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/kernel.h>
65 #include <sys/sysctl.h>
67 #include <sys/condvar.h>
74 #include <net/if_var.h>
75 #include <net/netisr.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_var.h>
83 #include <netinet/ip_options.h>
84 #include <netinet/igmp.h>
85 #include <netinet/igmp_var.h>
87 #include <machine/in_cksum.h>
89 #include <security/mac/mac_framework.h>
92 #define KTR_IGMPV3 KTR_INET
95 #define IGMP_SLOWHZ 2 /* 2 slow timeouts per second */
96 #define IGMP_FASTHZ 5 /* 5 fast timeouts per second */
97 #define IGMP_RESPONSE_BURST_INTERVAL (IGMP_FASTHZ / 2)
99 static struct igmp_ifsoftc *
100 igi_alloc_locked(struct ifnet *);
101 static void igi_delete_locked(const struct ifnet *);
102 static void igmp_dispatch_queue(struct mbufq *, int, const int);
103 static void igmp_fasttimo_vnet(void);
104 static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *);
105 static int igmp_handle_state_change(struct in_multi *,
106 struct igmp_ifsoftc *);
107 static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *);
108 static int igmp_input_v1_query(struct ifnet *, const struct ip *,
109 const struct igmp *);
110 static int igmp_input_v2_query(struct ifnet *, const struct ip *,
111 const struct igmp *);
112 static int igmp_input_v3_query(struct ifnet *, const struct ip *,
113 /*const*/ struct igmpv3 *);
114 static int igmp_input_v3_group_query(struct in_multi *,
115 struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *);
116 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
117 /*const*/ struct igmp *);
118 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
119 /*const*/ struct igmp *);
120 static void igmp_intr(struct mbuf *);
121 static int igmp_isgroupreported(const struct in_addr);
125 static char * igmp_rec_type_to_str(const int);
127 static void igmp_set_version(struct igmp_ifsoftc *, const int);
128 static void igmp_slowtimo_vnet(void);
129 static int igmp_v1v2_queue_report(struct in_multi *, const int);
130 static void igmp_v1v2_process_group_timer(struct in_multi *, const int);
131 static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *);
132 static void igmp_v2_update_group(struct in_multi *, const int);
133 static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *);
134 static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *);
136 igmp_v3_encap_report(struct ifnet *, struct mbuf *);
137 static int igmp_v3_enqueue_group_record(struct mbufq *,
138 struct in_multi *, const int, const int, const int);
139 static int igmp_v3_enqueue_filter_change(struct mbufq *,
141 static void igmp_v3_process_group_timers(struct in_multi_head *,
142 struct mbufq *, struct mbufq *, struct in_multi *,
144 static int igmp_v3_merge_state_changes(struct in_multi *,
146 static void igmp_v3_suppress_group_record(struct in_multi *);
147 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
148 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
149 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
150 static int sysctl_igmp_stat(SYSCTL_HANDLER_ARGS);
152 static const struct netisr_handler igmp_nh = {
154 .nh_handler = igmp_intr,
155 .nh_proto = NETISR_IGMP,
156 .nh_policy = NETISR_POLICY_SOURCE,
160 * System-wide globals.
162 * Unlocked access to these is OK, except for the global IGMP output
163 * queue. The IGMP subsystem lock ends up being system-wide for the moment,
164 * because all VIMAGEs have to share a global output queue, as netisrs
165 * themselves are not virtualized.
168 * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
169 * Any may be taken independently; if any are held at the same
170 * time, the above lock order must be followed.
171 * * All output is delegated to the netisr.
172 * Now that Giant has been eliminated, the netisr may be inlined.
173 * * IN_MULTI_LIST_LOCK covers in_multi.
174 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
175 * including the output queue.
176 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
177 * per-link state iterators.
178 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
179 * therefore it is not refcounted.
180 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
183 * * IGMP acquires its own reference every time an in_multi is passed to
184 * it and the group is being joined for the first time.
185 * * IGMP releases its reference(s) on in_multi in a deferred way,
186 * because the operations which process the release run as part of
187 * a loop whose control variables are directly affected by the release
188 * (that, and not recursing on the IF_ADDR_LOCK).
190 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
191 * to a vnet in ifp->if_vnet.
193 * SMPng: XXX We may potentially race operations on ifma_protospec.
194 * The problem is that we currently lack a clean way of taking the
195 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
196 * as anything which modifies ifma needs to be covered by that lock.
197 * So check for ifma_protospec being NULL before proceeding.
201 struct mbuf *m_raopt; /* Router Alert option */
202 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
205 * VIMAGE-wide globals.
207 * The IGMPv3 timers themselves need to run per-image, however, for
208 * historical reasons, timers run globally. This needs to be improved.
209 * An ifnet can only be in one vimage at a time, and the loopback
210 * ifnet, loif, is itself virtualized.
211 * It would otherwise be possible to seriously hose IGMP state,
212 * and create inconsistencies in upstream multicast routing, if you have
213 * multiple VIMAGEs running on the same link joining different multicast
214 * groups, UNLESS the "primary IP address" is different. This is because
215 * IGMP for IPv4 does not force link-local addresses to be used for each
216 * node, unlike MLD for IPv6.
217 * Obviously the IGMPv3 per-interface state has per-vimage granularity
220 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
221 * policy to control the address used by IGMP on the link.
223 VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general
225 VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change
227 VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host
228 * report; IGMPv3 g/sg
231 #define V_interface_timers_running VNET(interface_timers_running)
232 #define V_state_change_timers_running VNET(state_change_timers_running)
233 #define V_current_state_timers_running VNET(current_state_timers_running)
235 VNET_PCPUSTAT_DEFINE(struct igmpstat, igmpstat);
236 VNET_PCPUSTAT_SYSINIT(igmpstat);
237 VNET_PCPUSTAT_SYSUNINIT(igmpstat);
239 VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) =
240 LIST_HEAD_INITIALIZER(igi_head);
241 VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0};
243 #define V_igi_head VNET(igi_head)
244 #define V_igmp_gsrdelay VNET(igmp_gsrdelay)
246 VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1;
247 VNET_DEFINE_STATIC(int, igmp_sendra) = 1;
248 VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1;
249 VNET_DEFINE_STATIC(int, igmp_v1enable) = 1;
250 VNET_DEFINE_STATIC(int, igmp_v2enable) = 1;
251 VNET_DEFINE_STATIC(int, igmp_legacysupp);
252 VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3;
254 #define V_igmp_recvifkludge VNET(igmp_recvifkludge)
255 #define V_igmp_sendra VNET(igmp_sendra)
256 #define V_igmp_sendlocal VNET(igmp_sendlocal)
257 #define V_igmp_v1enable VNET(igmp_v1enable)
258 #define V_igmp_v2enable VNET(igmp_v2enable)
259 #define V_igmp_legacysupp VNET(igmp_legacysupp)
260 #define V_igmp_default_version VNET(igmp_default_version)
263 * Virtualized sysctls.
265 SYSCTL_PROC(_net_inet_igmp, IGMPCTL_STATS, stats,
266 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE,
267 &VNET_NAME(igmpstat), 0, sysctl_igmp_stat, "S,igmpstat",
268 "IGMP statistics (struct igmpstat, netinet/igmp_var.h)");
269 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
270 &VNET_NAME(igmp_recvifkludge), 0,
271 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
272 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
273 &VNET_NAME(igmp_sendra), 0,
274 "Send IP Router Alert option in IGMPv2/v3 messages");
275 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
276 &VNET_NAME(igmp_sendlocal), 0,
277 "Send IGMP membership reports for 224.0.0.0/24 groups");
278 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
279 &VNET_NAME(igmp_v1enable), 0,
280 "Enable backwards compatibility with IGMPv1");
281 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
282 &VNET_NAME(igmp_v2enable), 0,
283 "Enable backwards compatibility with IGMPv2");
284 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
285 &VNET_NAME(igmp_legacysupp), 0,
286 "Allow v1/v2 reports to suppress v3 group responses");
287 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
288 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
289 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
290 "Default version of IGMP to run on each interface");
291 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
292 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
293 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
294 "Rate limit for IGMPv3 Group-and-Source queries in seconds");
297 * Non-virtualized sysctls.
299 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
300 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
301 "Per-interface IGMPv3 state");
304 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
308 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
310 m->m_pkthdr.rcvif = ifp;
311 m->m_pkthdr.flowid = ifp->if_index;
315 igmp_scrub_context(struct mbuf *m)
318 m->m_pkthdr.PH_loc.ptr = NULL;
319 m->m_pkthdr.flowid = 0;
323 * Restore context from a queued IGMP output chain.
324 * Return saved ifindex.
326 * VIMAGE: The assertion is there to make sure that we
327 * actually called CURVNET_SET() with what's in the mbuf chain.
329 static __inline uint32_t
330 igmp_restore_context(struct mbuf *m)
334 #if defined(VIMAGE) && defined(INVARIANTS)
335 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
336 ("%s: called when curvnet was not restored", __func__));
339 return (m->m_pkthdr.flowid);
346 sysctl_igmp_stat(SYSCTL_HANDLER_ARGS)
348 struct igmpstat igps0;
352 error = sysctl_wire_old_buffer(req, sizeof(struct igmpstat));
356 if (req->oldptr != NULL) {
357 if (req->oldlen < sizeof(struct igmpstat))
361 * Copy the counters, and explicitly set the struct's
362 * version and length fields.
364 COUNTER_ARRAY_COPY(VNET(igmpstat), &igps0,
365 sizeof(struct igmpstat) / sizeof(uint64_t));
366 igps0.igps_version = IGPS_VERSION_3;
367 igps0.igps_len = IGPS_VERSION3_LEN;
368 error = SYSCTL_OUT(req, &igps0,
369 sizeof(struct igmpstat));
372 req->validlen = sizeof(struct igmpstat);
375 if (req->newptr != NULL) {
376 if (req->newlen < sizeof(struct igmpstat))
379 error = SYSCTL_IN(req, &igps0,
384 * igps0 must be "all zero".
387 while (p < (char *)&igps0 + sizeof(igps0) && *p == '\0')
389 if (p != (char *)&igps0 + sizeof(igps0)) {
393 COUNTER_ARRAY_ZERO(VNET(igmpstat),
394 sizeof(struct igmpstat) / sizeof(uint64_t));
401 * Retrieve or set default IGMP version.
403 * VIMAGE: Assume curvnet set by caller.
404 * SMPng: NOTE: Serialized by IGMP lock.
407 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
412 error = sysctl_wire_old_buffer(req, sizeof(int));
418 new = V_igmp_default_version;
420 error = sysctl_handle_int(oidp, &new, 0, req);
421 if (error || !req->newptr)
424 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
429 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
430 V_igmp_default_version, new);
432 V_igmp_default_version = new;
440 * Retrieve or set threshold between group-source queries in seconds.
442 * VIMAGE: Assume curvnet set by caller.
443 * SMPng: NOTE: Serialized by IGMP lock.
446 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
451 error = sysctl_wire_old_buffer(req, sizeof(int));
457 i = V_igmp_gsrdelay.tv_sec;
459 error = sysctl_handle_int(oidp, &i, 0, req);
460 if (error || !req->newptr)
463 if (i < -1 || i >= 60) {
468 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
469 V_igmp_gsrdelay.tv_sec, i);
470 V_igmp_gsrdelay.tv_sec = i;
478 * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
479 * For use by ifmcstat(8).
481 * SMPng: NOTE: Does an unlocked ifindex space read.
482 * VIMAGE: Assume curvnet set by caller. The node handler itself
483 * is not directly virtualized.
486 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
488 struct epoch_tracker et;
493 struct igmp_ifsoftc *igi;
498 if (req->newptr != NULL)
504 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
508 IN_MULTI_LIST_LOCK();
514 ifp = ifnet_byindex(name[0]);
519 LIST_FOREACH(igi, &V_igi_head, igi_link) {
520 if (ifp == igi->igi_ifp) {
521 struct igmp_ifinfo info;
523 info.igi_version = igi->igi_version;
524 info.igi_v1_timer = igi->igi_v1_timer;
525 info.igi_v2_timer = igi->igi_v2_timer;
526 info.igi_v3_timer = igi->igi_v3_timer;
527 info.igi_flags = igi->igi_flags;
528 info.igi_rv = igi->igi_rv;
529 info.igi_qi = igi->igi_qi;
530 info.igi_qri = igi->igi_qri;
531 info.igi_uri = igi->igi_uri;
532 error = SYSCTL_OUT(req, &info, sizeof(info));
539 IN_MULTI_LIST_UNLOCK();
544 * Dispatch an entire queue of pending packet chains
546 * VIMAGE: Assumes the vnet pointer has been set.
549 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
551 struct epoch_tracker et;
555 while ((m = mbufq_dequeue(mq)) != NULL) {
556 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
558 m->m_flags |= M_IGMP_LOOP;
559 netisr_dispatch(NETISR_IGMP, m);
567 * Filter outgoing IGMP report state by group.
569 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
570 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
571 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
572 * this may break certain IGMP snooping switches which rely on the old
575 * Return zero if the given group is one for which IGMP reports
576 * should be suppressed, or non-zero if reports should be issued.
579 igmp_isgroupreported(const struct in_addr addr)
582 if (in_allhosts(addr) ||
583 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
590 * Construct a Router Alert option to use in outgoing packets.
598 m = m_get(M_WAITOK, MT_DATA);
599 p = mtod(m, struct ipoption *);
600 p->ipopt_dst.s_addr = INADDR_ANY;
601 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */
602 p->ipopt_list[1] = 0x04; /* 4 bytes long */
603 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
604 p->ipopt_list[3] = 0x00; /* pad byte */
605 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
611 * Attach IGMP when PF_INET is attached to an interface.
613 struct igmp_ifsoftc *
614 igmp_domifattach(struct ifnet *ifp)
616 struct igmp_ifsoftc *igi;
618 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
619 __func__, ifp, ifp->if_xname);
623 igi = igi_alloc_locked(ifp);
624 if (!(ifp->if_flags & IFF_MULTICAST))
625 igi->igi_flags |= IGIF_SILENT;
633 * VIMAGE: assume curvnet set by caller.
635 static struct igmp_ifsoftc *
636 igi_alloc_locked(/*const*/ struct ifnet *ifp)
638 struct igmp_ifsoftc *igi;
642 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
647 igi->igi_version = V_igmp_default_version;
649 igi->igi_rv = IGMP_RV_INIT;
650 igi->igi_qi = IGMP_QI_INIT;
651 igi->igi_qri = IGMP_QRI_INIT;
652 igi->igi_uri = IGMP_URI_INIT;
653 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
655 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
657 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
667 * NOTE: Some finalization tasks need to run before the protocol domain
668 * is detached, but also before the link layer does its cleanup.
670 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
671 * XXX This is also bitten by unlocked ifma_protospec access.
674 igmp_ifdetach(struct ifnet *ifp)
676 struct igmp_ifsoftc *igi;
677 struct ifmultiaddr *ifma, *next;
678 struct in_multi *inm;
679 struct in_multi_head inm_free_tmp;
680 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
683 SLIST_INIT(&inm_free_tmp);
686 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
687 if (igi->igi_version == IGMP_VERSION_3) {
690 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
691 if (ifma->ifma_addr->sa_family != AF_INET ||
692 ifma->ifma_protospec == NULL)
694 inm = (struct in_multi *)ifma->ifma_protospec;
695 if (inm->inm_state == IGMP_LEAVING_MEMBER)
696 inm_rele_locked(&inm_free_tmp, inm);
697 inm_clear_recorded(inm);
698 if (__predict_false(ifma_restart)) {
699 ifma_restart = false;
703 IF_ADDR_WUNLOCK(ifp);
704 inm_release_list_deferred(&inm_free_tmp);
711 * Hook for domifdetach.
714 igmp_domifdetach(struct ifnet *ifp)
717 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
718 __func__, ifp, ifp->if_xname);
721 igi_delete_locked(ifp);
726 igi_delete_locked(const struct ifnet *ifp)
728 struct igmp_ifsoftc *igi, *tigi;
730 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
731 __func__, ifp, ifp->if_xname);
735 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
736 if (igi->igi_ifp == ifp) {
738 * Free deferred General Query responses.
740 mbufq_drain(&igi->igi_gq);
742 LIST_REMOVE(igi, igi_link);
750 * Process a received IGMPv1 query.
751 * Return non-zero if the message should be dropped.
753 * VIMAGE: The curvnet pointer is derived from the input ifp.
756 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
757 const struct igmp *igmp)
759 struct ifmultiaddr *ifma;
760 struct igmp_ifsoftc *igi;
761 struct in_multi *inm;
766 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
767 * 224.0.0.1. They are always treated as General Queries.
768 * igmp_group is always ignored. Do not drop it as a userland
769 * daemon may wish to see it.
770 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
772 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
773 IGMPSTAT_INC(igps_rcv_badqueries);
776 IGMPSTAT_INC(igps_rcv_gen_queries);
778 IN_MULTI_LIST_LOCK();
781 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
782 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
784 if (igi->igi_flags & IGIF_LOOPBACK) {
785 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
791 * Switch to IGMPv1 host compatibility mode.
793 igmp_set_version(igi, IGMP_VERSION_1);
795 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
798 * Start the timers in all of our group records
799 * for the interface on which the query arrived,
800 * except those which are already running.
802 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
803 if (ifma->ifma_addr->sa_family != AF_INET ||
804 ifma->ifma_protospec == NULL)
806 inm = (struct in_multi *)ifma->ifma_protospec;
807 if (inm->inm_timer != 0)
809 switch (inm->inm_state) {
810 case IGMP_NOT_MEMBER:
811 case IGMP_SILENT_MEMBER:
813 case IGMP_G_QUERY_PENDING_MEMBER:
814 case IGMP_SG_QUERY_PENDING_MEMBER:
815 case IGMP_REPORTING_MEMBER:
816 case IGMP_IDLE_MEMBER:
817 case IGMP_LAZY_MEMBER:
818 case IGMP_SLEEPING_MEMBER:
819 case IGMP_AWAKENING_MEMBER:
820 inm->inm_state = IGMP_REPORTING_MEMBER;
821 inm->inm_timer = IGMP_RANDOM_DELAY(
822 IGMP_V1V2_MAX_RI * IGMP_FASTHZ);
823 V_current_state_timers_running = 1;
825 case IGMP_LEAVING_MEMBER:
832 IN_MULTI_LIST_UNLOCK();
838 * Process a received IGMPv2 general or group-specific query.
841 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
842 const struct igmp *igmp)
844 struct ifmultiaddr *ifma;
845 struct igmp_ifsoftc *igi;
846 struct in_multi *inm;
847 int is_general_query;
852 is_general_query = 0;
855 * Validate address fields upfront.
856 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
858 if (in_nullhost(igmp->igmp_group)) {
860 * IGMPv2 General Query.
861 * If this was not sent to the all-hosts group, ignore it.
863 if (!in_allhosts(ip->ip_dst))
865 IGMPSTAT_INC(igps_rcv_gen_queries);
866 is_general_query = 1;
868 /* IGMPv2 Group-Specific Query. */
869 IGMPSTAT_INC(igps_rcv_group_queries);
872 IN_MULTI_LIST_LOCK();
875 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
876 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
878 if (igi->igi_flags & IGIF_LOOPBACK) {
879 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
885 * Ignore v2 query if in v1 Compatibility Mode.
887 if (igi->igi_version == IGMP_VERSION_1)
890 igmp_set_version(igi, IGMP_VERSION_2);
892 timer = igmp->igmp_code * IGMP_FASTHZ / IGMP_TIMER_SCALE;
896 if (is_general_query) {
898 * For each reporting group joined on this
899 * interface, kick the report timer.
901 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
903 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
904 if (ifma->ifma_addr->sa_family != AF_INET ||
905 ifma->ifma_protospec == NULL)
907 inm = (struct in_multi *)ifma->ifma_protospec;
908 igmp_v2_update_group(inm, timer);
912 * Group-specific IGMPv2 query, we need only
913 * look up the single group to process it.
915 inm = inm_lookup(ifp, igmp->igmp_group);
918 "process v2 query 0x%08x on ifp %p(%s)",
919 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
920 igmp_v2_update_group(inm, timer);
926 IN_MULTI_LIST_UNLOCK();
932 * Update the report timer on a group in response to an IGMPv2 query.
934 * If we are becoming the reporting member for this group, start the timer.
935 * If we already are the reporting member for this group, and timer is
936 * below the threshold, reset it.
938 * We may be updating the group for the first time since we switched
939 * to IGMPv3. If we are, then we must clear any recorded source lists,
940 * and transition to REPORTING state; the group timer is overloaded
941 * for group and group-source query responses.
943 * Unlike IGMPv3, the delay per group should be jittered
944 * to avoid bursts of IGMPv2 reports.
947 igmp_v2_update_group(struct in_multi *inm, const int timer)
950 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
951 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
953 IN_MULTI_LIST_LOCK_ASSERT();
955 switch (inm->inm_state) {
956 case IGMP_NOT_MEMBER:
957 case IGMP_SILENT_MEMBER:
959 case IGMP_REPORTING_MEMBER:
960 if (inm->inm_timer != 0 &&
961 inm->inm_timer <= timer) {
962 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
963 "skipping.", __func__);
967 case IGMP_SG_QUERY_PENDING_MEMBER:
968 case IGMP_G_QUERY_PENDING_MEMBER:
969 case IGMP_IDLE_MEMBER:
970 case IGMP_LAZY_MEMBER:
971 case IGMP_AWAKENING_MEMBER:
972 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
973 inm->inm_state = IGMP_REPORTING_MEMBER;
974 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
975 V_current_state_timers_running = 1;
977 case IGMP_SLEEPING_MEMBER:
978 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
979 inm->inm_state = IGMP_AWAKENING_MEMBER;
981 case IGMP_LEAVING_MEMBER:
987 * Process a received IGMPv3 general, group-specific or
988 * group-and-source-specific query.
989 * Assumes m has already been pulled up to the full IGMP message length.
990 * Return 0 if successful, otherwise an appropriate error code is returned.
993 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
994 /*const*/ struct igmpv3 *igmpv3)
996 struct igmp_ifsoftc *igi;
997 struct in_multi *inm;
998 int is_general_query;
999 uint32_t maxresp, nsrc, qqi;
1003 is_general_query = 0;
1005 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
1007 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
1008 if (maxresp >= 128) {
1009 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
1010 (IGMP_EXP(igmpv3->igmp_code) + 3);
1014 * Robustness must never be less than 2 for on-wire IGMPv3.
1015 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
1016 * an exception for interfaces whose IGMPv3 state changes
1017 * are redirected to loopback (e.g. MANET).
1019 qrv = IGMP_QRV(igmpv3->igmp_misc);
1021 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
1026 qqi = igmpv3->igmp_qqi;
1028 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
1029 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
1032 timer = maxresp * IGMP_FASTHZ / IGMP_TIMER_SCALE;
1036 nsrc = ntohs(igmpv3->igmp_numsrc);
1039 * Validate address fields and versions upfront before
1040 * accepting v3 query.
1041 * XXX SMPng: Unlocked access to igmpstat counters here.
1043 if (in_nullhost(igmpv3->igmp_group)) {
1045 * IGMPv3 General Query.
1047 * General Queries SHOULD be directed to 224.0.0.1.
1048 * A general query with a source list has undefined
1049 * behaviour; discard it.
1051 IGMPSTAT_INC(igps_rcv_gen_queries);
1052 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1053 IGMPSTAT_INC(igps_rcv_badqueries);
1056 is_general_query = 1;
1058 /* Group or group-source specific query. */
1060 IGMPSTAT_INC(igps_rcv_group_queries);
1062 IGMPSTAT_INC(igps_rcv_gsr_queries);
1065 IN_MULTI_LIST_LOCK();
1068 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1069 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1071 if (igi->igi_flags & IGIF_LOOPBACK) {
1072 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1073 ifp, ifp->if_xname);
1078 * Discard the v3 query if we're in Compatibility Mode.
1079 * The RFC is not obviously worded that hosts need to stay in
1080 * compatibility mode until the Old Version Querier Present
1083 if (igi->igi_version != IGMP_VERSION_3) {
1084 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1085 igi->igi_version, ifp, ifp->if_xname);
1089 igmp_set_version(igi, IGMP_VERSION_3);
1092 igi->igi_qri = maxresp;
1094 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1097 if (is_general_query) {
1099 * Schedule a current-state report on this ifp for
1100 * all groups, possibly containing source lists.
1101 * If there is a pending General Query response
1102 * scheduled earlier than the selected delay, do
1103 * not schedule any other reports.
1104 * Otherwise, reset the interface timer.
1106 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1107 ifp, ifp->if_xname);
1108 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1109 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1110 V_interface_timers_running = 1;
1114 * Group-source-specific queries are throttled on
1115 * a per-group basis to defeat denial-of-service attempts.
1116 * Queries for groups we are not a member of on this
1117 * link are simply ignored.
1119 inm = inm_lookup(ifp, igmpv3->igmp_group);
1123 if (!ratecheck(&inm->inm_lastgsrtv,
1124 &V_igmp_gsrdelay)) {
1125 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1127 IGMPSTAT_INC(igps_drop_gsr_queries);
1131 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1132 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1134 * If there is a pending General Query response
1135 * scheduled sooner than the selected delay, no
1136 * further report need be scheduled.
1137 * Otherwise, prepare to respond to the
1138 * group-specific or group-and-source query.
1140 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1141 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1146 IN_MULTI_LIST_UNLOCK();
1152 * Process a received IGMPv3 group-specific or group-and-source-specific
1154 * Return <0 if any error occurred. Currently this is ignored.
1157 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1158 int timer, /*const*/ struct igmpv3 *igmpv3)
1163 IN_MULTI_LIST_LOCK_ASSERT();
1168 switch (inm->inm_state) {
1169 case IGMP_NOT_MEMBER:
1170 case IGMP_SILENT_MEMBER:
1171 case IGMP_SLEEPING_MEMBER:
1172 case IGMP_LAZY_MEMBER:
1173 case IGMP_AWAKENING_MEMBER:
1174 case IGMP_IDLE_MEMBER:
1175 case IGMP_LEAVING_MEMBER:
1178 case IGMP_REPORTING_MEMBER:
1179 case IGMP_G_QUERY_PENDING_MEMBER:
1180 case IGMP_SG_QUERY_PENDING_MEMBER:
1184 nsrc = ntohs(igmpv3->igmp_numsrc);
1187 * Deal with group-specific queries upfront.
1188 * If any group query is already pending, purge any recorded
1189 * source-list state if it exists, and schedule a query response
1190 * for this group-specific query.
1193 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1194 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1195 inm_clear_recorded(inm);
1196 timer = min(inm->inm_timer, timer);
1198 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1199 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1200 V_current_state_timers_running = 1;
1205 * Deal with the case where a group-and-source-specific query has
1206 * been received but a group-specific query is already pending.
1208 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1209 timer = min(inm->inm_timer, timer);
1210 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1211 V_current_state_timers_running = 1;
1216 * Finally, deal with the case where a group-and-source-specific
1217 * query has been received, where a response to a previous g-s-r
1218 * query exists, or none exists.
1219 * In this case, we need to parse the source-list which the Querier
1220 * has provided us with and check if we have any source list filter
1221 * entries at T1 for these sources. If we do not, there is no need
1222 * schedule a report and the query may be dropped.
1223 * If we do, we must record them and schedule a current-state
1224 * report for those sources.
1225 * FIXME: Handling source lists larger than 1 mbuf requires that
1226 * we pass the mbuf chain pointer down to this function, and use
1227 * m_getptr() to walk the chain.
1229 if (inm->inm_nsrc > 0) {
1230 const struct in_addr *ap;
1233 ap = (const struct in_addr *)(igmpv3 + 1);
1235 for (i = 0; i < nsrc; i++, ap++) {
1236 retval = inm_record_source(inm, ap->s_addr);
1239 nrecorded += retval;
1241 if (nrecorded > 0) {
1243 "%s: schedule response to SG query", __func__);
1244 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1245 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1246 V_current_state_timers_running = 1;
1254 * Process a received IGMPv1 host membership report.
1256 * NOTE: 0.0.0.0 workaround breaks const correctness.
1259 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1260 /*const*/ struct igmp *igmp)
1262 struct in_ifaddr *ia;
1263 struct in_multi *inm;
1265 IGMPSTAT_INC(igps_rcv_reports);
1267 if (ifp->if_flags & IFF_LOOPBACK)
1270 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1271 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1272 IGMPSTAT_INC(igps_rcv_badreports);
1277 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1278 * Booting clients may use the source address 0.0.0.0. Some
1279 * IGMP daemons may not know how to use IP_RECVIF to determine
1280 * the interface upon which this message was received.
1281 * Replace 0.0.0.0 with the subnet address if told to do so.
1283 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1286 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1289 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1290 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1293 * IGMPv1 report suppression.
1294 * If we are a member of this group, and our membership should be
1295 * reported, stop our group timer and transition to the 'lazy' state.
1297 IN_MULTI_LIST_LOCK();
1298 inm = inm_lookup(ifp, igmp->igmp_group);
1300 struct igmp_ifsoftc *igi;
1304 KASSERT(igi != NULL,
1305 ("%s: no igi for ifp %p", __func__, ifp));
1309 IGMPSTAT_INC(igps_rcv_ourreports);
1312 * If we are in IGMPv3 host mode, do not allow the
1313 * other host's IGMPv1 report to suppress our reports
1314 * unless explicitly configured to do so.
1316 if (igi->igi_version == IGMP_VERSION_3) {
1317 if (V_igmp_legacysupp)
1318 igmp_v3_suppress_group_record(inm);
1324 switch (inm->inm_state) {
1325 case IGMP_NOT_MEMBER:
1326 case IGMP_SILENT_MEMBER:
1328 case IGMP_IDLE_MEMBER:
1329 case IGMP_LAZY_MEMBER:
1330 case IGMP_AWAKENING_MEMBER:
1332 "report suppressed for 0x%08x on ifp %p(%s)",
1333 ntohl(igmp->igmp_group.s_addr), ifp,
1335 case IGMP_SLEEPING_MEMBER:
1336 inm->inm_state = IGMP_SLEEPING_MEMBER;
1338 case IGMP_REPORTING_MEMBER:
1340 "report suppressed for 0x%08x on ifp %p(%s)",
1341 ntohl(igmp->igmp_group.s_addr), ifp,
1343 if (igi->igi_version == IGMP_VERSION_1)
1344 inm->inm_state = IGMP_LAZY_MEMBER;
1345 else if (igi->igi_version == IGMP_VERSION_2)
1346 inm->inm_state = IGMP_SLEEPING_MEMBER;
1348 case IGMP_G_QUERY_PENDING_MEMBER:
1349 case IGMP_SG_QUERY_PENDING_MEMBER:
1350 case IGMP_LEAVING_MEMBER:
1356 IN_MULTI_LIST_UNLOCK();
1362 * Process a received IGMPv2 host membership report.
1364 * NOTE: 0.0.0.0 workaround breaks const correctness.
1367 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1368 /*const*/ struct igmp *igmp)
1370 struct in_ifaddr *ia;
1371 struct in_multi *inm;
1374 * Make sure we don't hear our own membership report. Fast
1375 * leave requires knowing that we are the only member of a
1379 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1383 IGMPSTAT_INC(igps_rcv_reports);
1385 if (ifp->if_flags & IFF_LOOPBACK) {
1389 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1390 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1391 IGMPSTAT_INC(igps_rcv_badreports);
1396 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1397 * Booting clients may use the source address 0.0.0.0. Some
1398 * IGMP daemons may not know how to use IP_RECVIF to determine
1399 * the interface upon which this message was received.
1400 * Replace 0.0.0.0 with the subnet address if told to do so.
1402 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1404 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1407 CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1408 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1411 * IGMPv2 report suppression.
1412 * If we are a member of this group, and our membership should be
1413 * reported, and our group timer is pending or about to be reset,
1414 * stop our group timer by transitioning to the 'lazy' state.
1416 IN_MULTI_LIST_LOCK();
1417 inm = inm_lookup(ifp, igmp->igmp_group);
1419 struct igmp_ifsoftc *igi;
1422 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1424 IGMPSTAT_INC(igps_rcv_ourreports);
1427 * If we are in IGMPv3 host mode, do not allow the
1428 * other host's IGMPv1 report to suppress our reports
1429 * unless explicitly configured to do so.
1431 if (igi->igi_version == IGMP_VERSION_3) {
1432 if (V_igmp_legacysupp)
1433 igmp_v3_suppress_group_record(inm);
1439 switch (inm->inm_state) {
1440 case IGMP_NOT_MEMBER:
1441 case IGMP_SILENT_MEMBER:
1442 case IGMP_SLEEPING_MEMBER:
1444 case IGMP_REPORTING_MEMBER:
1445 case IGMP_IDLE_MEMBER:
1446 case IGMP_AWAKENING_MEMBER:
1448 "report suppressed for 0x%08x on ifp %p(%s)",
1449 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1450 case IGMP_LAZY_MEMBER:
1451 inm->inm_state = IGMP_LAZY_MEMBER;
1453 case IGMP_G_QUERY_PENDING_MEMBER:
1454 case IGMP_SG_QUERY_PENDING_MEMBER:
1455 case IGMP_LEAVING_MEMBER:
1461 IN_MULTI_LIST_UNLOCK();
1467 igmp_input(struct mbuf **mp, int *offp, int proto)
1478 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1481 ifp = m->m_pkthdr.rcvif;
1484 IGMPSTAT_INC(igps_rcv_total);
1486 ip = mtod(m, struct ip *);
1488 igmplen = ntohs(ip->ip_len) - iphlen;
1493 if (igmplen < IGMP_MINLEN) {
1494 IGMPSTAT_INC(igps_rcv_tooshort);
1496 return (IPPROTO_DONE);
1500 * Always pullup to the minimum size for v1/v2 or v3
1501 * to amortize calls to m_pullup().
1504 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1505 minlen += IGMP_V3_QUERY_MINLEN;
1507 minlen += IGMP_MINLEN;
1508 if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1509 (m = m_pullup(m, minlen)) == NULL) {
1510 IGMPSTAT_INC(igps_rcv_tooshort);
1511 return (IPPROTO_DONE);
1513 ip = mtod(m, struct ip *);
1516 * Validate checksum.
1518 m->m_data += iphlen;
1520 igmp = mtod(m, struct igmp *);
1521 if (in_cksum(m, igmplen)) {
1522 IGMPSTAT_INC(igps_rcv_badsum);
1524 return (IPPROTO_DONE);
1526 m->m_data -= iphlen;
1530 * IGMP control traffic is link-scope, and must have a TTL of 1.
1531 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1532 * probe packets may come from beyond the LAN.
1534 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1535 IGMPSTAT_INC(igps_rcv_badttl);
1537 return (IPPROTO_DONE);
1540 switch (igmp->igmp_type) {
1541 case IGMP_HOST_MEMBERSHIP_QUERY:
1542 if (igmplen == IGMP_MINLEN) {
1543 if (igmp->igmp_code == 0)
1544 queryver = IGMP_VERSION_1;
1546 queryver = IGMP_VERSION_2;
1547 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1548 queryver = IGMP_VERSION_3;
1550 IGMPSTAT_INC(igps_rcv_tooshort);
1552 return (IPPROTO_DONE);
1556 case IGMP_VERSION_1:
1557 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1558 if (!V_igmp_v1enable)
1560 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1562 return (IPPROTO_DONE);
1566 case IGMP_VERSION_2:
1567 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1568 if (!V_igmp_v2enable)
1570 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1572 return (IPPROTO_DONE);
1576 case IGMP_VERSION_3: {
1577 struct igmpv3 *igmpv3;
1581 IGMPSTAT_INC(igps_rcv_v3_queries);
1582 igmpv3 = (struct igmpv3 *)igmp;
1584 * Validate length based on source count.
1586 nsrc = ntohs(igmpv3->igmp_numsrc);
1587 if (nsrc * sizeof(in_addr_t) >
1588 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1589 IGMPSTAT_INC(igps_rcv_tooshort);
1591 return (IPPROTO_DONE);
1594 * m_pullup() may modify m, so pullup in
1597 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1598 sizeof(struct in_addr) * nsrc;
1599 if ((!M_WRITABLE(m) ||
1600 m->m_len < igmpv3len) &&
1601 (m = m_pullup(m, igmpv3len)) == NULL) {
1602 IGMPSTAT_INC(igps_rcv_tooshort);
1603 return (IPPROTO_DONE);
1605 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1607 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1609 return (IPPROTO_DONE);
1616 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1617 if (!V_igmp_v1enable)
1619 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1621 return (IPPROTO_DONE);
1625 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1626 if (!V_igmp_v2enable)
1628 if (!ip_checkrouteralert(m))
1629 IGMPSTAT_INC(igps_rcv_nora);
1630 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1632 return (IPPROTO_DONE);
1636 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1638 * Hosts do not need to process IGMPv3 membership reports,
1639 * as report suppression is no longer required.
1641 if (!ip_checkrouteralert(m))
1642 IGMPSTAT_INC(igps_rcv_nora);
1650 * Pass all valid IGMP packets up to any process(es) listening on a
1654 return (rip_input(mp, offp, proto));
1658 * Fast timeout handler (global).
1659 * VIMAGE: Timeout handlers are expected to service all vimages.
1661 static struct callout igmpfast_callout;
1663 igmp_fasttimo(void *arg __unused)
1665 struct epoch_tracker et;
1666 VNET_ITERATOR_DECL(vnet_iter);
1668 NET_EPOCH_ENTER(et);
1669 VNET_LIST_RLOCK_NOSLEEP();
1670 VNET_FOREACH(vnet_iter) {
1671 CURVNET_SET(vnet_iter);
1672 igmp_fasttimo_vnet();
1675 VNET_LIST_RUNLOCK_NOSLEEP();
1678 callout_reset(&igmpfast_callout, hz / IGMP_FASTHZ, igmp_fasttimo, NULL);
1682 * Fast timeout handler (per-vnet).
1683 * Sends are shuffled off to a netisr to deal with Giant.
1685 * VIMAGE: Assume caller has set up our curvnet.
1688 igmp_fasttimo_vnet(void)
1690 struct mbufq scq; /* State-change packets */
1691 struct mbufq qrq; /* Query response packets */
1693 struct igmp_ifsoftc *igi;
1694 struct ifmultiaddr *ifma, *next;
1695 struct in_multi *inm;
1696 struct in_multi_head inm_free_tmp;
1697 int loop, uri_fasthz;
1703 * Quick check to see if any work needs to be done, in order to
1704 * minimize the overhead of fasttimo processing.
1705 * SMPng: XXX Unlocked reads.
1707 if (!V_current_state_timers_running &&
1708 !V_interface_timers_running &&
1709 !V_state_change_timers_running)
1712 SLIST_INIT(&inm_free_tmp);
1713 IN_MULTI_LIST_LOCK();
1717 * IGMPv3 General Query response timer processing.
1719 if (V_interface_timers_running) {
1720 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1722 V_interface_timers_running = 0;
1723 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1724 if (igi->igi_v3_timer == 0) {
1726 } else if (--igi->igi_v3_timer == 0) {
1727 igmp_v3_dispatch_general_query(igi);
1729 V_interface_timers_running = 1;
1734 if (!V_current_state_timers_running &&
1735 !V_state_change_timers_running)
1738 V_current_state_timers_running = 0;
1739 V_state_change_timers_running = 0;
1741 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1744 * IGMPv1/v2/v3 host report and state-change timer processing.
1745 * Note: Processing a v3 group timer may remove a node.
1747 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1750 if (igi->igi_version == IGMP_VERSION_3) {
1751 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1752 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1754 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1755 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1760 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1761 if (ifma->ifma_addr->sa_family != AF_INET ||
1762 ifma->ifma_protospec == NULL)
1764 inm = (struct in_multi *)ifma->ifma_protospec;
1765 switch (igi->igi_version) {
1766 case IGMP_VERSION_1:
1767 case IGMP_VERSION_2:
1768 igmp_v1v2_process_group_timer(inm,
1771 case IGMP_VERSION_3:
1772 igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1773 &scq, inm, uri_fasthz);
1776 if (__predict_false(ifma_restart)) {
1777 ifma_restart = false;
1781 IF_ADDR_WUNLOCK(ifp);
1783 if (igi->igi_version == IGMP_VERSION_3) {
1784 igmp_dispatch_queue(&qrq, 0, loop);
1785 igmp_dispatch_queue(&scq, 0, loop);
1788 * Free the in_multi reference(s) for this
1791 inm_release_list_deferred(&inm_free_tmp);
1797 IN_MULTI_LIST_UNLOCK();
1801 * Update host report group timer for IGMPv1/v2.
1802 * Will update the global pending timer flags.
1805 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1807 int report_timer_expired;
1809 IN_MULTI_LIST_LOCK_ASSERT();
1812 if (inm->inm_timer == 0) {
1813 report_timer_expired = 0;
1814 } else if (--inm->inm_timer == 0) {
1815 report_timer_expired = 1;
1817 V_current_state_timers_running = 1;
1821 switch (inm->inm_state) {
1822 case IGMP_NOT_MEMBER:
1823 case IGMP_SILENT_MEMBER:
1824 case IGMP_IDLE_MEMBER:
1825 case IGMP_LAZY_MEMBER:
1826 case IGMP_SLEEPING_MEMBER:
1827 case IGMP_AWAKENING_MEMBER:
1829 case IGMP_REPORTING_MEMBER:
1830 if (report_timer_expired) {
1831 inm->inm_state = IGMP_IDLE_MEMBER;
1832 (void)igmp_v1v2_queue_report(inm,
1833 (version == IGMP_VERSION_2) ?
1834 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1835 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1838 case IGMP_G_QUERY_PENDING_MEMBER:
1839 case IGMP_SG_QUERY_PENDING_MEMBER:
1840 case IGMP_LEAVING_MEMBER:
1846 * Update a group's timers for IGMPv3.
1847 * Will update the global pending timer flags.
1848 * Note: Unlocked read from igi.
1851 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1852 struct mbufq *qrq, struct mbufq *scq,
1853 struct in_multi *inm, const int uri_fasthz)
1855 int query_response_timer_expired;
1856 int state_change_retransmit_timer_expired;
1858 IN_MULTI_LIST_LOCK_ASSERT();
1861 query_response_timer_expired = 0;
1862 state_change_retransmit_timer_expired = 0;
1865 * During a transition from v1/v2 compatibility mode back to v3,
1866 * a group record in REPORTING state may still have its group
1867 * timer active. This is a no-op in this function; it is easier
1868 * to deal with it here than to complicate the slow-timeout path.
1870 if (inm->inm_timer == 0) {
1871 query_response_timer_expired = 0;
1872 } else if (--inm->inm_timer == 0) {
1873 query_response_timer_expired = 1;
1875 V_current_state_timers_running = 1;
1878 if (inm->inm_sctimer == 0) {
1879 state_change_retransmit_timer_expired = 0;
1880 } else if (--inm->inm_sctimer == 0) {
1881 state_change_retransmit_timer_expired = 1;
1883 V_state_change_timers_running = 1;
1886 /* We are in fasttimo, so be quick about it. */
1887 if (!state_change_retransmit_timer_expired &&
1888 !query_response_timer_expired)
1891 switch (inm->inm_state) {
1892 case IGMP_NOT_MEMBER:
1893 case IGMP_SILENT_MEMBER:
1894 case IGMP_SLEEPING_MEMBER:
1895 case IGMP_LAZY_MEMBER:
1896 case IGMP_AWAKENING_MEMBER:
1897 case IGMP_IDLE_MEMBER:
1899 case IGMP_G_QUERY_PENDING_MEMBER:
1900 case IGMP_SG_QUERY_PENDING_MEMBER:
1902 * Respond to a previously pending Group-Specific
1903 * or Group-and-Source-Specific query by enqueueing
1904 * the appropriate Current-State report for
1905 * immediate transmission.
1907 if (query_response_timer_expired) {
1908 int retval __unused;
1910 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1911 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1912 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1914 inm->inm_state = IGMP_REPORTING_MEMBER;
1915 /* XXX Clear recorded sources for next time. */
1916 inm_clear_recorded(inm);
1919 case IGMP_REPORTING_MEMBER:
1920 case IGMP_LEAVING_MEMBER:
1921 if (state_change_retransmit_timer_expired) {
1923 * State-change retransmission timer fired.
1924 * If there are any further pending retransmissions,
1925 * set the global pending state-change flag, and
1928 if (--inm->inm_scrv > 0) {
1929 inm->inm_sctimer = uri_fasthz;
1930 V_state_change_timers_running = 1;
1933 * Retransmit the previously computed state-change
1934 * report. If there are no further pending
1935 * retransmissions, the mbuf queue will be consumed.
1936 * Update T0 state to T1 as we have now sent
1939 (void)igmp_v3_merge_state_changes(inm, scq);
1942 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1943 ntohl(inm->inm_addr.s_addr),
1944 inm->inm_ifp->if_xname);
1947 * If we are leaving the group for good, make sure
1948 * we release IGMP's reference to it.
1949 * This release must be deferred using a SLIST,
1950 * as we are called from a loop which traverses
1951 * the in_ifmultiaddr TAILQ.
1953 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1954 inm->inm_scrv == 0) {
1955 inm->inm_state = IGMP_NOT_MEMBER;
1956 inm_rele_locked(inmh, inm);
1964 * Suppress a group's pending response to a group or source/group query.
1966 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1967 * Do NOT update ST1/ST0 as this operation merely suppresses
1968 * the currently pending group record.
1969 * Do NOT suppress the response to a general query. It is possible but
1970 * it would require adding another state or flag.
1973 igmp_v3_suppress_group_record(struct in_multi *inm)
1976 IN_MULTI_LIST_LOCK_ASSERT();
1978 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1979 ("%s: not IGMPv3 mode on link", __func__));
1981 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1982 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1985 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1986 inm_clear_recorded(inm);
1989 inm->inm_state = IGMP_REPORTING_MEMBER;
1993 * Switch to a different IGMP version on the given interface,
1994 * as per Section 7.2.1.
1997 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1999 int old_version_timer;
2003 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
2004 version, igi->igi_ifp, igi->igi_ifp->if_xname);
2006 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
2008 * Compute the "Older Version Querier Present" timer as per
2011 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
2012 old_version_timer *= IGMP_SLOWHZ;
2014 if (version == IGMP_VERSION_1) {
2015 igi->igi_v1_timer = old_version_timer;
2016 igi->igi_v2_timer = 0;
2017 } else if (version == IGMP_VERSION_2) {
2018 igi->igi_v1_timer = 0;
2019 igi->igi_v2_timer = old_version_timer;
2023 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2024 if (igi->igi_version != IGMP_VERSION_2) {
2025 igi->igi_version = IGMP_VERSION_2;
2026 igmp_v3_cancel_link_timers(igi);
2028 } else if (igi->igi_v1_timer > 0) {
2029 if (igi->igi_version != IGMP_VERSION_1) {
2030 igi->igi_version = IGMP_VERSION_1;
2031 igmp_v3_cancel_link_timers(igi);
2037 * Cancel pending IGMPv3 timers for the given link and all groups
2038 * joined on it; state-change, general-query, and group-query timers.
2040 * Only ever called on a transition from v3 to Compatibility mode. Kill
2041 * the timers stone dead (this may be expensive for large N groups), they
2042 * will be restarted if Compatibility Mode deems that they must be due to
2046 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2048 struct ifmultiaddr *ifma, *ifmatmp;
2050 struct in_multi *inm;
2051 struct in_multi_head inm_free_tmp;
2053 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2054 igi->igi_ifp, igi->igi_ifp->if_xname);
2056 IN_MULTI_LIST_LOCK_ASSERT();
2060 SLIST_INIT(&inm_free_tmp);
2063 * Stop the v3 General Query Response on this link stone dead.
2064 * If fasttimo is woken up due to V_interface_timers_running,
2065 * the flag will be cleared if there are no pending link timers.
2067 igi->igi_v3_timer = 0;
2070 * Now clear the current-state and state-change report timers
2071 * for all memberships scoped to this link.
2075 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, ifmatmp) {
2076 if (ifma->ifma_addr->sa_family != AF_INET ||
2077 ifma->ifma_protospec == NULL)
2079 inm = (struct in_multi *)ifma->ifma_protospec;
2080 switch (inm->inm_state) {
2081 case IGMP_NOT_MEMBER:
2082 case IGMP_SILENT_MEMBER:
2083 case IGMP_IDLE_MEMBER:
2084 case IGMP_LAZY_MEMBER:
2085 case IGMP_SLEEPING_MEMBER:
2086 case IGMP_AWAKENING_MEMBER:
2088 * These states are either not relevant in v3 mode,
2089 * or are unreported. Do nothing.
2092 case IGMP_LEAVING_MEMBER:
2094 * If we are leaving the group and switching to
2095 * compatibility mode, we need to release the final
2096 * reference held for issuing the INCLUDE {}, and
2097 * transition to REPORTING to ensure the host leave
2098 * message is sent upstream to the old querier --
2099 * transition to NOT would lose the leave and race.
2101 inm_rele_locked(&inm_free_tmp, inm);
2103 case IGMP_G_QUERY_PENDING_MEMBER:
2104 case IGMP_SG_QUERY_PENDING_MEMBER:
2105 inm_clear_recorded(inm);
2107 case IGMP_REPORTING_MEMBER:
2108 inm->inm_state = IGMP_REPORTING_MEMBER;
2112 * Always clear state-change and group report timers.
2113 * Free any pending IGMPv3 state-change records.
2115 inm->inm_sctimer = 0;
2117 mbufq_drain(&inm->inm_scq);
2119 IF_ADDR_WUNLOCK(ifp);
2121 inm_release_list_deferred(&inm_free_tmp);
2125 * Update the Older Version Querier Present timers for a link.
2126 * See Section 7.2.1 of RFC 3376.
2129 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2134 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2136 * IGMPv1 and IGMPv2 Querier Present timers expired.
2140 if (igi->igi_version != IGMP_VERSION_3) {
2142 "%s: transition from v%d -> v%d on %p(%s)",
2143 __func__, igi->igi_version, IGMP_VERSION_3,
2144 igi->igi_ifp, igi->igi_ifp->if_xname);
2145 igi->igi_version = IGMP_VERSION_3;
2147 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2149 * IGMPv1 Querier Present timer expired,
2150 * IGMPv2 Querier Present timer running.
2151 * If IGMPv2 was disabled since last timeout,
2153 * If IGMPv2 is enabled, revert to IGMPv2.
2155 if (!V_igmp_v2enable) {
2157 "%s: transition from v%d -> v%d on %p(%s)",
2158 __func__, igi->igi_version, IGMP_VERSION_3,
2159 igi->igi_ifp, igi->igi_ifp->if_xname);
2160 igi->igi_v2_timer = 0;
2161 igi->igi_version = IGMP_VERSION_3;
2163 --igi->igi_v2_timer;
2164 if (igi->igi_version != IGMP_VERSION_2) {
2166 "%s: transition from v%d -> v%d on %p(%s)",
2167 __func__, igi->igi_version, IGMP_VERSION_2,
2168 igi->igi_ifp, igi->igi_ifp->if_xname);
2169 igi->igi_version = IGMP_VERSION_2;
2170 igmp_v3_cancel_link_timers(igi);
2173 } else if (igi->igi_v1_timer > 0) {
2175 * IGMPv1 Querier Present timer running.
2176 * Stop IGMPv2 timer if running.
2178 * If IGMPv1 was disabled since last timeout,
2180 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2182 if (!V_igmp_v1enable) {
2184 "%s: transition from v%d -> v%d on %p(%s)",
2185 __func__, igi->igi_version, IGMP_VERSION_3,
2186 igi->igi_ifp, igi->igi_ifp->if_xname);
2187 igi->igi_v1_timer = 0;
2188 igi->igi_version = IGMP_VERSION_3;
2190 --igi->igi_v1_timer;
2192 if (igi->igi_v2_timer > 0) {
2194 "%s: cancel v2 timer on %p(%s)",
2195 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2196 igi->igi_v2_timer = 0;
2202 * Global slowtimo handler.
2203 * VIMAGE: Timeout handlers are expected to service all vimages.
2205 static struct callout igmpslow_callout;
2207 igmp_slowtimo(void *arg __unused)
2209 struct epoch_tracker et;
2210 VNET_ITERATOR_DECL(vnet_iter);
2212 NET_EPOCH_ENTER(et);
2213 VNET_LIST_RLOCK_NOSLEEP();
2214 VNET_FOREACH(vnet_iter) {
2215 CURVNET_SET(vnet_iter);
2216 igmp_slowtimo_vnet();
2219 VNET_LIST_RUNLOCK_NOSLEEP();
2222 callout_reset(&igmpslow_callout, hz / IGMP_SLOWHZ, igmp_slowtimo, NULL);
2226 * Per-vnet slowtimo handler.
2229 igmp_slowtimo_vnet(void)
2231 struct igmp_ifsoftc *igi;
2235 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2236 igmp_v1v2_process_querier_timers(igi);
2243 * Dispatch an IGMPv1/v2 host report or leave message.
2244 * These are always small enough to fit inside a single mbuf.
2247 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2249 struct epoch_tracker et;
2255 IN_MULTI_LIST_LOCK_ASSERT();
2260 m = m_gethdr(M_NOWAIT, MT_DATA);
2263 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2265 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2267 m->m_data += sizeof(struct ip);
2268 m->m_len = sizeof(struct igmp);
2270 igmp = mtod(m, struct igmp *);
2271 igmp->igmp_type = type;
2272 igmp->igmp_code = 0;
2273 igmp->igmp_group = inm->inm_addr;
2274 igmp->igmp_cksum = 0;
2275 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2277 m->m_data -= sizeof(struct ip);
2278 m->m_len += sizeof(struct ip);
2280 ip = mtod(m, struct ip *);
2282 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2284 ip->ip_p = IPPROTO_IGMP;
2285 ip->ip_src.s_addr = INADDR_ANY;
2287 if (type == IGMP_HOST_LEAVE_MESSAGE)
2288 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2290 ip->ip_dst = inm->inm_addr;
2292 igmp_save_context(m, ifp);
2294 m->m_flags |= M_IGMPV2;
2295 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2296 m->m_flags |= M_IGMP_LOOP;
2298 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2299 NET_EPOCH_ENTER(et);
2300 netisr_dispatch(NETISR_IGMP, m);
2307 * Process a state change from the upper layer for the given IPv4 group.
2309 * Each socket holds a reference on the in_multi in its own ip_moptions.
2310 * The socket layer will have made the necessary updates to.the group
2311 * state, it is now up to IGMP to issue a state change report if there
2312 * has been any change between T0 (when the last state-change was issued)
2315 * We use the IGMPv3 state machine at group level. The IGMP module
2316 * however makes the decision as to which IGMP protocol version to speak.
2317 * A state change *from* INCLUDE {} always means an initial join.
2318 * A state change *to* INCLUDE {} always means a final leave.
2320 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2321 * save ourselves a bunch of work; any exclusive mode groups need not
2322 * compute source filter lists.
2324 * VIMAGE: curvnet should have been set by caller, as this routine
2325 * is called from the socket option handlers.
2328 igmp_change_state(struct in_multi *inm)
2330 struct igmp_ifsoftc *igi;
2335 IN_MULTI_LOCK_ASSERT();
2337 * Try to detect if the upper layer just asked us to change state
2338 * for an interface which has now gone away.
2340 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2341 ifp = inm->inm_ifma->ifma_ifp;
2343 * Sanity check that netinet's notion of ifp is the
2346 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2350 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2351 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2354 * If we detect a state transition to or from MCAST_UNDEFINED
2355 * for this group, then we are starting or finishing an IGMP
2356 * life cycle for this group.
2358 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2359 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2360 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2361 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2362 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2363 error = igmp_initial_join(inm, igi);
2365 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2366 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2367 igmp_final_leave(inm, igi);
2371 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2374 error = igmp_handle_state_change(inm, igi);
2382 * Perform the initial join for an IGMP group.
2384 * When joining a group:
2385 * If the group should have its IGMP traffic suppressed, do nothing.
2386 * IGMPv1 starts sending IGMPv1 host membership reports.
2387 * IGMPv2 starts sending IGMPv2 host membership reports.
2388 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2389 * initial state of the membership.
2392 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2396 int error, retval, syncstates;
2398 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2399 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2406 IN_MULTI_LOCK_ASSERT();
2409 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2412 * Groups joined on loopback or marked as 'not reported',
2413 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2414 * are never reported in any IGMP protocol exchanges.
2415 * All other groups enter the appropriate IGMP state machine
2416 * for the version in use on this link.
2417 * A link marked as IGIF_SILENT causes IGMP to be completely
2418 * disabled for the link.
2420 if ((ifp->if_flags & IFF_LOOPBACK) ||
2421 (igi->igi_flags & IGIF_SILENT) ||
2422 !igmp_isgroupreported(inm->inm_addr)) {
2424 "%s: not kicking state machine for silent group", __func__);
2425 inm->inm_state = IGMP_SILENT_MEMBER;
2429 * Deal with overlapping in_multi lifecycle.
2430 * If this group was LEAVING, then make sure
2431 * we drop the reference we picked up to keep the
2432 * group around for the final INCLUDE {} enqueue.
2434 if (igi->igi_version == IGMP_VERSION_3 &&
2435 inm->inm_state == IGMP_LEAVING_MEMBER) {
2436 MPASS(inm->inm_refcount > 1);
2437 inm_rele_locked(NULL, inm);
2439 inm->inm_state = IGMP_REPORTING_MEMBER;
2441 switch (igi->igi_version) {
2442 case IGMP_VERSION_1:
2443 case IGMP_VERSION_2:
2444 inm->inm_state = IGMP_IDLE_MEMBER;
2445 error = igmp_v1v2_queue_report(inm,
2446 (igi->igi_version == IGMP_VERSION_2) ?
2447 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2448 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2450 inm->inm_timer = IGMP_RANDOM_DELAY(
2451 IGMP_V1V2_MAX_RI * IGMP_FASTHZ);
2452 V_current_state_timers_running = 1;
2456 case IGMP_VERSION_3:
2458 * Defer update of T0 to T1, until the first copy
2459 * of the state change has been transmitted.
2464 * Immediately enqueue a State-Change Report for
2465 * this interface, freeing any previous reports.
2466 * Don't kick the timers if there is nothing to do,
2467 * or if an error occurred.
2471 retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2473 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2476 error = retval * -1;
2481 * Schedule transmission of pending state-change
2482 * report up to RV times for this link. The timer
2483 * will fire at the next igmp_fasttimo (~200ms),
2484 * giving us an opportunity to merge the reports.
2486 if (igi->igi_flags & IGIF_LOOPBACK) {
2489 KASSERT(igi->igi_rv > 1,
2490 ("%s: invalid robustness %d", __func__,
2492 inm->inm_scrv = igi->igi_rv;
2494 inm->inm_sctimer = 1;
2495 V_state_change_timers_running = 1;
2503 * Only update the T0 state if state change is atomic,
2504 * i.e. we don't need to wait for a timer to fire before we
2505 * can consider the state change to have been communicated.
2509 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2510 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2517 * Issue an intermediate state change during the IGMP life-cycle.
2520 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2525 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2526 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2530 IN_MULTI_LIST_LOCK_ASSERT();
2533 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2535 if ((ifp->if_flags & IFF_LOOPBACK) ||
2536 (igi->igi_flags & IGIF_SILENT) ||
2537 !igmp_isgroupreported(inm->inm_addr) ||
2538 (igi->igi_version != IGMP_VERSION_3)) {
2539 if (!igmp_isgroupreported(inm->inm_addr)) {
2541 "%s: not kicking state machine for silent group", __func__);
2543 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2545 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2546 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2550 mbufq_drain(&inm->inm_scq);
2552 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2553 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2558 * If record(s) were enqueued, start the state-change
2559 * report timer for this group.
2561 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2562 inm->inm_sctimer = 1;
2563 V_state_change_timers_running = 1;
2569 * Perform the final leave for an IGMP group.
2571 * When leaving a group:
2572 * IGMPv1 does nothing.
2573 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2574 * IGMPv3 enqueues a state-change report containing a transition
2575 * to INCLUDE {} for immediate transmission.
2578 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2584 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2585 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2586 inm->inm_ifp->if_xname);
2588 IN_MULTI_LIST_LOCK_ASSERT();
2591 switch (inm->inm_state) {
2592 case IGMP_NOT_MEMBER:
2593 case IGMP_SILENT_MEMBER:
2594 case IGMP_LEAVING_MEMBER:
2595 /* Already leaving or left; do nothing. */
2597 "%s: not kicking state machine for silent group", __func__);
2599 case IGMP_REPORTING_MEMBER:
2600 case IGMP_IDLE_MEMBER:
2601 case IGMP_G_QUERY_PENDING_MEMBER:
2602 case IGMP_SG_QUERY_PENDING_MEMBER:
2603 if (igi->igi_version == IGMP_VERSION_2) {
2605 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2606 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2607 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2610 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2611 inm->inm_state = IGMP_NOT_MEMBER;
2612 } else if (igi->igi_version == IGMP_VERSION_3) {
2614 * Stop group timer and all pending reports.
2615 * Immediately enqueue a state-change report
2616 * TO_IN {} to be sent on the next fast timeout,
2617 * giving us an opportunity to merge reports.
2619 mbufq_drain(&inm->inm_scq);
2621 if (igi->igi_flags & IGIF_LOOPBACK) {
2624 inm->inm_scrv = igi->igi_rv;
2626 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2627 "pending retransmissions.", __func__,
2628 ntohl(inm->inm_addr.s_addr),
2629 inm->inm_ifp->if_xname, inm->inm_scrv);
2630 if (inm->inm_scrv == 0) {
2631 inm->inm_state = IGMP_NOT_MEMBER;
2632 inm->inm_sctimer = 0;
2634 int retval __unused;
2636 inm_acquire_locked(inm);
2638 retval = igmp_v3_enqueue_group_record(
2639 &inm->inm_scq, inm, 1, 0, 0);
2640 KASSERT(retval != 0,
2641 ("%s: enqueue record = %d", __func__,
2644 inm->inm_state = IGMP_LEAVING_MEMBER;
2645 inm->inm_sctimer = 1;
2646 V_state_change_timers_running = 1;
2652 case IGMP_LAZY_MEMBER:
2653 case IGMP_SLEEPING_MEMBER:
2654 case IGMP_AWAKENING_MEMBER:
2655 /* Our reports are suppressed; do nothing. */
2661 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2662 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2663 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2664 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2665 __func__, ntohl(inm->inm_addr.s_addr),
2666 inm->inm_ifp->if_xname);
2671 * Enqueue an IGMPv3 group record to the given output queue.
2673 * XXX This function could do with having the allocation code
2674 * split out, and the multiple-tree-walks coalesced into a single
2675 * routine as has been done in igmp_v3_enqueue_filter_change().
2677 * If is_state_change is zero, a current-state record is appended.
2678 * If is_state_change is non-zero, a state-change report is appended.
2680 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2681 * If is_group_query is zero, and if there is a packet with free space
2682 * at the tail of the queue, it will be appended to providing there
2683 * is enough free space.
2684 * Otherwise a new mbuf packet chain is allocated.
2686 * If is_source_query is non-zero, each source is checked to see if
2687 * it was recorded for a Group-Source query, and will be omitted if
2688 * it is not both in-mode and recorded.
2690 * The function will attempt to allocate leading space in the packet
2691 * for the IP/IGMP header to be prepended without fragmenting the chain.
2693 * If successful the size of all data appended to the queue is returned,
2694 * otherwise an error code less than zero is returned, or zero if
2695 * no record(s) were appended.
2698 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2699 const int is_state_change, const int is_group_query,
2700 const int is_source_query)
2702 struct igmp_grouprec ig;
2703 struct igmp_grouprec *pig;
2705 struct ip_msource *ims, *nims;
2706 struct mbuf *m0, *m, *md;
2707 int is_filter_list_change;
2708 int minrec0len, m0srcs, msrcs, nbytes, off;
2709 int record_has_sources;
2715 IN_MULTI_LIST_LOCK_ASSERT();
2718 is_filter_list_change = 0;
2725 record_has_sources = 1;
2727 type = IGMP_DO_NOTHING;
2728 mode = inm->inm_st[1].iss_fmode;
2731 * If we did not transition out of ASM mode during t0->t1,
2732 * and there are no source nodes to process, we can skip
2733 * the generation of source records.
2735 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2737 record_has_sources = 0;
2739 if (is_state_change) {
2741 * Queue a state change record.
2742 * If the mode did not change, and there are non-ASM
2743 * listeners or source filters present,
2744 * we potentially need to issue two records for the group.
2745 * If we are transitioning to MCAST_UNDEFINED, we need
2746 * not send any sources.
2747 * If there are ASM listeners, and there was no filter
2748 * mode transition of any kind, do nothing.
2750 if (mode != inm->inm_st[0].iss_fmode) {
2751 if (mode == MCAST_EXCLUDE) {
2752 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2754 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2756 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2758 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2759 if (mode == MCAST_UNDEFINED)
2760 record_has_sources = 0;
2763 if (record_has_sources) {
2764 is_filter_list_change = 1;
2766 type = IGMP_DO_NOTHING;
2771 * Queue a current state record.
2773 if (mode == MCAST_EXCLUDE) {
2774 type = IGMP_MODE_IS_EXCLUDE;
2775 } else if (mode == MCAST_INCLUDE) {
2776 type = IGMP_MODE_IS_INCLUDE;
2777 KASSERT(inm->inm_st[1].iss_asm == 0,
2778 ("%s: inm %p is INCLUDE but ASM count is %d",
2779 __func__, inm, inm->inm_st[1].iss_asm));
2784 * Generate the filter list changes using a separate function.
2786 if (is_filter_list_change)
2787 return (igmp_v3_enqueue_filter_change(mq, inm));
2789 if (type == IGMP_DO_NOTHING) {
2790 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2791 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2796 * If any sources are present, we must be able to fit at least
2797 * one in the trailing space of the tail packet's mbuf,
2800 minrec0len = sizeof(struct igmp_grouprec);
2801 if (record_has_sources)
2802 minrec0len += sizeof(in_addr_t);
2804 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2805 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2806 inm->inm_ifp->if_xname);
2809 * Check if we have a packet in the tail of the queue for this
2810 * group into which the first group record for this group will fit.
2811 * Otherwise allocate a new packet.
2812 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2813 * Note: Group records for G/GSR query responses MUST be sent
2814 * in their own packet.
2816 m0 = mbufq_last(mq);
2817 if (!is_group_query &&
2819 (m0->m_pkthdr.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2820 (m0->m_pkthdr.len + minrec0len) <
2821 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2822 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2823 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2825 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2827 if (mbufq_full(mq)) {
2828 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2832 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2833 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2834 if (!is_state_change && !is_group_query) {
2835 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2837 m->m_data += IGMP_LEADINGSPACE;
2840 m = m_gethdr(M_NOWAIT, MT_DATA);
2842 M_ALIGN(m, IGMP_LEADINGSPACE);
2847 igmp_save_context(m, ifp);
2849 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2853 * Append group record.
2854 * If we have sources, we don't know how many yet.
2859 ig.ig_group = inm->inm_addr;
2860 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2863 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2866 nbytes += sizeof(struct igmp_grouprec);
2869 * Append as many sources as will fit in the first packet.
2870 * If we are appending to a new packet, the chain allocation
2871 * may potentially use clusters; use m_getptr() in this case.
2872 * If we are appending to an existing packet, we need to obtain
2873 * a pointer to the group record after m_append(), in case a new
2874 * mbuf was allocated.
2875 * Only append sources which are in-mode at t1. If we are
2876 * transitioning to MCAST_UNDEFINED state on the group, do not
2877 * include source entries.
2878 * Only report recorded sources in our filter set when responding
2879 * to a group-source query.
2881 if (record_has_sources) {
2884 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2885 md->m_len - nbytes);
2887 md = m_getptr(m, 0, &off);
2888 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2892 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2893 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2895 now = ims_get_mode(inm, ims, 1);
2896 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2897 if ((now != mode) ||
2898 (now == mode && mode == MCAST_UNDEFINED)) {
2899 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2902 if (is_source_query && ims->ims_stp == 0) {
2903 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2907 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2908 naddr = htonl(ims->ims_haddr);
2909 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2912 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2916 nbytes += sizeof(in_addr_t);
2918 if (msrcs == m0srcs)
2921 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2923 pig->ig_numsrc = htons(msrcs);
2924 nbytes += (msrcs * sizeof(in_addr_t));
2927 if (is_source_query && msrcs == 0) {
2928 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2935 * We are good to go with first packet.
2938 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2939 m->m_pkthdr.vt_nrecs = 1;
2940 mbufq_enqueue(mq, m);
2942 m->m_pkthdr.vt_nrecs++;
2945 * No further work needed if no source list in packet(s).
2947 if (!record_has_sources)
2951 * Whilst sources remain to be announced, we need to allocate
2952 * a new packet and fill out as many sources as will fit.
2953 * Always try for a cluster first.
2955 while (nims != NULL) {
2956 if (mbufq_full(mq)) {
2957 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2960 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2962 m->m_data += IGMP_LEADINGSPACE;
2964 m = m_gethdr(M_NOWAIT, MT_DATA);
2966 M_ALIGN(m, IGMP_LEADINGSPACE);
2970 igmp_save_context(m, ifp);
2971 md = m_getptr(m, 0, &off);
2972 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2973 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2975 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2978 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2981 m->m_pkthdr.vt_nrecs = 1;
2982 nbytes += sizeof(struct igmp_grouprec);
2984 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2985 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2988 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2989 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2991 now = ims_get_mode(inm, ims, 1);
2992 if ((now != mode) ||
2993 (now == mode && mode == MCAST_UNDEFINED)) {
2994 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2997 if (is_source_query && ims->ims_stp == 0) {
2998 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
3002 CTR1(KTR_IGMPV3, "%s: append node", __func__);
3003 naddr = htonl(ims->ims_haddr);
3004 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
3007 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
3012 if (msrcs == m0srcs)
3015 pig->ig_numsrc = htons(msrcs);
3016 nbytes += (msrcs * sizeof(in_addr_t));
3018 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
3019 mbufq_enqueue(mq, m);
3026 * Type used to mark record pass completion.
3027 * We exploit the fact we can cast to this easily from the
3028 * current filter modes on each ip_msource node.
3031 REC_NONE = 0x00, /* MCAST_UNDEFINED */
3032 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
3033 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
3034 REC_FULL = REC_ALLOW | REC_BLOCK
3038 * Enqueue an IGMPv3 filter list change to the given output queue.
3040 * Source list filter state is held in an RB-tree. When the filter list
3041 * for a group is changed without changing its mode, we need to compute
3042 * the deltas between T0 and T1 for each source in the filter set,
3043 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3045 * As we may potentially queue two record types, and the entire R-B tree
3046 * needs to be walked at once, we break this out into its own function
3047 * so we can generate a tightly packed queue of packets.
3049 * XXX This could be written to only use one tree walk, although that makes
3050 * serializing into the mbuf chains a bit harder. For now we do two walks
3051 * which makes things easier on us, and it may or may not be harder on
3054 * If successful the size of all data appended to the queue is returned,
3055 * otherwise an error code less than zero is returned, or zero if
3056 * no record(s) were appended.
3059 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3061 static const int MINRECLEN =
3062 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3064 struct igmp_grouprec ig;
3065 struct igmp_grouprec *pig;
3066 struct ip_msource *ims, *nims;
3067 struct mbuf *m, *m0, *md;
3069 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3073 uint8_t mode, now, then;
3074 rectype_t crt, drt, nrt;
3076 IN_MULTI_LIST_LOCK_ASSERT();
3078 if (inm->inm_nsrc == 0 ||
3079 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3082 ifp = inm->inm_ifp; /* interface */
3083 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3084 crt = REC_NONE; /* current group record type */
3085 drt = REC_NONE; /* mask of completed group record types */
3086 nrt = REC_NONE; /* record type for current node */
3087 m0srcs = 0; /* # source which will fit in current mbuf chain */
3088 nbytes = 0; /* # of bytes appended to group's state-change queue */
3089 npbytes = 0; /* # of bytes appended this packet */
3090 rsrcs = 0; /* # sources encoded in current record */
3091 schanged = 0; /* # nodes encoded in overall filter change */
3093 nallow = 0; /* # of source entries in ALLOW_NEW */
3094 nblock = 0; /* # of source entries in BLOCK_OLD */
3096 nims = NULL; /* next tree node pointer */
3099 * For each possible filter record mode.
3100 * The first kind of source we encounter tells us which
3101 * is the first kind of record we start appending.
3102 * If a node transitioned to UNDEFINED at t1, its mode is treated
3103 * as the inverse of the group's filter mode.
3105 while (drt != REC_FULL) {
3107 m0 = mbufq_last(mq);
3109 (m0->m_pkthdr.vt_nrecs + 1 <=
3110 IGMP_V3_REPORT_MAXRECS) &&
3111 (m0->m_pkthdr.len + MINRECLEN) <
3112 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3114 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3115 sizeof(struct igmp_grouprec)) /
3118 "%s: use previous packet", __func__);
3120 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3122 m->m_data += IGMP_LEADINGSPACE;
3124 m = m_gethdr(M_NOWAIT, MT_DATA);
3126 M_ALIGN(m, IGMP_LEADINGSPACE);
3130 "%s: m_get*() failed", __func__);
3133 m->m_pkthdr.vt_nrecs = 0;
3134 igmp_save_context(m, ifp);
3135 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3136 sizeof(struct igmp_grouprec)) /
3140 "%s: allocated new packet", __func__);
3143 * Append the IGMP group record header to the
3144 * current packet's data area.
3145 * Recalculate pointer to free space for next
3146 * group record, in case m_append() allocated
3147 * a new mbuf or cluster.
3149 memset(&ig, 0, sizeof(ig));
3150 ig.ig_group = inm->inm_addr;
3151 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3155 "%s: m_append() failed", __func__);
3158 npbytes += sizeof(struct igmp_grouprec);
3160 /* new packet; offset in c hain */
3161 md = m_getptr(m, npbytes -
3162 sizeof(struct igmp_grouprec), &off);
3163 pig = (struct igmp_grouprec *)(mtod(md,
3166 /* current packet; offset from last append */
3168 pig = (struct igmp_grouprec *)(mtod(md,
3169 uint8_t *) + md->m_len -
3170 sizeof(struct igmp_grouprec));
3173 * Begin walking the tree for this record type
3174 * pass, or continue from where we left off
3175 * previously if we had to allocate a new packet.
3176 * Only report deltas in-mode at t1.
3177 * We need not report included sources as allowed
3178 * if we are in inclusive mode on the group,
3179 * however the converse is not true.
3183 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3184 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3185 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3186 __func__, ims->ims_haddr);
3187 now = ims_get_mode(inm, ims, 1);
3188 then = ims_get_mode(inm, ims, 0);
3189 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3190 __func__, then, now);
3193 "%s: skip unchanged", __func__);
3196 if (mode == MCAST_EXCLUDE &&
3197 now == MCAST_INCLUDE) {
3199 "%s: skip IN src on EX group",
3203 nrt = (rectype_t)now;
3204 if (nrt == REC_NONE)
3205 nrt = (rectype_t)(~mode & REC_FULL);
3206 if (schanged++ == 0) {
3208 } else if (crt != nrt)
3210 naddr = htonl(ims->ims_haddr);
3211 if (!m_append(m, sizeof(in_addr_t),
3216 "%s: m_append() failed", __func__);
3220 nallow += !!(crt == REC_ALLOW);
3221 nblock += !!(crt == REC_BLOCK);
3223 if (++rsrcs == m0srcs)
3227 * If we did not append any tree nodes on this
3228 * pass, back out of allocations.
3231 npbytes -= sizeof(struct igmp_grouprec);
3234 "%s: m_free(m)", __func__);
3238 "%s: m_adj(m, -ig)", __func__);
3239 m_adj(m, -((int)sizeof(
3240 struct igmp_grouprec)));
3244 npbytes += (rsrcs * sizeof(in_addr_t));
3245 if (crt == REC_ALLOW)
3246 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3247 else if (crt == REC_BLOCK)
3248 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3249 pig->ig_numsrc = htons(rsrcs);
3251 * Count the new group record, and enqueue this
3252 * packet if it wasn't already queued.
3254 m->m_pkthdr.vt_nrecs++;
3256 mbufq_enqueue(mq, m);
3258 } while (nims != NULL);
3260 crt = (~crt & REC_FULL);
3263 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3270 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3273 struct mbuf *m; /* pending state-change */
3274 struct mbuf *m0; /* copy of pending state-change */
3275 struct mbuf *mt; /* last state-change in packet */
3276 int docopy, domerge;
3283 IN_MULTI_LIST_LOCK_ASSERT();
3287 * If there are further pending retransmissions, make a writable
3288 * copy of each queued state-change message before merging.
3290 if (inm->inm_scrv > 0)
3295 if (mbufq_first(gq) == NULL) {
3296 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3301 m = mbufq_first(gq);
3304 * Only merge the report into the current packet if
3305 * there is sufficient space to do so; an IGMPv3 report
3306 * packet may only contain 65,535 group records.
3307 * Always use a simple mbuf chain concatentation to do this,
3308 * as large state changes for single groups may have
3309 * allocated clusters.
3312 mt = mbufq_last(scq);
3314 recslen = m_length(m, NULL);
3316 if ((mt->m_pkthdr.vt_nrecs +
3317 m->m_pkthdr.vt_nrecs <=
3318 IGMP_V3_REPORT_MAXRECS) &&
3319 (mt->m_pkthdr.len + recslen <=
3320 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3324 if (!domerge && mbufq_full(gq)) {
3326 "%s: outbound queue full, skipping whole packet %p",
3336 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3337 m0 = mbufq_dequeue(gq);
3340 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3341 m0 = m_dup(m, M_NOWAIT);
3344 m0->m_nextpkt = NULL;
3349 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3351 mbufq_enqueue(scq, m0);
3353 struct mbuf *mtl; /* last mbuf of packet mt */
3355 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3359 m0->m_flags &= ~M_PKTHDR;
3360 mt->m_pkthdr.len += recslen;
3361 mt->m_pkthdr.vt_nrecs +=
3362 m0->m_pkthdr.vt_nrecs;
3372 * Respond to a pending IGMPv3 General Query.
3375 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3377 struct ifmultiaddr *ifma;
3379 struct in_multi *inm;
3380 int retval __unused, loop;
3382 IN_MULTI_LIST_LOCK_ASSERT();
3386 KASSERT(igi->igi_version == IGMP_VERSION_3,
3387 ("%s: called when version %d", __func__, igi->igi_version));
3390 * Check that there are some packets queued. If so, send them first.
3391 * For large number of groups the reply to general query can take
3392 * many packets, we should finish sending them before starting of
3393 * queuing the new reply.
3395 if (mbufq_len(&igi->igi_gq) != 0)
3400 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3401 if (ifma->ifma_addr->sa_family != AF_INET ||
3402 ifma->ifma_protospec == NULL)
3405 inm = (struct in_multi *)ifma->ifma_protospec;
3406 KASSERT(ifp == inm->inm_ifp,
3407 ("%s: inconsistent ifp", __func__));
3409 switch (inm->inm_state) {
3410 case IGMP_NOT_MEMBER:
3411 case IGMP_SILENT_MEMBER:
3413 case IGMP_REPORTING_MEMBER:
3414 case IGMP_IDLE_MEMBER:
3415 case IGMP_LAZY_MEMBER:
3416 case IGMP_SLEEPING_MEMBER:
3417 case IGMP_AWAKENING_MEMBER:
3418 inm->inm_state = IGMP_REPORTING_MEMBER;
3419 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3421 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3424 case IGMP_G_QUERY_PENDING_MEMBER:
3425 case IGMP_SG_QUERY_PENDING_MEMBER:
3426 case IGMP_LEAVING_MEMBER:
3432 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3433 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3436 * Slew transmission of bursts over 500ms intervals.
3438 if (mbufq_first(&igi->igi_gq) != NULL) {
3439 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3440 IGMP_RESPONSE_BURST_INTERVAL);
3441 V_interface_timers_running = 1;
3446 * Transmit the next pending IGMP message in the output queue.
3448 * We get called from netisr_processqueue(). A mutex private to igmpoq
3449 * will be acquired and released around this routine.
3451 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3452 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3453 * a link and uses a link-scope multicast address.
3456 igmp_intr(struct mbuf *m)
3458 struct ip_moptions imo;
3460 struct mbuf *ipopts, *m0;
3464 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3467 * Set VNET image pointer from enqueued mbuf chain
3468 * before doing anything else. Whilst we use interface
3469 * indexes to guard against interface detach, they are
3470 * unique to each VIMAGE and must be retrieved.
3472 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3473 ifindex = igmp_restore_context(m);
3476 * Check if the ifnet still exists. This limits the scope of
3477 * any race in the absence of a global ifp lock for low cost
3478 * (an array lookup).
3480 ifp = ifnet_byindex(ifindex);
3482 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3483 __func__, m, ifindex);
3485 IPSTAT_INC(ips_noroute);
3489 ipopts = V_igmp_sendra ? m_raopt : NULL;
3491 imo.imo_multicast_ttl = 1;
3492 imo.imo_multicast_vif = -1;
3493 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3496 * If the user requested that IGMP traffic be explicitly
3497 * redirected to the loopback interface (e.g. they are running a
3498 * MANET interface and the routing protocol needs to see the
3499 * updates), handle this now.
3501 if (m->m_flags & M_IGMP_LOOP)
3502 imo.imo_multicast_ifp = V_loif;
3504 imo.imo_multicast_ifp = ifp;
3506 if (m->m_flags & M_IGMPV2) {
3509 m0 = igmp_v3_encap_report(ifp, m);
3511 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3513 IPSTAT_INC(ips_odropped);
3518 igmp_scrub_context(m0);
3520 m0->m_pkthdr.rcvif = V_loif;
3522 mac_netinet_igmp_send(ifp, m0);
3524 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3526 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3530 IGMPSTAT_INC(igps_snd_reports);
3534 * We must restore the existing vnet pointer before
3535 * continuing as we are run from netisr context.
3541 * Encapsulate an IGMPv3 report.
3543 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3544 * chain has already had its IP/IGMPv3 header prepended. In this case
3545 * the function will not attempt to prepend; the lengths and checksums
3546 * will however be re-computed.
3548 * Returns a pointer to the new mbuf chain head, or NULL if the
3549 * allocation failed.
3551 static struct mbuf *
3552 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3554 struct igmp_report *igmp;
3556 int hdrlen, igmpreclen;
3558 KASSERT((m->m_flags & M_PKTHDR),
3559 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3561 igmpreclen = m_length(m, NULL);
3562 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3564 if (m->m_flags & M_IGMPV3_HDR) {
3565 igmpreclen -= hdrlen;
3567 M_PREPEND(m, hdrlen, M_NOWAIT);
3570 m->m_flags |= M_IGMPV3_HDR;
3573 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3575 m->m_data += sizeof(struct ip);
3576 m->m_len -= sizeof(struct ip);
3578 igmp = mtod(m, struct igmp_report *);
3579 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3582 igmp->ir_numgrps = htons(m->m_pkthdr.vt_nrecs);
3584 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3585 m->m_pkthdr.vt_nrecs = 0;
3587 m->m_data -= sizeof(struct ip);
3588 m->m_len += sizeof(struct ip);
3590 ip = mtod(m, struct ip *);
3591 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3592 ip->ip_len = htons(hdrlen + igmpreclen);
3593 ip->ip_off = htons(IP_DF);
3594 ip->ip_p = IPPROTO_IGMP;
3597 ip->ip_src.s_addr = INADDR_ANY;
3599 if (m->m_flags & M_IGMP_LOOP) {
3600 struct in_ifaddr *ia;
3604 ip->ip_src = ia->ia_addr.sin_addr;
3607 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3614 igmp_rec_type_to_str(const int type)
3618 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3621 case IGMP_CHANGE_TO_INCLUDE_MODE:
3624 case IGMP_MODE_IS_EXCLUDE:
3627 case IGMP_MODE_IS_INCLUDE:
3630 case IGMP_ALLOW_NEW_SOURCES:
3633 case IGMP_BLOCK_OLD_SOURCES:
3645 vnet_igmp_init(const void *unused __unused)
3648 netisr_register_vnet(&igmp_nh);
3650 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3651 vnet_igmp_init, NULL);
3654 vnet_igmp_uninit(const void *unused __unused)
3657 /* This can happen when we shutdown the entire network stack. */
3658 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3660 netisr_unregister_vnet(&igmp_nh);
3662 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3663 vnet_igmp_uninit, NULL);
3667 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3669 struct igmp_ifsoftc *igi, *tigi;
3670 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3673 db_printf("usage: show igi_list <addr>\n");
3676 igi_head = (struct _igi_list *)addr;
3678 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3679 db_printf("igmp_ifsoftc %p:\n", igi);
3680 db_printf(" ifp %p\n", igi->igi_ifp);
3681 db_printf(" version %u\n", igi->igi_version);
3682 db_printf(" v1_timer %u\n", igi->igi_v1_timer);
3683 db_printf(" v2_timer %u\n", igi->igi_v2_timer);
3684 db_printf(" v3_timer %u\n", igi->igi_v3_timer);
3685 db_printf(" flags %#x\n", igi->igi_flags);
3686 db_printf(" rv %u\n", igi->igi_rv);
3687 db_printf(" qi %u\n", igi->igi_qi);
3688 db_printf(" qri %u\n", igi->igi_qri);
3689 db_printf(" uri %u\n", igi->igi_uri);
3690 /* struct mbufq igi_gq; */
3697 igmp_modevent(module_t mod, int type, void *unused __unused)
3702 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3704 m_raopt = igmp_ra_alloc();
3705 netisr_register(&igmp_nh);
3706 callout_init(&igmpslow_callout, 1);
3707 callout_reset(&igmpslow_callout, hz / IGMP_SLOWHZ,
3708 igmp_slowtimo, NULL);
3709 callout_init(&igmpfast_callout, 1);
3710 callout_reset(&igmpfast_callout, hz / IGMP_FASTHZ,
3711 igmp_fasttimo, NULL);
3714 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3715 netisr_unregister(&igmp_nh);
3718 IGMP_LOCK_DESTROY();
3721 return (EOPNOTSUPP);
3726 static moduledata_t igmp_mod = {
3731 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);