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);
148 static int sysctl_igmp_stat(SYSCTL_HANDLER_ARGS);
150 static const struct netisr_handler igmp_nh = {
152 .nh_handler = igmp_intr,
153 .nh_proto = NETISR_IGMP,
154 .nh_policy = NETISR_POLICY_SOURCE,
158 * System-wide globals.
160 * Unlocked access to these is OK, except for the global IGMP output
161 * queue. The IGMP subsystem lock ends up being system-wide for the moment,
162 * because all VIMAGEs have to share a global output queue, as netisrs
163 * themselves are not virtualized.
166 * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
167 * Any may be taken independently; if any are held at the same
168 * time, the above lock order must be followed.
169 * * All output is delegated to the netisr.
170 * Now that Giant has been eliminated, the netisr may be inlined.
171 * * IN_MULTI_LIST_LOCK covers in_multi.
172 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file,
173 * including the output queue.
174 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
175 * per-link state iterators.
176 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface,
177 * therefore it is not refcounted.
178 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi.
181 * * IGMP acquires its own reference every time an in_multi is passed to
182 * it and the group is being joined for the first time.
183 * * IGMP releases its reference(s) on in_multi in a deferred way,
184 * because the operations which process the release run as part of
185 * a loop whose control variables are directly affected by the release
186 * (that, and not recursing on the IF_ADDR_LOCK).
188 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
189 * to a vnet in ifp->if_vnet.
191 * SMPng: XXX We may potentially race operations on ifma_protospec.
192 * The problem is that we currently lack a clean way of taking the
193 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
194 * as anything which modifies ifma needs to be covered by that lock.
195 * So check for ifma_protospec being NULL before proceeding.
199 struct mbuf *m_raopt; /* Router Alert option */
200 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
203 * VIMAGE-wide globals.
205 * The IGMPv3 timers themselves need to run per-image, however,
206 * protosw timers run globally (see tcp).
207 * An ifnet can only be in one vimage at a time, and the loopback
208 * ifnet, loif, is itself virtualized.
209 * It would otherwise be possible to seriously hose IGMP state,
210 * and create inconsistencies in upstream multicast routing, if you have
211 * multiple VIMAGEs running on the same link joining different multicast
212 * groups, UNLESS the "primary IP address" is different. This is because
213 * IGMP for IPv4 does not force link-local addresses to be used for each
214 * node, unlike MLD for IPv6.
215 * Obviously the IGMPv3 per-interface state has per-vimage granularity
218 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
219 * policy to control the address used by IGMP on the link.
221 VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general
223 VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change
225 VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host
226 * report; IGMPv3 g/sg
229 #define V_interface_timers_running VNET(interface_timers_running)
230 #define V_state_change_timers_running VNET(state_change_timers_running)
231 #define V_current_state_timers_running VNET(current_state_timers_running)
233 VNET_PCPUSTAT_DEFINE(struct igmpstat, igmpstat);
234 VNET_PCPUSTAT_SYSINIT(igmpstat);
235 VNET_PCPUSTAT_SYSUNINIT(igmpstat);
237 VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) =
238 LIST_HEAD_INITIALIZER(igi_head);
239 VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0};
241 #define V_igi_head VNET(igi_head)
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_PROC(_net_inet_igmp, IGMPCTL_STATS, stats,
264 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE,
265 &VNET_NAME(igmpstat), 0, sysctl_igmp_stat, "S,igmpstat",
266 "IGMP statistics (struct igmpstat, netinet/igmp_var.h)");
267 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW,
268 &VNET_NAME(igmp_recvifkludge), 0,
269 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
270 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW,
271 &VNET_NAME(igmp_sendra), 0,
272 "Send IP Router Alert option in IGMPv2/v3 messages");
273 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW,
274 &VNET_NAME(igmp_sendlocal), 0,
275 "Send IGMP membership reports for 224.0.0.0/24 groups");
276 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW,
277 &VNET_NAME(igmp_v1enable), 0,
278 "Enable backwards compatibility with IGMPv1");
279 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW,
280 &VNET_NAME(igmp_v2enable), 0,
281 "Enable backwards compatibility with IGMPv2");
282 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW,
283 &VNET_NAME(igmp_legacysupp), 0,
284 "Allow v1/v2 reports to suppress v3 group responses");
285 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version,
286 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
287 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I",
288 "Default version of IGMP to run on each interface");
289 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay,
290 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
291 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I",
292 "Rate limit for IGMPv3 Group-and-Source queries in seconds");
295 * Non-virtualized sysctls.
297 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo,
298 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo,
299 "Per-interface IGMPv3 state");
302 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
306 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
308 m->m_pkthdr.rcvif = ifp;
309 m->m_pkthdr.flowid = ifp->if_index;
313 igmp_scrub_context(struct mbuf *m)
316 m->m_pkthdr.PH_loc.ptr = NULL;
317 m->m_pkthdr.flowid = 0;
321 * Restore context from a queued IGMP output chain.
322 * Return saved ifindex.
324 * VIMAGE: The assertion is there to make sure that we
325 * actually called CURVNET_SET() with what's in the mbuf chain.
327 static __inline uint32_t
328 igmp_restore_context(struct mbuf *m)
332 #if defined(VIMAGE) && defined(INVARIANTS)
333 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr),
334 ("%s: called when curvnet was not restored", __func__));
337 return (m->m_pkthdr.flowid);
344 sysctl_igmp_stat(SYSCTL_HANDLER_ARGS)
346 struct igmpstat igps0;
350 error = sysctl_wire_old_buffer(req, sizeof(struct igmpstat));
354 if (req->oldptr != NULL) {
355 if (req->oldlen < sizeof(struct igmpstat))
359 * Copy the counters, and explicitly set the struct's
360 * version and length fields.
362 COUNTER_ARRAY_COPY(VNET(igmpstat), &igps0,
363 sizeof(struct igmpstat) / sizeof(uint64_t));
364 igps0.igps_version = IGPS_VERSION_3;
365 igps0.igps_len = IGPS_VERSION3_LEN;
366 error = SYSCTL_OUT(req, &igps0,
367 sizeof(struct igmpstat));
370 req->validlen = sizeof(struct igmpstat);
373 if (req->newptr != NULL) {
374 if (req->newlen < sizeof(struct igmpstat))
377 error = SYSCTL_IN(req, &igps0,
382 * igps0 must be "all zero".
385 while (p < (char *)&igps0 + sizeof(igps0) && *p == '\0')
387 if (p != (char *)&igps0 + sizeof(igps0)) {
391 COUNTER_ARRAY_ZERO(VNET(igmpstat),
392 sizeof(struct igmpstat) / sizeof(uint64_t));
399 * Retrieve or set default IGMP version.
401 * VIMAGE: Assume curvnet set by caller.
402 * SMPng: NOTE: Serialized by IGMP lock.
405 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
410 error = sysctl_wire_old_buffer(req, sizeof(int));
416 new = V_igmp_default_version;
418 error = sysctl_handle_int(oidp, &new, 0, req);
419 if (error || !req->newptr)
422 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
427 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
428 V_igmp_default_version, new);
430 V_igmp_default_version = new;
438 * Retrieve or set threshold between group-source queries in seconds.
440 * VIMAGE: Assume curvnet set by caller.
441 * SMPng: NOTE: Serialized by IGMP lock.
444 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
449 error = sysctl_wire_old_buffer(req, sizeof(int));
455 i = V_igmp_gsrdelay.tv_sec;
457 error = sysctl_handle_int(oidp, &i, 0, req);
458 if (error || !req->newptr)
461 if (i < -1 || i >= 60) {
466 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
467 V_igmp_gsrdelay.tv_sec, i);
468 V_igmp_gsrdelay.tv_sec = i;
476 * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
477 * For use by ifmcstat(8).
479 * SMPng: NOTE: Does an unlocked ifindex space read.
480 * VIMAGE: Assume curvnet set by caller. The node handler itself
481 * is not directly virtualized.
484 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
490 struct igmp_ifsoftc *igi;
495 if (req->newptr != NULL)
501 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
505 IN_MULTI_LIST_LOCK();
508 if (name[0] <= 0 || name[0] > V_if_index) {
515 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 epoch_tracker et;
677 struct igmp_ifsoftc *igi;
678 struct ifmultiaddr *ifma;
679 struct in_multi *inm;
680 struct in_multi_head inm_free_tmp;
681 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
684 SLIST_INIT(&inm_free_tmp);
687 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
688 if (igi->igi_version == IGMP_VERSION_3) {
691 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
692 inm = inm_ifmultiaddr_get_inm(ifma);
695 if (inm->inm_state == IGMP_LEAVING_MEMBER)
696 inm_rele_locked(&inm_free_tmp, inm);
697 inm_clear_recorded(inm);
700 IF_ADDR_WUNLOCK(ifp);
701 inm_release_list_deferred(&inm_free_tmp);
708 * Hook for domifdetach.
711 igmp_domifdetach(struct ifnet *ifp)
714 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
715 __func__, ifp, ifp->if_xname);
718 igi_delete_locked(ifp);
723 igi_delete_locked(const struct ifnet *ifp)
725 struct igmp_ifsoftc *igi, *tigi;
727 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
728 __func__, ifp, ifp->if_xname);
732 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
733 if (igi->igi_ifp == ifp) {
735 * Free deferred General Query responses.
737 mbufq_drain(&igi->igi_gq);
739 LIST_REMOVE(igi, igi_link);
747 * Process a received IGMPv1 query.
748 * Return non-zero if the message should be dropped.
750 * VIMAGE: The curvnet pointer is derived from the input ifp.
753 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
754 const struct igmp *igmp)
756 struct ifmultiaddr *ifma;
757 struct igmp_ifsoftc *igi;
758 struct in_multi *inm;
763 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
764 * 224.0.0.1. They are always treated as General Queries.
765 * igmp_group is always ignored. Do not drop it as a userland
766 * daemon may wish to see it.
767 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
769 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
770 IGMPSTAT_INC(igps_rcv_badqueries);
773 IGMPSTAT_INC(igps_rcv_gen_queries);
775 IN_MULTI_LIST_LOCK();
778 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
779 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
781 if (igi->igi_flags & IGIF_LOOPBACK) {
782 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
788 * Switch to IGMPv1 host compatibility mode.
790 igmp_set_version(igi, IGMP_VERSION_1);
792 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
795 * Start the timers in all of our group records
796 * for the interface on which the query arrived,
797 * except those which are already running.
799 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
800 inm = inm_ifmultiaddr_get_inm(ifma);
803 if (inm->inm_timer != 0)
805 switch (inm->inm_state) {
806 case IGMP_NOT_MEMBER:
807 case IGMP_SILENT_MEMBER:
809 case IGMP_G_QUERY_PENDING_MEMBER:
810 case IGMP_SG_QUERY_PENDING_MEMBER:
811 case IGMP_REPORTING_MEMBER:
812 case IGMP_IDLE_MEMBER:
813 case IGMP_LAZY_MEMBER:
814 case IGMP_SLEEPING_MEMBER:
815 case IGMP_AWAKENING_MEMBER:
816 inm->inm_state = IGMP_REPORTING_MEMBER;
817 inm->inm_timer = IGMP_RANDOM_DELAY(
818 IGMP_V1V2_MAX_RI * PR_FASTHZ);
819 V_current_state_timers_running = 1;
821 case IGMP_LEAVING_MEMBER:
828 IN_MULTI_LIST_UNLOCK();
834 * Process a received IGMPv2 general or group-specific query.
837 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
838 const struct igmp *igmp)
840 struct ifmultiaddr *ifma;
841 struct igmp_ifsoftc *igi;
842 struct in_multi *inm;
843 int is_general_query;
848 is_general_query = 0;
851 * Validate address fields upfront.
852 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
854 if (in_nullhost(igmp->igmp_group)) {
856 * IGMPv2 General Query.
857 * If this was not sent to the all-hosts group, ignore it.
859 if (!in_allhosts(ip->ip_dst))
861 IGMPSTAT_INC(igps_rcv_gen_queries);
862 is_general_query = 1;
864 /* IGMPv2 Group-Specific Query. */
865 IGMPSTAT_INC(igps_rcv_group_queries);
868 IN_MULTI_LIST_LOCK();
871 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
872 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
874 if (igi->igi_flags & IGIF_LOOPBACK) {
875 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
881 * Ignore v2 query if in v1 Compatibility Mode.
883 if (igi->igi_version == IGMP_VERSION_1)
886 igmp_set_version(igi, IGMP_VERSION_2);
888 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
892 if (is_general_query) {
894 * For each reporting group joined on this
895 * interface, kick the report timer.
897 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
899 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
900 inm = inm_ifmultiaddr_get_inm(ifma);
903 igmp_v2_update_group(inm, timer);
907 * Group-specific IGMPv2 query, we need only
908 * look up the single group to process it.
910 inm = inm_lookup(ifp, igmp->igmp_group);
913 "process v2 query 0x%08x on ifp %p(%s)",
914 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
915 igmp_v2_update_group(inm, timer);
921 IN_MULTI_LIST_UNLOCK();
927 * Update the report timer on a group in response to an IGMPv2 query.
929 * If we are becoming the reporting member for this group, start the timer.
930 * If we already are the reporting member for this group, and timer is
931 * below the threshold, reset it.
933 * We may be updating the group for the first time since we switched
934 * to IGMPv3. If we are, then we must clear any recorded source lists,
935 * and transition to REPORTING state; the group timer is overloaded
936 * for group and group-source query responses.
938 * Unlike IGMPv3, the delay per group should be jittered
939 * to avoid bursts of IGMPv2 reports.
942 igmp_v2_update_group(struct in_multi *inm, const int timer)
945 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__,
946 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer);
948 IN_MULTI_LIST_LOCK_ASSERT();
950 switch (inm->inm_state) {
951 case IGMP_NOT_MEMBER:
952 case IGMP_SILENT_MEMBER:
954 case IGMP_REPORTING_MEMBER:
955 if (inm->inm_timer != 0 &&
956 inm->inm_timer <= timer) {
957 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
958 "skipping.", __func__);
962 case IGMP_SG_QUERY_PENDING_MEMBER:
963 case IGMP_G_QUERY_PENDING_MEMBER:
964 case IGMP_IDLE_MEMBER:
965 case IGMP_LAZY_MEMBER:
966 case IGMP_AWAKENING_MEMBER:
967 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
968 inm->inm_state = IGMP_REPORTING_MEMBER;
969 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
970 V_current_state_timers_running = 1;
972 case IGMP_SLEEPING_MEMBER:
973 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
974 inm->inm_state = IGMP_AWAKENING_MEMBER;
976 case IGMP_LEAVING_MEMBER:
982 * Process a received IGMPv3 general, group-specific or
983 * group-and-source-specific query.
984 * Assumes m has already been pulled up to the full IGMP message length.
985 * Return 0 if successful, otherwise an appropriate error code is returned.
988 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
989 /*const*/ struct igmpv3 *igmpv3)
991 struct igmp_ifsoftc *igi;
992 struct in_multi *inm;
993 int is_general_query;
994 uint32_t maxresp, nsrc, qqi;
998 is_general_query = 0;
1000 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
1002 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
1003 if (maxresp >= 128) {
1004 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
1005 (IGMP_EXP(igmpv3->igmp_code) + 3);
1009 * Robustness must never be less than 2 for on-wire IGMPv3.
1010 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
1011 * an exception for interfaces whose IGMPv3 state changes
1012 * are redirected to loopback (e.g. MANET).
1014 qrv = IGMP_QRV(igmpv3->igmp_misc);
1016 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
1021 qqi = igmpv3->igmp_qqi;
1023 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
1024 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
1027 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
1031 nsrc = ntohs(igmpv3->igmp_numsrc);
1034 * Validate address fields and versions upfront before
1035 * accepting v3 query.
1036 * XXX SMPng: Unlocked access to igmpstat counters here.
1038 if (in_nullhost(igmpv3->igmp_group)) {
1040 * IGMPv3 General Query.
1042 * General Queries SHOULD be directed to 224.0.0.1.
1043 * A general query with a source list has undefined
1044 * behaviour; discard it.
1046 IGMPSTAT_INC(igps_rcv_gen_queries);
1047 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1048 IGMPSTAT_INC(igps_rcv_badqueries);
1051 is_general_query = 1;
1053 /* Group or group-source specific query. */
1055 IGMPSTAT_INC(igps_rcv_group_queries);
1057 IGMPSTAT_INC(igps_rcv_gsr_queries);
1060 IN_MULTI_LIST_LOCK();
1063 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1064 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
1066 if (igi->igi_flags & IGIF_LOOPBACK) {
1067 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1068 ifp, ifp->if_xname);
1073 * Discard the v3 query if we're in Compatibility Mode.
1074 * The RFC is not obviously worded that hosts need to stay in
1075 * compatibility mode until the Old Version Querier Present
1078 if (igi->igi_version != IGMP_VERSION_3) {
1079 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1080 igi->igi_version, ifp, ifp->if_xname);
1084 igmp_set_version(igi, IGMP_VERSION_3);
1087 igi->igi_qri = maxresp;
1089 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1092 if (is_general_query) {
1094 * Schedule a current-state report on this ifp for
1095 * all groups, possibly containing source lists.
1096 * If there is a pending General Query response
1097 * scheduled earlier than the selected delay, do
1098 * not schedule any other reports.
1099 * Otherwise, reset the interface timer.
1101 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1102 ifp, ifp->if_xname);
1103 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1104 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1105 V_interface_timers_running = 1;
1109 * Group-source-specific queries are throttled on
1110 * a per-group basis to defeat denial-of-service attempts.
1111 * Queries for groups we are not a member of on this
1112 * link are simply ignored.
1114 inm = inm_lookup(ifp, igmpv3->igmp_group);
1118 if (!ratecheck(&inm->inm_lastgsrtv,
1119 &V_igmp_gsrdelay)) {
1120 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1122 IGMPSTAT_INC(igps_drop_gsr_queries);
1126 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)",
1127 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname);
1129 * If there is a pending General Query response
1130 * scheduled sooner than the selected delay, no
1131 * further report need be scheduled.
1132 * Otherwise, prepare to respond to the
1133 * group-specific or group-and-source query.
1135 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1136 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1141 IN_MULTI_LIST_UNLOCK();
1147 * Process a received IGMPv3 group-specific or group-and-source-specific
1149 * Return <0 if any error occurred. Currently this is ignored.
1152 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi,
1153 int timer, /*const*/ struct igmpv3 *igmpv3)
1158 IN_MULTI_LIST_LOCK_ASSERT();
1163 switch (inm->inm_state) {
1164 case IGMP_NOT_MEMBER:
1165 case IGMP_SILENT_MEMBER:
1166 case IGMP_SLEEPING_MEMBER:
1167 case IGMP_LAZY_MEMBER:
1168 case IGMP_AWAKENING_MEMBER:
1169 case IGMP_IDLE_MEMBER:
1170 case IGMP_LEAVING_MEMBER:
1173 case IGMP_REPORTING_MEMBER:
1174 case IGMP_G_QUERY_PENDING_MEMBER:
1175 case IGMP_SG_QUERY_PENDING_MEMBER:
1179 nsrc = ntohs(igmpv3->igmp_numsrc);
1182 * Deal with group-specific queries upfront.
1183 * If any group query is already pending, purge any recorded
1184 * source-list state if it exists, and schedule a query response
1185 * for this group-specific query.
1188 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1189 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1190 inm_clear_recorded(inm);
1191 timer = min(inm->inm_timer, timer);
1193 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1194 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1195 V_current_state_timers_running = 1;
1200 * Deal with the case where a group-and-source-specific query has
1201 * been received but a group-specific query is already pending.
1203 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1204 timer = min(inm->inm_timer, timer);
1205 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1206 V_current_state_timers_running = 1;
1211 * Finally, deal with the case where a group-and-source-specific
1212 * query has been received, where a response to a previous g-s-r
1213 * query exists, or none exists.
1214 * In this case, we need to parse the source-list which the Querier
1215 * has provided us with and check if we have any source list filter
1216 * entries at T1 for these sources. If we do not, there is no need
1217 * schedule a report and the query may be dropped.
1218 * If we do, we must record them and schedule a current-state
1219 * report for those sources.
1220 * FIXME: Handling source lists larger than 1 mbuf requires that
1221 * we pass the mbuf chain pointer down to this function, and use
1222 * m_getptr() to walk the chain.
1224 if (inm->inm_nsrc > 0) {
1225 const struct in_addr *ap;
1228 ap = (const struct in_addr *)(igmpv3 + 1);
1230 for (i = 0; i < nsrc; i++, ap++) {
1231 retval = inm_record_source(inm, ap->s_addr);
1234 nrecorded += retval;
1236 if (nrecorded > 0) {
1238 "%s: schedule response to SG query", __func__);
1239 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1240 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1241 V_current_state_timers_running = 1;
1249 * Process a received IGMPv1 host membership report.
1251 * NOTE: 0.0.0.0 workaround breaks const correctness.
1254 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1255 /*const*/ struct igmp *igmp)
1257 struct rm_priotracker in_ifa_tracker;
1258 struct in_ifaddr *ia;
1259 struct in_multi *inm;
1261 IGMPSTAT_INC(igps_rcv_reports);
1263 if (ifp->if_flags & IFF_LOOPBACK)
1266 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1267 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1268 IGMPSTAT_INC(igps_rcv_badreports);
1273 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1274 * Booting clients may use the source address 0.0.0.0. Some
1275 * IGMP daemons may not know how to use IP_RECVIF to determine
1276 * the interface upon which this message was received.
1277 * Replace 0.0.0.0 with the subnet address if told to do so.
1279 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1280 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1282 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1285 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)",
1286 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1289 * IGMPv1 report suppression.
1290 * If we are a member of this group, and our membership should be
1291 * reported, stop our group timer and transition to the 'lazy' state.
1293 IN_MULTI_LIST_LOCK();
1294 inm = inm_lookup(ifp, igmp->igmp_group);
1296 struct igmp_ifsoftc *igi;
1300 KASSERT(igi != NULL,
1301 ("%s: no igi for ifp %p", __func__, ifp));
1305 IGMPSTAT_INC(igps_rcv_ourreports);
1308 * If we are in IGMPv3 host mode, do not allow the
1309 * other host's IGMPv1 report to suppress our reports
1310 * unless explicitly configured to do so.
1312 if (igi->igi_version == IGMP_VERSION_3) {
1313 if (V_igmp_legacysupp)
1314 igmp_v3_suppress_group_record(inm);
1320 switch (inm->inm_state) {
1321 case IGMP_NOT_MEMBER:
1322 case IGMP_SILENT_MEMBER:
1324 case IGMP_IDLE_MEMBER:
1325 case IGMP_LAZY_MEMBER:
1326 case IGMP_AWAKENING_MEMBER:
1328 "report suppressed for 0x%08x on ifp %p(%s)",
1329 ntohl(igmp->igmp_group.s_addr), ifp,
1331 case IGMP_SLEEPING_MEMBER:
1332 inm->inm_state = IGMP_SLEEPING_MEMBER;
1334 case IGMP_REPORTING_MEMBER:
1336 "report suppressed for 0x%08x on ifp %p(%s)",
1337 ntohl(igmp->igmp_group.s_addr), ifp,
1339 if (igi->igi_version == IGMP_VERSION_1)
1340 inm->inm_state = IGMP_LAZY_MEMBER;
1341 else if (igi->igi_version == IGMP_VERSION_2)
1342 inm->inm_state = IGMP_SLEEPING_MEMBER;
1344 case IGMP_G_QUERY_PENDING_MEMBER:
1345 case IGMP_SG_QUERY_PENDING_MEMBER:
1346 case IGMP_LEAVING_MEMBER:
1352 IN_MULTI_LIST_UNLOCK();
1358 * Process a received IGMPv2 host membership report.
1360 * NOTE: 0.0.0.0 workaround breaks const correctness.
1363 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1364 /*const*/ struct igmp *igmp)
1366 struct rm_priotracker in_ifa_tracker;
1367 struct in_ifaddr *ia;
1368 struct in_multi *inm;
1371 * Make sure we don't hear our own membership report. Fast
1372 * leave requires knowing that we are the only member of a
1375 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1376 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) {
1380 IGMPSTAT_INC(igps_rcv_reports);
1382 if (ifp->if_flags & IFF_LOOPBACK) {
1386 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1387 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1388 IGMPSTAT_INC(igps_rcv_badreports);
1393 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1394 * Booting clients may use the source address 0.0.0.0. Some
1395 * IGMP daemons may not know how to use IP_RECVIF to determine
1396 * the interface upon which this message was received.
1397 * Replace 0.0.0.0 with the subnet address if told to do so.
1399 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1401 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1404 CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)",
1405 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1408 * IGMPv2 report suppression.
1409 * If we are a member of this group, and our membership should be
1410 * reported, and our group timer is pending or about to be reset,
1411 * stop our group timer by transitioning to the 'lazy' state.
1413 IN_MULTI_LIST_LOCK();
1414 inm = inm_lookup(ifp, igmp->igmp_group);
1416 struct igmp_ifsoftc *igi;
1419 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1421 IGMPSTAT_INC(igps_rcv_ourreports);
1424 * If we are in IGMPv3 host mode, do not allow the
1425 * other host's IGMPv1 report to suppress our reports
1426 * unless explicitly configured to do so.
1428 if (igi->igi_version == IGMP_VERSION_3) {
1429 if (V_igmp_legacysupp)
1430 igmp_v3_suppress_group_record(inm);
1436 switch (inm->inm_state) {
1437 case IGMP_NOT_MEMBER:
1438 case IGMP_SILENT_MEMBER:
1439 case IGMP_SLEEPING_MEMBER:
1441 case IGMP_REPORTING_MEMBER:
1442 case IGMP_IDLE_MEMBER:
1443 case IGMP_AWAKENING_MEMBER:
1445 "report suppressed for 0x%08x on ifp %p(%s)",
1446 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname);
1447 case IGMP_LAZY_MEMBER:
1448 inm->inm_state = IGMP_LAZY_MEMBER;
1450 case IGMP_G_QUERY_PENDING_MEMBER:
1451 case IGMP_SG_QUERY_PENDING_MEMBER:
1452 case IGMP_LEAVING_MEMBER:
1458 IN_MULTI_LIST_UNLOCK();
1464 igmp_input(struct mbuf **mp, int *offp, int proto)
1475 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp);
1478 ifp = m->m_pkthdr.rcvif;
1481 IGMPSTAT_INC(igps_rcv_total);
1483 ip = mtod(m, struct ip *);
1485 igmplen = ntohs(ip->ip_len) - iphlen;
1490 if (igmplen < IGMP_MINLEN) {
1491 IGMPSTAT_INC(igps_rcv_tooshort);
1493 return (IPPROTO_DONE);
1497 * Always pullup to the minimum size for v1/v2 or v3
1498 * to amortize calls to m_pullup().
1501 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1502 minlen += IGMP_V3_QUERY_MINLEN;
1504 minlen += IGMP_MINLEN;
1505 if ((!M_WRITABLE(m) || m->m_len < minlen) &&
1506 (m = m_pullup(m, minlen)) == NULL) {
1507 IGMPSTAT_INC(igps_rcv_tooshort);
1508 return (IPPROTO_DONE);
1510 ip = mtod(m, struct ip *);
1513 * Validate checksum.
1515 m->m_data += iphlen;
1517 igmp = mtod(m, struct igmp *);
1518 if (in_cksum(m, igmplen)) {
1519 IGMPSTAT_INC(igps_rcv_badsum);
1521 return (IPPROTO_DONE);
1523 m->m_data -= iphlen;
1527 * IGMP control traffic is link-scope, and must have a TTL of 1.
1528 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception;
1529 * probe packets may come from beyond the LAN.
1531 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) {
1532 IGMPSTAT_INC(igps_rcv_badttl);
1534 return (IPPROTO_DONE);
1537 switch (igmp->igmp_type) {
1538 case IGMP_HOST_MEMBERSHIP_QUERY:
1539 if (igmplen == IGMP_MINLEN) {
1540 if (igmp->igmp_code == 0)
1541 queryver = IGMP_VERSION_1;
1543 queryver = IGMP_VERSION_2;
1544 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1545 queryver = IGMP_VERSION_3;
1547 IGMPSTAT_INC(igps_rcv_tooshort);
1549 return (IPPROTO_DONE);
1553 case IGMP_VERSION_1:
1554 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1555 if (!V_igmp_v1enable)
1557 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1559 return (IPPROTO_DONE);
1563 case IGMP_VERSION_2:
1564 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1565 if (!V_igmp_v2enable)
1567 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1569 return (IPPROTO_DONE);
1573 case IGMP_VERSION_3: {
1574 struct igmpv3 *igmpv3;
1578 IGMPSTAT_INC(igps_rcv_v3_queries);
1579 igmpv3 = (struct igmpv3 *)igmp;
1581 * Validate length based on source count.
1583 nsrc = ntohs(igmpv3->igmp_numsrc);
1584 if (nsrc * sizeof(in_addr_t) >
1585 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) {
1586 IGMPSTAT_INC(igps_rcv_tooshort);
1588 return (IPPROTO_DONE);
1591 * m_pullup() may modify m, so pullup in
1594 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1595 sizeof(struct in_addr) * nsrc;
1596 if ((!M_WRITABLE(m) ||
1597 m->m_len < igmpv3len) &&
1598 (m = m_pullup(m, igmpv3len)) == NULL) {
1599 IGMPSTAT_INC(igps_rcv_tooshort);
1600 return (IPPROTO_DONE);
1602 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1604 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1606 return (IPPROTO_DONE);
1613 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1614 if (!V_igmp_v1enable)
1616 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1618 return (IPPROTO_DONE);
1622 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1623 if (!V_igmp_v2enable)
1625 if (!ip_checkrouteralert(m))
1626 IGMPSTAT_INC(igps_rcv_nora);
1627 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1629 return (IPPROTO_DONE);
1633 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1635 * Hosts do not need to process IGMPv3 membership reports,
1636 * as report suppression is no longer required.
1638 if (!ip_checkrouteralert(m))
1639 IGMPSTAT_INC(igps_rcv_nora);
1647 * Pass all valid IGMP packets up to any process(es) listening on a
1651 return (rip_input(mp, offp, proto));
1655 * Fast timeout handler (global).
1656 * VIMAGE: Timeout handlers are expected to service all vimages.
1661 VNET_ITERATOR_DECL(vnet_iter);
1663 VNET_LIST_RLOCK_NOSLEEP();
1664 VNET_FOREACH(vnet_iter) {
1665 CURVNET_SET(vnet_iter);
1666 igmp_fasttimo_vnet();
1669 VNET_LIST_RUNLOCK_NOSLEEP();
1673 * Fast timeout handler (per-vnet).
1674 * Sends are shuffled off to a netisr to deal with Giant.
1676 * VIMAGE: Assume caller has set up our curvnet.
1679 igmp_fasttimo_vnet(void)
1681 struct mbufq scq; /* State-change packets */
1682 struct mbufq qrq; /* Query response packets */
1684 struct igmp_ifsoftc *igi;
1685 struct ifmultiaddr *ifma;
1686 struct in_multi *inm;
1687 struct in_multi_head inm_free_tmp;
1688 int loop, uri_fasthz;
1694 * Quick check to see if any work needs to be done, in order to
1695 * minimize the overhead of fasttimo processing.
1696 * SMPng: XXX Unlocked reads.
1698 if (!V_current_state_timers_running &&
1699 !V_interface_timers_running &&
1700 !V_state_change_timers_running)
1703 SLIST_INIT(&inm_free_tmp);
1704 IN_MULTI_LIST_LOCK();
1708 * IGMPv3 General Query response timer processing.
1710 if (V_interface_timers_running) {
1711 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1713 V_interface_timers_running = 0;
1714 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1715 if (igi->igi_v3_timer == 0) {
1717 } else if (--igi->igi_v3_timer == 0) {
1718 igmp_v3_dispatch_general_query(igi);
1720 V_interface_timers_running = 1;
1725 if (!V_current_state_timers_running &&
1726 !V_state_change_timers_running)
1729 V_current_state_timers_running = 0;
1730 V_state_change_timers_running = 0;
1732 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1735 * IGMPv1/v2/v3 host report and state-change timer processing.
1736 * Note: Processing a v3 group timer may remove a node.
1738 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1741 if (igi->igi_version == IGMP_VERSION_3) {
1742 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1743 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1745 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS);
1746 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1750 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1751 inm = inm_ifmultiaddr_get_inm(ifma);
1754 switch (igi->igi_version) {
1755 case IGMP_VERSION_1:
1756 case IGMP_VERSION_2:
1757 igmp_v1v2_process_group_timer(inm,
1760 case IGMP_VERSION_3:
1761 igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1762 &scq, inm, uri_fasthz);
1766 IF_ADDR_WUNLOCK(ifp);
1768 if (igi->igi_version == IGMP_VERSION_3) {
1769 igmp_dispatch_queue(&qrq, 0, loop);
1770 igmp_dispatch_queue(&scq, 0, loop);
1773 * Free the in_multi reference(s) for this
1776 inm_release_list_deferred(&inm_free_tmp);
1782 IN_MULTI_LIST_UNLOCK();
1786 * Update host report group timer for IGMPv1/v2.
1787 * Will update the global pending timer flags.
1790 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1792 int report_timer_expired;
1794 IN_MULTI_LIST_LOCK_ASSERT();
1797 if (inm->inm_timer == 0) {
1798 report_timer_expired = 0;
1799 } else if (--inm->inm_timer == 0) {
1800 report_timer_expired = 1;
1802 V_current_state_timers_running = 1;
1806 switch (inm->inm_state) {
1807 case IGMP_NOT_MEMBER:
1808 case IGMP_SILENT_MEMBER:
1809 case IGMP_IDLE_MEMBER:
1810 case IGMP_LAZY_MEMBER:
1811 case IGMP_SLEEPING_MEMBER:
1812 case IGMP_AWAKENING_MEMBER:
1814 case IGMP_REPORTING_MEMBER:
1815 if (report_timer_expired) {
1816 inm->inm_state = IGMP_IDLE_MEMBER;
1817 (void)igmp_v1v2_queue_report(inm,
1818 (version == IGMP_VERSION_2) ?
1819 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1820 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1823 case IGMP_G_QUERY_PENDING_MEMBER:
1824 case IGMP_SG_QUERY_PENDING_MEMBER:
1825 case IGMP_LEAVING_MEMBER:
1831 * Update a group's timers for IGMPv3.
1832 * Will update the global pending timer flags.
1833 * Note: Unlocked read from igi.
1836 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1837 struct mbufq *qrq, struct mbufq *scq,
1838 struct in_multi *inm, const int uri_fasthz)
1840 int query_response_timer_expired;
1841 int state_change_retransmit_timer_expired;
1843 IN_MULTI_LIST_LOCK_ASSERT();
1846 query_response_timer_expired = 0;
1847 state_change_retransmit_timer_expired = 0;
1850 * During a transition from v1/v2 compatibility mode back to v3,
1851 * a group record in REPORTING state may still have its group
1852 * timer active. This is a no-op in this function; it is easier
1853 * to deal with it here than to complicate the slow-timeout path.
1855 if (inm->inm_timer == 0) {
1856 query_response_timer_expired = 0;
1857 } else if (--inm->inm_timer == 0) {
1858 query_response_timer_expired = 1;
1860 V_current_state_timers_running = 1;
1863 if (inm->inm_sctimer == 0) {
1864 state_change_retransmit_timer_expired = 0;
1865 } else if (--inm->inm_sctimer == 0) {
1866 state_change_retransmit_timer_expired = 1;
1868 V_state_change_timers_running = 1;
1871 /* We are in fasttimo, so be quick about it. */
1872 if (!state_change_retransmit_timer_expired &&
1873 !query_response_timer_expired)
1876 switch (inm->inm_state) {
1877 case IGMP_NOT_MEMBER:
1878 case IGMP_SILENT_MEMBER:
1879 case IGMP_SLEEPING_MEMBER:
1880 case IGMP_LAZY_MEMBER:
1881 case IGMP_AWAKENING_MEMBER:
1882 case IGMP_IDLE_MEMBER:
1884 case IGMP_G_QUERY_PENDING_MEMBER:
1885 case IGMP_SG_QUERY_PENDING_MEMBER:
1887 * Respond to a previously pending Group-Specific
1888 * or Group-and-Source-Specific query by enqueueing
1889 * the appropriate Current-State report for
1890 * immediate transmission.
1892 if (query_response_timer_expired) {
1893 int retval __unused;
1895 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1896 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1897 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1899 inm->inm_state = IGMP_REPORTING_MEMBER;
1900 /* XXX Clear recorded sources for next time. */
1901 inm_clear_recorded(inm);
1904 case IGMP_REPORTING_MEMBER:
1905 case IGMP_LEAVING_MEMBER:
1906 if (state_change_retransmit_timer_expired) {
1908 * State-change retransmission timer fired.
1909 * If there are any further pending retransmissions,
1910 * set the global pending state-change flag, and
1913 if (--inm->inm_scrv > 0) {
1914 inm->inm_sctimer = uri_fasthz;
1915 V_state_change_timers_running = 1;
1918 * Retransmit the previously computed state-change
1919 * report. If there are no further pending
1920 * retransmissions, the mbuf queue will be consumed.
1921 * Update T0 state to T1 as we have now sent
1924 (void)igmp_v3_merge_state_changes(inm, scq);
1927 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1928 ntohl(inm->inm_addr.s_addr),
1929 inm->inm_ifp->if_xname);
1932 * If we are leaving the group for good, make sure
1933 * we release IGMP's reference to it.
1934 * This release must be deferred using a SLIST,
1935 * as we are called from a loop which traverses
1936 * the in_ifmultiaddr TAILQ.
1938 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1939 inm->inm_scrv == 0) {
1940 inm->inm_state = IGMP_NOT_MEMBER;
1941 inm_rele_locked(inmh, inm);
1949 * Suppress a group's pending response to a group or source/group query.
1951 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1952 * Do NOT update ST1/ST0 as this operation merely suppresses
1953 * the currently pending group record.
1954 * Do NOT suppress the response to a general query. It is possible but
1955 * it would require adding another state or flag.
1958 igmp_v3_suppress_group_record(struct in_multi *inm)
1961 IN_MULTI_LIST_LOCK_ASSERT();
1963 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1964 ("%s: not IGMPv3 mode on link", __func__));
1966 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1967 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1970 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1971 inm_clear_recorded(inm);
1974 inm->inm_state = IGMP_REPORTING_MEMBER;
1978 * Switch to a different IGMP version on the given interface,
1979 * as per Section 7.2.1.
1982 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1984 int old_version_timer;
1988 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1989 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1991 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1993 * Compute the "Older Version Querier Present" timer as per
1996 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1997 old_version_timer *= PR_SLOWHZ;
1999 if (version == IGMP_VERSION_1) {
2000 igi->igi_v1_timer = old_version_timer;
2001 igi->igi_v2_timer = 0;
2002 } else if (version == IGMP_VERSION_2) {
2003 igi->igi_v1_timer = 0;
2004 igi->igi_v2_timer = old_version_timer;
2008 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2009 if (igi->igi_version != IGMP_VERSION_2) {
2010 igi->igi_version = IGMP_VERSION_2;
2011 igmp_v3_cancel_link_timers(igi);
2013 } else if (igi->igi_v1_timer > 0) {
2014 if (igi->igi_version != IGMP_VERSION_1) {
2015 igi->igi_version = IGMP_VERSION_1;
2016 igmp_v3_cancel_link_timers(igi);
2022 * Cancel pending IGMPv3 timers for the given link and all groups
2023 * joined on it; state-change, general-query, and group-query timers.
2025 * Only ever called on a transition from v3 to Compatibility mode. Kill
2026 * the timers stone dead (this may be expensive for large N groups), they
2027 * will be restarted if Compatibility Mode deems that they must be due to
2031 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2033 struct ifmultiaddr *ifma;
2035 struct in_multi *inm;
2036 struct in_multi_head inm_free_tmp;
2038 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2039 igi->igi_ifp, igi->igi_ifp->if_xname);
2041 IN_MULTI_LIST_LOCK_ASSERT();
2045 SLIST_INIT(&inm_free_tmp);
2048 * Stop the v3 General Query Response on this link stone dead.
2049 * If fasttimo is woken up due to V_interface_timers_running,
2050 * the flag will be cleared if there are no pending link timers.
2052 igi->igi_v3_timer = 0;
2055 * Now clear the current-state and state-change report timers
2056 * for all memberships scoped to this link.
2060 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2061 inm = inm_ifmultiaddr_get_inm(ifma);
2064 switch (inm->inm_state) {
2065 case IGMP_NOT_MEMBER:
2066 case IGMP_SILENT_MEMBER:
2067 case IGMP_IDLE_MEMBER:
2068 case IGMP_LAZY_MEMBER:
2069 case IGMP_SLEEPING_MEMBER:
2070 case IGMP_AWAKENING_MEMBER:
2072 * These states are either not relevant in v3 mode,
2073 * or are unreported. Do nothing.
2076 case IGMP_LEAVING_MEMBER:
2078 * If we are leaving the group and switching to
2079 * compatibility mode, we need to release the final
2080 * reference held for issuing the INCLUDE {}, and
2081 * transition to REPORTING to ensure the host leave
2082 * message is sent upstream to the old querier --
2083 * transition to NOT would lose the leave and race.
2085 inm_rele_locked(&inm_free_tmp, inm);
2087 case IGMP_G_QUERY_PENDING_MEMBER:
2088 case IGMP_SG_QUERY_PENDING_MEMBER:
2089 inm_clear_recorded(inm);
2091 case IGMP_REPORTING_MEMBER:
2092 inm->inm_state = IGMP_REPORTING_MEMBER;
2096 * Always clear state-change and group report timers.
2097 * Free any pending IGMPv3 state-change records.
2099 inm->inm_sctimer = 0;
2101 mbufq_drain(&inm->inm_scq);
2103 IF_ADDR_WUNLOCK(ifp);
2105 inm_release_list_deferred(&inm_free_tmp);
2109 * Update the Older Version Querier Present timers for a link.
2110 * See Section 7.2.1 of RFC 3376.
2113 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2118 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2120 * IGMPv1 and IGMPv2 Querier Present timers expired.
2124 if (igi->igi_version != IGMP_VERSION_3) {
2126 "%s: transition from v%d -> v%d on %p(%s)",
2127 __func__, igi->igi_version, IGMP_VERSION_3,
2128 igi->igi_ifp, igi->igi_ifp->if_xname);
2129 igi->igi_version = IGMP_VERSION_3;
2131 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2133 * IGMPv1 Querier Present timer expired,
2134 * IGMPv2 Querier Present timer running.
2135 * If IGMPv2 was disabled since last timeout,
2137 * If IGMPv2 is enabled, revert to IGMPv2.
2139 if (!V_igmp_v2enable) {
2141 "%s: transition from v%d -> v%d on %p(%s)",
2142 __func__, igi->igi_version, IGMP_VERSION_3,
2143 igi->igi_ifp, igi->igi_ifp->if_xname);
2144 igi->igi_v2_timer = 0;
2145 igi->igi_version = IGMP_VERSION_3;
2147 --igi->igi_v2_timer;
2148 if (igi->igi_version != IGMP_VERSION_2) {
2150 "%s: transition from v%d -> v%d on %p(%s)",
2151 __func__, igi->igi_version, IGMP_VERSION_2,
2152 igi->igi_ifp, igi->igi_ifp->if_xname);
2153 igi->igi_version = IGMP_VERSION_2;
2154 igmp_v3_cancel_link_timers(igi);
2157 } else if (igi->igi_v1_timer > 0) {
2159 * IGMPv1 Querier Present timer running.
2160 * Stop IGMPv2 timer if running.
2162 * If IGMPv1 was disabled since last timeout,
2164 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2166 if (!V_igmp_v1enable) {
2168 "%s: transition from v%d -> v%d on %p(%s)",
2169 __func__, igi->igi_version, IGMP_VERSION_3,
2170 igi->igi_ifp, igi->igi_ifp->if_xname);
2171 igi->igi_v1_timer = 0;
2172 igi->igi_version = IGMP_VERSION_3;
2174 --igi->igi_v1_timer;
2176 if (igi->igi_v2_timer > 0) {
2178 "%s: cancel v2 timer on %p(%s)",
2179 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2180 igi->igi_v2_timer = 0;
2186 * Global slowtimo handler.
2187 * VIMAGE: Timeout handlers are expected to service all vimages.
2192 VNET_ITERATOR_DECL(vnet_iter);
2194 VNET_LIST_RLOCK_NOSLEEP();
2195 VNET_FOREACH(vnet_iter) {
2196 CURVNET_SET(vnet_iter);
2197 igmp_slowtimo_vnet();
2200 VNET_LIST_RUNLOCK_NOSLEEP();
2204 * Per-vnet slowtimo handler.
2207 igmp_slowtimo_vnet(void)
2209 struct igmp_ifsoftc *igi;
2213 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2214 igmp_v1v2_process_querier_timers(igi);
2221 * Dispatch an IGMPv1/v2 host report or leave message.
2222 * These are always small enough to fit inside a single mbuf.
2225 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2227 struct epoch_tracker et;
2233 IN_MULTI_LIST_LOCK_ASSERT();
2238 m = m_gethdr(M_NOWAIT, MT_DATA);
2241 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2243 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2245 m->m_data += sizeof(struct ip);
2246 m->m_len = sizeof(struct igmp);
2248 igmp = mtod(m, struct igmp *);
2249 igmp->igmp_type = type;
2250 igmp->igmp_code = 0;
2251 igmp->igmp_group = inm->inm_addr;
2252 igmp->igmp_cksum = 0;
2253 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2255 m->m_data -= sizeof(struct ip);
2256 m->m_len += sizeof(struct ip);
2258 ip = mtod(m, struct ip *);
2260 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2262 ip->ip_p = IPPROTO_IGMP;
2263 ip->ip_src.s_addr = INADDR_ANY;
2265 if (type == IGMP_HOST_LEAVE_MESSAGE)
2266 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2268 ip->ip_dst = inm->inm_addr;
2270 igmp_save_context(m, ifp);
2272 m->m_flags |= M_IGMPV2;
2273 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2274 m->m_flags |= M_IGMP_LOOP;
2276 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2277 NET_EPOCH_ENTER(et);
2278 netisr_dispatch(NETISR_IGMP, m);
2285 * Process a state change from the upper layer for the given IPv4 group.
2287 * Each socket holds a reference on the in_multi in its own ip_moptions.
2288 * The socket layer will have made the necessary updates to.the group
2289 * state, it is now up to IGMP to issue a state change report if there
2290 * has been any change between T0 (when the last state-change was issued)
2293 * We use the IGMPv3 state machine at group level. The IGMP module
2294 * however makes the decision as to which IGMP protocol version to speak.
2295 * A state change *from* INCLUDE {} always means an initial join.
2296 * A state change *to* INCLUDE {} always means a final leave.
2298 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2299 * save ourselves a bunch of work; any exclusive mode groups need not
2300 * compute source filter lists.
2302 * VIMAGE: curvnet should have been set by caller, as this routine
2303 * is called from the socket option handlers.
2306 igmp_change_state(struct in_multi *inm)
2308 struct igmp_ifsoftc *igi;
2313 IN_MULTI_LOCK_ASSERT();
2315 * Try to detect if the upper layer just asked us to change state
2316 * for an interface which has now gone away.
2318 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2319 ifp = inm->inm_ifma->ifma_ifp;
2323 * Sanity check that netinet's notion of ifp is the
2326 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2330 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2331 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2334 * If we detect a state transition to or from MCAST_UNDEFINED
2335 * for this group, then we are starting or finishing an IGMP
2336 * life cycle for this group.
2338 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2339 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2340 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2341 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2342 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2343 error = igmp_initial_join(inm, igi);
2345 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2346 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2347 igmp_final_leave(inm, igi);
2351 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2354 error = igmp_handle_state_change(inm, igi);
2362 * Perform the initial join for an IGMP group.
2364 * When joining a group:
2365 * If the group should have its IGMP traffic suppressed, do nothing.
2366 * IGMPv1 starts sending IGMPv1 host membership reports.
2367 * IGMPv2 starts sending IGMPv2 host membership reports.
2368 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2369 * initial state of the membership.
2372 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2376 int error, retval, syncstates;
2378 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2379 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2386 IN_MULTI_LOCK_ASSERT();
2389 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2392 * Groups joined on loopback or marked as 'not reported',
2393 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2394 * are never reported in any IGMP protocol exchanges.
2395 * All other groups enter the appropriate IGMP state machine
2396 * for the version in use on this link.
2397 * A link marked as IGIF_SILENT causes IGMP to be completely
2398 * disabled for the link.
2400 if ((ifp->if_flags & IFF_LOOPBACK) ||
2401 (igi->igi_flags & IGIF_SILENT) ||
2402 !igmp_isgroupreported(inm->inm_addr)) {
2404 "%s: not kicking state machine for silent group", __func__);
2405 inm->inm_state = IGMP_SILENT_MEMBER;
2409 * Deal with overlapping in_multi lifecycle.
2410 * If this group was LEAVING, then make sure
2411 * we drop the reference we picked up to keep the
2412 * group around for the final INCLUDE {} enqueue.
2414 if (igi->igi_version == IGMP_VERSION_3 &&
2415 inm->inm_state == IGMP_LEAVING_MEMBER) {
2416 MPASS(inm->inm_refcount > 1);
2417 inm_rele_locked(NULL, inm);
2419 inm->inm_state = IGMP_REPORTING_MEMBER;
2421 switch (igi->igi_version) {
2422 case IGMP_VERSION_1:
2423 case IGMP_VERSION_2:
2424 inm->inm_state = IGMP_IDLE_MEMBER;
2425 error = igmp_v1v2_queue_report(inm,
2426 (igi->igi_version == IGMP_VERSION_2) ?
2427 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2428 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2430 inm->inm_timer = IGMP_RANDOM_DELAY(
2431 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2432 V_current_state_timers_running = 1;
2436 case IGMP_VERSION_3:
2438 * Defer update of T0 to T1, until the first copy
2439 * of the state change has been transmitted.
2444 * Immediately enqueue a State-Change Report for
2445 * this interface, freeing any previous reports.
2446 * Don't kick the timers if there is nothing to do,
2447 * or if an error occurred.
2451 retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2453 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2456 error = retval * -1;
2461 * Schedule transmission of pending state-change
2462 * report up to RV times for this link. The timer
2463 * will fire at the next igmp_fasttimo (~200ms),
2464 * giving us an opportunity to merge the reports.
2466 if (igi->igi_flags & IGIF_LOOPBACK) {
2469 KASSERT(igi->igi_rv > 1,
2470 ("%s: invalid robustness %d", __func__,
2472 inm->inm_scrv = igi->igi_rv;
2474 inm->inm_sctimer = 1;
2475 V_state_change_timers_running = 1;
2483 * Only update the T0 state if state change is atomic,
2484 * i.e. we don't need to wait for a timer to fire before we
2485 * can consider the state change to have been communicated.
2489 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2490 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2497 * Issue an intermediate state change during the IGMP life-cycle.
2500 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2505 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2506 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2510 IN_MULTI_LIST_LOCK_ASSERT();
2513 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2515 if ((ifp->if_flags & IFF_LOOPBACK) ||
2516 (igi->igi_flags & IGIF_SILENT) ||
2517 !igmp_isgroupreported(inm->inm_addr) ||
2518 (igi->igi_version != IGMP_VERSION_3)) {
2519 if (!igmp_isgroupreported(inm->inm_addr)) {
2521 "%s: not kicking state machine for silent group", __func__);
2523 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2525 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2526 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2530 mbufq_drain(&inm->inm_scq);
2532 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2533 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2538 * If record(s) were enqueued, start the state-change
2539 * report timer for this group.
2541 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2542 inm->inm_sctimer = 1;
2543 V_state_change_timers_running = 1;
2549 * Perform the final leave for an IGMP group.
2551 * When leaving a group:
2552 * IGMPv1 does nothing.
2553 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2554 * IGMPv3 enqueues a state-change report containing a transition
2555 * to INCLUDE {} for immediate transmission.
2558 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2564 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2565 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2566 inm->inm_ifp->if_xname);
2568 IN_MULTI_LIST_LOCK_ASSERT();
2571 switch (inm->inm_state) {
2572 case IGMP_NOT_MEMBER:
2573 case IGMP_SILENT_MEMBER:
2574 case IGMP_LEAVING_MEMBER:
2575 /* Already leaving or left; do nothing. */
2577 "%s: not kicking state machine for silent group", __func__);
2579 case IGMP_REPORTING_MEMBER:
2580 case IGMP_IDLE_MEMBER:
2581 case IGMP_G_QUERY_PENDING_MEMBER:
2582 case IGMP_SG_QUERY_PENDING_MEMBER:
2583 if (igi->igi_version == IGMP_VERSION_2) {
2585 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2586 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2587 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2590 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2591 inm->inm_state = IGMP_NOT_MEMBER;
2592 } else if (igi->igi_version == IGMP_VERSION_3) {
2594 * Stop group timer and all pending reports.
2595 * Immediately enqueue a state-change report
2596 * TO_IN {} to be sent on the next fast timeout,
2597 * giving us an opportunity to merge reports.
2599 mbufq_drain(&inm->inm_scq);
2601 if (igi->igi_flags & IGIF_LOOPBACK) {
2604 inm->inm_scrv = igi->igi_rv;
2606 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2607 "pending retransmissions.", __func__,
2608 ntohl(inm->inm_addr.s_addr),
2609 inm->inm_ifp->if_xname, inm->inm_scrv);
2610 if (inm->inm_scrv == 0) {
2611 inm->inm_state = IGMP_NOT_MEMBER;
2612 inm->inm_sctimer = 0;
2614 int retval __unused;
2616 inm_acquire_locked(inm);
2618 retval = igmp_v3_enqueue_group_record(
2619 &inm->inm_scq, inm, 1, 0, 0);
2620 KASSERT(retval != 0,
2621 ("%s: enqueue record = %d", __func__,
2624 inm->inm_state = IGMP_LEAVING_MEMBER;
2625 inm->inm_sctimer = 1;
2626 V_state_change_timers_running = 1;
2632 case IGMP_LAZY_MEMBER:
2633 case IGMP_SLEEPING_MEMBER:
2634 case IGMP_AWAKENING_MEMBER:
2635 /* Our reports are suppressed; do nothing. */
2641 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2642 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2643 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2644 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2645 __func__, ntohl(inm->inm_addr.s_addr),
2646 inm->inm_ifp->if_xname);
2651 * Enqueue an IGMPv3 group record to the given output queue.
2653 * XXX This function could do with having the allocation code
2654 * split out, and the multiple-tree-walks coalesced into a single
2655 * routine as has been done in igmp_v3_enqueue_filter_change().
2657 * If is_state_change is zero, a current-state record is appended.
2658 * If is_state_change is non-zero, a state-change report is appended.
2660 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2661 * If is_group_query is zero, and if there is a packet with free space
2662 * at the tail of the queue, it will be appended to providing there
2663 * is enough free space.
2664 * Otherwise a new mbuf packet chain is allocated.
2666 * If is_source_query is non-zero, each source is checked to see if
2667 * it was recorded for a Group-Source query, and will be omitted if
2668 * it is not both in-mode and recorded.
2670 * The function will attempt to allocate leading space in the packet
2671 * for the IP/IGMP header to be prepended without fragmenting the chain.
2673 * If successful the size of all data appended to the queue is returned,
2674 * otherwise an error code less than zero is returned, or zero if
2675 * no record(s) were appended.
2678 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2679 const int is_state_change, const int is_group_query,
2680 const int is_source_query)
2682 struct igmp_grouprec ig;
2683 struct igmp_grouprec *pig;
2685 struct ip_msource *ims, *nims;
2686 struct mbuf *m0, *m, *md;
2687 int is_filter_list_change;
2688 int minrec0len, m0srcs, msrcs, nbytes, off;
2689 int record_has_sources;
2695 IN_MULTI_LIST_LOCK_ASSERT();
2698 is_filter_list_change = 0;
2705 record_has_sources = 1;
2707 type = IGMP_DO_NOTHING;
2708 mode = inm->inm_st[1].iss_fmode;
2711 * If we did not transition out of ASM mode during t0->t1,
2712 * and there are no source nodes to process, we can skip
2713 * the generation of source records.
2715 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2717 record_has_sources = 0;
2719 if (is_state_change) {
2721 * Queue a state change record.
2722 * If the mode did not change, and there are non-ASM
2723 * listeners or source filters present,
2724 * we potentially need to issue two records for the group.
2725 * If we are transitioning to MCAST_UNDEFINED, we need
2726 * not send any sources.
2727 * If there are ASM listeners, and there was no filter
2728 * mode transition of any kind, do nothing.
2730 if (mode != inm->inm_st[0].iss_fmode) {
2731 if (mode == MCAST_EXCLUDE) {
2732 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2734 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2736 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2738 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2739 if (mode == MCAST_UNDEFINED)
2740 record_has_sources = 0;
2743 if (record_has_sources) {
2744 is_filter_list_change = 1;
2746 type = IGMP_DO_NOTHING;
2751 * Queue a current state record.
2753 if (mode == MCAST_EXCLUDE) {
2754 type = IGMP_MODE_IS_EXCLUDE;
2755 } else if (mode == MCAST_INCLUDE) {
2756 type = IGMP_MODE_IS_INCLUDE;
2757 KASSERT(inm->inm_st[1].iss_asm == 0,
2758 ("%s: inm %p is INCLUDE but ASM count is %d",
2759 __func__, inm, inm->inm_st[1].iss_asm));
2764 * Generate the filter list changes using a separate function.
2766 if (is_filter_list_change)
2767 return (igmp_v3_enqueue_filter_change(mq, inm));
2769 if (type == IGMP_DO_NOTHING) {
2770 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2771 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2776 * If any sources are present, we must be able to fit at least
2777 * one in the trailing space of the tail packet's mbuf,
2780 minrec0len = sizeof(struct igmp_grouprec);
2781 if (record_has_sources)
2782 minrec0len += sizeof(in_addr_t);
2784 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2785 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2786 inm->inm_ifp->if_xname);
2789 * Check if we have a packet in the tail of the queue for this
2790 * group into which the first group record for this group will fit.
2791 * Otherwise allocate a new packet.
2792 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2793 * Note: Group records for G/GSR query responses MUST be sent
2794 * in their own packet.
2796 m0 = mbufq_last(mq);
2797 if (!is_group_query &&
2799 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2800 (m0->m_pkthdr.len + minrec0len) <
2801 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2802 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2803 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2805 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2807 if (mbufq_full(mq)) {
2808 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2812 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2813 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2814 if (!is_state_change && !is_group_query) {
2815 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2817 m->m_data += IGMP_LEADINGSPACE;
2820 m = m_gethdr(M_NOWAIT, MT_DATA);
2822 M_ALIGN(m, IGMP_LEADINGSPACE);
2827 igmp_save_context(m, ifp);
2829 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2833 * Append group record.
2834 * If we have sources, we don't know how many yet.
2839 ig.ig_group = inm->inm_addr;
2840 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2843 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2846 nbytes += sizeof(struct igmp_grouprec);
2849 * Append as many sources as will fit in the first packet.
2850 * If we are appending to a new packet, the chain allocation
2851 * may potentially use clusters; use m_getptr() in this case.
2852 * If we are appending to an existing packet, we need to obtain
2853 * a pointer to the group record after m_append(), in case a new
2854 * mbuf was allocated.
2855 * Only append sources which are in-mode at t1. If we are
2856 * transitioning to MCAST_UNDEFINED state on the group, do not
2857 * include source entries.
2858 * Only report recorded sources in our filter set when responding
2859 * to a group-source query.
2861 if (record_has_sources) {
2864 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2865 md->m_len - nbytes);
2867 md = m_getptr(m, 0, &off);
2868 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2872 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2873 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2875 now = ims_get_mode(inm, ims, 1);
2876 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2877 if ((now != mode) ||
2878 (now == mode && mode == MCAST_UNDEFINED)) {
2879 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2882 if (is_source_query && ims->ims_stp == 0) {
2883 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2887 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2888 naddr = htonl(ims->ims_haddr);
2889 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2892 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2896 nbytes += sizeof(in_addr_t);
2898 if (msrcs == m0srcs)
2901 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2903 pig->ig_numsrc = htons(msrcs);
2904 nbytes += (msrcs * sizeof(in_addr_t));
2907 if (is_source_query && msrcs == 0) {
2908 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2915 * We are good to go with first packet.
2918 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2919 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2920 mbufq_enqueue(mq, m);
2922 m->m_pkthdr.PH_vt.vt_nrecs++;
2925 * No further work needed if no source list in packet(s).
2927 if (!record_has_sources)
2931 * Whilst sources remain to be announced, we need to allocate
2932 * a new packet and fill out as many sources as will fit.
2933 * Always try for a cluster first.
2935 while (nims != NULL) {
2936 if (mbufq_full(mq)) {
2937 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2940 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2942 m->m_data += IGMP_LEADINGSPACE;
2944 m = m_gethdr(M_NOWAIT, MT_DATA);
2946 M_ALIGN(m, IGMP_LEADINGSPACE);
2950 igmp_save_context(m, ifp);
2951 md = m_getptr(m, 0, &off);
2952 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2953 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2955 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2958 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2961 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2962 nbytes += sizeof(struct igmp_grouprec);
2964 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2965 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2968 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2969 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2971 now = ims_get_mode(inm, ims, 1);
2972 if ((now != mode) ||
2973 (now == mode && mode == MCAST_UNDEFINED)) {
2974 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2977 if (is_source_query && ims->ims_stp == 0) {
2978 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2982 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2983 naddr = htonl(ims->ims_haddr);
2984 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2987 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2992 if (msrcs == m0srcs)
2995 pig->ig_numsrc = htons(msrcs);
2996 nbytes += (msrcs * sizeof(in_addr_t));
2998 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2999 mbufq_enqueue(mq, m);
3006 * Type used to mark record pass completion.
3007 * We exploit the fact we can cast to this easily from the
3008 * current filter modes on each ip_msource node.
3011 REC_NONE = 0x00, /* MCAST_UNDEFINED */
3012 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
3013 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
3014 REC_FULL = REC_ALLOW | REC_BLOCK
3018 * Enqueue an IGMPv3 filter list change to the given output queue.
3020 * Source list filter state is held in an RB-tree. When the filter list
3021 * for a group is changed without changing its mode, we need to compute
3022 * the deltas between T0 and T1 for each source in the filter set,
3023 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3025 * As we may potentially queue two record types, and the entire R-B tree
3026 * needs to be walked at once, we break this out into its own function
3027 * so we can generate a tightly packed queue of packets.
3029 * XXX This could be written to only use one tree walk, although that makes
3030 * serializing into the mbuf chains a bit harder. For now we do two walks
3031 * which makes things easier on us, and it may or may not be harder on
3034 * If successful the size of all data appended to the queue is returned,
3035 * otherwise an error code less than zero is returned, or zero if
3036 * no record(s) were appended.
3039 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3041 static const int MINRECLEN =
3042 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3044 struct igmp_grouprec ig;
3045 struct igmp_grouprec *pig;
3046 struct ip_msource *ims, *nims;
3047 struct mbuf *m, *m0, *md;
3049 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3051 uint8_t mode, now, then;
3052 rectype_t crt, drt, nrt;
3054 IN_MULTI_LIST_LOCK_ASSERT();
3056 if (inm->inm_nsrc == 0 ||
3057 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3060 ifp = inm->inm_ifp; /* interface */
3061 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3062 crt = REC_NONE; /* current group record type */
3063 drt = REC_NONE; /* mask of completed group record types */
3064 nrt = REC_NONE; /* record type for current node */
3065 m0srcs = 0; /* # source which will fit in current mbuf chain */
3066 nbytes = 0; /* # of bytes appended to group's state-change queue */
3067 npbytes = 0; /* # of bytes appended this packet */
3068 rsrcs = 0; /* # sources encoded in current record */
3069 schanged = 0; /* # nodes encoded in overall filter change */
3070 nallow = 0; /* # of source entries in ALLOW_NEW */
3071 nblock = 0; /* # of source entries in BLOCK_OLD */
3072 nims = NULL; /* next tree node pointer */
3075 * For each possible filter record mode.
3076 * The first kind of source we encounter tells us which
3077 * is the first kind of record we start appending.
3078 * If a node transitioned to UNDEFINED at t1, its mode is treated
3079 * as the inverse of the group's filter mode.
3081 while (drt != REC_FULL) {
3083 m0 = mbufq_last(mq);
3085 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3086 IGMP_V3_REPORT_MAXRECS) &&
3087 (m0->m_pkthdr.len + MINRECLEN) <
3088 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3090 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3091 sizeof(struct igmp_grouprec)) /
3094 "%s: use previous packet", __func__);
3096 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3098 m->m_data += IGMP_LEADINGSPACE;
3100 m = m_gethdr(M_NOWAIT, MT_DATA);
3102 M_ALIGN(m, IGMP_LEADINGSPACE);
3106 "%s: m_get*() failed", __func__);
3109 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3110 igmp_save_context(m, ifp);
3111 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3112 sizeof(struct igmp_grouprec)) /
3116 "%s: allocated new packet", __func__);
3119 * Append the IGMP group record header to the
3120 * current packet's data area.
3121 * Recalculate pointer to free space for next
3122 * group record, in case m_append() allocated
3123 * a new mbuf or cluster.
3125 memset(&ig, 0, sizeof(ig));
3126 ig.ig_group = inm->inm_addr;
3127 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3131 "%s: m_append() failed", __func__);
3134 npbytes += sizeof(struct igmp_grouprec);
3136 /* new packet; offset in c hain */
3137 md = m_getptr(m, npbytes -
3138 sizeof(struct igmp_grouprec), &off);
3139 pig = (struct igmp_grouprec *)(mtod(md,
3142 /* current packet; offset from last append */
3144 pig = (struct igmp_grouprec *)(mtod(md,
3145 uint8_t *) + md->m_len -
3146 sizeof(struct igmp_grouprec));
3149 * Begin walking the tree for this record type
3150 * pass, or continue from where we left off
3151 * previously if we had to allocate a new packet.
3152 * Only report deltas in-mode at t1.
3153 * We need not report included sources as allowed
3154 * if we are in inclusive mode on the group,
3155 * however the converse is not true.
3159 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3160 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3161 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3162 __func__, ims->ims_haddr);
3163 now = ims_get_mode(inm, ims, 1);
3164 then = ims_get_mode(inm, ims, 0);
3165 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3166 __func__, then, now);
3169 "%s: skip unchanged", __func__);
3172 if (mode == MCAST_EXCLUDE &&
3173 now == MCAST_INCLUDE) {
3175 "%s: skip IN src on EX group",
3179 nrt = (rectype_t)now;
3180 if (nrt == REC_NONE)
3181 nrt = (rectype_t)(~mode & REC_FULL);
3182 if (schanged++ == 0) {
3184 } else if (crt != nrt)
3186 naddr = htonl(ims->ims_haddr);
3187 if (!m_append(m, sizeof(in_addr_t),
3192 "%s: m_append() failed", __func__);
3195 nallow += !!(crt == REC_ALLOW);
3196 nblock += !!(crt == REC_BLOCK);
3197 if (++rsrcs == m0srcs)
3201 * If we did not append any tree nodes on this
3202 * pass, back out of allocations.
3205 npbytes -= sizeof(struct igmp_grouprec);
3208 "%s: m_free(m)", __func__);
3212 "%s: m_adj(m, -ig)", __func__);
3213 m_adj(m, -((int)sizeof(
3214 struct igmp_grouprec)));
3218 npbytes += (rsrcs * sizeof(in_addr_t));
3219 if (crt == REC_ALLOW)
3220 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3221 else if (crt == REC_BLOCK)
3222 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3223 pig->ig_numsrc = htons(rsrcs);
3225 * Count the new group record, and enqueue this
3226 * packet if it wasn't already queued.
3228 m->m_pkthdr.PH_vt.vt_nrecs++;
3230 mbufq_enqueue(mq, m);
3232 } while (nims != NULL);
3234 crt = (~crt & REC_FULL);
3237 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3244 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3247 struct mbuf *m; /* pending state-change */
3248 struct mbuf *m0; /* copy of pending state-change */
3249 struct mbuf *mt; /* last state-change in packet */
3250 int docopy, domerge;
3257 IN_MULTI_LIST_LOCK_ASSERT();
3261 * If there are further pending retransmissions, make a writable
3262 * copy of each queued state-change message before merging.
3264 if (inm->inm_scrv > 0)
3269 if (mbufq_first(gq) == NULL) {
3270 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3275 m = mbufq_first(gq);
3278 * Only merge the report into the current packet if
3279 * there is sufficient space to do so; an IGMPv3 report
3280 * packet may only contain 65,535 group records.
3281 * Always use a simple mbuf chain concatentation to do this,
3282 * as large state changes for single groups may have
3283 * allocated clusters.
3286 mt = mbufq_last(scq);
3288 recslen = m_length(m, NULL);
3290 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3291 m->m_pkthdr.PH_vt.vt_nrecs <=
3292 IGMP_V3_REPORT_MAXRECS) &&
3293 (mt->m_pkthdr.len + recslen <=
3294 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3298 if (!domerge && mbufq_full(gq)) {
3300 "%s: outbound queue full, skipping whole packet %p",
3310 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3311 m0 = mbufq_dequeue(gq);
3314 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3315 m0 = m_dup(m, M_NOWAIT);
3318 m0->m_nextpkt = NULL;
3323 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3325 mbufq_enqueue(scq, m0);
3327 struct mbuf *mtl; /* last mbuf of packet mt */
3329 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3333 m0->m_flags &= ~M_PKTHDR;
3334 mt->m_pkthdr.len += recslen;
3335 mt->m_pkthdr.PH_vt.vt_nrecs +=
3336 m0->m_pkthdr.PH_vt.vt_nrecs;
3346 * Respond to a pending IGMPv3 General Query.
3349 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3351 struct ifmultiaddr *ifma;
3353 struct in_multi *inm;
3354 int retval __unused, loop;
3356 IN_MULTI_LIST_LOCK_ASSERT();
3360 KASSERT(igi->igi_version == IGMP_VERSION_3,
3361 ("%s: called when version %d", __func__, igi->igi_version));
3364 * Check that there are some packets queued. If so, send them first.
3365 * For large number of groups the reply to general query can take
3366 * many packets, we should finish sending them before starting of
3367 * queuing the new reply.
3369 if (mbufq_len(&igi->igi_gq) != 0)
3374 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3375 inm = inm_ifmultiaddr_get_inm(ifma);
3378 KASSERT(ifp == inm->inm_ifp,
3379 ("%s: inconsistent ifp", __func__));
3381 switch (inm->inm_state) {
3382 case IGMP_NOT_MEMBER:
3383 case IGMP_SILENT_MEMBER:
3385 case IGMP_REPORTING_MEMBER:
3386 case IGMP_IDLE_MEMBER:
3387 case IGMP_LAZY_MEMBER:
3388 case IGMP_SLEEPING_MEMBER:
3389 case IGMP_AWAKENING_MEMBER:
3390 inm->inm_state = IGMP_REPORTING_MEMBER;
3391 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3393 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3396 case IGMP_G_QUERY_PENDING_MEMBER:
3397 case IGMP_SG_QUERY_PENDING_MEMBER:
3398 case IGMP_LEAVING_MEMBER:
3404 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3405 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3408 * Slew transmission of bursts over 500ms intervals.
3410 if (mbufq_first(&igi->igi_gq) != NULL) {
3411 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3412 IGMP_RESPONSE_BURST_INTERVAL);
3413 V_interface_timers_running = 1;
3418 * Transmit the next pending IGMP message in the output queue.
3420 * We get called from netisr_processqueue(). A mutex private to igmpoq
3421 * will be acquired and released around this routine.
3423 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3424 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3425 * a link and uses a link-scope multicast address.
3428 igmp_intr(struct mbuf *m)
3430 struct ip_moptions imo;
3432 struct mbuf *ipopts, *m0;
3436 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3439 * Set VNET image pointer from enqueued mbuf chain
3440 * before doing anything else. Whilst we use interface
3441 * indexes to guard against interface detach, they are
3442 * unique to each VIMAGE and must be retrieved.
3444 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3445 ifindex = igmp_restore_context(m);
3448 * Check if the ifnet still exists. This limits the scope of
3449 * any race in the absence of a global ifp lock for low cost
3450 * (an array lookup).
3452 ifp = ifnet_byindex(ifindex);
3454 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3455 __func__, m, ifindex);
3457 IPSTAT_INC(ips_noroute);
3461 ipopts = V_igmp_sendra ? m_raopt : NULL;
3463 imo.imo_multicast_ttl = 1;
3464 imo.imo_multicast_vif = -1;
3465 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3468 * If the user requested that IGMP traffic be explicitly
3469 * redirected to the loopback interface (e.g. they are running a
3470 * MANET interface and the routing protocol needs to see the
3471 * updates), handle this now.
3473 if (m->m_flags & M_IGMP_LOOP)
3474 imo.imo_multicast_ifp = V_loif;
3476 imo.imo_multicast_ifp = ifp;
3478 if (m->m_flags & M_IGMPV2) {
3481 m0 = igmp_v3_encap_report(ifp, m);
3483 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3485 IPSTAT_INC(ips_odropped);
3490 igmp_scrub_context(m0);
3492 m0->m_pkthdr.rcvif = V_loif;
3494 mac_netinet_igmp_send(ifp, m0);
3496 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3498 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3502 IGMPSTAT_INC(igps_snd_reports);
3506 * We must restore the existing vnet pointer before
3507 * continuing as we are run from netisr context.
3513 * Encapsulate an IGMPv3 report.
3515 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3516 * chain has already had its IP/IGMPv3 header prepended. In this case
3517 * the function will not attempt to prepend; the lengths and checksums
3518 * will however be re-computed.
3520 * Returns a pointer to the new mbuf chain head, or NULL if the
3521 * allocation failed.
3523 static struct mbuf *
3524 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3526 struct rm_priotracker in_ifa_tracker;
3527 struct igmp_report *igmp;
3529 int hdrlen, igmpreclen;
3531 KASSERT((m->m_flags & M_PKTHDR),
3532 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3534 igmpreclen = m_length(m, NULL);
3535 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3537 if (m->m_flags & M_IGMPV3_HDR) {
3538 igmpreclen -= hdrlen;
3540 M_PREPEND(m, hdrlen, M_NOWAIT);
3543 m->m_flags |= M_IGMPV3_HDR;
3546 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3548 m->m_data += sizeof(struct ip);
3549 m->m_len -= sizeof(struct ip);
3551 igmp = mtod(m, struct igmp_report *);
3552 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3555 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3557 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3558 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3560 m->m_data -= sizeof(struct ip);
3561 m->m_len += sizeof(struct ip);
3563 ip = mtod(m, struct ip *);
3564 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3565 ip->ip_len = htons(hdrlen + igmpreclen);
3566 ip->ip_off = htons(IP_DF);
3567 ip->ip_p = IPPROTO_IGMP;
3570 ip->ip_src.s_addr = INADDR_ANY;
3572 if (m->m_flags & M_IGMP_LOOP) {
3573 struct in_ifaddr *ia;
3575 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3577 ip->ip_src = ia->ia_addr.sin_addr;
3580 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3587 igmp_rec_type_to_str(const int type)
3591 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3594 case IGMP_CHANGE_TO_INCLUDE_MODE:
3597 case IGMP_MODE_IS_EXCLUDE:
3600 case IGMP_MODE_IS_INCLUDE:
3603 case IGMP_ALLOW_NEW_SOURCES:
3606 case IGMP_BLOCK_OLD_SOURCES:
3618 vnet_igmp_init(const void *unused __unused)
3621 netisr_register_vnet(&igmp_nh);
3623 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3624 vnet_igmp_init, NULL);
3627 vnet_igmp_uninit(const void *unused __unused)
3630 /* This can happen when we shutdown the entire network stack. */
3631 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3633 netisr_unregister_vnet(&igmp_nh);
3635 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3636 vnet_igmp_uninit, NULL);
3640 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3642 struct igmp_ifsoftc *igi, *tigi;
3643 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3646 db_printf("usage: show igi_list <addr>\n");
3649 igi_head = (struct _igi_list *)addr;
3651 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3652 db_printf("igmp_ifsoftc %p:\n", igi);
3653 db_printf(" ifp %p\n", igi->igi_ifp);
3654 db_printf(" version %u\n", igi->igi_version);
3655 db_printf(" v1_timer %u\n", igi->igi_v1_timer);
3656 db_printf(" v2_timer %u\n", igi->igi_v2_timer);
3657 db_printf(" v3_timer %u\n", igi->igi_v3_timer);
3658 db_printf(" flags %#x\n", igi->igi_flags);
3659 db_printf(" rv %u\n", igi->igi_rv);
3660 db_printf(" qi %u\n", igi->igi_qi);
3661 db_printf(" qri %u\n", igi->igi_qri);
3662 db_printf(" uri %u\n", igi->igi_uri);
3663 /* struct mbufq igi_gq; */
3670 igmp_modevent(module_t mod, int type, void *unused __unused)
3675 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3677 m_raopt = igmp_ra_alloc();
3678 netisr_register(&igmp_nh);
3681 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3682 netisr_unregister(&igmp_nh);
3685 IGMP_LOCK_DESTROY();
3688 return (EOPNOTSUPP);
3693 static moduledata_t igmp_mod = {
3698 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);