mld6.c revision 237992
1217309Snwhitehorn/*- 2251843Sbapt * Copyright (c) 2009 Bruce Simpson. 3217309Snwhitehorn * 4217309Snwhitehorn * Redistribution and use in source and binary forms, with or without 5217309Snwhitehorn * modification, are permitted provided that the following conditions 6251843Sbapt * are met: 7217309Snwhitehorn * 1. Redistributions of source code must retain the above copyright 8217309Snwhitehorn * notice, this list of conditions and the following disclaimer. 9217309Snwhitehorn * 2. Redistributions in binary form must reproduce the above copyright 10217309Snwhitehorn * notice, this list of conditions and the following disclaimer in the 11217309Snwhitehorn * documentation and/or other materials provided with the distribution. 12217309Snwhitehorn * 3. The name of the author may not be used to endorse or promote 13217309Snwhitehorn * products derived from this software without specific prior written 14217309Snwhitehorn * permission. 15217309Snwhitehorn * 16217309Snwhitehorn * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17217309Snwhitehorn * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18217309Snwhitehorn * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19217309Snwhitehorn * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20217309Snwhitehorn * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21217309Snwhitehorn * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22217309Snwhitehorn * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23217309Snwhitehorn * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24217309Snwhitehorn * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25217309Snwhitehorn * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26217309Snwhitehorn * SUCH DAMAGE. 27217309Snwhitehorn * 28217309Snwhitehorn * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $ 29217309Snwhitehorn */ 30217309Snwhitehorn 31217309Snwhitehorn/*- 32217309Snwhitehorn * Copyright (c) 1988 Stephen Deering. 33217309Snwhitehorn * Copyright (c) 1992, 1993 34217309Snwhitehorn * The Regents of the University of California. All rights reserved. 35217309Snwhitehorn * 36217309Snwhitehorn * This code is derived from software contributed to Berkeley by 37217309Snwhitehorn * Stephen Deering of Stanford University. 38217309Snwhitehorn * 39217309Snwhitehorn * Redistribution and use in source and binary forms, with or without 40217309Snwhitehorn * modification, are permitted provided that the following conditions 41217309Snwhitehorn * are met: 42217309Snwhitehorn * 1. Redistributions of source code must retain the above copyright 43217309Snwhitehorn * notice, this list of conditions and the following disclaimer. 44217309Snwhitehorn * 2. Redistributions in binary form must reproduce the above copyright 45217309Snwhitehorn * notice, this list of conditions and the following disclaimer in the 46217309Snwhitehorn * documentation and/or other materials provided with the distribution. 47217309Snwhitehorn * 4. Neither the name of the University nor the names of its contributors 48217309Snwhitehorn * may be used to endorse or promote products derived from this software 49217309Snwhitehorn * without specific prior written permission. 50217309Snwhitehorn * 51217309Snwhitehorn * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52217309Snwhitehorn * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53217309Snwhitehorn * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54217309Snwhitehorn * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55217309Snwhitehorn * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56217309Snwhitehorn * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57217309Snwhitehorn * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58217309Snwhitehorn * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59217309Snwhitehorn * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60217309Snwhitehorn * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61217309Snwhitehorn * SUCH DAMAGE. 62217309Snwhitehorn * 63217309Snwhitehorn * @(#)igmp.c 8.1 (Berkeley) 7/19/93 64217309Snwhitehorn */ 65217309Snwhitehorn 66217309Snwhitehorn#include <sys/cdefs.h> 67217309Snwhitehorn__FBSDID("$FreeBSD: stable/9/sys/netinet6/mld6.c 237992 2012-07-02 10:07:32Z bms $"); 68217309Snwhitehorn 69217309Snwhitehorn#include "opt_inet.h" 70217309Snwhitehorn#include "opt_inet6.h" 71217309Snwhitehorn 72217309Snwhitehorn#include <sys/param.h> 73217309Snwhitehorn#include <sys/systm.h> 74217309Snwhitehorn#include <sys/mbuf.h> 75217309Snwhitehorn#include <sys/socket.h> 76217309Snwhitehorn#include <sys/protosw.h> 77217309Snwhitehorn#include <sys/sysctl.h> 78217309Snwhitehorn#include <sys/kernel.h> 79217309Snwhitehorn#include <sys/callout.h> 80217309Snwhitehorn#include <sys/malloc.h> 81217309Snwhitehorn#include <sys/module.h> 82217309Snwhitehorn#include <sys/ktr.h> 83217309Snwhitehorn 84217309Snwhitehorn#include <net/if.h> 85217309Snwhitehorn#include <net/route.h> 86217309Snwhitehorn#include <net/vnet.h> 87217309Snwhitehorn 88217309Snwhitehorn#include <netinet/in.h> 89217309Snwhitehorn#include <netinet/in_var.h> 90217309Snwhitehorn#include <netinet6/in6_var.h> 91217309Snwhitehorn#include <netinet/ip6.h> 92217309Snwhitehorn#include <netinet6/ip6_var.h> 93217309Snwhitehorn#include <netinet6/scope6_var.h> 94217309Snwhitehorn#include <netinet/icmp6.h> 95217309Snwhitehorn#include <netinet6/mld6.h> 96217309Snwhitehorn#include <netinet6/mld6_var.h> 97217309Snwhitehorn 98217309Snwhitehorn#include <security/mac/mac_framework.h> 99217309Snwhitehorn 100217309Snwhitehorn#ifndef KTR_MLD 101217309Snwhitehorn#define KTR_MLD KTR_INET6 102217309Snwhitehorn#endif 103217309Snwhitehorn 104217309Snwhitehornstatic struct mld_ifinfo * 105217309Snwhitehorn mli_alloc_locked(struct ifnet *); 106217309Snwhitehornstatic void mli_delete_locked(const struct ifnet *); 107217309Snwhitehornstatic void mld_dispatch_packet(struct mbuf *); 108217309Snwhitehornstatic void mld_dispatch_queue(struct ifqueue *, int); 109217309Snwhitehornstatic void mld_final_leave(struct in6_multi *, struct mld_ifinfo *); 110217309Snwhitehornstatic void mld_fasttimo_vnet(void); 111217309Snwhitehornstatic int mld_handle_state_change(struct in6_multi *, 112217309Snwhitehorn struct mld_ifinfo *); 113217309Snwhitehornstatic int mld_initial_join(struct in6_multi *, struct mld_ifinfo *, 114217309Snwhitehorn const int); 115217309Snwhitehorn#ifdef KTR 116217309Snwhitehornstatic char * mld_rec_type_to_str(const int); 117217309Snwhitehorn#endif 118217309Snwhitehornstatic void mld_set_version(struct mld_ifinfo *, const int); 119217309Snwhitehornstatic void mld_slowtimo_vnet(void); 120217309Snwhitehornstatic int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, 121217309Snwhitehorn /*const*/ struct mld_hdr *); 122217309Snwhitehornstatic int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, 123217309Snwhitehorn /*const*/ struct mld_hdr *); 124217309Snwhitehornstatic void mld_v1_process_group_timer(struct mld_ifinfo *, 125217309Snwhitehorn struct in6_multi *); 126217309Snwhitehornstatic void mld_v1_process_querier_timers(struct mld_ifinfo *); 127217309Snwhitehornstatic int mld_v1_transmit_report(struct in6_multi *, const int); 128217309Snwhitehornstatic void mld_v1_update_group(struct in6_multi *, const int); 129217309Snwhitehornstatic void mld_v2_cancel_link_timers(struct mld_ifinfo *); 130217309Snwhitehornstatic void mld_v2_dispatch_general_query(struct mld_ifinfo *); 131217309Snwhitehornstatic struct mbuf * 132217309Snwhitehorn mld_v2_encap_report(struct ifnet *, struct mbuf *); 133217309Snwhitehornstatic int mld_v2_enqueue_filter_change(struct ifqueue *, 134217309Snwhitehorn struct in6_multi *); 135217309Snwhitehornstatic int mld_v2_enqueue_group_record(struct ifqueue *, 136217309Snwhitehorn struct in6_multi *, const int, const int, const int, 137217309Snwhitehorn const int); 138217309Snwhitehornstatic int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, 139217309Snwhitehorn struct mbuf *, const int, const int); 140217309Snwhitehornstatic int mld_v2_merge_state_changes(struct in6_multi *, 141217309Snwhitehorn struct ifqueue *); 142217309Snwhitehornstatic void mld_v2_process_group_timers(struct mld_ifinfo *, 143217309Snwhitehorn struct ifqueue *, struct ifqueue *, 144217309Snwhitehorn struct in6_multi *, const int); 145217309Snwhitehornstatic int mld_v2_process_group_query(struct in6_multi *, 146217309Snwhitehorn struct mld_ifinfo *mli, int, struct mbuf *, const int); 147217309Snwhitehornstatic int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 148217309Snwhitehornstatic int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); 149217309Snwhitehorn 150217309Snwhitehorn/* 151217309Snwhitehorn * Normative references: RFC 2710, RFC 3590, RFC 3810. 152217309Snwhitehorn * 153217309Snwhitehorn * Locking: 154217309Snwhitehorn * * The MLD subsystem lock ends up being system-wide for the moment, 155217309Snwhitehorn * but could be per-VIMAGE later on. 156217309Snwhitehorn * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 157217309Snwhitehorn * Any may be taken independently; if any are held at the same 158217309Snwhitehorn * time, the above lock order must be followed. 159217309Snwhitehorn * * IN6_MULTI_LOCK covers in_multi. 160217309Snwhitehorn * * MLD_LOCK covers per-link state and any global variables in this file. 161217309Snwhitehorn * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 162217309Snwhitehorn * per-link state iterators. 163217309Snwhitehorn * 164217309Snwhitehorn * XXX LOR PREVENTION 165217309Snwhitehorn * A special case for IPv6 is the in6_setscope() routine. ip6_output() 166217309Snwhitehorn * will not accept an ifp; it wants an embedded scope ID, unlike 167217309Snwhitehorn * ip_output(), which happily takes the ifp given to it. The embedded 168217309Snwhitehorn * scope ID is only used by MLD to select the outgoing interface. 169217309Snwhitehorn * 170217309Snwhitehorn * During interface attach and detach, MLD will take MLD_LOCK *after* 171217309Snwhitehorn * the IF_AFDATA_LOCK. 172217309Snwhitehorn * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call 173217309Snwhitehorn * it with MLD_LOCK held without triggering an LOR. A netisr with indirect 174217309Snwhitehorn * dispatch could work around this, but we'd rather not do that, as it 175217309Snwhitehorn * can introduce other races. 176217309Snwhitehorn * 177217309Snwhitehorn * As such, we exploit the fact that the scope ID is just the interface 178217309Snwhitehorn * index, and embed it in the IPv6 destination address accordingly. 179217309Snwhitehorn * This is potentially NOT VALID for MLDv1 reports, as they 180217309Snwhitehorn * are always sent to the multicast group itself; as MLDv2 181217309Snwhitehorn * reports are always sent to ff02::16, this is not an issue 182217309Snwhitehorn * when MLDv2 is in use. 183217309Snwhitehorn * 184217309Snwhitehorn * This does not however eliminate the LOR when ip6_output() itself 185217309Snwhitehorn * calls in6_setscope() internally whilst MLD_LOCK is held. This will 186217309Snwhitehorn * trigger a LOR warning in WITNESS when the ifnet is detached. 187217309Snwhitehorn * 188217309Snwhitehorn * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given 189217309Snwhitehorn * how it's used across the network stack. Here we're simply exploiting 190217309Snwhitehorn * the fact that MLD runs at a similar layer in the stack to scope6.c. 191217309Snwhitehorn * 192217309Snwhitehorn * VIMAGE: 193217309Snwhitehorn * * Each in6_multi corresponds to an ifp, and each ifp corresponds 194217309Snwhitehorn * to a vnet in ifp->if_vnet. 195217309Snwhitehorn */ 196217309Snwhitehornstatic struct mtx mld_mtx; 197217309SnwhitehornMALLOC_DEFINE(M_MLD, "mld", "mld state"); 198217309Snwhitehorn 199217309Snwhitehorn#define MLD_EMBEDSCOPE(pin6, zoneid) \ 200217309Snwhitehorn if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ 201217309Snwhitehorn IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ 202217309Snwhitehorn (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ 203217309Snwhitehorn 204217309Snwhitehorn/* 205217309Snwhitehorn * VIMAGE-wide globals. 206217309Snwhitehorn */ 207217309Snwhitehornstatic VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0}; 208217309Snwhitehornstatic VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head); 209217309Snwhitehornstatic VNET_DEFINE(int, interface_timers_running6); 210217309Snwhitehornstatic VNET_DEFINE(int, state_change_timers_running6); 211217309Snwhitehornstatic VNET_DEFINE(int, current_state_timers_running6); 212217309Snwhitehorn 213217309Snwhitehorn#define V_mld_gsrdelay VNET(mld_gsrdelay) 214217309Snwhitehorn#define V_mli_head VNET(mli_head) 215217309Snwhitehorn#define V_interface_timers_running6 VNET(interface_timers_running6) 216217309Snwhitehorn#define V_state_change_timers_running6 VNET(state_change_timers_running6) 217217309Snwhitehorn#define V_current_state_timers_running6 VNET(current_state_timers_running6) 218217309Snwhitehorn 219217309SnwhitehornSYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 220217309Snwhitehorn 221217309SnwhitehornSYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 222217309Snwhitehorn "IPv6 Multicast Listener Discovery"); 223217309Snwhitehorn 224217309Snwhitehorn/* 225217309Snwhitehorn * Virtualized sysctls. 226217309Snwhitehorn */ 227217309SnwhitehornSYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, 228251843Sbapt CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 229251843Sbapt &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", 230251843Sbapt "Rate limit for MLDv2 Group-and-Source queries in seconds"); 231251843Sbapt 232251843Sbapt/* 233251843Sbapt * Non-virtualized sysctls. 234217309Snwhitehorn */ 235251843SbaptSYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, 236217309Snwhitehorn sysctl_mld_ifinfo, "Per-interface MLDv2 state"); 237217309Snwhitehorn 238217309Snwhitehornstatic int mld_v1enable = 1; 239217309SnwhitehornSYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW, 240217309Snwhitehorn &mld_v1enable, 0, "Enable fallback to MLDv1"); 241217309SnwhitehornTUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable); 242217309Snwhitehorn 243217309Snwhitehornstatic int mld_use_allow = 1; 244217309SnwhitehornSYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW, 245217309Snwhitehorn &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); 246217309SnwhitehornTUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow); 247217309Snwhitehorn 248217309Snwhitehorn/* 249217309Snwhitehorn * Packed Router Alert option structure declaration. 250217309Snwhitehorn */ 251251843Sbaptstruct mld_raopt { 252217309Snwhitehorn struct ip6_hbh hbh; 253217309Snwhitehorn struct ip6_opt pad; 254217309Snwhitehorn struct ip6_opt_router ra; 255217309Snwhitehorn} __packed; 256217309Snwhitehorn 257217309Snwhitehorn/* 258217309Snwhitehorn * Router Alert hop-by-hop option header. 259217309Snwhitehorn */ 260217309Snwhitehornstatic struct mld_raopt mld_ra = { 261217309Snwhitehorn .hbh = { 0, 0 }, 262251843Sbapt .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 263217309Snwhitehorn .ra = { 264251843Sbapt .ip6or_type = IP6OPT_ROUTER_ALERT, 265217309Snwhitehorn .ip6or_len = IP6OPT_RTALERT_LEN - 2, 266217309Snwhitehorn .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 267217309Snwhitehorn .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 268217309Snwhitehorn } 269217309Snwhitehorn}; 270217309Snwhitehornstatic struct ip6_pktopts mld_po; 271217309Snwhitehorn 272217309Snwhitehornstatic __inline void 273217309Snwhitehornmld_save_context(struct mbuf *m, struct ifnet *ifp) 274217309Snwhitehorn{ 275217309Snwhitehorn 276217309Snwhitehorn#ifdef VIMAGE 277217309Snwhitehorn m->m_pkthdr.header = ifp->if_vnet; 278217309Snwhitehorn#endif /* VIMAGE */ 279217309Snwhitehorn m->m_pkthdr.flowid = ifp->if_index; 280217309Snwhitehorn} 281217309Snwhitehorn 282217309Snwhitehornstatic __inline void 283217309Snwhitehornmld_scrub_context(struct mbuf *m) 284217309Snwhitehorn{ 285217309Snwhitehorn 286217309Snwhitehorn m->m_pkthdr.header = NULL; 287217309Snwhitehorn m->m_pkthdr.flowid = 0; 288217309Snwhitehorn} 289217309Snwhitehorn 290217309Snwhitehorn/* 291217309Snwhitehorn * Restore context from a queued output chain. 292217309Snwhitehorn * Return saved ifindex. 293217309Snwhitehorn * 294217309Snwhitehorn * VIMAGE: The assertion is there to make sure that we 295217309Snwhitehorn * actually called CURVNET_SET() with what's in the mbuf chain. 296217309Snwhitehorn */ 297217309Snwhitehornstatic __inline uint32_t 298217309Snwhitehornmld_restore_context(struct mbuf *m) 299251843Sbapt{ 300217309Snwhitehorn 301251843Sbapt#if defined(VIMAGE) && defined(INVARIANTS) 302251843Sbapt KASSERT(curvnet == m->m_pkthdr.header, 303251843Sbapt ("%s: called when curvnet was not restored", __func__)); 304251843Sbapt#endif 305251843Sbapt return (m->m_pkthdr.flowid); 306251843Sbapt} 307251843Sbapt 308217309Snwhitehorn/* 309217309Snwhitehorn * Retrieve or set threshold between group-source queries in seconds. 310217309Snwhitehorn * 311217309Snwhitehorn * VIMAGE: Assume curvnet set by caller. 312217309Snwhitehorn * SMPng: NOTE: Serialized by MLD lock. 313217309Snwhitehorn */ 314217309Snwhitehornstatic int 315217309Snwhitehornsysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 316217309Snwhitehorn{ 317217309Snwhitehorn int error; 318217309Snwhitehorn int i; 319217309Snwhitehorn 320217309Snwhitehorn error = sysctl_wire_old_buffer(req, sizeof(int)); 321251843Sbapt if (error) 322217309Snwhitehorn return (error); 323251843Sbapt 324251843Sbapt MLD_LOCK(); 325251843Sbapt 326251843Sbapt i = V_mld_gsrdelay.tv_sec; 327251843Sbapt 328251843Sbapt error = sysctl_handle_int(oidp, &i, 0, req); 329251843Sbapt if (error || !req->newptr) 330217309Snwhitehorn goto out_locked; 331217309Snwhitehorn 332217309Snwhitehorn if (i < -1 || i >= 60) { 333217309Snwhitehorn error = EINVAL; 334217309Snwhitehorn goto out_locked; 335217309Snwhitehorn } 336217309Snwhitehorn 337217309Snwhitehorn CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 338217309Snwhitehorn V_mld_gsrdelay.tv_sec, i); 339217309Snwhitehorn V_mld_gsrdelay.tv_sec = i; 340217309Snwhitehorn 341217309Snwhitehornout_locked: 342217309Snwhitehorn MLD_UNLOCK(); 343217309Snwhitehorn return (error); 344217309Snwhitehorn} 345217309Snwhitehorn 346217309Snwhitehorn/* 347217309Snwhitehorn * Expose struct mld_ifinfo to userland, keyed by ifindex. 348217309Snwhitehorn * For use by ifmcstat(8). 349217309Snwhitehorn * 350217309Snwhitehorn * SMPng: NOTE: Does an unlocked ifindex space read. 351217309Snwhitehorn * VIMAGE: Assume curvnet set by caller. The node handler itself 352217309Snwhitehorn * is not directly virtualized. 353217309Snwhitehorn */ 354217309Snwhitehornstatic int 355217309Snwhitehornsysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 356217309Snwhitehorn{ 357217309Snwhitehorn int *name; 358217309Snwhitehorn int error; 359217309Snwhitehorn u_int namelen; 360217309Snwhitehorn struct ifnet *ifp; 361217309Snwhitehorn struct mld_ifinfo *mli; 362217309Snwhitehorn 363217309Snwhitehorn name = (int *)arg1; 364217309Snwhitehorn namelen = arg2; 365217309Snwhitehorn 366217309Snwhitehorn if (req->newptr != NULL) 367217309Snwhitehorn return (EPERM); 368217309Snwhitehorn 369217309Snwhitehorn if (namelen != 1) 370217309Snwhitehorn return (EINVAL); 371217309Snwhitehorn 372217309Snwhitehorn error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); 373217309Snwhitehorn if (error) 374217309Snwhitehorn return (error); 375217309Snwhitehorn 376217309Snwhitehorn IN6_MULTI_LOCK(); 377217309Snwhitehorn MLD_LOCK(); 378217309Snwhitehorn 379217309Snwhitehorn if (name[0] <= 0 || name[0] > V_if_index) { 380217309Snwhitehorn error = ENOENT; 381217309Snwhitehorn goto out_locked; 382217309Snwhitehorn } 383217309Snwhitehorn 384217309Snwhitehorn error = ENOENT; 385217309Snwhitehorn 386217309Snwhitehorn ifp = ifnet_byindex(name[0]); 387217309Snwhitehorn if (ifp == NULL) 388217309Snwhitehorn goto out_locked; 389217309Snwhitehorn 390217309Snwhitehorn LIST_FOREACH(mli, &V_mli_head, mli_link) { 391217309Snwhitehorn if (ifp == mli->mli_ifp) { 392217309Snwhitehorn error = SYSCTL_OUT(req, mli, 393217309Snwhitehorn sizeof(struct mld_ifinfo)); 394217309Snwhitehorn break; 395217309Snwhitehorn } 396217309Snwhitehorn } 397224014Snwhitehorn 398217309Snwhitehornout_locked: 399217309Snwhitehorn MLD_UNLOCK(); 400217309Snwhitehorn IN6_MULTI_UNLOCK(); 401217309Snwhitehorn return (error); 402217309Snwhitehorn} 403217309Snwhitehorn 404217309Snwhitehorn/* 405217309Snwhitehorn * Dispatch an entire queue of pending packet chains. 406217309Snwhitehorn * VIMAGE: Assumes the vnet pointer has been set. 407217309Snwhitehorn */ 408217309Snwhitehornstatic void 409217309Snwhitehornmld_dispatch_queue(struct ifqueue *ifq, int limit) 410217309Snwhitehorn{ 411217309Snwhitehorn struct mbuf *m; 412217309Snwhitehorn 413217309Snwhitehorn for (;;) { 414217309Snwhitehorn _IF_DEQUEUE(ifq, m); 415217309Snwhitehorn if (m == NULL) 416217309Snwhitehorn break; 417217309Snwhitehorn CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m); 418217309Snwhitehorn mld_dispatch_packet(m); 419217309Snwhitehorn if (--limit == 0) 420217309Snwhitehorn break; 421217309Snwhitehorn } 422217309Snwhitehorn} 423217309Snwhitehorn 424217309Snwhitehorn/* 425217309Snwhitehorn * Filter outgoing MLD report state by group. 426217309Snwhitehorn * 427217309Snwhitehorn * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) 428217309Snwhitehorn * and node-local addresses. However, kernel and socket consumers 429217309Snwhitehorn * always embed the KAME scope ID in the address provided, so strip it 430217309Snwhitehorn * when performing comparison. 431217309Snwhitehorn * Note: This is not the same as the *multicast* scope. 432217309Snwhitehorn * 433217309Snwhitehorn * Return zero if the given group is one for which MLD reports 434217309Snwhitehorn * should be suppressed, or non-zero if reports should be issued. 435217309Snwhitehorn */ 436217309Snwhitehornstatic __inline int 437217309Snwhitehornmld_is_addr_reported(const struct in6_addr *addr) 438217309Snwhitehorn{ 439251843Sbapt 440217309Snwhitehorn KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); 441217309Snwhitehorn 442217309Snwhitehorn if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) 443217309Snwhitehorn return (0); 444217309Snwhitehorn 445217309Snwhitehorn if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { 446217309Snwhitehorn struct in6_addr tmp = *addr; 447217309Snwhitehorn in6_clearscope(&tmp); 448217309Snwhitehorn if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) 449217309Snwhitehorn return (0); 450217309Snwhitehorn } 451217309Snwhitehorn 452217309Snwhitehorn return (1); 453217309Snwhitehorn} 454217309Snwhitehorn 455217309Snwhitehorn/* 456217309Snwhitehorn * Attach MLD when PF_INET6 is attached to an interface. 457217309Snwhitehorn * 458217309Snwhitehorn * SMPng: Normally called with IF_AFDATA_LOCK held. 459217309Snwhitehorn */ 460217309Snwhitehornstruct mld_ifinfo * 461217309Snwhitehornmld_domifattach(struct ifnet *ifp) 462217309Snwhitehorn{ 463217309Snwhitehorn struct mld_ifinfo *mli; 464217309Snwhitehorn 465217309Snwhitehorn CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 466217309Snwhitehorn __func__, ifp, ifp->if_xname); 467217309Snwhitehorn 468217309Snwhitehorn MLD_LOCK(); 469217309Snwhitehorn 470217309Snwhitehorn mli = mli_alloc_locked(ifp); 471217309Snwhitehorn if (!(ifp->if_flags & IFF_MULTICAST)) 472217309Snwhitehorn mli->mli_flags |= MLIF_SILENT; 473217309Snwhitehorn if (mld_use_allow) 474217309Snwhitehorn mli->mli_flags |= MLIF_USEALLOW; 475217309Snwhitehorn 476217309Snwhitehorn MLD_UNLOCK(); 477217309Snwhitehorn 478217309Snwhitehorn return (mli); 479217309Snwhitehorn} 480217309Snwhitehorn 481217309Snwhitehorn/* 482217309Snwhitehorn * VIMAGE: assume curvnet set by caller. 483217309Snwhitehorn */ 484217309Snwhitehornstatic struct mld_ifinfo * 485217309Snwhitehornmli_alloc_locked(/*const*/ struct ifnet *ifp) 486217309Snwhitehorn{ 487217309Snwhitehorn struct mld_ifinfo *mli; 488217309Snwhitehorn 489217309Snwhitehorn MLD_LOCK_ASSERT(); 490217309Snwhitehorn 491217309Snwhitehorn mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO); 492217309Snwhitehorn if (mli == NULL) 493217309Snwhitehorn goto out; 494217309Snwhitehorn 495217309Snwhitehorn mli->mli_ifp = ifp; 496217309Snwhitehorn mli->mli_version = MLD_VERSION_2; 497217309Snwhitehorn mli->mli_flags = 0; 498217309Snwhitehorn mli->mli_rv = MLD_RV_INIT; 499217309Snwhitehorn mli->mli_qi = MLD_QI_INIT; 500251843Sbapt mli->mli_qri = MLD_QRI_INIT; 501251843Sbapt mli->mli_uri = MLD_URI_INIT; 502217309Snwhitehorn 503217309Snwhitehorn SLIST_INIT(&mli->mli_relinmhead); 504251843Sbapt 505217309Snwhitehorn /* 506217309Snwhitehorn * Responses to general queries are subject to bounds. 507217309Snwhitehorn */ 508217309Snwhitehorn IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 509217309Snwhitehorn 510217309Snwhitehorn LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 511217309Snwhitehorn 512217309Snwhitehorn CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)", 513217309Snwhitehorn ifp, ifp->if_xname); 514217309Snwhitehorn 515217309Snwhitehornout: 516217309Snwhitehorn return (mli); 517217309Snwhitehorn} 518217309Snwhitehorn 519217309Snwhitehorn/* 520217309Snwhitehorn * Hook for ifdetach. 521217309Snwhitehorn * 522217309Snwhitehorn * NOTE: Some finalization tasks need to run before the protocol domain 523217309Snwhitehorn * is detached, but also before the link layer does its cleanup. 524217309Snwhitehorn * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 525217309Snwhitehorn * 526217309Snwhitehorn * SMPng: Caller must hold IN6_MULTI_LOCK(). 527217309Snwhitehorn * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 528217309Snwhitehorn * XXX This routine is also bitten by unlocked ifma_protospec access. 529217309Snwhitehorn */ 530217309Snwhitehornvoid 531217309Snwhitehornmld_ifdetach(struct ifnet *ifp) 532217309Snwhitehorn{ 533217309Snwhitehorn struct mld_ifinfo *mli; 534217309Snwhitehorn struct ifmultiaddr *ifma; 535217309Snwhitehorn struct in6_multi *inm, *tinm; 536217309Snwhitehorn 537217309Snwhitehorn CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 538217309Snwhitehorn ifp->if_xname); 539217309Snwhitehorn 540217309Snwhitehorn IN6_MULTI_LOCK_ASSERT(); 541217309Snwhitehorn MLD_LOCK(); 542217309Snwhitehorn 543217309Snwhitehorn mli = MLD_IFINFO(ifp); 544217309Snwhitehorn if (mli->mli_version == MLD_VERSION_2) { 545217309Snwhitehorn IF_ADDR_RLOCK(ifp); 546217309Snwhitehorn TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 547217309Snwhitehorn if (ifma->ifma_addr->sa_family != AF_INET6 || 548251843Sbapt ifma->ifma_protospec == NULL) 549217309Snwhitehorn continue; 550217309Snwhitehorn inm = (struct in6_multi *)ifma->ifma_protospec; 551217309Snwhitehorn if (inm->in6m_state == MLD_LEAVING_MEMBER) { 552217309Snwhitehorn SLIST_INSERT_HEAD(&mli->mli_relinmhead, 553217309Snwhitehorn inm, in6m_nrele); 554217309Snwhitehorn } 555217309Snwhitehorn in6m_clear_recorded(inm); 556217309Snwhitehorn } 557217309Snwhitehorn IF_ADDR_RUNLOCK(ifp); 558217309Snwhitehorn SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, 559217309Snwhitehorn tinm) { 560217309Snwhitehorn SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 561217309Snwhitehorn in6m_release_locked(inm); 562217309Snwhitehorn } 563217309Snwhitehorn } 564217309Snwhitehorn 565217309Snwhitehorn MLD_UNLOCK(); 566217309Snwhitehorn} 567217309Snwhitehorn 568217309Snwhitehorn/* 569217309Snwhitehorn * Hook for domifdetach. 570217309Snwhitehorn * Runs after link-layer cleanup; free MLD state. 571217309Snwhitehorn * 572217309Snwhitehorn * SMPng: Normally called with IF_AFDATA_LOCK held. 573217309Snwhitehorn */ 574217309Snwhitehornvoid 575217309Snwhitehornmld_domifdetach(struct ifnet *ifp) 576217309Snwhitehorn{ 577217309Snwhitehorn 578217309Snwhitehorn CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 579217309Snwhitehorn __func__, ifp, ifp->if_xname); 580217309Snwhitehorn 581217309Snwhitehorn MLD_LOCK(); 582224014Snwhitehorn mli_delete_locked(ifp); 583217309Snwhitehorn MLD_UNLOCK(); 584217309Snwhitehorn} 585217309Snwhitehorn 586217309Snwhitehornstatic void 587217309Snwhitehornmli_delete_locked(const struct ifnet *ifp) 588217309Snwhitehorn{ 589217309Snwhitehorn struct mld_ifinfo *mli, *tmli; 590217309Snwhitehorn 591217309Snwhitehorn CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)", 592217309Snwhitehorn __func__, ifp, ifp->if_xname); 593217309Snwhitehorn 594217309Snwhitehorn MLD_LOCK_ASSERT(); 595217309Snwhitehorn 596217309Snwhitehorn LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { 597217309Snwhitehorn if (mli->mli_ifp == ifp) { 598217309Snwhitehorn /* 599217309Snwhitehorn * Free deferred General Query responses. 600217309Snwhitehorn */ 601217309Snwhitehorn _IF_DRAIN(&mli->mli_gq); 602217309Snwhitehorn 603217309Snwhitehorn LIST_REMOVE(mli, mli_link); 604217309Snwhitehorn 605217309Snwhitehorn KASSERT(SLIST_EMPTY(&mli->mli_relinmhead), 606217309Snwhitehorn ("%s: there are dangling in_multi references", 607217309Snwhitehorn __func__)); 608217309Snwhitehorn 609217309Snwhitehorn free(mli, M_MLD); 610217309Snwhitehorn return; 611217309Snwhitehorn } 612217309Snwhitehorn } 613217309Snwhitehorn#ifdef INVARIANTS 614217309Snwhitehorn panic("%s: mld_ifinfo not found for ifp %p\n", __func__, ifp); 615217309Snwhitehorn#endif 616217309Snwhitehorn} 617217309Snwhitehorn 618217309Snwhitehorn/* 619217309Snwhitehorn * Process a received MLDv1 general or address-specific query. 620217309Snwhitehorn * Assumes that the query header has been pulled up to sizeof(mld_hdr). 621217309Snwhitehorn * 622217309Snwhitehorn * NOTE: Can't be fully const correct as we temporarily embed scope ID in 623217309Snwhitehorn * mld_addr. This is OK as we own the mbuf chain. 624217309Snwhitehorn */ 625217309Snwhitehornstatic int 626217309Snwhitehornmld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 627217309Snwhitehorn /*const*/ struct mld_hdr *mld) 628217309Snwhitehorn{ 629217309Snwhitehorn struct ifmultiaddr *ifma; 630217309Snwhitehorn struct mld_ifinfo *mli; 631217309Snwhitehorn struct in6_multi *inm; 632217309Snwhitehorn int is_general_query; 633217309Snwhitehorn uint16_t timer; 634217309Snwhitehorn#ifdef KTR 635217309Snwhitehorn char ip6tbuf[INET6_ADDRSTRLEN]; 636217309Snwhitehorn#endif 637217309Snwhitehorn 638217309Snwhitehorn is_general_query = 0; 639217309Snwhitehorn 640217309Snwhitehorn if (!mld_v1enable) { 641217309Snwhitehorn CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", 642217309Snwhitehorn ip6_sprintf(ip6tbuf, &mld->mld_addr), 643217309Snwhitehorn ifp, ifp->if_xname); 644217309Snwhitehorn return (0); 645217309Snwhitehorn } 646217309Snwhitehorn 647217309Snwhitehorn /* 648217309Snwhitehorn * RFC3810 Section 6.2: MLD queries must originate from 649217309Snwhitehorn * a router's link-local address. 650217309Snwhitehorn */ 651217309Snwhitehorn if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 652217309Snwhitehorn CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 653217309Snwhitehorn ip6_sprintf(ip6tbuf, &ip6->ip6_src), 654217309Snwhitehorn ifp, ifp->if_xname); 655217309Snwhitehorn return (0); 656217309Snwhitehorn } 657217309Snwhitehorn 658217309Snwhitehorn /* 659217309Snwhitehorn * Do address field validation upfront before we accept 660217309Snwhitehorn * the query. 661217309Snwhitehorn */ 662217309Snwhitehorn if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 663217309Snwhitehorn /* 664217309Snwhitehorn * MLDv1 General Query. 665217309Snwhitehorn * If this was not sent to the all-nodes group, ignore it. 666217309Snwhitehorn */ 667217309Snwhitehorn struct in6_addr dst; 668217309Snwhitehorn 669217309Snwhitehorn dst = ip6->ip6_dst; 670217309Snwhitehorn in6_clearscope(&dst); 671217309Snwhitehorn if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) 672220749Snwhitehorn return (EINVAL); 673220749Snwhitehorn is_general_query = 1; 674220749Snwhitehorn } else { 675220749Snwhitehorn /* 676217309Snwhitehorn * Embed scope ID of receiving interface in MLD query for 677217309Snwhitehorn * lookup whilst we don't hold other locks. 678217309Snwhitehorn */ 679217309Snwhitehorn in6_setscope(&mld->mld_addr, ifp, NULL); 680217309Snwhitehorn } 681217309Snwhitehorn 682217309Snwhitehorn IN6_MULTI_LOCK(); 683217309Snwhitehorn MLD_LOCK(); 684251843Sbapt 685217309Snwhitehorn /* 686217309Snwhitehorn * Switch to MLDv1 host compatibility mode. 687217309Snwhitehorn */ 688 mli = MLD_IFINFO(ifp); 689 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 690 mld_set_version(mli, MLD_VERSION_1); 691 692 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 693 if (timer == 0) 694 timer = 1; 695 696 IF_ADDR_RLOCK(ifp); 697 if (is_general_query) { 698 /* 699 * For each reporting group joined on this 700 * interface, kick the report timer. 701 */ 702 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", 703 ifp, ifp->if_xname); 704 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 705 if (ifma->ifma_addr->sa_family != AF_INET6 || 706 ifma->ifma_protospec == NULL) 707 continue; 708 inm = (struct in6_multi *)ifma->ifma_protospec; 709 mld_v1_update_group(inm, timer); 710 } 711 } else { 712 /* 713 * MLDv1 Group-Specific Query. 714 * If this is a group-specific MLDv1 query, we need only 715 * look up the single group to process it. 716 */ 717 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 718 if (inm != NULL) { 719 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", 720 ip6_sprintf(ip6tbuf, &mld->mld_addr), 721 ifp, ifp->if_xname); 722 mld_v1_update_group(inm, timer); 723 } 724 /* XXX Clear embedded scope ID as userland won't expect it. */ 725 in6_clearscope(&mld->mld_addr); 726 } 727 728 IF_ADDR_RUNLOCK(ifp); 729 MLD_UNLOCK(); 730 IN6_MULTI_UNLOCK(); 731 732 return (0); 733} 734 735/* 736 * Update the report timer on a group in response to an MLDv1 query. 737 * 738 * If we are becoming the reporting member for this group, start the timer. 739 * If we already are the reporting member for this group, and timer is 740 * below the threshold, reset it. 741 * 742 * We may be updating the group for the first time since we switched 743 * to MLDv2. If we are, then we must clear any recorded source lists, 744 * and transition to REPORTING state; the group timer is overloaded 745 * for group and group-source query responses. 746 * 747 * Unlike MLDv2, the delay per group should be jittered 748 * to avoid bursts of MLDv1 reports. 749 */ 750static void 751mld_v1_update_group(struct in6_multi *inm, const int timer) 752{ 753#ifdef KTR 754 char ip6tbuf[INET6_ADDRSTRLEN]; 755#endif 756 757 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 758 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 759 inm->in6m_ifp->if_xname, timer); 760 761 IN6_MULTI_LOCK_ASSERT(); 762 763 switch (inm->in6m_state) { 764 case MLD_NOT_MEMBER: 765 case MLD_SILENT_MEMBER: 766 break; 767 case MLD_REPORTING_MEMBER: 768 if (inm->in6m_timer != 0 && 769 inm->in6m_timer <= timer) { 770 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 771 "skipping.", __func__); 772 break; 773 } 774 /* FALLTHROUGH */ 775 case MLD_SG_QUERY_PENDING_MEMBER: 776 case MLD_G_QUERY_PENDING_MEMBER: 777 case MLD_IDLE_MEMBER: 778 case MLD_LAZY_MEMBER: 779 case MLD_AWAKENING_MEMBER: 780 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 781 inm->in6m_state = MLD_REPORTING_MEMBER; 782 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 783 V_current_state_timers_running6 = 1; 784 break; 785 case MLD_SLEEPING_MEMBER: 786 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 787 inm->in6m_state = MLD_AWAKENING_MEMBER; 788 break; 789 case MLD_LEAVING_MEMBER: 790 break; 791 } 792} 793 794/* 795 * Process a received MLDv2 general, group-specific or 796 * group-and-source-specific query. 797 * 798 * Assumes that the query header has been pulled up to sizeof(mldv2_query). 799 * 800 * Return 0 if successful, otherwise an appropriate error code is returned. 801 */ 802static int 803mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 804 struct mbuf *m, const int off, const int icmp6len) 805{ 806 struct mld_ifinfo *mli; 807 struct mldv2_query *mld; 808 struct in6_multi *inm; 809 uint32_t maxdelay, nsrc, qqi; 810 int is_general_query; 811 uint16_t timer; 812 uint8_t qrv; 813#ifdef KTR 814 char ip6tbuf[INET6_ADDRSTRLEN]; 815#endif 816 817 is_general_query = 0; 818 819 /* 820 * RFC3810 Section 6.2: MLD queries must originate from 821 * a router's link-local address. 822 */ 823 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 824 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 825 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 826 ifp, ifp->if_xname); 827 return (0); 828 } 829 830 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname); 831 832 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); 833 834 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 835 if (maxdelay >= 32768) { 836 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 837 (MLD_MRC_EXP(maxdelay) + 3); 838 } 839 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 840 if (timer == 0) 841 timer = 1; 842 843 qrv = MLD_QRV(mld->mld_misc); 844 if (qrv < 2) { 845 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 846 qrv, MLD_RV_INIT); 847 qrv = MLD_RV_INIT; 848 } 849 850 qqi = mld->mld_qqi; 851 if (qqi >= 128) { 852 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 853 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 854 } 855 856 nsrc = ntohs(mld->mld_numsrc); 857 if (nsrc > MLD_MAX_GS_SOURCES) 858 return (EMSGSIZE); 859 if (icmp6len < sizeof(struct mldv2_query) + 860 (nsrc * sizeof(struct in6_addr))) 861 return (EMSGSIZE); 862 863 /* 864 * Do further input validation upfront to avoid resetting timers 865 * should we need to discard this query. 866 */ 867 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 868 /* 869 * A general query with a source list has undefined 870 * behaviour; discard it. 871 */ 872 if (nsrc > 0) 873 return (EINVAL); 874 is_general_query = 1; 875 } else { 876 /* 877 * Embed scope ID of receiving interface in MLD query for 878 * lookup whilst we don't hold other locks (due to KAME 879 * locking lameness). We own this mbuf chain just now. 880 */ 881 in6_setscope(&mld->mld_addr, ifp, NULL); 882 } 883 884 IN6_MULTI_LOCK(); 885 MLD_LOCK(); 886 887 mli = MLD_IFINFO(ifp); 888 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 889 890 /* 891 * Discard the v2 query if we're in Compatibility Mode. 892 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 893 * until the Old Version Querier Present timer expires. 894 */ 895 if (mli->mli_version != MLD_VERSION_2) 896 goto out_locked; 897 898 mld_set_version(mli, MLD_VERSION_2); 899 mli->mli_rv = qrv; 900 mli->mli_qi = qqi; 901 mli->mli_qri = maxdelay; 902 903 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 904 maxdelay); 905 906 if (is_general_query) { 907 /* 908 * MLDv2 General Query. 909 * 910 * Schedule a current-state report on this ifp for 911 * all groups, possibly containing source lists. 912 * 913 * If there is a pending General Query response 914 * scheduled earlier than the selected delay, do 915 * not schedule any other reports. 916 * Otherwise, reset the interface timer. 917 */ 918 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 919 ifp, ifp->if_xname); 920 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 921 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 922 V_interface_timers_running6 = 1; 923 } 924 } else { 925 /* 926 * MLDv2 Group-specific or Group-and-source-specific Query. 927 * 928 * Group-source-specific queries are throttled on 929 * a per-group basis to defeat denial-of-service attempts. 930 * Queries for groups we are not a member of on this 931 * link are simply ignored. 932 */ 933 IF_ADDR_RLOCK(ifp); 934 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 935 if (inm == NULL) { 936 IF_ADDR_RUNLOCK(ifp); 937 goto out_locked; 938 } 939 if (nsrc > 0) { 940 if (!ratecheck(&inm->in6m_lastgsrtv, 941 &V_mld_gsrdelay)) { 942 CTR1(KTR_MLD, "%s: GS query throttled.", 943 __func__); 944 IF_ADDR_RUNLOCK(ifp); 945 goto out_locked; 946 } 947 } 948 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 949 ifp, ifp->if_xname); 950 /* 951 * If there is a pending General Query response 952 * scheduled sooner than the selected delay, no 953 * further report need be scheduled. 954 * Otherwise, prepare to respond to the 955 * group-specific or group-and-source query. 956 */ 957 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 958 mld_v2_process_group_query(inm, mli, timer, m, off); 959 960 /* XXX Clear embedded scope ID as userland won't expect it. */ 961 in6_clearscope(&mld->mld_addr); 962 IF_ADDR_RUNLOCK(ifp); 963 } 964 965out_locked: 966 MLD_UNLOCK(); 967 IN6_MULTI_UNLOCK(); 968 969 return (0); 970} 971 972/* 973 * Process a recieved MLDv2 group-specific or group-and-source-specific 974 * query. 975 * Return <0 if any error occured. Currently this is ignored. 976 */ 977static int 978mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli, 979 int timer, struct mbuf *m0, const int off) 980{ 981 struct mldv2_query *mld; 982 int retval; 983 uint16_t nsrc; 984 985 IN6_MULTI_LOCK_ASSERT(); 986 MLD_LOCK_ASSERT(); 987 988 retval = 0; 989 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); 990 991 switch (inm->in6m_state) { 992 case MLD_NOT_MEMBER: 993 case MLD_SILENT_MEMBER: 994 case MLD_SLEEPING_MEMBER: 995 case MLD_LAZY_MEMBER: 996 case MLD_AWAKENING_MEMBER: 997 case MLD_IDLE_MEMBER: 998 case MLD_LEAVING_MEMBER: 999 return (retval); 1000 break; 1001 case MLD_REPORTING_MEMBER: 1002 case MLD_G_QUERY_PENDING_MEMBER: 1003 case MLD_SG_QUERY_PENDING_MEMBER: 1004 break; 1005 } 1006 1007 nsrc = ntohs(mld->mld_numsrc); 1008 1009 /* 1010 * Deal with group-specific queries upfront. 1011 * If any group query is already pending, purge any recorded 1012 * source-list state if it exists, and schedule a query response 1013 * for this group-specific query. 1014 */ 1015 if (nsrc == 0) { 1016 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1017 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1018 in6m_clear_recorded(inm); 1019 timer = min(inm->in6m_timer, timer); 1020 } 1021 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1022 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1023 V_current_state_timers_running6 = 1; 1024 return (retval); 1025 } 1026 1027 /* 1028 * Deal with the case where a group-and-source-specific query has 1029 * been received but a group-specific query is already pending. 1030 */ 1031 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1032 timer = min(inm->in6m_timer, timer); 1033 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1034 V_current_state_timers_running6 = 1; 1035 return (retval); 1036 } 1037 1038 /* 1039 * Finally, deal with the case where a group-and-source-specific 1040 * query has been received, where a response to a previous g-s-r 1041 * query exists, or none exists. 1042 * In this case, we need to parse the source-list which the Querier 1043 * has provided us with and check if we have any source list filter 1044 * entries at T1 for these sources. If we do not, there is no need 1045 * schedule a report and the query may be dropped. 1046 * If we do, we must record them and schedule a current-state 1047 * report for those sources. 1048 */ 1049 if (inm->in6m_nsrc > 0) { 1050 struct mbuf *m; 1051 uint8_t *sp; 1052 int i, nrecorded; 1053 int soff; 1054 1055 m = m0; 1056 soff = off + sizeof(struct mldv2_query); 1057 nrecorded = 0; 1058 for (i = 0; i < nsrc; i++) { 1059 sp = mtod(m, uint8_t *) + soff; 1060 retval = in6m_record_source(inm, 1061 (const struct in6_addr *)sp); 1062 if (retval < 0) 1063 break; 1064 nrecorded += retval; 1065 soff += sizeof(struct in6_addr); 1066 if (soff >= m->m_len) { 1067 soff = soff - m->m_len; 1068 m = m->m_next; 1069 if (m == NULL) 1070 break; 1071 } 1072 } 1073 if (nrecorded > 0) { 1074 CTR1(KTR_MLD, 1075 "%s: schedule response to SG query", __func__); 1076 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1077 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1078 V_current_state_timers_running6 = 1; 1079 } 1080 } 1081 1082 return (retval); 1083} 1084 1085/* 1086 * Process a received MLDv1 host membership report. 1087 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1088 * 1089 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1090 * mld_addr. This is OK as we own the mbuf chain. 1091 */ 1092static int 1093mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1094 /*const*/ struct mld_hdr *mld) 1095{ 1096 struct in6_addr src, dst; 1097 struct in6_ifaddr *ia; 1098 struct in6_multi *inm; 1099#ifdef KTR 1100 char ip6tbuf[INET6_ADDRSTRLEN]; 1101#endif 1102 1103 if (!mld_v1enable) { 1104 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1105 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1106 ifp, ifp->if_xname); 1107 return (0); 1108 } 1109 1110 if (ifp->if_flags & IFF_LOOPBACK) 1111 return (0); 1112 1113 /* 1114 * MLDv1 reports must originate from a host's link-local address, 1115 * or the unspecified address (when booting). 1116 */ 1117 src = ip6->ip6_src; 1118 in6_clearscope(&src); 1119 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1120 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1121 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1122 ifp, ifp->if_xname); 1123 return (EINVAL); 1124 } 1125 1126 /* 1127 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1128 * group, and must be directed to the group itself. 1129 */ 1130 dst = ip6->ip6_dst; 1131 in6_clearscope(&dst); 1132 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1133 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1134 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1135 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1136 ifp, ifp->if_xname); 1137 return (EINVAL); 1138 } 1139 1140 /* 1141 * Make sure we don't hear our own membership report, as fast 1142 * leave requires knowing that we are the only member of a 1143 * group. Assume we used the link-local address if available, 1144 * otherwise look for ::. 1145 * 1146 * XXX Note that scope ID comparison is needed for the address 1147 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1148 * performed for the on-wire address. 1149 */ 1150 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1151 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1152 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { 1153 if (ia != NULL) 1154 ifa_free(&ia->ia_ifa); 1155 return (0); 1156 } 1157 if (ia != NULL) 1158 ifa_free(&ia->ia_ifa); 1159 1160 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1161 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname); 1162 1163 /* 1164 * Embed scope ID of receiving interface in MLD query for lookup 1165 * whilst we don't hold other locks (due to KAME locking lameness). 1166 */ 1167 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1168 in6_setscope(&mld->mld_addr, ifp, NULL); 1169 1170 IN6_MULTI_LOCK(); 1171 MLD_LOCK(); 1172 IF_ADDR_RLOCK(ifp); 1173 1174 /* 1175 * MLDv1 report suppression. 1176 * If we are a member of this group, and our membership should be 1177 * reported, and our group timer is pending or about to be reset, 1178 * stop our group timer by transitioning to the 'lazy' state. 1179 */ 1180 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1181 if (inm != NULL) { 1182 struct mld_ifinfo *mli; 1183 1184 mli = inm->in6m_mli; 1185 KASSERT(mli != NULL, 1186 ("%s: no mli for ifp %p", __func__, ifp)); 1187 1188 /* 1189 * If we are in MLDv2 host mode, do not allow the 1190 * other host's MLDv1 report to suppress our reports. 1191 */ 1192 if (mli->mli_version == MLD_VERSION_2) 1193 goto out_locked; 1194 1195 inm->in6m_timer = 0; 1196 1197 switch (inm->in6m_state) { 1198 case MLD_NOT_MEMBER: 1199 case MLD_SILENT_MEMBER: 1200 case MLD_SLEEPING_MEMBER: 1201 break; 1202 case MLD_REPORTING_MEMBER: 1203 case MLD_IDLE_MEMBER: 1204 case MLD_AWAKENING_MEMBER: 1205 CTR3(KTR_MLD, 1206 "report suppressed for %s on ifp %p(%s)", 1207 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1208 ifp, ifp->if_xname); 1209 case MLD_LAZY_MEMBER: 1210 inm->in6m_state = MLD_LAZY_MEMBER; 1211 break; 1212 case MLD_G_QUERY_PENDING_MEMBER: 1213 case MLD_SG_QUERY_PENDING_MEMBER: 1214 case MLD_LEAVING_MEMBER: 1215 break; 1216 } 1217 } 1218 1219out_locked: 1220 IF_ADDR_RUNLOCK(ifp); 1221 MLD_UNLOCK(); 1222 IN6_MULTI_UNLOCK(); 1223 1224 /* XXX Clear embedded scope ID as userland won't expect it. */ 1225 in6_clearscope(&mld->mld_addr); 1226 1227 return (0); 1228} 1229 1230/* 1231 * MLD input path. 1232 * 1233 * Assume query messages which fit in a single ICMPv6 message header 1234 * have been pulled up. 1235 * Assume that userland will want to see the message, even if it 1236 * otherwise fails kernel input validation; do not free it. 1237 * Pullup may however free the mbuf chain m if it fails. 1238 * 1239 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1240 */ 1241int 1242mld_input(struct mbuf *m, int off, int icmp6len) 1243{ 1244 struct ifnet *ifp; 1245 struct ip6_hdr *ip6; 1246 struct mld_hdr *mld; 1247 int mldlen; 1248 1249 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1250 1251 ifp = m->m_pkthdr.rcvif; 1252 1253 ip6 = mtod(m, struct ip6_hdr *); 1254 1255 /* Pullup to appropriate size. */ 1256 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1257 if (mld->mld_type == MLD_LISTENER_QUERY && 1258 icmp6len >= sizeof(struct mldv2_query)) { 1259 mldlen = sizeof(struct mldv2_query); 1260 } else { 1261 mldlen = sizeof(struct mld_hdr); 1262 } 1263 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1264 if (mld == NULL) { 1265 ICMP6STAT_INC(icp6s_badlen); 1266 return (IPPROTO_DONE); 1267 } 1268 1269 /* 1270 * Userland needs to see all of this traffic for implementing 1271 * the endpoint discovery portion of multicast routing. 1272 */ 1273 switch (mld->mld_type) { 1274 case MLD_LISTENER_QUERY: 1275 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1276 if (icmp6len == sizeof(struct mld_hdr)) { 1277 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1278 return (0); 1279 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1280 if (mld_v2_input_query(ifp, ip6, m, off, 1281 icmp6len) != 0) 1282 return (0); 1283 } 1284 break; 1285 case MLD_LISTENER_REPORT: 1286 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1287 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1288 return (0); 1289 break; 1290 case MLDV2_LISTENER_REPORT: 1291 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1292 break; 1293 case MLD_LISTENER_DONE: 1294 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1295 break; 1296 default: 1297 break; 1298 } 1299 1300 return (0); 1301} 1302 1303/* 1304 * Fast timeout handler (global). 1305 * VIMAGE: Timeout handlers are expected to service all vimages. 1306 */ 1307void 1308mld_fasttimo(void) 1309{ 1310 VNET_ITERATOR_DECL(vnet_iter); 1311 1312 VNET_LIST_RLOCK_NOSLEEP(); 1313 VNET_FOREACH(vnet_iter) { 1314 CURVNET_SET(vnet_iter); 1315 mld_fasttimo_vnet(); 1316 CURVNET_RESTORE(); 1317 } 1318 VNET_LIST_RUNLOCK_NOSLEEP(); 1319} 1320 1321/* 1322 * Fast timeout handler (per-vnet). 1323 * 1324 * VIMAGE: Assume caller has set up our curvnet. 1325 */ 1326static void 1327mld_fasttimo_vnet(void) 1328{ 1329 struct ifqueue scq; /* State-change packets */ 1330 struct ifqueue qrq; /* Query response packets */ 1331 struct ifnet *ifp; 1332 struct mld_ifinfo *mli; 1333 struct ifmultiaddr *ifma; 1334 struct in6_multi *inm, *tinm; 1335 int uri_fasthz; 1336 1337 uri_fasthz = 0; 1338 1339 /* 1340 * Quick check to see if any work needs to be done, in order to 1341 * minimize the overhead of fasttimo processing. 1342 * SMPng: XXX Unlocked reads. 1343 */ 1344 if (!V_current_state_timers_running6 && 1345 !V_interface_timers_running6 && 1346 !V_state_change_timers_running6) 1347 return; 1348 1349 IN6_MULTI_LOCK(); 1350 MLD_LOCK(); 1351 1352 /* 1353 * MLDv2 General Query response timer processing. 1354 */ 1355 if (V_interface_timers_running6) { 1356 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1357 1358 V_interface_timers_running6 = 0; 1359 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1360 if (mli->mli_v2_timer == 0) { 1361 /* Do nothing. */ 1362 } else if (--mli->mli_v2_timer == 0) { 1363 mld_v2_dispatch_general_query(mli); 1364 } else { 1365 V_interface_timers_running6 = 1; 1366 } 1367 } 1368 } 1369 1370 if (!V_current_state_timers_running6 && 1371 !V_state_change_timers_running6) 1372 goto out_locked; 1373 1374 V_current_state_timers_running6 = 0; 1375 V_state_change_timers_running6 = 0; 1376 1377 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1378 1379 /* 1380 * MLD host report and state-change timer processing. 1381 * Note: Processing a v2 group timer may remove a node. 1382 */ 1383 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1384 ifp = mli->mli_ifp; 1385 1386 if (mli->mli_version == MLD_VERSION_2) { 1387 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1388 PR_FASTHZ); 1389 1390 memset(&qrq, 0, sizeof(struct ifqueue)); 1391 IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS); 1392 1393 memset(&scq, 0, sizeof(struct ifqueue)); 1394 IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1395 } 1396 1397 IF_ADDR_RLOCK(ifp); 1398 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1399 if (ifma->ifma_addr->sa_family != AF_INET6 || 1400 ifma->ifma_protospec == NULL) 1401 continue; 1402 inm = (struct in6_multi *)ifma->ifma_protospec; 1403 switch (mli->mli_version) { 1404 case MLD_VERSION_1: 1405 mld_v1_process_group_timer(mli, inm); 1406 break; 1407 case MLD_VERSION_2: 1408 mld_v2_process_group_timers(mli, &qrq, 1409 &scq, inm, uri_fasthz); 1410 break; 1411 } 1412 } 1413 IF_ADDR_RUNLOCK(ifp); 1414 1415 switch (mli->mli_version) { 1416 case MLD_VERSION_1: 1417 /* 1418 * Transmit reports for this lifecycle. This 1419 * is done while not holding IF_ADDR_LOCK 1420 * since this can call 1421 * in6ifa_ifpforlinklocal() which locks 1422 * IF_ADDR_LOCK internally as well as 1423 * ip6_output() to transmit a packet. 1424 */ 1425 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1426 in6m_nrele, tinm) { 1427 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1428 in6m_nrele); 1429 (void)mld_v1_transmit_report(inm, 1430 MLD_LISTENER_REPORT); 1431 } 1432 break; 1433 case MLD_VERSION_2: 1434 mld_dispatch_queue(&qrq, 0); 1435 mld_dispatch_queue(&scq, 0); 1436 1437 /* 1438 * Free the in_multi reference(s) for 1439 * this lifecycle. 1440 */ 1441 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1442 in6m_nrele, tinm) { 1443 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1444 in6m_nrele); 1445 in6m_release_locked(inm); 1446 } 1447 break; 1448 } 1449 } 1450 1451out_locked: 1452 MLD_UNLOCK(); 1453 IN6_MULTI_UNLOCK(); 1454} 1455 1456/* 1457 * Update host report group timer. 1458 * Will update the global pending timer flags. 1459 */ 1460static void 1461mld_v1_process_group_timer(struct mld_ifinfo *mli, struct in6_multi *inm) 1462{ 1463 int report_timer_expired; 1464 1465 IN6_MULTI_LOCK_ASSERT(); 1466 MLD_LOCK_ASSERT(); 1467 1468 if (inm->in6m_timer == 0) { 1469 report_timer_expired = 0; 1470 } else if (--inm->in6m_timer == 0) { 1471 report_timer_expired = 1; 1472 } else { 1473 V_current_state_timers_running6 = 1; 1474 return; 1475 } 1476 1477 switch (inm->in6m_state) { 1478 case MLD_NOT_MEMBER: 1479 case MLD_SILENT_MEMBER: 1480 case MLD_IDLE_MEMBER: 1481 case MLD_LAZY_MEMBER: 1482 case MLD_SLEEPING_MEMBER: 1483 case MLD_AWAKENING_MEMBER: 1484 break; 1485 case MLD_REPORTING_MEMBER: 1486 if (report_timer_expired) { 1487 inm->in6m_state = MLD_IDLE_MEMBER; 1488 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1489 in6m_nrele); 1490 } 1491 break; 1492 case MLD_G_QUERY_PENDING_MEMBER: 1493 case MLD_SG_QUERY_PENDING_MEMBER: 1494 case MLD_LEAVING_MEMBER: 1495 break; 1496 } 1497} 1498 1499/* 1500 * Update a group's timers for MLDv2. 1501 * Will update the global pending timer flags. 1502 * Note: Unlocked read from mli. 1503 */ 1504static void 1505mld_v2_process_group_timers(struct mld_ifinfo *mli, 1506 struct ifqueue *qrq, struct ifqueue *scq, 1507 struct in6_multi *inm, const int uri_fasthz) 1508{ 1509 int query_response_timer_expired; 1510 int state_change_retransmit_timer_expired; 1511#ifdef KTR 1512 char ip6tbuf[INET6_ADDRSTRLEN]; 1513#endif 1514 1515 IN6_MULTI_LOCK_ASSERT(); 1516 MLD_LOCK_ASSERT(); 1517 1518 query_response_timer_expired = 0; 1519 state_change_retransmit_timer_expired = 0; 1520 1521 /* 1522 * During a transition from compatibility mode back to MLDv2, 1523 * a group record in REPORTING state may still have its group 1524 * timer active. This is a no-op in this function; it is easier 1525 * to deal with it here than to complicate the slow-timeout path. 1526 */ 1527 if (inm->in6m_timer == 0) { 1528 query_response_timer_expired = 0; 1529 } else if (--inm->in6m_timer == 0) { 1530 query_response_timer_expired = 1; 1531 } else { 1532 V_current_state_timers_running6 = 1; 1533 } 1534 1535 if (inm->in6m_sctimer == 0) { 1536 state_change_retransmit_timer_expired = 0; 1537 } else if (--inm->in6m_sctimer == 0) { 1538 state_change_retransmit_timer_expired = 1; 1539 } else { 1540 V_state_change_timers_running6 = 1; 1541 } 1542 1543 /* We are in fasttimo, so be quick about it. */ 1544 if (!state_change_retransmit_timer_expired && 1545 !query_response_timer_expired) 1546 return; 1547 1548 switch (inm->in6m_state) { 1549 case MLD_NOT_MEMBER: 1550 case MLD_SILENT_MEMBER: 1551 case MLD_SLEEPING_MEMBER: 1552 case MLD_LAZY_MEMBER: 1553 case MLD_AWAKENING_MEMBER: 1554 case MLD_IDLE_MEMBER: 1555 break; 1556 case MLD_G_QUERY_PENDING_MEMBER: 1557 case MLD_SG_QUERY_PENDING_MEMBER: 1558 /* 1559 * Respond to a previously pending Group-Specific 1560 * or Group-and-Source-Specific query by enqueueing 1561 * the appropriate Current-State report for 1562 * immediate transmission. 1563 */ 1564 if (query_response_timer_expired) { 1565 int retval; 1566 1567 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1568 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 1569 0); 1570 CTR2(KTR_MLD, "%s: enqueue record = %d", 1571 __func__, retval); 1572 inm->in6m_state = MLD_REPORTING_MEMBER; 1573 in6m_clear_recorded(inm); 1574 } 1575 /* FALLTHROUGH */ 1576 case MLD_REPORTING_MEMBER: 1577 case MLD_LEAVING_MEMBER: 1578 if (state_change_retransmit_timer_expired) { 1579 /* 1580 * State-change retransmission timer fired. 1581 * If there are any further pending retransmissions, 1582 * set the global pending state-change flag, and 1583 * reset the timer. 1584 */ 1585 if (--inm->in6m_scrv > 0) { 1586 inm->in6m_sctimer = uri_fasthz; 1587 V_state_change_timers_running6 = 1; 1588 } 1589 /* 1590 * Retransmit the previously computed state-change 1591 * report. If there are no further pending 1592 * retransmissions, the mbuf queue will be consumed. 1593 * Update T0 state to T1 as we have now sent 1594 * a state-change. 1595 */ 1596 (void)mld_v2_merge_state_changes(inm, scq); 1597 1598 in6m_commit(inm); 1599 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1600 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1601 inm->in6m_ifp->if_xname); 1602 1603 /* 1604 * If we are leaving the group for good, make sure 1605 * we release MLD's reference to it. 1606 * This release must be deferred using a SLIST, 1607 * as we are called from a loop which traverses 1608 * the in_ifmultiaddr TAILQ. 1609 */ 1610 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1611 inm->in6m_scrv == 0) { 1612 inm->in6m_state = MLD_NOT_MEMBER; 1613 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 1614 inm, in6m_nrele); 1615 } 1616 } 1617 break; 1618 } 1619} 1620 1621/* 1622 * Switch to a different version on the given interface, 1623 * as per Section 9.12. 1624 */ 1625static void 1626mld_set_version(struct mld_ifinfo *mli, const int version) 1627{ 1628 int old_version_timer; 1629 1630 MLD_LOCK_ASSERT(); 1631 1632 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1633 version, mli->mli_ifp, mli->mli_ifp->if_xname); 1634 1635 if (version == MLD_VERSION_1) { 1636 /* 1637 * Compute the "Older Version Querier Present" timer as per 1638 * Section 9.12. 1639 */ 1640 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1641 old_version_timer *= PR_SLOWHZ; 1642 mli->mli_v1_timer = old_version_timer; 1643 } 1644 1645 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1646 mli->mli_version = MLD_VERSION_1; 1647 mld_v2_cancel_link_timers(mli); 1648 } 1649} 1650 1651/* 1652 * Cancel pending MLDv2 timers for the given link and all groups 1653 * joined on it; state-change, general-query, and group-query timers. 1654 */ 1655static void 1656mld_v2_cancel_link_timers(struct mld_ifinfo *mli) 1657{ 1658 struct ifmultiaddr *ifma; 1659 struct ifnet *ifp; 1660 struct in6_multi *inm, *tinm; 1661 1662 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1663 mli->mli_ifp, mli->mli_ifp->if_xname); 1664 1665 IN6_MULTI_LOCK_ASSERT(); 1666 MLD_LOCK_ASSERT(); 1667 1668 /* 1669 * Fast-track this potentially expensive operation 1670 * by checking all the global 'timer pending' flags. 1671 */ 1672 if (!V_interface_timers_running6 && 1673 !V_state_change_timers_running6 && 1674 !V_current_state_timers_running6) 1675 return; 1676 1677 mli->mli_v2_timer = 0; 1678 1679 ifp = mli->mli_ifp; 1680 1681 IF_ADDR_RLOCK(ifp); 1682 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1683 if (ifma->ifma_addr->sa_family != AF_INET6) 1684 continue; 1685 inm = (struct in6_multi *)ifma->ifma_protospec; 1686 switch (inm->in6m_state) { 1687 case MLD_NOT_MEMBER: 1688 case MLD_SILENT_MEMBER: 1689 case MLD_IDLE_MEMBER: 1690 case MLD_LAZY_MEMBER: 1691 case MLD_SLEEPING_MEMBER: 1692 case MLD_AWAKENING_MEMBER: 1693 break; 1694 case MLD_LEAVING_MEMBER: 1695 /* 1696 * If we are leaving the group and switching 1697 * version, we need to release the final 1698 * reference held for issuing the INCLUDE {}. 1699 */ 1700 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1701 in6m_nrele); 1702 /* FALLTHROUGH */ 1703 case MLD_G_QUERY_PENDING_MEMBER: 1704 case MLD_SG_QUERY_PENDING_MEMBER: 1705 in6m_clear_recorded(inm); 1706 /* FALLTHROUGH */ 1707 case MLD_REPORTING_MEMBER: 1708 inm->in6m_sctimer = 0; 1709 inm->in6m_timer = 0; 1710 inm->in6m_state = MLD_REPORTING_MEMBER; 1711 /* 1712 * Free any pending MLDv2 state-change records. 1713 */ 1714 _IF_DRAIN(&inm->in6m_scq); 1715 break; 1716 } 1717 } 1718 IF_ADDR_RUNLOCK(ifp); 1719 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) { 1720 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 1721 in6m_release_locked(inm); 1722 } 1723} 1724 1725/* 1726 * Global slowtimo handler. 1727 * VIMAGE: Timeout handlers are expected to service all vimages. 1728 */ 1729void 1730mld_slowtimo(void) 1731{ 1732 VNET_ITERATOR_DECL(vnet_iter); 1733 1734 VNET_LIST_RLOCK_NOSLEEP(); 1735 VNET_FOREACH(vnet_iter) { 1736 CURVNET_SET(vnet_iter); 1737 mld_slowtimo_vnet(); 1738 CURVNET_RESTORE(); 1739 } 1740 VNET_LIST_RUNLOCK_NOSLEEP(); 1741} 1742 1743/* 1744 * Per-vnet slowtimo handler. 1745 */ 1746static void 1747mld_slowtimo_vnet(void) 1748{ 1749 struct mld_ifinfo *mli; 1750 1751 MLD_LOCK(); 1752 1753 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1754 mld_v1_process_querier_timers(mli); 1755 } 1756 1757 MLD_UNLOCK(); 1758} 1759 1760/* 1761 * Update the Older Version Querier Present timers for a link. 1762 * See Section 9.12 of RFC 3810. 1763 */ 1764static void 1765mld_v1_process_querier_timers(struct mld_ifinfo *mli) 1766{ 1767 1768 MLD_LOCK_ASSERT(); 1769 1770 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { 1771 /* 1772 * MLDv1 Querier Present timer expired; revert to MLDv2. 1773 */ 1774 CTR5(KTR_MLD, 1775 "%s: transition from v%d -> v%d on %p(%s)", 1776 __func__, mli->mli_version, MLD_VERSION_2, 1777 mli->mli_ifp, mli->mli_ifp->if_xname); 1778 mli->mli_version = MLD_VERSION_2; 1779 } 1780} 1781 1782/* 1783 * Transmit an MLDv1 report immediately. 1784 */ 1785static int 1786mld_v1_transmit_report(struct in6_multi *in6m, const int type) 1787{ 1788 struct ifnet *ifp; 1789 struct in6_ifaddr *ia; 1790 struct ip6_hdr *ip6; 1791 struct mbuf *mh, *md; 1792 struct mld_hdr *mld; 1793 1794 IN6_MULTI_LOCK_ASSERT(); 1795 MLD_LOCK_ASSERT(); 1796 1797 ifp = in6m->in6m_ifp; 1798 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1799 /* ia may be NULL if link-local address is tentative. */ 1800 1801 MGETHDR(mh, M_DONTWAIT, MT_HEADER); 1802 if (mh == NULL) { 1803 if (ia != NULL) 1804 ifa_free(&ia->ia_ifa); 1805 return (ENOMEM); 1806 } 1807 MGET(md, M_DONTWAIT, MT_DATA); 1808 if (md == NULL) { 1809 m_free(mh); 1810 if (ia != NULL) 1811 ifa_free(&ia->ia_ifa); 1812 return (ENOMEM); 1813 } 1814 mh->m_next = md; 1815 1816 /* 1817 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so 1818 * that ether_output() does not need to allocate another mbuf 1819 * for the header in the most common case. 1820 */ 1821 MH_ALIGN(mh, sizeof(struct ip6_hdr)); 1822 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); 1823 mh->m_len = sizeof(struct ip6_hdr); 1824 1825 ip6 = mtod(mh, struct ip6_hdr *); 1826 ip6->ip6_flow = 0; 1827 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 1828 ip6->ip6_vfc |= IPV6_VERSION; 1829 ip6->ip6_nxt = IPPROTO_ICMPV6; 1830 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 1831 ip6->ip6_dst = in6m->in6m_addr; 1832 1833 md->m_len = sizeof(struct mld_hdr); 1834 mld = mtod(md, struct mld_hdr *); 1835 mld->mld_type = type; 1836 mld->mld_code = 0; 1837 mld->mld_cksum = 0; 1838 mld->mld_maxdelay = 0; 1839 mld->mld_reserved = 0; 1840 mld->mld_addr = in6m->in6m_addr; 1841 in6_clearscope(&mld->mld_addr); 1842 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 1843 sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); 1844 1845 mld_save_context(mh, ifp); 1846 mh->m_flags |= M_MLDV1; 1847 1848 mld_dispatch_packet(mh); 1849 1850 if (ia != NULL) 1851 ifa_free(&ia->ia_ifa); 1852 return (0); 1853} 1854 1855/* 1856 * Process a state change from the upper layer for the given IPv6 group. 1857 * 1858 * Each socket holds a reference on the in_multi in its own ip_moptions. 1859 * The socket layer will have made the necessary updates to.the group 1860 * state, it is now up to MLD to issue a state change report if there 1861 * has been any change between T0 (when the last state-change was issued) 1862 * and T1 (now). 1863 * 1864 * We use the MLDv2 state machine at group level. The MLd module 1865 * however makes the decision as to which MLD protocol version to speak. 1866 * A state change *from* INCLUDE {} always means an initial join. 1867 * A state change *to* INCLUDE {} always means a final leave. 1868 * 1869 * If delay is non-zero, and the state change is an initial multicast 1870 * join, the state change report will be delayed by 'delay' ticks 1871 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1872 * the initial MLDv2 state change report will be delayed by whichever 1873 * is sooner, a pending state-change timer or delay itself. 1874 * 1875 * VIMAGE: curvnet should have been set by caller, as this routine 1876 * is called from the socket option handlers. 1877 */ 1878int 1879mld_change_state(struct in6_multi *inm, const int delay) 1880{ 1881 struct mld_ifinfo *mli; 1882 struct ifnet *ifp; 1883 int error; 1884 1885 IN6_MULTI_LOCK_ASSERT(); 1886 1887 error = 0; 1888 1889 /* 1890 * Try to detect if the upper layer just asked us to change state 1891 * for an interface which has now gone away. 1892 */ 1893 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1894 ifp = inm->in6m_ifma->ifma_ifp; 1895 if (ifp != NULL) { 1896 /* 1897 * Sanity check that netinet6's notion of ifp is the 1898 * same as net's. 1899 */ 1900 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1901 } 1902 1903 MLD_LOCK(); 1904 1905 mli = MLD_IFINFO(ifp); 1906 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 1907 1908 /* 1909 * If we detect a state transition to or from MCAST_UNDEFINED 1910 * for this group, then we are starting or finishing an MLD 1911 * life cycle for this group. 1912 */ 1913 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1914 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1915 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1916 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1917 CTR1(KTR_MLD, "%s: initial join", __func__); 1918 error = mld_initial_join(inm, mli, delay); 1919 goto out_locked; 1920 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1921 CTR1(KTR_MLD, "%s: final leave", __func__); 1922 mld_final_leave(inm, mli); 1923 goto out_locked; 1924 } 1925 } else { 1926 CTR1(KTR_MLD, "%s: filter set change", __func__); 1927 } 1928 1929 error = mld_handle_state_change(inm, mli); 1930 1931out_locked: 1932 MLD_UNLOCK(); 1933 return (error); 1934} 1935 1936/* 1937 * Perform the initial join for an MLD group. 1938 * 1939 * When joining a group: 1940 * If the group should have its MLD traffic suppressed, do nothing. 1941 * MLDv1 starts sending MLDv1 host membership reports. 1942 * MLDv2 will schedule an MLDv2 state-change report containing the 1943 * initial state of the membership. 1944 * 1945 * If the delay argument is non-zero, then we must delay sending the 1946 * initial state change for delay ticks (in units of PR_FASTHZ). 1947 */ 1948static int 1949mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli, 1950 const int delay) 1951{ 1952 struct ifnet *ifp; 1953 struct ifqueue *ifq; 1954 int error, retval, syncstates; 1955 int odelay; 1956#ifdef KTR 1957 char ip6tbuf[INET6_ADDRSTRLEN]; 1958#endif 1959 1960 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1961 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1962 inm->in6m_ifp, inm->in6m_ifp->if_xname); 1963 1964 error = 0; 1965 syncstates = 1; 1966 1967 ifp = inm->in6m_ifp; 1968 1969 IN6_MULTI_LOCK_ASSERT(); 1970 MLD_LOCK_ASSERT(); 1971 1972 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1973 1974 /* 1975 * Groups joined on loopback or marked as 'not reported', 1976 * enter the MLD_SILENT_MEMBER state and 1977 * are never reported in any protocol exchanges. 1978 * All other groups enter the appropriate state machine 1979 * for the version in use on this link. 1980 * A link marked as MLIF_SILENT causes MLD to be completely 1981 * disabled for the link. 1982 */ 1983 if ((ifp->if_flags & IFF_LOOPBACK) || 1984 (mli->mli_flags & MLIF_SILENT) || 1985 !mld_is_addr_reported(&inm->in6m_addr)) { 1986 CTR1(KTR_MLD, 1987"%s: not kicking state machine for silent group", __func__); 1988 inm->in6m_state = MLD_SILENT_MEMBER; 1989 inm->in6m_timer = 0; 1990 } else { 1991 /* 1992 * Deal with overlapping in_multi lifecycle. 1993 * If this group was LEAVING, then make sure 1994 * we drop the reference we picked up to keep the 1995 * group around for the final INCLUDE {} enqueue. 1996 */ 1997 if (mli->mli_version == MLD_VERSION_2 && 1998 inm->in6m_state == MLD_LEAVING_MEMBER) 1999 in6m_release_locked(inm); 2000 2001 inm->in6m_state = MLD_REPORTING_MEMBER; 2002 2003 switch (mli->mli_version) { 2004 case MLD_VERSION_1: 2005 /* 2006 * If a delay was provided, only use it if 2007 * it is greater than the delay normally 2008 * used for an MLDv1 state change report, 2009 * and delay sending the initial MLDv1 report 2010 * by not transitioning to the IDLE state. 2011 */ 2012 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2013 if (delay) { 2014 inm->in6m_timer = max(delay, odelay); 2015 V_current_state_timers_running6 = 1; 2016 } else { 2017 inm->in6m_state = MLD_IDLE_MEMBER; 2018 error = mld_v1_transmit_report(inm, 2019 MLD_LISTENER_REPORT); 2020 if (error == 0) { 2021 inm->in6m_timer = odelay; 2022 V_current_state_timers_running6 = 1; 2023 } 2024 } 2025 break; 2026 2027 case MLD_VERSION_2: 2028 /* 2029 * Defer update of T0 to T1, until the first copy 2030 * of the state change has been transmitted. 2031 */ 2032 syncstates = 0; 2033 2034 /* 2035 * Immediately enqueue a State-Change Report for 2036 * this interface, freeing any previous reports. 2037 * Don't kick the timers if there is nothing to do, 2038 * or if an error occurred. 2039 */ 2040 ifq = &inm->in6m_scq; 2041 _IF_DRAIN(ifq); 2042 retval = mld_v2_enqueue_group_record(ifq, inm, 1, 2043 0, 0, (mli->mli_flags & MLIF_USEALLOW)); 2044 CTR2(KTR_MLD, "%s: enqueue record = %d", 2045 __func__, retval); 2046 if (retval <= 0) { 2047 error = retval * -1; 2048 break; 2049 } 2050 2051 /* 2052 * Schedule transmission of pending state-change 2053 * report up to RV times for this link. The timer 2054 * will fire at the next mld_fasttimo (~200ms), 2055 * giving us an opportunity to merge the reports. 2056 * 2057 * If a delay was provided to this function, only 2058 * use this delay if sooner than the existing one. 2059 */ 2060 KASSERT(mli->mli_rv > 1, 2061 ("%s: invalid robustness %d", __func__, 2062 mli->mli_rv)); 2063 inm->in6m_scrv = mli->mli_rv; 2064 if (delay) { 2065 if (inm->in6m_sctimer > 1) { 2066 inm->in6m_sctimer = 2067 min(inm->in6m_sctimer, delay); 2068 } else 2069 inm->in6m_sctimer = delay; 2070 } else 2071 inm->in6m_sctimer = 1; 2072 V_state_change_timers_running6 = 1; 2073 2074 error = 0; 2075 break; 2076 } 2077 } 2078 2079 /* 2080 * Only update the T0 state if state change is atomic, 2081 * i.e. we don't need to wait for a timer to fire before we 2082 * can consider the state change to have been communicated. 2083 */ 2084 if (syncstates) { 2085 in6m_commit(inm); 2086 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2087 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2088 inm->in6m_ifp->if_xname); 2089 } 2090 2091 return (error); 2092} 2093 2094/* 2095 * Issue an intermediate state change during the life-cycle. 2096 */ 2097static int 2098mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli) 2099{ 2100 struct ifnet *ifp; 2101 int retval; 2102#ifdef KTR 2103 char ip6tbuf[INET6_ADDRSTRLEN]; 2104#endif 2105 2106 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", 2107 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2108 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2109 2110 ifp = inm->in6m_ifp; 2111 2112 IN6_MULTI_LOCK_ASSERT(); 2113 MLD_LOCK_ASSERT(); 2114 2115 KASSERT(mli && mli->mli_ifp == ifp, 2116 ("%s: inconsistent ifp", __func__)); 2117 2118 if ((ifp->if_flags & IFF_LOOPBACK) || 2119 (mli->mli_flags & MLIF_SILENT) || 2120 !mld_is_addr_reported(&inm->in6m_addr) || 2121 (mli->mli_version != MLD_VERSION_2)) { 2122 if (!mld_is_addr_reported(&inm->in6m_addr)) { 2123 CTR1(KTR_MLD, 2124"%s: not kicking state machine for silent group", __func__); 2125 } 2126 CTR1(KTR_MLD, "%s: nothing to do", __func__); 2127 in6m_commit(inm); 2128 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2129 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2130 inm->in6m_ifp->if_xname); 2131 return (0); 2132 } 2133 2134 _IF_DRAIN(&inm->in6m_scq); 2135 2136 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, 2137 (mli->mli_flags & MLIF_USEALLOW)); 2138 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2139 if (retval <= 0) 2140 return (-retval); 2141 2142 /* 2143 * If record(s) were enqueued, start the state-change 2144 * report timer for this group. 2145 */ 2146 inm->in6m_scrv = mli->mli_rv; 2147 inm->in6m_sctimer = 1; 2148 V_state_change_timers_running6 = 1; 2149 2150 return (0); 2151} 2152 2153/* 2154 * Perform the final leave for a multicast address. 2155 * 2156 * When leaving a group: 2157 * MLDv1 sends a DONE message, if and only if we are the reporter. 2158 * MLDv2 enqueues a state-change report containing a transition 2159 * to INCLUDE {} for immediate transmission. 2160 */ 2161static void 2162mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli) 2163{ 2164 int syncstates; 2165#ifdef KTR 2166 char ip6tbuf[INET6_ADDRSTRLEN]; 2167#endif 2168 2169 syncstates = 1; 2170 2171 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", 2172 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2173 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2174 2175 IN6_MULTI_LOCK_ASSERT(); 2176 MLD_LOCK_ASSERT(); 2177 2178 switch (inm->in6m_state) { 2179 case MLD_NOT_MEMBER: 2180 case MLD_SILENT_MEMBER: 2181 case MLD_LEAVING_MEMBER: 2182 /* Already leaving or left; do nothing. */ 2183 CTR1(KTR_MLD, 2184"%s: not kicking state machine for silent group", __func__); 2185 break; 2186 case MLD_REPORTING_MEMBER: 2187 case MLD_IDLE_MEMBER: 2188 case MLD_G_QUERY_PENDING_MEMBER: 2189 case MLD_SG_QUERY_PENDING_MEMBER: 2190 if (mli->mli_version == MLD_VERSION_1) { 2191#ifdef INVARIANTS 2192 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 2193 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) 2194 panic("%s: MLDv2 state reached, not MLDv2 mode", 2195 __func__); 2196#endif 2197 mld_v1_transmit_report(inm, MLD_LISTENER_DONE); 2198 inm->in6m_state = MLD_NOT_MEMBER; 2199 } else if (mli->mli_version == MLD_VERSION_2) { 2200 /* 2201 * Stop group timer and all pending reports. 2202 * Immediately enqueue a state-change report 2203 * TO_IN {} to be sent on the next fast timeout, 2204 * giving us an opportunity to merge reports. 2205 */ 2206 _IF_DRAIN(&inm->in6m_scq); 2207 inm->in6m_timer = 0; 2208 inm->in6m_scrv = mli->mli_rv; 2209 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2210 "pending retransmissions.", __func__, 2211 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2212 inm->in6m_ifp->if_xname, inm->in6m_scrv); 2213 if (inm->in6m_scrv == 0) { 2214 inm->in6m_state = MLD_NOT_MEMBER; 2215 inm->in6m_sctimer = 0; 2216 } else { 2217 int retval; 2218 2219 in6m_acquire_locked(inm); 2220 2221 retval = mld_v2_enqueue_group_record( 2222 &inm->in6m_scq, inm, 1, 0, 0, 2223 (mli->mli_flags & MLIF_USEALLOW)); 2224 KASSERT(retval != 0, 2225 ("%s: enqueue record = %d", __func__, 2226 retval)); 2227 2228 inm->in6m_state = MLD_LEAVING_MEMBER; 2229 inm->in6m_sctimer = 1; 2230 V_state_change_timers_running6 = 1; 2231 syncstates = 0; 2232 } 2233 break; 2234 } 2235 break; 2236 case MLD_LAZY_MEMBER: 2237 case MLD_SLEEPING_MEMBER: 2238 case MLD_AWAKENING_MEMBER: 2239 /* Our reports are suppressed; do nothing. */ 2240 break; 2241 } 2242 2243 if (syncstates) { 2244 in6m_commit(inm); 2245 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2246 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2247 inm->in6m_ifp->if_xname); 2248 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2249 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2250 __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname); 2251 } 2252} 2253 2254/* 2255 * Enqueue an MLDv2 group record to the given output queue. 2256 * 2257 * If is_state_change is zero, a current-state record is appended. 2258 * If is_state_change is non-zero, a state-change report is appended. 2259 * 2260 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2261 * If is_group_query is zero, and if there is a packet with free space 2262 * at the tail of the queue, it will be appended to providing there 2263 * is enough free space. 2264 * Otherwise a new mbuf packet chain is allocated. 2265 * 2266 * If is_source_query is non-zero, each source is checked to see if 2267 * it was recorded for a Group-Source query, and will be omitted if 2268 * it is not both in-mode and recorded. 2269 * 2270 * If use_block_allow is non-zero, state change reports for initial join 2271 * and final leave, on an inclusive mode group with a source list, will be 2272 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. 2273 * 2274 * The function will attempt to allocate leading space in the packet 2275 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2276 * 2277 * If successful the size of all data appended to the queue is returned, 2278 * otherwise an error code less than zero is returned, or zero if 2279 * no record(s) were appended. 2280 */ 2281static int 2282mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm, 2283 const int is_state_change, const int is_group_query, 2284 const int is_source_query, const int use_block_allow) 2285{ 2286 struct mldv2_record mr; 2287 struct mldv2_record *pmr; 2288 struct ifnet *ifp; 2289 struct ip6_msource *ims, *nims; 2290 struct mbuf *m0, *m, *md; 2291 int error, is_filter_list_change; 2292 int minrec0len, m0srcs, msrcs, nbytes, off; 2293 int record_has_sources; 2294 int now; 2295 int type; 2296 uint8_t mode; 2297#ifdef KTR 2298 char ip6tbuf[INET6_ADDRSTRLEN]; 2299#endif 2300 2301 IN6_MULTI_LOCK_ASSERT(); 2302 2303 error = 0; 2304 ifp = inm->in6m_ifp; 2305 is_filter_list_change = 0; 2306 m = NULL; 2307 m0 = NULL; 2308 m0srcs = 0; 2309 msrcs = 0; 2310 nbytes = 0; 2311 nims = NULL; 2312 record_has_sources = 1; 2313 pmr = NULL; 2314 type = MLD_DO_NOTHING; 2315 mode = inm->in6m_st[1].iss_fmode; 2316 2317 /* 2318 * If we did not transition out of ASM mode during t0->t1, 2319 * and there are no source nodes to process, we can skip 2320 * the generation of source records. 2321 */ 2322 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2323 inm->in6m_nsrc == 0) 2324 record_has_sources = 0; 2325 2326 if (is_state_change) { 2327 /* 2328 * Queue a state change record. 2329 * If the mode did not change, and there are non-ASM 2330 * listeners or source filters present, 2331 * we potentially need to issue two records for the group. 2332 * If there are ASM listeners, and there was no filter 2333 * mode transition of any kind, do nothing. 2334 * 2335 * If we are transitioning to MCAST_UNDEFINED, we need 2336 * not send any sources. A transition to/from this state is 2337 * considered inclusive with some special treatment. 2338 * 2339 * If we are rewriting initial joins/leaves to use 2340 * ALLOW/BLOCK, and the group's membership is inclusive, 2341 * we need to send sources in all cases. 2342 */ 2343 if (mode != inm->in6m_st[0].iss_fmode) { 2344 if (mode == MCAST_EXCLUDE) { 2345 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2346 __func__); 2347 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2348 } else { 2349 CTR1(KTR_MLD, "%s: change to INCLUDE", 2350 __func__); 2351 if (use_block_allow) { 2352 /* 2353 * XXX 2354 * Here we're interested in state 2355 * edges either direction between 2356 * MCAST_UNDEFINED and MCAST_INCLUDE. 2357 * Perhaps we should just check 2358 * the group state, rather than 2359 * the filter mode. 2360 */ 2361 if (mode == MCAST_UNDEFINED) { 2362 type = MLD_BLOCK_OLD_SOURCES; 2363 } else { 2364 type = MLD_ALLOW_NEW_SOURCES; 2365 } 2366 } else { 2367 type = MLD_CHANGE_TO_INCLUDE_MODE; 2368 if (mode == MCAST_UNDEFINED) 2369 record_has_sources = 0; 2370 } 2371 } 2372 } else { 2373 if (record_has_sources) { 2374 is_filter_list_change = 1; 2375 } else { 2376 type = MLD_DO_NOTHING; 2377 } 2378 } 2379 } else { 2380 /* 2381 * Queue a current state record. 2382 */ 2383 if (mode == MCAST_EXCLUDE) { 2384 type = MLD_MODE_IS_EXCLUDE; 2385 } else if (mode == MCAST_INCLUDE) { 2386 type = MLD_MODE_IS_INCLUDE; 2387 KASSERT(inm->in6m_st[1].iss_asm == 0, 2388 ("%s: inm %p is INCLUDE but ASM count is %d", 2389 __func__, inm, inm->in6m_st[1].iss_asm)); 2390 } 2391 } 2392 2393 /* 2394 * Generate the filter list changes using a separate function. 2395 */ 2396 if (is_filter_list_change) 2397 return (mld_v2_enqueue_filter_change(ifq, inm)); 2398 2399 if (type == MLD_DO_NOTHING) { 2400 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2401 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2402 inm->in6m_ifp->if_xname); 2403 return (0); 2404 } 2405 2406 /* 2407 * If any sources are present, we must be able to fit at least 2408 * one in the trailing space of the tail packet's mbuf, 2409 * ideally more. 2410 */ 2411 minrec0len = sizeof(struct mldv2_record); 2412 if (record_has_sources) 2413 minrec0len += sizeof(struct in6_addr); 2414 2415 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2416 mld_rec_type_to_str(type), 2417 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2418 inm->in6m_ifp->if_xname); 2419 2420 /* 2421 * Check if we have a packet in the tail of the queue for this 2422 * group into which the first group record for this group will fit. 2423 * Otherwise allocate a new packet. 2424 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2425 * Note: Group records for G/GSR query responses MUST be sent 2426 * in their own packet. 2427 */ 2428 m0 = ifq->ifq_tail; 2429 if (!is_group_query && 2430 m0 != NULL && 2431 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2432 (m0->m_pkthdr.len + minrec0len) < 2433 (ifp->if_mtu - MLD_MTUSPACE)) { 2434 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2435 sizeof(struct mldv2_record)) / 2436 sizeof(struct in6_addr); 2437 m = m0; 2438 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2439 } else { 2440 if (_IF_QFULL(ifq)) { 2441 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2442 return (-ENOMEM); 2443 } 2444 m = NULL; 2445 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2446 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2447 if (!is_state_change && !is_group_query) 2448 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2449 if (m == NULL) 2450 m = m_gethdr(M_DONTWAIT, MT_DATA); 2451 if (m == NULL) 2452 return (-ENOMEM); 2453 2454 mld_save_context(m, ifp); 2455 2456 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2457 } 2458 2459 /* 2460 * Append group record. 2461 * If we have sources, we don't know how many yet. 2462 */ 2463 mr.mr_type = type; 2464 mr.mr_datalen = 0; 2465 mr.mr_numsrc = 0; 2466 mr.mr_addr = inm->in6m_addr; 2467 in6_clearscope(&mr.mr_addr); 2468 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2469 if (m != m0) 2470 m_freem(m); 2471 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2472 return (-ENOMEM); 2473 } 2474 nbytes += sizeof(struct mldv2_record); 2475 2476 /* 2477 * Append as many sources as will fit in the first packet. 2478 * If we are appending to a new packet, the chain allocation 2479 * may potentially use clusters; use m_getptr() in this case. 2480 * If we are appending to an existing packet, we need to obtain 2481 * a pointer to the group record after m_append(), in case a new 2482 * mbuf was allocated. 2483 * 2484 * Only append sources which are in-mode at t1. If we are 2485 * transitioning to MCAST_UNDEFINED state on the group, and 2486 * use_block_allow is zero, do not include source entries. 2487 * Otherwise, we need to include this source in the report. 2488 * 2489 * Only report recorded sources in our filter set when responding 2490 * to a group-source query. 2491 */ 2492 if (record_has_sources) { 2493 if (m == m0) { 2494 md = m_last(m); 2495 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2496 md->m_len - nbytes); 2497 } else { 2498 md = m_getptr(m, 0, &off); 2499 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2500 off); 2501 } 2502 msrcs = 0; 2503 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2504 nims) { 2505 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2506 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2507 now = im6s_get_mode(inm, ims, 1); 2508 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2509 if ((now != mode) || 2510 (now == mode && 2511 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2512 CTR1(KTR_MLD, "%s: skip node", __func__); 2513 continue; 2514 } 2515 if (is_source_query && ims->im6s_stp == 0) { 2516 CTR1(KTR_MLD, "%s: skip unrecorded node", 2517 __func__); 2518 continue; 2519 } 2520 CTR1(KTR_MLD, "%s: append node", __func__); 2521 if (!m_append(m, sizeof(struct in6_addr), 2522 (void *)&ims->im6s_addr)) { 2523 if (m != m0) 2524 m_freem(m); 2525 CTR1(KTR_MLD, "%s: m_append() failed.", 2526 __func__); 2527 return (-ENOMEM); 2528 } 2529 nbytes += sizeof(struct in6_addr); 2530 ++msrcs; 2531 if (msrcs == m0srcs) 2532 break; 2533 } 2534 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2535 msrcs); 2536 pmr->mr_numsrc = htons(msrcs); 2537 nbytes += (msrcs * sizeof(struct in6_addr)); 2538 } 2539 2540 if (is_source_query && msrcs == 0) { 2541 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2542 if (m != m0) 2543 m_freem(m); 2544 return (0); 2545 } 2546 2547 /* 2548 * We are good to go with first packet. 2549 */ 2550 if (m != m0) { 2551 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2552 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2553 _IF_ENQUEUE(ifq, m); 2554 } else 2555 m->m_pkthdr.PH_vt.vt_nrecs++; 2556 2557 /* 2558 * No further work needed if no source list in packet(s). 2559 */ 2560 if (!record_has_sources) 2561 return (nbytes); 2562 2563 /* 2564 * Whilst sources remain to be announced, we need to allocate 2565 * a new packet and fill out as many sources as will fit. 2566 * Always try for a cluster first. 2567 */ 2568 while (nims != NULL) { 2569 if (_IF_QFULL(ifq)) { 2570 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2571 return (-ENOMEM); 2572 } 2573 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2574 if (m == NULL) 2575 m = m_gethdr(M_DONTWAIT, MT_DATA); 2576 if (m == NULL) 2577 return (-ENOMEM); 2578 mld_save_context(m, ifp); 2579 md = m_getptr(m, 0, &off); 2580 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2581 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2582 2583 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2584 if (m != m0) 2585 m_freem(m); 2586 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2587 return (-ENOMEM); 2588 } 2589 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2590 nbytes += sizeof(struct mldv2_record); 2591 2592 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2593 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2594 2595 msrcs = 0; 2596 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2597 CTR2(KTR_MLD, "%s: visit node %s", 2598 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2599 now = im6s_get_mode(inm, ims, 1); 2600 if ((now != mode) || 2601 (now == mode && 2602 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2603 CTR1(KTR_MLD, "%s: skip node", __func__); 2604 continue; 2605 } 2606 if (is_source_query && ims->im6s_stp == 0) { 2607 CTR1(KTR_MLD, "%s: skip unrecorded node", 2608 __func__); 2609 continue; 2610 } 2611 CTR1(KTR_MLD, "%s: append node", __func__); 2612 if (!m_append(m, sizeof(struct in6_addr), 2613 (void *)&ims->im6s_addr)) { 2614 if (m != m0) 2615 m_freem(m); 2616 CTR1(KTR_MLD, "%s: m_append() failed.", 2617 __func__); 2618 return (-ENOMEM); 2619 } 2620 ++msrcs; 2621 if (msrcs == m0srcs) 2622 break; 2623 } 2624 pmr->mr_numsrc = htons(msrcs); 2625 nbytes += (msrcs * sizeof(struct in6_addr)); 2626 2627 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2628 _IF_ENQUEUE(ifq, m); 2629 } 2630 2631 return (nbytes); 2632} 2633 2634/* 2635 * Type used to mark record pass completion. 2636 * We exploit the fact we can cast to this easily from the 2637 * current filter modes on each ip_msource node. 2638 */ 2639typedef enum { 2640 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2641 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2642 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2643 REC_FULL = REC_ALLOW | REC_BLOCK 2644} rectype_t; 2645 2646/* 2647 * Enqueue an MLDv2 filter list change to the given output queue. 2648 * 2649 * Source list filter state is held in an RB-tree. When the filter list 2650 * for a group is changed without changing its mode, we need to compute 2651 * the deltas between T0 and T1 for each source in the filter set, 2652 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2653 * 2654 * As we may potentially queue two record types, and the entire R-B tree 2655 * needs to be walked at once, we break this out into its own function 2656 * so we can generate a tightly packed queue of packets. 2657 * 2658 * XXX This could be written to only use one tree walk, although that makes 2659 * serializing into the mbuf chains a bit harder. For now we do two walks 2660 * which makes things easier on us, and it may or may not be harder on 2661 * the L2 cache. 2662 * 2663 * If successful the size of all data appended to the queue is returned, 2664 * otherwise an error code less than zero is returned, or zero if 2665 * no record(s) were appended. 2666 */ 2667static int 2668mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm) 2669{ 2670 static const int MINRECLEN = 2671 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2672 struct ifnet *ifp; 2673 struct mldv2_record mr; 2674 struct mldv2_record *pmr; 2675 struct ip6_msource *ims, *nims; 2676 struct mbuf *m, *m0, *md; 2677 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2678 int nallow, nblock; 2679 uint8_t mode, now, then; 2680 rectype_t crt, drt, nrt; 2681#ifdef KTR 2682 char ip6tbuf[INET6_ADDRSTRLEN]; 2683#endif 2684 2685 IN6_MULTI_LOCK_ASSERT(); 2686 2687 if (inm->in6m_nsrc == 0 || 2688 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2689 return (0); 2690 2691 ifp = inm->in6m_ifp; /* interface */ 2692 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2693 crt = REC_NONE; /* current group record type */ 2694 drt = REC_NONE; /* mask of completed group record types */ 2695 nrt = REC_NONE; /* record type for current node */ 2696 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2697 npbytes = 0; /* # of bytes appended this packet */ 2698 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2699 rsrcs = 0; /* # sources encoded in current record */ 2700 schanged = 0; /* # nodes encoded in overall filter change */ 2701 nallow = 0; /* # of source entries in ALLOW_NEW */ 2702 nblock = 0; /* # of source entries in BLOCK_OLD */ 2703 nims = NULL; /* next tree node pointer */ 2704 2705 /* 2706 * For each possible filter record mode. 2707 * The first kind of source we encounter tells us which 2708 * is the first kind of record we start appending. 2709 * If a node transitioned to UNDEFINED at t1, its mode is treated 2710 * as the inverse of the group's filter mode. 2711 */ 2712 while (drt != REC_FULL) { 2713 do { 2714 m0 = ifq->ifq_tail; 2715 if (m0 != NULL && 2716 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2717 MLD_V2_REPORT_MAXRECS) && 2718 (m0->m_pkthdr.len + MINRECLEN) < 2719 (ifp->if_mtu - MLD_MTUSPACE)) { 2720 m = m0; 2721 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2722 sizeof(struct mldv2_record)) / 2723 sizeof(struct in6_addr); 2724 CTR1(KTR_MLD, 2725 "%s: use previous packet", __func__); 2726 } else { 2727 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2728 if (m == NULL) 2729 m = m_gethdr(M_DONTWAIT, MT_DATA); 2730 if (m == NULL) { 2731 CTR1(KTR_MLD, 2732 "%s: m_get*() failed", __func__); 2733 return (-ENOMEM); 2734 } 2735 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2736 mld_save_context(m, ifp); 2737 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2738 sizeof(struct mldv2_record)) / 2739 sizeof(struct in6_addr); 2740 npbytes = 0; 2741 CTR1(KTR_MLD, 2742 "%s: allocated new packet", __func__); 2743 } 2744 /* 2745 * Append the MLD group record header to the 2746 * current packet's data area. 2747 * Recalculate pointer to free space for next 2748 * group record, in case m_append() allocated 2749 * a new mbuf or cluster. 2750 */ 2751 memset(&mr, 0, sizeof(mr)); 2752 mr.mr_addr = inm->in6m_addr; 2753 in6_clearscope(&mr.mr_addr); 2754 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2755 if (m != m0) 2756 m_freem(m); 2757 CTR1(KTR_MLD, 2758 "%s: m_append() failed", __func__); 2759 return (-ENOMEM); 2760 } 2761 npbytes += sizeof(struct mldv2_record); 2762 if (m != m0) { 2763 /* new packet; offset in chain */ 2764 md = m_getptr(m, npbytes - 2765 sizeof(struct mldv2_record), &off); 2766 pmr = (struct mldv2_record *)(mtod(md, 2767 uint8_t *) + off); 2768 } else { 2769 /* current packet; offset from last append */ 2770 md = m_last(m); 2771 pmr = (struct mldv2_record *)(mtod(md, 2772 uint8_t *) + md->m_len - 2773 sizeof(struct mldv2_record)); 2774 } 2775 /* 2776 * Begin walking the tree for this record type 2777 * pass, or continue from where we left off 2778 * previously if we had to allocate a new packet. 2779 * Only report deltas in-mode at t1. 2780 * We need not report included sources as allowed 2781 * if we are in inclusive mode on the group, 2782 * however the converse is not true. 2783 */ 2784 rsrcs = 0; 2785 if (nims == NULL) { 2786 nims = RB_MIN(ip6_msource_tree, 2787 &inm->in6m_srcs); 2788 } 2789 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2790 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2791 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2792 now = im6s_get_mode(inm, ims, 1); 2793 then = im6s_get_mode(inm, ims, 0); 2794 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2795 __func__, then, now); 2796 if (now == then) { 2797 CTR1(KTR_MLD, 2798 "%s: skip unchanged", __func__); 2799 continue; 2800 } 2801 if (mode == MCAST_EXCLUDE && 2802 now == MCAST_INCLUDE) { 2803 CTR1(KTR_MLD, 2804 "%s: skip IN src on EX group", 2805 __func__); 2806 continue; 2807 } 2808 nrt = (rectype_t)now; 2809 if (nrt == REC_NONE) 2810 nrt = (rectype_t)(~mode & REC_FULL); 2811 if (schanged++ == 0) { 2812 crt = nrt; 2813 } else if (crt != nrt) 2814 continue; 2815 if (!m_append(m, sizeof(struct in6_addr), 2816 (void *)&ims->im6s_addr)) { 2817 if (m != m0) 2818 m_freem(m); 2819 CTR1(KTR_MLD, 2820 "%s: m_append() failed", __func__); 2821 return (-ENOMEM); 2822 } 2823 nallow += !!(crt == REC_ALLOW); 2824 nblock += !!(crt == REC_BLOCK); 2825 if (++rsrcs == m0srcs) 2826 break; 2827 } 2828 /* 2829 * If we did not append any tree nodes on this 2830 * pass, back out of allocations. 2831 */ 2832 if (rsrcs == 0) { 2833 npbytes -= sizeof(struct mldv2_record); 2834 if (m != m0) { 2835 CTR1(KTR_MLD, 2836 "%s: m_free(m)", __func__); 2837 m_freem(m); 2838 } else { 2839 CTR1(KTR_MLD, 2840 "%s: m_adj(m, -mr)", __func__); 2841 m_adj(m, -((int)sizeof( 2842 struct mldv2_record))); 2843 } 2844 continue; 2845 } 2846 npbytes += (rsrcs * sizeof(struct in6_addr)); 2847 if (crt == REC_ALLOW) 2848 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2849 else if (crt == REC_BLOCK) 2850 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2851 pmr->mr_numsrc = htons(rsrcs); 2852 /* 2853 * Count the new group record, and enqueue this 2854 * packet if it wasn't already queued. 2855 */ 2856 m->m_pkthdr.PH_vt.vt_nrecs++; 2857 if (m != m0) 2858 _IF_ENQUEUE(ifq, m); 2859 nbytes += npbytes; 2860 } while (nims != NULL); 2861 drt |= crt; 2862 crt = (~crt & REC_FULL); 2863 } 2864 2865 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2866 nallow, nblock); 2867 2868 return (nbytes); 2869} 2870 2871static int 2872mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq) 2873{ 2874 struct ifqueue *gq; 2875 struct mbuf *m; /* pending state-change */ 2876 struct mbuf *m0; /* copy of pending state-change */ 2877 struct mbuf *mt; /* last state-change in packet */ 2878 int docopy, domerge; 2879 u_int recslen; 2880 2881 docopy = 0; 2882 domerge = 0; 2883 recslen = 0; 2884 2885 IN6_MULTI_LOCK_ASSERT(); 2886 MLD_LOCK_ASSERT(); 2887 2888 /* 2889 * If there are further pending retransmissions, make a writable 2890 * copy of each queued state-change message before merging. 2891 */ 2892 if (inm->in6m_scrv > 0) 2893 docopy = 1; 2894 2895 gq = &inm->in6m_scq; 2896#ifdef KTR 2897 if (gq->ifq_head == NULL) { 2898 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2899 __func__, inm); 2900 } 2901#endif 2902 2903 m = gq->ifq_head; 2904 while (m != NULL) { 2905 /* 2906 * Only merge the report into the current packet if 2907 * there is sufficient space to do so; an MLDv2 report 2908 * packet may only contain 65,535 group records. 2909 * Always use a simple mbuf chain concatentation to do this, 2910 * as large state changes for single groups may have 2911 * allocated clusters. 2912 */ 2913 domerge = 0; 2914 mt = ifscq->ifq_tail; 2915 if (mt != NULL) { 2916 recslen = m_length(m, NULL); 2917 2918 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2919 m->m_pkthdr.PH_vt.vt_nrecs <= 2920 MLD_V2_REPORT_MAXRECS) && 2921 (mt->m_pkthdr.len + recslen <= 2922 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2923 domerge = 1; 2924 } 2925 2926 if (!domerge && _IF_QFULL(gq)) { 2927 CTR2(KTR_MLD, 2928 "%s: outbound queue full, skipping whole packet %p", 2929 __func__, m); 2930 mt = m->m_nextpkt; 2931 if (!docopy) 2932 m_freem(m); 2933 m = mt; 2934 continue; 2935 } 2936 2937 if (!docopy) { 2938 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2939 _IF_DEQUEUE(gq, m0); 2940 m = m0->m_nextpkt; 2941 } else { 2942 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2943 m0 = m_dup(m, M_NOWAIT); 2944 if (m0 == NULL) 2945 return (ENOMEM); 2946 m0->m_nextpkt = NULL; 2947 m = m->m_nextpkt; 2948 } 2949 2950 if (!domerge) { 2951 CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)", 2952 __func__, m0, ifscq); 2953 _IF_ENQUEUE(ifscq, m0); 2954 } else { 2955 struct mbuf *mtl; /* last mbuf of packet mt */ 2956 2957 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2958 __func__, m0, mt); 2959 2960 mtl = m_last(mt); 2961 m0->m_flags &= ~M_PKTHDR; 2962 mt->m_pkthdr.len += recslen; 2963 mt->m_pkthdr.PH_vt.vt_nrecs += 2964 m0->m_pkthdr.PH_vt.vt_nrecs; 2965 2966 mtl->m_next = m0; 2967 } 2968 } 2969 2970 return (0); 2971} 2972 2973/* 2974 * Respond to a pending MLDv2 General Query. 2975 */ 2976static void 2977mld_v2_dispatch_general_query(struct mld_ifinfo *mli) 2978{ 2979 struct ifmultiaddr *ifma; 2980 struct ifnet *ifp; 2981 struct in6_multi *inm; 2982 int retval; 2983 2984 IN6_MULTI_LOCK_ASSERT(); 2985 MLD_LOCK_ASSERT(); 2986 2987 KASSERT(mli->mli_version == MLD_VERSION_2, 2988 ("%s: called when version %d", __func__, mli->mli_version)); 2989 2990 ifp = mli->mli_ifp; 2991 2992 IF_ADDR_RLOCK(ifp); 2993 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2994 if (ifma->ifma_addr->sa_family != AF_INET6 || 2995 ifma->ifma_protospec == NULL) 2996 continue; 2997 2998 inm = (struct in6_multi *)ifma->ifma_protospec; 2999 KASSERT(ifp == inm->in6m_ifp, 3000 ("%s: inconsistent ifp", __func__)); 3001 3002 switch (inm->in6m_state) { 3003 case MLD_NOT_MEMBER: 3004 case MLD_SILENT_MEMBER: 3005 break; 3006 case MLD_REPORTING_MEMBER: 3007 case MLD_IDLE_MEMBER: 3008 case MLD_LAZY_MEMBER: 3009 case MLD_SLEEPING_MEMBER: 3010 case MLD_AWAKENING_MEMBER: 3011 inm->in6m_state = MLD_REPORTING_MEMBER; 3012 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 3013 inm, 0, 0, 0, 0); 3014 CTR2(KTR_MLD, "%s: enqueue record = %d", 3015 __func__, retval); 3016 break; 3017 case MLD_G_QUERY_PENDING_MEMBER: 3018 case MLD_SG_QUERY_PENDING_MEMBER: 3019 case MLD_LEAVING_MEMBER: 3020 break; 3021 } 3022 } 3023 IF_ADDR_RUNLOCK(ifp); 3024 3025 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 3026 3027 /* 3028 * Slew transmission of bursts over 500ms intervals. 3029 */ 3030 if (mli->mli_gq.ifq_head != NULL) { 3031 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 3032 MLD_RESPONSE_BURST_INTERVAL); 3033 V_interface_timers_running6 = 1; 3034 } 3035} 3036 3037/* 3038 * Transmit the next pending message in the output queue. 3039 * 3040 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3041 * MRT: Nothing needs to be done, as MLD traffic is always local to 3042 * a link and uses a link-scope multicast address. 3043 */ 3044static void 3045mld_dispatch_packet(struct mbuf *m) 3046{ 3047 struct ip6_moptions im6o; 3048 struct ifnet *ifp; 3049 struct ifnet *oifp; 3050 struct mbuf *m0; 3051 struct mbuf *md; 3052 struct ip6_hdr *ip6; 3053 struct mld_hdr *mld; 3054 int error; 3055 int off; 3056 int type; 3057 uint32_t ifindex; 3058 3059 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3060 3061 /* 3062 * Set VNET image pointer from enqueued mbuf chain 3063 * before doing anything else. Whilst we use interface 3064 * indexes to guard against interface detach, they are 3065 * unique to each VIMAGE and must be retrieved. 3066 */ 3067 ifindex = mld_restore_context(m); 3068 3069 /* 3070 * Check if the ifnet still exists. This limits the scope of 3071 * any race in the absence of a global ifp lock for low cost 3072 * (an array lookup). 3073 */ 3074 ifp = ifnet_byindex(ifindex); 3075 if (ifp == NULL) { 3076 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3077 __func__, m, ifindex); 3078 m_freem(m); 3079 IP6STAT_INC(ip6s_noroute); 3080 goto out; 3081 } 3082 3083 im6o.im6o_multicast_hlim = 1; 3084 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3085 im6o.im6o_multicast_ifp = ifp; 3086 3087 if (m->m_flags & M_MLDV1) { 3088 m0 = m; 3089 } else { 3090 m0 = mld_v2_encap_report(ifp, m); 3091 if (m0 == NULL) { 3092 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3093 IP6STAT_INC(ip6s_odropped); 3094 goto out; 3095 } 3096 } 3097 3098 mld_scrub_context(m0); 3099 m->m_flags &= ~(M_PROTOFLAGS); 3100 m0->m_pkthdr.rcvif = V_loif; 3101 3102 ip6 = mtod(m0, struct ip6_hdr *); 3103#if 0 3104 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ 3105#else 3106 /* 3107 * XXX XXX Break some KPI rules to prevent an LOR which would 3108 * occur if we called in6_setscope() at transmission. 3109 * See comments at top of file. 3110 */ 3111 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); 3112#endif 3113 3114 /* 3115 * Retrieve the ICMPv6 type before handoff to ip6_output(), 3116 * so we can bump the stats. 3117 */ 3118 md = m_getptr(m0, sizeof(struct ip6_hdr), &off); 3119 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); 3120 type = mld->mld_type; 3121 3122 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, 3123 &oifp, NULL); 3124 if (error) { 3125 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); 3126 goto out; 3127 } 3128 ICMP6STAT_INC(icp6s_outhist[type]); 3129 if (oifp != NULL) { 3130 icmp6_ifstat_inc(oifp, ifs6_out_msg); 3131 switch (type) { 3132 case MLD_LISTENER_REPORT: 3133 case MLDV2_LISTENER_REPORT: 3134 icmp6_ifstat_inc(oifp, ifs6_out_mldreport); 3135 break; 3136 case MLD_LISTENER_DONE: 3137 icmp6_ifstat_inc(oifp, ifs6_out_mlddone); 3138 break; 3139 } 3140 } 3141out: 3142 return; 3143} 3144 3145/* 3146 * Encapsulate an MLDv2 report. 3147 * 3148 * KAME IPv6 requires that hop-by-hop options be passed separately, 3149 * and that the IPv6 header be prepended in a separate mbuf. 3150 * 3151 * Returns a pointer to the new mbuf chain head, or NULL if the 3152 * allocation failed. 3153 */ 3154static struct mbuf * 3155mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) 3156{ 3157 struct mbuf *mh; 3158 struct mldv2_report *mld; 3159 struct ip6_hdr *ip6; 3160 struct in6_ifaddr *ia; 3161 int mldreclen; 3162 3163 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 3164 KASSERT((m->m_flags & M_PKTHDR), 3165 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3166 3167 /* 3168 * RFC3590: OK to send as :: or tentative during DAD. 3169 */ 3170 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 3171 if (ia == NULL) 3172 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); 3173 3174 MGETHDR(mh, M_DONTWAIT, MT_HEADER); 3175 if (mh == NULL) { 3176 if (ia != NULL) 3177 ifa_free(&ia->ia_ifa); 3178 m_freem(m); 3179 return (NULL); 3180 } 3181 MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); 3182 3183 mldreclen = m_length(m, NULL); 3184 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); 3185 3186 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); 3187 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + 3188 sizeof(struct mldv2_report) + mldreclen; 3189 3190 ip6 = mtod(mh, struct ip6_hdr *); 3191 ip6->ip6_flow = 0; 3192 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3193 ip6->ip6_vfc |= IPV6_VERSION; 3194 ip6->ip6_nxt = IPPROTO_ICMPV6; 3195 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 3196 if (ia != NULL) 3197 ifa_free(&ia->ia_ifa); 3198 ip6->ip6_dst = in6addr_linklocal_allv2routers; 3199 /* scope ID will be set in netisr */ 3200 3201 mld = (struct mldv2_report *)(ip6 + 1); 3202 mld->mld_type = MLDV2_LISTENER_REPORT; 3203 mld->mld_code = 0; 3204 mld->mld_cksum = 0; 3205 mld->mld_v2_reserved = 0; 3206 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3207 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3208 3209 mh->m_next = m; 3210 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 3211 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); 3212 return (mh); 3213} 3214 3215#ifdef KTR 3216static char * 3217mld_rec_type_to_str(const int type) 3218{ 3219 3220 switch (type) { 3221 case MLD_CHANGE_TO_EXCLUDE_MODE: 3222 return "TO_EX"; 3223 break; 3224 case MLD_CHANGE_TO_INCLUDE_MODE: 3225 return "TO_IN"; 3226 break; 3227 case MLD_MODE_IS_EXCLUDE: 3228 return "MODE_EX"; 3229 break; 3230 case MLD_MODE_IS_INCLUDE: 3231 return "MODE_IN"; 3232 break; 3233 case MLD_ALLOW_NEW_SOURCES: 3234 return "ALLOW_NEW"; 3235 break; 3236 case MLD_BLOCK_OLD_SOURCES: 3237 return "BLOCK_OLD"; 3238 break; 3239 default: 3240 break; 3241 } 3242 return "unknown"; 3243} 3244#endif 3245 3246static void 3247mld_init(void *unused __unused) 3248{ 3249 3250 CTR1(KTR_MLD, "%s: initializing", __func__); 3251 MLD_LOCK_INIT(); 3252 3253 ip6_initpktopts(&mld_po); 3254 mld_po.ip6po_hlim = 1; 3255 mld_po.ip6po_hbh = &mld_ra.hbh; 3256 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; 3257 mld_po.ip6po_flags = IP6PO_DONTFRAG; 3258} 3259SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL); 3260 3261static void 3262mld_uninit(void *unused __unused) 3263{ 3264 3265 CTR1(KTR_MLD, "%s: tearing down", __func__); 3266 MLD_LOCK_DESTROY(); 3267} 3268SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL); 3269 3270static void 3271vnet_mld_init(const void *unused __unused) 3272{ 3273 3274 CTR1(KTR_MLD, "%s: initializing", __func__); 3275 3276 LIST_INIT(&V_mli_head); 3277} 3278VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init, 3279 NULL); 3280 3281static void 3282vnet_mld_uninit(const void *unused __unused) 3283{ 3284 3285 CTR1(KTR_MLD, "%s: tearing down", __func__); 3286 3287 KASSERT(LIST_EMPTY(&V_mli_head), 3288 ("%s: mli list not empty; ifnets not detached?", __func__)); 3289} 3290VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit, 3291 NULL); 3292 3293static int 3294mld_modevent(module_t mod, int type, void *unused __unused) 3295{ 3296 3297 switch (type) { 3298 case MOD_LOAD: 3299 case MOD_UNLOAD: 3300 break; 3301 default: 3302 return (EOPNOTSUPP); 3303 } 3304 return (0); 3305} 3306 3307static moduledata_t mld_mod = { 3308 "mld", 3309 mld_modevent, 3310 0 3311}; 3312DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3313