2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Bruce Simpson.
5 * Copyright (c) 1988 Stephen Deering.
6 * Copyright (c) 1992, 1993
7 * The Regents of the University of California. All rights reserved.
9 * This code is derived from software contributed to Berkeley by
10 * Stephen Deering of Stanford University.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)igmp.c 8.1 (Berkeley) 7/19/93
40 * Internet Group Management Protocol (IGMP) routines.
41 * [RFC1112, RFC2236, RFC3376]
43 * Written by Steve Deering, Stanford, May 1988.
44 * Modified by Rosen Sharma, Stanford, Aug 1994.
45 * Modified by Bill Fenner, Xerox PARC, Feb 1995.
46 * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995.
47 * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson.
49 * MULTICAST Revision: 3.5.1.4
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
57 #include <sys/param.h>
58 #include <sys/systm.h>
59 #include <sys/module.h>
60 #include <sys/malloc.h>
62 #include <sys/socket.h>
63 #include <sys/protosw.h>
64 #include <sys/kernel.h>
66 #include <sys/rmlock.h>
67 #include <sys/sysctl.h>
69 #include <sys/condvar.h>
76 #include <net/if_var.h>
77 #include <net/netisr.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 #include <netinet/in_systm.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_var.h>
85 #include <netinet/ip_options.h>
86 #include <netinet/igmp.h>
87 #include <netinet/igmp_var.h>
89 #include <machine/in_cksum.h>
91 #include <security/mac/mac_framework.h>
94 #define KTR_IGMPV3 KTR_INET
97 static struct igmp_ifsoftc *
98 igi_alloc_locked(struct ifnet *);
99 static void igi_delete_locked(const struct ifnet *);
100 static void igmp_dispatch_queue(struct mbufq *, int, const int);
101 static void igmp_fasttimo_vnet(void);
102 static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
103 static int igmp_handle_state_change(struct in_multi *,
104 struct igmp_ifsoftc *);
105 static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
106 static int igmp_input_v1_query(struct ifnet *, const struct ip *,
107 const struct igmp *);
108 static int igmp_input_v2_query(struct ifnet *, const struct ip *,
109 const struct igmp *);
110 static int igmp_input_v3_query(struct ifnet *, const struct ip *,
111 /*const*/ struct igmpv3 *);
112 static int igmp_input_v3_group_query(struct in_multi *,
113 struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
114 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
115 /*const*/ struct igmp *);
116 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
117 /*const*/ struct igmp *);
118 static void igmp_intr(struct mbuf *);
119 static int igmp_isgroupreported(const struct in_addr);
123 static char * igmp_rec_type_to_str(const int);
125 static void igmp_set_version(struct igmp_ifsoftc *, const int);
126 static void igmp_slowtimo_vnet(void);
127 static int igmp_v1v2_queue_report(struct in_multi *, const int);
128 static void igmp_v1v2_process_group_timer(struct in_multi *, const int);
129 static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
130 static void igmp_v2_update_group(struct in_multi *, const int);
131 static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
132 static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
134 igmp_v3_encap_report(struct ifnet *, struct mbuf *);
135 static int igmp_v3_enqueue_group_record(struct mbufq *,
136 struct in_multi *, const int, const int, const int);
137 static int igmp_v3_enqueue_filter_change(struct mbufq *,
139 static void igmp_v3_process_group_timers(struct in_multi_head *,
140 struct mbufq *, struct mbufq *, struct in_multi *,
142 static int igmp_v3_merge_state_changes(struct in_multi *,
144 static void igmp_v3_suppress_group_record(struct in_multi *);
145 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
146 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
147 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
149 static const struct netisr_handler igmp_nh = {
151 .nh_handler = igmp_intr,
152 .nh_proto = NETISR_IGMP,
153 .nh_policy = NETISR_POLICY_SOURCE,
157 * System-wide globals.
159 * Unlocked access to these is OK, except for the global IGMP output
160 * queue. The IGMP subsystem lock ends up being system-wide for the moment,
161 * because all VIMAGEs have to share a global output queue, as netisrs
162 * themselves are not virtualized.
165 * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
166 * Any may be taken independently; if any are held at the same
167 * time, the above lock order must be followed.
168 * * All output is delegated to the netisr.
169 * Now that Giant has been eliminated, the netisr may be inlined.
170 * * IN_MULTI_LIST_LOCK covers in_multi.
171 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
172 * including the output queue.
173 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
174 * per-link state iterators.
175 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
176 * therefore it is not refcounted.
177 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
180 * * IGMP acquires its own reference every time an in_multi is passed to
181 * it and the group is being joined for the first time.
182 * * IGMP releases its reference(s) on in_multi in a deferred way,
183 * because the operations which process the release run as part of
184 * a loop whose control variables are directly affected by the release
185 * (that, and not recursing on the IF_ADDR_LOCK).
187 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
188 * to a vnet in ifp->if_vnet.
190 * SMPng: XXX We may potentially race operations on ifma_protospec.
191 * The problem is that we currently lack a clean way of taking the
192 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
193 * as anything which modifies ifma needs to be covered by that lock.
194 * So check for ifma_protospec being NULL before proceeding.
198 struct mbuf *m_raopt; /* Router Alert option */
199 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
202 * VIMAGE-wide globals.
204 * The IGMPv3 timers themselves need to run per-image, however,
205 * protosw timers run globally (see tcp).
206 * An ifnet can only be in one vimage at a time, and the loopback
207 * ifnet, loif, is itself virtualized.
208 * It would otherwise be possible to seriously hose IGMP state,
209 * and create inconsistencies in upstream multicast routing, if you have
210 * multiple VIMAGEs running on the same link joining different multicast
211 * groups, UNLESS the "primary IP address" is different. This is because
212 * IGMP for IPv4 does not force link-local addresses to be used for each
213 * node, unlike MLD for IPv6.
214 * Obviously the IGMPv3 per-interface state has per-vimage granularity
217 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
218 * policy to control the address used by IGMP on the link.
220 VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general
222 VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change
224 VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host
225 * report; IGMPv3 g/sg
228 #define V_interface_timers_running VNET(interface_timers_running)
229 #define V_state_change_timers_running VNET(state_change_timers_running)
230 #define V_current_state_timers_running VNET(current_state_timers_running)
232 VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) =
233 LIST_HEAD_INITIALIZER(igi_head);
234 VNET_DEFINE_STATIC(struct igmpstat, igmpstat) = {
235 .igps_version = IGPS_VERSION_3,
236 .igps_len = sizeof(struct igmpstat),
238 VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0};
240 #define V_igi_head VNET(igi_head)
241 #define V_igmpstat VNET(igmpstat)
242 #define V_igmp_gsrdelay VNET(igmp_gsrdelay)
244 VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1;
245 VNET_DEFINE_STATIC(int, igmp_sendra) = 1;
246 VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1;
247 VNET_DEFINE_STATIC(int, igmp_v1enable) = 1;
248 VNET_DEFINE_STATIC(int, igmp_v2enable) = 1;
249 VNET_DEFINE_STATIC(int, igmp_legacysupp);
250 VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3;
252 #define V_igmp_recvifkludge VNET(igmp_recvifkludge)
253 #define V_igmp_sendra VNET(igmp_sendra)
254 #define V_igmp_sendlocal VNET(igmp_sendlocal)
255 #define V_igmp_v1enable VNET(igmp_v1enable)
256 #define V_igmp_v2enable VNET(igmp_v2enable)
257 #define V_igmp_legacysupp VNET(igmp_legacysupp)
258 #define V_igmp_default_version VNET(igmp_default_version)
261 * Virtualized sysctls.
263 SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW,
264 &VNET_NAME(igmpstat), igmpstat, "");
265 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
266 &VNET_NAME(igmp_recvifkludge), 0,
267 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
268 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
269 &VNET_NAME(igmp_sendra), 0,
270 "Send IP Router Alert option in IGMPv2/v3 messages");
271 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
272 &VNET_NAME(igmp_sendlocal), 0,
273 "Send IGMP membership reports for 224.0.0.0/24 groups");
274 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
275 &VNET_NAME(igmp_v1enable), 0,
276 "Enable backwards compatibility with IGMPv1");
277 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
278 &VNET_NAME(igmp_v2enable), 0,
279 "Enable backwards compatibility with IGMPv2");
280 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
281 &VNET_NAME(igmp_legacysupp), 0,
282 "Allow v1/v2 reports to suppress v3 group responses");
283 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
284 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
285 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
286 "Default version of IGMP to run on each interface");
287 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
288 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
289 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
290 "Rate limit for IGMPv3 Group-and-Source queries in seconds");
293 * Non-virtualized sysctls.
295 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
296 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
297 "Per-interface IGMPv3 state");
300 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
304 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
306 m->m_pkthdr.rcvif = ifp;
307 m->m_pkthdr.flowid = ifp->if_index;
311 igmp_scrub_context(struct mbuf *m)
314 m->m_pkthdr.PH_loc.ptr = NULL;
315 m->m_pkthdr.flowid = 0;
319 * Restore context from a queued IGMP output chain.
320 * Return saved ifindex.
322 * VIMAGE: The assertion is there to make sure that we
323 * actually called CURVNET_SET() with what's in the mbuf chain.
325 static __inline uint32_t
326 igmp_restore_context(struct mbuf *m)
330 #if defined(VIMAGE) && defined(INVARIANTS)
331 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
332 ("%s: called when curvnet was not restored", __func__));
335 return (m->m_pkthdr.flowid);
339 * Retrieve or set default IGMP version.
341 * VIMAGE: Assume curvnet set by caller.
342 * SMPng: NOTE: Serialized by IGMP lock.
345 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
350 error = sysctl_wire_old_buffer(req, sizeof(int));
356 new = V_igmp_default_version;
358 error = sysctl_handle_int(oidp, &new, 0, req);
359 if (error || !req->newptr)
362 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
367 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
368 V_igmp_default_version, new);
370 V_igmp_default_version = new;
378 * Retrieve or set threshold between group-source queries in seconds.
380 * VIMAGE: Assume curvnet set by caller.
381 * SMPng: NOTE: Serialized by IGMP lock.
384 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
389 error = sysctl_wire_old_buffer(req, sizeof(int));
395 i = V_igmp_gsrdelay.tv_sec;
397 error = sysctl_handle_int(oidp, &i, 0, req);
398 if (error || !req->newptr)
401 if (i < -1 || i >= 60) {
406 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
407 V_igmp_gsrdelay.tv_sec, i);
408 V_igmp_gsrdelay.tv_sec = i;
416 * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
417 * For use by ifmcstat(8).
419 * SMPng: NOTE: Does an unlocked ifindex space read.
420 * VIMAGE: Assume curvnet set by caller. The node handler itself
421 * is not directly virtualized.
424 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
430 struct igmp_ifsoftc *igi;
435 if (req->newptr != NULL)
441 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
445 IN_MULTI_LIST_LOCK();
448 if (name[0] <= 0 || name[0] > V_if_index) {
455 ifp = ifnet_byindex(name[0]);
459 LIST_FOREACH(igi, &V_igi_head, igi_link) {
460 if (ifp == igi->igi_ifp) {
461 struct igmp_ifinfo info;
463 info.igi_version = igi->igi_version;
464 info.igi_v1_timer = igi->igi_v1_timer;
465 info.igi_v2_timer = igi->igi_v2_timer;
466 info.igi_v3_timer = igi->igi_v3_timer;
467 info.igi_flags = igi->igi_flags;
468 info.igi_rv = igi->igi_rv;
469 info.igi_qi = igi->igi_qi;
470 info.igi_qri = igi->igi_qri;
471 info.igi_uri = igi->igi_uri;
472 error = SYSCTL_OUT(req, &info, sizeof(info));
479 IN_MULTI_LIST_UNLOCK();
484 * Dispatch an entire queue of pending packet chains
486 * VIMAGE: Assumes the vnet pointer has been set.
489 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
491 struct epoch_tracker et;
495 while ((m = mbufq_dequeue(mq)) != NULL) {
496 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
498 m->m_flags |= M_IGMP_LOOP;
499 netisr_dispatch(NETISR_IGMP, m);
507 * Filter outgoing IGMP report state by group.
509 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
510 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
511 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
512 * this may break certain IGMP snooping switches which rely on the old
515 * Return zero if the given group is one for which IGMP reports
516 * should be suppressed, or non-zero if reports should be issued.
519 igmp_isgroupreported(const struct in_addr addr)
522 if (in_allhosts(addr) ||
523 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
530 * Construct a Router Alert option to use in outgoing packets.
538 m = m_get(M_WAITOK, MT_DATA);
539 p = mtod(m, struct ipoption *);
540 p->ipopt_dst.s_addr = INADDR_ANY;
541 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */
542 p->ipopt_list[1] = 0x04; /* 4 bytes long */
543 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
544 p->ipopt_list[3] = 0x00; /* pad byte */
545 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
551 * Attach IGMP when PF_INET is attached to an interface.
553 struct igmp_ifsoftc *
554 igmp_domifattach(struct ifnet *ifp)
556 struct igmp_ifsoftc *igi;
558 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
559 __func__, ifp, ifp->if_xname);
563 igi = igi_alloc_locked(ifp);
564 if (!(ifp->if_flags & IFF_MULTICAST))
565 igi->igi_flags |= IGIF_SILENT;
573 * VIMAGE: assume curvnet set by caller.
575 static struct igmp_ifsoftc *
576 igi_alloc_locked(/*const*/ struct ifnet *ifp)
578 struct igmp_ifsoftc *igi;
582 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
587 igi->igi_version = V_igmp_default_version;
589 igi->igi_rv = IGMP_RV_INIT;
590 igi->igi_qi = IGMP_QI_INIT;
591 igi->igi_qri = IGMP_QRI_INIT;
592 igi->igi_uri = IGMP_URI_INIT;
593 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
595 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
597 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
607 * NOTE: Some finalization tasks need to run before the protocol domain
608 * is detached, but also before the link layer does its cleanup.
610 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
611 * XXX This is also bitten by unlocked ifma_protospec access.
614 igmp_ifdetach(struct ifnet *ifp)
616 struct igmp_ifsoftc *igi;
617 struct ifmultiaddr *ifma, *next;
618 struct in_multi *inm;
619 struct in_multi_head inm_free_tmp;
620 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
623 SLIST_INIT(&inm_free_tmp);
626 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
627 if (igi->igi_version == IGMP_VERSION_3) {
630 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
631 if (ifma->ifma_addr->sa_family != AF_INET ||
632 ifma->ifma_protospec == NULL)
634 inm = (struct in_multi *)ifma->ifma_protospec;
635 if (inm->inm_state == IGMP_LEAVING_MEMBER)
636 inm_rele_locked(&inm_free_tmp, inm);
637 inm_clear_recorded(inm);
638 if (__predict_false(ifma_restart)) {
639 ifma_restart = false;
643 IF_ADDR_WUNLOCK(ifp);
644 inm_release_list_deferred(&inm_free_tmp);
651 * Hook for domifdetach.
654 igmp_domifdetach(struct ifnet *ifp)
657 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
658 __func__, ifp, ifp->if_xname);
661 igi_delete_locked(ifp);
666 igi_delete_locked(const struct ifnet *ifp)
668 struct igmp_ifsoftc *igi, *tigi;
670 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
671 __func__, ifp, ifp->if_xname);
675 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
676 if (igi->igi_ifp == ifp) {
678 * Free deferred General Query responses.
680 mbufq_drain(&igi->igi_gq);
682 LIST_REMOVE(igi, igi_link);
690 * Process a received IGMPv1 query.
691 * Return non-zero if the message should be dropped.
693 * VIMAGE: The curvnet pointer is derived from the input ifp.
696 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
697 const struct igmp *igmp)
699 struct ifmultiaddr *ifma;
700 struct igmp_ifsoftc *igi;
701 struct in_multi *inm;
706 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
707 * 224.0.0.1. They are always treated as General Queries.
708 * igmp_group is always ignored. Do not drop it as a userland
709 * daemon may wish to see it.
710 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
712 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
713 IGMPSTAT_INC(igps_rcv_badqueries);
716 IGMPSTAT_INC(igps_rcv_gen_queries);
718 IN_MULTI_LIST_LOCK();
721 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
722 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
724 if (igi->igi_flags & IGIF_LOOPBACK) {
725 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
731 * Switch to IGMPv1 host compatibility mode.
733 igmp_set_version(igi, IGMP_VERSION_1);
735 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
738 * Start the timers in all of our group records
739 * for the interface on which the query arrived,
740 * except those which are already running.
742 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
743 if (ifma->ifma_addr->sa_family != AF_INET ||
744 ifma->ifma_protospec == NULL)
746 inm = (struct in_multi *)ifma->ifma_protospec;
747 if (inm->inm_timer != 0)
749 switch (inm->inm_state) {
750 case IGMP_NOT_MEMBER:
751 case IGMP_SILENT_MEMBER:
753 case IGMP_G_QUERY_PENDING_MEMBER:
754 case IGMP_SG_QUERY_PENDING_MEMBER:
755 case IGMP_REPORTING_MEMBER:
756 case IGMP_IDLE_MEMBER:
757 case IGMP_LAZY_MEMBER:
758 case IGMP_SLEEPING_MEMBER:
759 case IGMP_AWAKENING_MEMBER:
760 inm->inm_state = IGMP_REPORTING_MEMBER;
761 inm->inm_timer = IGMP_RANDOM_DELAY(
762 IGMP_V1V2_MAX_RI * PR_FASTHZ);
763 V_current_state_timers_running = 1;
765 case IGMP_LEAVING_MEMBER:
772 IN_MULTI_LIST_UNLOCK();
778 * Process a received IGMPv2 general or group-specific query.
781 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
782 const struct igmp *igmp)
784 struct ifmultiaddr *ifma;
785 struct igmp_ifsoftc *igi;
786 struct in_multi *inm;
787 int is_general_query;
792 is_general_query = 0;
795 * Validate address fields upfront.
796 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
798 if (in_nullhost(igmp->igmp_group)) {
800 * IGMPv2 General Query.
801 * If this was not sent to the all-hosts group, ignore it.
803 if (!in_allhosts(ip->ip_dst))
805 IGMPSTAT_INC(igps_rcv_gen_queries);
806 is_general_query = 1;
808 /* IGMPv2 Group-Specific Query. */
809 IGMPSTAT_INC(igps_rcv_group_queries);
812 IN_MULTI_LIST_LOCK();
815 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
816 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
818 if (igi->igi_flags & IGIF_LOOPBACK) {
819 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
825 * Ignore v2 query if in v1 Compatibility Mode.
827 if (igi->igi_version == IGMP_VERSION_1)
830 igmp_set_version(igi, IGMP_VERSION_2);
832 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
836 if (is_general_query) {
838 * For each reporting group joined on this
839 * interface, kick the report timer.
841 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
843 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
844 if (ifma->ifma_addr->sa_family != AF_INET ||
845 ifma->ifma_protospec == NULL)
847 inm = (struct in_multi *)ifma->ifma_protospec;
848 igmp_v2_update_group(inm, timer);
852 * Group-specific IGMPv2 query, we need only
853 * look up the single group to process it.
855 inm = inm_lookup(ifp, igmp->igmp_group);
858 "process v2 query 0x%08x on ifp %p(%s)",
859 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
860 igmp_v2_update_group(inm, timer);
866 IN_MULTI_LIST_UNLOCK();
872 * Update the report timer on a group in response to an IGMPv2 query.
874 * If we are becoming the reporting member for this group, start the timer.
875 * If we already are the reporting member for this group, and timer is
876 * below the threshold, reset it.
878 * We may be updating the group for the first time since we switched
879 * to IGMPv3. If we are, then we must clear any recorded source lists,
880 * and transition to REPORTING state; the group timer is overloaded
881 * for group and group-source query responses.
883 * Unlike IGMPv3, the delay per group should be jittered
884 * to avoid bursts of IGMPv2 reports.
887 igmp_v2_update_group(struct in_multi *inm, const int timer)
890 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
891 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
893 IN_MULTI_LIST_LOCK_ASSERT();
895 switch (inm->inm_state) {
896 case IGMP_NOT_MEMBER:
897 case IGMP_SILENT_MEMBER:
899 case IGMP_REPORTING_MEMBER:
900 if (inm->inm_timer != 0 &&
901 inm->inm_timer <= timer) {
902 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
903 "skipping.", __func__);
907 case IGMP_SG_QUERY_PENDING_MEMBER:
908 case IGMP_G_QUERY_PENDING_MEMBER:
909 case IGMP_IDLE_MEMBER:
910 case IGMP_LAZY_MEMBER:
911 case IGMP_AWAKENING_MEMBER:
912 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
913 inm->inm_state = IGMP_REPORTING_MEMBER;
914 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
915 V_current_state_timers_running = 1;
917 case IGMP_SLEEPING_MEMBER:
918 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
919 inm->inm_state = IGMP_AWAKENING_MEMBER;
921 case IGMP_LEAVING_MEMBER:
927 * Process a received IGMPv3 general, group-specific or
928 * group-and-source-specific query.
929 * Assumes m has already been pulled up to the full IGMP message length.
930 * Return 0 if successful, otherwise an appropriate error code is returned.
933 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
934 /*const*/ struct igmpv3 *igmpv3)
936 struct igmp_ifsoftc *igi;
937 struct in_multi *inm;
938 int is_general_query;
939 uint32_t maxresp, nsrc, qqi;
943 is_general_query = 0;
945 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
947 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
948 if (maxresp >= 128) {
949 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
950 (IGMP_EXP(igmpv3->igmp_code) + 3);
954 * Robustness must never be less than 2 for on-wire IGMPv3.
955 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
956 * an exception for interfaces whose IGMPv3 state changes
957 * are redirected to loopback (e.g. MANET).
959 qrv = IGMP_QRV(igmpv3->igmp_misc);
961 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
966 qqi = igmpv3->igmp_qqi;
968 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
969 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
972 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
976 nsrc = ntohs(igmpv3->igmp_numsrc);
979 * Validate address fields and versions upfront before
980 * accepting v3 query.
981 * XXX SMPng: Unlocked access to igmpstat counters here.
983 if (in_nullhost(igmpv3->igmp_group)) {
985 * IGMPv3 General Query.
987 * General Queries SHOULD be directed to 224.0.0.1.
988 * A general query with a source list has undefined
989 * behaviour; discard it.
991 IGMPSTAT_INC(igps_rcv_gen_queries);
992 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
993 IGMPSTAT_INC(igps_rcv_badqueries);
996 is_general_query = 1;
998 /* Group or group-source specific query. */
1000 IGMPSTAT_INC(igps_rcv_group_queries);
1002 IGMPSTAT_INC(igps_rcv_gsr_queries);
1005 IN_MULTI_LIST_LOCK();
1008 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1009 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1011 if (igi->igi_flags & IGIF_LOOPBACK) {
1012 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1013 ifp, ifp->if_xname);
1018 * Discard the v3 query if we're in Compatibility Mode.
1019 * The RFC is not obviously worded that hosts need to stay in
1020 * compatibility mode until the Old Version Querier Present
1023 if (igi->igi_version != IGMP_VERSION_3) {
1024 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1025 igi->igi_version, ifp, ifp->if_xname);
1029 igmp_set_version(igi, IGMP_VERSION_3);
1032 igi->igi_qri = maxresp;
1034 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1037 if (is_general_query) {
1039 * Schedule a current-state report on this ifp for
1040 * all groups, possibly containing source lists.
1041 * If there is a pending General Query response
1042 * scheduled earlier than the selected delay, do
1043 * not schedule any other reports.
1044 * Otherwise, reset the interface timer.
1046 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1047 ifp, ifp->if_xname);
1048 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1049 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1050 V_interface_timers_running = 1;
1054 * Group-source-specific queries are throttled on
1055 * a per-group basis to defeat denial-of-service attempts.
1056 * Queries for groups we are not a member of on this
1057 * link are simply ignored.
1059 inm = inm_lookup(ifp, igmpv3->igmp_group);
1063 if (!ratecheck(&inm->inm_lastgsrtv,
1064 &V_igmp_gsrdelay)) {
1065 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1067 IGMPSTAT_INC(igps_drop_gsr_queries);
1071 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1072 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1074 * If there is a pending General Query response
1075 * scheduled sooner than the selected delay, no
1076 * further report need be scheduled.
1077 * Otherwise, prepare to respond to the
1078 * group-specific or group-and-source query.
1080 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1081 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1086 IN_MULTI_LIST_UNLOCK();
1092 * Process a received IGMPv3 group-specific or group-and-source-specific
1094 * Return <0 if any error occurred. Currently this is ignored.
1097 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1098 int timer, /*const*/ struct igmpv3 *igmpv3)
1103 IN_MULTI_LIST_LOCK_ASSERT();
1108 switch (inm->inm_state) {
1109 case IGMP_NOT_MEMBER:
1110 case IGMP_SILENT_MEMBER:
1111 case IGMP_SLEEPING_MEMBER:
1112 case IGMP_LAZY_MEMBER:
1113 case IGMP_AWAKENING_MEMBER:
1114 case IGMP_IDLE_MEMBER:
1115 case IGMP_LEAVING_MEMBER:
1118 case IGMP_REPORTING_MEMBER:
1119 case IGMP_G_QUERY_PENDING_MEMBER:
1120 case IGMP_SG_QUERY_PENDING_MEMBER:
1124 nsrc = ntohs(igmpv3->igmp_numsrc);
1127 * Deal with group-specific queries upfront.
1128 * If any group query is already pending, purge any recorded
1129 * source-list state if it exists, and schedule a query response
1130 * for this group-specific query.
1133 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1134 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1135 inm_clear_recorded(inm);
1136 timer = min(inm->inm_timer, timer);
1138 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1139 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1140 V_current_state_timers_running = 1;
1145 * Deal with the case where a group-and-source-specific query has
1146 * been received but a group-specific query is already pending.
1148 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1149 timer = min(inm->inm_timer, timer);
1150 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1151 V_current_state_timers_running = 1;
1156 * Finally, deal with the case where a group-and-source-specific
1157 * query has been received, where a response to a previous g-s-r
1158 * query exists, or none exists.
1159 * In this case, we need to parse the source-list which the Querier
1160 * has provided us with and check if we have any source list filter
1161 * entries at T1 for these sources. If we do not, there is no need
1162 * schedule a report and the query may be dropped.
1163 * If we do, we must record them and schedule a current-state
1164 * report for those sources.
1165 * FIXME: Handling source lists larger than 1 mbuf requires that
1166 * we pass the mbuf chain pointer down to this function, and use
1167 * m_getptr() to walk the chain.
1169 if (inm->inm_nsrc > 0) {
1170 const struct in_addr *ap;
1173 ap = (const struct in_addr *)(igmpv3 + 1);
1175 for (i = 0; i < nsrc; i++, ap++) {
1176 retval = inm_record_source(inm, ap->s_addr);
1179 nrecorded += retval;
1181 if (nrecorded > 0) {
1183 "%s: schedule response to SG query", __func__);
1184 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1185 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1186 V_current_state_timers_running = 1;
1194 * Process a received IGMPv1 host membership report.
1196 * NOTE: 0.0.0.0 workaround breaks const correctness.
1199 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1200 /*const*/ struct igmp *igmp)
1202 struct rm_priotracker in_ifa_tracker;
1203 struct in_ifaddr *ia;
1204 struct in_multi *inm;
1206 IGMPSTAT_INC(igps_rcv_reports);
1208 if (ifp->if_flags & IFF_LOOPBACK)
1211 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1212 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1213 IGMPSTAT_INC(igps_rcv_badreports);
1218 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1219 * Booting clients may use the source address 0.0.0.0. Some
1220 * IGMP daemons may not know how to use IP_RECVIF to determine
1221 * the interface upon which this message was received.
1222 * Replace 0.0.0.0 with the subnet address if told to do so.
1224 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1225 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1227 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1230 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1231 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1234 * IGMPv1 report suppression.
1235 * If we are a member of this group, and our membership should be
1236 * reported, stop our group timer and transition to the 'lazy' state.
1238 IN_MULTI_LIST_LOCK();
1239 inm = inm_lookup(ifp, igmp->igmp_group);
1241 struct igmp_ifsoftc *igi;
1245 KASSERT(igi != NULL,
1246 ("%s: no igi for ifp %p", __func__, ifp));
1250 IGMPSTAT_INC(igps_rcv_ourreports);
1253 * If we are in IGMPv3 host mode, do not allow the
1254 * other host's IGMPv1 report to suppress our reports
1255 * unless explicitly configured to do so.
1257 if (igi->igi_version == IGMP_VERSION_3) {
1258 if (V_igmp_legacysupp)
1259 igmp_v3_suppress_group_record(inm);
1265 switch (inm->inm_state) {
1266 case IGMP_NOT_MEMBER:
1267 case IGMP_SILENT_MEMBER:
1269 case IGMP_IDLE_MEMBER:
1270 case IGMP_LAZY_MEMBER:
1271 case IGMP_AWAKENING_MEMBER:
1273 "report suppressed for 0x%08x on ifp %p(%s)",
1274 ntohl(igmp->igmp_group.s_addr), ifp,
1276 case IGMP_SLEEPING_MEMBER:
1277 inm->inm_state = IGMP_SLEEPING_MEMBER;
1279 case IGMP_REPORTING_MEMBER:
1281 "report suppressed for 0x%08x on ifp %p(%s)",
1282 ntohl(igmp->igmp_group.s_addr), ifp,
1284 if (igi->igi_version == IGMP_VERSION_1)
1285 inm->inm_state = IGMP_LAZY_MEMBER;
1286 else if (igi->igi_version == IGMP_VERSION_2)
1287 inm->inm_state = IGMP_SLEEPING_MEMBER;
1289 case IGMP_G_QUERY_PENDING_MEMBER:
1290 case IGMP_SG_QUERY_PENDING_MEMBER:
1291 case IGMP_LEAVING_MEMBER:
1297 IN_MULTI_LIST_UNLOCK();
1303 * Process a received IGMPv2 host membership report.
1305 * NOTE: 0.0.0.0 workaround breaks const correctness.
1308 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1309 /*const*/ struct igmp *igmp)
1311 struct rm_priotracker in_ifa_tracker;
1312 struct in_ifaddr *ia;
1313 struct in_multi *inm;
1316 * Make sure we don't hear our own membership report. Fast
1317 * leave requires knowing that we are the only member of a
1320 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1321 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1325 IGMPSTAT_INC(igps_rcv_reports);
1327 if (ifp->if_flags & IFF_LOOPBACK) {
1331 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1332 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1333 IGMPSTAT_INC(igps_rcv_badreports);
1338 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1339 * Booting clients may use the source address 0.0.0.0. Some
1340 * IGMP daemons may not know how to use IP_RECVIF to determine
1341 * the interface upon which this message was received.
1342 * Replace 0.0.0.0 with the subnet address if told to do so.
1344 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1346 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1349 CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1350 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1353 * IGMPv2 report suppression.
1354 * If we are a member of this group, and our membership should be
1355 * reported, and our group timer is pending or about to be reset,
1356 * stop our group timer by transitioning to the 'lazy' state.
1358 IN_MULTI_LIST_LOCK();
1359 inm = inm_lookup(ifp, igmp->igmp_group);
1361 struct igmp_ifsoftc *igi;
1364 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1366 IGMPSTAT_INC(igps_rcv_ourreports);
1369 * If we are in IGMPv3 host mode, do not allow the
1370 * other host's IGMPv1 report to suppress our reports
1371 * unless explicitly configured to do so.
1373 if (igi->igi_version == IGMP_VERSION_3) {
1374 if (V_igmp_legacysupp)
1375 igmp_v3_suppress_group_record(inm);
1381 switch (inm->inm_state) {
1382 case IGMP_NOT_MEMBER:
1383 case IGMP_SILENT_MEMBER:
1384 case IGMP_SLEEPING_MEMBER:
1386 case IGMP_REPORTING_MEMBER:
1387 case IGMP_IDLE_MEMBER:
1388 case IGMP_AWAKENING_MEMBER:
1390 "report suppressed for 0x%08x on ifp %p(%s)",
1391 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1392 case IGMP_LAZY_MEMBER:
1393 inm->inm_state = IGMP_LAZY_MEMBER;
1395 case IGMP_G_QUERY_PENDING_MEMBER:
1396 case IGMP_SG_QUERY_PENDING_MEMBER:
1397 case IGMP_LEAVING_MEMBER:
1403 IN_MULTI_LIST_UNLOCK();
1409 igmp_input(struct mbuf **mp, int *offp, int proto)
1420 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1423 ifp = m->m_pkthdr.rcvif;
1426 IGMPSTAT_INC(igps_rcv_total);
1428 ip = mtod(m, struct ip *);
1430 igmplen = ntohs(ip->ip_len) - iphlen;
1435 if (igmplen < IGMP_MINLEN) {
1436 IGMPSTAT_INC(igps_rcv_tooshort);
1438 return (IPPROTO_DONE);
1442 * Always pullup to the minimum size for v1/v2 or v3
1443 * to amortize calls to m_pullup().
1446 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1447 minlen += IGMP_V3_QUERY_MINLEN;
1449 minlen += IGMP_MINLEN;
1450 if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1451 (m = m_pullup(m, minlen)) == NULL) {
1452 IGMPSTAT_INC(igps_rcv_tooshort);
1453 return (IPPROTO_DONE);
1455 ip = mtod(m, struct ip *);
1458 * Validate checksum.
1460 m->m_data += iphlen;
1462 igmp = mtod(m, struct igmp *);
1463 if (in_cksum(m, igmplen)) {
1464 IGMPSTAT_INC(igps_rcv_badsum);
1466 return (IPPROTO_DONE);
1468 m->m_data -= iphlen;
1472 * IGMP control traffic is link-scope, and must have a TTL of 1.
1473 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1474 * probe packets may come from beyond the LAN.
1476 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1477 IGMPSTAT_INC(igps_rcv_badttl);
1479 return (IPPROTO_DONE);
1482 switch (igmp->igmp_type) {
1483 case IGMP_HOST_MEMBERSHIP_QUERY:
1484 if (igmplen == IGMP_MINLEN) {
1485 if (igmp->igmp_code == 0)
1486 queryver = IGMP_VERSION_1;
1488 queryver = IGMP_VERSION_2;
1489 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1490 queryver = IGMP_VERSION_3;
1492 IGMPSTAT_INC(igps_rcv_tooshort);
1494 return (IPPROTO_DONE);
1498 case IGMP_VERSION_1:
1499 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1500 if (!V_igmp_v1enable)
1502 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1504 return (IPPROTO_DONE);
1508 case IGMP_VERSION_2:
1509 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1510 if (!V_igmp_v2enable)
1512 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1514 return (IPPROTO_DONE);
1518 case IGMP_VERSION_3: {
1519 struct igmpv3 *igmpv3;
1523 IGMPSTAT_INC(igps_rcv_v3_queries);
1524 igmpv3 = (struct igmpv3 *)igmp;
1526 * Validate length based on source count.
1528 nsrc = ntohs(igmpv3->igmp_numsrc);
1529 if (nsrc * sizeof(in_addr_t) >
1530 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1531 IGMPSTAT_INC(igps_rcv_tooshort);
1532 return (IPPROTO_DONE);
1535 * m_pullup() may modify m, so pullup in
1538 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1539 sizeof(struct in_addr) * nsrc;
1540 if ((!M_WRITABLE(m) ||
1541 m->m_len < igmpv3len) &&
1542 (m = m_pullup(m, igmpv3len)) == NULL) {
1543 IGMPSTAT_INC(igps_rcv_tooshort);
1544 return (IPPROTO_DONE);
1546 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1548 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1550 return (IPPROTO_DONE);
1557 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1558 if (!V_igmp_v1enable)
1560 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1562 return (IPPROTO_DONE);
1566 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1567 if (!V_igmp_v2enable)
1569 if (!ip_checkrouteralert(m))
1570 IGMPSTAT_INC(igps_rcv_nora);
1571 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1573 return (IPPROTO_DONE);
1577 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1579 * Hosts do not need to process IGMPv3 membership reports,
1580 * as report suppression is no longer required.
1582 if (!ip_checkrouteralert(m))
1583 IGMPSTAT_INC(igps_rcv_nora);
1591 * Pass all valid IGMP packets up to any process(es) listening on a
1595 return (rip_input(mp, offp, proto));
1600 * Fast timeout handler (global).
1601 * VIMAGE: Timeout handlers are expected to service all vimages.
1606 VNET_ITERATOR_DECL(vnet_iter);
1608 VNET_LIST_RLOCK_NOSLEEP();
1609 VNET_FOREACH(vnet_iter) {
1610 CURVNET_SET(vnet_iter);
1611 igmp_fasttimo_vnet();
1614 VNET_LIST_RUNLOCK_NOSLEEP();
1618 * Fast timeout handler (per-vnet).
1619 * Sends are shuffled off to a netisr to deal with Giant.
1621 * VIMAGE: Assume caller has set up our curvnet.
1624 igmp_fasttimo_vnet(void)
1626 struct mbufq scq; /* State-change packets */
1627 struct mbufq qrq; /* Query response packets */
1629 struct igmp_ifsoftc *igi;
1630 struct ifmultiaddr *ifma, *next;
1631 struct in_multi *inm;
1632 struct in_multi_head inm_free_tmp;
1633 int loop, uri_fasthz;
1639 * Quick check to see if any work needs to be done, in order to
1640 * minimize the overhead of fasttimo processing.
1641 * SMPng: XXX Unlocked reads.
1643 if (!V_current_state_timers_running &&
1644 !V_interface_timers_running &&
1645 !V_state_change_timers_running)
1648 SLIST_INIT(&inm_free_tmp);
1649 IN_MULTI_LIST_LOCK();
1653 * IGMPv3 General Query response timer processing.
1655 if (V_interface_timers_running) {
1656 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1658 V_interface_timers_running = 0;
1659 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1660 if (igi->igi_v3_timer == 0) {
1662 } else if (--igi->igi_v3_timer == 0) {
1663 igmp_v3_dispatch_general_query(igi);
1665 V_interface_timers_running = 1;
1670 if (!V_current_state_timers_running &&
1671 !V_state_change_timers_running)
1674 V_current_state_timers_running = 0;
1675 V_state_change_timers_running = 0;
1677 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1680 * IGMPv1/v2/v3 host report and state-change timer processing.
1681 * Note: Processing a v3 group timer may remove a node.
1683 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1686 if (igi->igi_version == IGMP_VERSION_3) {
1687 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1688 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1690 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1691 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1696 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1697 if (ifma->ifma_addr->sa_family != AF_INET ||
1698 ifma->ifma_protospec == NULL)
1700 inm = (struct in_multi *)ifma->ifma_protospec;
1701 switch (igi->igi_version) {
1702 case IGMP_VERSION_1:
1703 case IGMP_VERSION_2:
1704 igmp_v1v2_process_group_timer(inm,
1707 case IGMP_VERSION_3:
1708 igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1709 &scq, inm, uri_fasthz);
1712 if (__predict_false(ifma_restart)) {
1713 ifma_restart = false;
1717 IF_ADDR_WUNLOCK(ifp);
1719 if (igi->igi_version == IGMP_VERSION_3) {
1720 igmp_dispatch_queue(&qrq, 0, loop);
1721 igmp_dispatch_queue(&scq, 0, loop);
1724 * Free the in_multi reference(s) for this
1727 inm_release_list_deferred(&inm_free_tmp);
1733 IN_MULTI_LIST_UNLOCK();
1737 * Update host report group timer for IGMPv1/v2.
1738 * Will update the global pending timer flags.
1741 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1743 int report_timer_expired;
1745 IN_MULTI_LIST_LOCK_ASSERT();
1748 if (inm->inm_timer == 0) {
1749 report_timer_expired = 0;
1750 } else if (--inm->inm_timer == 0) {
1751 report_timer_expired = 1;
1753 V_current_state_timers_running = 1;
1757 switch (inm->inm_state) {
1758 case IGMP_NOT_MEMBER:
1759 case IGMP_SILENT_MEMBER:
1760 case IGMP_IDLE_MEMBER:
1761 case IGMP_LAZY_MEMBER:
1762 case IGMP_SLEEPING_MEMBER:
1763 case IGMP_AWAKENING_MEMBER:
1765 case IGMP_REPORTING_MEMBER:
1766 if (report_timer_expired) {
1767 inm->inm_state = IGMP_IDLE_MEMBER;
1768 (void)igmp_v1v2_queue_report(inm,
1769 (version == IGMP_VERSION_2) ?
1770 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1771 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1774 case IGMP_G_QUERY_PENDING_MEMBER:
1775 case IGMP_SG_QUERY_PENDING_MEMBER:
1776 case IGMP_LEAVING_MEMBER:
1782 * Update a group's timers for IGMPv3.
1783 * Will update the global pending timer flags.
1784 * Note: Unlocked read from igi.
1787 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1788 struct mbufq *qrq, struct mbufq *scq,
1789 struct in_multi *inm, const int uri_fasthz)
1791 int query_response_timer_expired;
1792 int state_change_retransmit_timer_expired;
1794 IN_MULTI_LIST_LOCK_ASSERT();
1797 query_response_timer_expired = 0;
1798 state_change_retransmit_timer_expired = 0;
1801 * During a transition from v1/v2 compatibility mode back to v3,
1802 * a group record in REPORTING state may still have its group
1803 * timer active. This is a no-op in this function; it is easier
1804 * to deal with it here than to complicate the slow-timeout path.
1806 if (inm->inm_timer == 0) {
1807 query_response_timer_expired = 0;
1808 } else if (--inm->inm_timer == 0) {
1809 query_response_timer_expired = 1;
1811 V_current_state_timers_running = 1;
1814 if (inm->inm_sctimer == 0) {
1815 state_change_retransmit_timer_expired = 0;
1816 } else if (--inm->inm_sctimer == 0) {
1817 state_change_retransmit_timer_expired = 1;
1819 V_state_change_timers_running = 1;
1822 /* We are in fasttimo, so be quick about it. */
1823 if (!state_change_retransmit_timer_expired &&
1824 !query_response_timer_expired)
1827 switch (inm->inm_state) {
1828 case IGMP_NOT_MEMBER:
1829 case IGMP_SILENT_MEMBER:
1830 case IGMP_SLEEPING_MEMBER:
1831 case IGMP_LAZY_MEMBER:
1832 case IGMP_AWAKENING_MEMBER:
1833 case IGMP_IDLE_MEMBER:
1835 case IGMP_G_QUERY_PENDING_MEMBER:
1836 case IGMP_SG_QUERY_PENDING_MEMBER:
1838 * Respond to a previously pending Group-Specific
1839 * or Group-and-Source-Specific query by enqueueing
1840 * the appropriate Current-State report for
1841 * immediate transmission.
1843 if (query_response_timer_expired) {
1844 int retval __unused;
1846 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1847 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1848 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1850 inm->inm_state = IGMP_REPORTING_MEMBER;
1851 /* XXX Clear recorded sources for next time. */
1852 inm_clear_recorded(inm);
1855 case IGMP_REPORTING_MEMBER:
1856 case IGMP_LEAVING_MEMBER:
1857 if (state_change_retransmit_timer_expired) {
1859 * State-change retransmission timer fired.
1860 * If there are any further pending retransmissions,
1861 * set the global pending state-change flag, and
1864 if (--inm->inm_scrv > 0) {
1865 inm->inm_sctimer = uri_fasthz;
1866 V_state_change_timers_running = 1;
1869 * Retransmit the previously computed state-change
1870 * report. If there are no further pending
1871 * retransmissions, the mbuf queue will be consumed.
1872 * Update T0 state to T1 as we have now sent
1875 (void)igmp_v3_merge_state_changes(inm, scq);
1878 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1879 ntohl(inm->inm_addr.s_addr),
1880 inm->inm_ifp->if_xname);
1883 * If we are leaving the group for good, make sure
1884 * we release IGMP's reference to it.
1885 * This release must be deferred using a SLIST,
1886 * as we are called from a loop which traverses
1887 * the in_ifmultiaddr TAILQ.
1889 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1890 inm->inm_scrv == 0) {
1891 inm->inm_state = IGMP_NOT_MEMBER;
1892 inm_rele_locked(inmh, inm);
1901 * Suppress a group's pending response to a group or source/group query.
1903 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1904 * Do NOT update ST1/ST0 as this operation merely suppresses
1905 * the currently pending group record.
1906 * Do NOT suppress the response to a general query. It is possible but
1907 * it would require adding another state or flag.
1910 igmp_v3_suppress_group_record(struct in_multi *inm)
1913 IN_MULTI_LIST_LOCK_ASSERT();
1915 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1916 ("%s: not IGMPv3 mode on link", __func__));
1918 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1919 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1922 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1923 inm_clear_recorded(inm);
1926 inm->inm_state = IGMP_REPORTING_MEMBER;
1930 * Switch to a different IGMP version on the given interface,
1931 * as per Section 7.2.1.
1934 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1936 int old_version_timer;
1940 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1941 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1943 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1945 * Compute the "Older Version Querier Present" timer as per
1948 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1949 old_version_timer *= PR_SLOWHZ;
1951 if (version == IGMP_VERSION_1) {
1952 igi->igi_v1_timer = old_version_timer;
1953 igi->igi_v2_timer = 0;
1954 } else if (version == IGMP_VERSION_2) {
1955 igi->igi_v1_timer = 0;
1956 igi->igi_v2_timer = old_version_timer;
1960 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1961 if (igi->igi_version != IGMP_VERSION_2) {
1962 igi->igi_version = IGMP_VERSION_2;
1963 igmp_v3_cancel_link_timers(igi);
1965 } else if (igi->igi_v1_timer > 0) {
1966 if (igi->igi_version != IGMP_VERSION_1) {
1967 igi->igi_version = IGMP_VERSION_1;
1968 igmp_v3_cancel_link_timers(igi);
1974 * Cancel pending IGMPv3 timers for the given link and all groups
1975 * joined on it; state-change, general-query, and group-query timers.
1977 * Only ever called on a transition from v3 to Compatibility mode. Kill
1978 * the timers stone dead (this may be expensive for large N groups), they
1979 * will be restarted if Compatibility Mode deems that they must be due to
1983 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
1985 struct ifmultiaddr *ifma, *ifmatmp;
1987 struct in_multi *inm;
1988 struct in_multi_head inm_free_tmp;
1990 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
1991 igi->igi_ifp, igi->igi_ifp->if_xname);
1993 IN_MULTI_LIST_LOCK_ASSERT();
1997 SLIST_INIT(&inm_free_tmp);
2000 * Stop the v3 General Query Response on this link stone dead.
2001 * If fasttimo is woken up due to V_interface_timers_running,
2002 * the flag will be cleared if there are no pending link timers.
2004 igi->igi_v3_timer = 0;
2007 * Now clear the current-state and state-change report timers
2008 * for all memberships scoped to this link.
2012 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, ifmatmp) {
2013 if (ifma->ifma_addr->sa_family != AF_INET ||
2014 ifma->ifma_protospec == NULL)
2016 inm = (struct in_multi *)ifma->ifma_protospec;
2017 switch (inm->inm_state) {
2018 case IGMP_NOT_MEMBER:
2019 case IGMP_SILENT_MEMBER:
2020 case IGMP_IDLE_MEMBER:
2021 case IGMP_LAZY_MEMBER:
2022 case IGMP_SLEEPING_MEMBER:
2023 case IGMP_AWAKENING_MEMBER:
2025 * These states are either not relevant in v3 mode,
2026 * or are unreported. Do nothing.
2029 case IGMP_LEAVING_MEMBER:
2031 * If we are leaving the group and switching to
2032 * compatibility mode, we need to release the final
2033 * reference held for issuing the INCLUDE {}, and
2034 * transition to REPORTING to ensure the host leave
2035 * message is sent upstream to the old querier --
2036 * transition to NOT would lose the leave and race.
2038 inm_rele_locked(&inm_free_tmp, inm);
2040 case IGMP_G_QUERY_PENDING_MEMBER:
2041 case IGMP_SG_QUERY_PENDING_MEMBER:
2042 inm_clear_recorded(inm);
2044 case IGMP_REPORTING_MEMBER:
2045 inm->inm_state = IGMP_REPORTING_MEMBER;
2049 * Always clear state-change and group report timers.
2050 * Free any pending IGMPv3 state-change records.
2052 inm->inm_sctimer = 0;
2054 mbufq_drain(&inm->inm_scq);
2056 IF_ADDR_WUNLOCK(ifp);
2058 inm_release_list_deferred(&inm_free_tmp);
2062 * Update the Older Version Querier Present timers for a link.
2063 * See Section 7.2.1 of RFC 3376.
2066 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2071 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2073 * IGMPv1 and IGMPv2 Querier Present timers expired.
2077 if (igi->igi_version != IGMP_VERSION_3) {
2079 "%s: transition from v%d -> v%d on %p(%s)",
2080 __func__, igi->igi_version, IGMP_VERSION_3,
2081 igi->igi_ifp, igi->igi_ifp->if_xname);
2082 igi->igi_version = IGMP_VERSION_3;
2084 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2086 * IGMPv1 Querier Present timer expired,
2087 * IGMPv2 Querier Present timer running.
2088 * If IGMPv2 was disabled since last timeout,
2090 * If IGMPv2 is enabled, revert to IGMPv2.
2092 if (!V_igmp_v2enable) {
2094 "%s: transition from v%d -> v%d on %p(%s)",
2095 __func__, igi->igi_version, IGMP_VERSION_3,
2096 igi->igi_ifp, igi->igi_ifp->if_xname);
2097 igi->igi_v2_timer = 0;
2098 igi->igi_version = IGMP_VERSION_3;
2100 --igi->igi_v2_timer;
2101 if (igi->igi_version != IGMP_VERSION_2) {
2103 "%s: transition from v%d -> v%d on %p(%s)",
2104 __func__, igi->igi_version, IGMP_VERSION_2,
2105 igi->igi_ifp, igi->igi_ifp->if_xname);
2106 igi->igi_version = IGMP_VERSION_2;
2107 igmp_v3_cancel_link_timers(igi);
2110 } else if (igi->igi_v1_timer > 0) {
2112 * IGMPv1 Querier Present timer running.
2113 * Stop IGMPv2 timer if running.
2115 * If IGMPv1 was disabled since last timeout,
2117 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2119 if (!V_igmp_v1enable) {
2121 "%s: transition from v%d -> v%d on %p(%s)",
2122 __func__, igi->igi_version, IGMP_VERSION_3,
2123 igi->igi_ifp, igi->igi_ifp->if_xname);
2124 igi->igi_v1_timer = 0;
2125 igi->igi_version = IGMP_VERSION_3;
2127 --igi->igi_v1_timer;
2129 if (igi->igi_v2_timer > 0) {
2131 "%s: cancel v2 timer on %p(%s)",
2132 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2133 igi->igi_v2_timer = 0;
2139 * Global slowtimo handler.
2140 * VIMAGE: Timeout handlers are expected to service all vimages.
2145 VNET_ITERATOR_DECL(vnet_iter);
2147 VNET_LIST_RLOCK_NOSLEEP();
2148 VNET_FOREACH(vnet_iter) {
2149 CURVNET_SET(vnet_iter);
2150 igmp_slowtimo_vnet();
2153 VNET_LIST_RUNLOCK_NOSLEEP();
2157 * Per-vnet slowtimo handler.
2160 igmp_slowtimo_vnet(void)
2162 struct igmp_ifsoftc *igi;
2166 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2167 igmp_v1v2_process_querier_timers(igi);
2174 * Dispatch an IGMPv1/v2 host report or leave message.
2175 * These are always small enough to fit inside a single mbuf.
2178 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2180 struct epoch_tracker et;
2186 IN_MULTI_LIST_LOCK_ASSERT();
2191 m = m_gethdr(M_NOWAIT, MT_DATA);
2194 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2196 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2198 m->m_data += sizeof(struct ip);
2199 m->m_len = sizeof(struct igmp);
2201 igmp = mtod(m, struct igmp *);
2202 igmp->igmp_type = type;
2203 igmp->igmp_code = 0;
2204 igmp->igmp_group = inm->inm_addr;
2205 igmp->igmp_cksum = 0;
2206 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2208 m->m_data -= sizeof(struct ip);
2209 m->m_len += sizeof(struct ip);
2211 ip = mtod(m, struct ip *);
2213 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2215 ip->ip_p = IPPROTO_IGMP;
2216 ip->ip_src.s_addr = INADDR_ANY;
2218 if (type == IGMP_HOST_LEAVE_MESSAGE)
2219 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2221 ip->ip_dst = inm->inm_addr;
2223 igmp_save_context(m, ifp);
2225 m->m_flags |= M_IGMPV2;
2226 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2227 m->m_flags |= M_IGMP_LOOP;
2229 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2230 NET_EPOCH_ENTER(et);
2231 netisr_dispatch(NETISR_IGMP, m);
2238 * Process a state change from the upper layer for the given IPv4 group.
2240 * Each socket holds a reference on the in_multi in its own ip_moptions.
2241 * The socket layer will have made the necessary updates to.the group
2242 * state, it is now up to IGMP to issue a state change report if there
2243 * has been any change between T0 (when the last state-change was issued)
2246 * We use the IGMPv3 state machine at group level. The IGMP module
2247 * however makes the decision as to which IGMP protocol version to speak.
2248 * A state change *from* INCLUDE {} always means an initial join.
2249 * A state change *to* INCLUDE {} always means a final leave.
2251 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2252 * save ourselves a bunch of work; any exclusive mode groups need not
2253 * compute source filter lists.
2255 * VIMAGE: curvnet should have been set by caller, as this routine
2256 * is called from the socket option handlers.
2259 igmp_change_state(struct in_multi *inm)
2261 struct igmp_ifsoftc *igi;
2266 IN_MULTI_LOCK_ASSERT();
2268 * Try to detect if the upper layer just asked us to change state
2269 * for an interface which has now gone away.
2271 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2272 ifp = inm->inm_ifma->ifma_ifp;
2274 * Sanity check that netinet's notion of ifp is the
2277 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2281 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2282 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2285 * If we detect a state transition to or from MCAST_UNDEFINED
2286 * for this group, then we are starting or finishing an IGMP
2287 * life cycle for this group.
2289 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2290 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2291 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2292 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2293 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2294 error = igmp_initial_join(inm, igi);
2296 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2297 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2298 igmp_final_leave(inm, igi);
2302 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2305 error = igmp_handle_state_change(inm, igi);
2313 * Perform the initial join for an IGMP group.
2315 * When joining a group:
2316 * If the group should have its IGMP traffic suppressed, do nothing.
2317 * IGMPv1 starts sending IGMPv1 host membership reports.
2318 * IGMPv2 starts sending IGMPv2 host membership reports.
2319 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2320 * initial state of the membership.
2323 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2327 int error, retval, syncstates;
2329 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2330 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2337 IN_MULTI_LOCK_ASSERT();
2340 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2343 * Groups joined on loopback or marked as 'not reported',
2344 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2345 * are never reported in any IGMP protocol exchanges.
2346 * All other groups enter the appropriate IGMP state machine
2347 * for the version in use on this link.
2348 * A link marked as IGIF_SILENT causes IGMP to be completely
2349 * disabled for the link.
2351 if ((ifp->if_flags & IFF_LOOPBACK) ||
2352 (igi->igi_flags & IGIF_SILENT) ||
2353 !igmp_isgroupreported(inm->inm_addr)) {
2355 "%s: not kicking state machine for silent group", __func__);
2356 inm->inm_state = IGMP_SILENT_MEMBER;
2360 * Deal with overlapping in_multi lifecycle.
2361 * If this group was LEAVING, then make sure
2362 * we drop the reference we picked up to keep the
2363 * group around for the final INCLUDE {} enqueue.
2365 if (igi->igi_version == IGMP_VERSION_3 &&
2366 inm->inm_state == IGMP_LEAVING_MEMBER) {
2367 MPASS(inm->inm_refcount > 1);
2368 inm_rele_locked(NULL, inm);
2370 inm->inm_state = IGMP_REPORTING_MEMBER;
2372 switch (igi->igi_version) {
2373 case IGMP_VERSION_1:
2374 case IGMP_VERSION_2:
2375 inm->inm_state = IGMP_IDLE_MEMBER;
2376 error = igmp_v1v2_queue_report(inm,
2377 (igi->igi_version == IGMP_VERSION_2) ?
2378 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2379 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2381 inm->inm_timer = IGMP_RANDOM_DELAY(
2382 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2383 V_current_state_timers_running = 1;
2387 case IGMP_VERSION_3:
2389 * Defer update of T0 to T1, until the first copy
2390 * of the state change has been transmitted.
2395 * Immediately enqueue a State-Change Report for
2396 * this interface, freeing any previous reports.
2397 * Don't kick the timers if there is nothing to do,
2398 * or if an error occurred.
2402 retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2404 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2407 error = retval * -1;
2412 * Schedule transmission of pending state-change
2413 * report up to RV times for this link. The timer
2414 * will fire at the next igmp_fasttimo (~200ms),
2415 * giving us an opportunity to merge the reports.
2417 if (igi->igi_flags & IGIF_LOOPBACK) {
2420 KASSERT(igi->igi_rv > 1,
2421 ("%s: invalid robustness %d", __func__,
2423 inm->inm_scrv = igi->igi_rv;
2425 inm->inm_sctimer = 1;
2426 V_state_change_timers_running = 1;
2434 * Only update the T0 state if state change is atomic,
2435 * i.e. we don't need to wait for a timer to fire before we
2436 * can consider the state change to have been communicated.
2440 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2441 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2448 * Issue an intermediate state change during the IGMP life-cycle.
2451 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2456 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2457 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2461 IN_MULTI_LIST_LOCK_ASSERT();
2464 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2466 if ((ifp->if_flags & IFF_LOOPBACK) ||
2467 (igi->igi_flags & IGIF_SILENT) ||
2468 !igmp_isgroupreported(inm->inm_addr) ||
2469 (igi->igi_version != IGMP_VERSION_3)) {
2470 if (!igmp_isgroupreported(inm->inm_addr)) {
2472 "%s: not kicking state machine for silent group", __func__);
2474 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2476 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2477 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2481 mbufq_drain(&inm->inm_scq);
2483 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2484 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2489 * If record(s) were enqueued, start the state-change
2490 * report timer for this group.
2492 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2493 inm->inm_sctimer = 1;
2494 V_state_change_timers_running = 1;
2500 * Perform the final leave for an IGMP group.
2502 * When leaving a group:
2503 * IGMPv1 does nothing.
2504 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2505 * IGMPv3 enqueues a state-change report containing a transition
2506 * to INCLUDE {} for immediate transmission.
2509 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2515 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2516 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2517 inm->inm_ifp->if_xname);
2519 IN_MULTI_LIST_LOCK_ASSERT();
2522 switch (inm->inm_state) {
2523 case IGMP_NOT_MEMBER:
2524 case IGMP_SILENT_MEMBER:
2525 case IGMP_LEAVING_MEMBER:
2526 /* Already leaving or left; do nothing. */
2528 "%s: not kicking state machine for silent group", __func__);
2530 case IGMP_REPORTING_MEMBER:
2531 case IGMP_IDLE_MEMBER:
2532 case IGMP_G_QUERY_PENDING_MEMBER:
2533 case IGMP_SG_QUERY_PENDING_MEMBER:
2534 if (igi->igi_version == IGMP_VERSION_2) {
2536 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2537 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2538 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2541 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2542 inm->inm_state = IGMP_NOT_MEMBER;
2543 } else if (igi->igi_version == IGMP_VERSION_3) {
2545 * Stop group timer and all pending reports.
2546 * Immediately enqueue a state-change report
2547 * TO_IN {} to be sent on the next fast timeout,
2548 * giving us an opportunity to merge reports.
2550 mbufq_drain(&inm->inm_scq);
2552 if (igi->igi_flags & IGIF_LOOPBACK) {
2555 inm->inm_scrv = igi->igi_rv;
2557 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2558 "pending retransmissions.", __func__,
2559 ntohl(inm->inm_addr.s_addr),
2560 inm->inm_ifp->if_xname, inm->inm_scrv);
2561 if (inm->inm_scrv == 0) {
2562 inm->inm_state = IGMP_NOT_MEMBER;
2563 inm->inm_sctimer = 0;
2565 int retval __unused;
2567 inm_acquire_locked(inm);
2569 retval = igmp_v3_enqueue_group_record(
2570 &inm->inm_scq, inm, 1, 0, 0);
2571 KASSERT(retval != 0,
2572 ("%s: enqueue record = %d", __func__,
2575 inm->inm_state = IGMP_LEAVING_MEMBER;
2576 inm->inm_sctimer = 1;
2577 V_state_change_timers_running = 1;
2583 case IGMP_LAZY_MEMBER:
2584 case IGMP_SLEEPING_MEMBER:
2585 case IGMP_AWAKENING_MEMBER:
2586 /* Our reports are suppressed; do nothing. */
2592 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2593 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2594 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2595 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2596 __func__, ntohl(inm->inm_addr.s_addr),
2597 inm->inm_ifp->if_xname);
2602 * Enqueue an IGMPv3 group record to the given output queue.
2604 * XXX This function could do with having the allocation code
2605 * split out, and the multiple-tree-walks coalesced into a single
2606 * routine as has been done in igmp_v3_enqueue_filter_change().
2608 * If is_state_change is zero, a current-state record is appended.
2609 * If is_state_change is non-zero, a state-change report is appended.
2611 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2612 * If is_group_query is zero, and if there is a packet with free space
2613 * at the tail of the queue, it will be appended to providing there
2614 * is enough free space.
2615 * Otherwise a new mbuf packet chain is allocated.
2617 * If is_source_query is non-zero, each source is checked to see if
2618 * it was recorded for a Group-Source query, and will be omitted if
2619 * it is not both in-mode and recorded.
2621 * The function will attempt to allocate leading space in the packet
2622 * for the IP/IGMP header to be prepended without fragmenting the chain.
2624 * If successful the size of all data appended to the queue is returned,
2625 * otherwise an error code less than zero is returned, or zero if
2626 * no record(s) were appended.
2629 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2630 const int is_state_change, const int is_group_query,
2631 const int is_source_query)
2633 struct igmp_grouprec ig;
2634 struct igmp_grouprec *pig;
2636 struct ip_msource *ims, *nims;
2637 struct mbuf *m0, *m, *md;
2638 int is_filter_list_change;
2639 int minrec0len, m0srcs, msrcs, nbytes, off;
2640 int record_has_sources;
2646 IN_MULTI_LIST_LOCK_ASSERT();
2649 is_filter_list_change = 0;
2656 record_has_sources = 1;
2658 type = IGMP_DO_NOTHING;
2659 mode = inm->inm_st[1].iss_fmode;
2662 * If we did not transition out of ASM mode during t0->t1,
2663 * and there are no source nodes to process, we can skip
2664 * the generation of source records.
2666 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2668 record_has_sources = 0;
2670 if (is_state_change) {
2672 * Queue a state change record.
2673 * If the mode did not change, and there are non-ASM
2674 * listeners or source filters present,
2675 * we potentially need to issue two records for the group.
2676 * If we are transitioning to MCAST_UNDEFINED, we need
2677 * not send any sources.
2678 * If there are ASM listeners, and there was no filter
2679 * mode transition of any kind, do nothing.
2681 if (mode != inm->inm_st[0].iss_fmode) {
2682 if (mode == MCAST_EXCLUDE) {
2683 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2685 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2687 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2689 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2690 if (mode == MCAST_UNDEFINED)
2691 record_has_sources = 0;
2694 if (record_has_sources) {
2695 is_filter_list_change = 1;
2697 type = IGMP_DO_NOTHING;
2702 * Queue a current state record.
2704 if (mode == MCAST_EXCLUDE) {
2705 type = IGMP_MODE_IS_EXCLUDE;
2706 } else if (mode == MCAST_INCLUDE) {
2707 type = IGMP_MODE_IS_INCLUDE;
2708 KASSERT(inm->inm_st[1].iss_asm == 0,
2709 ("%s: inm %p is INCLUDE but ASM count is %d",
2710 __func__, inm, inm->inm_st[1].iss_asm));
2715 * Generate the filter list changes using a separate function.
2717 if (is_filter_list_change)
2718 return (igmp_v3_enqueue_filter_change(mq, inm));
2720 if (type == IGMP_DO_NOTHING) {
2721 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2722 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2727 * If any sources are present, we must be able to fit at least
2728 * one in the trailing space of the tail packet's mbuf,
2731 minrec0len = sizeof(struct igmp_grouprec);
2732 if (record_has_sources)
2733 minrec0len += sizeof(in_addr_t);
2735 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2736 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2737 inm->inm_ifp->if_xname);
2740 * Check if we have a packet in the tail of the queue for this
2741 * group into which the first group record for this group will fit.
2742 * Otherwise allocate a new packet.
2743 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2744 * Note: Group records for G/GSR query responses MUST be sent
2745 * in their own packet.
2747 m0 = mbufq_last(mq);
2748 if (!is_group_query &&
2750 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2751 (m0->m_pkthdr.len + minrec0len) <
2752 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2753 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2754 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2756 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2758 if (mbufq_full(mq)) {
2759 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2763 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2764 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2765 if (!is_state_change && !is_group_query) {
2766 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2768 m->m_data += IGMP_LEADINGSPACE;
2771 m = m_gethdr(M_NOWAIT, MT_DATA);
2773 M_ALIGN(m, IGMP_LEADINGSPACE);
2778 igmp_save_context(m, ifp);
2780 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2784 * Append group record.
2785 * If we have sources, we don't know how many yet.
2790 ig.ig_group = inm->inm_addr;
2791 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2794 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2797 nbytes += sizeof(struct igmp_grouprec);
2800 * Append as many sources as will fit in the first packet.
2801 * If we are appending to a new packet, the chain allocation
2802 * may potentially use clusters; use m_getptr() in this case.
2803 * If we are appending to an existing packet, we need to obtain
2804 * a pointer to the group record after m_append(), in case a new
2805 * mbuf was allocated.
2806 * Only append sources which are in-mode at t1. If we are
2807 * transitioning to MCAST_UNDEFINED state on the group, do not
2808 * include source entries.
2809 * Only report recorded sources in our filter set when responding
2810 * to a group-source query.
2812 if (record_has_sources) {
2815 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2816 md->m_len - nbytes);
2818 md = m_getptr(m, 0, &off);
2819 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2823 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2824 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2826 now = ims_get_mode(inm, ims, 1);
2827 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2828 if ((now != mode) ||
2829 (now == mode && mode == MCAST_UNDEFINED)) {
2830 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2833 if (is_source_query && ims->ims_stp == 0) {
2834 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2838 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2839 naddr = htonl(ims->ims_haddr);
2840 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2843 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2847 nbytes += sizeof(in_addr_t);
2849 if (msrcs == m0srcs)
2852 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2854 pig->ig_numsrc = htons(msrcs);
2855 nbytes += (msrcs * sizeof(in_addr_t));
2858 if (is_source_query && msrcs == 0) {
2859 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2866 * We are good to go with first packet.
2869 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2870 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2871 mbufq_enqueue(mq, m);
2873 m->m_pkthdr.PH_vt.vt_nrecs++;
2876 * No further work needed if no source list in packet(s).
2878 if (!record_has_sources)
2882 * Whilst sources remain to be announced, we need to allocate
2883 * a new packet and fill out as many sources as will fit.
2884 * Always try for a cluster first.
2886 while (nims != NULL) {
2887 if (mbufq_full(mq)) {
2888 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2891 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2893 m->m_data += IGMP_LEADINGSPACE;
2895 m = m_gethdr(M_NOWAIT, MT_DATA);
2897 M_ALIGN(m, IGMP_LEADINGSPACE);
2901 igmp_save_context(m, ifp);
2902 md = m_getptr(m, 0, &off);
2903 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2904 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2906 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2909 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2912 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2913 nbytes += sizeof(struct igmp_grouprec);
2915 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2916 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2919 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2920 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2922 now = ims_get_mode(inm, ims, 1);
2923 if ((now != mode) ||
2924 (now == mode && mode == MCAST_UNDEFINED)) {
2925 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2928 if (is_source_query && ims->ims_stp == 0) {
2929 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2933 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2934 naddr = htonl(ims->ims_haddr);
2935 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2938 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2943 if (msrcs == m0srcs)
2946 pig->ig_numsrc = htons(msrcs);
2947 nbytes += (msrcs * sizeof(in_addr_t));
2949 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2950 mbufq_enqueue(mq, m);
2957 * Type used to mark record pass completion.
2958 * We exploit the fact we can cast to this easily from the
2959 * current filter modes on each ip_msource node.
2962 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2963 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2964 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2965 REC_FULL = REC_ALLOW | REC_BLOCK
2969 * Enqueue an IGMPv3 filter list change to the given output queue.
2971 * Source list filter state is held in an RB-tree. When the filter list
2972 * for a group is changed without changing its mode, we need to compute
2973 * the deltas between T0 and T1 for each source in the filter set,
2974 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2976 * As we may potentially queue two record types, and the entire R-B tree
2977 * needs to be walked at once, we break this out into its own function
2978 * so we can generate a tightly packed queue of packets.
2980 * XXX This could be written to only use one tree walk, although that makes
2981 * serializing into the mbuf chains a bit harder. For now we do two walks
2982 * which makes things easier on us, and it may or may not be harder on
2985 * If successful the size of all data appended to the queue is returned,
2986 * otherwise an error code less than zero is returned, or zero if
2987 * no record(s) were appended.
2990 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
2992 static const int MINRECLEN =
2993 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
2995 struct igmp_grouprec ig;
2996 struct igmp_grouprec *pig;
2997 struct ip_msource *ims, *nims;
2998 struct mbuf *m, *m0, *md;
3000 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3002 uint8_t mode, now, then;
3003 rectype_t crt, drt, nrt;
3005 IN_MULTI_LIST_LOCK_ASSERT();
3007 if (inm->inm_nsrc == 0 ||
3008 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3011 ifp = inm->inm_ifp; /* interface */
3012 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3013 crt = REC_NONE; /* current group record type */
3014 drt = REC_NONE; /* mask of completed group record types */
3015 nrt = REC_NONE; /* record type for current node */
3016 m0srcs = 0; /* # source which will fit in current mbuf chain */
3017 nbytes = 0; /* # of bytes appended to group's state-change queue */
3018 npbytes = 0; /* # of bytes appended this packet */
3019 rsrcs = 0; /* # sources encoded in current record */
3020 schanged = 0; /* # nodes encoded in overall filter change */
3021 nallow = 0; /* # of source entries in ALLOW_NEW */
3022 nblock = 0; /* # of source entries in BLOCK_OLD */
3023 nims = NULL; /* next tree node pointer */
3026 * For each possible filter record mode.
3027 * The first kind of source we encounter tells us which
3028 * is the first kind of record we start appending.
3029 * If a node transitioned to UNDEFINED at t1, its mode is treated
3030 * as the inverse of the group's filter mode.
3032 while (drt != REC_FULL) {
3034 m0 = mbufq_last(mq);
3036 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3037 IGMP_V3_REPORT_MAXRECS) &&
3038 (m0->m_pkthdr.len + MINRECLEN) <
3039 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3041 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3042 sizeof(struct igmp_grouprec)) /
3045 "%s: use previous packet", __func__);
3047 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3049 m->m_data += IGMP_LEADINGSPACE;
3051 m = m_gethdr(M_NOWAIT, MT_DATA);
3053 M_ALIGN(m, IGMP_LEADINGSPACE);
3057 "%s: m_get*() failed", __func__);
3060 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3061 igmp_save_context(m, ifp);
3062 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3063 sizeof(struct igmp_grouprec)) /
3067 "%s: allocated new packet", __func__);
3070 * Append the IGMP group record header to the
3071 * current packet's data area.
3072 * Recalculate pointer to free space for next
3073 * group record, in case m_append() allocated
3074 * a new mbuf or cluster.
3076 memset(&ig, 0, sizeof(ig));
3077 ig.ig_group = inm->inm_addr;
3078 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3082 "%s: m_append() failed", __func__);
3085 npbytes += sizeof(struct igmp_grouprec);
3087 /* new packet; offset in c hain */
3088 md = m_getptr(m, npbytes -
3089 sizeof(struct igmp_grouprec), &off);
3090 pig = (struct igmp_grouprec *)(mtod(md,
3093 /* current packet; offset from last append */
3095 pig = (struct igmp_grouprec *)(mtod(md,
3096 uint8_t *) + md->m_len -
3097 sizeof(struct igmp_grouprec));
3100 * Begin walking the tree for this record type
3101 * pass, or continue from where we left off
3102 * previously if we had to allocate a new packet.
3103 * Only report deltas in-mode at t1.
3104 * We need not report included sources as allowed
3105 * if we are in inclusive mode on the group,
3106 * however the converse is not true.
3110 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3111 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3112 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3113 __func__, ims->ims_haddr);
3114 now = ims_get_mode(inm, ims, 1);
3115 then = ims_get_mode(inm, ims, 0);
3116 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3117 __func__, then, now);
3120 "%s: skip unchanged", __func__);
3123 if (mode == MCAST_EXCLUDE &&
3124 now == MCAST_INCLUDE) {
3126 "%s: skip IN src on EX group",
3130 nrt = (rectype_t)now;
3131 if (nrt == REC_NONE)
3132 nrt = (rectype_t)(~mode & REC_FULL);
3133 if (schanged++ == 0) {
3135 } else if (crt != nrt)
3137 naddr = htonl(ims->ims_haddr);
3138 if (!m_append(m, sizeof(in_addr_t),
3143 "%s: m_append() failed", __func__);
3146 nallow += !!(crt == REC_ALLOW);
3147 nblock += !!(crt == REC_BLOCK);
3148 if (++rsrcs == m0srcs)
3152 * If we did not append any tree nodes on this
3153 * pass, back out of allocations.
3156 npbytes -= sizeof(struct igmp_grouprec);
3159 "%s: m_free(m)", __func__);
3163 "%s: m_adj(m, -ig)", __func__);
3164 m_adj(m, -((int)sizeof(
3165 struct igmp_grouprec)));
3169 npbytes += (rsrcs * sizeof(in_addr_t));
3170 if (crt == REC_ALLOW)
3171 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3172 else if (crt == REC_BLOCK)
3173 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3174 pig->ig_numsrc = htons(rsrcs);
3176 * Count the new group record, and enqueue this
3177 * packet if it wasn't already queued.
3179 m->m_pkthdr.PH_vt.vt_nrecs++;
3181 mbufq_enqueue(mq, m);
3183 } while (nims != NULL);
3185 crt = (~crt & REC_FULL);
3188 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3195 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3198 struct mbuf *m; /* pending state-change */
3199 struct mbuf *m0; /* copy of pending state-change */
3200 struct mbuf *mt; /* last state-change in packet */
3201 int docopy, domerge;
3208 IN_MULTI_LIST_LOCK_ASSERT();
3212 * If there are further pending retransmissions, make a writable
3213 * copy of each queued state-change message before merging.
3215 if (inm->inm_scrv > 0)
3220 if (mbufq_first(gq) == NULL) {
3221 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3226 m = mbufq_first(gq);
3229 * Only merge the report into the current packet if
3230 * there is sufficient space to do so; an IGMPv3 report
3231 * packet may only contain 65,535 group records.
3232 * Always use a simple mbuf chain concatentation to do this,
3233 * as large state changes for single groups may have
3234 * allocated clusters.
3237 mt = mbufq_last(scq);
3239 recslen = m_length(m, NULL);
3241 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3242 m->m_pkthdr.PH_vt.vt_nrecs <=
3243 IGMP_V3_REPORT_MAXRECS) &&
3244 (mt->m_pkthdr.len + recslen <=
3245 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3249 if (!domerge && mbufq_full(gq)) {
3251 "%s: outbound queue full, skipping whole packet %p",
3261 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3262 m0 = mbufq_dequeue(gq);
3265 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3266 m0 = m_dup(m, M_NOWAIT);
3269 m0->m_nextpkt = NULL;
3274 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3276 mbufq_enqueue(scq, m0);
3278 struct mbuf *mtl; /* last mbuf of packet mt */
3280 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3284 m0->m_flags &= ~M_PKTHDR;
3285 mt->m_pkthdr.len += recslen;
3286 mt->m_pkthdr.PH_vt.vt_nrecs +=
3287 m0->m_pkthdr.PH_vt.vt_nrecs;
3297 * Respond to a pending IGMPv3 General Query.
3300 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3302 struct ifmultiaddr *ifma;
3304 struct in_multi *inm;
3305 int retval __unused, loop;
3307 IN_MULTI_LIST_LOCK_ASSERT();
3311 KASSERT(igi->igi_version == IGMP_VERSION_3,
3312 ("%s: called when version %d", __func__, igi->igi_version));
3315 * Check that there are some packets queued. If so, send them first.
3316 * For large number of groups the reply to general query can take
3317 * many packets, we should finish sending them before starting of
3318 * queuing the new reply.
3320 if (mbufq_len(&igi->igi_gq) != 0)
3325 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3326 if (ifma->ifma_addr->sa_family != AF_INET ||
3327 ifma->ifma_protospec == NULL)
3330 inm = (struct in_multi *)ifma->ifma_protospec;
3331 KASSERT(ifp == inm->inm_ifp,
3332 ("%s: inconsistent ifp", __func__));
3334 switch (inm->inm_state) {
3335 case IGMP_NOT_MEMBER:
3336 case IGMP_SILENT_MEMBER:
3338 case IGMP_REPORTING_MEMBER:
3339 case IGMP_IDLE_MEMBER:
3340 case IGMP_LAZY_MEMBER:
3341 case IGMP_SLEEPING_MEMBER:
3342 case IGMP_AWAKENING_MEMBER:
3343 inm->inm_state = IGMP_REPORTING_MEMBER;
3344 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3346 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3349 case IGMP_G_QUERY_PENDING_MEMBER:
3350 case IGMP_SG_QUERY_PENDING_MEMBER:
3351 case IGMP_LEAVING_MEMBER:
3357 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3358 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3361 * Slew transmission of bursts over 500ms intervals.
3363 if (mbufq_first(&igi->igi_gq) != NULL) {
3364 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3365 IGMP_RESPONSE_BURST_INTERVAL);
3366 V_interface_timers_running = 1;
3371 * Transmit the next pending IGMP message in the output queue.
3373 * We get called from netisr_processqueue(). A mutex private to igmpoq
3374 * will be acquired and released around this routine.
3376 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3377 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3378 * a link and uses a link-scope multicast address.
3381 igmp_intr(struct mbuf *m)
3383 struct ip_moptions imo;
3385 struct mbuf *ipopts, *m0;
3389 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3392 * Set VNET image pointer from enqueued mbuf chain
3393 * before doing anything else. Whilst we use interface
3394 * indexes to guard against interface detach, they are
3395 * unique to each VIMAGE and must be retrieved.
3397 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3398 ifindex = igmp_restore_context(m);
3401 * Check if the ifnet still exists. This limits the scope of
3402 * any race in the absence of a global ifp lock for low cost
3403 * (an array lookup).
3405 ifp = ifnet_byindex(ifindex);
3407 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3408 __func__, m, ifindex);
3410 IPSTAT_INC(ips_noroute);
3414 ipopts = V_igmp_sendra ? m_raopt : NULL;
3416 imo.imo_multicast_ttl = 1;
3417 imo.imo_multicast_vif = -1;
3418 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3421 * If the user requested that IGMP traffic be explicitly
3422 * redirected to the loopback interface (e.g. they are running a
3423 * MANET interface and the routing protocol needs to see the
3424 * updates), handle this now.
3426 if (m->m_flags & M_IGMP_LOOP)
3427 imo.imo_multicast_ifp = V_loif;
3429 imo.imo_multicast_ifp = ifp;
3431 if (m->m_flags & M_IGMPV2) {
3434 m0 = igmp_v3_encap_report(ifp, m);
3436 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3438 IPSTAT_INC(ips_odropped);
3443 igmp_scrub_context(m0);
3445 m0->m_pkthdr.rcvif = V_loif;
3447 mac_netinet_igmp_send(ifp, m0);
3449 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3451 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3455 IGMPSTAT_INC(igps_snd_reports);
3459 * We must restore the existing vnet pointer before
3460 * continuing as we are run from netisr context.
3466 * Encapsulate an IGMPv3 report.
3468 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3469 * chain has already had its IP/IGMPv3 header prepended. In this case
3470 * the function will not attempt to prepend; the lengths and checksums
3471 * will however be re-computed.
3473 * Returns a pointer to the new mbuf chain head, or NULL if the
3474 * allocation failed.
3476 static struct mbuf *
3477 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3479 struct rm_priotracker in_ifa_tracker;
3480 struct igmp_report *igmp;
3482 int hdrlen, igmpreclen;
3484 KASSERT((m->m_flags & M_PKTHDR),
3485 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3487 igmpreclen = m_length(m, NULL);
3488 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3490 if (m->m_flags & M_IGMPV3_HDR) {
3491 igmpreclen -= hdrlen;
3493 M_PREPEND(m, hdrlen, M_NOWAIT);
3496 m->m_flags |= M_IGMPV3_HDR;
3499 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3501 m->m_data += sizeof(struct ip);
3502 m->m_len -= sizeof(struct ip);
3504 igmp = mtod(m, struct igmp_report *);
3505 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3508 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3510 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3511 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3513 m->m_data -= sizeof(struct ip);
3514 m->m_len += sizeof(struct ip);
3516 ip = mtod(m, struct ip *);
3517 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3518 ip->ip_len = htons(hdrlen + igmpreclen);
3519 ip->ip_off = htons(IP_DF);
3520 ip->ip_p = IPPROTO_IGMP;
3523 ip->ip_src.s_addr = INADDR_ANY;
3525 if (m->m_flags & M_IGMP_LOOP) {
3526 struct in_ifaddr *ia;
3528 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3530 ip->ip_src = ia->ia_addr.sin_addr;
3533 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3540 igmp_rec_type_to_str(const int type)
3544 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3547 case IGMP_CHANGE_TO_INCLUDE_MODE:
3550 case IGMP_MODE_IS_EXCLUDE:
3553 case IGMP_MODE_IS_INCLUDE:
3556 case IGMP_ALLOW_NEW_SOURCES:
3559 case IGMP_BLOCK_OLD_SOURCES:
3571 vnet_igmp_init(const void *unused __unused)
3574 netisr_register_vnet(&igmp_nh);
3576 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3577 vnet_igmp_init, NULL);
3580 vnet_igmp_uninit(const void *unused __unused)
3583 /* This can happen when we shutdown the entire network stack. */
3584 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3586 netisr_unregister_vnet(&igmp_nh);
3588 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3589 vnet_igmp_uninit, NULL);
3593 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3595 struct igmp_ifsoftc *igi, *tigi;
3596 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3599 db_printf("usage: show igi_list <addr>\n");
3602 igi_head = (struct _igi_list *)addr;
3604 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3605 db_printf("igmp_ifsoftc %p:\n", igi);
3606 db_printf(" ifp %p\n", igi->igi_ifp);
3607 db_printf(" version %u\n", igi->igi_version);
3608 db_printf(" v1_timer %u\n", igi->igi_v1_timer);
3609 db_printf(" v2_timer %u\n", igi->igi_v2_timer);
3610 db_printf(" v3_timer %u\n", igi->igi_v3_timer);
3611 db_printf(" flags %#x\n", igi->igi_flags);
3612 db_printf(" rv %u\n", igi->igi_rv);
3613 db_printf(" qi %u\n", igi->igi_qi);
3614 db_printf(" qri %u\n", igi->igi_qri);
3615 db_printf(" uri %u\n", igi->igi_uri);
3616 /* struct mbufq igi_gq; */
3623 igmp_modevent(module_t mod, int type, void *unused __unused)
3628 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3630 m_raopt = igmp_ra_alloc();
3631 netisr_register(&igmp_nh);
3634 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3635 netisr_unregister(&igmp_nh);
3638 IGMP_LOCK_DESTROY();
3641 return (EOPNOTSUPP);
3646 static moduledata_t igmp_mod = {
3651 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);