
897 * below the threshold, reset it.
899 * We may be updating the group for the first time since we switched
900 * to IGMPv3. If we are, then we must clear any recorded source lists,
901 * and transition to REPORTING state; the group timer is overloaded
902 * for group and group-source query responses.
904 * Unlike IGMPv3, the delay per group should be jittered
905 * to avoid bursts of IGMPv2 reports.
908 igmp_v2_update_group(struct in_multi *inm, const int timer)
911 char addrbuf[INET_ADDRSTRLEN];
914 CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
915 inet_ntoa_r(inm->inm_addr, addrbuf), inm->inm_ifp->if_xname, timer);
917 IN_MULTI_LOCK_ASSERT();
919 switch (inm->inm_state) {
920 case IGMP_NOT_MEMBER:
921 case IGMP_SILENT_MEMBER:
923 case IGMP_REPORTING_MEMBER:
924 if (inm->inm_timer != 0 &&
925 inm->inm_timer <= timer) {
926 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
927 "skipping.", __func__);
931 case IGMP_SG_QUERY_PENDING_MEMBER:
932 case IGMP_G_QUERY_PENDING_MEMBER:
933 case IGMP_IDLE_MEMBER:
934 case IGMP_LAZY_MEMBER:
935 case IGMP_AWAKENING_MEMBER:
936 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
937 inm->inm_state = IGMP_REPORTING_MEMBER;
938 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
939 V_current_state_timers_running = 1;
941 case IGMP_SLEEPING_MEMBER:
942 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
943 inm->inm_state = IGMP_AWAKENING_MEMBER;
945 case IGMP_LEAVING_MEMBER:
951 * Process a received IGMPv3 general, group-specific or
952 * group-and-source-specific query.
953 * Assumes m has already been pulled up to the full IGMP message length.
954 * Return 0 if successful, otherwise an appropriate error code is returned.
957 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
958 /*const*/ struct igmpv3 *igmpv3)
960 struct igmp_ifsoftc *igi;
961 struct in_multi *inm;
962 int is_general_query;
963 uint32_t maxresp, nsrc, qqi;
967 char addrbuf[INET_ADDRSTRLEN];
970 is_general_query = 0;
972 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
974 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
975 if (maxresp >= 128) {
976 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
977 (IGMP_EXP(igmpv3->igmp_code) + 3);
981 * Robustness must never be less than 2 for on-wire IGMPv3.
982 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
983 * an exception for interfaces whose IGMPv3 state changes
984 * are redirected to loopback (e.g. MANET).
986 qrv = IGMP_QRV(igmpv3->igmp_misc);
988 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
993 qqi = igmpv3->igmp_qqi;
995 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
996 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
999 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
1003 nsrc = ntohs(igmpv3->igmp_numsrc);
1006 * Validate address fields and versions upfront before
1007 * accepting v3 query.
1008 * XXX SMPng: Unlocked access to igmpstat counters here.
1010 if (in_nullhost(igmpv3->igmp_group)) {
1012 * IGMPv3 General Query.
1014 * General Queries SHOULD be directed to 224.0.0.1.
1015 * A general query with a source list has undefined
1016 * behaviour; discard it.
1018 IGMPSTAT_INC(igps_rcv_gen_queries);
1019 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1020 IGMPSTAT_INC(igps_rcv_badqueries);
1023 is_general_query = 1;
1025 /* Group or group-source specific query. */
1027 IGMPSTAT_INC(igps_rcv_group_queries);
1029 IGMPSTAT_INC(igps_rcv_gsr_queries);
1035 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1036 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1038 if (igi->igi_flags & IGIF_LOOPBACK) {
1039 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1040 ifp, ifp->if_xname);
1045 * Discard the v3 query if we're in Compatibility Mode.
1046 * The RFC is not obviously worded that hosts need to stay in
1047 * compatibility mode until the Old Version Querier Present
1050 if (igi->igi_version != IGMP_VERSION_3) {
1051 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1052 igi->igi_version, ifp, ifp->if_xname);
1056 igmp_set_version(igi, IGMP_VERSION_3);
1059 igi->igi_qri = maxresp;
1061 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1064 if (is_general_query) {
1066 * Schedule a current-state report on this ifp for
1067 * all groups, possibly containing source lists.
1068 * If there is a pending General Query response
1069 * scheduled earlier than the selected delay, do
1070 * not schedule any other reports.
1071 * Otherwise, reset the interface timer.
1073 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1074 ifp, ifp->if_xname);
1075 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1076 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1077 V_interface_timers_running = 1;
1081 * Group-source-specific queries are throttled on
1082 * a per-group basis to defeat denial-of-service attempts.
1083 * Queries for groups we are not a member of on this
1084 * link are simply ignored.
1086 inm = inm_lookup(ifp, igmpv3->igmp_group);
1090 if (!ratecheck(&inm->inm_lastgsrtv,
1091 &V_igmp_gsrdelay)) {
1092 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1094 IGMPSTAT_INC(igps_drop_gsr_queries);
1098 CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1099 inet_ntoa_r(igmpv3->igmp_group, addrbuf), ifp,
1102 * If there is a pending General Query response
1103 * scheduled sooner than the selected delay, no
1104 * further report need be scheduled.
1105 * Otherwise, prepare to respond to the
1106 * group-specific or group-and-source query.
1108 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1109 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1120 * Process a received IGMPv3 group-specific or group-and-source-specific
1122 * Return <0 if any error occurred. Currently this is ignored.
1125 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1126 int timer, /*const*/ struct igmpv3 *igmpv3)
1131 IN_MULTI_LOCK_ASSERT();
1136 switch (inm->inm_state) {
1137 case IGMP_NOT_MEMBER:
1138 case IGMP_SILENT_MEMBER:
1139 case IGMP_SLEEPING_MEMBER:
1140 case IGMP_LAZY_MEMBER:
1141 case IGMP_AWAKENING_MEMBER:
1142 case IGMP_IDLE_MEMBER:
1143 case IGMP_LEAVING_MEMBER:
1146 case IGMP_REPORTING_MEMBER:
1147 case IGMP_G_QUERY_PENDING_MEMBER:
1148 case IGMP_SG_QUERY_PENDING_MEMBER:
1152 nsrc = ntohs(igmpv3->igmp_numsrc);
1155 * Deal with group-specific queries upfront.
1156 * If any group query is already pending, purge any recorded
1157 * source-list state if it exists, and schedule a query response
1158 * for this group-specific query.
1161 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1162 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1163 inm_clear_recorded(inm);
1164 timer = min(inm->inm_timer, timer);
1166 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1167 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1168 V_current_state_timers_running = 1;
1173 * Deal with the case where a group-and-source-specific query has
1174 * been received but a group-specific query is already pending.
1176 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1177 timer = min(inm->inm_timer, timer);
1178 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1179 V_current_state_timers_running = 1;
1184 * Finally, deal with the case where a group-and-source-specific
1185 * query has been received, where a response to a previous g-s-r
1186 * query exists, or none exists.
1187 * In this case, we need to parse the source-list which the Querier
1188 * has provided us with and check if we have any source list filter
1189 * entries at T1 for these sources. If we do not, there is no need
1190 * schedule a report and the query may be dropped.
1191 * If we do, we must record them and schedule a current-state
1192 * report for those sources.
1193 * FIXME: Handling source lists larger than 1 mbuf requires that
1194 * we pass the mbuf chain pointer down to this function, and use
1195 * m_getptr() to walk the chain.
1197 if (inm->inm_nsrc > 0) {
1198 const struct in_addr *ap;
1201 ap = (const struct in_addr *)(igmpv3 + 1);
1203 for (i = 0; i < nsrc; i++, ap++) {
1204 retval = inm_record_source(inm, ap->s_addr);
1207 nrecorded += retval;
1209 if (nrecorded > 0) {
1211 "%s: schedule response to SG query", __func__);
1212 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1213 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1214 V_current_state_timers_running = 1;
1222 * Process a received IGMPv1 host membership report.
1224 * NOTE: 0.0.0.0 workaround breaks const correctness.
1227 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1228 /*const*/ struct igmp *igmp)
1230 struct rm_priotracker in_ifa_tracker;
1231 struct in_ifaddr *ia;
1232 struct in_multi *inm;
1234 char addrbuf[INET_ADDRSTRLEN];
1237 IGMPSTAT_INC(igps_rcv_reports);
1239 if (ifp->if_flags & IFF_LOOPBACK)
1242 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1243 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1244 IGMPSTAT_INC(igps_rcv_badreports);
1249 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1250 * Booting clients may use the source address 0.0.0.0. Some
1251 * IGMP daemons may not know how to use IP_RECVIF to determine
1252 * the interface upon which this message was received.
1253 * Replace 0.0.0.0 with the subnet address if told to do so.
1255 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1256 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1258 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1259 ifa_free(&ia->ia_ifa);
1263 CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1264 inet_ntoa_r(igmp->igmp_group, addrbuf), ifp, ifp->if_xname);
1267 * IGMPv1 report suppression.
1268 * If we are a member of this group, and our membership should be
1269 * reported, stop our group timer and transition to the 'lazy' state.
1272 inm = inm_lookup(ifp, igmp->igmp_group);
1274 struct igmp_ifsoftc *igi;
1278 KASSERT(igi != NULL,
1279 ("%s: no igi for ifp %p", __func__, ifp));
1283 IGMPSTAT_INC(igps_rcv_ourreports);
1286 * If we are in IGMPv3 host mode, do not allow the
1287 * other host's IGMPv1 report to suppress our reports
1288 * unless explicitly configured to do so.
1290 if (igi->igi_version == IGMP_VERSION_3) {
1291 if (V_igmp_legacysupp)
1292 igmp_v3_suppress_group_record(inm);
1298 switch (inm->inm_state) {
1299 case IGMP_NOT_MEMBER:
1300 case IGMP_SILENT_MEMBER:
1302 case IGMP_IDLE_MEMBER:
1303 case IGMP_LAZY_MEMBER:
1304 case IGMP_AWAKENING_MEMBER:
1306 "report suppressed for %s on ifp %p(%s)",
1307 inet_ntoa_r(igmp->igmp_group, addrbuf), ifp,
1309 case IGMP_SLEEPING_MEMBER:
1310 inm->inm_state = IGMP_SLEEPING_MEMBER;
1312 case IGMP_REPORTING_MEMBER:
1314 "report suppressed for %s on ifp %p(%s)",
1315 inet_ntoa_r(igmp->igmp_group, addrbuf), ifp,
1317 if (igi->igi_version == IGMP_VERSION_1)
1318 inm->inm_state = IGMP_LAZY_MEMBER;
1319 else if (igi->igi_version == IGMP_VERSION_2)
1320 inm->inm_state = IGMP_SLEEPING_MEMBER;
1322 case IGMP_G_QUERY_PENDING_MEMBER:
1323 case IGMP_SG_QUERY_PENDING_MEMBER:
1324 case IGMP_LEAVING_MEMBER:
1336 * Process a received IGMPv2 host membership report.
1338 * NOTE: 0.0.0.0 workaround breaks const correctness.
1341 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1342 /*const*/ struct igmp *igmp)
1344 struct rm_priotracker in_ifa_tracker;
1345 struct in_ifaddr *ia;
1346 struct in_multi *inm;
1348 char addrbuf[INET_ADDRSTRLEN];
1352 * Make sure we don't hear our own membership report. Fast
1353 * leave requires knowing that we are the only member of a
1356 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1357 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1358 ifa_free(&ia->ia_ifa);
1362 IGMPSTAT_INC(igps_rcv_reports);
1364 if (ifp->if_flags & IFF_LOOPBACK) {
1366 ifa_free(&ia->ia_ifa);
1370 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1371 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1373 ifa_free(&ia->ia_ifa);
1374 IGMPSTAT_INC(igps_rcv_badreports);
1379 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1380 * Booting clients may use the source address 0.0.0.0. Some
1381 * IGMP daemons may not know how to use IP_RECVIF to determine
1382 * the interface upon which this message was received.
1383 * Replace 0.0.0.0 with the subnet address if told to do so.
1385 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1387 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1390 ifa_free(&ia->ia_ifa);
1392 CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1393 inet_ntoa_r(igmp->igmp_group, addrbuf), ifp, ifp->if_xname);
1396 * IGMPv2 report suppression.
1397 * If we are a member of this group, and our membership should be
1398 * reported, and our group timer is pending or about to be reset,
1399 * stop our group timer by transitioning to the 'lazy' state.
1402 inm = inm_lookup(ifp, igmp->igmp_group);
1404 struct igmp_ifsoftc *igi;
1407 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1409 IGMPSTAT_INC(igps_rcv_ourreports);
1412 * If we are in IGMPv3 host mode, do not allow the
1413 * other host's IGMPv1 report to suppress our reports
1414 * unless explicitly configured to do so.
1416 if (igi->igi_version == IGMP_VERSION_3) {
1417 if (V_igmp_legacysupp)
1418 igmp_v3_suppress_group_record(inm);
1424 switch (inm->inm_state) {
1425 case IGMP_NOT_MEMBER:
1426 case IGMP_SILENT_MEMBER:
1427 case IGMP_SLEEPING_MEMBER:
1429 case IGMP_REPORTING_MEMBER:
1430 case IGMP_IDLE_MEMBER:
1431 case IGMP_AWAKENING_MEMBER:
1433 "report suppressed for %s on ifp %p(%s)",
1434 inet_ntoa_r(igmp->igmp_group, addrbuf), ifp,
1436 case IGMP_LAZY_MEMBER:
1437 inm->inm_state = IGMP_LAZY_MEMBER;
1439 case IGMP_G_QUERY_PENDING_MEMBER:
1440 case IGMP_SG_QUERY_PENDING_MEMBER:
1441 case IGMP_LEAVING_MEMBER:
1453 igmp_input(struct mbuf **mp, int *offp, int proto)
1464 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1467 ifp = m->m_pkthdr.rcvif;
1470 IGMPSTAT_INC(igps_rcv_total);
1472 ip = mtod(m, struct ip *);
1474 igmplen = ntohs(ip->ip_len) - iphlen;
1479 if (igmplen < IGMP_MINLEN) {
1480 IGMPSTAT_INC(igps_rcv_tooshort);
1482 return (IPPROTO_DONE);
1486 * Always pullup to the minimum size for v1/v2 or v3
1487 * to amortize calls to m_pullup().
1490 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1491 minlen += IGMP_V3_QUERY_MINLEN;
1493 minlen += IGMP_MINLEN;
1494 if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1495 (m = m_pullup(m, minlen)) == NULL) {
1496 IGMPSTAT_INC(igps_rcv_tooshort);
1497 return (IPPROTO_DONE);
1499 ip = mtod(m, struct ip *);
1502 * Validate checksum.
1504 m->m_data += iphlen;
1506 igmp = mtod(m, struct igmp *);
1507 if (in_cksum(m, igmplen)) {
1508 IGMPSTAT_INC(igps_rcv_badsum);
1510 return (IPPROTO_DONE);
1512 m->m_data -= iphlen;
1516 * IGMP control traffic is link-scope, and must have a TTL of 1.
1517 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1518 * probe packets may come from beyond the LAN.
1520 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1521 IGMPSTAT_INC(igps_rcv_badttl);
1523 return (IPPROTO_DONE);
1526 switch (igmp->igmp_type) {
1527 case IGMP_HOST_MEMBERSHIP_QUERY:
1528 if (igmplen == IGMP_MINLEN) {
1529 if (igmp->igmp_code == 0)
1530 queryver = IGMP_VERSION_1;
1532 queryver = IGMP_VERSION_2;
1533 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1534 queryver = IGMP_VERSION_3;
1536 IGMPSTAT_INC(igps_rcv_tooshort);
1538 return (IPPROTO_DONE);
1542 case IGMP_VERSION_1:
1543 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1544 if (!V_igmp_v1enable)
1546 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1548 return (IPPROTO_DONE);
1552 case IGMP_VERSION_2:
1553 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1554 if (!V_igmp_v2enable)
1556 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1558 return (IPPROTO_DONE);
1562 case IGMP_VERSION_3: {
1563 struct igmpv3 *igmpv3;
1567 IGMPSTAT_INC(igps_rcv_v3_queries);
1568 igmpv3 = (struct igmpv3 *)igmp;
1570 * Validate length based on source count.
1572 nsrc = ntohs(igmpv3->igmp_numsrc);
1573 if (nsrc * sizeof(in_addr_t) >
1574 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1575 IGMPSTAT_INC(igps_rcv_tooshort);
1576 return (IPPROTO_DONE);
1579 * m_pullup() may modify m, so pullup in
1582 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1583 sizeof(struct in_addr) * nsrc;
1584 if ((!M_WRITABLE(m) ||
1585 m->m_len < igmpv3len) &&
1586 (m = m_pullup(m, igmpv3len)) == NULL) {
1587 IGMPSTAT_INC(igps_rcv_tooshort);
1588 return (IPPROTO_DONE);
1590 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1592 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1594 return (IPPROTO_DONE);
1601 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1602 if (!V_igmp_v1enable)
1604 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1606 return (IPPROTO_DONE);
1610 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1611 if (!V_igmp_v2enable)
1613 if (!ip_checkrouteralert(m))
1614 IGMPSTAT_INC(igps_rcv_nora);
1615 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1617 return (IPPROTO_DONE);
1621 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1623 * Hosts do not need to process IGMPv3 membership reports,
1624 * as report suppression is no longer required.
1626 if (!ip_checkrouteralert(m))
1627 IGMPSTAT_INC(igps_rcv_nora);
1635 * Pass all valid IGMP packets up to any process(es) listening on a
1639 return (rip_input(mp, offp, proto));
1644 * Fast timeout handler (global).
1645 * VIMAGE: Timeout handlers are expected to service all vimages.
1650 VNET_ITERATOR_DECL(vnet_iter);
1652 VNET_LIST_RLOCK_NOSLEEP();
1653 VNET_FOREACH(vnet_iter) {
1654 CURVNET_SET(vnet_iter);
1655 igmp_fasttimo_vnet();
1658 VNET_LIST_RUNLOCK_NOSLEEP();
1662 * Fast timeout handler (per-vnet).
1663 * Sends are shuffled off to a netisr to deal with Giant.
1665 * VIMAGE: Assume caller has set up our curvnet.
1668 igmp_fasttimo_vnet(void)
1670 struct mbufq scq; /* State-change packets */
1671 struct mbufq qrq; /* Query response packets */
1673 struct igmp_ifsoftc *igi;
1674 struct ifmultiaddr *ifma;
1675 struct in_multi *inm;
1676 int loop, uri_fasthz;
1682 * Quick check to see if any work needs to be done, in order to
1683 * minimize the overhead of fasttimo processing.
1684 * SMPng: XXX Unlocked reads.
1686 if (!V_current_state_timers_running &&
1687 !V_interface_timers_running &&
1688 !V_state_change_timers_running)
1695 * IGMPv3 General Query response timer processing.
1697 if (V_interface_timers_running) {
1698 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1700 V_interface_timers_running = 0;
1701 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1702 if (igi->igi_v3_timer == 0) {
1704 } else if (--igi->igi_v3_timer == 0) {
1705 igmp_v3_dispatch_general_query(igi);
1707 V_interface_timers_running = 1;
1712 if (!V_current_state_timers_running &&
1713 !V_state_change_timers_running)
1716 V_current_state_timers_running = 0;
1717 V_state_change_timers_running = 0;
1719 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1722 * IGMPv1/v2/v3 host report and state-change timer processing.
1723 * Note: Processing a v3 group timer may remove a node.
1725 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1728 if (igi->igi_version == IGMP_VERSION_3) {
1729 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1730 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1732 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1733 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1737 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1738 if (ifma->ifma_addr->sa_family != AF_INET ||
1739 ifma->ifma_protospec == NULL)
1741 inm = (struct in_multi *)ifma->ifma_protospec;
1742 switch (igi->igi_version) {
1743 case IGMP_VERSION_1:
1744 case IGMP_VERSION_2:
1745 igmp_v1v2_process_group_timer(inm,
1748 case IGMP_VERSION_3:
1749 igmp_v3_process_group_timers(igi, &qrq,
1750 &scq, inm, uri_fasthz);
1754 IF_ADDR_RUNLOCK(ifp);
1756 if (igi->igi_version == IGMP_VERSION_3) {
1757 struct in_multi *tinm;
1759 igmp_dispatch_queue(&qrq, 0, loop);
1760 igmp_dispatch_queue(&scq, 0, loop);
1763 * Free the in_multi reference(s) for this
1766 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1768 SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1770 inm_release_locked(inm);
1781 * Update host report group timer for IGMPv1/v2.
1782 * Will update the global pending timer flags.
1785 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1787 int report_timer_expired;
1789 IN_MULTI_LOCK_ASSERT();
1792 if (inm->inm_timer == 0) {
1793 report_timer_expired = 0;
1794 } else if (--inm->inm_timer == 0) {
1795 report_timer_expired = 1;
1797 V_current_state_timers_running = 1;
1801 switch (inm->inm_state) {
1802 case IGMP_NOT_MEMBER:
1803 case IGMP_SILENT_MEMBER:
1804 case IGMP_IDLE_MEMBER:
1805 case IGMP_LAZY_MEMBER:
1806 case IGMP_SLEEPING_MEMBER:
1807 case IGMP_AWAKENING_MEMBER:
1809 case IGMP_REPORTING_MEMBER:
1810 if (report_timer_expired) {
1811 inm->inm_state = IGMP_IDLE_MEMBER;
1812 (void)igmp_v1v2_queue_report(inm,
1813 (version == IGMP_VERSION_2) ?
1814 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1815 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1818 case IGMP_G_QUERY_PENDING_MEMBER:
1819 case IGMP_SG_QUERY_PENDING_MEMBER:
1820 case IGMP_LEAVING_MEMBER:
1826 * Update a group's timers for IGMPv3.
1827 * Will update the global pending timer flags.
1828 * Note: Unlocked read from igi.
1831 igmp_v3_process_group_timers(struct igmp_ifsoftc *igi,
1832 struct mbufq *qrq, struct mbufq *scq,
1833 struct in_multi *inm, const int uri_fasthz)
1835 int query_response_timer_expired;
1836 int state_change_retransmit_timer_expired;
1838 char addrbuf[INET_ADDRSTRLEN];
1841 IN_MULTI_LOCK_ASSERT();
1844 query_response_timer_expired = 0;
1845 state_change_retransmit_timer_expired = 0;
1848 * During a transition from v1/v2 compatibility mode back to v3,
1849 * a group record in REPORTING state may still have its group
1850 * timer active. This is a no-op in this function; it is easier
1851 * to deal with it here than to complicate the slow-timeout path.
1853 if (inm->inm_timer == 0) {
1854 query_response_timer_expired = 0;
1855 } else if (--inm->inm_timer == 0) {
1856 query_response_timer_expired = 1;
1858 V_current_state_timers_running = 1;
1861 if (inm->inm_sctimer == 0) {
1862 state_change_retransmit_timer_expired = 0;
1863 } else if (--inm->inm_sctimer == 0) {
1864 state_change_retransmit_timer_expired = 1;
1866 V_state_change_timers_running = 1;
1869 /* We are in fasttimo, so be quick about it. */
1870 if (!state_change_retransmit_timer_expired &&
1871 !query_response_timer_expired)
1874 switch (inm->inm_state) {
1875 case IGMP_NOT_MEMBER:
1876 case IGMP_SILENT_MEMBER:
1877 case IGMP_SLEEPING_MEMBER:
1878 case IGMP_LAZY_MEMBER:
1879 case IGMP_AWAKENING_MEMBER:
1880 case IGMP_IDLE_MEMBER:
1882 case IGMP_G_QUERY_PENDING_MEMBER:
1883 case IGMP_SG_QUERY_PENDING_MEMBER:
1885 * Respond to a previously pending Group-Specific
1886 * or Group-and-Source-Specific query by enqueueing
1887 * the appropriate Current-State report for
1888 * immediate transmission.
1890 if (query_response_timer_expired) {
1893 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1894 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1895 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1897 inm->inm_state = IGMP_REPORTING_MEMBER;
1898 /* XXX Clear recorded sources for next time. */
1899 inm_clear_recorded(inm);
1902 case IGMP_REPORTING_MEMBER:
1903 case IGMP_LEAVING_MEMBER:
1904 if (state_change_retransmit_timer_expired) {
1906 * State-change retransmission timer fired.
1907 * If there are any further pending retransmissions,
1908 * set the global pending state-change flag, and
1911 if (--inm->inm_scrv > 0) {
1912 inm->inm_sctimer = uri_fasthz;
1913 V_state_change_timers_running = 1;
1916 * Retransmit the previously computed state-change
1917 * report. If there are no further pending
1918 * retransmissions, the mbuf queue will be consumed.
1919 * Update T0 state to T1 as we have now sent
1922 (void)igmp_v3_merge_state_changes(inm, scq);
1925 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1926 inet_ntoa_r(inm->inm_addr, addrbuf),
1927 inm->inm_ifp->if_xname);
1930 * If we are leaving the group for good, make sure
1931 * we release IGMP's reference to it.
1932 * This release must be deferred using a SLIST,
1933 * as we are called from a loop which traverses
1934 * the in_ifmultiaddr TAILQ.
1936 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1937 inm->inm_scrv == 0) {
1938 inm->inm_state = IGMP_NOT_MEMBER;
1939 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1949 * Suppress a group's pending response to a group or source/group query.
1951 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1952 * Do NOT update ST1/ST0 as this operation merely suppresses
1953 * the currently pending group record.
1954 * Do NOT suppress the response to a general query. It is possible but
1955 * it would require adding another state or flag.
1958 igmp_v3_suppress_group_record(struct in_multi *inm)
1961 IN_MULTI_LOCK_ASSERT();
1963 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1964 ("%s: not IGMPv3 mode on link", __func__));
1966 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1967 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1970 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1971 inm_clear_recorded(inm);
1974 inm->inm_state = IGMP_REPORTING_MEMBER;
1978 * Switch to a different IGMP version on the given interface,
1979 * as per Section 7.2.1.
1982 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1984 int old_version_timer;
1988 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1989 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1991 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1993 * Compute the "Older Version Querier Present" timer as per
1996 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1997 old_version_timer *= PR_SLOWHZ;
1999 if (version == IGMP_VERSION_1) {
2000 igi->igi_v1_timer = old_version_timer;
2001 igi->igi_v2_timer = 0;
2002 } else if (version == IGMP_VERSION_2) {
2003 igi->igi_v1_timer = 0;
2004 igi->igi_v2_timer = old_version_timer;
2008 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2009 if (igi->igi_version != IGMP_VERSION_2) {
2010 igi->igi_version = IGMP_VERSION_2;
2011 igmp_v3_cancel_link_timers(igi);
2013 } else if (igi->igi_v1_timer > 0) {
2014 if (igi->igi_version != IGMP_VERSION_1) {
2015 igi->igi_version = IGMP_VERSION_1;
2016 igmp_v3_cancel_link_timers(igi);
2022 * Cancel pending IGMPv3 timers for the given link and all groups
2023 * joined on it; state-change, general-query, and group-query timers.
2025 * Only ever called on a transition from v3 to Compatibility mode. Kill
2026 * the timers stone dead (this may be expensive for large N groups), they
2027 * will be restarted if Compatibility Mode deems that they must be due to
2031 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2033 struct ifmultiaddr *ifma;
2035 struct in_multi *inm, *tinm;
2037 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2038 igi->igi_ifp, igi->igi_ifp->if_xname);
2040 IN_MULTI_LOCK_ASSERT();
2044 * Stop the v3 General Query Response on this link stone dead.
2045 * If fasttimo is woken up due to V_interface_timers_running,
2046 * the flag will be cleared if there are no pending link timers.
2048 igi->igi_v3_timer = 0;
2051 * Now clear the current-state and state-change report timers
2052 * for all memberships scoped to this link.
2056 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2057 if (ifma->ifma_addr->sa_family != AF_INET ||
2058 ifma->ifma_protospec == NULL)
2060 inm = (struct in_multi *)ifma->ifma_protospec;
2061 switch (inm->inm_state) {
2062 case IGMP_NOT_MEMBER:
2063 case IGMP_SILENT_MEMBER:
2064 case IGMP_IDLE_MEMBER:
2065 case IGMP_LAZY_MEMBER:
2066 case IGMP_SLEEPING_MEMBER:
2067 case IGMP_AWAKENING_MEMBER:
2069 * These states are either not relevant in v3 mode,
2070 * or are unreported. Do nothing.
2073 case IGMP_LEAVING_MEMBER:
2075 * If we are leaving the group and switching to
2076 * compatibility mode, we need to release the final
2077 * reference held for issuing the INCLUDE {}, and
2078 * transition to REPORTING to ensure the host leave
2079 * message is sent upstream to the old querier --
2080 * transition to NOT would lose the leave and race.
2082 SLIST_INSERT_HEAD(&igi->igi_relinmhead, inm, inm_nrele);
2084 case IGMP_G_QUERY_PENDING_MEMBER:
2085 case IGMP_SG_QUERY_PENDING_MEMBER:
2086 inm_clear_recorded(inm);
2088 case IGMP_REPORTING_MEMBER:
2089 inm->inm_state = IGMP_REPORTING_MEMBER;
2093 * Always clear state-change and group report timers.
2094 * Free any pending IGMPv3 state-change records.
2096 inm->inm_sctimer = 0;
2098 mbufq_drain(&inm->inm_scq);
2100 IF_ADDR_RUNLOCK(ifp);
2101 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, tinm) {
2102 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
2103 inm_release_locked(inm);
2108 * Update the Older Version Querier Present timers for a link.
2109 * See Section 7.2.1 of RFC 3376.
2112 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2117 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2119 * IGMPv1 and IGMPv2 Querier Present timers expired.
2123 if (igi->igi_version != IGMP_VERSION_3) {
2125 "%s: transition from v%d -> v%d on %p(%s)",
2126 __func__, igi->igi_version, IGMP_VERSION_3,
2127 igi->igi_ifp, igi->igi_ifp->if_xname);
2128 igi->igi_version = IGMP_VERSION_3;
2130 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2132 * IGMPv1 Querier Present timer expired,
2133 * IGMPv2 Querier Present timer running.
2134 * If IGMPv2 was disabled since last timeout,
2136 * If IGMPv2 is enabled, revert to IGMPv2.
2138 if (!V_igmp_v2enable) {
2140 "%s: transition from v%d -> v%d on %p(%s)",
2141 __func__, igi->igi_version, IGMP_VERSION_3,
2142 igi->igi_ifp, igi->igi_ifp->if_xname);
2143 igi->igi_v2_timer = 0;
2144 igi->igi_version = IGMP_VERSION_3;
2146 --igi->igi_v2_timer;
2147 if (igi->igi_version != IGMP_VERSION_2) {
2149 "%s: transition from v%d -> v%d on %p(%s)",
2150 __func__, igi->igi_version, IGMP_VERSION_2,
2151 igi->igi_ifp, igi->igi_ifp->if_xname);
2152 igi->igi_version = IGMP_VERSION_2;
2153 igmp_v3_cancel_link_timers(igi);
2156 } else if (igi->igi_v1_timer > 0) {
2158 * IGMPv1 Querier Present timer running.
2159 * Stop IGMPv2 timer if running.
2161 * If IGMPv1 was disabled since last timeout,
2163 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2165 if (!V_igmp_v1enable) {
2167 "%s: transition from v%d -> v%d on %p(%s)",
2168 __func__, igi->igi_version, IGMP_VERSION_3,
2169 igi->igi_ifp, igi->igi_ifp->if_xname);
2170 igi->igi_v1_timer = 0;
2171 igi->igi_version = IGMP_VERSION_3;
2173 --igi->igi_v1_timer;
2175 if (igi->igi_v2_timer > 0) {
2177 "%s: cancel v2 timer on %p(%s)",
2178 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2179 igi->igi_v2_timer = 0;
2185 * Global slowtimo handler.
2186 * VIMAGE: Timeout handlers are expected to service all vimages.
2191 VNET_ITERATOR_DECL(vnet_iter);
2193 VNET_LIST_RLOCK_NOSLEEP();
2194 VNET_FOREACH(vnet_iter) {
2195 CURVNET_SET(vnet_iter);
2196 igmp_slowtimo_vnet();
2199 VNET_LIST_RUNLOCK_NOSLEEP();
2203 * Per-vnet slowtimo handler.
2206 igmp_slowtimo_vnet(void)
2208 struct igmp_ifsoftc *igi;
2212 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2213 igmp_v1v2_process_querier_timers(igi);
2220 * Dispatch an IGMPv1/v2 host report or leave message.
2221 * These are always small enough to fit inside a single mbuf.
2224 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2231 IN_MULTI_LOCK_ASSERT();
2236 m = m_gethdr(M_NOWAIT, MT_DATA);
2239 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2241 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2243 m->m_data += sizeof(struct ip);
2244 m->m_len = sizeof(struct igmp);
2246 igmp = mtod(m, struct igmp *);
2247 igmp->igmp_type = type;
2248 igmp->igmp_code = 0;
2249 igmp->igmp_group = inm->inm_addr;
2250 igmp->igmp_cksum = 0;
2251 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2253 m->m_data -= sizeof(struct ip);
2254 m->m_len += sizeof(struct ip);
2256 ip = mtod(m, struct ip *);
2258 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2260 ip->ip_p = IPPROTO_IGMP;
2261 ip->ip_src.s_addr = INADDR_ANY;
2263 if (type == IGMP_HOST_LEAVE_MESSAGE)
2264 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2266 ip->ip_dst = inm->inm_addr;
2268 igmp_save_context(m, ifp);
2270 m->m_flags |= M_IGMPV2;
2271 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2272 m->m_flags |= M_IGMP_LOOP;
2274 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2275 netisr_dispatch(NETISR_IGMP, m);
2281 * Process a state change from the upper layer for the given IPv4 group.
2283 * Each socket holds a reference on the in_multi in its own ip_moptions.
2284 * The socket layer will have made the necessary updates to.the group
2285 * state, it is now up to IGMP to issue a state change report if there
2286 * has been any change between T0 (when the last state-change was issued)
2289 * We use the IGMPv3 state machine at group level. The IGMP module
2290 * however makes the decision as to which IGMP protocol version to speak.
2291 * A state change *from* INCLUDE {} always means an initial join.
2292 * A state change *to* INCLUDE {} always means a final leave.
2294 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2295 * save ourselves a bunch of work; any exclusive mode groups need not
2296 * compute source filter lists.
2298 * VIMAGE: curvnet should have been set by caller, as this routine
2299 * is called from the socket option handlers.
2302 igmp_change_state(struct in_multi *inm)
2304 struct igmp_ifsoftc *igi;
2308 IN_MULTI_LOCK_ASSERT();
2313 * Try to detect if the upper layer just asked us to change state
2314 * for an interface which has now gone away.
2316 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2317 ifp = inm->inm_ifma->ifma_ifp;
2319 * Sanity check that netinet's notion of ifp is the
2322 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2326 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2327 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2330 * If we detect a state transition to or from MCAST_UNDEFINED
2331 * for this group, then we are starting or finishing an IGMP
2332 * life cycle for this group.
2334 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2335 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2336 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2337 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2338 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2339 error = igmp_initial_join(inm, igi);
2341 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2342 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2343 igmp_final_leave(inm, igi);
2347 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2350 error = igmp_handle_state_change(inm, igi);
2358 * Perform the initial join for an IGMP group.
2360 * When joining a group:
2361 * If the group should have its IGMP traffic suppressed, do nothing.
2362 * IGMPv1 starts sending IGMPv1 host membership reports.
2363 * IGMPv2 starts sending IGMPv2 host membership reports.
2364 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2365 * initial state of the membership.
2368 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2372 int error, retval, syncstates;
2374 char addrbuf[INET_ADDRSTRLEN];
2377 CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2378 __func__, inet_ntoa_r(inm->inm_addr, addrbuf), inm->inm_ifp,
2379 inm->inm_ifp->if_xname);
2386 IN_MULTI_LOCK_ASSERT();
2389 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2392 * Groups joined on loopback or marked as 'not reported',
2393 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2394 * are never reported in any IGMP protocol exchanges.
2395 * All other groups enter the appropriate IGMP state machine
2396 * for the version in use on this link.
2397 * A link marked as IGIF_SILENT causes IGMP to be completely
2398 * disabled for the link.
2400 if ((ifp->if_flags & IFF_LOOPBACK) ||
2401 (igi->igi_flags & IGIF_SILENT) ||
2402 !igmp_isgroupreported(inm->inm_addr)) {
2404 "%s: not kicking state machine for silent group", __func__);
2405 inm->inm_state = IGMP_SILENT_MEMBER;
2409 * Deal with overlapping in_multi lifecycle.
2410 * If this group was LEAVING, then make sure
2411 * we drop the reference we picked up to keep the
2412 * group around for the final INCLUDE {} enqueue.
2414 if (igi->igi_version == IGMP_VERSION_3 &&
2415 inm->inm_state == IGMP_LEAVING_MEMBER)
2416 inm_release_locked(inm);
2418 inm->inm_state = IGMP_REPORTING_MEMBER;
2420 switch (igi->igi_version) {
2421 case IGMP_VERSION_1:
2422 case IGMP_VERSION_2:
2423 inm->inm_state = IGMP_IDLE_MEMBER;
2424 error = igmp_v1v2_queue_report(inm,
2425 (igi->igi_version == IGMP_VERSION_2) ?
2426 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2427 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2429 inm->inm_timer = IGMP_RANDOM_DELAY(
2430 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2431 V_current_state_timers_running = 1;
2435 case IGMP_VERSION_3:
2437 * Defer update of T0 to T1, until the first copy
2438 * of the state change has been transmitted.
2443 * Immediately enqueue a State-Change Report for
2444 * this interface, freeing any previous reports.
2445 * Don't kick the timers if there is nothing to do,
2446 * or if an error occurred.
2450 retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2452 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2455 error = retval * -1;
2460 * Schedule transmission of pending state-change
2461 * report up to RV times for this link. The timer
2462 * will fire at the next igmp_fasttimo (~200ms),
2463 * giving us an opportunity to merge the reports.
2465 if (igi->igi_flags & IGIF_LOOPBACK) {
2468 KASSERT(igi->igi_rv > 1,
2469 ("%s: invalid robustness %d", __func__,
2471 inm->inm_scrv = igi->igi_rv;
2473 inm->inm_sctimer = 1;
2474 V_state_change_timers_running = 1;
2482 * Only update the T0 state if state change is atomic,
2483 * i.e. we don't need to wait for a timer to fire before we
2484 * can consider the state change to have been communicated.
2488 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2489 inet_ntoa_r(inm->inm_addr, addrbuf),
2490 inm->inm_ifp->if_xname);
2497 * Issue an intermediate state change during the IGMP life-cycle.
2500 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2505 char addrbuf[INET_ADDRSTRLEN];
2508 CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2509 __func__, inet_ntoa_r(inm->inm_addr, addrbuf), inm->inm_ifp,
2510 inm->inm_ifp->if_xname);
2514 IN_MULTI_LOCK_ASSERT();
2517 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2519 if ((ifp->if_flags & IFF_LOOPBACK) ||
2520 (igi->igi_flags & IGIF_SILENT) ||
2521 !igmp_isgroupreported(inm->inm_addr) ||
2522 (igi->igi_version != IGMP_VERSION_3)) {
2523 if (!igmp_isgroupreported(inm->inm_addr)) {
2525 "%s: not kicking state machine for silent group", __func__);
2527 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2529 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2530 inet_ntoa_r(inm->inm_addr, addrbuf),
2531 inm->inm_ifp->if_xname);
2535 mbufq_drain(&inm->inm_scq);
2537 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2538 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2543 * If record(s) were enqueued, start the state-change
2544 * report timer for this group.
2546 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2547 inm->inm_sctimer = 1;
2548 V_state_change_timers_running = 1;
2554 * Perform the final leave for an IGMP group.
2556 * When leaving a group:
2557 * IGMPv1 does nothing.
2558 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2559 * IGMPv3 enqueues a state-change report containing a transition
2560 * to INCLUDE {} for immediate transmission.
2563 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2567 char addrbuf[INET_ADDRSTRLEN];
2572 CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2573 __func__, inet_ntoa_r(inm->inm_addr, addrbuf), inm->inm_ifp,
2574 inm->inm_ifp->if_xname);
2576 IN_MULTI_LOCK_ASSERT();
2579 switch (inm->inm_state) {
2580 case IGMP_NOT_MEMBER:
2581 case IGMP_SILENT_MEMBER:
2582 case IGMP_LEAVING_MEMBER:
2583 /* Already leaving or left; do nothing. */
2585 "%s: not kicking state machine for silent group", __func__);
2587 case IGMP_REPORTING_MEMBER:
2588 case IGMP_IDLE_MEMBER:
2589 case IGMP_G_QUERY_PENDING_MEMBER:
2590 case IGMP_SG_QUERY_PENDING_MEMBER:
2591 if (igi->igi_version == IGMP_VERSION_2) {
2593 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2594 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2595 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2598 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2599 inm->inm_state = IGMP_NOT_MEMBER;
2600 } else if (igi->igi_version == IGMP_VERSION_3) {
2602 * Stop group timer and all pending reports.
2603 * Immediately enqueue a state-change report
2604 * TO_IN {} to be sent on the next fast timeout,
2605 * giving us an opportunity to merge reports.
2607 mbufq_drain(&inm->inm_scq);
2609 if (igi->igi_flags & IGIF_LOOPBACK) {
2612 inm->inm_scrv = igi->igi_rv;
2614 CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2615 "pending retransmissions.", __func__,
2616 inet_ntoa_r(inm->inm_addr, addrbuf),
2617 inm->inm_ifp->if_xname, inm->inm_scrv);
2618 if (inm->inm_scrv == 0) {
2619 inm->inm_state = IGMP_NOT_MEMBER;
2620 inm->inm_sctimer = 0;
2624 inm_acquire_locked(inm);
2626 retval = igmp_v3_enqueue_group_record(
2627 &inm->inm_scq, inm, 1, 0, 0);
2628 KASSERT(retval != 0,
2629 ("%s: enqueue record = %d", __func__,
2632 inm->inm_state = IGMP_LEAVING_MEMBER;
2633 inm->inm_sctimer = 1;
2634 V_state_change_timers_running = 1;
2640 case IGMP_LAZY_MEMBER:
2641 case IGMP_SLEEPING_MEMBER:
2642 case IGMP_AWAKENING_MEMBER:
2643 /* Our reports are suppressed; do nothing. */
2649 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2650 inet_ntoa_r(inm->inm_addr, addrbuf),
2651 inm->inm_ifp->if_xname);
2652 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2653 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2654 __func__, inet_ntoa_r(inm->inm_addr, addrbuf),
2655 inm->inm_ifp->if_xname);
2660 * Enqueue an IGMPv3 group record to the given output queue.
2662 * XXX This function could do with having the allocation code
2663 * split out, and the multiple-tree-walks coalesced into a single
2664 * routine as has been done in igmp_v3_enqueue_filter_change().
2666 * If is_state_change is zero, a current-state record is appended.
2667 * If is_state_change is non-zero, a state-change report is appended.
2669 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2670 * If is_group_query is zero, and if there is a packet with free space
2671 * at the tail of the queue, it will be appended to providing there
2672 * is enough free space.
2673 * Otherwise a new mbuf packet chain is allocated.
2675 * If is_source_query is non-zero, each source is checked to see if
2676 * it was recorded for a Group-Source query, and will be omitted if
2677 * it is not both in-mode and recorded.
2679 * The function will attempt to allocate leading space in the packet
2680 * for the IP/IGMP header to be prepended without fragmenting the chain.
2682 * If successful the size of all data appended to the queue is returned,
2683 * otherwise an error code less than zero is returned, or zero if
2684 * no record(s) were appended.
2687 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2688 const int is_state_change, const int is_group_query,
2689 const int is_source_query)
2691 struct igmp_grouprec ig;
2692 struct igmp_grouprec *pig;
2694 struct ip_msource *ims, *nims;
2695 struct mbuf *m0, *m, *md;
2696 int error, is_filter_list_change;
2697 int minrec0len, m0srcs, msrcs, nbytes, off;
2698 int record_has_sources;
2704 char addrbuf[INET_ADDRSTRLEN];
2707 IN_MULTI_LOCK_ASSERT();
2711 is_filter_list_change = 0;
2718 record_has_sources = 1;
2720 type = IGMP_DO_NOTHING;
2721 mode = inm->inm_st[1].iss_fmode;
2724 * If we did not transition out of ASM mode during t0->t1,
2725 * and there are no source nodes to process, we can skip
2726 * the generation of source records.
2728 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2730 record_has_sources = 0;
2732 if (is_state_change) {
2734 * Queue a state change record.
2735 * If the mode did not change, and there are non-ASM
2736 * listeners or source filters present,
2737 * we potentially need to issue two records for the group.
2738 * If we are transitioning to MCAST_UNDEFINED, we need
2739 * not send any sources.
2740 * If there are ASM listeners, and there was no filter
2741 * mode transition of any kind, do nothing.
2743 if (mode != inm->inm_st[0].iss_fmode) {
2744 if (mode == MCAST_EXCLUDE) {
2745 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2747 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2749 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2751 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2752 if (mode == MCAST_UNDEFINED)
2753 record_has_sources = 0;
2756 if (record_has_sources) {
2757 is_filter_list_change = 1;
2759 type = IGMP_DO_NOTHING;
2764 * Queue a current state record.
2766 if (mode == MCAST_EXCLUDE) {
2767 type = IGMP_MODE_IS_EXCLUDE;
2768 } else if (mode == MCAST_INCLUDE) {
2769 type = IGMP_MODE_IS_INCLUDE;
2770 KASSERT(inm->inm_st[1].iss_asm == 0,
2771 ("%s: inm %p is INCLUDE but ASM count is %d",
2772 __func__, inm, inm->inm_st[1].iss_asm));
2777 * Generate the filter list changes using a separate function.
2779 if (is_filter_list_change)
2780 return (igmp_v3_enqueue_filter_change(mq, inm));
2782 if (type == IGMP_DO_NOTHING) {
2783 CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2784 __func__, inet_ntoa_r(inm->inm_addr, addrbuf),
2785 inm->inm_ifp->if_xname);
2790 * If any sources are present, we must be able to fit at least
2791 * one in the trailing space of the tail packet's mbuf,
2794 minrec0len = sizeof(struct igmp_grouprec);
2795 if (record_has_sources)
2796 minrec0len += sizeof(in_addr_t);
2798 CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2799 igmp_rec_type_to_str(type), inet_ntoa_r(inm->inm_addr, addrbuf),
2800 inm->inm_ifp->if_xname);
2803 * Check if we have a packet in the tail of the queue for this
2804 * group into which the first group record for this group will fit.
2805 * Otherwise allocate a new packet.
2806 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2807 * Note: Group records for G/GSR query responses MUST be sent
2808 * in their own packet.
2810 m0 = mbufq_last(mq);
2811 if (!is_group_query &&
2813 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2814 (m0->m_pkthdr.len + minrec0len) <
2815 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2816 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2817 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2819 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2821 if (mbufq_full(mq)) {
2822 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2826 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2827 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2828 if (!is_state_change && !is_group_query) {
2829 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2831 m->m_data += IGMP_LEADINGSPACE;
2834 m = m_gethdr(M_NOWAIT, MT_DATA);
2836 M_ALIGN(m, IGMP_LEADINGSPACE);
2841 igmp_save_context(m, ifp);
2843 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2847 * Append group record.
2848 * If we have sources, we don't know how many yet.
2853 ig.ig_group = inm->inm_addr;
2854 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2857 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2860 nbytes += sizeof(struct igmp_grouprec);
2863 * Append as many sources as will fit in the first packet.
2864 * If we are appending to a new packet, the chain allocation
2865 * may potentially use clusters; use m_getptr() in this case.
2866 * If we are appending to an existing packet, we need to obtain
2867 * a pointer to the group record after m_append(), in case a new
2868 * mbuf was allocated.
2869 * Only append sources which are in-mode at t1. If we are
2870 * transitioning to MCAST_UNDEFINED state on the group, do not
2871 * include source entries.
2872 * Only report recorded sources in our filter set when responding
2873 * to a group-source query.
2875 if (record_has_sources) {
2878 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2879 md->m_len - nbytes);
2881 md = m_getptr(m, 0, &off);
2882 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2886 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2887 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2888 inet_ntoa_haddr(ims->ims_haddr, addrbuf));
2889 now = ims_get_mode(inm, ims, 1);
2890 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2891 if ((now != mode) ||
2892 (now == mode && mode == MCAST_UNDEFINED)) {
2893 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2896 if (is_source_query && ims->ims_stp == 0) {
2897 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2901 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2902 naddr = htonl(ims->ims_haddr);
2903 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2906 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2910 nbytes += sizeof(in_addr_t);
2912 if (msrcs == m0srcs)
2915 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2917 pig->ig_numsrc = htons(msrcs);
2918 nbytes += (msrcs * sizeof(in_addr_t));
2921 if (is_source_query && msrcs == 0) {
2922 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2929 * We are good to go with first packet.
2932 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2933 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2934 mbufq_enqueue(mq, m);
2936 m->m_pkthdr.PH_vt.vt_nrecs++;
2939 * No further work needed if no source list in packet(s).
2941 if (!record_has_sources)
2945 * Whilst sources remain to be announced, we need to allocate
2946 * a new packet and fill out as many sources as will fit.
2947 * Always try for a cluster first.
2949 while (nims != NULL) {
2950 if (mbufq_full(mq)) {
2951 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2954 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2956 m->m_data += IGMP_LEADINGSPACE;
2958 m = m_gethdr(M_NOWAIT, MT_DATA);
2960 M_ALIGN(m, IGMP_LEADINGSPACE);
2964 igmp_save_context(m, ifp);
2965 md = m_getptr(m, 0, &off);
2966 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2967 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2969 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2972 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2975 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2976 nbytes += sizeof(struct igmp_grouprec);
2978 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2979 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2982 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2983 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2984 inet_ntoa_haddr(ims->ims_haddr, addrbuf));
2985 now = ims_get_mode(inm, ims, 1);
2986 if ((now != mode) ||
2987 (now == mode && mode == MCAST_UNDEFINED)) {
2988 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2991 if (is_source_query && ims->ims_stp == 0) {
2992 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2996 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2997 naddr = htonl(ims->ims_haddr);
2998 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
3001 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
3006 if (msrcs == m0srcs)
3009 pig->ig_numsrc = htons(msrcs);
3010 nbytes += (msrcs * sizeof(in_addr_t));
3012 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
3013 mbufq_enqueue(mq, m);
3020 * Type used to mark record pass completion.
3021 * We exploit the fact we can cast to this easily from the
3022 * current filter modes on each ip_msource node.
3025 REC_NONE = 0x00, /* MCAST_UNDEFINED */
3026 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
3027 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
3028 REC_FULL = REC_ALLOW | REC_BLOCK
3032 * Enqueue an IGMPv3 filter list change to the given output queue.
3034 * Source list filter state is held in an RB-tree. When the filter list
3035 * for a group is changed without changing its mode, we need to compute
3036 * the deltas between T0 and T1 for each source in the filter set,
3037 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3039 * As we may potentially queue two record types, and the entire R-B tree
3040 * needs to be walked at once, we break this out into its own function
3041 * so we can generate a tightly packed queue of packets.
3043 * XXX This could be written to only use one tree walk, although that makes
3044 * serializing into the mbuf chains a bit harder. For now we do two walks
3045 * which makes things easier on us, and it may or may not be harder on
3048 * If successful the size of all data appended to the queue is returned,
3049 * otherwise an error code less than zero is returned, or zero if
3050 * no record(s) were appended.
3053 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3055 static const int MINRECLEN =
3056 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3058 struct igmp_grouprec ig;
3059 struct igmp_grouprec *pig;
3060 struct ip_msource *ims, *nims;
3061 struct mbuf *m, *m0, *md;
3063 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3065 uint8_t mode, now, then;
3066 rectype_t crt, drt, nrt;
3068 char addrbuf[INET_ADDRSTRLEN];
3071 IN_MULTI_LOCK_ASSERT();
3073 if (inm->inm_nsrc == 0 ||
3074 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3077 ifp = inm->inm_ifp; /* interface */
3078 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3079 crt = REC_NONE; /* current group record type */
3080 drt = REC_NONE; /* mask of completed group record types */
3081 nrt = REC_NONE; /* record type for current node */
3082 m0srcs = 0; /* # source which will fit in current mbuf chain */
3083 nbytes = 0; /* # of bytes appended to group's state-change queue */
3084 npbytes = 0; /* # of bytes appended this packet */
3085 rsrcs = 0; /* # sources encoded in current record */
3086 schanged = 0; /* # nodes encoded in overall filter change */
3087 nallow = 0; /* # of source entries in ALLOW_NEW */
3088 nblock = 0; /* # of source entries in BLOCK_OLD */
3089 nims = NULL; /* next tree node pointer */
3092 * For each possible filter record mode.
3093 * The first kind of source we encounter tells us which
3094 * is the first kind of record we start appending.
3095 * If a node transitioned to UNDEFINED at t1, its mode is treated
3096 * as the inverse of the group's filter mode.
3098 while (drt != REC_FULL) {
3100 m0 = mbufq_last(mq);
3102 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3103 IGMP_V3_REPORT_MAXRECS) &&
3104 (m0->m_pkthdr.len + MINRECLEN) <
3105 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3107 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3108 sizeof(struct igmp_grouprec)) /
3111 "%s: use previous packet", __func__);
3113 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3115 m->m_data += IGMP_LEADINGSPACE;
3117 m = m_gethdr(M_NOWAIT, MT_DATA);
3119 M_ALIGN(m, IGMP_LEADINGSPACE);
3123 "%s: m_get*() failed", __func__);
3126 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3127 igmp_save_context(m, ifp);
3128 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3129 sizeof(struct igmp_grouprec)) /
3133 "%s: allocated new packet", __func__);
3136 * Append the IGMP group record header to the
3137 * current packet's data area.
3138 * Recalculate pointer to free space for next
3139 * group record, in case m_append() allocated
3140 * a new mbuf or cluster.
3142 memset(&ig, 0, sizeof(ig));
3143 ig.ig_group = inm->inm_addr;
3144 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3148 "%s: m_append() failed", __func__);
3151 npbytes += sizeof(struct igmp_grouprec);
3153 /* new packet; offset in c hain */
3154 md = m_getptr(m, npbytes -
3155 sizeof(struct igmp_grouprec), &off);
3156 pig = (struct igmp_grouprec *)(mtod(md,
3159 /* current packet; offset from last append */
3161 pig = (struct igmp_grouprec *)(mtod(md,
3162 uint8_t *) + md->m_len -
3163 sizeof(struct igmp_grouprec));
3166 * Begin walking the tree for this record type
3167 * pass, or continue from where we left off
3168 * previously if we had to allocate a new packet.
3169 * Only report deltas in-mode at t1.
3170 * We need not report included sources as allowed
3171 * if we are in inclusive mode on the group,
3172 * however the converse is not true.
3176 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3177 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3178 CTR2(KTR_IGMPV3, "%s: visit node %s",
3180 inet_ntoa_haddr(ims->ims_haddr, addrbuf));
3181 now = ims_get_mode(inm, ims, 1);
3182 then = ims_get_mode(inm, ims, 0);
3183 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3184 __func__, then, now);
3187 "%s: skip unchanged", __func__);
3190 if (mode == MCAST_EXCLUDE &&
3191 now == MCAST_INCLUDE) {
3193 "%s: skip IN src on EX group",
3197 nrt = (rectype_t)now;
3198 if (nrt == REC_NONE)
3199 nrt = (rectype_t)(~mode & REC_FULL);
3200 if (schanged++ == 0) {
3202 } else if (crt != nrt)
3204 naddr = htonl(ims->ims_haddr);
3205 if (!m_append(m, sizeof(in_addr_t),
3210 "%s: m_append() failed", __func__);
3213 nallow += !!(crt == REC_ALLOW);
3214 nblock += !!(crt == REC_BLOCK);
3215 if (++rsrcs == m0srcs)
3219 * If we did not append any tree nodes on this
3220 * pass, back out of allocations.
3223 npbytes -= sizeof(struct igmp_grouprec);
3226 "%s: m_free(m)", __func__);
3230 "%s: m_adj(m, -ig)", __func__);
3231 m_adj(m, -((int)sizeof(
3232 struct igmp_grouprec)));
3236 npbytes += (rsrcs * sizeof(in_addr_t));
3237 if (crt == REC_ALLOW)
3238 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3239 else if (crt == REC_BLOCK)
3240 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3241 pig->ig_numsrc = htons(rsrcs);
3243 * Count the new group record, and enqueue this
3244 * packet if it wasn't already queued.
3246 m->m_pkthdr.PH_vt.vt_nrecs++;
3248 mbufq_enqueue(mq, m);
3250 } while (nims != NULL);
3252 crt = (~crt & REC_FULL);
3255 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3262 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3265 struct mbuf *m; /* pending state-change */
3266 struct mbuf *m0; /* copy of pending state-change */
3267 struct mbuf *mt; /* last state-change in packet */
3268 int docopy, domerge;
3275 IN_MULTI_LOCK_ASSERT();
3279 * If there are further pending retransmissions, make a writable
3280 * copy of each queued state-change message before merging.
3282 if (inm->inm_scrv > 0)
3287 if (mbufq_first(gq) == NULL) {
3288 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3293 m = mbufq_first(gq);
3296 * Only merge the report into the current packet if
3297 * there is sufficient space to do so; an IGMPv3 report
3298 * packet may only contain 65,535 group records.
3299 * Always use a simple mbuf chain concatentation to do this,
3300 * as large state changes for single groups may have
3301 * allocated clusters.
3304 mt = mbufq_last(scq);
3306 recslen = m_length(m, NULL);
3308 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3309 m->m_pkthdr.PH_vt.vt_nrecs <=
3310 IGMP_V3_REPORT_MAXRECS) &&
3311 (mt->m_pkthdr.len + recslen <=
3312 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3316 if (!domerge && mbufq_full(gq)) {
3318 "%s: outbound queue full, skipping whole packet %p",
3328 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3329 m0 = mbufq_dequeue(gq);
3332 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3333 m0 = m_dup(m, M_NOWAIT);
3336 m0->m_nextpkt = NULL;
3341 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3343 mbufq_enqueue(scq, m0);
3345 struct mbuf *mtl; /* last mbuf of packet mt */
3347 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3351 m0->m_flags &= ~M_PKTHDR;
3352 mt->m_pkthdr.len += recslen;
3353 mt->m_pkthdr.PH_vt.vt_nrecs +=
3354 m0->m_pkthdr.PH_vt.vt_nrecs;
3364 * Respond to a pending IGMPv3 General Query.
3367 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3369 struct ifmultiaddr *ifma;
3371 struct in_multi *inm;
3374 IN_MULTI_LOCK_ASSERT();
3377 KASSERT(igi->igi_version == IGMP_VERSION_3,
3378 ("%s: called when version %d", __func__, igi->igi_version));
3381 * Check that there are some packets queued. If so, send them first.
3382 * For large number of groups the reply to general query can take
3383 * many packets, we should finish sending them before starting of
3384 * queuing the new reply.
3386 if (mbufq_len(&igi->igi_gq) != 0)
3392 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3393 if (ifma->ifma_addr->sa_family != AF_INET ||
3394 ifma->ifma_protospec == NULL)
3397 inm = (struct in_multi *)ifma->ifma_protospec;
3398 KASSERT(ifp == inm->inm_ifp,
3399 ("%s: inconsistent ifp", __func__));
3401 switch (inm->inm_state) {
3402 case IGMP_NOT_MEMBER:
3403 case IGMP_SILENT_MEMBER:
3405 case IGMP_REPORTING_MEMBER:
3406 case IGMP_IDLE_MEMBER:
3407 case IGMP_LAZY_MEMBER:
3408 case IGMP_SLEEPING_MEMBER:
3409 case IGMP_AWAKENING_MEMBER:
3410 inm->inm_state = IGMP_REPORTING_MEMBER;
3411 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3413 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3416 case IGMP_G_QUERY_PENDING_MEMBER:
3417 case IGMP_SG_QUERY_PENDING_MEMBER:
3418 case IGMP_LEAVING_MEMBER:
3422 IF_ADDR_RUNLOCK(ifp);
3425 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3426 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3429 * Slew transmission of bursts over 500ms intervals.
3431 if (mbufq_first(&igi->igi_gq) != NULL) {
3432 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3433 IGMP_RESPONSE_BURST_INTERVAL);
3434 V_interface_timers_running = 1;
3439 * Transmit the next pending IGMP message in the output queue.
3441 * We get called from netisr_processqueue(). A mutex private to igmpoq
3442 * will be acquired and released around this routine.
3444 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3445 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3446 * a link and uses a link-scope multicast address.
3449 igmp_intr(struct mbuf *m)
3451 struct ip_moptions imo;
3453 struct mbuf *ipopts, *m0;
3457 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3460 * Set VNET image pointer from enqueued mbuf chain
3461 * before doing anything else. Whilst we use interface
3462 * indexes to guard against interface detach, they are
3463 * unique to each VIMAGE and must be retrieved.
3465 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3466 ifindex = igmp_restore_context(m);
3469 * Check if the ifnet still exists. This limits the scope of
3470 * any race in the absence of a global ifp lock for low cost
3471 * (an array lookup).
3473 ifp = ifnet_byindex(ifindex);
3475 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3476 __func__, m, ifindex);
3478 IPSTAT_INC(ips_noroute);
3482 ipopts = V_igmp_sendra ? m_raopt : NULL;
3484 imo.imo_multicast_ttl = 1;
3485 imo.imo_multicast_vif = -1;
3486 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3489 * If the user requested that IGMP traffic be explicitly
3490 * redirected to the loopback interface (e.g. they are running a
3491 * MANET interface and the routing protocol needs to see the
3492 * updates), handle this now.
3494 if (m->m_flags & M_IGMP_LOOP)
3495 imo.imo_multicast_ifp = V_loif;
3497 imo.imo_multicast_ifp = ifp;
3499 if (m->m_flags & M_IGMPV2) {
3502 m0 = igmp_v3_encap_report(ifp, m);
3504 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3506 IPSTAT_INC(ips_odropped);
3511 igmp_scrub_context(m0);
3513 m0->m_pkthdr.rcvif = V_loif;
3515 mac_netinet_igmp_send(ifp, m0);
3517 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3519 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3523 IGMPSTAT_INC(igps_snd_reports);
3527 * We must restore the existing vnet pointer before
3528 * continuing as we are run from netisr context.
3534 * Encapsulate an IGMPv3 report.
3536 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3537 * chain has already had its IP/IGMPv3 header prepended. In this case
3538 * the function will not attempt to prepend; the lengths and checksums
3539 * will however be re-computed.
3541 * Returns a pointer to the new mbuf chain head, or NULL if the
3542 * allocation failed.
3544 static struct mbuf *
3545 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3547 struct rm_priotracker in_ifa_tracker;
3548 struct igmp_report *igmp;
3550 int hdrlen, igmpreclen;
3552 KASSERT((m->m_flags & M_PKTHDR),
3553 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3555 igmpreclen = m_length(m, NULL);
3556 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3558 if (m->m_flags & M_IGMPV3_HDR) {
3559 igmpreclen -= hdrlen;
3561 M_PREPEND(m, hdrlen, M_NOWAIT);
3564 m->m_flags |= M_IGMPV3_HDR;
3567 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3569 m->m_data += sizeof(struct ip);
3570 m->m_len -= sizeof(struct ip);
3572 igmp = mtod(m, struct igmp_report *);
3573 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3576 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3578 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3579 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3581 m->m_data -= sizeof(struct ip);
3582 m->m_len += sizeof(struct ip);
3584 ip = mtod(m, struct ip *);
3585 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3586 ip->ip_len = htons(hdrlen + igmpreclen);
3587 ip->ip_off = htons(IP_DF);
3588 ip->ip_p = IPPROTO_IGMP;
3591 ip->ip_src.s_addr = INADDR_ANY;
3593 if (m->m_flags & M_IGMP_LOOP) {
3594 struct in_ifaddr *ia;
3596 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3598 ip->ip_src = ia->ia_addr.sin_addr;
3599 ifa_free(&ia->ia_ifa);
3603 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3610 igmp_rec_type_to_str(const int type)
3614 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3617 case IGMP_CHANGE_TO_INCLUDE_MODE:
3620 case IGMP_MODE_IS_EXCLUDE:
3623 case IGMP_MODE_IS_INCLUDE:
3626 case IGMP_ALLOW_NEW_SOURCES:
3629 case IGMP_BLOCK_OLD_SOURCES:
3641 vnet_igmp_init(const void *unused __unused)
3644 netisr_register_vnet(&igmp_nh);
3646 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3647 vnet_igmp_init, NULL);
3650 vnet_igmp_uninit(const void *unused __unused)
3653 /* This can happen when we shutdown the entire network stack. */
3654 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3656 netisr_unregister_vnet(&igmp_nh);
3658 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3659 vnet_igmp_uninit, NULL);
3663 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3665 struct igmp_ifsoftc *igi, *tigi;
3666 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3669 db_printf("usage: show igi_list <addr>\n");
3672 igi_head = (struct _igi_list *)addr;
3674 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3675 db_printf("igmp_ifsoftc %p:\n", igi);
3676 db_printf(" ifp %p\n", igi->igi_ifp);
3677 db_printf(" version %u\n", igi->igi_version);
3678 db_printf(" v1_timer %u\n", igi->igi_v1_timer);
3679 db_printf(" v2_timer %u\n", igi->igi_v2_timer);
3680 db_printf(" v3_timer %u\n", igi->igi_v3_timer);
3681 db_printf(" flags %#x\n", igi->igi_flags);
3682 db_printf(" rv %u\n", igi->igi_rv);
3683 db_printf(" qi %u\n", igi->igi_qi);
3684 db_printf(" qri %u\n", igi->igi_qri);
3685 db_printf(" uri %u\n", igi->igi_uri);
3686 /* SLIST_HEAD(,in_multi) igi_relinmhead */
3687 /* struct mbufq igi_gq; */
3694 igmp_modevent(module_t mod, int type, void *unused __unused)
3699 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3701 m_raopt = igmp_ra_alloc();
3702 netisr_register(&igmp_nh);
3705 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3706 netisr_unregister(&igmp_nh);
3709 IGMP_LOCK_DESTROY();
3712 return (EOPNOTSUPP);
3717 static moduledata_t igmp_mod = {
3722 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);