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.flowid = ifp->if_index;
310 igmp_scrub_context(struct mbuf *m)
313 m->m_pkthdr.PH_loc.ptr = NULL;
314 m->m_pkthdr.flowid = 0;
318 * Restore context from a queued IGMP output chain.
319 * Return saved ifindex.
321 * VIMAGE: The assertion is there to make sure that we
322 * actually called CURVNET_SET() with what's in the mbuf chain.
324 static __inline uint32_t
325 igmp_restore_context(struct mbuf *m)
329 #if defined(VIMAGE) && defined(INVARIANTS)
330 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
331 ("%s: called when curvnet was not restored", __func__));
334 return (m->m_pkthdr.flowid);
338 * Retrieve or set default IGMP version.
340 * VIMAGE: Assume curvnet set by caller.
341 * SMPng: NOTE: Serialized by IGMP lock.
344 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
349 error = sysctl_wire_old_buffer(req, sizeof(int));
355 new = V_igmp_default_version;
357 error = sysctl_handle_int(oidp, &new, 0, req);
358 if (error || !req->newptr)
361 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
366 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
367 V_igmp_default_version, new);
369 V_igmp_default_version = new;
377 * Retrieve or set threshold between group-source queries in seconds.
379 * VIMAGE: Assume curvnet set by caller.
380 * SMPng: NOTE: Serialized by IGMP lock.
383 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
388 error = sysctl_wire_old_buffer(req, sizeof(int));
394 i = V_igmp_gsrdelay.tv_sec;
396 error = sysctl_handle_int(oidp, &i, 0, req);
397 if (error || !req->newptr)
400 if (i < -1 || i >= 60) {
405 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
406 V_igmp_gsrdelay.tv_sec, i);
407 V_igmp_gsrdelay.tv_sec = i;
415 * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
416 * For use by ifmcstat(8).
418 * SMPng: NOTE: Does an unlocked ifindex space read.
419 * VIMAGE: Assume curvnet set by caller. The node handler itself
420 * is not directly virtualized.
423 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
429 struct igmp_ifsoftc *igi;
434 if (req->newptr != NULL)
440 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
444 IN_MULTI_LIST_LOCK();
447 if (name[0] <= 0 || name[0] > V_if_index) {
454 ifp = ifnet_byindex(name[0]);
458 LIST_FOREACH(igi, &V_igi_head, igi_link) {
459 if (ifp == igi->igi_ifp) {
460 struct igmp_ifinfo info;
462 info.igi_version = igi->igi_version;
463 info.igi_v1_timer = igi->igi_v1_timer;
464 info.igi_v2_timer = igi->igi_v2_timer;
465 info.igi_v3_timer = igi->igi_v3_timer;
466 info.igi_flags = igi->igi_flags;
467 info.igi_rv = igi->igi_rv;
468 info.igi_qi = igi->igi_qi;
469 info.igi_qri = igi->igi_qri;
470 info.igi_uri = igi->igi_uri;
471 error = SYSCTL_OUT(req, &info, sizeof(info));
478 IN_MULTI_LIST_UNLOCK();
483 * Dispatch an entire queue of pending packet chains
485 * VIMAGE: Assumes the vnet pointer has been set.
488 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
490 struct epoch_tracker et;
494 while ((m = mbufq_dequeue(mq)) != NULL) {
495 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
497 m->m_flags |= M_IGMP_LOOP;
498 netisr_dispatch(NETISR_IGMP, m);
506 * Filter outgoing IGMP report state by group.
508 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
509 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
510 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
511 * this may break certain IGMP snooping switches which rely on the old
514 * Return zero if the given group is one for which IGMP reports
515 * should be suppressed, or non-zero if reports should be issued.
518 igmp_isgroupreported(const struct in_addr addr)
521 if (in_allhosts(addr) ||
522 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
529 * Construct a Router Alert option to use in outgoing packets.
537 m = m_get(M_WAITOK, MT_DATA);
538 p = mtod(m, struct ipoption *);
539 p->ipopt_dst.s_addr = INADDR_ANY;
540 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */
541 p->ipopt_list[1] = 0x04; /* 4 bytes long */
542 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
543 p->ipopt_list[3] = 0x00; /* pad byte */
544 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
550 * Attach IGMP when PF_INET is attached to an interface.
552 struct igmp_ifsoftc *
553 igmp_domifattach(struct ifnet *ifp)
555 struct igmp_ifsoftc *igi;
557 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
558 __func__, ifp, ifp->if_xname);
562 igi = igi_alloc_locked(ifp);
563 if (!(ifp->if_flags & IFF_MULTICAST))
564 igi->igi_flags |= IGIF_SILENT;
572 * VIMAGE: assume curvnet set by caller.
574 static struct igmp_ifsoftc *
575 igi_alloc_locked(/*const*/ struct ifnet *ifp)
577 struct igmp_ifsoftc *igi;
581 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
586 igi->igi_version = V_igmp_default_version;
588 igi->igi_rv = IGMP_RV_INIT;
589 igi->igi_qi = IGMP_QI_INIT;
590 igi->igi_qri = IGMP_QRI_INIT;
591 igi->igi_uri = IGMP_URI_INIT;
592 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
594 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
596 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
606 * NOTE: Some finalization tasks need to run before the protocol domain
607 * is detached, but also before the link layer does its cleanup.
609 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
610 * XXX This is also bitten by unlocked ifma_protospec access.
613 igmp_ifdetach(struct ifnet *ifp)
615 struct igmp_ifsoftc *igi;
616 struct ifmultiaddr *ifma, *next;
617 struct in_multi *inm;
618 struct in_multi_head inm_free_tmp;
619 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
622 SLIST_INIT(&inm_free_tmp);
625 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
626 if (igi->igi_version == IGMP_VERSION_3) {
629 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
630 if (ifma->ifma_addr->sa_family != AF_INET ||
631 ifma->ifma_protospec == NULL)
633 inm = (struct in_multi *)ifma->ifma_protospec;
634 if (inm->inm_state == IGMP_LEAVING_MEMBER)
635 inm_rele_locked(&inm_free_tmp, inm);
636 inm_clear_recorded(inm);
637 if (__predict_false(ifma_restart)) {
638 ifma_restart = false;
642 IF_ADDR_WUNLOCK(ifp);
643 inm_release_list_deferred(&inm_free_tmp);
650 * Hook for domifdetach.
653 igmp_domifdetach(struct ifnet *ifp)
656 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
657 __func__, ifp, ifp->if_xname);
660 igi_delete_locked(ifp);
665 igi_delete_locked(const struct ifnet *ifp)
667 struct igmp_ifsoftc *igi, *tigi;
669 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
670 __func__, ifp, ifp->if_xname);
674 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
675 if (igi->igi_ifp == ifp) {
677 * Free deferred General Query responses.
679 mbufq_drain(&igi->igi_gq);
681 LIST_REMOVE(igi, igi_link);
689 * Process a received IGMPv1 query.
690 * Return non-zero if the message should be dropped.
692 * VIMAGE: The curvnet pointer is derived from the input ifp.
695 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
696 const struct igmp *igmp)
698 struct ifmultiaddr *ifma;
699 struct igmp_ifsoftc *igi;
700 struct in_multi *inm;
705 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
706 * 224.0.0.1. They are always treated as General Queries.
707 * igmp_group is always ignored. Do not drop it as a userland
708 * daemon may wish to see it.
709 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
711 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
712 IGMPSTAT_INC(igps_rcv_badqueries);
715 IGMPSTAT_INC(igps_rcv_gen_queries);
717 IN_MULTI_LIST_LOCK();
720 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
721 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
723 if (igi->igi_flags & IGIF_LOOPBACK) {
724 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
730 * Switch to IGMPv1 host compatibility mode.
732 igmp_set_version(igi, IGMP_VERSION_1);
734 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
737 * Start the timers in all of our group records
738 * for the interface on which the query arrived,
739 * except those which are already running.
741 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
742 if (ifma->ifma_addr->sa_family != AF_INET ||
743 ifma->ifma_protospec == NULL)
745 inm = (struct in_multi *)ifma->ifma_protospec;
746 if (inm->inm_timer != 0)
748 switch (inm->inm_state) {
749 case IGMP_NOT_MEMBER:
750 case IGMP_SILENT_MEMBER:
752 case IGMP_G_QUERY_PENDING_MEMBER:
753 case IGMP_SG_QUERY_PENDING_MEMBER:
754 case IGMP_REPORTING_MEMBER:
755 case IGMP_IDLE_MEMBER:
756 case IGMP_LAZY_MEMBER:
757 case IGMP_SLEEPING_MEMBER:
758 case IGMP_AWAKENING_MEMBER:
759 inm->inm_state = IGMP_REPORTING_MEMBER;
760 inm->inm_timer = IGMP_RANDOM_DELAY(
761 IGMP_V1V2_MAX_RI * PR_FASTHZ);
762 V_current_state_timers_running = 1;
764 case IGMP_LEAVING_MEMBER:
771 IN_MULTI_LIST_UNLOCK();
777 * Process a received IGMPv2 general or group-specific query.
780 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
781 const struct igmp *igmp)
783 struct ifmultiaddr *ifma;
784 struct igmp_ifsoftc *igi;
785 struct in_multi *inm;
786 int is_general_query;
791 is_general_query = 0;
794 * Validate address fields upfront.
795 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
797 if (in_nullhost(igmp->igmp_group)) {
799 * IGMPv2 General Query.
800 * If this was not sent to the all-hosts group, ignore it.
802 if (!in_allhosts(ip->ip_dst))
804 IGMPSTAT_INC(igps_rcv_gen_queries);
805 is_general_query = 1;
807 /* IGMPv2 Group-Specific Query. */
808 IGMPSTAT_INC(igps_rcv_group_queries);
811 IN_MULTI_LIST_LOCK();
814 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
815 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
817 if (igi->igi_flags & IGIF_LOOPBACK) {
818 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
824 * Ignore v2 query if in v1 Compatibility Mode.
826 if (igi->igi_version == IGMP_VERSION_1)
829 igmp_set_version(igi, IGMP_VERSION_2);
831 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
835 if (is_general_query) {
837 * For each reporting group joined on this
838 * interface, kick the report timer.
840 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
842 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
843 if (ifma->ifma_addr->sa_family != AF_INET ||
844 ifma->ifma_protospec == NULL)
846 inm = (struct in_multi *)ifma->ifma_protospec;
847 igmp_v2_update_group(inm, timer);
851 * Group-specific IGMPv2 query, we need only
852 * look up the single group to process it.
854 inm = inm_lookup(ifp, igmp->igmp_group);
857 "process v2 query 0x%08x on ifp %p(%s)",
858 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
859 igmp_v2_update_group(inm, timer);
865 IN_MULTI_LIST_UNLOCK();
871 * Update the report timer on a group in response to an IGMPv2 query.
873 * If we are becoming the reporting member for this group, start the timer.
874 * If we already are the reporting member for this group, and timer is
875 * below the threshold, reset it.
877 * We may be updating the group for the first time since we switched
878 * to IGMPv3. If we are, then we must clear any recorded source lists,
879 * and transition to REPORTING state; the group timer is overloaded
880 * for group and group-source query responses.
882 * Unlike IGMPv3, the delay per group should be jittered
883 * to avoid bursts of IGMPv2 reports.
886 igmp_v2_update_group(struct in_multi *inm, const int timer)
889 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
890 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
892 IN_MULTI_LIST_LOCK_ASSERT();
894 switch (inm->inm_state) {
895 case IGMP_NOT_MEMBER:
896 case IGMP_SILENT_MEMBER:
898 case IGMP_REPORTING_MEMBER:
899 if (inm->inm_timer != 0 &&
900 inm->inm_timer <= timer) {
901 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
902 "skipping.", __func__);
906 case IGMP_SG_QUERY_PENDING_MEMBER:
907 case IGMP_G_QUERY_PENDING_MEMBER:
908 case IGMP_IDLE_MEMBER:
909 case IGMP_LAZY_MEMBER:
910 case IGMP_AWAKENING_MEMBER:
911 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
912 inm->inm_state = IGMP_REPORTING_MEMBER;
913 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
914 V_current_state_timers_running = 1;
916 case IGMP_SLEEPING_MEMBER:
917 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
918 inm->inm_state = IGMP_AWAKENING_MEMBER;
920 case IGMP_LEAVING_MEMBER:
926 * Process a received IGMPv3 general, group-specific or
927 * group-and-source-specific query.
928 * Assumes m has already been pulled up to the full IGMP message length.
929 * Return 0 if successful, otherwise an appropriate error code is returned.
932 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
933 /*const*/ struct igmpv3 *igmpv3)
935 struct igmp_ifsoftc *igi;
936 struct in_multi *inm;
937 int is_general_query;
938 uint32_t maxresp, nsrc, qqi;
942 is_general_query = 0;
944 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
946 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
947 if (maxresp >= 128) {
948 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
949 (IGMP_EXP(igmpv3->igmp_code) + 3);
953 * Robustness must never be less than 2 for on-wire IGMPv3.
954 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
955 * an exception for interfaces whose IGMPv3 state changes
956 * are redirected to loopback (e.g. MANET).
958 qrv = IGMP_QRV(igmpv3->igmp_misc);
960 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
965 qqi = igmpv3->igmp_qqi;
967 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
968 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
971 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
975 nsrc = ntohs(igmpv3->igmp_numsrc);
978 * Validate address fields and versions upfront before
979 * accepting v3 query.
980 * XXX SMPng: Unlocked access to igmpstat counters here.
982 if (in_nullhost(igmpv3->igmp_group)) {
984 * IGMPv3 General Query.
986 * General Queries SHOULD be directed to 224.0.0.1.
987 * A general query with a source list has undefined
988 * behaviour; discard it.
990 IGMPSTAT_INC(igps_rcv_gen_queries);
991 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
992 IGMPSTAT_INC(igps_rcv_badqueries);
995 is_general_query = 1;
997 /* Group or group-source specific query. */
999 IGMPSTAT_INC(igps_rcv_group_queries);
1001 IGMPSTAT_INC(igps_rcv_gsr_queries);
1004 IN_MULTI_LIST_LOCK();
1007 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1008 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1010 if (igi->igi_flags & IGIF_LOOPBACK) {
1011 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1012 ifp, ifp->if_xname);
1017 * Discard the v3 query if we're in Compatibility Mode.
1018 * The RFC is not obviously worded that hosts need to stay in
1019 * compatibility mode until the Old Version Querier Present
1022 if (igi->igi_version != IGMP_VERSION_3) {
1023 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1024 igi->igi_version, ifp, ifp->if_xname);
1028 igmp_set_version(igi, IGMP_VERSION_3);
1031 igi->igi_qri = maxresp;
1033 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1036 if (is_general_query) {
1038 * Schedule a current-state report on this ifp for
1039 * all groups, possibly containing source lists.
1040 * If there is a pending General Query response
1041 * scheduled earlier than the selected delay, do
1042 * not schedule any other reports.
1043 * Otherwise, reset the interface timer.
1045 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1046 ifp, ifp->if_xname);
1047 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1048 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1049 V_interface_timers_running = 1;
1053 * Group-source-specific queries are throttled on
1054 * a per-group basis to defeat denial-of-service attempts.
1055 * Queries for groups we are not a member of on this
1056 * link are simply ignored.
1058 inm = inm_lookup(ifp, igmpv3->igmp_group);
1062 if (!ratecheck(&inm->inm_lastgsrtv,
1063 &V_igmp_gsrdelay)) {
1064 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1066 IGMPSTAT_INC(igps_drop_gsr_queries);
1070 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1071 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1073 * If there is a pending General Query response
1074 * scheduled sooner than the selected delay, no
1075 * further report need be scheduled.
1076 * Otherwise, prepare to respond to the
1077 * group-specific or group-and-source query.
1079 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1080 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1085 IN_MULTI_LIST_UNLOCK();
1091 * Process a received IGMPv3 group-specific or group-and-source-specific
1093 * Return <0 if any error occurred. Currently this is ignored.
1096 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1097 int timer, /*const*/ struct igmpv3 *igmpv3)
1102 IN_MULTI_LIST_LOCK_ASSERT();
1107 switch (inm->inm_state) {
1108 case IGMP_NOT_MEMBER:
1109 case IGMP_SILENT_MEMBER:
1110 case IGMP_SLEEPING_MEMBER:
1111 case IGMP_LAZY_MEMBER:
1112 case IGMP_AWAKENING_MEMBER:
1113 case IGMP_IDLE_MEMBER:
1114 case IGMP_LEAVING_MEMBER:
1117 case IGMP_REPORTING_MEMBER:
1118 case IGMP_G_QUERY_PENDING_MEMBER:
1119 case IGMP_SG_QUERY_PENDING_MEMBER:
1123 nsrc = ntohs(igmpv3->igmp_numsrc);
1126 * Deal with group-specific queries upfront.
1127 * If any group query is already pending, purge any recorded
1128 * source-list state if it exists, and schedule a query response
1129 * for this group-specific query.
1132 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1133 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1134 inm_clear_recorded(inm);
1135 timer = min(inm->inm_timer, timer);
1137 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1138 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1139 V_current_state_timers_running = 1;
1144 * Deal with the case where a group-and-source-specific query has
1145 * been received but a group-specific query is already pending.
1147 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1148 timer = min(inm->inm_timer, timer);
1149 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1150 V_current_state_timers_running = 1;
1155 * Finally, deal with the case where a group-and-source-specific
1156 * query has been received, where a response to a previous g-s-r
1157 * query exists, or none exists.
1158 * In this case, we need to parse the source-list which the Querier
1159 * has provided us with and check if we have any source list filter
1160 * entries at T1 for these sources. If we do not, there is no need
1161 * schedule a report and the query may be dropped.
1162 * If we do, we must record them and schedule a current-state
1163 * report for those sources.
1164 * FIXME: Handling source lists larger than 1 mbuf requires that
1165 * we pass the mbuf chain pointer down to this function, and use
1166 * m_getptr() to walk the chain.
1168 if (inm->inm_nsrc > 0) {
1169 const struct in_addr *ap;
1172 ap = (const struct in_addr *)(igmpv3 + 1);
1174 for (i = 0; i < nsrc; i++, ap++) {
1175 retval = inm_record_source(inm, ap->s_addr);
1178 nrecorded += retval;
1180 if (nrecorded > 0) {
1182 "%s: schedule response to SG query", __func__);
1183 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1184 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1185 V_current_state_timers_running = 1;
1193 * Process a received IGMPv1 host membership report.
1195 * NOTE: 0.0.0.0 workaround breaks const correctness.
1198 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1199 /*const*/ struct igmp *igmp)
1201 struct rm_priotracker in_ifa_tracker;
1202 struct in_ifaddr *ia;
1203 struct in_multi *inm;
1205 IGMPSTAT_INC(igps_rcv_reports);
1207 if (ifp->if_flags & IFF_LOOPBACK)
1210 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1211 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1212 IGMPSTAT_INC(igps_rcv_badreports);
1217 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1218 * Booting clients may use the source address 0.0.0.0. Some
1219 * IGMP daemons may not know how to use IP_RECVIF to determine
1220 * the interface upon which this message was received.
1221 * Replace 0.0.0.0 with the subnet address if told to do so.
1223 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1224 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1226 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1229 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1230 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1233 * IGMPv1 report suppression.
1234 * If we are a member of this group, and our membership should be
1235 * reported, stop our group timer and transition to the 'lazy' state.
1237 IN_MULTI_LIST_LOCK();
1238 inm = inm_lookup(ifp, igmp->igmp_group);
1240 struct igmp_ifsoftc *igi;
1244 KASSERT(igi != NULL,
1245 ("%s: no igi for ifp %p", __func__, ifp));
1249 IGMPSTAT_INC(igps_rcv_ourreports);
1252 * If we are in IGMPv3 host mode, do not allow the
1253 * other host's IGMPv1 report to suppress our reports
1254 * unless explicitly configured to do so.
1256 if (igi->igi_version == IGMP_VERSION_3) {
1257 if (V_igmp_legacysupp)
1258 igmp_v3_suppress_group_record(inm);
1264 switch (inm->inm_state) {
1265 case IGMP_NOT_MEMBER:
1266 case IGMP_SILENT_MEMBER:
1268 case IGMP_IDLE_MEMBER:
1269 case IGMP_LAZY_MEMBER:
1270 case IGMP_AWAKENING_MEMBER:
1272 "report suppressed for 0x%08x on ifp %p(%s)",
1273 ntohl(igmp->igmp_group.s_addr), ifp,
1275 case IGMP_SLEEPING_MEMBER:
1276 inm->inm_state = IGMP_SLEEPING_MEMBER;
1278 case IGMP_REPORTING_MEMBER:
1280 "report suppressed for 0x%08x on ifp %p(%s)",
1281 ntohl(igmp->igmp_group.s_addr), ifp,
1283 if (igi->igi_version == IGMP_VERSION_1)
1284 inm->inm_state = IGMP_LAZY_MEMBER;
1285 else if (igi->igi_version == IGMP_VERSION_2)
1286 inm->inm_state = IGMP_SLEEPING_MEMBER;
1288 case IGMP_G_QUERY_PENDING_MEMBER:
1289 case IGMP_SG_QUERY_PENDING_MEMBER:
1290 case IGMP_LEAVING_MEMBER:
1296 IN_MULTI_LIST_UNLOCK();
1302 * Process a received IGMPv2 host membership report.
1304 * NOTE: 0.0.0.0 workaround breaks const correctness.
1307 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1308 /*const*/ struct igmp *igmp)
1310 struct rm_priotracker in_ifa_tracker;
1311 struct in_ifaddr *ia;
1312 struct in_multi *inm;
1315 * Make sure we don't hear our own membership report. Fast
1316 * leave requires knowing that we are the only member of a
1319 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1320 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1324 IGMPSTAT_INC(igps_rcv_reports);
1326 if (ifp->if_flags & IFF_LOOPBACK) {
1330 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1331 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1332 IGMPSTAT_INC(igps_rcv_badreports);
1337 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1338 * Booting clients may use the source address 0.0.0.0. Some
1339 * IGMP daemons may not know how to use IP_RECVIF to determine
1340 * the interface upon which this message was received.
1341 * Replace 0.0.0.0 with the subnet address if told to do so.
1343 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1345 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1348 CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1349 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1352 * IGMPv2 report suppression.
1353 * If we are a member of this group, and our membership should be
1354 * reported, and our group timer is pending or about to be reset,
1355 * stop our group timer by transitioning to the 'lazy' state.
1357 IN_MULTI_LIST_LOCK();
1358 inm = inm_lookup(ifp, igmp->igmp_group);
1360 struct igmp_ifsoftc *igi;
1363 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1365 IGMPSTAT_INC(igps_rcv_ourreports);
1368 * If we are in IGMPv3 host mode, do not allow the
1369 * other host's IGMPv1 report to suppress our reports
1370 * unless explicitly configured to do so.
1372 if (igi->igi_version == IGMP_VERSION_3) {
1373 if (V_igmp_legacysupp)
1374 igmp_v3_suppress_group_record(inm);
1380 switch (inm->inm_state) {
1381 case IGMP_NOT_MEMBER:
1382 case IGMP_SILENT_MEMBER:
1383 case IGMP_SLEEPING_MEMBER:
1385 case IGMP_REPORTING_MEMBER:
1386 case IGMP_IDLE_MEMBER:
1387 case IGMP_AWAKENING_MEMBER:
1389 "report suppressed for 0x%08x on ifp %p(%s)",
1390 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1391 case IGMP_LAZY_MEMBER:
1392 inm->inm_state = IGMP_LAZY_MEMBER;
1394 case IGMP_G_QUERY_PENDING_MEMBER:
1395 case IGMP_SG_QUERY_PENDING_MEMBER:
1396 case IGMP_LEAVING_MEMBER:
1402 IN_MULTI_LIST_UNLOCK();
1408 igmp_input(struct mbuf **mp, int *offp, int proto)
1419 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1422 ifp = m->m_pkthdr.rcvif;
1425 IGMPSTAT_INC(igps_rcv_total);
1427 ip = mtod(m, struct ip *);
1429 igmplen = ntohs(ip->ip_len) - iphlen;
1434 if (igmplen < IGMP_MINLEN) {
1435 IGMPSTAT_INC(igps_rcv_tooshort);
1437 return (IPPROTO_DONE);
1441 * Always pullup to the minimum size for v1/v2 or v3
1442 * to amortize calls to m_pullup().
1445 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1446 minlen += IGMP_V3_QUERY_MINLEN;
1448 minlen += IGMP_MINLEN;
1449 if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1450 (m = m_pullup(m, minlen)) == NULL) {
1451 IGMPSTAT_INC(igps_rcv_tooshort);
1452 return (IPPROTO_DONE);
1454 ip = mtod(m, struct ip *);
1457 * Validate checksum.
1459 m->m_data += iphlen;
1461 igmp = mtod(m, struct igmp *);
1462 if (in_cksum(m, igmplen)) {
1463 IGMPSTAT_INC(igps_rcv_badsum);
1465 return (IPPROTO_DONE);
1467 m->m_data -= iphlen;
1471 * IGMP control traffic is link-scope, and must have a TTL of 1.
1472 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1473 * probe packets may come from beyond the LAN.
1475 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1476 IGMPSTAT_INC(igps_rcv_badttl);
1478 return (IPPROTO_DONE);
1481 switch (igmp->igmp_type) {
1482 case IGMP_HOST_MEMBERSHIP_QUERY:
1483 if (igmplen == IGMP_MINLEN) {
1484 if (igmp->igmp_code == 0)
1485 queryver = IGMP_VERSION_1;
1487 queryver = IGMP_VERSION_2;
1488 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1489 queryver = IGMP_VERSION_3;
1491 IGMPSTAT_INC(igps_rcv_tooshort);
1493 return (IPPROTO_DONE);
1497 case IGMP_VERSION_1:
1498 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1499 if (!V_igmp_v1enable)
1501 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1503 return (IPPROTO_DONE);
1507 case IGMP_VERSION_2:
1508 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1509 if (!V_igmp_v2enable)
1511 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1513 return (IPPROTO_DONE);
1517 case IGMP_VERSION_3: {
1518 struct igmpv3 *igmpv3;
1522 IGMPSTAT_INC(igps_rcv_v3_queries);
1523 igmpv3 = (struct igmpv3 *)igmp;
1525 * Validate length based on source count.
1527 nsrc = ntohs(igmpv3->igmp_numsrc);
1528 if (nsrc * sizeof(in_addr_t) >
1529 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1530 IGMPSTAT_INC(igps_rcv_tooshort);
1531 return (IPPROTO_DONE);
1534 * m_pullup() may modify m, so pullup in
1537 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1538 sizeof(struct in_addr) * nsrc;
1539 if ((!M_WRITABLE(m) ||
1540 m->m_len < igmpv3len) &&
1541 (m = m_pullup(m, igmpv3len)) == NULL) {
1542 IGMPSTAT_INC(igps_rcv_tooshort);
1543 return (IPPROTO_DONE);
1545 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1547 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1549 return (IPPROTO_DONE);
1556 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1557 if (!V_igmp_v1enable)
1559 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1561 return (IPPROTO_DONE);
1565 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1566 if (!V_igmp_v2enable)
1568 if (!ip_checkrouteralert(m))
1569 IGMPSTAT_INC(igps_rcv_nora);
1570 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1572 return (IPPROTO_DONE);
1576 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1578 * Hosts do not need to process IGMPv3 membership reports,
1579 * as report suppression is no longer required.
1581 if (!ip_checkrouteralert(m))
1582 IGMPSTAT_INC(igps_rcv_nora);
1590 * Pass all valid IGMP packets up to any process(es) listening on a
1594 return (rip_input(mp, offp, proto));
1599 * Fast timeout handler (global).
1600 * VIMAGE: Timeout handlers are expected to service all vimages.
1605 VNET_ITERATOR_DECL(vnet_iter);
1607 VNET_LIST_RLOCK_NOSLEEP();
1608 VNET_FOREACH(vnet_iter) {
1609 CURVNET_SET(vnet_iter);
1610 igmp_fasttimo_vnet();
1613 VNET_LIST_RUNLOCK_NOSLEEP();
1617 * Fast timeout handler (per-vnet).
1618 * Sends are shuffled off to a netisr to deal with Giant.
1620 * VIMAGE: Assume caller has set up our curvnet.
1623 igmp_fasttimo_vnet(void)
1625 struct mbufq scq; /* State-change packets */
1626 struct mbufq qrq; /* Query response packets */
1628 struct igmp_ifsoftc *igi;
1629 struct ifmultiaddr *ifma, *next;
1630 struct in_multi *inm;
1631 struct in_multi_head inm_free_tmp;
1632 int loop, uri_fasthz;
1638 * Quick check to see if any work needs to be done, in order to
1639 * minimize the overhead of fasttimo processing.
1640 * SMPng: XXX Unlocked reads.
1642 if (!V_current_state_timers_running &&
1643 !V_interface_timers_running &&
1644 !V_state_change_timers_running)
1647 SLIST_INIT(&inm_free_tmp);
1648 IN_MULTI_LIST_LOCK();
1652 * IGMPv3 General Query response timer processing.
1654 if (V_interface_timers_running) {
1655 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1657 V_interface_timers_running = 0;
1658 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1659 if (igi->igi_v3_timer == 0) {
1661 } else if (--igi->igi_v3_timer == 0) {
1662 igmp_v3_dispatch_general_query(igi);
1664 V_interface_timers_running = 1;
1669 if (!V_current_state_timers_running &&
1670 !V_state_change_timers_running)
1673 V_current_state_timers_running = 0;
1674 V_state_change_timers_running = 0;
1676 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1679 * IGMPv1/v2/v3 host report and state-change timer processing.
1680 * Note: Processing a v3 group timer may remove a node.
1682 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1685 if (igi->igi_version == IGMP_VERSION_3) {
1686 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1687 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1689 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1690 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1695 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1696 if (ifma->ifma_addr->sa_family != AF_INET ||
1697 ifma->ifma_protospec == NULL)
1699 inm = (struct in_multi *)ifma->ifma_protospec;
1700 switch (igi->igi_version) {
1701 case IGMP_VERSION_1:
1702 case IGMP_VERSION_2:
1703 igmp_v1v2_process_group_timer(inm,
1706 case IGMP_VERSION_3:
1707 igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1708 &scq, inm, uri_fasthz);
1711 if (__predict_false(ifma_restart)) {
1712 ifma_restart = false;
1716 IF_ADDR_WUNLOCK(ifp);
1718 if (igi->igi_version == IGMP_VERSION_3) {
1719 igmp_dispatch_queue(&qrq, 0, loop);
1720 igmp_dispatch_queue(&scq, 0, loop);
1723 * Free the in_multi reference(s) for this
1726 inm_release_list_deferred(&inm_free_tmp);
1732 IN_MULTI_LIST_UNLOCK();
1736 * Update host report group timer for IGMPv1/v2.
1737 * Will update the global pending timer flags.
1740 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1742 int report_timer_expired;
1744 IN_MULTI_LIST_LOCK_ASSERT();
1747 if (inm->inm_timer == 0) {
1748 report_timer_expired = 0;
1749 } else if (--inm->inm_timer == 0) {
1750 report_timer_expired = 1;
1752 V_current_state_timers_running = 1;
1756 switch (inm->inm_state) {
1757 case IGMP_NOT_MEMBER:
1758 case IGMP_SILENT_MEMBER:
1759 case IGMP_IDLE_MEMBER:
1760 case IGMP_LAZY_MEMBER:
1761 case IGMP_SLEEPING_MEMBER:
1762 case IGMP_AWAKENING_MEMBER:
1764 case IGMP_REPORTING_MEMBER:
1765 if (report_timer_expired) {
1766 inm->inm_state = IGMP_IDLE_MEMBER;
1767 (void)igmp_v1v2_queue_report(inm,
1768 (version == IGMP_VERSION_2) ?
1769 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1770 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1773 case IGMP_G_QUERY_PENDING_MEMBER:
1774 case IGMP_SG_QUERY_PENDING_MEMBER:
1775 case IGMP_LEAVING_MEMBER:
1781 * Update a group's timers for IGMPv3.
1782 * Will update the global pending timer flags.
1783 * Note: Unlocked read from igi.
1786 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1787 struct mbufq *qrq, struct mbufq *scq,
1788 struct in_multi *inm, const int uri_fasthz)
1790 int query_response_timer_expired;
1791 int state_change_retransmit_timer_expired;
1793 IN_MULTI_LIST_LOCK_ASSERT();
1796 query_response_timer_expired = 0;
1797 state_change_retransmit_timer_expired = 0;
1800 * During a transition from v1/v2 compatibility mode back to v3,
1801 * a group record in REPORTING state may still have its group
1802 * timer active. This is a no-op in this function; it is easier
1803 * to deal with it here than to complicate the slow-timeout path.
1805 if (inm->inm_timer == 0) {
1806 query_response_timer_expired = 0;
1807 } else if (--inm->inm_timer == 0) {
1808 query_response_timer_expired = 1;
1810 V_current_state_timers_running = 1;
1813 if (inm->inm_sctimer == 0) {
1814 state_change_retransmit_timer_expired = 0;
1815 } else if (--inm->inm_sctimer == 0) {
1816 state_change_retransmit_timer_expired = 1;
1818 V_state_change_timers_running = 1;
1821 /* We are in fasttimo, so be quick about it. */
1822 if (!state_change_retransmit_timer_expired &&
1823 !query_response_timer_expired)
1826 switch (inm->inm_state) {
1827 case IGMP_NOT_MEMBER:
1828 case IGMP_SILENT_MEMBER:
1829 case IGMP_SLEEPING_MEMBER:
1830 case IGMP_LAZY_MEMBER:
1831 case IGMP_AWAKENING_MEMBER:
1832 case IGMP_IDLE_MEMBER:
1834 case IGMP_G_QUERY_PENDING_MEMBER:
1835 case IGMP_SG_QUERY_PENDING_MEMBER:
1837 * Respond to a previously pending Group-Specific
1838 * or Group-and-Source-Specific query by enqueueing
1839 * the appropriate Current-State report for
1840 * immediate transmission.
1842 if (query_response_timer_expired) {
1843 int retval __unused;
1845 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1846 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1847 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1849 inm->inm_state = IGMP_REPORTING_MEMBER;
1850 /* XXX Clear recorded sources for next time. */
1851 inm_clear_recorded(inm);
1854 case IGMP_REPORTING_MEMBER:
1855 case IGMP_LEAVING_MEMBER:
1856 if (state_change_retransmit_timer_expired) {
1858 * State-change retransmission timer fired.
1859 * If there are any further pending retransmissions,
1860 * set the global pending state-change flag, and
1863 if (--inm->inm_scrv > 0) {
1864 inm->inm_sctimer = uri_fasthz;
1865 V_state_change_timers_running = 1;
1868 * Retransmit the previously computed state-change
1869 * report. If there are no further pending
1870 * retransmissions, the mbuf queue will be consumed.
1871 * Update T0 state to T1 as we have now sent
1874 (void)igmp_v3_merge_state_changes(inm, scq);
1877 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1878 ntohl(inm->inm_addr.s_addr),
1879 inm->inm_ifp->if_xname);
1882 * If we are leaving the group for good, make sure
1883 * we release IGMP's reference to it.
1884 * This release must be deferred using a SLIST,
1885 * as we are called from a loop which traverses
1886 * the in_ifmultiaddr TAILQ.
1888 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1889 inm->inm_scrv == 0) {
1890 inm->inm_state = IGMP_NOT_MEMBER;
1891 inm_rele_locked(inmh, inm);
1900 * Suppress a group's pending response to a group or source/group query.
1902 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1903 * Do NOT update ST1/ST0 as this operation merely suppresses
1904 * the currently pending group record.
1905 * Do NOT suppress the response to a general query. It is possible but
1906 * it would require adding another state or flag.
1909 igmp_v3_suppress_group_record(struct in_multi *inm)
1912 IN_MULTI_LIST_LOCK_ASSERT();
1914 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1915 ("%s: not IGMPv3 mode on link", __func__));
1917 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1918 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1921 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1922 inm_clear_recorded(inm);
1925 inm->inm_state = IGMP_REPORTING_MEMBER;
1929 * Switch to a different IGMP version on the given interface,
1930 * as per Section 7.2.1.
1933 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1935 int old_version_timer;
1939 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1940 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1942 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1944 * Compute the "Older Version Querier Present" timer as per
1947 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1948 old_version_timer *= PR_SLOWHZ;
1950 if (version == IGMP_VERSION_1) {
1951 igi->igi_v1_timer = old_version_timer;
1952 igi->igi_v2_timer = 0;
1953 } else if (version == IGMP_VERSION_2) {
1954 igi->igi_v1_timer = 0;
1955 igi->igi_v2_timer = old_version_timer;
1959 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1960 if (igi->igi_version != IGMP_VERSION_2) {
1961 igi->igi_version = IGMP_VERSION_2;
1962 igmp_v3_cancel_link_timers(igi);
1964 } else if (igi->igi_v1_timer > 0) {
1965 if (igi->igi_version != IGMP_VERSION_1) {
1966 igi->igi_version = IGMP_VERSION_1;
1967 igmp_v3_cancel_link_timers(igi);
1973 * Cancel pending IGMPv3 timers for the given link and all groups
1974 * joined on it; state-change, general-query, and group-query timers.
1976 * Only ever called on a transition from v3 to Compatibility mode. Kill
1977 * the timers stone dead (this may be expensive for large N groups), they
1978 * will be restarted if Compatibility Mode deems that they must be due to
1982 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
1984 struct ifmultiaddr *ifma, *ifmatmp;
1986 struct in_multi *inm;
1987 struct in_multi_head inm_free_tmp;
1989 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
1990 igi->igi_ifp, igi->igi_ifp->if_xname);
1992 IN_MULTI_LIST_LOCK_ASSERT();
1996 SLIST_INIT(&inm_free_tmp);
1999 * Stop the v3 General Query Response on this link stone dead.
2000 * If fasttimo is woken up due to V_interface_timers_running,
2001 * the flag will be cleared if there are no pending link timers.
2003 igi->igi_v3_timer = 0;
2006 * Now clear the current-state and state-change report timers
2007 * for all memberships scoped to this link.
2011 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, ifmatmp) {
2012 if (ifma->ifma_addr->sa_family != AF_INET ||
2013 ifma->ifma_protospec == NULL)
2015 inm = (struct in_multi *)ifma->ifma_protospec;
2016 switch (inm->inm_state) {
2017 case IGMP_NOT_MEMBER:
2018 case IGMP_SILENT_MEMBER:
2019 case IGMP_IDLE_MEMBER:
2020 case IGMP_LAZY_MEMBER:
2021 case IGMP_SLEEPING_MEMBER:
2022 case IGMP_AWAKENING_MEMBER:
2024 * These states are either not relevant in v3 mode,
2025 * or are unreported. Do nothing.
2028 case IGMP_LEAVING_MEMBER:
2030 * If we are leaving the group and switching to
2031 * compatibility mode, we need to release the final
2032 * reference held for issuing the INCLUDE {}, and
2033 * transition to REPORTING to ensure the host leave
2034 * message is sent upstream to the old querier --
2035 * transition to NOT would lose the leave and race.
2037 inm_rele_locked(&inm_free_tmp, inm);
2039 case IGMP_G_QUERY_PENDING_MEMBER:
2040 case IGMP_SG_QUERY_PENDING_MEMBER:
2041 inm_clear_recorded(inm);
2043 case IGMP_REPORTING_MEMBER:
2044 inm->inm_state = IGMP_REPORTING_MEMBER;
2048 * Always clear state-change and group report timers.
2049 * Free any pending IGMPv3 state-change records.
2051 inm->inm_sctimer = 0;
2053 mbufq_drain(&inm->inm_scq);
2055 IF_ADDR_WUNLOCK(ifp);
2057 inm_release_list_deferred(&inm_free_tmp);
2061 * Update the Older Version Querier Present timers for a link.
2062 * See Section 7.2.1 of RFC 3376.
2065 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2070 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2072 * IGMPv1 and IGMPv2 Querier Present timers expired.
2076 if (igi->igi_version != IGMP_VERSION_3) {
2078 "%s: transition from v%d -> v%d on %p(%s)",
2079 __func__, igi->igi_version, IGMP_VERSION_3,
2080 igi->igi_ifp, igi->igi_ifp->if_xname);
2081 igi->igi_version = IGMP_VERSION_3;
2083 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2085 * IGMPv1 Querier Present timer expired,
2086 * IGMPv2 Querier Present timer running.
2087 * If IGMPv2 was disabled since last timeout,
2089 * If IGMPv2 is enabled, revert to IGMPv2.
2091 if (!V_igmp_v2enable) {
2093 "%s: transition from v%d -> v%d on %p(%s)",
2094 __func__, igi->igi_version, IGMP_VERSION_3,
2095 igi->igi_ifp, igi->igi_ifp->if_xname);
2096 igi->igi_v2_timer = 0;
2097 igi->igi_version = IGMP_VERSION_3;
2099 --igi->igi_v2_timer;
2100 if (igi->igi_version != IGMP_VERSION_2) {
2102 "%s: transition from v%d -> v%d on %p(%s)",
2103 __func__, igi->igi_version, IGMP_VERSION_2,
2104 igi->igi_ifp, igi->igi_ifp->if_xname);
2105 igi->igi_version = IGMP_VERSION_2;
2106 igmp_v3_cancel_link_timers(igi);
2109 } else if (igi->igi_v1_timer > 0) {
2111 * IGMPv1 Querier Present timer running.
2112 * Stop IGMPv2 timer if running.
2114 * If IGMPv1 was disabled since last timeout,
2116 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2118 if (!V_igmp_v1enable) {
2120 "%s: transition from v%d -> v%d on %p(%s)",
2121 __func__, igi->igi_version, IGMP_VERSION_3,
2122 igi->igi_ifp, igi->igi_ifp->if_xname);
2123 igi->igi_v1_timer = 0;
2124 igi->igi_version = IGMP_VERSION_3;
2126 --igi->igi_v1_timer;
2128 if (igi->igi_v2_timer > 0) {
2130 "%s: cancel v2 timer on %p(%s)",
2131 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2132 igi->igi_v2_timer = 0;
2138 * Global slowtimo handler.
2139 * VIMAGE: Timeout handlers are expected to service all vimages.
2144 VNET_ITERATOR_DECL(vnet_iter);
2146 VNET_LIST_RLOCK_NOSLEEP();
2147 VNET_FOREACH(vnet_iter) {
2148 CURVNET_SET(vnet_iter);
2149 igmp_slowtimo_vnet();
2152 VNET_LIST_RUNLOCK_NOSLEEP();
2156 * Per-vnet slowtimo handler.
2159 igmp_slowtimo_vnet(void)
2161 struct igmp_ifsoftc *igi;
2165 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2166 igmp_v1v2_process_querier_timers(igi);
2173 * Dispatch an IGMPv1/v2 host report or leave message.
2174 * These are always small enough to fit inside a single mbuf.
2177 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2179 struct epoch_tracker et;
2185 IN_MULTI_LIST_LOCK_ASSERT();
2190 m = m_gethdr(M_NOWAIT, MT_DATA);
2193 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2195 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2197 m->m_data += sizeof(struct ip);
2198 m->m_len = sizeof(struct igmp);
2200 igmp = mtod(m, struct igmp *);
2201 igmp->igmp_type = type;
2202 igmp->igmp_code = 0;
2203 igmp->igmp_group = inm->inm_addr;
2204 igmp->igmp_cksum = 0;
2205 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2207 m->m_data -= sizeof(struct ip);
2208 m->m_len += sizeof(struct ip);
2210 ip = mtod(m, struct ip *);
2212 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2214 ip->ip_p = IPPROTO_IGMP;
2215 ip->ip_src.s_addr = INADDR_ANY;
2217 if (type == IGMP_HOST_LEAVE_MESSAGE)
2218 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2220 ip->ip_dst = inm->inm_addr;
2222 igmp_save_context(m, ifp);
2224 m->m_flags |= M_IGMPV2;
2225 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2226 m->m_flags |= M_IGMP_LOOP;
2228 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2229 NET_EPOCH_ENTER(et);
2230 netisr_dispatch(NETISR_IGMP, m);
2237 * Process a state change from the upper layer for the given IPv4 group.
2239 * Each socket holds a reference on the in_multi in its own ip_moptions.
2240 * The socket layer will have made the necessary updates to.the group
2241 * state, it is now up to IGMP to issue a state change report if there
2242 * has been any change between T0 (when the last state-change was issued)
2245 * We use the IGMPv3 state machine at group level. The IGMP module
2246 * however makes the decision as to which IGMP protocol version to speak.
2247 * A state change *from* INCLUDE {} always means an initial join.
2248 * A state change *to* INCLUDE {} always means a final leave.
2250 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2251 * save ourselves a bunch of work; any exclusive mode groups need not
2252 * compute source filter lists.
2254 * VIMAGE: curvnet should have been set by caller, as this routine
2255 * is called from the socket option handlers.
2258 igmp_change_state(struct in_multi *inm)
2260 struct igmp_ifsoftc *igi;
2265 IN_MULTI_LOCK_ASSERT();
2267 * Try to detect if the upper layer just asked us to change state
2268 * for an interface which has now gone away.
2270 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2271 ifp = inm->inm_ifma->ifma_ifp;
2273 * Sanity check that netinet's notion of ifp is the
2276 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2280 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2281 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2284 * If we detect a state transition to or from MCAST_UNDEFINED
2285 * for this group, then we are starting or finishing an IGMP
2286 * life cycle for this group.
2288 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2289 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2290 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2291 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2292 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2293 error = igmp_initial_join(inm, igi);
2295 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2296 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2297 igmp_final_leave(inm, igi);
2301 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2304 error = igmp_handle_state_change(inm, igi);
2312 * Perform the initial join for an IGMP group.
2314 * When joining a group:
2315 * If the group should have its IGMP traffic suppressed, do nothing.
2316 * IGMPv1 starts sending IGMPv1 host membership reports.
2317 * IGMPv2 starts sending IGMPv2 host membership reports.
2318 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2319 * initial state of the membership.
2322 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2326 int error, retval, syncstates;
2328 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2329 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2336 IN_MULTI_LOCK_ASSERT();
2339 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2342 * Groups joined on loopback or marked as 'not reported',
2343 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2344 * are never reported in any IGMP protocol exchanges.
2345 * All other groups enter the appropriate IGMP state machine
2346 * for the version in use on this link.
2347 * A link marked as IGIF_SILENT causes IGMP to be completely
2348 * disabled for the link.
2350 if ((ifp->if_flags & IFF_LOOPBACK) ||
2351 (igi->igi_flags & IGIF_SILENT) ||
2352 !igmp_isgroupreported(inm->inm_addr)) {
2354 "%s: not kicking state machine for silent group", __func__);
2355 inm->inm_state = IGMP_SILENT_MEMBER;
2359 * Deal with overlapping in_multi lifecycle.
2360 * If this group was LEAVING, then make sure
2361 * we drop the reference we picked up to keep the
2362 * group around for the final INCLUDE {} enqueue.
2364 if (igi->igi_version == IGMP_VERSION_3 &&
2365 inm->inm_state == IGMP_LEAVING_MEMBER) {
2366 MPASS(inm->inm_refcount > 1);
2367 inm_rele_locked(NULL, inm);
2369 inm->inm_state = IGMP_REPORTING_MEMBER;
2371 switch (igi->igi_version) {
2372 case IGMP_VERSION_1:
2373 case IGMP_VERSION_2:
2374 inm->inm_state = IGMP_IDLE_MEMBER;
2375 error = igmp_v1v2_queue_report(inm,
2376 (igi->igi_version == IGMP_VERSION_2) ?
2377 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2378 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2380 inm->inm_timer = IGMP_RANDOM_DELAY(
2381 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2382 V_current_state_timers_running = 1;
2386 case IGMP_VERSION_3:
2388 * Defer update of T0 to T1, until the first copy
2389 * of the state change has been transmitted.
2394 * Immediately enqueue a State-Change Report for
2395 * this interface, freeing any previous reports.
2396 * Don't kick the timers if there is nothing to do,
2397 * or if an error occurred.
2401 retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2403 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2406 error = retval * -1;
2411 * Schedule transmission of pending state-change
2412 * report up to RV times for this link. The timer
2413 * will fire at the next igmp_fasttimo (~200ms),
2414 * giving us an opportunity to merge the reports.
2416 if (igi->igi_flags & IGIF_LOOPBACK) {
2419 KASSERT(igi->igi_rv > 1,
2420 ("%s: invalid robustness %d", __func__,
2422 inm->inm_scrv = igi->igi_rv;
2424 inm->inm_sctimer = 1;
2425 V_state_change_timers_running = 1;
2433 * Only update the T0 state if state change is atomic,
2434 * i.e. we don't need to wait for a timer to fire before we
2435 * can consider the state change to have been communicated.
2439 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2440 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2447 * Issue an intermediate state change during the IGMP life-cycle.
2450 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2455 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2456 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2460 IN_MULTI_LIST_LOCK_ASSERT();
2463 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2465 if ((ifp->if_flags & IFF_LOOPBACK) ||
2466 (igi->igi_flags & IGIF_SILENT) ||
2467 !igmp_isgroupreported(inm->inm_addr) ||
2468 (igi->igi_version != IGMP_VERSION_3)) {
2469 if (!igmp_isgroupreported(inm->inm_addr)) {
2471 "%s: not kicking state machine for silent group", __func__);
2473 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2475 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2476 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2480 mbufq_drain(&inm->inm_scq);
2482 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2483 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2488 * If record(s) were enqueued, start the state-change
2489 * report timer for this group.
2491 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2492 inm->inm_sctimer = 1;
2493 V_state_change_timers_running = 1;
2499 * Perform the final leave for an IGMP group.
2501 * When leaving a group:
2502 * IGMPv1 does nothing.
2503 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2504 * IGMPv3 enqueues a state-change report containing a transition
2505 * to INCLUDE {} for immediate transmission.
2508 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2514 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2515 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2516 inm->inm_ifp->if_xname);
2518 IN_MULTI_LIST_LOCK_ASSERT();
2521 switch (inm->inm_state) {
2522 case IGMP_NOT_MEMBER:
2523 case IGMP_SILENT_MEMBER:
2524 case IGMP_LEAVING_MEMBER:
2525 /* Already leaving or left; do nothing. */
2527 "%s: not kicking state machine for silent group", __func__);
2529 case IGMP_REPORTING_MEMBER:
2530 case IGMP_IDLE_MEMBER:
2531 case IGMP_G_QUERY_PENDING_MEMBER:
2532 case IGMP_SG_QUERY_PENDING_MEMBER:
2533 if (igi->igi_version == IGMP_VERSION_2) {
2535 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2536 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2537 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2540 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2541 inm->inm_state = IGMP_NOT_MEMBER;
2542 } else if (igi->igi_version == IGMP_VERSION_3) {
2544 * Stop group timer and all pending reports.
2545 * Immediately enqueue a state-change report
2546 * TO_IN {} to be sent on the next fast timeout,
2547 * giving us an opportunity to merge reports.
2549 mbufq_drain(&inm->inm_scq);
2551 if (igi->igi_flags & IGIF_LOOPBACK) {
2554 inm->inm_scrv = igi->igi_rv;
2556 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2557 "pending retransmissions.", __func__,
2558 ntohl(inm->inm_addr.s_addr),
2559 inm->inm_ifp->if_xname, inm->inm_scrv);
2560 if (inm->inm_scrv == 0) {
2561 inm->inm_state = IGMP_NOT_MEMBER;
2562 inm->inm_sctimer = 0;
2564 int retval __unused;
2566 inm_acquire_locked(inm);
2568 retval = igmp_v3_enqueue_group_record(
2569 &inm->inm_scq, inm, 1, 0, 0);
2570 KASSERT(retval != 0,
2571 ("%s: enqueue record = %d", __func__,
2574 inm->inm_state = IGMP_LEAVING_MEMBER;
2575 inm->inm_sctimer = 1;
2576 V_state_change_timers_running = 1;
2582 case IGMP_LAZY_MEMBER:
2583 case IGMP_SLEEPING_MEMBER:
2584 case IGMP_AWAKENING_MEMBER:
2585 /* Our reports are suppressed; do nothing. */
2591 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2592 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2593 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2594 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2595 __func__, ntohl(inm->inm_addr.s_addr),
2596 inm->inm_ifp->if_xname);
2601 * Enqueue an IGMPv3 group record to the given output queue.
2603 * XXX This function could do with having the allocation code
2604 * split out, and the multiple-tree-walks coalesced into a single
2605 * routine as has been done in igmp_v3_enqueue_filter_change().
2607 * If is_state_change is zero, a current-state record is appended.
2608 * If is_state_change is non-zero, a state-change report is appended.
2610 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2611 * If is_group_query is zero, and if there is a packet with free space
2612 * at the tail of the queue, it will be appended to providing there
2613 * is enough free space.
2614 * Otherwise a new mbuf packet chain is allocated.
2616 * If is_source_query is non-zero, each source is checked to see if
2617 * it was recorded for a Group-Source query, and will be omitted if
2618 * it is not both in-mode and recorded.
2620 * The function will attempt to allocate leading space in the packet
2621 * for the IP/IGMP header to be prepended without fragmenting the chain.
2623 * If successful the size of all data appended to the queue is returned,
2624 * otherwise an error code less than zero is returned, or zero if
2625 * no record(s) were appended.
2628 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2629 const int is_state_change, const int is_group_query,
2630 const int is_source_query)
2632 struct igmp_grouprec ig;
2633 struct igmp_grouprec *pig;
2635 struct ip_msource *ims, *nims;
2636 struct mbuf *m0, *m, *md;
2637 int is_filter_list_change;
2638 int minrec0len, m0srcs, msrcs, nbytes, off;
2639 int record_has_sources;
2645 IN_MULTI_LIST_LOCK_ASSERT();
2648 is_filter_list_change = 0;
2655 record_has_sources = 1;
2657 type = IGMP_DO_NOTHING;
2658 mode = inm->inm_st[1].iss_fmode;
2661 * If we did not transition out of ASM mode during t0->t1,
2662 * and there are no source nodes to process, we can skip
2663 * the generation of source records.
2665 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2667 record_has_sources = 0;
2669 if (is_state_change) {
2671 * Queue a state change record.
2672 * If the mode did not change, and there are non-ASM
2673 * listeners or source filters present,
2674 * we potentially need to issue two records for the group.
2675 * If we are transitioning to MCAST_UNDEFINED, we need
2676 * not send any sources.
2677 * If there are ASM listeners, and there was no filter
2678 * mode transition of any kind, do nothing.
2680 if (mode != inm->inm_st[0].iss_fmode) {
2681 if (mode == MCAST_EXCLUDE) {
2682 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2684 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2686 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2688 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2689 if (mode == MCAST_UNDEFINED)
2690 record_has_sources = 0;
2693 if (record_has_sources) {
2694 is_filter_list_change = 1;
2696 type = IGMP_DO_NOTHING;
2701 * Queue a current state record.
2703 if (mode == MCAST_EXCLUDE) {
2704 type = IGMP_MODE_IS_EXCLUDE;
2705 } else if (mode == MCAST_INCLUDE) {
2706 type = IGMP_MODE_IS_INCLUDE;
2707 KASSERT(inm->inm_st[1].iss_asm == 0,
2708 ("%s: inm %p is INCLUDE but ASM count is %d",
2709 __func__, inm, inm->inm_st[1].iss_asm));
2714 * Generate the filter list changes using a separate function.
2716 if (is_filter_list_change)
2717 return (igmp_v3_enqueue_filter_change(mq, inm));
2719 if (type == IGMP_DO_NOTHING) {
2720 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2721 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2726 * If any sources are present, we must be able to fit at least
2727 * one in the trailing space of the tail packet's mbuf,
2730 minrec0len = sizeof(struct igmp_grouprec);
2731 if (record_has_sources)
2732 minrec0len += sizeof(in_addr_t);
2734 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2735 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2736 inm->inm_ifp->if_xname);
2739 * Check if we have a packet in the tail of the queue for this
2740 * group into which the first group record for this group will fit.
2741 * Otherwise allocate a new packet.
2742 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2743 * Note: Group records for G/GSR query responses MUST be sent
2744 * in their own packet.
2746 m0 = mbufq_last(mq);
2747 if (!is_group_query &&
2749 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2750 (m0->m_pkthdr.len + minrec0len) <
2751 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2752 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2753 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2755 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2757 if (mbufq_full(mq)) {
2758 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2762 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2763 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2764 if (!is_state_change && !is_group_query) {
2765 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2767 m->m_data += IGMP_LEADINGSPACE;
2770 m = m_gethdr(M_NOWAIT, MT_DATA);
2772 M_ALIGN(m, IGMP_LEADINGSPACE);
2777 igmp_save_context(m, ifp);
2779 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2783 * Append group record.
2784 * If we have sources, we don't know how many yet.
2789 ig.ig_group = inm->inm_addr;
2790 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2793 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2796 nbytes += sizeof(struct igmp_grouprec);
2799 * Append as many sources as will fit in the first packet.
2800 * If we are appending to a new packet, the chain allocation
2801 * may potentially use clusters; use m_getptr() in this case.
2802 * If we are appending to an existing packet, we need to obtain
2803 * a pointer to the group record after m_append(), in case a new
2804 * mbuf was allocated.
2805 * Only append sources which are in-mode at t1. If we are
2806 * transitioning to MCAST_UNDEFINED state on the group, do not
2807 * include source entries.
2808 * Only report recorded sources in our filter set when responding
2809 * to a group-source query.
2811 if (record_has_sources) {
2814 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2815 md->m_len - nbytes);
2817 md = m_getptr(m, 0, &off);
2818 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2822 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2823 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2825 now = ims_get_mode(inm, ims, 1);
2826 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2827 if ((now != mode) ||
2828 (now == mode && mode == MCAST_UNDEFINED)) {
2829 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2832 if (is_source_query && ims->ims_stp == 0) {
2833 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2837 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2838 naddr = htonl(ims->ims_haddr);
2839 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2842 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2846 nbytes += sizeof(in_addr_t);
2848 if (msrcs == m0srcs)
2851 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2853 pig->ig_numsrc = htons(msrcs);
2854 nbytes += (msrcs * sizeof(in_addr_t));
2857 if (is_source_query && msrcs == 0) {
2858 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2865 * We are good to go with first packet.
2868 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2869 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2870 mbufq_enqueue(mq, m);
2872 m->m_pkthdr.PH_vt.vt_nrecs++;
2875 * No further work needed if no source list in packet(s).
2877 if (!record_has_sources)
2881 * Whilst sources remain to be announced, we need to allocate
2882 * a new packet and fill out as many sources as will fit.
2883 * Always try for a cluster first.
2885 while (nims != NULL) {
2886 if (mbufq_full(mq)) {
2887 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2890 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2892 m->m_data += IGMP_LEADINGSPACE;
2894 m = m_gethdr(M_NOWAIT, MT_DATA);
2896 M_ALIGN(m, IGMP_LEADINGSPACE);
2900 igmp_save_context(m, ifp);
2901 md = m_getptr(m, 0, &off);
2902 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2903 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2905 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2908 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2911 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2912 nbytes += sizeof(struct igmp_grouprec);
2914 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2915 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2918 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2919 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2921 now = ims_get_mode(inm, ims, 1);
2922 if ((now != mode) ||
2923 (now == mode && mode == MCAST_UNDEFINED)) {
2924 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2927 if (is_source_query && ims->ims_stp == 0) {
2928 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2932 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2933 naddr = htonl(ims->ims_haddr);
2934 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2937 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2942 if (msrcs == m0srcs)
2945 pig->ig_numsrc = htons(msrcs);
2946 nbytes += (msrcs * sizeof(in_addr_t));
2948 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2949 mbufq_enqueue(mq, m);
2956 * Type used to mark record pass completion.
2957 * We exploit the fact we can cast to this easily from the
2958 * current filter modes on each ip_msource node.
2961 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2962 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2963 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2964 REC_FULL = REC_ALLOW | REC_BLOCK
2968 * Enqueue an IGMPv3 filter list change to the given output queue.
2970 * Source list filter state is held in an RB-tree. When the filter list
2971 * for a group is changed without changing its mode, we need to compute
2972 * the deltas between T0 and T1 for each source in the filter set,
2973 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2975 * As we may potentially queue two record types, and the entire R-B tree
2976 * needs to be walked at once, we break this out into its own function
2977 * so we can generate a tightly packed queue of packets.
2979 * XXX This could be written to only use one tree walk, although that makes
2980 * serializing into the mbuf chains a bit harder. For now we do two walks
2981 * which makes things easier on us, and it may or may not be harder on
2984 * If successful the size of all data appended to the queue is returned,
2985 * otherwise an error code less than zero is returned, or zero if
2986 * no record(s) were appended.
2989 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
2991 static const int MINRECLEN =
2992 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
2994 struct igmp_grouprec ig;
2995 struct igmp_grouprec *pig;
2996 struct ip_msource *ims, *nims;
2997 struct mbuf *m, *m0, *md;
2999 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3001 uint8_t mode, now, then;
3002 rectype_t crt, drt, nrt;
3004 IN_MULTI_LIST_LOCK_ASSERT();
3006 if (inm->inm_nsrc == 0 ||
3007 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3010 ifp = inm->inm_ifp; /* interface */
3011 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3012 crt = REC_NONE; /* current group record type */
3013 drt = REC_NONE; /* mask of completed group record types */
3014 nrt = REC_NONE; /* record type for current node */
3015 m0srcs = 0; /* # source which will fit in current mbuf chain */
3016 nbytes = 0; /* # of bytes appended to group's state-change queue */
3017 npbytes = 0; /* # of bytes appended this packet */
3018 rsrcs = 0; /* # sources encoded in current record */
3019 schanged = 0; /* # nodes encoded in overall filter change */
3020 nallow = 0; /* # of source entries in ALLOW_NEW */
3021 nblock = 0; /* # of source entries in BLOCK_OLD */
3022 nims = NULL; /* next tree node pointer */
3025 * For each possible filter record mode.
3026 * The first kind of source we encounter tells us which
3027 * is the first kind of record we start appending.
3028 * If a node transitioned to UNDEFINED at t1, its mode is treated
3029 * as the inverse of the group's filter mode.
3031 while (drt != REC_FULL) {
3033 m0 = mbufq_last(mq);
3035 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3036 IGMP_V3_REPORT_MAXRECS) &&
3037 (m0->m_pkthdr.len + MINRECLEN) <
3038 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3040 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3041 sizeof(struct igmp_grouprec)) /
3044 "%s: use previous packet", __func__);
3046 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3048 m->m_data += IGMP_LEADINGSPACE;
3050 m = m_gethdr(M_NOWAIT, MT_DATA);
3052 M_ALIGN(m, IGMP_LEADINGSPACE);
3056 "%s: m_get*() failed", __func__);
3059 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3060 igmp_save_context(m, ifp);
3061 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3062 sizeof(struct igmp_grouprec)) /
3066 "%s: allocated new packet", __func__);
3069 * Append the IGMP group record header to the
3070 * current packet's data area.
3071 * Recalculate pointer to free space for next
3072 * group record, in case m_append() allocated
3073 * a new mbuf or cluster.
3075 memset(&ig, 0, sizeof(ig));
3076 ig.ig_group = inm->inm_addr;
3077 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3081 "%s: m_append() failed", __func__);
3084 npbytes += sizeof(struct igmp_grouprec);
3086 /* new packet; offset in c hain */
3087 md = m_getptr(m, npbytes -
3088 sizeof(struct igmp_grouprec), &off);
3089 pig = (struct igmp_grouprec *)(mtod(md,
3092 /* current packet; offset from last append */
3094 pig = (struct igmp_grouprec *)(mtod(md,
3095 uint8_t *) + md->m_len -
3096 sizeof(struct igmp_grouprec));
3099 * Begin walking the tree for this record type
3100 * pass, or continue from where we left off
3101 * previously if we had to allocate a new packet.
3102 * Only report deltas in-mode at t1.
3103 * We need not report included sources as allowed
3104 * if we are in inclusive mode on the group,
3105 * however the converse is not true.
3109 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3110 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3111 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3112 __func__, ims->ims_haddr);
3113 now = ims_get_mode(inm, ims, 1);
3114 then = ims_get_mode(inm, ims, 0);
3115 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3116 __func__, then, now);
3119 "%s: skip unchanged", __func__);
3122 if (mode == MCAST_EXCLUDE &&
3123 now == MCAST_INCLUDE) {
3125 "%s: skip IN src on EX group",
3129 nrt = (rectype_t)now;
3130 if (nrt == REC_NONE)
3131 nrt = (rectype_t)(~mode & REC_FULL);
3132 if (schanged++ == 0) {
3134 } else if (crt != nrt)
3136 naddr = htonl(ims->ims_haddr);
3137 if (!m_append(m, sizeof(in_addr_t),
3142 "%s: m_append() failed", __func__);
3145 nallow += !!(crt == REC_ALLOW);
3146 nblock += !!(crt == REC_BLOCK);
3147 if (++rsrcs == m0srcs)
3151 * If we did not append any tree nodes on this
3152 * pass, back out of allocations.
3155 npbytes -= sizeof(struct igmp_grouprec);
3158 "%s: m_free(m)", __func__);
3162 "%s: m_adj(m, -ig)", __func__);
3163 m_adj(m, -((int)sizeof(
3164 struct igmp_grouprec)));
3168 npbytes += (rsrcs * sizeof(in_addr_t));
3169 if (crt == REC_ALLOW)
3170 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3171 else if (crt == REC_BLOCK)
3172 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3173 pig->ig_numsrc = htons(rsrcs);
3175 * Count the new group record, and enqueue this
3176 * packet if it wasn't already queued.
3178 m->m_pkthdr.PH_vt.vt_nrecs++;
3180 mbufq_enqueue(mq, m);
3182 } while (nims != NULL);
3184 crt = (~crt & REC_FULL);
3187 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3194 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3197 struct mbuf *m; /* pending state-change */
3198 struct mbuf *m0; /* copy of pending state-change */
3199 struct mbuf *mt; /* last state-change in packet */
3200 int docopy, domerge;
3207 IN_MULTI_LIST_LOCK_ASSERT();
3211 * If there are further pending retransmissions, make a writable
3212 * copy of each queued state-change message before merging.
3214 if (inm->inm_scrv > 0)
3219 if (mbufq_first(gq) == NULL) {
3220 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3225 m = mbufq_first(gq);
3228 * Only merge the report into the current packet if
3229 * there is sufficient space to do so; an IGMPv3 report
3230 * packet may only contain 65,535 group records.
3231 * Always use a simple mbuf chain concatentation to do this,
3232 * as large state changes for single groups may have
3233 * allocated clusters.
3236 mt = mbufq_last(scq);
3238 recslen = m_length(m, NULL);
3240 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3241 m->m_pkthdr.PH_vt.vt_nrecs <=
3242 IGMP_V3_REPORT_MAXRECS) &&
3243 (mt->m_pkthdr.len + recslen <=
3244 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3248 if (!domerge && mbufq_full(gq)) {
3250 "%s: outbound queue full, skipping whole packet %p",
3260 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3261 m0 = mbufq_dequeue(gq);
3264 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3265 m0 = m_dup(m, M_NOWAIT);
3268 m0->m_nextpkt = NULL;
3273 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3275 mbufq_enqueue(scq, m0);
3277 struct mbuf *mtl; /* last mbuf of packet mt */
3279 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3283 m0->m_flags &= ~M_PKTHDR;
3284 mt->m_pkthdr.len += recslen;
3285 mt->m_pkthdr.PH_vt.vt_nrecs +=
3286 m0->m_pkthdr.PH_vt.vt_nrecs;
3296 * Respond to a pending IGMPv3 General Query.
3299 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3301 struct ifmultiaddr *ifma;
3303 struct in_multi *inm;
3304 int retval __unused, loop;
3306 IN_MULTI_LIST_LOCK_ASSERT();
3310 KASSERT(igi->igi_version == IGMP_VERSION_3,
3311 ("%s: called when version %d", __func__, igi->igi_version));
3314 * Check that there are some packets queued. If so, send them first.
3315 * For large number of groups the reply to general query can take
3316 * many packets, we should finish sending them before starting of
3317 * queuing the new reply.
3319 if (mbufq_len(&igi->igi_gq) != 0)
3324 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3325 if (ifma->ifma_addr->sa_family != AF_INET ||
3326 ifma->ifma_protospec == NULL)
3329 inm = (struct in_multi *)ifma->ifma_protospec;
3330 KASSERT(ifp == inm->inm_ifp,
3331 ("%s: inconsistent ifp", __func__));
3333 switch (inm->inm_state) {
3334 case IGMP_NOT_MEMBER:
3335 case IGMP_SILENT_MEMBER:
3337 case IGMP_REPORTING_MEMBER:
3338 case IGMP_IDLE_MEMBER:
3339 case IGMP_LAZY_MEMBER:
3340 case IGMP_SLEEPING_MEMBER:
3341 case IGMP_AWAKENING_MEMBER:
3342 inm->inm_state = IGMP_REPORTING_MEMBER;
3343 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3345 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3348 case IGMP_G_QUERY_PENDING_MEMBER:
3349 case IGMP_SG_QUERY_PENDING_MEMBER:
3350 case IGMP_LEAVING_MEMBER:
3356 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3357 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3360 * Slew transmission of bursts over 500ms intervals.
3362 if (mbufq_first(&igi->igi_gq) != NULL) {
3363 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3364 IGMP_RESPONSE_BURST_INTERVAL);
3365 V_interface_timers_running = 1;
3370 * Transmit the next pending IGMP message in the output queue.
3372 * We get called from netisr_processqueue(). A mutex private to igmpoq
3373 * will be acquired and released around this routine.
3375 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3376 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3377 * a link and uses a link-scope multicast address.
3380 igmp_intr(struct mbuf *m)
3382 struct ip_moptions imo;
3384 struct mbuf *ipopts, *m0;
3388 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3391 * Set VNET image pointer from enqueued mbuf chain
3392 * before doing anything else. Whilst we use interface
3393 * indexes to guard against interface detach, they are
3394 * unique to each VIMAGE and must be retrieved.
3396 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3397 ifindex = igmp_restore_context(m);
3400 * Check if the ifnet still exists. This limits the scope of
3401 * any race in the absence of a global ifp lock for low cost
3402 * (an array lookup).
3404 ifp = ifnet_byindex(ifindex);
3406 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3407 __func__, m, ifindex);
3409 IPSTAT_INC(ips_noroute);
3413 ipopts = V_igmp_sendra ? m_raopt : NULL;
3415 imo.imo_multicast_ttl = 1;
3416 imo.imo_multicast_vif = -1;
3417 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3420 * If the user requested that IGMP traffic be explicitly
3421 * redirected to the loopback interface (e.g. they are running a
3422 * MANET interface and the routing protocol needs to see the
3423 * updates), handle this now.
3425 if (m->m_flags & M_IGMP_LOOP)
3426 imo.imo_multicast_ifp = V_loif;
3428 imo.imo_multicast_ifp = ifp;
3430 if (m->m_flags & M_IGMPV2) {
3433 m0 = igmp_v3_encap_report(ifp, m);
3435 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3437 IPSTAT_INC(ips_odropped);
3442 igmp_scrub_context(m0);
3444 m0->m_pkthdr.rcvif = V_loif;
3446 mac_netinet_igmp_send(ifp, m0);
3448 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3450 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3454 IGMPSTAT_INC(igps_snd_reports);
3458 * We must restore the existing vnet pointer before
3459 * continuing as we are run from netisr context.
3465 * Encapsulate an IGMPv3 report.
3467 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3468 * chain has already had its IP/IGMPv3 header prepended. In this case
3469 * the function will not attempt to prepend; the lengths and checksums
3470 * will however be re-computed.
3472 * Returns a pointer to the new mbuf chain head, or NULL if the
3473 * allocation failed.
3475 static struct mbuf *
3476 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3478 struct rm_priotracker in_ifa_tracker;
3479 struct igmp_report *igmp;
3481 int hdrlen, igmpreclen;
3483 KASSERT((m->m_flags & M_PKTHDR),
3484 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3486 igmpreclen = m_length(m, NULL);
3487 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3489 if (m->m_flags & M_IGMPV3_HDR) {
3490 igmpreclen -= hdrlen;
3492 M_PREPEND(m, hdrlen, M_NOWAIT);
3495 m->m_flags |= M_IGMPV3_HDR;
3498 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3500 m->m_data += sizeof(struct ip);
3501 m->m_len -= sizeof(struct ip);
3503 igmp = mtod(m, struct igmp_report *);
3504 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3507 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3509 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3510 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3512 m->m_data -= sizeof(struct ip);
3513 m->m_len += sizeof(struct ip);
3515 ip = mtod(m, struct ip *);
3516 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3517 ip->ip_len = htons(hdrlen + igmpreclen);
3518 ip->ip_off = htons(IP_DF);
3519 ip->ip_p = IPPROTO_IGMP;
3522 ip->ip_src.s_addr = INADDR_ANY;
3524 if (m->m_flags & M_IGMP_LOOP) {
3525 struct in_ifaddr *ia;
3527 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3529 ip->ip_src = ia->ia_addr.sin_addr;
3532 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3539 igmp_rec_type_to_str(const int type)
3543 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3546 case IGMP_CHANGE_TO_INCLUDE_MODE:
3549 case IGMP_MODE_IS_EXCLUDE:
3552 case IGMP_MODE_IS_INCLUDE:
3555 case IGMP_ALLOW_NEW_SOURCES:
3558 case IGMP_BLOCK_OLD_SOURCES:
3570 vnet_igmp_init(const void *unused __unused)
3573 netisr_register_vnet(&igmp_nh);
3575 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3576 vnet_igmp_init, NULL);
3579 vnet_igmp_uninit(const void *unused __unused)
3582 /* This can happen when we shutdown the entire network stack. */
3583 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3585 netisr_unregister_vnet(&igmp_nh);
3587 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3588 vnet_igmp_uninit, NULL);
3592 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3594 struct igmp_ifsoftc *igi, *tigi;
3595 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3598 db_printf("usage: show igi_list <addr>\n");
3601 igi_head = (struct _igi_list *)addr;
3603 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3604 db_printf("igmp_ifsoftc %p:\n", igi);
3605 db_printf(" ifp %p\n", igi->igi_ifp);
3606 db_printf(" version %u\n", igi->igi_version);
3607 db_printf(" v1_timer %u\n", igi->igi_v1_timer);
3608 db_printf(" v2_timer %u\n", igi->igi_v2_timer);
3609 db_printf(" v3_timer %u\n", igi->igi_v3_timer);
3610 db_printf(" flags %#x\n", igi->igi_flags);
3611 db_printf(" rv %u\n", igi->igi_rv);
3612 db_printf(" qi %u\n", igi->igi_qi);
3613 db_printf(" qri %u\n", igi->igi_qri);
3614 db_printf(" uri %u\n", igi->igi_uri);
3615 /* struct mbufq igi_gq; */
3622 igmp_modevent(module_t mod, int type, void *unused __unused)
3627 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3629 m_raopt = igmp_ra_alloc();
3630 netisr_register(&igmp_nh);
3633 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3634 netisr_unregister(&igmp_nh);
3637 IGMP_LOCK_DESTROY();
3640 return (EOPNOTSUPP);
3645 static moduledata_t igmp_mod = {
3650 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);