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_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) =
234 LIST_HEAD_INITIALIZER(igi_head);
235 VNET_DEFINE_STATIC(struct igmpstat, igmpstat) = {
236 .igps_version = IGPS_VERSION_3,
237 .igps_len = sizeof(struct igmpstat),
239 VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0};
241 #define V_igi_head VNET(igi_head)
242 #define V_igmpstat VNET(igmpstat)
243 #define V_igmp_gsrdelay VNET(igmp_gsrdelay)
245 VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1;
246 VNET_DEFINE_STATIC(int, igmp_sendra) = 1;
247 VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1;
248 VNET_DEFINE_STATIC(int, igmp_v1enable) = 1;
249 VNET_DEFINE_STATIC(int, igmp_v2enable) = 1;
250 VNET_DEFINE_STATIC(int, igmp_legacysupp);
251 VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3;
253 #define V_igmp_recvifkludge VNET(igmp_recvifkludge)
254 #define V_igmp_sendra VNET(igmp_sendra)
255 #define V_igmp_sendlocal VNET(igmp_sendlocal)
256 #define V_igmp_v1enable VNET(igmp_v1enable)
257 #define V_igmp_v2enable VNET(igmp_v2enable)
258 #define V_igmp_legacysupp VNET(igmp_legacysupp)
259 #define V_igmp_default_version VNET(igmp_default_version)
262 * Virtualized sysctls.
264 SYSCTL_PROC(_net_inet_igmp, IGMPCTL_STATS, stats,
265 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE,
266 &VNET_NAME(igmpstat), 0, sysctl_igmp_stat, "S,igmpstat", "");
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(V_igmpstat));
354 if (req->oldptr != NULL) {
355 if (req->oldlen < sizeof(V_igmpstat))
358 error = SYSCTL_OUT(req, &V_igmpstat,
361 req->validlen = sizeof(V_igmpstat);
364 if (req->newptr != NULL) {
365 if (req->newlen < sizeof(V_igmpstat))
368 error = SYSCTL_IN(req, &igps0,
373 * igps0 must be "all zero".
376 while (*p == '\0' && p < (char *)&igps0 + sizeof(igps0))
378 if (p != (char *)&igps0 + sizeof(igps0)) {
383 * Avoid overwrite of the version and length field.
385 igps0.igps_version = V_igmpstat.igps_version;
386 igps0.igps_len = V_igmpstat.igps_len;
387 bcopy(&igps0, &V_igmpstat, sizeof(V_igmpstat));
394 * Retrieve or set default IGMP version.
396 * VIMAGE: Assume curvnet set by caller.
397 * SMPng: NOTE: Serialized by IGMP lock.
400 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
405 error = sysctl_wire_old_buffer(req, sizeof(int));
411 new = V_igmp_default_version;
413 error = sysctl_handle_int(oidp, &new, 0, req);
414 if (error || !req->newptr)
417 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
422 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
423 V_igmp_default_version, new);
425 V_igmp_default_version = new;
433 * Retrieve or set threshold between group-source queries in seconds.
435 * VIMAGE: Assume curvnet set by caller.
436 * SMPng: NOTE: Serialized by IGMP lock.
439 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
444 error = sysctl_wire_old_buffer(req, sizeof(int));
450 i = V_igmp_gsrdelay.tv_sec;
452 error = sysctl_handle_int(oidp, &i, 0, req);
453 if (error || !req->newptr)
456 if (i < -1 || i >= 60) {
461 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
462 V_igmp_gsrdelay.tv_sec, i);
463 V_igmp_gsrdelay.tv_sec = i;
471 * Expose struct igmp_ifsoftc to userland, keyed by ifindex.
472 * For use by ifmcstat(8).
474 * SMPng: NOTE: Does an unlocked ifindex space read.
475 * VIMAGE: Assume curvnet set by caller. The node handler itself
476 * is not directly virtualized.
479 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
485 struct igmp_ifsoftc *igi;
490 if (req->newptr != NULL)
496 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
500 IN_MULTI_LIST_LOCK();
503 if (name[0] <= 0 || name[0] > V_if_index) {
510 ifp = ifnet_byindex(name[0]);
514 LIST_FOREACH(igi, &V_igi_head, igi_link) {
515 if (ifp == igi->igi_ifp) {
516 struct igmp_ifinfo info;
518 info.igi_version = igi->igi_version;
519 info.igi_v1_timer = igi->igi_v1_timer;
520 info.igi_v2_timer = igi->igi_v2_timer;
521 info.igi_v3_timer = igi->igi_v3_timer;
522 info.igi_flags = igi->igi_flags;
523 info.igi_rv = igi->igi_rv;
524 info.igi_qi = igi->igi_qi;
525 info.igi_qri = igi->igi_qri;
526 info.igi_uri = igi->igi_uri;
527 error = SYSCTL_OUT(req, &info, sizeof(info));
534 IN_MULTI_LIST_UNLOCK();
539 * Dispatch an entire queue of pending packet chains
541 * VIMAGE: Assumes the vnet pointer has been set.
544 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop)
546 struct epoch_tracker et;
550 while ((m = mbufq_dequeue(mq)) != NULL) {
551 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m);
553 m->m_flags |= M_IGMP_LOOP;
554 netisr_dispatch(NETISR_IGMP, m);
562 * Filter outgoing IGMP report state by group.
564 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
565 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
566 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
567 * this may break certain IGMP snooping switches which rely on the old
570 * Return zero if the given group is one for which IGMP reports
571 * should be suppressed, or non-zero if reports should be issued.
574 igmp_isgroupreported(const struct in_addr addr)
577 if (in_allhosts(addr) ||
578 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
585 * Construct a Router Alert option to use in outgoing packets.
593 m = m_get(M_WAITOK, MT_DATA);
594 p = mtod(m, struct ipoption *);
595 p->ipopt_dst.s_addr = INADDR_ANY;
596 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */
597 p->ipopt_list[1] = 0x04; /* 4 bytes long */
598 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
599 p->ipopt_list[3] = 0x00; /* pad byte */
600 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
606 * Attach IGMP when PF_INET is attached to an interface.
608 struct igmp_ifsoftc *
609 igmp_domifattach(struct ifnet *ifp)
611 struct igmp_ifsoftc *igi;
613 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
614 __func__, ifp, ifp->if_xname);
618 igi = igi_alloc_locked(ifp);
619 if (!(ifp->if_flags & IFF_MULTICAST))
620 igi->igi_flags |= IGIF_SILENT;
628 * VIMAGE: assume curvnet set by caller.
630 static struct igmp_ifsoftc *
631 igi_alloc_locked(/*const*/ struct ifnet *ifp)
633 struct igmp_ifsoftc *igi;
637 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO);
642 igi->igi_version = V_igmp_default_version;
644 igi->igi_rv = IGMP_RV_INIT;
645 igi->igi_qi = IGMP_QI_INIT;
646 igi->igi_qri = IGMP_QRI_INIT;
647 igi->igi_uri = IGMP_URI_INIT;
648 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
650 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
652 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)",
662 * NOTE: Some finalization tasks need to run before the protocol domain
663 * is detached, but also before the link layer does its cleanup.
665 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
666 * XXX This is also bitten by unlocked ifma_protospec access.
669 igmp_ifdetach(struct ifnet *ifp)
671 struct igmp_ifsoftc *igi;
672 struct ifmultiaddr *ifma, *next;
673 struct in_multi *inm;
674 struct in_multi_head inm_free_tmp;
675 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
678 SLIST_INIT(&inm_free_tmp);
681 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
682 if (igi->igi_version == IGMP_VERSION_3) {
685 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
686 if (ifma->ifma_addr->sa_family != AF_INET ||
687 ifma->ifma_protospec == NULL)
689 inm = (struct in_multi *)ifma->ifma_protospec;
690 if (inm->inm_state == IGMP_LEAVING_MEMBER)
691 inm_rele_locked(&inm_free_tmp, inm);
692 inm_clear_recorded(inm);
693 if (__predict_false(ifma_restart)) {
694 ifma_restart = false;
698 IF_ADDR_WUNLOCK(ifp);
699 inm_release_list_deferred(&inm_free_tmp);
706 * Hook for domifdetach.
709 igmp_domifdetach(struct ifnet *ifp)
712 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
713 __func__, ifp, ifp->if_xname);
716 igi_delete_locked(ifp);
721 igi_delete_locked(const struct ifnet *ifp)
723 struct igmp_ifsoftc *igi, *tigi;
725 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)",
726 __func__, ifp, ifp->if_xname);
730 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
731 if (igi->igi_ifp == ifp) {
733 * Free deferred General Query responses.
735 mbufq_drain(&igi->igi_gq);
737 LIST_REMOVE(igi, igi_link);
745 * Process a received IGMPv1 query.
746 * Return non-zero if the message should be dropped.
748 * VIMAGE: The curvnet pointer is derived from the input ifp.
751 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
752 const struct igmp *igmp)
754 struct ifmultiaddr *ifma;
755 struct igmp_ifsoftc *igi;
756 struct in_multi *inm;
761 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
762 * 224.0.0.1. They are always treated as General Queries.
763 * igmp_group is always ignored. Do not drop it as a userland
764 * daemon may wish to see it.
765 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
767 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
768 IGMPSTAT_INC(igps_rcv_badqueries);
771 IGMPSTAT_INC(igps_rcv_gen_queries);
773 IN_MULTI_LIST_LOCK();
776 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
777 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
779 if (igi->igi_flags & IGIF_LOOPBACK) {
780 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
786 * Switch to IGMPv1 host compatibility mode.
788 igmp_set_version(igi, IGMP_VERSION_1);
790 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
793 * Start the timers in all of our group records
794 * for the interface on which the query arrived,
795 * except those which are already running.
797 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
798 if (ifma->ifma_addr->sa_family != AF_INET ||
799 ifma->ifma_protospec == NULL)
801 inm = (struct in_multi *)ifma->ifma_protospec;
802 if (inm->inm_timer != 0)
804 switch (inm->inm_state) {
805 case IGMP_NOT_MEMBER:
806 case IGMP_SILENT_MEMBER:
808 case IGMP_G_QUERY_PENDING_MEMBER:
809 case IGMP_SG_QUERY_PENDING_MEMBER:
810 case IGMP_REPORTING_MEMBER:
811 case IGMP_IDLE_MEMBER:
812 case IGMP_LAZY_MEMBER:
813 case IGMP_SLEEPING_MEMBER:
814 case IGMP_AWAKENING_MEMBER:
815 inm->inm_state = IGMP_REPORTING_MEMBER;
816 inm->inm_timer = IGMP_RANDOM_DELAY(
817 IGMP_V1V2_MAX_RI * PR_FASTHZ);
818 V_current_state_timers_running = 1;
820 case IGMP_LEAVING_MEMBER:
827 IN_MULTI_LIST_UNLOCK();
833 * Process a received IGMPv2 general or group-specific query.
836 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
837 const struct igmp *igmp)
839 struct ifmultiaddr *ifma;
840 struct igmp_ifsoftc *igi;
841 struct in_multi *inm;
842 int is_general_query;
847 is_general_query = 0;
850 * Validate address fields upfront.
851 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
853 if (in_nullhost(igmp->igmp_group)) {
855 * IGMPv2 General Query.
856 * If this was not sent to the all-hosts group, ignore it.
858 if (!in_allhosts(ip->ip_dst))
860 IGMPSTAT_INC(igps_rcv_gen_queries);
861 is_general_query = 1;
863 /* IGMPv2 Group-Specific Query. */
864 IGMPSTAT_INC(igps_rcv_group_queries);
867 IN_MULTI_LIST_LOCK();
870 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
871 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
873 if (igi->igi_flags & IGIF_LOOPBACK) {
874 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
880 * Ignore v2 query if in v1 Compatibility Mode.
882 if (igi->igi_version == IGMP_VERSION_1)
885 igmp_set_version(igi, IGMP_VERSION_2);
887 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
891 if (is_general_query) {
893 * For each reporting group joined on this
894 * interface, kick the report timer.
896 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
898 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
899 if (ifma->ifma_addr->sa_family != AF_INET ||
900 ifma->ifma_protospec == NULL)
902 inm = (struct in_multi *)ifma->ifma_protospec;
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);
1587 return (IPPROTO_DONE);
1590 * m_pullup() may modify m, so pullup in
1593 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1594 sizeof(struct in_addr) * nsrc;
1595 if ((!M_WRITABLE(m) ||
1596 m->m_len < igmpv3len) &&
1597 (m = m_pullup(m, igmpv3len)) == NULL) {
1598 IGMPSTAT_INC(igps_rcv_tooshort);
1599 return (IPPROTO_DONE);
1601 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1603 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1605 return (IPPROTO_DONE);
1612 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1613 if (!V_igmp_v1enable)
1615 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1617 return (IPPROTO_DONE);
1621 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1622 if (!V_igmp_v2enable)
1624 if (!ip_checkrouteralert(m))
1625 IGMPSTAT_INC(igps_rcv_nora);
1626 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1628 return (IPPROTO_DONE);
1632 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1634 * Hosts do not need to process IGMPv3 membership reports,
1635 * as report suppression is no longer required.
1637 if (!ip_checkrouteralert(m))
1638 IGMPSTAT_INC(igps_rcv_nora);
1646 * Pass all valid IGMP packets up to any process(es) listening on a
1650 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, *next;
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);
1751 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1752 if (ifma->ifma_addr->sa_family != AF_INET ||
1753 ifma->ifma_protospec == NULL)
1755 inm = (struct in_multi *)ifma->ifma_protospec;
1756 switch (igi->igi_version) {
1757 case IGMP_VERSION_1:
1758 case IGMP_VERSION_2:
1759 igmp_v1v2_process_group_timer(inm,
1762 case IGMP_VERSION_3:
1763 igmp_v3_process_group_timers(&inm_free_tmp, &qrq,
1764 &scq, inm, uri_fasthz);
1767 if (__predict_false(ifma_restart)) {
1768 ifma_restart = false;
1772 IF_ADDR_WUNLOCK(ifp);
1774 if (igi->igi_version == IGMP_VERSION_3) {
1775 igmp_dispatch_queue(&qrq, 0, loop);
1776 igmp_dispatch_queue(&scq, 0, loop);
1779 * Free the in_multi reference(s) for this
1782 inm_release_list_deferred(&inm_free_tmp);
1788 IN_MULTI_LIST_UNLOCK();
1792 * Update host report group timer for IGMPv1/v2.
1793 * Will update the global pending timer flags.
1796 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1798 int report_timer_expired;
1800 IN_MULTI_LIST_LOCK_ASSERT();
1803 if (inm->inm_timer == 0) {
1804 report_timer_expired = 0;
1805 } else if (--inm->inm_timer == 0) {
1806 report_timer_expired = 1;
1808 V_current_state_timers_running = 1;
1812 switch (inm->inm_state) {
1813 case IGMP_NOT_MEMBER:
1814 case IGMP_SILENT_MEMBER:
1815 case IGMP_IDLE_MEMBER:
1816 case IGMP_LAZY_MEMBER:
1817 case IGMP_SLEEPING_MEMBER:
1818 case IGMP_AWAKENING_MEMBER:
1820 case IGMP_REPORTING_MEMBER:
1821 if (report_timer_expired) {
1822 inm->inm_state = IGMP_IDLE_MEMBER;
1823 (void)igmp_v1v2_queue_report(inm,
1824 (version == IGMP_VERSION_2) ?
1825 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1826 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1829 case IGMP_G_QUERY_PENDING_MEMBER:
1830 case IGMP_SG_QUERY_PENDING_MEMBER:
1831 case IGMP_LEAVING_MEMBER:
1837 * Update a group's timers for IGMPv3.
1838 * Will update the global pending timer flags.
1839 * Note: Unlocked read from igi.
1842 igmp_v3_process_group_timers(struct in_multi_head *inmh,
1843 struct mbufq *qrq, struct mbufq *scq,
1844 struct in_multi *inm, const int uri_fasthz)
1846 int query_response_timer_expired;
1847 int state_change_retransmit_timer_expired;
1849 IN_MULTI_LIST_LOCK_ASSERT();
1852 query_response_timer_expired = 0;
1853 state_change_retransmit_timer_expired = 0;
1856 * During a transition from v1/v2 compatibility mode back to v3,
1857 * a group record in REPORTING state may still have its group
1858 * timer active. This is a no-op in this function; it is easier
1859 * to deal with it here than to complicate the slow-timeout path.
1861 if (inm->inm_timer == 0) {
1862 query_response_timer_expired = 0;
1863 } else if (--inm->inm_timer == 0) {
1864 query_response_timer_expired = 1;
1866 V_current_state_timers_running = 1;
1869 if (inm->inm_sctimer == 0) {
1870 state_change_retransmit_timer_expired = 0;
1871 } else if (--inm->inm_sctimer == 0) {
1872 state_change_retransmit_timer_expired = 1;
1874 V_state_change_timers_running = 1;
1877 /* We are in fasttimo, so be quick about it. */
1878 if (!state_change_retransmit_timer_expired &&
1879 !query_response_timer_expired)
1882 switch (inm->inm_state) {
1883 case IGMP_NOT_MEMBER:
1884 case IGMP_SILENT_MEMBER:
1885 case IGMP_SLEEPING_MEMBER:
1886 case IGMP_LAZY_MEMBER:
1887 case IGMP_AWAKENING_MEMBER:
1888 case IGMP_IDLE_MEMBER:
1890 case IGMP_G_QUERY_PENDING_MEMBER:
1891 case IGMP_SG_QUERY_PENDING_MEMBER:
1893 * Respond to a previously pending Group-Specific
1894 * or Group-and-Source-Specific query by enqueueing
1895 * the appropriate Current-State report for
1896 * immediate transmission.
1898 if (query_response_timer_expired) {
1899 int retval __unused;
1901 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1902 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1903 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1905 inm->inm_state = IGMP_REPORTING_MEMBER;
1906 /* XXX Clear recorded sources for next time. */
1907 inm_clear_recorded(inm);
1910 case IGMP_REPORTING_MEMBER:
1911 case IGMP_LEAVING_MEMBER:
1912 if (state_change_retransmit_timer_expired) {
1914 * State-change retransmission timer fired.
1915 * If there are any further pending retransmissions,
1916 * set the global pending state-change flag, and
1919 if (--inm->inm_scrv > 0) {
1920 inm->inm_sctimer = uri_fasthz;
1921 V_state_change_timers_running = 1;
1924 * Retransmit the previously computed state-change
1925 * report. If there are no further pending
1926 * retransmissions, the mbuf queue will be consumed.
1927 * Update T0 state to T1 as we have now sent
1930 (void)igmp_v3_merge_state_changes(inm, scq);
1933 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
1934 ntohl(inm->inm_addr.s_addr),
1935 inm->inm_ifp->if_xname);
1938 * If we are leaving the group for good, make sure
1939 * we release IGMP's reference to it.
1940 * This release must be deferred using a SLIST,
1941 * as we are called from a loop which traverses
1942 * the in_ifmultiaddr TAILQ.
1944 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1945 inm->inm_scrv == 0) {
1946 inm->inm_state = IGMP_NOT_MEMBER;
1947 inm_rele_locked(inmh, inm);
1956 * Suppress a group's pending response to a group or source/group query.
1958 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1959 * Do NOT update ST1/ST0 as this operation merely suppresses
1960 * the currently pending group record.
1961 * Do NOT suppress the response to a general query. It is possible but
1962 * it would require adding another state or flag.
1965 igmp_v3_suppress_group_record(struct in_multi *inm)
1968 IN_MULTI_LIST_LOCK_ASSERT();
1970 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1971 ("%s: not IGMPv3 mode on link", __func__));
1973 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1974 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1977 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1978 inm_clear_recorded(inm);
1981 inm->inm_state = IGMP_REPORTING_MEMBER;
1985 * Switch to a different IGMP version on the given interface,
1986 * as per Section 7.2.1.
1989 igmp_set_version(struct igmp_ifsoftc *igi, const int version)
1991 int old_version_timer;
1995 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1996 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1998 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
2000 * Compute the "Older Version Querier Present" timer as per
2003 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
2004 old_version_timer *= PR_SLOWHZ;
2006 if (version == IGMP_VERSION_1) {
2007 igi->igi_v1_timer = old_version_timer;
2008 igi->igi_v2_timer = 0;
2009 } else if (version == IGMP_VERSION_2) {
2010 igi->igi_v1_timer = 0;
2011 igi->igi_v2_timer = old_version_timer;
2015 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2016 if (igi->igi_version != IGMP_VERSION_2) {
2017 igi->igi_version = IGMP_VERSION_2;
2018 igmp_v3_cancel_link_timers(igi);
2020 } else if (igi->igi_v1_timer > 0) {
2021 if (igi->igi_version != IGMP_VERSION_1) {
2022 igi->igi_version = IGMP_VERSION_1;
2023 igmp_v3_cancel_link_timers(igi);
2029 * Cancel pending IGMPv3 timers for the given link and all groups
2030 * joined on it; state-change, general-query, and group-query timers.
2032 * Only ever called on a transition from v3 to Compatibility mode. Kill
2033 * the timers stone dead (this may be expensive for large N groups), they
2034 * will be restarted if Compatibility Mode deems that they must be due to
2038 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi)
2040 struct ifmultiaddr *ifma, *ifmatmp;
2042 struct in_multi *inm;
2043 struct in_multi_head inm_free_tmp;
2045 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
2046 igi->igi_ifp, igi->igi_ifp->if_xname);
2048 IN_MULTI_LIST_LOCK_ASSERT();
2052 SLIST_INIT(&inm_free_tmp);
2055 * Stop the v3 General Query Response on this link stone dead.
2056 * If fasttimo is woken up due to V_interface_timers_running,
2057 * the flag will be cleared if there are no pending link timers.
2059 igi->igi_v3_timer = 0;
2062 * Now clear the current-state and state-change report timers
2063 * for all memberships scoped to this link.
2067 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, ifmatmp) {
2068 if (ifma->ifma_addr->sa_family != AF_INET ||
2069 ifma->ifma_protospec == NULL)
2071 inm = (struct in_multi *)ifma->ifma_protospec;
2072 switch (inm->inm_state) {
2073 case IGMP_NOT_MEMBER:
2074 case IGMP_SILENT_MEMBER:
2075 case IGMP_IDLE_MEMBER:
2076 case IGMP_LAZY_MEMBER:
2077 case IGMP_SLEEPING_MEMBER:
2078 case IGMP_AWAKENING_MEMBER:
2080 * These states are either not relevant in v3 mode,
2081 * or are unreported. Do nothing.
2084 case IGMP_LEAVING_MEMBER:
2086 * If we are leaving the group and switching to
2087 * compatibility mode, we need to release the final
2088 * reference held for issuing the INCLUDE {}, and
2089 * transition to REPORTING to ensure the host leave
2090 * message is sent upstream to the old querier --
2091 * transition to NOT would lose the leave and race.
2093 inm_rele_locked(&inm_free_tmp, inm);
2095 case IGMP_G_QUERY_PENDING_MEMBER:
2096 case IGMP_SG_QUERY_PENDING_MEMBER:
2097 inm_clear_recorded(inm);
2099 case IGMP_REPORTING_MEMBER:
2100 inm->inm_state = IGMP_REPORTING_MEMBER;
2104 * Always clear state-change and group report timers.
2105 * Free any pending IGMPv3 state-change records.
2107 inm->inm_sctimer = 0;
2109 mbufq_drain(&inm->inm_scq);
2111 IF_ADDR_WUNLOCK(ifp);
2113 inm_release_list_deferred(&inm_free_tmp);
2117 * Update the Older Version Querier Present timers for a link.
2118 * See Section 7.2.1 of RFC 3376.
2121 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi)
2126 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2128 * IGMPv1 and IGMPv2 Querier Present timers expired.
2132 if (igi->igi_version != IGMP_VERSION_3) {
2134 "%s: transition from v%d -> v%d on %p(%s)",
2135 __func__, igi->igi_version, IGMP_VERSION_3,
2136 igi->igi_ifp, igi->igi_ifp->if_xname);
2137 igi->igi_version = IGMP_VERSION_3;
2139 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2141 * IGMPv1 Querier Present timer expired,
2142 * IGMPv2 Querier Present timer running.
2143 * If IGMPv2 was disabled since last timeout,
2145 * If IGMPv2 is enabled, revert to IGMPv2.
2147 if (!V_igmp_v2enable) {
2149 "%s: transition from v%d -> v%d on %p(%s)",
2150 __func__, igi->igi_version, IGMP_VERSION_3,
2151 igi->igi_ifp, igi->igi_ifp->if_xname);
2152 igi->igi_v2_timer = 0;
2153 igi->igi_version = IGMP_VERSION_3;
2155 --igi->igi_v2_timer;
2156 if (igi->igi_version != IGMP_VERSION_2) {
2158 "%s: transition from v%d -> v%d on %p(%s)",
2159 __func__, igi->igi_version, IGMP_VERSION_2,
2160 igi->igi_ifp, igi->igi_ifp->if_xname);
2161 igi->igi_version = IGMP_VERSION_2;
2162 igmp_v3_cancel_link_timers(igi);
2165 } else if (igi->igi_v1_timer > 0) {
2167 * IGMPv1 Querier Present timer running.
2168 * Stop IGMPv2 timer if running.
2170 * If IGMPv1 was disabled since last timeout,
2172 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2174 if (!V_igmp_v1enable) {
2176 "%s: transition from v%d -> v%d on %p(%s)",
2177 __func__, igi->igi_version, IGMP_VERSION_3,
2178 igi->igi_ifp, igi->igi_ifp->if_xname);
2179 igi->igi_v1_timer = 0;
2180 igi->igi_version = IGMP_VERSION_3;
2182 --igi->igi_v1_timer;
2184 if (igi->igi_v2_timer > 0) {
2186 "%s: cancel v2 timer on %p(%s)",
2187 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2188 igi->igi_v2_timer = 0;
2194 * Global slowtimo handler.
2195 * VIMAGE: Timeout handlers are expected to service all vimages.
2200 VNET_ITERATOR_DECL(vnet_iter);
2202 VNET_LIST_RLOCK_NOSLEEP();
2203 VNET_FOREACH(vnet_iter) {
2204 CURVNET_SET(vnet_iter);
2205 igmp_slowtimo_vnet();
2208 VNET_LIST_RUNLOCK_NOSLEEP();
2212 * Per-vnet slowtimo handler.
2215 igmp_slowtimo_vnet(void)
2217 struct igmp_ifsoftc *igi;
2221 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2222 igmp_v1v2_process_querier_timers(igi);
2229 * Dispatch an IGMPv1/v2 host report or leave message.
2230 * These are always small enough to fit inside a single mbuf.
2233 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2235 struct epoch_tracker et;
2241 IN_MULTI_LIST_LOCK_ASSERT();
2246 m = m_gethdr(M_NOWAIT, MT_DATA);
2249 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2251 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2253 m->m_data += sizeof(struct ip);
2254 m->m_len = sizeof(struct igmp);
2256 igmp = mtod(m, struct igmp *);
2257 igmp->igmp_type = type;
2258 igmp->igmp_code = 0;
2259 igmp->igmp_group = inm->inm_addr;
2260 igmp->igmp_cksum = 0;
2261 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2263 m->m_data -= sizeof(struct ip);
2264 m->m_len += sizeof(struct ip);
2266 ip = mtod(m, struct ip *);
2268 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp));
2270 ip->ip_p = IPPROTO_IGMP;
2271 ip->ip_src.s_addr = INADDR_ANY;
2273 if (type == IGMP_HOST_LEAVE_MESSAGE)
2274 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2276 ip->ip_dst = inm->inm_addr;
2278 igmp_save_context(m, ifp);
2280 m->m_flags |= M_IGMPV2;
2281 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2282 m->m_flags |= M_IGMP_LOOP;
2284 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2285 NET_EPOCH_ENTER(et);
2286 netisr_dispatch(NETISR_IGMP, m);
2293 * Process a state change from the upper layer for the given IPv4 group.
2295 * Each socket holds a reference on the in_multi in its own ip_moptions.
2296 * The socket layer will have made the necessary updates to.the group
2297 * state, it is now up to IGMP to issue a state change report if there
2298 * has been any change between T0 (when the last state-change was issued)
2301 * We use the IGMPv3 state machine at group level. The IGMP module
2302 * however makes the decision as to which IGMP protocol version to speak.
2303 * A state change *from* INCLUDE {} always means an initial join.
2304 * A state change *to* INCLUDE {} always means a final leave.
2306 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2307 * save ourselves a bunch of work; any exclusive mode groups need not
2308 * compute source filter lists.
2310 * VIMAGE: curvnet should have been set by caller, as this routine
2311 * is called from the socket option handlers.
2314 igmp_change_state(struct in_multi *inm)
2316 struct igmp_ifsoftc *igi;
2321 IN_MULTI_LOCK_ASSERT();
2323 * Try to detect if the upper layer just asked us to change state
2324 * for an interface which has now gone away.
2326 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2327 ifp = inm->inm_ifma->ifma_ifp;
2329 * Sanity check that netinet's notion of ifp is the
2332 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2336 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2337 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp));
2340 * If we detect a state transition to or from MCAST_UNDEFINED
2341 * for this group, then we are starting or finishing an IGMP
2342 * life cycle for this group.
2344 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2345 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2346 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2347 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2348 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2349 error = igmp_initial_join(inm, igi);
2351 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2352 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2353 igmp_final_leave(inm, igi);
2357 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2360 error = igmp_handle_state_change(inm, igi);
2368 * Perform the initial join for an IGMP group.
2370 * When joining a group:
2371 * If the group should have its IGMP traffic suppressed, do nothing.
2372 * IGMPv1 starts sending IGMPv1 host membership reports.
2373 * IGMPv2 starts sending IGMPv2 host membership reports.
2374 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2375 * initial state of the membership.
2378 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi)
2382 int error, retval, syncstates;
2384 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__,
2385 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2392 IN_MULTI_LOCK_ASSERT();
2395 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2398 * Groups joined on loopback or marked as 'not reported',
2399 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2400 * are never reported in any IGMP protocol exchanges.
2401 * All other groups enter the appropriate IGMP state machine
2402 * for the version in use on this link.
2403 * A link marked as IGIF_SILENT causes IGMP to be completely
2404 * disabled for the link.
2406 if ((ifp->if_flags & IFF_LOOPBACK) ||
2407 (igi->igi_flags & IGIF_SILENT) ||
2408 !igmp_isgroupreported(inm->inm_addr)) {
2410 "%s: not kicking state machine for silent group", __func__);
2411 inm->inm_state = IGMP_SILENT_MEMBER;
2415 * Deal with overlapping in_multi lifecycle.
2416 * If this group was LEAVING, then make sure
2417 * we drop the reference we picked up to keep the
2418 * group around for the final INCLUDE {} enqueue.
2420 if (igi->igi_version == IGMP_VERSION_3 &&
2421 inm->inm_state == IGMP_LEAVING_MEMBER) {
2422 MPASS(inm->inm_refcount > 1);
2423 inm_rele_locked(NULL, inm);
2425 inm->inm_state = IGMP_REPORTING_MEMBER;
2427 switch (igi->igi_version) {
2428 case IGMP_VERSION_1:
2429 case IGMP_VERSION_2:
2430 inm->inm_state = IGMP_IDLE_MEMBER;
2431 error = igmp_v1v2_queue_report(inm,
2432 (igi->igi_version == IGMP_VERSION_2) ?
2433 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2434 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2436 inm->inm_timer = IGMP_RANDOM_DELAY(
2437 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2438 V_current_state_timers_running = 1;
2442 case IGMP_VERSION_3:
2444 * Defer update of T0 to T1, until the first copy
2445 * of the state change has been transmitted.
2450 * Immediately enqueue a State-Change Report for
2451 * this interface, freeing any previous reports.
2452 * Don't kick the timers if there is nothing to do,
2453 * or if an error occurred.
2457 retval = igmp_v3_enqueue_group_record(mq, inm, 1,
2459 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2462 error = retval * -1;
2467 * Schedule transmission of pending state-change
2468 * report up to RV times for this link. The timer
2469 * will fire at the next igmp_fasttimo (~200ms),
2470 * giving us an opportunity to merge the reports.
2472 if (igi->igi_flags & IGIF_LOOPBACK) {
2475 KASSERT(igi->igi_rv > 1,
2476 ("%s: invalid robustness %d", __func__,
2478 inm->inm_scrv = igi->igi_rv;
2480 inm->inm_sctimer = 1;
2481 V_state_change_timers_running = 1;
2489 * Only update the T0 state if state change is atomic,
2490 * i.e. we don't need to wait for a timer to fire before we
2491 * can consider the state change to have been communicated.
2495 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2496 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2503 * Issue an intermediate state change during the IGMP life-cycle.
2506 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi)
2511 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__,
2512 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname);
2516 IN_MULTI_LIST_LOCK_ASSERT();
2519 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2521 if ((ifp->if_flags & IFF_LOOPBACK) ||
2522 (igi->igi_flags & IGIF_SILENT) ||
2523 !igmp_isgroupreported(inm->inm_addr) ||
2524 (igi->igi_version != IGMP_VERSION_3)) {
2525 if (!igmp_isgroupreported(inm->inm_addr)) {
2527 "%s: not kicking state machine for silent group", __func__);
2529 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2531 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2532 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2536 mbufq_drain(&inm->inm_scq);
2538 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2539 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2544 * If record(s) were enqueued, start the state-change
2545 * report timer for this group.
2547 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2548 inm->inm_sctimer = 1;
2549 V_state_change_timers_running = 1;
2555 * Perform the final leave for an IGMP group.
2557 * When leaving a group:
2558 * IGMPv1 does nothing.
2559 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2560 * IGMPv3 enqueues a state-change report containing a transition
2561 * to INCLUDE {} for immediate transmission.
2564 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi)
2570 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)",
2571 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp,
2572 inm->inm_ifp->if_xname);
2574 IN_MULTI_LIST_LOCK_ASSERT();
2577 switch (inm->inm_state) {
2578 case IGMP_NOT_MEMBER:
2579 case IGMP_SILENT_MEMBER:
2580 case IGMP_LEAVING_MEMBER:
2581 /* Already leaving or left; do nothing. */
2583 "%s: not kicking state machine for silent group", __func__);
2585 case IGMP_REPORTING_MEMBER:
2586 case IGMP_IDLE_MEMBER:
2587 case IGMP_G_QUERY_PENDING_MEMBER:
2588 case IGMP_SG_QUERY_PENDING_MEMBER:
2589 if (igi->igi_version == IGMP_VERSION_2) {
2591 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2592 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2593 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2596 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2597 inm->inm_state = IGMP_NOT_MEMBER;
2598 } else if (igi->igi_version == IGMP_VERSION_3) {
2600 * Stop group timer and all pending reports.
2601 * Immediately enqueue a state-change report
2602 * TO_IN {} to be sent on the next fast timeout,
2603 * giving us an opportunity to merge reports.
2605 mbufq_drain(&inm->inm_scq);
2607 if (igi->igi_flags & IGIF_LOOPBACK) {
2610 inm->inm_scrv = igi->igi_rv;
2612 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d "
2613 "pending retransmissions.", __func__,
2614 ntohl(inm->inm_addr.s_addr),
2615 inm->inm_ifp->if_xname, inm->inm_scrv);
2616 if (inm->inm_scrv == 0) {
2617 inm->inm_state = IGMP_NOT_MEMBER;
2618 inm->inm_sctimer = 0;
2620 int retval __unused;
2622 inm_acquire_locked(inm);
2624 retval = igmp_v3_enqueue_group_record(
2625 &inm->inm_scq, inm, 1, 0, 0);
2626 KASSERT(retval != 0,
2627 ("%s: enqueue record = %d", __func__,
2630 inm->inm_state = IGMP_LEAVING_MEMBER;
2631 inm->inm_sctimer = 1;
2632 V_state_change_timers_running = 1;
2638 case IGMP_LAZY_MEMBER:
2639 case IGMP_SLEEPING_MEMBER:
2640 case IGMP_AWAKENING_MEMBER:
2641 /* Our reports are suppressed; do nothing. */
2647 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__,
2648 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2649 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2650 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s",
2651 __func__, ntohl(inm->inm_addr.s_addr),
2652 inm->inm_ifp->if_xname);
2657 * Enqueue an IGMPv3 group record to the given output queue.
2659 * XXX This function could do with having the allocation code
2660 * split out, and the multiple-tree-walks coalesced into a single
2661 * routine as has been done in igmp_v3_enqueue_filter_change().
2663 * If is_state_change is zero, a current-state record is appended.
2664 * If is_state_change is non-zero, a state-change report is appended.
2666 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2667 * If is_group_query is zero, and if there is a packet with free space
2668 * at the tail of the queue, it will be appended to providing there
2669 * is enough free space.
2670 * Otherwise a new mbuf packet chain is allocated.
2672 * If is_source_query is non-zero, each source is checked to see if
2673 * it was recorded for a Group-Source query, and will be omitted if
2674 * it is not both in-mode and recorded.
2676 * The function will attempt to allocate leading space in the packet
2677 * for the IP/IGMP header to be prepended without fragmenting the chain.
2679 * If successful the size of all data appended to the queue is returned,
2680 * otherwise an error code less than zero is returned, or zero if
2681 * no record(s) were appended.
2684 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm,
2685 const int is_state_change, const int is_group_query,
2686 const int is_source_query)
2688 struct igmp_grouprec ig;
2689 struct igmp_grouprec *pig;
2691 struct ip_msource *ims, *nims;
2692 struct mbuf *m0, *m, *md;
2693 int is_filter_list_change;
2694 int minrec0len, m0srcs, msrcs, nbytes, off;
2695 int record_has_sources;
2701 IN_MULTI_LIST_LOCK_ASSERT();
2704 is_filter_list_change = 0;
2711 record_has_sources = 1;
2713 type = IGMP_DO_NOTHING;
2714 mode = inm->inm_st[1].iss_fmode;
2717 * If we did not transition out of ASM mode during t0->t1,
2718 * and there are no source nodes to process, we can skip
2719 * the generation of source records.
2721 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2723 record_has_sources = 0;
2725 if (is_state_change) {
2727 * Queue a state change record.
2728 * If the mode did not change, and there are non-ASM
2729 * listeners or source filters present,
2730 * we potentially need to issue two records for the group.
2731 * If we are transitioning to MCAST_UNDEFINED, we need
2732 * not send any sources.
2733 * If there are ASM listeners, and there was no filter
2734 * mode transition of any kind, do nothing.
2736 if (mode != inm->inm_st[0].iss_fmode) {
2737 if (mode == MCAST_EXCLUDE) {
2738 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2740 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2742 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2744 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2745 if (mode == MCAST_UNDEFINED)
2746 record_has_sources = 0;
2749 if (record_has_sources) {
2750 is_filter_list_change = 1;
2752 type = IGMP_DO_NOTHING;
2757 * Queue a current state record.
2759 if (mode == MCAST_EXCLUDE) {
2760 type = IGMP_MODE_IS_EXCLUDE;
2761 } else if (mode == MCAST_INCLUDE) {
2762 type = IGMP_MODE_IS_INCLUDE;
2763 KASSERT(inm->inm_st[1].iss_asm == 0,
2764 ("%s: inm %p is INCLUDE but ASM count is %d",
2765 __func__, inm, inm->inm_st[1].iss_asm));
2770 * Generate the filter list changes using a separate function.
2772 if (is_filter_list_change)
2773 return (igmp_v3_enqueue_filter_change(mq, inm));
2775 if (type == IGMP_DO_NOTHING) {
2776 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__,
2777 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname);
2782 * If any sources are present, we must be able to fit at least
2783 * one in the trailing space of the tail packet's mbuf,
2786 minrec0len = sizeof(struct igmp_grouprec);
2787 if (record_has_sources)
2788 minrec0len += sizeof(in_addr_t);
2790 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__,
2791 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr),
2792 inm->inm_ifp->if_xname);
2795 * Check if we have a packet in the tail of the queue for this
2796 * group into which the first group record for this group will fit.
2797 * Otherwise allocate a new packet.
2798 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2799 * Note: Group records for G/GSR query responses MUST be sent
2800 * in their own packet.
2802 m0 = mbufq_last(mq);
2803 if (!is_group_query &&
2805 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2806 (m0->m_pkthdr.len + minrec0len) <
2807 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2808 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2809 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2811 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2813 if (mbufq_full(mq)) {
2814 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2818 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2819 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2820 if (!is_state_change && !is_group_query) {
2821 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2823 m->m_data += IGMP_LEADINGSPACE;
2826 m = m_gethdr(M_NOWAIT, MT_DATA);
2828 M_ALIGN(m, IGMP_LEADINGSPACE);
2833 igmp_save_context(m, ifp);
2835 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2839 * Append group record.
2840 * If we have sources, we don't know how many yet.
2845 ig.ig_group = inm->inm_addr;
2846 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2849 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2852 nbytes += sizeof(struct igmp_grouprec);
2855 * Append as many sources as will fit in the first packet.
2856 * If we are appending to a new packet, the chain allocation
2857 * may potentially use clusters; use m_getptr() in this case.
2858 * If we are appending to an existing packet, we need to obtain
2859 * a pointer to the group record after m_append(), in case a new
2860 * mbuf was allocated.
2861 * Only append sources which are in-mode at t1. If we are
2862 * transitioning to MCAST_UNDEFINED state on the group, do not
2863 * include source entries.
2864 * Only report recorded sources in our filter set when responding
2865 * to a group-source query.
2867 if (record_has_sources) {
2870 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2871 md->m_len - nbytes);
2873 md = m_getptr(m, 0, &off);
2874 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2878 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2879 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2881 now = ims_get_mode(inm, ims, 1);
2882 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2883 if ((now != mode) ||
2884 (now == mode && mode == MCAST_UNDEFINED)) {
2885 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2888 if (is_source_query && ims->ims_stp == 0) {
2889 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2893 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2894 naddr = htonl(ims->ims_haddr);
2895 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2898 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2902 nbytes += sizeof(in_addr_t);
2904 if (msrcs == m0srcs)
2907 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2909 pig->ig_numsrc = htons(msrcs);
2910 nbytes += (msrcs * sizeof(in_addr_t));
2913 if (is_source_query && msrcs == 0) {
2914 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2921 * We are good to go with first packet.
2924 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2925 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2926 mbufq_enqueue(mq, m);
2928 m->m_pkthdr.PH_vt.vt_nrecs++;
2931 * No further work needed if no source list in packet(s).
2933 if (!record_has_sources)
2937 * Whilst sources remain to be announced, we need to allocate
2938 * a new packet and fill out as many sources as will fit.
2939 * Always try for a cluster first.
2941 while (nims != NULL) {
2942 if (mbufq_full(mq)) {
2943 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2946 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2948 m->m_data += IGMP_LEADINGSPACE;
2950 m = m_gethdr(M_NOWAIT, MT_DATA);
2952 M_ALIGN(m, IGMP_LEADINGSPACE);
2956 igmp_save_context(m, ifp);
2957 md = m_getptr(m, 0, &off);
2958 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2959 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2961 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2964 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2967 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2968 nbytes += sizeof(struct igmp_grouprec);
2970 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2971 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2974 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2975 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2977 now = ims_get_mode(inm, ims, 1);
2978 if ((now != mode) ||
2979 (now == mode && mode == MCAST_UNDEFINED)) {
2980 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2983 if (is_source_query && ims->ims_stp == 0) {
2984 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2988 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2989 naddr = htonl(ims->ims_haddr);
2990 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2993 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2998 if (msrcs == m0srcs)
3001 pig->ig_numsrc = htons(msrcs);
3002 nbytes += (msrcs * sizeof(in_addr_t));
3004 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
3005 mbufq_enqueue(mq, m);
3012 * Type used to mark record pass completion.
3013 * We exploit the fact we can cast to this easily from the
3014 * current filter modes on each ip_msource node.
3017 REC_NONE = 0x00, /* MCAST_UNDEFINED */
3018 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
3019 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
3020 REC_FULL = REC_ALLOW | REC_BLOCK
3024 * Enqueue an IGMPv3 filter list change to the given output queue.
3026 * Source list filter state is held in an RB-tree. When the filter list
3027 * for a group is changed without changing its mode, we need to compute
3028 * the deltas between T0 and T1 for each source in the filter set,
3029 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3031 * As we may potentially queue two record types, and the entire R-B tree
3032 * needs to be walked at once, we break this out into its own function
3033 * so we can generate a tightly packed queue of packets.
3035 * XXX This could be written to only use one tree walk, although that makes
3036 * serializing into the mbuf chains a bit harder. For now we do two walks
3037 * which makes things easier on us, and it may or may not be harder on
3040 * If successful the size of all data appended to the queue is returned,
3041 * otherwise an error code less than zero is returned, or zero if
3042 * no record(s) were appended.
3045 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm)
3047 static const int MINRECLEN =
3048 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3050 struct igmp_grouprec ig;
3051 struct igmp_grouprec *pig;
3052 struct ip_msource *ims, *nims;
3053 struct mbuf *m, *m0, *md;
3055 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3057 uint8_t mode, now, then;
3058 rectype_t crt, drt, nrt;
3060 IN_MULTI_LIST_LOCK_ASSERT();
3062 if (inm->inm_nsrc == 0 ||
3063 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3066 ifp = inm->inm_ifp; /* interface */
3067 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3068 crt = REC_NONE; /* current group record type */
3069 drt = REC_NONE; /* mask of completed group record types */
3070 nrt = REC_NONE; /* record type for current node */
3071 m0srcs = 0; /* # source which will fit in current mbuf chain */
3072 nbytes = 0; /* # of bytes appended to group's state-change queue */
3073 npbytes = 0; /* # of bytes appended this packet */
3074 rsrcs = 0; /* # sources encoded in current record */
3075 schanged = 0; /* # nodes encoded in overall filter change */
3076 nallow = 0; /* # of source entries in ALLOW_NEW */
3077 nblock = 0; /* # of source entries in BLOCK_OLD */
3078 nims = NULL; /* next tree node pointer */
3081 * For each possible filter record mode.
3082 * The first kind of source we encounter tells us which
3083 * is the first kind of record we start appending.
3084 * If a node transitioned to UNDEFINED at t1, its mode is treated
3085 * as the inverse of the group's filter mode.
3087 while (drt != REC_FULL) {
3089 m0 = mbufq_last(mq);
3091 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3092 IGMP_V3_REPORT_MAXRECS) &&
3093 (m0->m_pkthdr.len + MINRECLEN) <
3094 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3096 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3097 sizeof(struct igmp_grouprec)) /
3100 "%s: use previous packet", __func__);
3102 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
3104 m->m_data += IGMP_LEADINGSPACE;
3106 m = m_gethdr(M_NOWAIT, MT_DATA);
3108 M_ALIGN(m, IGMP_LEADINGSPACE);
3112 "%s: m_get*() failed", __func__);
3115 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3116 igmp_save_context(m, ifp);
3117 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3118 sizeof(struct igmp_grouprec)) /
3122 "%s: allocated new packet", __func__);
3125 * Append the IGMP group record header to the
3126 * current packet's data area.
3127 * Recalculate pointer to free space for next
3128 * group record, in case m_append() allocated
3129 * a new mbuf or cluster.
3131 memset(&ig, 0, sizeof(ig));
3132 ig.ig_group = inm->inm_addr;
3133 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3137 "%s: m_append() failed", __func__);
3140 npbytes += sizeof(struct igmp_grouprec);
3142 /* new packet; offset in c hain */
3143 md = m_getptr(m, npbytes -
3144 sizeof(struct igmp_grouprec), &off);
3145 pig = (struct igmp_grouprec *)(mtod(md,
3148 /* current packet; offset from last append */
3150 pig = (struct igmp_grouprec *)(mtod(md,
3151 uint8_t *) + md->m_len -
3152 sizeof(struct igmp_grouprec));
3155 * Begin walking the tree for this record type
3156 * pass, or continue from where we left off
3157 * previously if we had to allocate a new packet.
3158 * Only report deltas in-mode at t1.
3159 * We need not report included sources as allowed
3160 * if we are in inclusive mode on the group,
3161 * however the converse is not true.
3165 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3166 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3167 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x",
3168 __func__, ims->ims_haddr);
3169 now = ims_get_mode(inm, ims, 1);
3170 then = ims_get_mode(inm, ims, 0);
3171 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3172 __func__, then, now);
3175 "%s: skip unchanged", __func__);
3178 if (mode == MCAST_EXCLUDE &&
3179 now == MCAST_INCLUDE) {
3181 "%s: skip IN src on EX group",
3185 nrt = (rectype_t)now;
3186 if (nrt == REC_NONE)
3187 nrt = (rectype_t)(~mode & REC_FULL);
3188 if (schanged++ == 0) {
3190 } else if (crt != nrt)
3192 naddr = htonl(ims->ims_haddr);
3193 if (!m_append(m, sizeof(in_addr_t),
3198 "%s: m_append() failed", __func__);
3201 nallow += !!(crt == REC_ALLOW);
3202 nblock += !!(crt == REC_BLOCK);
3203 if (++rsrcs == m0srcs)
3207 * If we did not append any tree nodes on this
3208 * pass, back out of allocations.
3211 npbytes -= sizeof(struct igmp_grouprec);
3214 "%s: m_free(m)", __func__);
3218 "%s: m_adj(m, -ig)", __func__);
3219 m_adj(m, -((int)sizeof(
3220 struct igmp_grouprec)));
3224 npbytes += (rsrcs * sizeof(in_addr_t));
3225 if (crt == REC_ALLOW)
3226 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3227 else if (crt == REC_BLOCK)
3228 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3229 pig->ig_numsrc = htons(rsrcs);
3231 * Count the new group record, and enqueue this
3232 * packet if it wasn't already queued.
3234 m->m_pkthdr.PH_vt.vt_nrecs++;
3236 mbufq_enqueue(mq, m);
3238 } while (nims != NULL);
3240 crt = (~crt & REC_FULL);
3243 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3250 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq)
3253 struct mbuf *m; /* pending state-change */
3254 struct mbuf *m0; /* copy of pending state-change */
3255 struct mbuf *mt; /* last state-change in packet */
3256 int docopy, domerge;
3263 IN_MULTI_LIST_LOCK_ASSERT();
3267 * If there are further pending retransmissions, make a writable
3268 * copy of each queued state-change message before merging.
3270 if (inm->inm_scrv > 0)
3275 if (mbufq_first(gq) == NULL) {
3276 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3281 m = mbufq_first(gq);
3284 * Only merge the report into the current packet if
3285 * there is sufficient space to do so; an IGMPv3 report
3286 * packet may only contain 65,535 group records.
3287 * Always use a simple mbuf chain concatentation to do this,
3288 * as large state changes for single groups may have
3289 * allocated clusters.
3292 mt = mbufq_last(scq);
3294 recslen = m_length(m, NULL);
3296 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3297 m->m_pkthdr.PH_vt.vt_nrecs <=
3298 IGMP_V3_REPORT_MAXRECS) &&
3299 (mt->m_pkthdr.len + recslen <=
3300 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3304 if (!domerge && mbufq_full(gq)) {
3306 "%s: outbound queue full, skipping whole packet %p",
3316 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3317 m0 = mbufq_dequeue(gq);
3320 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3321 m0 = m_dup(m, M_NOWAIT);
3324 m0->m_nextpkt = NULL;
3329 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)",
3331 mbufq_enqueue(scq, m0);
3333 struct mbuf *mtl; /* last mbuf of packet mt */
3335 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)",
3339 m0->m_flags &= ~M_PKTHDR;
3340 mt->m_pkthdr.len += recslen;
3341 mt->m_pkthdr.PH_vt.vt_nrecs +=
3342 m0->m_pkthdr.PH_vt.vt_nrecs;
3352 * Respond to a pending IGMPv3 General Query.
3355 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi)
3357 struct ifmultiaddr *ifma;
3359 struct in_multi *inm;
3360 int retval __unused, loop;
3362 IN_MULTI_LIST_LOCK_ASSERT();
3366 KASSERT(igi->igi_version == IGMP_VERSION_3,
3367 ("%s: called when version %d", __func__, igi->igi_version));
3370 * Check that there are some packets queued. If so, send them first.
3371 * For large number of groups the reply to general query can take
3372 * many packets, we should finish sending them before starting of
3373 * queuing the new reply.
3375 if (mbufq_len(&igi->igi_gq) != 0)
3380 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3381 if (ifma->ifma_addr->sa_family != AF_INET ||
3382 ifma->ifma_protospec == NULL)
3385 inm = (struct in_multi *)ifma->ifma_protospec;
3386 KASSERT(ifp == inm->inm_ifp,
3387 ("%s: inconsistent ifp", __func__));
3389 switch (inm->inm_state) {
3390 case IGMP_NOT_MEMBER:
3391 case IGMP_SILENT_MEMBER:
3393 case IGMP_REPORTING_MEMBER:
3394 case IGMP_IDLE_MEMBER:
3395 case IGMP_LAZY_MEMBER:
3396 case IGMP_SLEEPING_MEMBER:
3397 case IGMP_AWAKENING_MEMBER:
3398 inm->inm_state = IGMP_REPORTING_MEMBER;
3399 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3401 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3404 case IGMP_G_QUERY_PENDING_MEMBER:
3405 case IGMP_SG_QUERY_PENDING_MEMBER:
3406 case IGMP_LEAVING_MEMBER:
3412 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3413 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3416 * Slew transmission of bursts over 500ms intervals.
3418 if (mbufq_first(&igi->igi_gq) != NULL) {
3419 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3420 IGMP_RESPONSE_BURST_INTERVAL);
3421 V_interface_timers_running = 1;
3426 * Transmit the next pending IGMP message in the output queue.
3428 * We get called from netisr_processqueue(). A mutex private to igmpoq
3429 * will be acquired and released around this routine.
3431 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3432 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3433 * a link and uses a link-scope multicast address.
3436 igmp_intr(struct mbuf *m)
3438 struct ip_moptions imo;
3440 struct mbuf *ipopts, *m0;
3444 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3447 * Set VNET image pointer from enqueued mbuf chain
3448 * before doing anything else. Whilst we use interface
3449 * indexes to guard against interface detach, they are
3450 * unique to each VIMAGE and must be retrieved.
3452 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr));
3453 ifindex = igmp_restore_context(m);
3456 * Check if the ifnet still exists. This limits the scope of
3457 * any race in the absence of a global ifp lock for low cost
3458 * (an array lookup).
3460 ifp = ifnet_byindex(ifindex);
3462 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3463 __func__, m, ifindex);
3465 IPSTAT_INC(ips_noroute);
3469 ipopts = V_igmp_sendra ? m_raopt : NULL;
3471 imo.imo_multicast_ttl = 1;
3472 imo.imo_multicast_vif = -1;
3473 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3476 * If the user requested that IGMP traffic be explicitly
3477 * redirected to the loopback interface (e.g. they are running a
3478 * MANET interface and the routing protocol needs to see the
3479 * updates), handle this now.
3481 if (m->m_flags & M_IGMP_LOOP)
3482 imo.imo_multicast_ifp = V_loif;
3484 imo.imo_multicast_ifp = ifp;
3486 if (m->m_flags & M_IGMPV2) {
3489 m0 = igmp_v3_encap_report(ifp, m);
3491 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3493 IPSTAT_INC(ips_odropped);
3498 igmp_scrub_context(m0);
3500 m0->m_pkthdr.rcvif = V_loif;
3502 mac_netinet_igmp_send(ifp, m0);
3504 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3506 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3510 IGMPSTAT_INC(igps_snd_reports);
3514 * We must restore the existing vnet pointer before
3515 * continuing as we are run from netisr context.
3521 * Encapsulate an IGMPv3 report.
3523 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3524 * chain has already had its IP/IGMPv3 header prepended. In this case
3525 * the function will not attempt to prepend; the lengths and checksums
3526 * will however be re-computed.
3528 * Returns a pointer to the new mbuf chain head, or NULL if the
3529 * allocation failed.
3531 static struct mbuf *
3532 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3534 struct rm_priotracker in_ifa_tracker;
3535 struct igmp_report *igmp;
3537 int hdrlen, igmpreclen;
3539 KASSERT((m->m_flags & M_PKTHDR),
3540 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3542 igmpreclen = m_length(m, NULL);
3543 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3545 if (m->m_flags & M_IGMPV3_HDR) {
3546 igmpreclen -= hdrlen;
3548 M_PREPEND(m, hdrlen, M_NOWAIT);
3551 m->m_flags |= M_IGMPV3_HDR;
3554 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3556 m->m_data += sizeof(struct ip);
3557 m->m_len -= sizeof(struct ip);
3559 igmp = mtod(m, struct igmp_report *);
3560 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3563 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3565 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3566 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3568 m->m_data -= sizeof(struct ip);
3569 m->m_len += sizeof(struct ip);
3571 ip = mtod(m, struct ip *);
3572 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3573 ip->ip_len = htons(hdrlen + igmpreclen);
3574 ip->ip_off = htons(IP_DF);
3575 ip->ip_p = IPPROTO_IGMP;
3578 ip->ip_src.s_addr = INADDR_ANY;
3580 if (m->m_flags & M_IGMP_LOOP) {
3581 struct in_ifaddr *ia;
3583 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
3585 ip->ip_src = ia->ia_addr.sin_addr;
3588 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3595 igmp_rec_type_to_str(const int type)
3599 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3602 case IGMP_CHANGE_TO_INCLUDE_MODE:
3605 case IGMP_MODE_IS_EXCLUDE:
3608 case IGMP_MODE_IS_INCLUDE:
3611 case IGMP_ALLOW_NEW_SOURCES:
3614 case IGMP_BLOCK_OLD_SOURCES:
3626 vnet_igmp_init(const void *unused __unused)
3629 netisr_register_vnet(&igmp_nh);
3631 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3632 vnet_igmp_init, NULL);
3635 vnet_igmp_uninit(const void *unused __unused)
3638 /* This can happen when we shutdown the entire network stack. */
3639 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3641 netisr_unregister_vnet(&igmp_nh);
3643 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY,
3644 vnet_igmp_uninit, NULL);
3648 DB_SHOW_COMMAND(igi_list, db_show_igi_list)
3650 struct igmp_ifsoftc *igi, *tigi;
3651 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head;
3654 db_printf("usage: show igi_list <addr>\n");
3657 igi_head = (struct _igi_list *)addr;
3659 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) {
3660 db_printf("igmp_ifsoftc %p:\n", igi);
3661 db_printf(" ifp %p\n", igi->igi_ifp);
3662 db_printf(" version %u\n", igi->igi_version);
3663 db_printf(" v1_timer %u\n", igi->igi_v1_timer);
3664 db_printf(" v2_timer %u\n", igi->igi_v2_timer);
3665 db_printf(" v3_timer %u\n", igi->igi_v3_timer);
3666 db_printf(" flags %#x\n", igi->igi_flags);
3667 db_printf(" rv %u\n", igi->igi_rv);
3668 db_printf(" qi %u\n", igi->igi_qi);
3669 db_printf(" qri %u\n", igi->igi_qri);
3670 db_printf(" uri %u\n", igi->igi_uri);
3671 /* struct mbufq igi_gq; */
3678 igmp_modevent(module_t mod, int type, void *unused __unused)
3683 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3685 m_raopt = igmp_ra_alloc();
3686 netisr_register(&igmp_nh);
3689 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3690 netisr_unregister(&igmp_nh);
3693 IGMP_LOCK_DESTROY();
3696 return (EOPNOTSUPP);
3701 static moduledata_t igmp_mod = {
3706 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE);