2 * Copyright (c) 2007-2009 Bruce Simpson.
3 * Copyright (c) 1988 Stephen Deering.
4 * Copyright (c) 1992, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Stephen Deering of Stanford University.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)igmp.c 8.1 (Berkeley) 7/19/93
38 * Internet Group Management Protocol (IGMP) routines.
39 * [RFC1112, RFC2236, RFC3376]
41 * Written by Steve Deering, Stanford, May 1988.
42 * Modified by Rosen Sharma, Stanford, Aug 1994.
43 * Modified by Bill Fenner, Xerox PARC, Feb 1995.
44 * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995.
45 * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson.
47 * MULTICAST Revision: 3.5.1.4
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/module.h>
56 #include <sys/malloc.h>
58 #include <sys/socket.h>
59 #include <sys/protosw.h>
60 #include <sys/kernel.h>
61 #include <sys/sysctl.h>
62 #include <sys/vimage.h>
64 #include <sys/condvar.h>
67 #include <net/netisr.h>
68 #include <net/route.h>
71 #include <netinet/in.h>
72 #include <netinet/in_var.h>
73 #include <netinet/in_systm.h>
74 #include <netinet/ip.h>
75 #include <netinet/ip_var.h>
76 #include <netinet/ip_options.h>
77 #include <netinet/igmp.h>
78 #include <netinet/igmp_var.h>
79 #include <netinet/vinet.h>
81 #include <machine/in_cksum.h>
83 #include <security/mac/mac_framework.h>
86 #define KTR_IGMPV3 KTR_INET
89 static struct igmp_ifinfo *
90 igi_alloc_locked(struct ifnet *);
91 static void igi_delete_locked(const struct ifnet *);
92 static void igmp_dispatch_queue(struct ifqueue *, int, const int);
93 static void igmp_fasttimo_vnet(void);
94 static void igmp_final_leave(struct in_multi *, struct igmp_ifinfo *);
95 static int igmp_handle_state_change(struct in_multi *,
96 struct igmp_ifinfo *);
97 static int igmp_initial_join(struct in_multi *, struct igmp_ifinfo *);
98 static int igmp_input_v1_query(struct ifnet *, const struct ip *,
100 static int igmp_input_v2_query(struct ifnet *, const struct ip *,
101 const struct igmp *);
102 static int igmp_input_v3_query(struct ifnet *, const struct ip *,
103 /*const*/ struct igmpv3 *);
104 static int igmp_input_v3_group_query(struct in_multi *,
105 struct igmp_ifinfo *, int, /*const*/ struct igmpv3 *);
106 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *,
107 /*const*/ struct igmp *);
108 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *,
109 /*const*/ struct igmp *);
110 static void igmp_intr(struct mbuf *);
111 static int igmp_isgroupreported(const struct in_addr);
115 static char * igmp_rec_type_to_str(const int);
117 static void igmp_set_version(struct igmp_ifinfo *, const int);
118 static void igmp_slowtimo_vnet(void);
119 static void igmp_sysinit(void);
120 static int igmp_v1v2_queue_report(struct in_multi *, const int);
121 static void igmp_v1v2_process_group_timer(struct in_multi *, const int);
122 static void igmp_v1v2_process_querier_timers(struct igmp_ifinfo *);
123 static void igmp_v2_update_group(struct in_multi *, const int);
124 static void igmp_v3_cancel_link_timers(struct igmp_ifinfo *);
125 static void igmp_v3_dispatch_general_query(struct igmp_ifinfo *);
127 igmp_v3_encap_report(struct ifnet *, struct mbuf *);
128 static int igmp_v3_enqueue_group_record(struct ifqueue *,
129 struct in_multi *, const int, const int, const int);
130 static int igmp_v3_enqueue_filter_change(struct ifqueue *,
132 static void igmp_v3_process_group_timers(struct igmp_ifinfo *,
133 struct ifqueue *, struct ifqueue *, struct in_multi *,
135 static int igmp_v3_merge_state_changes(struct in_multi *,
137 static void igmp_v3_suppress_group_record(struct in_multi *);
138 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS);
139 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS);
140 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS);
142 static vnet_attach_fn vnet_igmp_iattach;
143 static vnet_detach_fn vnet_igmp_idetach;
145 static const struct netisr_handler igmp_nh = {
147 .nh_handler = igmp_intr,
148 .nh_proto = NETISR_IGMP,
149 .nh_policy = NETISR_POLICY_SOURCE,
153 * System-wide globals.
155 * Unlocked access to these is OK, except for the global IGMP output
156 * queue. The IGMP subsystem lock ends up being system-wide for the moment,
157 * because all VIMAGEs have to share a global output queue, as netisrs
158 * themselves are not virtualized.
161 * * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
162 * Any may be taken independently; if any are held at the same
163 * time, the above lock order must be followed.
164 * * All output is delegated to the netisr.
165 * Now that Giant has been eliminated, the netisr may be inlined.
166 * * IN_MULTI_LOCK covers in_multi.
167 * * IGMP_LOCK covers igmp_ifinfo and any global variables in this file,
168 * including the output queue.
169 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
170 * per-link state iterators.
171 * * igmp_ifinfo is valid as long as PF_INET is attached to the interface,
172 * therefore it is not refcounted.
173 * We allow unlocked reads of igmp_ifinfo when accessed via in_multi.
176 * * IGMP acquires its own reference every time an in_multi is passed to
177 * it and the group is being joined for the first time.
178 * * IGMP releases its reference(s) on in_multi in a deferred way,
179 * because the operations which process the release run as part of
180 * a loop whose control variables are directly affected by the release
181 * (that, and not recursing on the IF_ADDR_LOCK).
183 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds
184 * to a vnet in ifp->if_vnet.
186 * SMPng: XXX We may potentially race operations on ifma_protospec.
187 * The problem is that we currently lack a clean way of taking the
188 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing,
189 * as anything which modifies ifma needs to be covered by that lock.
190 * So check for ifma_protospec being NULL before proceeding.
194 struct mbuf *m_raopt; /* Router Alert option */
195 MALLOC_DEFINE(M_IGMP, "igmp", "igmp state");
198 * VIMAGE-wide globals.
200 * The IGMPv3 timers themselves need to run per-image, however,
201 * protosw timers run globally (see tcp).
202 * An ifnet can only be in one vimage at a time, and the loopback
203 * ifnet, loif, is itself virtualized.
204 * It would otherwise be possible to seriously hose IGMP state,
205 * and create inconsistencies in upstream multicast routing, if you have
206 * multiple VIMAGEs running on the same link joining different multicast
207 * groups, UNLESS the "primary IP address" is different. This is because
208 * IGMP for IPv4 does not force link-local addresses to be used for each
209 * node, unlike MLD for IPv6.
210 * Obviously the IGMPv3 per-interface state has per-vimage granularity
213 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection
214 * policy to control the address used by IGMP on the link.
216 #ifdef VIMAGE_GLOBALS
217 int interface_timers_running; /* IGMPv3 general query response */
218 int state_change_timers_running; /* IGMPv3 state-change retransmit */
219 int current_state_timers_running; /* IGMPv1/v2 host report;
220 * IGMPv3 g/sg query response */
222 LIST_HEAD(, igmp_ifinfo) igi_head;
223 struct igmpstat igmpstat;
224 struct timeval igmp_gsrdelay;
226 int igmp_recvifkludge;
232 int igmp_default_version;
233 #endif /* VIMAGE_GLOBALS */
236 * Virtualized sysctls.
238 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_igmp, IGMPCTL_STATS, stats,
239 CTLFLAG_RW, igmpstat, igmpstat, "");
240 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, recvifkludge,
241 CTLFLAG_RW, igmp_recvifkludge, 0,
242 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address");
243 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, sendra,
244 CTLFLAG_RW, igmp_sendra, 0,
245 "Send IP Router Alert option in IGMPv2/v3 messages");
246 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, sendlocal,
247 CTLFLAG_RW, igmp_sendlocal, 0,
248 "Send IGMP membership reports for 224.0.0.0/24 groups");
249 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, v1enable,
250 CTLFLAG_RW, igmp_v1enable, 0,
251 "Enable backwards compatibility with IGMPv1");
252 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, v2enable,
253 CTLFLAG_RW, igmp_v2enable, 0,
254 "Enable backwards compatibility with IGMPv2");
255 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, legacysupp,
256 CTLFLAG_RW, igmp_legacysupp, 0,
257 "Allow v1/v2 reports to suppress v3 group responses");
258 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, default_version,
259 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, igmp_default_version, 0,
260 sysctl_igmp_default_version, "I",
261 "Default version of IGMP to run on each interface");
262 SYSCTL_V_PROC(V_NET, vnet_inet, _net_inet_igmp, OID_AUTO, gsrdelay,
263 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, igmp_gsrdelay.tv_sec, 0,
264 sysctl_igmp_gsr, "I",
265 "Rate limit for IGMPv3 Group-and-Source queries in seconds");
268 * Non-virtualized sysctls.
270 SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE,
271 sysctl_igmp_ifinfo, "Per-interface IGMPv3 state");
274 igmp_save_context(struct mbuf *m, struct ifnet *ifp)
278 m->m_pkthdr.header = ifp->if_vnet;
280 m->m_pkthdr.flowid = ifp->if_index;
284 igmp_scrub_context(struct mbuf *m)
287 m->m_pkthdr.header = NULL;
288 m->m_pkthdr.flowid = 0;
292 static __inline char *
293 inet_ntoa_haddr(in_addr_t haddr)
297 ia.s_addr = htonl(haddr);
298 return (inet_ntoa(ia));
303 * Restore context from a queued IGMP output chain.
304 * Return saved ifindex.
306 * VIMAGE: The assertion is there to make sure that we
307 * actually called CURVNET_SET() with what's in the mbuf chain.
309 static __inline uint32_t
310 igmp_restore_context(struct mbuf *m)
314 #if defined(VIMAGE) && defined(INVARIANTS)
315 KASSERT(curvnet == (m->m_pkthdr.header),
316 ("%s: called when curvnet was not restored", __func__));
319 return (m->m_pkthdr.flowid);
323 * Retrieve or set default IGMP version.
325 * VIMAGE: Assume curvnet set by caller.
326 * SMPng: NOTE: Serialized by IGMP lock.
329 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS)
331 INIT_VNET_INET(curvnet);
335 error = sysctl_wire_old_buffer(req, sizeof(int));
341 new = V_igmp_default_version;
343 error = sysctl_handle_int(oidp, &new, 0, req);
344 if (error || !req->newptr)
347 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) {
352 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d",
353 V_igmp_default_version, new);
355 V_igmp_default_version = new;
363 * Retrieve or set threshold between group-source queries in seconds.
365 * VIMAGE: Assume curvnet set by caller.
366 * SMPng: NOTE: Serialized by IGMP lock.
369 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS)
371 INIT_VNET_INET(curvnet);
375 error = sysctl_wire_old_buffer(req, sizeof(int));
381 i = V_igmp_gsrdelay.tv_sec;
383 error = sysctl_handle_int(oidp, &i, 0, req);
384 if (error || !req->newptr)
387 if (i < -1 || i >= 60) {
392 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d",
393 V_igmp_gsrdelay.tv_sec, i);
394 V_igmp_gsrdelay.tv_sec = i;
402 * Expose struct igmp_ifinfo to userland, keyed by ifindex.
403 * For use by ifmcstat(8).
405 * SMPng: NOTE: Does an unlocked ifindex space read.
406 * VIMAGE: Assume curvnet set by caller. The node handler itself
407 * is not directly virtualized.
410 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS)
412 INIT_VNET_NET(curvnet);
413 INIT_VNET_INET(curvnet);
418 struct igmp_ifinfo *igi;
423 if (req->newptr != NULL)
429 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo));
436 if (name[0] <= 0 || name[0] > V_if_index) {
443 ifp = ifnet_byindex(name[0]);
447 LIST_FOREACH(igi, &V_igi_head, igi_link) {
448 if (ifp == igi->igi_ifp) {
449 error = SYSCTL_OUT(req, igi,
450 sizeof(struct igmp_ifinfo));
462 * Dispatch an entire queue of pending packet chains
464 * VIMAGE: Assumes the vnet pointer has been set.
467 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop)
475 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m);
477 m->m_flags |= M_IGMP_LOOP;
478 netisr_dispatch(NETISR_IGMP, m);
485 * Filter outgoing IGMP report state by group.
487 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1).
488 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are
489 * disabled for all groups in the 224.0.0.0/24 link-local scope. However,
490 * this may break certain IGMP snooping switches which rely on the old
493 * Return zero if the given group is one for which IGMP reports
494 * should be suppressed, or non-zero if reports should be issued.
497 igmp_isgroupreported(const struct in_addr addr)
499 INIT_VNET_INET(curvnet);
501 if (in_allhosts(addr) ||
502 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr)))))
509 * Construct a Router Alert option to use in outgoing packets.
517 MGET(m, M_DONTWAIT, MT_DATA);
518 p = mtod(m, struct ipoption *);
519 p->ipopt_dst.s_addr = INADDR_ANY;
520 p->ipopt_list[0] = IPOPT_RA; /* Router Alert Option */
521 p->ipopt_list[1] = 0x04; /* 4 bytes long */
522 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */
523 p->ipopt_list[3] = 0x00; /* pad byte */
524 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1];
530 * Attach IGMP when PF_INET is attached to an interface.
533 igmp_domifattach(struct ifnet *ifp)
535 struct igmp_ifinfo *igi;
537 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
538 __func__, ifp, ifp->if_xname);
542 igi = igi_alloc_locked(ifp);
543 if (!(ifp->if_flags & IFF_MULTICAST))
544 igi->igi_flags |= IGIF_SILENT;
552 * VIMAGE: assume curvnet set by caller.
554 static struct igmp_ifinfo *
555 igi_alloc_locked(/*const*/ struct ifnet *ifp)
557 INIT_VNET_INET(ifp->if_vnet);
558 struct igmp_ifinfo *igi;
562 igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO);
567 igi->igi_version = V_igmp_default_version;
569 igi->igi_rv = IGMP_RV_INIT;
570 igi->igi_qi = IGMP_QI_INIT;
571 igi->igi_qri = IGMP_QRI_INIT;
572 igi->igi_uri = IGMP_URI_INIT;
574 SLIST_INIT(&igi->igi_relinmhead);
577 * Responses to general queries are subject to bounds.
579 IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS);
581 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link);
583 CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)",
593 * NOTE: Some finalization tasks need to run before the protocol domain
594 * is detached, but also before the link layer does its cleanup.
596 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK().
597 * XXX This is also bitten by unlocked ifma_protospec access.
600 igmp_ifdetach(struct ifnet *ifp)
602 struct igmp_ifinfo *igi;
603 struct ifmultiaddr *ifma;
604 struct in_multi *inm, *tinm;
606 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp,
611 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
612 if (igi->igi_version == IGMP_VERSION_3) {
614 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
615 if (ifma->ifma_addr->sa_family != AF_INET ||
616 ifma->ifma_protospec == NULL)
619 KASSERT(ifma->ifma_protospec != NULL,
620 ("%s: ifma_protospec is NULL", __func__));
622 inm = (struct in_multi *)ifma->ifma_protospec;
623 if (inm->inm_state == IGMP_LEAVING_MEMBER) {
624 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
627 inm_clear_recorded(inm);
631 * Free the in_multi reference(s) for this IGMP lifecycle.
633 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele,
635 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele);
636 inm_release_locked(inm);
644 * Hook for domifdetach.
647 igmp_domifdetach(struct ifnet *ifp)
649 struct igmp_ifinfo *igi;
651 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)",
652 __func__, ifp, ifp->if_xname);
656 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
657 igi_delete_locked(ifp);
663 igi_delete_locked(const struct ifnet *ifp)
665 INIT_VNET_INET(ifp->if_vnet);
666 struct igmp_ifinfo *igi, *tigi;
668 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)",
669 __func__, ifp, ifp->if_xname);
673 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) {
674 if (igi->igi_ifp == ifp) {
676 * Free deferred General Query responses.
678 _IF_DRAIN(&igi->igi_gq);
680 LIST_REMOVE(igi, igi_link);
682 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead),
683 ("%s: there are dangling in_multi references",
692 panic("%s: igmp_ifinfo not found for ifp %p\n", __func__, ifp);
697 * Process a received IGMPv1 query.
698 * Return non-zero if the message should be dropped.
700 * VIMAGE: The curvnet pointer is derived from the input ifp.
703 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip,
704 const struct igmp *igmp)
706 INIT_VNET_INET(ifp->if_vnet);
707 struct ifmultiaddr *ifma;
708 struct igmp_ifinfo *igi;
709 struct in_multi *inm;
712 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to
713 * 224.0.0.1. They are always treated as General Queries.
714 * igmp_group is always ignored. Do not drop it as a userland
715 * daemon may wish to see it.
716 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
718 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) {
719 IGMPSTAT_INC(igps_rcv_badqueries);
722 IGMPSTAT_INC(igps_rcv_gen_queries);
727 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
728 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
730 if (igi->igi_flags & IGIF_LOOPBACK) {
731 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)",
737 * Switch to IGMPv1 host compatibility mode.
739 igmp_set_version(igi, IGMP_VERSION_1);
741 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname);
744 * Start the timers in all of our group records
745 * for the interface on which the query arrived,
746 * except those which are already running.
749 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
750 if (ifma->ifma_addr->sa_family != AF_INET ||
751 ifma->ifma_protospec == NULL)
753 inm = (struct in_multi *)ifma->ifma_protospec;
754 if (inm->inm_timer != 0)
756 switch (inm->inm_state) {
757 case IGMP_NOT_MEMBER:
758 case IGMP_SILENT_MEMBER:
760 case IGMP_G_QUERY_PENDING_MEMBER:
761 case IGMP_SG_QUERY_PENDING_MEMBER:
762 case IGMP_REPORTING_MEMBER:
763 case IGMP_IDLE_MEMBER:
764 case IGMP_LAZY_MEMBER:
765 case IGMP_SLEEPING_MEMBER:
766 case IGMP_AWAKENING_MEMBER:
767 inm->inm_state = IGMP_REPORTING_MEMBER;
768 inm->inm_timer = IGMP_RANDOM_DELAY(
769 IGMP_V1V2_MAX_RI * PR_FASTHZ);
770 V_current_state_timers_running = 1;
772 case IGMP_LEAVING_MEMBER:
786 * Process a received IGMPv2 general or group-specific query.
789 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip,
790 const struct igmp *igmp)
792 INIT_VNET_INET(ifp->if_vnet);
793 struct ifmultiaddr *ifma;
794 struct igmp_ifinfo *igi;
795 struct in_multi *inm;
796 int is_general_query;
799 is_general_query = 0;
802 * Validate address fields upfront.
803 * XXX SMPng: unlocked increments in igmpstat assumed atomic.
805 if (in_nullhost(igmp->igmp_group)) {
807 * IGMPv2 General Query.
808 * If this was not sent to the all-hosts group, ignore it.
810 if (!in_allhosts(ip->ip_dst))
812 IGMPSTAT_INC(igps_rcv_gen_queries);
813 is_general_query = 1;
815 /* IGMPv2 Group-Specific Query. */
816 IGMPSTAT_INC(igps_rcv_group_queries);
822 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
823 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
825 if (igi->igi_flags & IGIF_LOOPBACK) {
826 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)",
832 * Ignore v2 query if in v1 Compatibility Mode.
834 if (igi->igi_version == IGMP_VERSION_1)
837 igmp_set_version(igi, IGMP_VERSION_2);
839 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE;
843 if (is_general_query) {
845 * For each reporting group joined on this
846 * interface, kick the report timer.
848 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)",
851 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
852 if (ifma->ifma_addr->sa_family != AF_INET ||
853 ifma->ifma_protospec == NULL)
855 inm = (struct in_multi *)ifma->ifma_protospec;
856 igmp_v2_update_group(inm, timer);
861 * Group-specific IGMPv2 query, we need only
862 * look up the single group to process it.
864 inm = inm_lookup(ifp, igmp->igmp_group);
866 CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)",
867 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
868 igmp_v2_update_group(inm, timer);
880 * Update the report timer on a group in response to an IGMPv2 query.
882 * If we are becoming the reporting member for this group, start the timer.
883 * If we already are the reporting member for this group, and timer is
884 * below the threshold, reset it.
886 * We may be updating the group for the first time since we switched
887 * to IGMPv3. If we are, then we must clear any recorded source lists,
888 * and transition to REPORTING state; the group timer is overloaded
889 * for group and group-source query responses.
891 * Unlike IGMPv3, the delay per group should be jittered
892 * to avoid bursts of IGMPv2 reports.
895 igmp_v2_update_group(struct in_multi *inm, const int timer)
897 INIT_VNET_INET(curvnet);
899 CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__,
900 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer);
902 IN_MULTI_LOCK_ASSERT();
904 switch (inm->inm_state) {
905 case IGMP_NOT_MEMBER:
906 case IGMP_SILENT_MEMBER:
908 case IGMP_REPORTING_MEMBER:
909 if (inm->inm_timer != 0 &&
910 inm->inm_timer <= timer) {
911 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, "
912 "skipping.", __func__);
916 case IGMP_SG_QUERY_PENDING_MEMBER:
917 case IGMP_G_QUERY_PENDING_MEMBER:
918 case IGMP_IDLE_MEMBER:
919 case IGMP_LAZY_MEMBER:
920 case IGMP_AWAKENING_MEMBER:
921 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__);
922 inm->inm_state = IGMP_REPORTING_MEMBER;
923 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
924 V_current_state_timers_running = 1;
926 case IGMP_SLEEPING_MEMBER:
927 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__);
928 inm->inm_state = IGMP_AWAKENING_MEMBER;
930 case IGMP_LEAVING_MEMBER:
936 * Process a received IGMPv3 general, group-specific or
937 * group-and-source-specific query.
938 * Assumes m has already been pulled up to the full IGMP message length.
939 * Return 0 if successful, otherwise an appropriate error code is returned.
942 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip,
943 /*const*/ struct igmpv3 *igmpv3)
945 INIT_VNET_INET(ifp->if_vnet);
946 struct igmp_ifinfo *igi;
947 struct in_multi *inm;
948 int is_general_query;
949 uint32_t maxresp, nsrc, qqi;
953 is_general_query = 0;
955 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname);
957 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */
958 if (maxresp >= 128) {
959 maxresp = IGMP_MANT(igmpv3->igmp_code) <<
960 (IGMP_EXP(igmpv3->igmp_code) + 3);
964 * Robustness must never be less than 2 for on-wire IGMPv3.
965 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make
966 * an exception for interfaces whose IGMPv3 state changes
967 * are redirected to loopback (e.g. MANET).
969 qrv = IGMP_QRV(igmpv3->igmp_misc);
971 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__,
976 qqi = igmpv3->igmp_qqi;
978 qqi = IGMP_MANT(igmpv3->igmp_qqi) <<
979 (IGMP_EXP(igmpv3->igmp_qqi) + 3);
982 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE;
986 nsrc = ntohs(igmpv3->igmp_numsrc);
989 * Validate address fields and versions upfront before
990 * accepting v3 query.
991 * XXX SMPng: Unlocked access to igmpstat counters here.
993 if (in_nullhost(igmpv3->igmp_group)) {
995 * IGMPv3 General Query.
997 * General Queries SHOULD be directed to 224.0.0.1.
998 * A general query with a source list has undefined
999 * behaviour; discard it.
1001 IGMPSTAT_INC(igps_rcv_gen_queries);
1002 if (!in_allhosts(ip->ip_dst) || nsrc > 0) {
1003 IGMPSTAT_INC(igps_rcv_badqueries);
1006 is_general_query = 1;
1008 /* Group or group-source specific query. */
1010 IGMPSTAT_INC(igps_rcv_group_queries);
1012 IGMPSTAT_INC(igps_rcv_gsr_queries);
1018 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
1019 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
1021 if (igi->igi_flags & IGIF_LOOPBACK) {
1022 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)",
1023 ifp, ifp->if_xname);
1028 * Discard the v3 query if we're in Compatibility Mode.
1029 * The RFC is not obviously worded that hosts need to stay in
1030 * compatibility mode until the Old Version Querier Present
1033 if (igi->igi_version != IGMP_VERSION_3) {
1034 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)",
1035 igi->igi_version, ifp, ifp->if_xname);
1039 igmp_set_version(igi, IGMP_VERSION_3);
1042 igi->igi_qri = maxresp;
1044 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi,
1047 if (is_general_query) {
1049 * Schedule a current-state report on this ifp for
1050 * all groups, possibly containing source lists.
1051 * If there is a pending General Query response
1052 * scheduled earlier than the selected delay, do
1053 * not schedule any other reports.
1054 * Otherwise, reset the interface timer.
1056 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)",
1057 ifp, ifp->if_xname);
1058 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) {
1059 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer);
1060 V_interface_timers_running = 1;
1064 * Group-source-specific queries are throttled on
1065 * a per-group basis to defeat denial-of-service attempts.
1066 * Queries for groups we are not a member of on this
1067 * link are simply ignored.
1069 inm = inm_lookup(ifp, igmpv3->igmp_group);
1073 if (!ratecheck(&inm->inm_lastgsrtv,
1074 &V_igmp_gsrdelay)) {
1075 CTR1(KTR_IGMPV3, "%s: GS query throttled.",
1077 IGMPSTAT_INC(igps_drop_gsr_queries);
1081 CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)",
1082 inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname);
1084 * If there is a pending General Query response
1085 * scheduled sooner than the selected delay, no
1086 * further report need be scheduled.
1087 * Otherwise, prepare to respond to the
1088 * group-specific or group-and-source query.
1090 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer)
1091 igmp_input_v3_group_query(inm, igi, timer, igmpv3);
1102 * Process a recieved IGMPv3 group-specific or group-and-source-specific
1104 * Return <0 if any error occured. Currently this is ignored.
1107 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi,
1108 int timer, /*const*/ struct igmpv3 *igmpv3)
1110 INIT_VNET_INET(curvnet);
1114 IN_MULTI_LOCK_ASSERT();
1119 switch (inm->inm_state) {
1120 case IGMP_NOT_MEMBER:
1121 case IGMP_SILENT_MEMBER:
1122 case IGMP_SLEEPING_MEMBER:
1123 case IGMP_LAZY_MEMBER:
1124 case IGMP_AWAKENING_MEMBER:
1125 case IGMP_IDLE_MEMBER:
1126 case IGMP_LEAVING_MEMBER:
1129 case IGMP_REPORTING_MEMBER:
1130 case IGMP_G_QUERY_PENDING_MEMBER:
1131 case IGMP_SG_QUERY_PENDING_MEMBER:
1135 nsrc = ntohs(igmpv3->igmp_numsrc);
1138 * Deal with group-specific queries upfront.
1139 * If any group query is already pending, purge any recorded
1140 * source-list state if it exists, and schedule a query response
1141 * for this group-specific query.
1144 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
1145 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) {
1146 inm_clear_recorded(inm);
1147 timer = min(inm->inm_timer, timer);
1149 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER;
1150 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1151 V_current_state_timers_running = 1;
1156 * Deal with the case where a group-and-source-specific query has
1157 * been received but a group-specific query is already pending.
1159 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) {
1160 timer = min(inm->inm_timer, timer);
1161 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1162 V_current_state_timers_running = 1;
1167 * Finally, deal with the case where a group-and-source-specific
1168 * query has been received, where a response to a previous g-s-r
1169 * query exists, or none exists.
1170 * In this case, we need to parse the source-list which the Querier
1171 * has provided us with and check if we have any source list filter
1172 * entries at T1 for these sources. If we do not, there is no need
1173 * schedule a report and the query may be dropped.
1174 * If we do, we must record them and schedule a current-state
1175 * report for those sources.
1176 * FIXME: Handling source lists larger than 1 mbuf requires that
1177 * we pass the mbuf chain pointer down to this function, and use
1178 * m_getptr() to walk the chain.
1180 if (inm->inm_nsrc > 0) {
1181 const struct in_addr *ap;
1184 ap = (const struct in_addr *)(igmpv3 + 1);
1186 for (i = 0; i < nsrc; i++, ap++) {
1187 retval = inm_record_source(inm, ap->s_addr);
1190 nrecorded += retval;
1192 if (nrecorded > 0) {
1194 "%s: schedule response to SG query", __func__);
1195 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER;
1196 inm->inm_timer = IGMP_RANDOM_DELAY(timer);
1197 V_current_state_timers_running = 1;
1205 * Process a received IGMPv1 host membership report.
1207 * NOTE: 0.0.0.0 workaround breaks const correctness.
1210 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1211 /*const*/ struct igmp *igmp)
1213 INIT_VNET_INET(ifp->if_vnet);
1214 struct in_ifaddr *ia;
1215 struct in_multi *inm;
1217 IGMPSTAT_INC(igps_rcv_reports);
1219 if (ifp->if_flags & IFF_LOOPBACK)
1222 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr) ||
1223 !in_hosteq(igmp->igmp_group, ip->ip_dst))) {
1224 IGMPSTAT_INC(igps_rcv_badreports);
1229 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1230 * Booting clients may use the source address 0.0.0.0. Some
1231 * IGMP daemons may not know how to use IP_RECVIF to determine
1232 * the interface upon which this message was received.
1233 * Replace 0.0.0.0 with the subnet address if told to do so.
1235 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1238 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1241 CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)",
1242 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1245 * IGMPv1 report suppression.
1246 * If we are a member of this group, and our membership should be
1247 * reported, stop our group timer and transition to the 'lazy' state.
1250 inm = inm_lookup(ifp, igmp->igmp_group);
1252 struct igmp_ifinfo *igi;
1256 KASSERT(igi != NULL,
1257 ("%s: no igi for ifp %p", __func__, ifp));
1261 IGMPSTAT_INC(igps_rcv_ourreports);
1264 * If we are in IGMPv3 host mode, do not allow the
1265 * other host's IGMPv1 report to suppress our reports
1266 * unless explicitly configured to do so.
1268 if (igi->igi_version == IGMP_VERSION_3) {
1269 if (V_igmp_legacysupp)
1270 igmp_v3_suppress_group_record(inm);
1276 switch (inm->inm_state) {
1277 case IGMP_NOT_MEMBER:
1278 case IGMP_SILENT_MEMBER:
1280 case IGMP_IDLE_MEMBER:
1281 case IGMP_LAZY_MEMBER:
1282 case IGMP_AWAKENING_MEMBER:
1284 "report suppressed for %s on ifp %p(%s)",
1285 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1286 case IGMP_SLEEPING_MEMBER:
1287 inm->inm_state = IGMP_SLEEPING_MEMBER;
1289 case IGMP_REPORTING_MEMBER:
1291 "report suppressed for %s on ifp %p(%s)",
1292 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1293 if (igi->igi_version == IGMP_VERSION_1)
1294 inm->inm_state = IGMP_LAZY_MEMBER;
1295 else if (igi->igi_version == IGMP_VERSION_2)
1296 inm->inm_state = IGMP_SLEEPING_MEMBER;
1298 case IGMP_G_QUERY_PENDING_MEMBER:
1299 case IGMP_SG_QUERY_PENDING_MEMBER:
1300 case IGMP_LEAVING_MEMBER:
1312 * Process a received IGMPv2 host membership report.
1314 * NOTE: 0.0.0.0 workaround breaks const correctness.
1317 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip,
1318 /*const*/ struct igmp *igmp)
1320 INIT_VNET_INET(ifp->if_vnet);
1321 struct in_ifaddr *ia;
1322 struct in_multi *inm;
1325 * Make sure we don't hear our own membership report. Fast
1326 * leave requires knowing that we are the only member of a
1330 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr))
1333 IGMPSTAT_INC(igps_rcv_reports);
1335 if (ifp->if_flags & IFF_LOOPBACK)
1338 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) ||
1339 !in_hosteq(igmp->igmp_group, ip->ip_dst)) {
1340 IGMPSTAT_INC(igps_rcv_badreports);
1345 * RFC 3376, Section 4.2.13, 9.2, 9.3:
1346 * Booting clients may use the source address 0.0.0.0. Some
1347 * IGMP daemons may not know how to use IP_RECVIF to determine
1348 * the interface upon which this message was received.
1349 * Replace 0.0.0.0 with the subnet address if told to do so.
1351 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) {
1353 ip->ip_src.s_addr = htonl(ia->ia_subnet);
1356 CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)",
1357 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1360 * IGMPv2 report suppression.
1361 * If we are a member of this group, and our membership should be
1362 * reported, and our group timer is pending or about to be reset,
1363 * stop our group timer by transitioning to the 'lazy' state.
1366 inm = inm_lookup(ifp, igmp->igmp_group);
1368 struct igmp_ifinfo *igi;
1371 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp));
1373 IGMPSTAT_INC(igps_rcv_ourreports);
1376 * If we are in IGMPv3 host mode, do not allow the
1377 * other host's IGMPv1 report to suppress our reports
1378 * unless explicitly configured to do so.
1380 if (igi->igi_version == IGMP_VERSION_3) {
1381 if (V_igmp_legacysupp)
1382 igmp_v3_suppress_group_record(inm);
1388 switch (inm->inm_state) {
1389 case IGMP_NOT_MEMBER:
1390 case IGMP_SILENT_MEMBER:
1391 case IGMP_SLEEPING_MEMBER:
1393 case IGMP_REPORTING_MEMBER:
1394 case IGMP_IDLE_MEMBER:
1395 case IGMP_AWAKENING_MEMBER:
1397 "report suppressed for %s on ifp %p(%s)",
1398 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname);
1399 case IGMP_LAZY_MEMBER:
1400 inm->inm_state = IGMP_LAZY_MEMBER;
1402 case IGMP_G_QUERY_PENDING_MEMBER:
1403 case IGMP_SG_QUERY_PENDING_MEMBER:
1404 case IGMP_LEAVING_MEMBER:
1416 igmp_input(struct mbuf *m, int off)
1426 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1428 ifp = m->m_pkthdr.rcvif;
1429 INIT_VNET_INET(ifp->if_vnet);
1431 IGMPSTAT_INC(igps_rcv_total);
1433 ip = mtod(m, struct ip *);
1435 igmplen = ip->ip_len;
1440 if (igmplen < IGMP_MINLEN) {
1441 IGMPSTAT_INC(igps_rcv_tooshort);
1447 * Always pullup to the minimum size for v1/v2 or v3
1448 * to amortize calls to m_pullup().
1451 if (igmplen >= IGMP_V3_QUERY_MINLEN)
1452 minlen += IGMP_V3_QUERY_MINLEN;
1454 minlen += IGMP_MINLEN;
1455 if ((m->m_flags & M_EXT || m->m_len < minlen) &&
1456 (m = m_pullup(m, minlen)) == 0) {
1457 IGMPSTAT_INC(igps_rcv_tooshort);
1460 ip = mtod(m, struct ip *);
1462 if (ip->ip_ttl != 1) {
1463 IGMPSTAT_INC(igps_rcv_badttl);
1469 * Validate checksum.
1471 m->m_data += iphlen;
1473 igmp = mtod(m, struct igmp *);
1474 if (in_cksum(m, igmplen)) {
1475 IGMPSTAT_INC(igps_rcv_badsum);
1479 m->m_data -= iphlen;
1482 switch (igmp->igmp_type) {
1483 case IGMP_HOST_MEMBERSHIP_QUERY:
1484 if (igmplen == IGMP_MINLEN) {
1485 if (igmp->igmp_code == 0)
1486 queryver = IGMP_VERSION_1;
1488 queryver = IGMP_VERSION_2;
1489 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) {
1490 queryver = IGMP_VERSION_3;
1492 IGMPSTAT_INC(igps_rcv_tooshort);
1498 case IGMP_VERSION_1:
1499 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1500 if (!V_igmp_v1enable)
1502 if (igmp_input_v1_query(ifp, ip, igmp) != 0) {
1508 case IGMP_VERSION_2:
1509 IGMPSTAT_INC(igps_rcv_v1v2_queries);
1510 if (!V_igmp_v2enable)
1512 if (igmp_input_v2_query(ifp, ip, igmp) != 0) {
1518 case IGMP_VERSION_3: {
1519 struct igmpv3 *igmpv3;
1524 IGMPSTAT_INC(igps_rcv_v3_queries);
1525 igmpv3 = (struct igmpv3 *)igmp;
1527 * Validate length based on source count.
1529 nsrc = ntohs(igmpv3->igmp_numsrc);
1530 srclen = sizeof(struct in_addr) * nsrc;
1531 if (nsrc * sizeof(in_addr_t) > srclen) {
1532 IGMPSTAT_INC(igps_rcv_tooshort);
1536 * m_pullup() may modify m, so pullup in
1539 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN +
1541 if ((m->m_flags & M_EXT ||
1542 m->m_len < igmpv3len) &&
1543 (m = m_pullup(m, igmpv3len)) == NULL) {
1544 IGMPSTAT_INC(igps_rcv_tooshort);
1547 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *)
1549 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) {
1558 case IGMP_v1_HOST_MEMBERSHIP_REPORT:
1559 if (!V_igmp_v1enable)
1561 if (igmp_input_v1_report(ifp, ip, igmp) != 0) {
1567 case IGMP_v2_HOST_MEMBERSHIP_REPORT:
1568 if (!V_igmp_v2enable)
1570 if (!ip_checkrouteralert(m))
1571 IGMPSTAT_INC(igps_rcv_nora);
1572 if (igmp_input_v2_report(ifp, ip, igmp) != 0) {
1578 case IGMP_v3_HOST_MEMBERSHIP_REPORT:
1580 * Hosts do not need to process IGMPv3 membership reports,
1581 * as report suppression is no longer required.
1583 if (!ip_checkrouteralert(m))
1584 IGMPSTAT_INC(igps_rcv_nora);
1592 * Pass all valid IGMP packets up to any process(es) listening on a
1600 * Fast timeout handler (global).
1601 * VIMAGE: Timeout handlers are expected to service all vimages.
1606 VNET_ITERATOR_DECL(vnet_iter);
1609 VNET_FOREACH(vnet_iter) {
1610 CURVNET_SET(vnet_iter);
1611 igmp_fasttimo_vnet();
1614 VNET_LIST_RUNLOCK();
1618 * Fast timeout handler (per-vnet).
1619 * Sends are shuffled off to a netisr to deal with Giant.
1621 * VIMAGE: Assume caller has set up our curvnet.
1624 igmp_fasttimo_vnet(void)
1626 INIT_VNET_INET(curvnet);
1627 struct ifqueue scq; /* State-change packets */
1628 struct ifqueue qrq; /* Query response packets */
1630 struct igmp_ifinfo *igi;
1631 struct ifmultiaddr *ifma, *tifma;
1632 struct in_multi *inm;
1633 int loop, uri_fasthz;
1639 * Quick check to see if any work needs to be done, in order to
1640 * minimize the overhead of fasttimo processing.
1641 * SMPng: XXX Unlocked reads.
1643 if (!V_current_state_timers_running &&
1644 !V_interface_timers_running &&
1645 !V_state_change_timers_running)
1652 * IGMPv3 General Query response timer processing.
1654 if (V_interface_timers_running) {
1655 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__);
1657 V_interface_timers_running = 0;
1658 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1659 if (igi->igi_v3_timer == 0) {
1661 } else if (--igi->igi_v3_timer == 0) {
1662 igmp_v3_dispatch_general_query(igi);
1664 V_interface_timers_running = 1;
1669 if (!V_current_state_timers_running &&
1670 !V_state_change_timers_running)
1673 V_current_state_timers_running = 0;
1674 V_state_change_timers_running = 0;
1676 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__);
1679 * IGMPv1/v2/v3 host report and state-change timer processing.
1680 * Note: Processing a v3 group timer may remove a node.
1682 LIST_FOREACH(igi, &V_igi_head, igi_link) {
1685 if (igi->igi_version == IGMP_VERSION_3) {
1686 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
1687 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri *
1690 memset(&qrq, 0, sizeof(struct ifqueue));
1691 IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS);
1693 memset(&scq, 0, sizeof(struct ifqueue));
1694 IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS);
1698 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link,
1700 if (ifma->ifma_addr->sa_family != AF_INET ||
1701 ifma->ifma_protospec == NULL)
1703 inm = (struct in_multi *)ifma->ifma_protospec;
1704 switch (igi->igi_version) {
1705 case IGMP_VERSION_1:
1706 case IGMP_VERSION_2:
1707 igmp_v1v2_process_group_timer(inm,
1710 case IGMP_VERSION_3:
1711 igmp_v3_process_group_timers(igi, &qrq,
1712 &scq, inm, uri_fasthz);
1716 IF_ADDR_UNLOCK(ifp);
1718 if (igi->igi_version == IGMP_VERSION_3) {
1719 struct in_multi *tinm;
1721 igmp_dispatch_queue(&qrq, 0, loop);
1722 igmp_dispatch_queue(&scq, 0, loop);
1725 * Free the in_multi reference(s) for this
1728 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead,
1730 SLIST_REMOVE_HEAD(&igi->igi_relinmhead,
1732 inm_release_locked(inm);
1743 * Update host report group timer for IGMPv1/v2.
1744 * Will update the global pending timer flags.
1747 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version)
1749 INIT_VNET_INET(curvnet);
1750 int report_timer_expired;
1752 IN_MULTI_LOCK_ASSERT();
1755 if (inm->inm_timer == 0) {
1756 report_timer_expired = 0;
1757 } else if (--inm->inm_timer == 0) {
1758 report_timer_expired = 1;
1760 V_current_state_timers_running = 1;
1764 switch (inm->inm_state) {
1765 case IGMP_NOT_MEMBER:
1766 case IGMP_SILENT_MEMBER:
1767 case IGMP_IDLE_MEMBER:
1768 case IGMP_LAZY_MEMBER:
1769 case IGMP_SLEEPING_MEMBER:
1770 case IGMP_AWAKENING_MEMBER:
1772 case IGMP_REPORTING_MEMBER:
1773 if (report_timer_expired) {
1774 inm->inm_state = IGMP_IDLE_MEMBER;
1775 (void)igmp_v1v2_queue_report(inm,
1776 (version == IGMP_VERSION_2) ?
1777 IGMP_v2_HOST_MEMBERSHIP_REPORT :
1778 IGMP_v1_HOST_MEMBERSHIP_REPORT);
1781 case IGMP_G_QUERY_PENDING_MEMBER:
1782 case IGMP_SG_QUERY_PENDING_MEMBER:
1783 case IGMP_LEAVING_MEMBER:
1789 * Update a group's timers for IGMPv3.
1790 * Will update the global pending timer flags.
1791 * Note: Unlocked read from igi.
1794 igmp_v3_process_group_timers(struct igmp_ifinfo *igi,
1795 struct ifqueue *qrq, struct ifqueue *scq,
1796 struct in_multi *inm, const int uri_fasthz)
1798 INIT_VNET_INET(curvnet);
1799 int query_response_timer_expired;
1800 int state_change_retransmit_timer_expired;
1802 IN_MULTI_LOCK_ASSERT();
1805 query_response_timer_expired = 0;
1806 state_change_retransmit_timer_expired = 0;
1809 * During a transition from v1/v2 compatibility mode back to v3,
1810 * a group record in REPORTING state may still have its group
1811 * timer active. This is a no-op in this function; it is easier
1812 * to deal with it here than to complicate the slow-timeout path.
1814 if (inm->inm_timer == 0) {
1815 query_response_timer_expired = 0;
1816 } else if (--inm->inm_timer == 0) {
1817 query_response_timer_expired = 1;
1819 V_current_state_timers_running = 1;
1822 if (inm->inm_sctimer == 0) {
1823 state_change_retransmit_timer_expired = 0;
1824 } else if (--inm->inm_sctimer == 0) {
1825 state_change_retransmit_timer_expired = 1;
1827 V_state_change_timers_running = 1;
1830 /* We are in fasttimo, so be quick about it. */
1831 if (!state_change_retransmit_timer_expired &&
1832 !query_response_timer_expired)
1835 switch (inm->inm_state) {
1836 case IGMP_NOT_MEMBER:
1837 case IGMP_SILENT_MEMBER:
1838 case IGMP_SLEEPING_MEMBER:
1839 case IGMP_LAZY_MEMBER:
1840 case IGMP_AWAKENING_MEMBER:
1841 case IGMP_IDLE_MEMBER:
1843 case IGMP_G_QUERY_PENDING_MEMBER:
1844 case IGMP_SG_QUERY_PENDING_MEMBER:
1846 * Respond to a previously pending Group-Specific
1847 * or Group-and-Source-Specific query by enqueueing
1848 * the appropriate Current-State report for
1849 * immediate transmission.
1851 if (query_response_timer_expired) {
1854 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1,
1855 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER));
1856 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
1858 inm->inm_state = IGMP_REPORTING_MEMBER;
1859 /* XXX Clear recorded sources for next time. */
1860 inm_clear_recorded(inm);
1863 case IGMP_REPORTING_MEMBER:
1864 case IGMP_LEAVING_MEMBER:
1865 if (state_change_retransmit_timer_expired) {
1867 * State-change retransmission timer fired.
1868 * If there are any further pending retransmissions,
1869 * set the global pending state-change flag, and
1872 if (--inm->inm_scrv > 0) {
1873 inm->inm_sctimer = uri_fasthz;
1874 V_state_change_timers_running = 1;
1877 * Retransmit the previously computed state-change
1878 * report. If there are no further pending
1879 * retransmissions, the mbuf queue will be consumed.
1880 * Update T0 state to T1 as we have now sent
1883 (void)igmp_v3_merge_state_changes(inm, scq);
1886 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
1887 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
1890 * If we are leaving the group for good, make sure
1891 * we release IGMP's reference to it.
1892 * This release must be deferred using a SLIST,
1893 * as we are called from a loop which traverses
1894 * the in_ifmultiaddr TAILQ.
1896 if (inm->inm_state == IGMP_LEAVING_MEMBER &&
1897 inm->inm_scrv == 0) {
1898 inm->inm_state = IGMP_NOT_MEMBER;
1899 SLIST_INSERT_HEAD(&igi->igi_relinmhead,
1909 * Suppress a group's pending response to a group or source/group query.
1911 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency.
1912 * Do NOT update ST1/ST0 as this operation merely suppresses
1913 * the currently pending group record.
1914 * Do NOT suppress the response to a general query. It is possible but
1915 * it would require adding another state or flag.
1918 igmp_v3_suppress_group_record(struct in_multi *inm)
1921 IN_MULTI_LOCK_ASSERT();
1923 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3,
1924 ("%s: not IGMPv3 mode on link", __func__));
1926 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER ||
1927 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER)
1930 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
1931 inm_clear_recorded(inm);
1934 inm->inm_state = IGMP_REPORTING_MEMBER;
1938 * Switch to a different IGMP version on the given interface,
1939 * as per Section 7.2.1.
1942 igmp_set_version(struct igmp_ifinfo *igi, const int version)
1944 int old_version_timer;
1948 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__,
1949 version, igi->igi_ifp, igi->igi_ifp->if_xname);
1951 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) {
1953 * Compute the "Older Version Querier Present" timer as per
1956 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri;
1957 old_version_timer *= PR_SLOWHZ;
1959 if (version == IGMP_VERSION_1) {
1960 igi->igi_v1_timer = old_version_timer;
1961 igi->igi_v2_timer = 0;
1962 } else if (version == IGMP_VERSION_2) {
1963 igi->igi_v1_timer = 0;
1964 igi->igi_v2_timer = old_version_timer;
1968 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
1969 if (igi->igi_version != IGMP_VERSION_2) {
1970 igi->igi_version = IGMP_VERSION_2;
1971 igmp_v3_cancel_link_timers(igi);
1973 } else if (igi->igi_v1_timer > 0) {
1974 if (igi->igi_version != IGMP_VERSION_1) {
1975 igi->igi_version = IGMP_VERSION_1;
1976 igmp_v3_cancel_link_timers(igi);
1982 * Cancel pending IGMPv3 timers for the given link and all groups
1983 * joined on it; state-change, general-query, and group-query timers.
1985 * Only ever called on a transition from v3 to Compatibility mode. Kill
1986 * the timers stone dead (this may be expensive for large N groups), they
1987 * will be restarted if Compatibility Mode deems that they must be due to
1991 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi)
1993 struct ifmultiaddr *ifma;
1995 struct in_multi *inm;
1997 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__,
1998 igi->igi_ifp, igi->igi_ifp->if_xname);
2000 IN_MULTI_LOCK_ASSERT();
2004 * Stop the v3 General Query Response on this link stone dead.
2005 * If fasttimo is woken up due to V_interface_timers_running,
2006 * the flag will be cleared if there are no pending link timers.
2008 igi->igi_v3_timer = 0;
2011 * Now clear the current-state and state-change report timers
2012 * for all memberships scoped to this link.
2016 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2017 if (ifma->ifma_addr->sa_family != AF_INET ||
2018 ifma->ifma_protospec == NULL)
2020 inm = (struct in_multi *)ifma->ifma_protospec;
2021 switch (inm->inm_state) {
2022 case IGMP_NOT_MEMBER:
2023 case IGMP_SILENT_MEMBER:
2024 case IGMP_IDLE_MEMBER:
2025 case IGMP_LAZY_MEMBER:
2026 case IGMP_SLEEPING_MEMBER:
2027 case IGMP_AWAKENING_MEMBER:
2029 * These states are either not relevant in v3 mode,
2030 * or are unreported. Do nothing.
2033 case IGMP_LEAVING_MEMBER:
2035 * If we are leaving the group and switching to
2036 * compatibility mode, we need to release the final
2037 * reference held for issuing the INCLUDE {}, and
2038 * transition to REPORTING to ensure the host leave
2039 * message is sent upstream to the old querier --
2040 * transition to NOT would lose the leave and race.
2042 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK
2043 * around inm_release_locked(), as it is not
2044 * a recursive mutex.
2046 IF_ADDR_UNLOCK(ifp);
2047 inm_release_locked(inm);
2050 case IGMP_G_QUERY_PENDING_MEMBER:
2051 case IGMP_SG_QUERY_PENDING_MEMBER:
2052 inm_clear_recorded(inm);
2054 case IGMP_REPORTING_MEMBER:
2055 inm->inm_state = IGMP_REPORTING_MEMBER;
2059 * Always clear state-change and group report timers.
2060 * Free any pending IGMPv3 state-change records.
2062 inm->inm_sctimer = 0;
2064 _IF_DRAIN(&inm->inm_scq);
2066 IF_ADDR_UNLOCK(ifp);
2070 * Update the Older Version Querier Present timers for a link.
2071 * See Section 7.2.1 of RFC 3376.
2074 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi)
2076 INIT_VNET_INET(curvnet);
2080 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) {
2082 * IGMPv1 and IGMPv2 Querier Present timers expired.
2086 if (igi->igi_version != IGMP_VERSION_3) {
2088 "%s: transition from v%d -> v%d on %p(%s)",
2089 __func__, igi->igi_version, IGMP_VERSION_3,
2090 igi->igi_ifp, igi->igi_ifp->if_xname);
2091 igi->igi_version = IGMP_VERSION_3;
2093 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) {
2095 * IGMPv1 Querier Present timer expired,
2096 * IGMPv2 Querier Present timer running.
2097 * If IGMPv2 was disabled since last timeout,
2099 * If IGMPv2 is enabled, revert to IGMPv2.
2101 if (!V_igmp_v2enable) {
2103 "%s: transition from v%d -> v%d on %p(%s)",
2104 __func__, igi->igi_version, IGMP_VERSION_3,
2105 igi->igi_ifp, igi->igi_ifp->if_xname);
2106 igi->igi_v2_timer = 0;
2107 igi->igi_version = IGMP_VERSION_3;
2109 --igi->igi_v2_timer;
2110 if (igi->igi_version != IGMP_VERSION_2) {
2112 "%s: transition from v%d -> v%d on %p(%s)",
2113 __func__, igi->igi_version, IGMP_VERSION_2,
2114 igi->igi_ifp, igi->igi_ifp->if_xname);
2115 igi->igi_version = IGMP_VERSION_2;
2118 } else if (igi->igi_v1_timer > 0) {
2120 * IGMPv1 Querier Present timer running.
2121 * Stop IGMPv2 timer if running.
2123 * If IGMPv1 was disabled since last timeout,
2125 * If IGMPv1 is enabled, reset IGMPv2 timer if running.
2127 if (!V_igmp_v1enable) {
2129 "%s: transition from v%d -> v%d on %p(%s)",
2130 __func__, igi->igi_version, IGMP_VERSION_3,
2131 igi->igi_ifp, igi->igi_ifp->if_xname);
2132 igi->igi_v1_timer = 0;
2133 igi->igi_version = IGMP_VERSION_3;
2135 --igi->igi_v1_timer;
2137 if (igi->igi_v2_timer > 0) {
2139 "%s: cancel v2 timer on %p(%s)",
2140 __func__, igi->igi_ifp, igi->igi_ifp->if_xname);
2141 igi->igi_v2_timer = 0;
2147 * Global slowtimo handler.
2148 * VIMAGE: Timeout handlers are expected to service all vimages.
2153 VNET_ITERATOR_DECL(vnet_iter);
2156 VNET_FOREACH(vnet_iter) {
2157 CURVNET_SET(vnet_iter);
2158 igmp_slowtimo_vnet();
2161 VNET_LIST_RUNLOCK();
2165 * Per-vnet slowtimo handler.
2168 igmp_slowtimo_vnet(void)
2170 INIT_VNET_INET(curvnet);
2171 struct igmp_ifinfo *igi;
2175 LIST_FOREACH(igi, &V_igi_head, igi_link) {
2176 igmp_v1v2_process_querier_timers(igi);
2183 * Dispatch an IGMPv1/v2 host report or leave message.
2184 * These are always small enough to fit inside a single mbuf.
2187 igmp_v1v2_queue_report(struct in_multi *inm, const int type)
2194 IN_MULTI_LOCK_ASSERT();
2199 MGETHDR(m, M_DONTWAIT, MT_DATA);
2202 MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp));
2204 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp);
2206 m->m_data += sizeof(struct ip);
2207 m->m_len = sizeof(struct igmp);
2209 igmp = mtod(m, struct igmp *);
2210 igmp->igmp_type = type;
2211 igmp->igmp_code = 0;
2212 igmp->igmp_group = inm->inm_addr;
2213 igmp->igmp_cksum = 0;
2214 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp));
2216 m->m_data -= sizeof(struct ip);
2217 m->m_len += sizeof(struct ip);
2219 ip = mtod(m, struct ip *);
2221 ip->ip_len = sizeof(struct ip) + sizeof(struct igmp);
2223 ip->ip_p = IPPROTO_IGMP;
2224 ip->ip_src.s_addr = INADDR_ANY;
2226 if (type == IGMP_HOST_LEAVE_MESSAGE)
2227 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP);
2229 ip->ip_dst = inm->inm_addr;
2231 igmp_save_context(m, ifp);
2233 m->m_flags |= M_IGMPV2;
2234 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK)
2235 m->m_flags |= M_IGMP_LOOP;
2237 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m);
2238 netisr_dispatch(NETISR_IGMP, m);
2244 * Process a state change from the upper layer for the given IPv4 group.
2246 * Each socket holds a reference on the in_multi in its own ip_moptions.
2247 * The socket layer will have made the necessary updates to.the group
2248 * state, it is now up to IGMP to issue a state change report if there
2249 * has been any change between T0 (when the last state-change was issued)
2252 * We use the IGMPv3 state machine at group level. The IGMP module
2253 * however makes the decision as to which IGMP protocol version to speak.
2254 * A state change *from* INCLUDE {} always means an initial join.
2255 * A state change *to* INCLUDE {} always means a final leave.
2257 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can
2258 * save ourselves a bunch of work; any exclusive mode groups need not
2259 * compute source filter lists.
2261 * VIMAGE: curvnet should have been set by caller, as this routine
2262 * is called from the socket option handlers.
2265 igmp_change_state(struct in_multi *inm)
2267 struct igmp_ifinfo *igi;
2271 IN_MULTI_LOCK_ASSERT();
2276 * Try to detect if the upper layer just asked us to change state
2277 * for an interface which has now gone away.
2279 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
2280 ifp = inm->inm_ifma->ifma_ifp;
2283 * Sanity check that netinet's notion of ifp is the
2286 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
2291 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp;
2292 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp));
2295 * If we detect a state transition to or from MCAST_UNDEFINED
2296 * for this group, then we are starting or finishing an IGMP
2297 * life cycle for this group.
2299 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) {
2300 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__,
2301 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode);
2302 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) {
2303 CTR1(KTR_IGMPV3, "%s: initial join", __func__);
2304 error = igmp_initial_join(inm, igi);
2306 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) {
2307 CTR1(KTR_IGMPV3, "%s: final leave", __func__);
2308 igmp_final_leave(inm, igi);
2312 CTR1(KTR_IGMPV3, "%s: filter set change", __func__);
2315 error = igmp_handle_state_change(inm, igi);
2323 * Perform the initial join for an IGMP group.
2325 * When joining a group:
2326 * If the group should have its IGMP traffic suppressed, do nothing.
2327 * IGMPv1 starts sending IGMPv1 host membership reports.
2328 * IGMPv2 starts sending IGMPv2 host membership reports.
2329 * IGMPv3 will schedule an IGMPv3 state-change report containing the
2330 * initial state of the membership.
2333 igmp_initial_join(struct in_multi *inm, struct igmp_ifinfo *igi)
2335 INIT_VNET_INET(curvnet);
2337 struct ifqueue *ifq;
2338 int error, retval, syncstates;
2340 CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)",
2341 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2342 inm->inm_ifp->if_xname);
2349 IN_MULTI_LOCK_ASSERT();
2352 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2355 * Groups joined on loopback or marked as 'not reported',
2356 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and
2357 * are never reported in any IGMP protocol exchanges.
2358 * All other groups enter the appropriate IGMP state machine
2359 * for the version in use on this link.
2360 * A link marked as IGIF_SILENT causes IGMP to be completely
2361 * disabled for the link.
2363 if ((ifp->if_flags & IFF_LOOPBACK) ||
2364 (igi->igi_flags & IGIF_SILENT) ||
2365 !igmp_isgroupreported(inm->inm_addr)) {
2367 "%s: not kicking state machine for silent group", __func__);
2368 inm->inm_state = IGMP_SILENT_MEMBER;
2372 * Deal with overlapping in_multi lifecycle.
2373 * If this group was LEAVING, then make sure
2374 * we drop the reference we picked up to keep the
2375 * group around for the final INCLUDE {} enqueue.
2377 if (igi->igi_version == IGMP_VERSION_3 &&
2378 inm->inm_state == IGMP_LEAVING_MEMBER)
2379 inm_release_locked(inm);
2381 inm->inm_state = IGMP_REPORTING_MEMBER;
2383 switch (igi->igi_version) {
2384 case IGMP_VERSION_1:
2385 case IGMP_VERSION_2:
2386 inm->inm_state = IGMP_IDLE_MEMBER;
2387 error = igmp_v1v2_queue_report(inm,
2388 (igi->igi_version == IGMP_VERSION_2) ?
2389 IGMP_v2_HOST_MEMBERSHIP_REPORT :
2390 IGMP_v1_HOST_MEMBERSHIP_REPORT);
2392 inm->inm_timer = IGMP_RANDOM_DELAY(
2393 IGMP_V1V2_MAX_RI * PR_FASTHZ);
2394 V_current_state_timers_running = 1;
2398 case IGMP_VERSION_3:
2400 * Defer update of T0 to T1, until the first copy
2401 * of the state change has been transmitted.
2406 * Immediately enqueue a State-Change Report for
2407 * this interface, freeing any previous reports.
2408 * Don't kick the timers if there is nothing to do,
2409 * or if an error occurred.
2411 ifq = &inm->inm_scq;
2413 retval = igmp_v3_enqueue_group_record(ifq, inm, 1,
2415 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
2418 error = retval * -1;
2423 * Schedule transmission of pending state-change
2424 * report up to RV times for this link. The timer
2425 * will fire at the next igmp_fasttimo (~200ms),
2426 * giving us an opportunity to merge the reports.
2428 if (igi->igi_flags & IGIF_LOOPBACK) {
2431 KASSERT(igi->igi_rv > 1,
2432 ("%s: invalid robustness %d", __func__,
2434 inm->inm_scrv = igi->igi_rv;
2436 inm->inm_sctimer = 1;
2437 V_state_change_timers_running = 1;
2445 * Only update the T0 state if state change is atomic,
2446 * i.e. we don't need to wait for a timer to fire before we
2447 * can consider the state change to have been communicated.
2451 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2452 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2459 * Issue an intermediate state change during the IGMP life-cycle.
2462 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifinfo *igi)
2464 INIT_VNET_INET(curvnet);
2468 CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)",
2469 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2470 inm->inm_ifp->if_xname);
2474 IN_MULTI_LOCK_ASSERT();
2477 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__));
2479 if ((ifp->if_flags & IFF_LOOPBACK) ||
2480 (igi->igi_flags & IGIF_SILENT) ||
2481 !igmp_isgroupreported(inm->inm_addr) ||
2482 (igi->igi_version != IGMP_VERSION_3)) {
2483 if (!igmp_isgroupreported(inm->inm_addr)) {
2485 "%s: not kicking state machine for silent group", __func__);
2487 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__);
2489 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2490 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2494 _IF_DRAIN(&inm->inm_scq);
2496 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0);
2497 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval);
2502 * If record(s) were enqueued, start the state-change
2503 * report timer for this group.
2505 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv);
2506 inm->inm_sctimer = 1;
2507 V_state_change_timers_running = 1;
2513 * Perform the final leave for an IGMP group.
2515 * When leaving a group:
2516 * IGMPv1 does nothing.
2517 * IGMPv2 sends a host leave message, if and only if we are the reporter.
2518 * IGMPv3 enqueues a state-change report containing a transition
2519 * to INCLUDE {} for immediate transmission.
2522 igmp_final_leave(struct in_multi *inm, struct igmp_ifinfo *igi)
2524 INIT_VNET_INET(curvnet);
2529 CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)",
2530 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp,
2531 inm->inm_ifp->if_xname);
2533 IN_MULTI_LOCK_ASSERT();
2536 switch (inm->inm_state) {
2537 case IGMP_NOT_MEMBER:
2538 case IGMP_SILENT_MEMBER:
2539 case IGMP_LEAVING_MEMBER:
2540 /* Already leaving or left; do nothing. */
2542 "%s: not kicking state machine for silent group", __func__);
2544 case IGMP_REPORTING_MEMBER:
2545 case IGMP_IDLE_MEMBER:
2546 case IGMP_G_QUERY_PENDING_MEMBER:
2547 case IGMP_SG_QUERY_PENDING_MEMBER:
2548 if (igi->igi_version == IGMP_VERSION_2) {
2550 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER ||
2551 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)
2552 panic("%s: IGMPv3 state reached, not IGMPv3 mode",
2555 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE);
2556 inm->inm_state = IGMP_NOT_MEMBER;
2557 } else if (igi->igi_version == IGMP_VERSION_3) {
2559 * Stop group timer and all pending reports.
2560 * Immediately enqueue a state-change report
2561 * TO_IN {} to be sent on the next fast timeout,
2562 * giving us an opportunity to merge reports.
2564 _IF_DRAIN(&inm->inm_scq);
2566 if (igi->igi_flags & IGIF_LOOPBACK) {
2569 inm->inm_scrv = igi->igi_rv;
2571 CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d "
2572 "pending retransmissions.", __func__,
2573 inet_ntoa(inm->inm_addr),
2574 inm->inm_ifp->if_xname, inm->inm_scrv);
2575 if (inm->inm_scrv == 0) {
2576 inm->inm_state = IGMP_NOT_MEMBER;
2577 inm->inm_sctimer = 0;
2581 inm_acquire_locked(inm);
2583 retval = igmp_v3_enqueue_group_record(
2584 &inm->inm_scq, inm, 1, 0, 0);
2585 KASSERT(retval != 0,
2586 ("%s: enqueue record = %d", __func__,
2589 inm->inm_state = IGMP_LEAVING_MEMBER;
2590 inm->inm_sctimer = 1;
2591 V_state_change_timers_running = 1;
2597 case IGMP_LAZY_MEMBER:
2598 case IGMP_SLEEPING_MEMBER:
2599 case IGMP_AWAKENING_MEMBER:
2600 /* Our reports are suppressed; do nothing. */
2606 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__,
2607 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2608 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
2609 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s",
2610 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname);
2615 * Enqueue an IGMPv3 group record to the given output queue.
2617 * XXX This function could do with having the allocation code
2618 * split out, and the multiple-tree-walks coalesced into a single
2619 * routine as has been done in igmp_v3_enqueue_filter_change().
2621 * If is_state_change is zero, a current-state record is appended.
2622 * If is_state_change is non-zero, a state-change report is appended.
2624 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2625 * If is_group_query is zero, and if there is a packet with free space
2626 * at the tail of the queue, it will be appended to providing there
2627 * is enough free space.
2628 * Otherwise a new mbuf packet chain is allocated.
2630 * If is_source_query is non-zero, each source is checked to see if
2631 * it was recorded for a Group-Source query, and will be omitted if
2632 * it is not both in-mode and recorded.
2634 * The function will attempt to allocate leading space in the packet
2635 * for the IP/IGMP header to be prepended without fragmenting the chain.
2637 * If successful the size of all data appended to the queue is returned,
2638 * otherwise an error code less than zero is returned, or zero if
2639 * no record(s) were appended.
2642 igmp_v3_enqueue_group_record(struct ifqueue *ifq, struct in_multi *inm,
2643 const int is_state_change, const int is_group_query,
2644 const int is_source_query)
2646 struct igmp_grouprec ig;
2647 struct igmp_grouprec *pig;
2649 struct ip_msource *ims, *nims;
2650 struct mbuf *m0, *m, *md;
2651 int error, is_filter_list_change;
2652 int minrec0len, m0srcs, msrcs, nbytes, off;
2653 int record_has_sources;
2659 IN_MULTI_LOCK_ASSERT();
2663 is_filter_list_change = 0;
2670 record_has_sources = 1;
2672 type = IGMP_DO_NOTHING;
2673 mode = inm->inm_st[1].iss_fmode;
2676 * If we did not transition out of ASM mode during t0->t1,
2677 * and there are no source nodes to process, we can skip
2678 * the generation of source records.
2680 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 &&
2682 record_has_sources = 0;
2684 if (is_state_change) {
2686 * Queue a state change record.
2687 * If the mode did not change, and there are non-ASM
2688 * listeners or source filters present,
2689 * we potentially need to issue two records for the group.
2690 * If we are transitioning to MCAST_UNDEFINED, we need
2691 * not send any sources.
2692 * If there are ASM listeners, and there was no filter
2693 * mode transition of any kind, do nothing.
2695 if (mode != inm->inm_st[0].iss_fmode) {
2696 if (mode == MCAST_EXCLUDE) {
2697 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE",
2699 type = IGMP_CHANGE_TO_EXCLUDE_MODE;
2701 CTR1(KTR_IGMPV3, "%s: change to INCLUDE",
2703 type = IGMP_CHANGE_TO_INCLUDE_MODE;
2704 if (mode == MCAST_UNDEFINED)
2705 record_has_sources = 0;
2708 if (record_has_sources) {
2709 is_filter_list_change = 1;
2711 type = IGMP_DO_NOTHING;
2716 * Queue a current state record.
2718 if (mode == MCAST_EXCLUDE) {
2719 type = IGMP_MODE_IS_EXCLUDE;
2720 } else if (mode == MCAST_INCLUDE) {
2721 type = IGMP_MODE_IS_INCLUDE;
2722 KASSERT(inm->inm_st[1].iss_asm == 0,
2723 ("%s: inm %p is INCLUDE but ASM count is %d",
2724 __func__, inm, inm->inm_st[1].iss_asm));
2729 * Generate the filter list changes using a separate function.
2731 if (is_filter_list_change)
2732 return (igmp_v3_enqueue_filter_change(ifq, inm));
2734 if (type == IGMP_DO_NOTHING) {
2735 CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s",
2736 __func__, inet_ntoa(inm->inm_addr),
2737 inm->inm_ifp->if_xname);
2742 * If any sources are present, we must be able to fit at least
2743 * one in the trailing space of the tail packet's mbuf,
2746 minrec0len = sizeof(struct igmp_grouprec);
2747 if (record_has_sources)
2748 minrec0len += sizeof(in_addr_t);
2750 CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__,
2751 igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr),
2752 inm->inm_ifp->if_xname);
2755 * Check if we have a packet in the tail of the queue for this
2756 * group into which the first group record for this group will fit.
2757 * Otherwise allocate a new packet.
2758 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT.
2759 * Note: Group records for G/GSR query responses MUST be sent
2760 * in their own packet.
2763 if (!is_group_query &&
2765 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) &&
2766 (m0->m_pkthdr.len + minrec0len) <
2767 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
2768 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2769 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2771 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__);
2773 if (_IF_QFULL(ifq)) {
2774 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2778 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2779 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2780 if (!is_state_change && !is_group_query) {
2781 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2783 m->m_data += IGMP_LEADINGSPACE;
2786 m = m_gethdr(M_DONTWAIT, MT_DATA);
2788 MH_ALIGN(m, IGMP_LEADINGSPACE);
2793 igmp_save_context(m, ifp);
2795 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__);
2799 * Append group record.
2800 * If we have sources, we don't know how many yet.
2805 ig.ig_group = inm->inm_addr;
2806 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2809 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2812 nbytes += sizeof(struct igmp_grouprec);
2815 * Append as many sources as will fit in the first packet.
2816 * If we are appending to a new packet, the chain allocation
2817 * may potentially use clusters; use m_getptr() in this case.
2818 * If we are appending to an existing packet, we need to obtain
2819 * a pointer to the group record after m_append(), in case a new
2820 * mbuf was allocated.
2821 * Only append sources which are in-mode at t1. If we are
2822 * transitioning to MCAST_UNDEFINED state on the group, do not
2823 * include source entries.
2824 * Only report recorded sources in our filter set when responding
2825 * to a group-source query.
2827 if (record_has_sources) {
2830 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2831 md->m_len - nbytes);
2833 md = m_getptr(m, 0, &off);
2834 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) +
2838 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) {
2839 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2840 inet_ntoa_haddr(ims->ims_haddr));
2841 now = ims_get_mode(inm, ims, 1);
2842 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now);
2843 if ((now != mode) ||
2844 (now == mode && mode == MCAST_UNDEFINED)) {
2845 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2848 if (is_source_query && ims->ims_stp == 0) {
2849 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2853 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2854 naddr = htonl(ims->ims_haddr);
2855 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2858 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2862 nbytes += sizeof(in_addr_t);
2864 if (msrcs == m0srcs)
2867 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__,
2869 pig->ig_numsrc = htons(msrcs);
2870 nbytes += (msrcs * sizeof(in_addr_t));
2873 if (is_source_query && msrcs == 0) {
2874 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__);
2881 * We are good to go with first packet.
2884 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__);
2885 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2886 _IF_ENQUEUE(ifq, m);
2888 m->m_pkthdr.PH_vt.vt_nrecs++;
2891 * No further work needed if no source list in packet(s).
2893 if (!record_has_sources)
2897 * Whilst sources remain to be announced, we need to allocate
2898 * a new packet and fill out as many sources as will fit.
2899 * Always try for a cluster first.
2901 while (nims != NULL) {
2902 if (_IF_QFULL(ifq)) {
2903 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__);
2906 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2908 m->m_data += IGMP_LEADINGSPACE;
2910 m = m_gethdr(M_DONTWAIT, MT_DATA);
2912 MH_ALIGN(m, IGMP_LEADINGSPACE);
2916 igmp_save_context(m, ifp);
2917 md = m_getptr(m, 0, &off);
2918 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off);
2919 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__);
2921 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) {
2924 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__);
2927 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2928 nbytes += sizeof(struct igmp_grouprec);
2930 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
2931 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t);
2934 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
2935 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2936 inet_ntoa_haddr(ims->ims_haddr));
2937 now = ims_get_mode(inm, ims, 1);
2938 if ((now != mode) ||
2939 (now == mode && mode == MCAST_UNDEFINED)) {
2940 CTR1(KTR_IGMPV3, "%s: skip node", __func__);
2943 if (is_source_query && ims->ims_stp == 0) {
2944 CTR1(KTR_IGMPV3, "%s: skip unrecorded node",
2948 CTR1(KTR_IGMPV3, "%s: append node", __func__);
2949 naddr = htonl(ims->ims_haddr);
2950 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) {
2953 CTR1(KTR_IGMPV3, "%s: m_append() failed.",
2958 if (msrcs == m0srcs)
2961 pig->ig_numsrc = htons(msrcs);
2962 nbytes += (msrcs * sizeof(in_addr_t));
2964 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__);
2965 _IF_ENQUEUE(ifq, m);
2972 * Type used to mark record pass completion.
2973 * We exploit the fact we can cast to this easily from the
2974 * current filter modes on each ip_msource node.
2977 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2978 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2979 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2980 REC_FULL = REC_ALLOW | REC_BLOCK
2984 * Enqueue an IGMPv3 filter list change to the given output queue.
2986 * Source list filter state is held in an RB-tree. When the filter list
2987 * for a group is changed without changing its mode, we need to compute
2988 * the deltas between T0 and T1 for each source in the filter set,
2989 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2991 * As we may potentially queue two record types, and the entire R-B tree
2992 * needs to be walked at once, we break this out into its own function
2993 * so we can generate a tightly packed queue of packets.
2995 * XXX This could be written to only use one tree walk, although that makes
2996 * serializing into the mbuf chains a bit harder. For now we do two walks
2997 * which makes things easier on us, and it may or may not be harder on
3000 * If successful the size of all data appended to the queue is returned,
3001 * otherwise an error code less than zero is returned, or zero if
3002 * no record(s) were appended.
3005 igmp_v3_enqueue_filter_change(struct ifqueue *ifq, struct in_multi *inm)
3007 static const int MINRECLEN =
3008 sizeof(struct igmp_grouprec) + sizeof(in_addr_t);
3010 struct igmp_grouprec ig;
3011 struct igmp_grouprec *pig;
3012 struct ip_msource *ims, *nims;
3013 struct mbuf *m, *m0, *md;
3015 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3017 uint8_t mode, now, then;
3018 rectype_t crt, drt, nrt;
3020 IN_MULTI_LOCK_ASSERT();
3022 if (inm->inm_nsrc == 0 ||
3023 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0))
3026 ifp = inm->inm_ifp; /* interface */
3027 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */
3028 crt = REC_NONE; /* current group record type */
3029 drt = REC_NONE; /* mask of completed group record types */
3030 nrt = REC_NONE; /* record type for current node */
3031 m0srcs = 0; /* # source which will fit in current mbuf chain */
3032 nbytes = 0; /* # of bytes appended to group's state-change queue */
3033 npbytes = 0; /* # of bytes appended this packet */
3034 rsrcs = 0; /* # sources encoded in current record */
3035 schanged = 0; /* # nodes encoded in overall filter change */
3036 nallow = 0; /* # of source entries in ALLOW_NEW */
3037 nblock = 0; /* # of source entries in BLOCK_OLD */
3038 nims = NULL; /* next tree node pointer */
3041 * For each possible filter record mode.
3042 * The first kind of source we encounter tells us which
3043 * is the first kind of record we start appending.
3044 * If a node transitioned to UNDEFINED at t1, its mode is treated
3045 * as the inverse of the group's filter mode.
3047 while (drt != REC_FULL) {
3051 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
3052 IGMP_V3_REPORT_MAXRECS) &&
3053 (m0->m_pkthdr.len + MINRECLEN) <
3054 (ifp->if_mtu - IGMP_LEADINGSPACE)) {
3056 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3057 sizeof(struct igmp_grouprec)) /
3060 "%s: use previous packet", __func__);
3062 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3064 m->m_data += IGMP_LEADINGSPACE;
3066 m = m_gethdr(M_DONTWAIT, MT_DATA);
3068 MH_ALIGN(m, IGMP_LEADINGSPACE);
3072 "%s: m_get*() failed", __func__);
3075 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3076 igmp_save_context(m, ifp);
3077 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE -
3078 sizeof(struct igmp_grouprec)) /
3082 "%s: allocated new packet", __func__);
3085 * Append the IGMP group record header to the
3086 * current packet's data area.
3087 * Recalculate pointer to free space for next
3088 * group record, in case m_append() allocated
3089 * a new mbuf or cluster.
3091 memset(&ig, 0, sizeof(ig));
3092 ig.ig_group = inm->inm_addr;
3093 if (!m_append(m, sizeof(ig), (void *)&ig)) {
3097 "%s: m_append() failed", __func__);
3100 npbytes += sizeof(struct igmp_grouprec);
3102 /* new packet; offset in c hain */
3103 md = m_getptr(m, npbytes -
3104 sizeof(struct igmp_grouprec), &off);
3105 pig = (struct igmp_grouprec *)(mtod(md,
3108 /* current packet; offset from last append */
3110 pig = (struct igmp_grouprec *)(mtod(md,
3111 uint8_t *) + md->m_len -
3112 sizeof(struct igmp_grouprec));
3115 * Begin walking the tree for this record type
3116 * pass, or continue from where we left off
3117 * previously if we had to allocate a new packet.
3118 * Only report deltas in-mode at t1.
3119 * We need not report included sources as allowed
3120 * if we are in inclusive mode on the group,
3121 * however the converse is not true.
3125 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs);
3126 RB_FOREACH_FROM(ims, ip_msource_tree, nims) {
3127 CTR2(KTR_IGMPV3, "%s: visit node %s",
3128 __func__, inet_ntoa_haddr(ims->ims_haddr));
3129 now = ims_get_mode(inm, ims, 1);
3130 then = ims_get_mode(inm, ims, 0);
3131 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d",
3132 __func__, then, now);
3135 "%s: skip unchanged", __func__);
3138 if (mode == MCAST_EXCLUDE &&
3139 now == MCAST_INCLUDE) {
3141 "%s: skip IN src on EX group",
3145 nrt = (rectype_t)now;
3146 if (nrt == REC_NONE)
3147 nrt = (rectype_t)(~mode & REC_FULL);
3148 if (schanged++ == 0) {
3150 } else if (crt != nrt)
3152 naddr = htonl(ims->ims_haddr);
3153 if (!m_append(m, sizeof(in_addr_t),
3158 "%s: m_append() failed", __func__);
3161 nallow += !!(crt == REC_ALLOW);
3162 nblock += !!(crt == REC_BLOCK);
3163 if (++rsrcs == m0srcs)
3167 * If we did not append any tree nodes on this
3168 * pass, back out of allocations.
3171 npbytes -= sizeof(struct igmp_grouprec);
3174 "%s: m_free(m)", __func__);
3178 "%s: m_adj(m, -ig)", __func__);
3179 m_adj(m, -((int)sizeof(
3180 struct igmp_grouprec)));
3184 npbytes += (rsrcs * sizeof(in_addr_t));
3185 if (crt == REC_ALLOW)
3186 pig->ig_type = IGMP_ALLOW_NEW_SOURCES;
3187 else if (crt == REC_BLOCK)
3188 pig->ig_type = IGMP_BLOCK_OLD_SOURCES;
3189 pig->ig_numsrc = htons(rsrcs);
3191 * Count the new group record, and enqueue this
3192 * packet if it wasn't already queued.
3194 m->m_pkthdr.PH_vt.vt_nrecs++;
3196 _IF_ENQUEUE(ifq, m);
3198 } while (nims != NULL);
3200 crt = (~crt & REC_FULL);
3203 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
3210 igmp_v3_merge_state_changes(struct in_multi *inm, struct ifqueue *ifscq)
3213 struct mbuf *m; /* pending state-change */
3214 struct mbuf *m0; /* copy of pending state-change */
3215 struct mbuf *mt; /* last state-change in packet */
3216 int docopy, domerge;
3223 IN_MULTI_LOCK_ASSERT();
3227 * If there are further pending retransmissions, make a writable
3228 * copy of each queued state-change message before merging.
3230 if (inm->inm_scrv > 0)
3235 if (gq->ifq_head == NULL) {
3236 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty",
3244 * Only merge the report into the current packet if
3245 * there is sufficient space to do so; an IGMPv3 report
3246 * packet may only contain 65,535 group records.
3247 * Always use a simple mbuf chain concatentation to do this,
3248 * as large state changes for single groups may have
3249 * allocated clusters.
3252 mt = ifscq->ifq_tail;
3254 recslen = m_length(m, NULL);
3256 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
3257 m->m_pkthdr.PH_vt.vt_nrecs <=
3258 IGMP_V3_REPORT_MAXRECS) &&
3259 (mt->m_pkthdr.len + recslen <=
3260 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE)))
3264 if (!domerge && _IF_QFULL(gq)) {
3266 "%s: outbound queue full, skipping whole packet %p",
3276 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m);
3277 _IF_DEQUEUE(gq, m0);
3280 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m);
3281 m0 = m_dup(m, M_NOWAIT);
3284 m0->m_nextpkt = NULL;
3289 CTR3(KTR_IGMPV3, "%s: queueing %p to ifscq %p)",
3290 __func__, m0, ifscq);
3291 _IF_ENQUEUE(ifscq, m0);
3293 struct mbuf *mtl; /* last mbuf of packet mt */
3295 CTR3(KTR_IGMPV3, "%s: merging %p with ifscq tail %p)",
3299 m0->m_flags &= ~M_PKTHDR;
3300 mt->m_pkthdr.len += recslen;
3301 mt->m_pkthdr.PH_vt.vt_nrecs +=
3302 m0->m_pkthdr.PH_vt.vt_nrecs;
3312 * Respond to a pending IGMPv3 General Query.
3315 igmp_v3_dispatch_general_query(struct igmp_ifinfo *igi)
3317 INIT_VNET_INET(curvnet);
3318 struct ifmultiaddr *ifma, *tifma;
3320 struct in_multi *inm;
3323 IN_MULTI_LOCK_ASSERT();
3326 KASSERT(igi->igi_version == IGMP_VERSION_3,
3327 ("%s: called when version %d", __func__, igi->igi_version));
3332 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) {
3333 if (ifma->ifma_addr->sa_family != AF_INET ||
3334 ifma->ifma_protospec == NULL)
3337 inm = (struct in_multi *)ifma->ifma_protospec;
3338 KASSERT(ifp == inm->inm_ifp,
3339 ("%s: inconsistent ifp", __func__));
3341 switch (inm->inm_state) {
3342 case IGMP_NOT_MEMBER:
3343 case IGMP_SILENT_MEMBER:
3345 case IGMP_REPORTING_MEMBER:
3346 case IGMP_IDLE_MEMBER:
3347 case IGMP_LAZY_MEMBER:
3348 case IGMP_SLEEPING_MEMBER:
3349 case IGMP_AWAKENING_MEMBER:
3350 inm->inm_state = IGMP_REPORTING_MEMBER;
3351 retval = igmp_v3_enqueue_group_record(&igi->igi_gq,
3353 CTR2(KTR_IGMPV3, "%s: enqueue record = %d",
3356 case IGMP_G_QUERY_PENDING_MEMBER:
3357 case IGMP_SG_QUERY_PENDING_MEMBER:
3358 case IGMP_LEAVING_MEMBER:
3362 IF_ADDR_UNLOCK(ifp);
3364 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0;
3365 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop);
3368 * Slew transmission of bursts over 500ms intervals.
3370 if (igi->igi_gq.ifq_head != NULL) {
3371 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY(
3372 IGMP_RESPONSE_BURST_INTERVAL);
3373 V_interface_timers_running = 1;
3378 * Transmit the next pending IGMP message in the output queue.
3380 * We get called from netisr_processqueue(). A mutex private to igmpoq
3381 * will be acquired and released around this routine.
3383 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3384 * MRT: Nothing needs to be done, as IGMP traffic is always local to
3385 * a link and uses a link-scope multicast address.
3388 igmp_intr(struct mbuf *m)
3390 struct ip_moptions imo;
3392 struct mbuf *ipopts, *m0;
3396 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m);
3399 * Set VNET image pointer from enqueued mbuf chain
3400 * before doing anything else. Whilst we use interface
3401 * indexes to guard against interface detach, they are
3402 * unique to each VIMAGE and must be retrieved.
3404 CURVNET_SET((struct vnet *)(m->m_pkthdr.header));
3405 INIT_VNET_NET(curvnet);
3406 INIT_VNET_INET(curvnet);
3407 ifindex = igmp_restore_context(m);
3410 * Check if the ifnet still exists. This limits the scope of
3411 * any race in the absence of a global ifp lock for low cost
3412 * (an array lookup).
3414 ifp = ifnet_byindex(ifindex);
3416 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.",
3417 __func__, m, ifindex);
3419 IPSTAT_INC(ips_noroute);
3423 ipopts = V_igmp_sendra ? m_raopt : NULL;
3425 imo.imo_multicast_ttl = 1;
3426 imo.imo_multicast_vif = -1;
3427 imo.imo_multicast_loop = (V_ip_mrouter != NULL);
3430 * If the user requested that IGMP traffic be explicitly
3431 * redirected to the loopback interface (e.g. they are running a
3432 * MANET interface and the routing protocol needs to see the
3433 * updates), handle this now.
3435 if (m->m_flags & M_IGMP_LOOP)
3436 imo.imo_multicast_ifp = V_loif;
3438 imo.imo_multicast_ifp = ifp;
3440 if (m->m_flags & M_IGMPV2) {
3443 m0 = igmp_v3_encap_report(ifp, m);
3445 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m);
3447 IPSTAT_INC(ips_odropped);
3452 igmp_scrub_context(m0);
3453 m->m_flags &= ~(M_PROTOFLAGS);
3454 m0->m_pkthdr.rcvif = V_loif;
3456 mac_netinet_igmp_send(ifp, m0);
3458 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL);
3460 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error);
3464 IGMPSTAT_INC(igps_snd_reports);
3468 * We must restore the existing vnet pointer before
3469 * continuing as we are run from netisr context.
3475 * Encapsulate an IGMPv3 report.
3477 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf
3478 * chain has already had its IP/IGMPv3 header prepended. In this case
3479 * the function will not attempt to prepend; the lengths and checksums
3480 * will however be re-computed.
3482 * Returns a pointer to the new mbuf chain head, or NULL if the
3483 * allocation failed.
3485 static struct mbuf *
3486 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m)
3488 INIT_VNET_INET(curvnet);
3489 struct igmp_report *igmp;
3491 int hdrlen, igmpreclen;
3493 KASSERT((m->m_flags & M_PKTHDR),
3494 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3496 igmpreclen = m_length(m, NULL);
3497 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report);
3499 if (m->m_flags & M_IGMPV3_HDR) {
3500 igmpreclen -= hdrlen;
3502 M_PREPEND(m, hdrlen, M_DONTWAIT);
3505 m->m_flags |= M_IGMPV3_HDR;
3508 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen);
3510 m->m_data += sizeof(struct ip);
3511 m->m_len -= sizeof(struct ip);
3513 igmp = mtod(m, struct igmp_report *);
3514 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT;
3517 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3519 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen);
3520 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3522 m->m_data -= sizeof(struct ip);
3523 m->m_len += sizeof(struct ip);
3525 ip = mtod(m, struct ip *);
3526 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL;
3527 ip->ip_len = hdrlen + igmpreclen;
3529 ip->ip_p = IPPROTO_IGMP;
3532 ip->ip_src.s_addr = INADDR_ANY;
3534 if (m->m_flags & M_IGMP_LOOP) {
3535 struct in_ifaddr *ia;
3539 ip->ip_src = ia->ia_addr.sin_addr;
3542 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP);
3549 igmp_rec_type_to_str(const int type)
3553 case IGMP_CHANGE_TO_EXCLUDE_MODE:
3556 case IGMP_CHANGE_TO_INCLUDE_MODE:
3559 case IGMP_MODE_IS_EXCLUDE:
3562 case IGMP_MODE_IS_INCLUDE:
3565 case IGMP_ALLOW_NEW_SOURCES:
3568 case IGMP_BLOCK_OLD_SOURCES:
3582 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3586 m_raopt = igmp_ra_alloc();
3588 netisr_register(&igmp_nh);
3592 igmp_sysuninit(void)
3595 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3597 netisr_unregister(&igmp_nh);
3602 IGMP_LOCK_DESTROY();
3606 * Initialize an IGMPv3 instance.
3607 * VIMAGE: Assumes curvnet set by caller and called per vimage.
3610 vnet_igmp_iattach(const void *unused __unused)
3612 INIT_VNET_INET(curvnet);
3614 CTR1(KTR_IGMPV3, "%s: initializing", __func__);
3616 LIST_INIT(&V_igi_head);
3618 V_current_state_timers_running = 0;
3619 V_state_change_timers_running = 0;
3620 V_interface_timers_running = 0;
3623 * Initialize sysctls to default values.
3625 V_igmp_recvifkludge = 1;
3627 V_igmp_sendlocal = 1;
3628 V_igmp_v1enable = 1;
3629 V_igmp_v2enable = 1;
3630 V_igmp_legacysupp = 0;
3631 V_igmp_default_version = IGMP_VERSION_3;
3632 V_igmp_gsrdelay.tv_sec = 10;
3633 V_igmp_gsrdelay.tv_usec = 0;
3635 memset(&V_igmpstat, 0, sizeof(struct igmpstat));
3636 V_igmpstat.igps_version = IGPS_VERSION_3;
3637 V_igmpstat.igps_len = sizeof(struct igmpstat);
3643 vnet_igmp_idetach(const void *unused __unused)
3646 INIT_VNET_INET(curvnet);
3649 CTR1(KTR_IGMPV3, "%s: tearing down", __func__);
3651 KASSERT(LIST_EMPTY(&V_igi_head),
3652 ("%s: igi list not empty; ifnets not detached?", __func__));
3657 #ifndef VIMAGE_GLOBALS
3658 static vnet_modinfo_t vnet_igmp_modinfo = {
3659 .vmi_id = VNET_MOD_IGMP,
3661 .vmi_dependson = VNET_MOD_INET,
3662 .vmi_iattach = vnet_igmp_iattach,
3663 .vmi_idetach = vnet_igmp_idetach
3668 igmp_modevent(module_t mod, int type, void *unused __unused)
3674 #ifndef VIMAGE_GLOBALS
3675 vnet_mod_register(&vnet_igmp_modinfo);
3677 vnet_igmp_iattach(NULL);
3681 #ifndef VIMAGE_GLOBALS
3682 vnet_mod_deregister(&vnet_igmp_modinfo);
3684 vnet_igmp_idetach(NULL);
3689 return (EOPNOTSUPP);
3694 static moduledata_t igmp_mod = {
3699 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);