1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * IP PACKET CLASSIFIER 28 * 29 * The IP packet classifier provides mapping between IP packets and persistent 30 * connection state for connection-oriented protocols. It also provides 31 * interface for managing connection states. 32 * 33 * The connection state is kept in conn_t data structure and contains, among 34 * other things: 35 * 36 * o local/remote address and ports 37 * o Transport protocol 38 * o squeue for the connection (for TCP only) 39 * o reference counter 40 * o Connection state 41 * o hash table linkage 42 * o interface/ire information 43 * o credentials 44 * o ipsec policy 45 * o send and receive functions. 46 * o mutex lock. 47 * 48 * Connections use a reference counting scheme. They are freed when the 49 * reference counter drops to zero. A reference is incremented when connection 50 * is placed in a list or table, when incoming packet for the connection arrives 51 * and when connection is processed via squeue (squeue processing may be 52 * asynchronous and the reference protects the connection from being destroyed 53 * before its processing is finished). 54 * 55 * conn_recv is used to pass up packets to the ULP. 56 * For TCP conn_recv changes. It is tcp_input_listener_unbound initially for 57 * a listener, and changes to tcp_input_listener as the listener has picked a 58 * good squeue. For other cases it is set to tcp_input_data. 59 * 60 * conn_recvicmp is used to pass up ICMP errors to the ULP. 61 * 62 * Classifier uses several hash tables: 63 * 64 * ipcl_conn_fanout: contains all TCP connections in CONNECTED state 65 * ipcl_bind_fanout: contains all connections in BOUND state 66 * ipcl_proto_fanout: IPv4 protocol fanout 67 * ipcl_proto_fanout_v6: IPv6 protocol fanout 68 * ipcl_udp_fanout: contains all UDP connections 69 * ipcl_iptun_fanout: contains all IP tunnel connections 70 * ipcl_globalhash_fanout: contains all connections 71 * 72 * The ipcl_globalhash_fanout is used for any walkers (like snmp and Clustering) 73 * which need to view all existing connections. 74 * 75 * All tables are protected by per-bucket locks. When both per-bucket lock and 76 * connection lock need to be held, the per-bucket lock should be acquired 77 * first, followed by the connection lock. 78 * 79 * All functions doing search in one of these tables increment a reference 80 * counter on the connection found (if any). This reference should be dropped 81 * when the caller has finished processing the connection. 82 * 83 * 84 * INTERFACES: 85 * =========== 86 * 87 * Connection Lookup: 88 * ------------------ 89 * 90 * conn_t *ipcl_classify_v4(mp, protocol, hdr_len, ira, ip_stack) 91 * conn_t *ipcl_classify_v6(mp, protocol, hdr_len, ira, ip_stack) 92 * 93 * Finds connection for an incoming IPv4 or IPv6 packet. Returns NULL if 94 * it can't find any associated connection. If the connection is found, its 95 * reference counter is incremented. 96 * 97 * mp: mblock, containing packet header. The full header should fit 98 * into a single mblock. It should also contain at least full IP 99 * and TCP or UDP header. 100 * 101 * protocol: Either IPPROTO_TCP or IPPROTO_UDP. 102 * 103 * hdr_len: The size of IP header. It is used to find TCP or UDP header in 104 * the packet. 105 * 106 * ira->ira_zoneid: The zone in which the returned connection must be; the 107 * zoneid corresponding to the ire_zoneid on the IRE located for 108 * the packet's destination address. 109 * 110 * ira->ira_flags: Contains the IRAF_TX_MAC_EXEMPTABLE and 111 * IRAF_TX_SHARED_ADDR flags 112 * 113 * For TCP connections, the lookup order is as follows: 114 * 5-tuple {src, dst, protocol, local port, remote port} 115 * lookup in ipcl_conn_fanout table. 116 * 3-tuple {dst, remote port, protocol} lookup in 117 * ipcl_bind_fanout table. 118 * 119 * For UDP connections, a 5-tuple {src, dst, protocol, local port, 120 * remote port} lookup is done on ipcl_udp_fanout. Note that, 121 * these interfaces do not handle cases where a packets belongs 122 * to multiple UDP clients, which is handled in IP itself. 123 * 124 * If the destination IRE is ALL_ZONES (indicated by zoneid), then we must 125 * determine which actual zone gets the segment. This is used only in a 126 * labeled environment. The matching rules are: 127 * 128 * - If it's not a multilevel port, then the label on the packet selects 129 * the zone. Unlabeled packets are delivered to the global zone. 130 * 131 * - If it's a multilevel port, then only the zone registered to receive 132 * packets on that port matches. 133 * 134 * Also, in a labeled environment, packet labels need to be checked. For fully 135 * bound TCP connections, we can assume that the packet label was checked 136 * during connection establishment, and doesn't need to be checked on each 137 * packet. For others, though, we need to check for strict equality or, for 138 * multilevel ports, membership in the range or set. This part currently does 139 * a tnrh lookup on each packet, but could be optimized to use cached results 140 * if that were necessary. (SCTP doesn't come through here, but if it did, 141 * we would apply the same rules as TCP.) 142 * 143 * An implication of the above is that fully-bound TCP sockets must always use 144 * distinct 4-tuples; they can't be discriminated by label alone. 145 * 146 * Note that we cannot trust labels on packets sent to fully-bound UDP sockets, 147 * as there's no connection set-up handshake and no shared state. 148 * 149 * Labels on looped-back packets within a single zone do not need to be 150 * checked, as all processes in the same zone have the same label. 151 * 152 * Finally, for unlabeled packets received by a labeled system, special rules 153 * apply. We consider only the MLP if there is one. Otherwise, we prefer a 154 * socket in the zone whose label matches the default label of the sender, if 155 * any. In any event, the receiving socket must have SO_MAC_EXEMPT set and the 156 * receiver's label must dominate the sender's default label. 157 * 158 * conn_t *ipcl_tcp_lookup_reversed_ipv4(ipha_t *, tcpha_t *, int, ip_stack); 159 * conn_t *ipcl_tcp_lookup_reversed_ipv6(ip6_t *, tcpha_t *, int, uint_t, 160 * ip_stack); 161 * 162 * Lookup routine to find a exact match for {src, dst, local port, 163 * remote port) for TCP connections in ipcl_conn_fanout. The address and 164 * ports are read from the IP and TCP header respectively. 165 * 166 * conn_t *ipcl_lookup_listener_v4(lport, laddr, protocol, 167 * zoneid, ip_stack); 168 * conn_t *ipcl_lookup_listener_v6(lport, laddr, protocol, ifindex, 169 * zoneid, ip_stack); 170 * 171 * Lookup routine to find a listener with the tuple {lport, laddr, 172 * protocol} in the ipcl_bind_fanout table. For IPv6, an additional 173 * parameter interface index is also compared. 174 * 175 * void ipcl_walk(func, arg, ip_stack) 176 * 177 * Apply 'func' to every connection available. The 'func' is called as 178 * (*func)(connp, arg). The walk is non-atomic so connections may be 179 * created and destroyed during the walk. The CONN_CONDEMNED and 180 * CONN_INCIPIENT flags ensure that connections which are newly created 181 * or being destroyed are not selected by the walker. 182 * 183 * Table Updates 184 * ------------- 185 * 186 * int ipcl_conn_insert(connp); 187 * int ipcl_conn_insert_v4(connp); 188 * int ipcl_conn_insert_v6(connp); 189 * 190 * Insert 'connp' in the ipcl_conn_fanout. 191 * Arguements : 192 * connp conn_t to be inserted 193 * 194 * Return value : 195 * 0 if connp was inserted 196 * EADDRINUSE if the connection with the same tuple 197 * already exists. 198 * 199 * int ipcl_bind_insert(connp); 200 * int ipcl_bind_insert_v4(connp); 201 * int ipcl_bind_insert_v6(connp); 202 * 203 * Insert 'connp' in ipcl_bind_fanout. 204 * Arguements : 205 * connp conn_t to be inserted 206 * 207 * 208 * void ipcl_hash_remove(connp); 209 * 210 * Removes the 'connp' from the connection fanout table. 211 * 212 * Connection Creation/Destruction 213 * ------------------------------- 214 * 215 * conn_t *ipcl_conn_create(type, sleep, netstack_t *) 216 * 217 * Creates a new conn based on the type flag, inserts it into 218 * globalhash table. 219 * 220 * type: This flag determines the type of conn_t which needs to be 221 * created i.e., which kmem_cache it comes from. 222 * IPCL_TCPCONN indicates a TCP connection 223 * IPCL_SCTPCONN indicates a SCTP connection 224 * IPCL_UDPCONN indicates a UDP conn_t. 225 * IPCL_RAWIPCONN indicates a RAWIP/ICMP conn_t. 226 * IPCL_RTSCONN indicates a RTS conn_t. 227 * IPCL_IPCCONN indicates all other connections. 228 * 229 * void ipcl_conn_destroy(connp) 230 * 231 * Destroys the connection state, removes it from the global 232 * connection hash table and frees its memory. 233 */ 234 235 #include <sys/types.h> 236 #include <sys/stream.h> 237 #include <sys/stropts.h> 238 #include <sys/sysmacros.h> 239 #include <sys/strsubr.h> 240 #include <sys/strsun.h> 241 #define _SUN_TPI_VERSION 2 242 #include <sys/ddi.h> 243 #include <sys/cmn_err.h> 244 #include <sys/debug.h> 245 246 #include <sys/systm.h> 247 #include <sys/param.h> 248 #include <sys/kmem.h> 249 #include <sys/isa_defs.h> 250 #include <inet/common.h> 251 #include <netinet/ip6.h> 252 #include <netinet/icmp6.h> 253 254 #include <inet/ip.h> 255 #include <inet/ip_if.h> 256 #include <inet/ip_ire.h> 257 #include <inet/ip6.h> 258 #include <inet/ip_ndp.h> 259 #include <inet/ip_impl.h> 260 #include <inet/udp_impl.h> 261 #include <inet/sctp_ip.h> 262 #include <inet/sctp/sctp_impl.h> 263 #include <inet/rawip_impl.h> 264 #include <inet/rts_impl.h> 265 #include <inet/iptun/iptun_impl.h> 266 267 #include <sys/cpuvar.h> 268 269 #include <inet/ipclassifier.h> 270 #include <inet/tcp.h> 271 #include <inet/ipsec_impl.h> 272 273 #include <sys/tsol/tnet.h> 274 #include <sys/sockio.h> 275 276 /* Old value for compatibility. Setable in /etc/system */ 277 uint_t tcp_conn_hash_size = 0; 278 279 /* New value. Zero means choose automatically. Setable in /etc/system */ 280 uint_t ipcl_conn_hash_size = 0; 281 uint_t ipcl_conn_hash_memfactor = 8192; 282 uint_t ipcl_conn_hash_maxsize = 82500; 283 284 /* bind/udp fanout table size */ 285 uint_t ipcl_bind_fanout_size = 512; 286 uint_t ipcl_udp_fanout_size = 16384; 287 288 /* Raw socket fanout size. Must be a power of 2. */ 289 uint_t ipcl_raw_fanout_size = 256; 290 291 /* 292 * The IPCL_IPTUN_HASH() function works best with a prime table size. We 293 * expect that most large deployments would have hundreds of tunnels, and 294 * thousands in the extreme case. 295 */ 296 uint_t ipcl_iptun_fanout_size = 6143; 297 298 /* 299 * Power of 2^N Primes useful for hashing for N of 0-28, 300 * these primes are the nearest prime <= 2^N - 2^(N-2). 301 */ 302 303 #define P2Ps() {0, 0, 0, 5, 11, 23, 47, 89, 191, 383, 761, 1531, 3067, \ 304 6143, 12281, 24571, 49139, 98299, 196597, 393209, \ 305 786431, 1572853, 3145721, 6291449, 12582893, 25165813, \ 306 50331599, 100663291, 201326557, 0} 307 308 /* 309 * wrapper structure to ensure that conn and what follows it (tcp_t, etc) 310 * are aligned on cache lines. 311 */ 312 typedef union itc_s { 313 conn_t itc_conn; 314 char itcu_filler[CACHE_ALIGN(conn_s)]; 315 } itc_t; 316 317 struct kmem_cache *tcp_conn_cache; 318 struct kmem_cache *ip_conn_cache; 319 extern struct kmem_cache *sctp_conn_cache; 320 struct kmem_cache *udp_conn_cache; 321 struct kmem_cache *rawip_conn_cache; 322 struct kmem_cache *rts_conn_cache; 323 324 extern void tcp_timermp_free(tcp_t *); 325 extern mblk_t *tcp_timermp_alloc(int); 326 327 static int ip_conn_constructor(void *, void *, int); 328 static void ip_conn_destructor(void *, void *); 329 330 static int tcp_conn_constructor(void *, void *, int); 331 static void tcp_conn_destructor(void *, void *); 332 333 static int udp_conn_constructor(void *, void *, int); 334 static void udp_conn_destructor(void *, void *); 335 336 static int rawip_conn_constructor(void *, void *, int); 337 static void rawip_conn_destructor(void *, void *); 338 339 static int rts_conn_constructor(void *, void *, int); 340 static void rts_conn_destructor(void *, void *); 341 342 /* 343 * Global (for all stack instances) init routine 344 */ 345 void 346 ipcl_g_init(void) 347 { 348 ip_conn_cache = kmem_cache_create("ip_conn_cache", 349 sizeof (conn_t), CACHE_ALIGN_SIZE, 350 ip_conn_constructor, ip_conn_destructor, 351 NULL, NULL, NULL, 0); 352 353 tcp_conn_cache = kmem_cache_create("tcp_conn_cache", 354 sizeof (itc_t) + sizeof (tcp_t), CACHE_ALIGN_SIZE, 355 tcp_conn_constructor, tcp_conn_destructor, 356 tcp_conn_reclaim, NULL, NULL, 0); 357 358 udp_conn_cache = kmem_cache_create("udp_conn_cache", 359 sizeof (itc_t) + sizeof (udp_t), CACHE_ALIGN_SIZE, 360 udp_conn_constructor, udp_conn_destructor, 361 NULL, NULL, NULL, 0); 362 363 rawip_conn_cache = kmem_cache_create("rawip_conn_cache", 364 sizeof (itc_t) + sizeof (icmp_t), CACHE_ALIGN_SIZE, 365 rawip_conn_constructor, rawip_conn_destructor, 366 NULL, NULL, NULL, 0); 367 368 rts_conn_cache = kmem_cache_create("rts_conn_cache", 369 sizeof (itc_t) + sizeof (rts_t), CACHE_ALIGN_SIZE, 370 rts_conn_constructor, rts_conn_destructor, 371 NULL, NULL, NULL, 0); 372 } 373 374 /* 375 * ipclassifier intialization routine, sets up hash tables. 376 */ 377 void 378 ipcl_init(ip_stack_t *ipst) 379 { 380 int i; 381 int sizes[] = P2Ps(); 382 383 /* 384 * Calculate size of conn fanout table from /etc/system settings 385 */ 386 if (ipcl_conn_hash_size != 0) { 387 ipst->ips_ipcl_conn_fanout_size = ipcl_conn_hash_size; 388 } else if (tcp_conn_hash_size != 0) { 389 ipst->ips_ipcl_conn_fanout_size = tcp_conn_hash_size; 390 } else { 391 extern pgcnt_t freemem; 392 393 ipst->ips_ipcl_conn_fanout_size = 394 (freemem * PAGESIZE) / ipcl_conn_hash_memfactor; 395 396 if (ipst->ips_ipcl_conn_fanout_size > ipcl_conn_hash_maxsize) { 397 ipst->ips_ipcl_conn_fanout_size = 398 ipcl_conn_hash_maxsize; 399 } 400 } 401 402 for (i = 9; i < sizeof (sizes) / sizeof (*sizes) - 1; i++) { 403 if (sizes[i] >= ipst->ips_ipcl_conn_fanout_size) { 404 break; 405 } 406 } 407 if ((ipst->ips_ipcl_conn_fanout_size = sizes[i]) == 0) { 408 /* Out of range, use the 2^16 value */ 409 ipst->ips_ipcl_conn_fanout_size = sizes[16]; 410 } 411 412 /* Take values from /etc/system */ 413 ipst->ips_ipcl_bind_fanout_size = ipcl_bind_fanout_size; 414 ipst->ips_ipcl_udp_fanout_size = ipcl_udp_fanout_size; 415 ipst->ips_ipcl_raw_fanout_size = ipcl_raw_fanout_size; 416 ipst->ips_ipcl_iptun_fanout_size = ipcl_iptun_fanout_size; 417 418 ASSERT(ipst->ips_ipcl_conn_fanout == NULL); 419 420 ipst->ips_ipcl_conn_fanout = kmem_zalloc( 421 ipst->ips_ipcl_conn_fanout_size * sizeof (connf_t), KM_SLEEP); 422 423 for (i = 0; i < ipst->ips_ipcl_conn_fanout_size; i++) { 424 mutex_init(&ipst->ips_ipcl_conn_fanout[i].connf_lock, NULL, 425 MUTEX_DEFAULT, NULL); 426 } 427 428 ipst->ips_ipcl_bind_fanout = kmem_zalloc( 429 ipst->ips_ipcl_bind_fanout_size * sizeof (connf_t), KM_SLEEP); 430 431 for (i = 0; i < ipst->ips_ipcl_bind_fanout_size; i++) { 432 mutex_init(&ipst->ips_ipcl_bind_fanout[i].connf_lock, NULL, 433 MUTEX_DEFAULT, NULL); 434 } 435 436 ipst->ips_ipcl_proto_fanout_v4 = kmem_zalloc(IPPROTO_MAX * 437 sizeof (connf_t), KM_SLEEP); 438 for (i = 0; i < IPPROTO_MAX; i++) { 439 mutex_init(&ipst->ips_ipcl_proto_fanout_v4[i].connf_lock, NULL, 440 MUTEX_DEFAULT, NULL); 441 } 442 443 ipst->ips_ipcl_proto_fanout_v6 = kmem_zalloc(IPPROTO_MAX * 444 sizeof (connf_t), KM_SLEEP); 445 for (i = 0; i < IPPROTO_MAX; i++) { 446 mutex_init(&ipst->ips_ipcl_proto_fanout_v6[i].connf_lock, NULL, 447 MUTEX_DEFAULT, NULL); 448 } 449 450 ipst->ips_rts_clients = kmem_zalloc(sizeof (connf_t), KM_SLEEP); 451 mutex_init(&ipst->ips_rts_clients->connf_lock, 452 NULL, MUTEX_DEFAULT, NULL); 453 454 ipst->ips_ipcl_udp_fanout = kmem_zalloc( 455 ipst->ips_ipcl_udp_fanout_size * sizeof (connf_t), KM_SLEEP); 456 for (i = 0; i < ipst->ips_ipcl_udp_fanout_size; i++) { 457 mutex_init(&ipst->ips_ipcl_udp_fanout[i].connf_lock, NULL, 458 MUTEX_DEFAULT, NULL); 459 } 460 461 ipst->ips_ipcl_iptun_fanout = kmem_zalloc( 462 ipst->ips_ipcl_iptun_fanout_size * sizeof (connf_t), KM_SLEEP); 463 for (i = 0; i < ipst->ips_ipcl_iptun_fanout_size; i++) { 464 mutex_init(&ipst->ips_ipcl_iptun_fanout[i].connf_lock, NULL, 465 MUTEX_DEFAULT, NULL); 466 } 467 468 ipst->ips_ipcl_raw_fanout = kmem_zalloc( 469 ipst->ips_ipcl_raw_fanout_size * sizeof (connf_t), KM_SLEEP); 470 for (i = 0; i < ipst->ips_ipcl_raw_fanout_size; i++) { 471 mutex_init(&ipst->ips_ipcl_raw_fanout[i].connf_lock, NULL, 472 MUTEX_DEFAULT, NULL); 473 } 474 475 ipst->ips_ipcl_globalhash_fanout = kmem_zalloc( 476 sizeof (connf_t) * CONN_G_HASH_SIZE, KM_SLEEP); 477 for (i = 0; i < CONN_G_HASH_SIZE; i++) { 478 mutex_init(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock, 479 NULL, MUTEX_DEFAULT, NULL); 480 } 481 } 482 483 void 484 ipcl_g_destroy(void) 485 { 486 kmem_cache_destroy(ip_conn_cache); 487 kmem_cache_destroy(tcp_conn_cache); 488 kmem_cache_destroy(udp_conn_cache); 489 kmem_cache_destroy(rawip_conn_cache); 490 kmem_cache_destroy(rts_conn_cache); 491 } 492 493 /* 494 * All user-level and kernel use of the stack must be gone 495 * by now. 496 */ 497 void 498 ipcl_destroy(ip_stack_t *ipst) 499 { 500 int i; 501 502 for (i = 0; i < ipst->ips_ipcl_conn_fanout_size; i++) { 503 ASSERT(ipst->ips_ipcl_conn_fanout[i].connf_head == NULL); 504 mutex_destroy(&ipst->ips_ipcl_conn_fanout[i].connf_lock); 505 } 506 kmem_free(ipst->ips_ipcl_conn_fanout, ipst->ips_ipcl_conn_fanout_size * 507 sizeof (connf_t)); 508 ipst->ips_ipcl_conn_fanout = NULL; 509 510 for (i = 0; i < ipst->ips_ipcl_bind_fanout_size; i++) { 511 ASSERT(ipst->ips_ipcl_bind_fanout[i].connf_head == NULL); 512 mutex_destroy(&ipst->ips_ipcl_bind_fanout[i].connf_lock); 513 } 514 kmem_free(ipst->ips_ipcl_bind_fanout, ipst->ips_ipcl_bind_fanout_size * 515 sizeof (connf_t)); 516 ipst->ips_ipcl_bind_fanout = NULL; 517 518 for (i = 0; i < IPPROTO_MAX; i++) { 519 ASSERT(ipst->ips_ipcl_proto_fanout_v4[i].connf_head == NULL); 520 mutex_destroy(&ipst->ips_ipcl_proto_fanout_v4[i].connf_lock); 521 } 522 kmem_free(ipst->ips_ipcl_proto_fanout_v4, 523 IPPROTO_MAX * sizeof (connf_t)); 524 ipst->ips_ipcl_proto_fanout_v4 = NULL; 525 526 for (i = 0; i < IPPROTO_MAX; i++) { 527 ASSERT(ipst->ips_ipcl_proto_fanout_v6[i].connf_head == NULL); 528 mutex_destroy(&ipst->ips_ipcl_proto_fanout_v6[i].connf_lock); 529 } 530 kmem_free(ipst->ips_ipcl_proto_fanout_v6, 531 IPPROTO_MAX * sizeof (connf_t)); 532 ipst->ips_ipcl_proto_fanout_v6 = NULL; 533 534 for (i = 0; i < ipst->ips_ipcl_udp_fanout_size; i++) { 535 ASSERT(ipst->ips_ipcl_udp_fanout[i].connf_head == NULL); 536 mutex_destroy(&ipst->ips_ipcl_udp_fanout[i].connf_lock); 537 } 538 kmem_free(ipst->ips_ipcl_udp_fanout, ipst->ips_ipcl_udp_fanout_size * 539 sizeof (connf_t)); 540 ipst->ips_ipcl_udp_fanout = NULL; 541 542 for (i = 0; i < ipst->ips_ipcl_iptun_fanout_size; i++) { 543 ASSERT(ipst->ips_ipcl_iptun_fanout[i].connf_head == NULL); 544 mutex_destroy(&ipst->ips_ipcl_iptun_fanout[i].connf_lock); 545 } 546 kmem_free(ipst->ips_ipcl_iptun_fanout, 547 ipst->ips_ipcl_iptun_fanout_size * sizeof (connf_t)); 548 ipst->ips_ipcl_iptun_fanout = NULL; 549 550 for (i = 0; i < ipst->ips_ipcl_raw_fanout_size; i++) { 551 ASSERT(ipst->ips_ipcl_raw_fanout[i].connf_head == NULL); 552 mutex_destroy(&ipst->ips_ipcl_raw_fanout[i].connf_lock); 553 } 554 kmem_free(ipst->ips_ipcl_raw_fanout, ipst->ips_ipcl_raw_fanout_size * 555 sizeof (connf_t)); 556 ipst->ips_ipcl_raw_fanout = NULL; 557 558 for (i = 0; i < CONN_G_HASH_SIZE; i++) { 559 ASSERT(ipst->ips_ipcl_globalhash_fanout[i].connf_head == NULL); 560 mutex_destroy(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock); 561 } 562 kmem_free(ipst->ips_ipcl_globalhash_fanout, 563 sizeof (connf_t) * CONN_G_HASH_SIZE); 564 ipst->ips_ipcl_globalhash_fanout = NULL; 565 566 ASSERT(ipst->ips_rts_clients->connf_head == NULL); 567 mutex_destroy(&ipst->ips_rts_clients->connf_lock); 568 kmem_free(ipst->ips_rts_clients, sizeof (connf_t)); 569 ipst->ips_rts_clients = NULL; 570 } 571 572 /* 573 * conn creation routine. initialize the conn, sets the reference 574 * and inserts it in the global hash table. 575 */ 576 conn_t * 577 ipcl_conn_create(uint32_t type, int sleep, netstack_t *ns) 578 { 579 conn_t *connp; 580 struct kmem_cache *conn_cache; 581 582 switch (type) { 583 case IPCL_SCTPCONN: 584 if ((connp = kmem_cache_alloc(sctp_conn_cache, sleep)) == NULL) 585 return (NULL); 586 sctp_conn_init(connp); 587 netstack_hold(ns); 588 connp->conn_netstack = ns; 589 connp->conn_ixa->ixa_ipst = ns->netstack_ip; 590 ipcl_globalhash_insert(connp); 591 return (connp); 592 593 case IPCL_TCPCONN: 594 conn_cache = tcp_conn_cache; 595 break; 596 597 case IPCL_UDPCONN: 598 conn_cache = udp_conn_cache; 599 break; 600 601 case IPCL_RAWIPCONN: 602 conn_cache = rawip_conn_cache; 603 break; 604 605 case IPCL_RTSCONN: 606 conn_cache = rts_conn_cache; 607 break; 608 609 case IPCL_IPCCONN: 610 conn_cache = ip_conn_cache; 611 break; 612 613 default: 614 connp = NULL; 615 ASSERT(0); 616 } 617 618 if ((connp = kmem_cache_alloc(conn_cache, sleep)) == NULL) 619 return (NULL); 620 621 connp->conn_ref = 1; 622 netstack_hold(ns); 623 connp->conn_netstack = ns; 624 connp->conn_ixa->ixa_ipst = ns->netstack_ip; 625 ipcl_globalhash_insert(connp); 626 return (connp); 627 } 628 629 void 630 ipcl_conn_destroy(conn_t *connp) 631 { 632 mblk_t *mp; 633 netstack_t *ns = connp->conn_netstack; 634 635 ASSERT(!MUTEX_HELD(&connp->conn_lock)); 636 ASSERT(connp->conn_ref == 0); 637 638 DTRACE_PROBE1(conn__destroy, conn_t *, connp); 639 640 if (connp->conn_cred != NULL) { 641 crfree(connp->conn_cred); 642 connp->conn_cred = NULL; 643 /* ixa_cred done in ipcl_conn_cleanup below */ 644 } 645 646 if (connp->conn_ht_iphc != NULL) { 647 kmem_free(connp->conn_ht_iphc, connp->conn_ht_iphc_allocated); 648 connp->conn_ht_iphc = NULL; 649 connp->conn_ht_iphc_allocated = 0; 650 connp->conn_ht_iphc_len = 0; 651 connp->conn_ht_ulp = NULL; 652 connp->conn_ht_ulp_len = 0; 653 } 654 ip_pkt_free(&connp->conn_xmit_ipp); 655 656 ipcl_globalhash_remove(connp); 657 658 if (connp->conn_latch != NULL) { 659 IPLATCH_REFRELE(connp->conn_latch); 660 connp->conn_latch = NULL; 661 } 662 if (connp->conn_latch_in_policy != NULL) { 663 IPPOL_REFRELE(connp->conn_latch_in_policy); 664 connp->conn_latch_in_policy = NULL; 665 } 666 if (connp->conn_latch_in_action != NULL) { 667 IPACT_REFRELE(connp->conn_latch_in_action); 668 connp->conn_latch_in_action = NULL; 669 } 670 if (connp->conn_policy != NULL) { 671 IPPH_REFRELE(connp->conn_policy, ns); 672 connp->conn_policy = NULL; 673 } 674 675 if (connp->conn_ipsec_opt_mp != NULL) { 676 freemsg(connp->conn_ipsec_opt_mp); 677 connp->conn_ipsec_opt_mp = NULL; 678 } 679 680 if (connp->conn_flags & IPCL_TCPCONN) { 681 tcp_t *tcp = connp->conn_tcp; 682 683 tcp_free(tcp); 684 mp = tcp->tcp_timercache; 685 686 tcp->tcp_tcps = NULL; 687 688 if (tcp->tcp_sack_info != NULL) { 689 bzero(tcp->tcp_sack_info, sizeof (tcp_sack_info_t)); 690 kmem_cache_free(tcp_sack_info_cache, 691 tcp->tcp_sack_info); 692 } 693 694 /* 695 * tcp_rsrv_mp can be NULL if tcp_get_conn() fails to allocate 696 * the mblk. 697 */ 698 if (tcp->tcp_rsrv_mp != NULL) { 699 freeb(tcp->tcp_rsrv_mp); 700 tcp->tcp_rsrv_mp = NULL; 701 mutex_destroy(&tcp->tcp_rsrv_mp_lock); 702 } 703 704 ipcl_conn_cleanup(connp); 705 connp->conn_flags = IPCL_TCPCONN; 706 if (ns != NULL) { 707 ASSERT(tcp->tcp_tcps == NULL); 708 connp->conn_netstack = NULL; 709 connp->conn_ixa->ixa_ipst = NULL; 710 netstack_rele(ns); 711 } 712 713 bzero(tcp, sizeof (tcp_t)); 714 715 tcp->tcp_timercache = mp; 716 tcp->tcp_connp = connp; 717 kmem_cache_free(tcp_conn_cache, connp); 718 return; 719 } 720 721 if (connp->conn_flags & IPCL_SCTPCONN) { 722 ASSERT(ns != NULL); 723 sctp_free(connp); 724 return; 725 } 726 727 ipcl_conn_cleanup(connp); 728 if (ns != NULL) { 729 connp->conn_netstack = NULL; 730 connp->conn_ixa->ixa_ipst = NULL; 731 netstack_rele(ns); 732 } 733 734 /* leave conn_priv aka conn_udp, conn_icmp, etc in place. */ 735 if (connp->conn_flags & IPCL_UDPCONN) { 736 connp->conn_flags = IPCL_UDPCONN; 737 kmem_cache_free(udp_conn_cache, connp); 738 } else if (connp->conn_flags & IPCL_RAWIPCONN) { 739 connp->conn_flags = IPCL_RAWIPCONN; 740 connp->conn_proto = IPPROTO_ICMP; 741 connp->conn_ixa->ixa_protocol = connp->conn_proto; 742 kmem_cache_free(rawip_conn_cache, connp); 743 } else if (connp->conn_flags & IPCL_RTSCONN) { 744 connp->conn_flags = IPCL_RTSCONN; 745 kmem_cache_free(rts_conn_cache, connp); 746 } else { 747 connp->conn_flags = IPCL_IPCCONN; 748 ASSERT(connp->conn_flags & IPCL_IPCCONN); 749 ASSERT(connp->conn_priv == NULL); 750 kmem_cache_free(ip_conn_cache, connp); 751 } 752 } 753 754 /* 755 * Running in cluster mode - deregister listener information 756 */ 757 static void 758 ipcl_conn_unlisten(conn_t *connp) 759 { 760 ASSERT((connp->conn_flags & IPCL_CL_LISTENER) != 0); 761 ASSERT(connp->conn_lport != 0); 762 763 if (cl_inet_unlisten != NULL) { 764 sa_family_t addr_family; 765 uint8_t *laddrp; 766 767 if (connp->conn_ipversion == IPV6_VERSION) { 768 addr_family = AF_INET6; 769 laddrp = (uint8_t *)&connp->conn_bound_addr_v6; 770 } else { 771 addr_family = AF_INET; 772 laddrp = (uint8_t *)&connp->conn_bound_addr_v4; 773 } 774 (*cl_inet_unlisten)(connp->conn_netstack->netstack_stackid, 775 IPPROTO_TCP, addr_family, laddrp, connp->conn_lport, NULL); 776 } 777 connp->conn_flags &= ~IPCL_CL_LISTENER; 778 } 779 780 /* 781 * We set the IPCL_REMOVED flag (instead of clearing the flag indicating 782 * which table the conn belonged to). So for debugging we can see which hash 783 * table this connection was in. 784 */ 785 #define IPCL_HASH_REMOVE(connp) { \ 786 connf_t *connfp = (connp)->conn_fanout; \ 787 ASSERT(!MUTEX_HELD(&((connp)->conn_lock))); \ 788 if (connfp != NULL) { \ 789 mutex_enter(&connfp->connf_lock); \ 790 if ((connp)->conn_next != NULL) \ 791 (connp)->conn_next->conn_prev = \ 792 (connp)->conn_prev; \ 793 if ((connp)->conn_prev != NULL) \ 794 (connp)->conn_prev->conn_next = \ 795 (connp)->conn_next; \ 796 else \ 797 connfp->connf_head = (connp)->conn_next; \ 798 (connp)->conn_fanout = NULL; \ 799 (connp)->conn_next = NULL; \ 800 (connp)->conn_prev = NULL; \ 801 (connp)->conn_flags |= IPCL_REMOVED; \ 802 if (((connp)->conn_flags & IPCL_CL_LISTENER) != 0) \ 803 ipcl_conn_unlisten((connp)); \ 804 CONN_DEC_REF((connp)); \ 805 mutex_exit(&connfp->connf_lock); \ 806 } \ 807 } 808 809 void 810 ipcl_hash_remove(conn_t *connp) 811 { 812 uint8_t protocol = connp->conn_proto; 813 814 IPCL_HASH_REMOVE(connp); 815 if (protocol == IPPROTO_RSVP) 816 ill_set_inputfn_all(connp->conn_netstack->netstack_ip); 817 } 818 819 /* 820 * The whole purpose of this function is allow removal of 821 * a conn_t from the connected hash for timewait reclaim. 822 * This is essentially a TW reclaim fastpath where timewait 823 * collector checks under fanout lock (so no one else can 824 * get access to the conn_t) that refcnt is 2 i.e. one for 825 * TCP and one for the classifier hash list. If ref count 826 * is indeed 2, we can just remove the conn under lock and 827 * avoid cleaning up the conn under squeue. This gives us 828 * improved performance. 829 */ 830 void 831 ipcl_hash_remove_locked(conn_t *connp, connf_t *connfp) 832 { 833 ASSERT(MUTEX_HELD(&connfp->connf_lock)); 834 ASSERT(MUTEX_HELD(&connp->conn_lock)); 835 ASSERT((connp->conn_flags & IPCL_CL_LISTENER) == 0); 836 837 if ((connp)->conn_next != NULL) { 838 (connp)->conn_next->conn_prev = (connp)->conn_prev; 839 } 840 if ((connp)->conn_prev != NULL) { 841 (connp)->conn_prev->conn_next = (connp)->conn_next; 842 } else { 843 connfp->connf_head = (connp)->conn_next; 844 } 845 (connp)->conn_fanout = NULL; 846 (connp)->conn_next = NULL; 847 (connp)->conn_prev = NULL; 848 (connp)->conn_flags |= IPCL_REMOVED; 849 ASSERT((connp)->conn_ref == 2); 850 (connp)->conn_ref--; 851 } 852 853 #define IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp) { \ 854 ASSERT((connp)->conn_fanout == NULL); \ 855 ASSERT((connp)->conn_next == NULL); \ 856 ASSERT((connp)->conn_prev == NULL); \ 857 if ((connfp)->connf_head != NULL) { \ 858 (connfp)->connf_head->conn_prev = (connp); \ 859 (connp)->conn_next = (connfp)->connf_head; \ 860 } \ 861 (connp)->conn_fanout = (connfp); \ 862 (connfp)->connf_head = (connp); \ 863 (connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) | \ 864 IPCL_CONNECTED; \ 865 CONN_INC_REF(connp); \ 866 } 867 868 #define IPCL_HASH_INSERT_CONNECTED(connfp, connp) { \ 869 IPCL_HASH_REMOVE((connp)); \ 870 mutex_enter(&(connfp)->connf_lock); \ 871 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp); \ 872 mutex_exit(&(connfp)->connf_lock); \ 873 } 874 875 #define IPCL_HASH_INSERT_BOUND(connfp, connp) { \ 876 conn_t *pconnp = NULL, *nconnp; \ 877 IPCL_HASH_REMOVE((connp)); \ 878 mutex_enter(&(connfp)->connf_lock); \ 879 nconnp = (connfp)->connf_head; \ 880 while (nconnp != NULL && \ 881 !_IPCL_V4_MATCH_ANY(nconnp->conn_laddr_v6)) { \ 882 pconnp = nconnp; \ 883 nconnp = nconnp->conn_next; \ 884 } \ 885 if (pconnp != NULL) { \ 886 pconnp->conn_next = (connp); \ 887 (connp)->conn_prev = pconnp; \ 888 } else { \ 889 (connfp)->connf_head = (connp); \ 890 } \ 891 if (nconnp != NULL) { \ 892 (connp)->conn_next = nconnp; \ 893 nconnp->conn_prev = (connp); \ 894 } \ 895 (connp)->conn_fanout = (connfp); \ 896 (connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) | \ 897 IPCL_BOUND; \ 898 CONN_INC_REF(connp); \ 899 mutex_exit(&(connfp)->connf_lock); \ 900 } 901 902 #define IPCL_HASH_INSERT_WILDCARD(connfp, connp) { \ 903 conn_t **list, *prev, *next; \ 904 boolean_t isv4mapped = \ 905 IN6_IS_ADDR_V4MAPPED(&(connp)->conn_laddr_v6); \ 906 IPCL_HASH_REMOVE((connp)); \ 907 mutex_enter(&(connfp)->connf_lock); \ 908 list = &(connfp)->connf_head; \ 909 prev = NULL; \ 910 while ((next = *list) != NULL) { \ 911 if (isv4mapped && \ 912 IN6_IS_ADDR_UNSPECIFIED(&next->conn_laddr_v6) && \ 913 connp->conn_zoneid == next->conn_zoneid) { \ 914 (connp)->conn_next = next; \ 915 if (prev != NULL) \ 916 prev = next->conn_prev; \ 917 next->conn_prev = (connp); \ 918 break; \ 919 } \ 920 list = &next->conn_next; \ 921 prev = next; \ 922 } \ 923 (connp)->conn_prev = prev; \ 924 *list = (connp); \ 925 (connp)->conn_fanout = (connfp); \ 926 (connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) | \ 927 IPCL_BOUND; \ 928 CONN_INC_REF((connp)); \ 929 mutex_exit(&(connfp)->connf_lock); \ 930 } 931 932 void 933 ipcl_hash_insert_wildcard(connf_t *connfp, conn_t *connp) 934 { 935 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 936 } 937 938 /* 939 * Because the classifier is used to classify inbound packets, the destination 940 * address is meant to be our local tunnel address (tunnel source), and the 941 * source the remote tunnel address (tunnel destination). 942 * 943 * Note that conn_proto can't be used for fanout since the upper protocol 944 * can be both 41 and 4 when IPv6 and IPv4 are over the same tunnel. 945 */ 946 conn_t * 947 ipcl_iptun_classify_v4(ipaddr_t *src, ipaddr_t *dst, ip_stack_t *ipst) 948 { 949 connf_t *connfp; 950 conn_t *connp; 951 952 /* first look for IPv4 tunnel links */ 953 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH(*dst, *src)]; 954 mutex_enter(&connfp->connf_lock); 955 for (connp = connfp->connf_head; connp != NULL; 956 connp = connp->conn_next) { 957 if (IPCL_IPTUN_MATCH(connp, *dst, *src)) 958 break; 959 } 960 if (connp != NULL) 961 goto done; 962 963 mutex_exit(&connfp->connf_lock); 964 965 /* We didn't find an IPv4 tunnel, try a 6to4 tunnel */ 966 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH(*dst, 967 INADDR_ANY)]; 968 mutex_enter(&connfp->connf_lock); 969 for (connp = connfp->connf_head; connp != NULL; 970 connp = connp->conn_next) { 971 if (IPCL_IPTUN_MATCH(connp, *dst, INADDR_ANY)) 972 break; 973 } 974 done: 975 if (connp != NULL) 976 CONN_INC_REF(connp); 977 mutex_exit(&connfp->connf_lock); 978 return (connp); 979 } 980 981 conn_t * 982 ipcl_iptun_classify_v6(in6_addr_t *src, in6_addr_t *dst, ip_stack_t *ipst) 983 { 984 connf_t *connfp; 985 conn_t *connp; 986 987 /* Look for an IPv6 tunnel link */ 988 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH_V6(dst, src)]; 989 mutex_enter(&connfp->connf_lock); 990 for (connp = connfp->connf_head; connp != NULL; 991 connp = connp->conn_next) { 992 if (IPCL_IPTUN_MATCH_V6(connp, dst, src)) { 993 CONN_INC_REF(connp); 994 break; 995 } 996 } 997 mutex_exit(&connfp->connf_lock); 998 return (connp); 999 } 1000 1001 /* 1002 * This function is used only for inserting SCTP raw socket now. 1003 * This may change later. 1004 * 1005 * Note that only one raw socket can be bound to a port. The param 1006 * lport is in network byte order. 1007 */ 1008 static int 1009 ipcl_sctp_hash_insert(conn_t *connp, in_port_t lport) 1010 { 1011 connf_t *connfp; 1012 conn_t *oconnp; 1013 ip_stack_t *ipst = connp->conn_netstack->netstack_ip; 1014 1015 connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(ntohs(lport), ipst)]; 1016 1017 /* Check for existing raw socket already bound to the port. */ 1018 mutex_enter(&connfp->connf_lock); 1019 for (oconnp = connfp->connf_head; oconnp != NULL; 1020 oconnp = oconnp->conn_next) { 1021 if (oconnp->conn_lport == lport && 1022 oconnp->conn_zoneid == connp->conn_zoneid && 1023 oconnp->conn_family == connp->conn_family && 1024 ((IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) || 1025 IN6_IS_ADDR_UNSPECIFIED(&oconnp->conn_laddr_v6) || 1026 IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_laddr_v6) || 1027 IN6_IS_ADDR_V4MAPPED_ANY(&oconnp->conn_laddr_v6)) || 1028 IN6_ARE_ADDR_EQUAL(&oconnp->conn_laddr_v6, 1029 &connp->conn_laddr_v6))) { 1030 break; 1031 } 1032 } 1033 mutex_exit(&connfp->connf_lock); 1034 if (oconnp != NULL) 1035 return (EADDRNOTAVAIL); 1036 1037 if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) || 1038 IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) { 1039 if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) || 1040 IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_laddr_v6)) { 1041 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 1042 } else { 1043 IPCL_HASH_INSERT_BOUND(connfp, connp); 1044 } 1045 } else { 1046 IPCL_HASH_INSERT_CONNECTED(connfp, connp); 1047 } 1048 return (0); 1049 } 1050 1051 static int 1052 ipcl_iptun_hash_insert(conn_t *connp, ip_stack_t *ipst) 1053 { 1054 connf_t *connfp; 1055 conn_t *tconnp; 1056 ipaddr_t laddr = connp->conn_laddr_v4; 1057 ipaddr_t faddr = connp->conn_faddr_v4; 1058 1059 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH(laddr, faddr)]; 1060 mutex_enter(&connfp->connf_lock); 1061 for (tconnp = connfp->connf_head; tconnp != NULL; 1062 tconnp = tconnp->conn_next) { 1063 if (IPCL_IPTUN_MATCH(tconnp, laddr, faddr)) { 1064 /* A tunnel is already bound to these addresses. */ 1065 mutex_exit(&connfp->connf_lock); 1066 return (EADDRINUSE); 1067 } 1068 } 1069 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp); 1070 mutex_exit(&connfp->connf_lock); 1071 return (0); 1072 } 1073 1074 static int 1075 ipcl_iptun_hash_insert_v6(conn_t *connp, ip_stack_t *ipst) 1076 { 1077 connf_t *connfp; 1078 conn_t *tconnp; 1079 in6_addr_t *laddr = &connp->conn_laddr_v6; 1080 in6_addr_t *faddr = &connp->conn_faddr_v6; 1081 1082 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH_V6(laddr, faddr)]; 1083 mutex_enter(&connfp->connf_lock); 1084 for (tconnp = connfp->connf_head; tconnp != NULL; 1085 tconnp = tconnp->conn_next) { 1086 if (IPCL_IPTUN_MATCH_V6(tconnp, laddr, faddr)) { 1087 /* A tunnel is already bound to these addresses. */ 1088 mutex_exit(&connfp->connf_lock); 1089 return (EADDRINUSE); 1090 } 1091 } 1092 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp); 1093 mutex_exit(&connfp->connf_lock); 1094 return (0); 1095 } 1096 1097 /* 1098 * Check for a MAC exemption conflict on a labeled system. Note that for 1099 * protocols that use port numbers (UDP, TCP, SCTP), we do this check up in the 1100 * transport layer. This check is for binding all other protocols. 1101 * 1102 * Returns true if there's a conflict. 1103 */ 1104 static boolean_t 1105 check_exempt_conflict_v4(conn_t *connp, ip_stack_t *ipst) 1106 { 1107 connf_t *connfp; 1108 conn_t *tconn; 1109 1110 connfp = &ipst->ips_ipcl_proto_fanout_v4[connp->conn_proto]; 1111 mutex_enter(&connfp->connf_lock); 1112 for (tconn = connfp->connf_head; tconn != NULL; 1113 tconn = tconn->conn_next) { 1114 /* We don't allow v4 fallback for v6 raw socket */ 1115 if (connp->conn_family != tconn->conn_family) 1116 continue; 1117 /* If neither is exempt, then there's no conflict */ 1118 if ((connp->conn_mac_mode == CONN_MAC_DEFAULT) && 1119 (tconn->conn_mac_mode == CONN_MAC_DEFAULT)) 1120 continue; 1121 /* We are only concerned about sockets for a different zone */ 1122 if (connp->conn_zoneid == tconn->conn_zoneid) 1123 continue; 1124 /* If both are bound to different specific addrs, ok */ 1125 if (connp->conn_laddr_v4 != INADDR_ANY && 1126 tconn->conn_laddr_v4 != INADDR_ANY && 1127 connp->conn_laddr_v4 != tconn->conn_laddr_v4) 1128 continue; 1129 /* These two conflict; fail */ 1130 break; 1131 } 1132 mutex_exit(&connfp->connf_lock); 1133 return (tconn != NULL); 1134 } 1135 1136 static boolean_t 1137 check_exempt_conflict_v6(conn_t *connp, ip_stack_t *ipst) 1138 { 1139 connf_t *connfp; 1140 conn_t *tconn; 1141 1142 connfp = &ipst->ips_ipcl_proto_fanout_v6[connp->conn_proto]; 1143 mutex_enter(&connfp->connf_lock); 1144 for (tconn = connfp->connf_head; tconn != NULL; 1145 tconn = tconn->conn_next) { 1146 /* We don't allow v4 fallback for v6 raw socket */ 1147 if (connp->conn_family != tconn->conn_family) 1148 continue; 1149 /* If neither is exempt, then there's no conflict */ 1150 if ((connp->conn_mac_mode == CONN_MAC_DEFAULT) && 1151 (tconn->conn_mac_mode == CONN_MAC_DEFAULT)) 1152 continue; 1153 /* We are only concerned about sockets for a different zone */ 1154 if (connp->conn_zoneid == tconn->conn_zoneid) 1155 continue; 1156 /* If both are bound to different addrs, ok */ 1157 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) && 1158 !IN6_IS_ADDR_UNSPECIFIED(&tconn->conn_laddr_v6) && 1159 !IN6_ARE_ADDR_EQUAL(&connp->conn_laddr_v6, 1160 &tconn->conn_laddr_v6)) 1161 continue; 1162 /* These two conflict; fail */ 1163 break; 1164 } 1165 mutex_exit(&connfp->connf_lock); 1166 return (tconn != NULL); 1167 } 1168 1169 /* 1170 * (v4, v6) bind hash insertion routines 1171 * The caller has already setup the conn (conn_proto, conn_laddr_v6, conn_lport) 1172 */ 1173 1174 int 1175 ipcl_bind_insert(conn_t *connp) 1176 { 1177 if (connp->conn_ipversion == IPV6_VERSION) 1178 return (ipcl_bind_insert_v6(connp)); 1179 else 1180 return (ipcl_bind_insert_v4(connp)); 1181 } 1182 1183 int 1184 ipcl_bind_insert_v4(conn_t *connp) 1185 { 1186 connf_t *connfp; 1187 int ret = 0; 1188 ip_stack_t *ipst = connp->conn_netstack->netstack_ip; 1189 uint16_t lport = connp->conn_lport; 1190 uint8_t protocol = connp->conn_proto; 1191 1192 if (IPCL_IS_IPTUN(connp)) 1193 return (ipcl_iptun_hash_insert(connp, ipst)); 1194 1195 switch (protocol) { 1196 default: 1197 if (is_system_labeled() && 1198 check_exempt_conflict_v4(connp, ipst)) 1199 return (EADDRINUSE); 1200 /* FALLTHROUGH */ 1201 case IPPROTO_UDP: 1202 if (protocol == IPPROTO_UDP) { 1203 connfp = &ipst->ips_ipcl_udp_fanout[ 1204 IPCL_UDP_HASH(lport, ipst)]; 1205 } else { 1206 connfp = &ipst->ips_ipcl_proto_fanout_v4[protocol]; 1207 } 1208 1209 if (connp->conn_faddr_v4 != INADDR_ANY) { 1210 IPCL_HASH_INSERT_CONNECTED(connfp, connp); 1211 } else if (connp->conn_laddr_v4 != INADDR_ANY) { 1212 IPCL_HASH_INSERT_BOUND(connfp, connp); 1213 } else { 1214 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 1215 } 1216 if (protocol == IPPROTO_RSVP) 1217 ill_set_inputfn_all(ipst); 1218 break; 1219 1220 case IPPROTO_TCP: 1221 /* Insert it in the Bind Hash */ 1222 ASSERT(connp->conn_zoneid != ALL_ZONES); 1223 connfp = &ipst->ips_ipcl_bind_fanout[ 1224 IPCL_BIND_HASH(lport, ipst)]; 1225 if (connp->conn_laddr_v4 != INADDR_ANY) { 1226 IPCL_HASH_INSERT_BOUND(connfp, connp); 1227 } else { 1228 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 1229 } 1230 if (cl_inet_listen != NULL) { 1231 ASSERT(connp->conn_ipversion == IPV4_VERSION); 1232 connp->conn_flags |= IPCL_CL_LISTENER; 1233 (*cl_inet_listen)( 1234 connp->conn_netstack->netstack_stackid, 1235 IPPROTO_TCP, AF_INET, 1236 (uint8_t *)&connp->conn_bound_addr_v4, lport, NULL); 1237 } 1238 break; 1239 1240 case IPPROTO_SCTP: 1241 ret = ipcl_sctp_hash_insert(connp, lport); 1242 break; 1243 } 1244 1245 return (ret); 1246 } 1247 1248 int 1249 ipcl_bind_insert_v6(conn_t *connp) 1250 { 1251 connf_t *connfp; 1252 int ret = 0; 1253 ip_stack_t *ipst = connp->conn_netstack->netstack_ip; 1254 uint16_t lport = connp->conn_lport; 1255 uint8_t protocol = connp->conn_proto; 1256 1257 if (IPCL_IS_IPTUN(connp)) { 1258 return (ipcl_iptun_hash_insert_v6(connp, ipst)); 1259 } 1260 1261 switch (protocol) { 1262 default: 1263 if (is_system_labeled() && 1264 check_exempt_conflict_v6(connp, ipst)) 1265 return (EADDRINUSE); 1266 /* FALLTHROUGH */ 1267 case IPPROTO_UDP: 1268 if (protocol == IPPROTO_UDP) { 1269 connfp = &ipst->ips_ipcl_udp_fanout[ 1270 IPCL_UDP_HASH(lport, ipst)]; 1271 } else { 1272 connfp = &ipst->ips_ipcl_proto_fanout_v6[protocol]; 1273 } 1274 1275 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6)) { 1276 IPCL_HASH_INSERT_CONNECTED(connfp, connp); 1277 } else if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6)) { 1278 IPCL_HASH_INSERT_BOUND(connfp, connp); 1279 } else { 1280 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 1281 } 1282 break; 1283 1284 case IPPROTO_TCP: 1285 /* Insert it in the Bind Hash */ 1286 ASSERT(connp->conn_zoneid != ALL_ZONES); 1287 connfp = &ipst->ips_ipcl_bind_fanout[ 1288 IPCL_BIND_HASH(lport, ipst)]; 1289 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6)) { 1290 IPCL_HASH_INSERT_BOUND(connfp, connp); 1291 } else { 1292 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 1293 } 1294 if (cl_inet_listen != NULL) { 1295 sa_family_t addr_family; 1296 uint8_t *laddrp; 1297 1298 if (connp->conn_ipversion == IPV6_VERSION) { 1299 addr_family = AF_INET6; 1300 laddrp = 1301 (uint8_t *)&connp->conn_bound_addr_v6; 1302 } else { 1303 addr_family = AF_INET; 1304 laddrp = (uint8_t *)&connp->conn_bound_addr_v4; 1305 } 1306 connp->conn_flags |= IPCL_CL_LISTENER; 1307 (*cl_inet_listen)( 1308 connp->conn_netstack->netstack_stackid, 1309 IPPROTO_TCP, addr_family, laddrp, lport, NULL); 1310 } 1311 break; 1312 1313 case IPPROTO_SCTP: 1314 ret = ipcl_sctp_hash_insert(connp, lport); 1315 break; 1316 } 1317 1318 return (ret); 1319 } 1320 1321 /* 1322 * ipcl_conn_hash insertion routines. 1323 * The caller has already set conn_proto and the addresses/ports in the conn_t. 1324 */ 1325 1326 int 1327 ipcl_conn_insert(conn_t *connp) 1328 { 1329 if (connp->conn_ipversion == IPV6_VERSION) 1330 return (ipcl_conn_insert_v6(connp)); 1331 else 1332 return (ipcl_conn_insert_v4(connp)); 1333 } 1334 1335 int 1336 ipcl_conn_insert_v4(conn_t *connp) 1337 { 1338 connf_t *connfp; 1339 conn_t *tconnp; 1340 int ret = 0; 1341 ip_stack_t *ipst = connp->conn_netstack->netstack_ip; 1342 uint16_t lport = connp->conn_lport; 1343 uint8_t protocol = connp->conn_proto; 1344 1345 if (IPCL_IS_IPTUN(connp)) 1346 return (ipcl_iptun_hash_insert(connp, ipst)); 1347 1348 switch (protocol) { 1349 case IPPROTO_TCP: 1350 /* 1351 * For TCP, we check whether the connection tuple already 1352 * exists before allowing the connection to proceed. We 1353 * also allow indexing on the zoneid. This is to allow 1354 * multiple shared stack zones to have the same tcp 1355 * connection tuple. In practice this only happens for 1356 * INADDR_LOOPBACK as it's the only local address which 1357 * doesn't have to be unique. 1358 */ 1359 connfp = &ipst->ips_ipcl_conn_fanout[ 1360 IPCL_CONN_HASH(connp->conn_faddr_v4, 1361 connp->conn_ports, ipst)]; 1362 mutex_enter(&connfp->connf_lock); 1363 for (tconnp = connfp->connf_head; tconnp != NULL; 1364 tconnp = tconnp->conn_next) { 1365 if (IPCL_CONN_MATCH(tconnp, connp->conn_proto, 1366 connp->conn_faddr_v4, connp->conn_laddr_v4, 1367 connp->conn_ports) && 1368 IPCL_ZONE_MATCH(tconnp, connp->conn_zoneid)) { 1369 /* Already have a conn. bail out */ 1370 mutex_exit(&connfp->connf_lock); 1371 return (EADDRINUSE); 1372 } 1373 } 1374 if (connp->conn_fanout != NULL) { 1375 /* 1376 * Probably a XTI/TLI application trying to do a 1377 * rebind. Let it happen. 1378 */ 1379 mutex_exit(&connfp->connf_lock); 1380 IPCL_HASH_REMOVE(connp); 1381 mutex_enter(&connfp->connf_lock); 1382 } 1383 1384 ASSERT(connp->conn_recv != NULL); 1385 ASSERT(connp->conn_recvicmp != NULL); 1386 1387 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp); 1388 mutex_exit(&connfp->connf_lock); 1389 break; 1390 1391 case IPPROTO_SCTP: 1392 /* 1393 * The raw socket may have already been bound, remove it 1394 * from the hash first. 1395 */ 1396 IPCL_HASH_REMOVE(connp); 1397 ret = ipcl_sctp_hash_insert(connp, lport); 1398 break; 1399 1400 default: 1401 /* 1402 * Check for conflicts among MAC exempt bindings. For 1403 * transports with port numbers, this is done by the upper 1404 * level per-transport binding logic. For all others, it's 1405 * done here. 1406 */ 1407 if (is_system_labeled() && 1408 check_exempt_conflict_v4(connp, ipst)) 1409 return (EADDRINUSE); 1410 /* FALLTHROUGH */ 1411 1412 case IPPROTO_UDP: 1413 if (protocol == IPPROTO_UDP) { 1414 connfp = &ipst->ips_ipcl_udp_fanout[ 1415 IPCL_UDP_HASH(lport, ipst)]; 1416 } else { 1417 connfp = &ipst->ips_ipcl_proto_fanout_v4[protocol]; 1418 } 1419 1420 if (connp->conn_faddr_v4 != INADDR_ANY) { 1421 IPCL_HASH_INSERT_CONNECTED(connfp, connp); 1422 } else if (connp->conn_laddr_v4 != INADDR_ANY) { 1423 IPCL_HASH_INSERT_BOUND(connfp, connp); 1424 } else { 1425 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 1426 } 1427 break; 1428 } 1429 1430 return (ret); 1431 } 1432 1433 int 1434 ipcl_conn_insert_v6(conn_t *connp) 1435 { 1436 connf_t *connfp; 1437 conn_t *tconnp; 1438 int ret = 0; 1439 ip_stack_t *ipst = connp->conn_netstack->netstack_ip; 1440 uint16_t lport = connp->conn_lport; 1441 uint8_t protocol = connp->conn_proto; 1442 uint_t ifindex = connp->conn_bound_if; 1443 1444 if (IPCL_IS_IPTUN(connp)) 1445 return (ipcl_iptun_hash_insert_v6(connp, ipst)); 1446 1447 switch (protocol) { 1448 case IPPROTO_TCP: 1449 1450 /* 1451 * For tcp, we check whether the connection tuple already 1452 * exists before allowing the connection to proceed. We 1453 * also allow indexing on the zoneid. This is to allow 1454 * multiple shared stack zones to have the same tcp 1455 * connection tuple. In practice this only happens for 1456 * ipv6_loopback as it's the only local address which 1457 * doesn't have to be unique. 1458 */ 1459 connfp = &ipst->ips_ipcl_conn_fanout[ 1460 IPCL_CONN_HASH_V6(connp->conn_faddr_v6, connp->conn_ports, 1461 ipst)]; 1462 mutex_enter(&connfp->connf_lock); 1463 for (tconnp = connfp->connf_head; tconnp != NULL; 1464 tconnp = tconnp->conn_next) { 1465 /* NOTE: need to match zoneid. Bug in onnv-gate */ 1466 if (IPCL_CONN_MATCH_V6(tconnp, connp->conn_proto, 1467 connp->conn_faddr_v6, connp->conn_laddr_v6, 1468 connp->conn_ports) && 1469 (tconnp->conn_bound_if == 0 || 1470 tconnp->conn_bound_if == ifindex) && 1471 IPCL_ZONE_MATCH(tconnp, connp->conn_zoneid)) { 1472 /* Already have a conn. bail out */ 1473 mutex_exit(&connfp->connf_lock); 1474 return (EADDRINUSE); 1475 } 1476 } 1477 if (connp->conn_fanout != NULL) { 1478 /* 1479 * Probably a XTI/TLI application trying to do a 1480 * rebind. Let it happen. 1481 */ 1482 mutex_exit(&connfp->connf_lock); 1483 IPCL_HASH_REMOVE(connp); 1484 mutex_enter(&connfp->connf_lock); 1485 } 1486 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp); 1487 mutex_exit(&connfp->connf_lock); 1488 break; 1489 1490 case IPPROTO_SCTP: 1491 IPCL_HASH_REMOVE(connp); 1492 ret = ipcl_sctp_hash_insert(connp, lport); 1493 break; 1494 1495 default: 1496 if (is_system_labeled() && 1497 check_exempt_conflict_v6(connp, ipst)) 1498 return (EADDRINUSE); 1499 /* FALLTHROUGH */ 1500 case IPPROTO_UDP: 1501 if (protocol == IPPROTO_UDP) { 1502 connfp = &ipst->ips_ipcl_udp_fanout[ 1503 IPCL_UDP_HASH(lport, ipst)]; 1504 } else { 1505 connfp = &ipst->ips_ipcl_proto_fanout_v6[protocol]; 1506 } 1507 1508 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6)) { 1509 IPCL_HASH_INSERT_CONNECTED(connfp, connp); 1510 } else if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6)) { 1511 IPCL_HASH_INSERT_BOUND(connfp, connp); 1512 } else { 1513 IPCL_HASH_INSERT_WILDCARD(connfp, connp); 1514 } 1515 break; 1516 } 1517 1518 return (ret); 1519 } 1520 1521 /* 1522 * v4 packet classifying function. looks up the fanout table to 1523 * find the conn, the packet belongs to. returns the conn with 1524 * the reference held, null otherwise. 1525 * 1526 * If zoneid is ALL_ZONES, then the search rules described in the "Connection 1527 * Lookup" comment block are applied. Labels are also checked as described 1528 * above. If the packet is from the inside (looped back), and is from the same 1529 * zone, then label checks are omitted. 1530 */ 1531 conn_t * 1532 ipcl_classify_v4(mblk_t *mp, uint8_t protocol, uint_t hdr_len, 1533 ip_recv_attr_t *ira, ip_stack_t *ipst) 1534 { 1535 ipha_t *ipha; 1536 connf_t *connfp, *bind_connfp; 1537 uint16_t lport; 1538 uint16_t fport; 1539 uint32_t ports; 1540 conn_t *connp; 1541 uint16_t *up; 1542 zoneid_t zoneid = ira->ira_zoneid; 1543 1544 ipha = (ipha_t *)mp->b_rptr; 1545 up = (uint16_t *)((uchar_t *)ipha + hdr_len + TCP_PORTS_OFFSET); 1546 1547 switch (protocol) { 1548 case IPPROTO_TCP: 1549 ports = *(uint32_t *)up; 1550 connfp = 1551 &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_src, 1552 ports, ipst)]; 1553 mutex_enter(&connfp->connf_lock); 1554 for (connp = connfp->connf_head; connp != NULL; 1555 connp = connp->conn_next) { 1556 if (IPCL_CONN_MATCH(connp, protocol, 1557 ipha->ipha_src, ipha->ipha_dst, ports) && 1558 (connp->conn_zoneid == zoneid || 1559 connp->conn_allzones || 1560 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) && 1561 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) && 1562 (ira->ira_flags & IRAF_TX_SHARED_ADDR)))) 1563 break; 1564 } 1565 1566 if (connp != NULL) { 1567 /* 1568 * We have a fully-bound TCP connection. 1569 * 1570 * For labeled systems, there's no need to check the 1571 * label here. It's known to be good as we checked 1572 * before allowing the connection to become bound. 1573 */ 1574 CONN_INC_REF(connp); 1575 mutex_exit(&connfp->connf_lock); 1576 return (connp); 1577 } 1578 1579 mutex_exit(&connfp->connf_lock); 1580 lport = up[1]; 1581 bind_connfp = 1582 &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)]; 1583 mutex_enter(&bind_connfp->connf_lock); 1584 for (connp = bind_connfp->connf_head; connp != NULL; 1585 connp = connp->conn_next) { 1586 if (IPCL_BIND_MATCH(connp, protocol, ipha->ipha_dst, 1587 lport) && 1588 (connp->conn_zoneid == zoneid || 1589 connp->conn_allzones || 1590 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) && 1591 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) && 1592 (ira->ira_flags & IRAF_TX_SHARED_ADDR)))) 1593 break; 1594 } 1595 1596 /* 1597 * If the matching connection is SLP on a private address, then 1598 * the label on the packet must match the local zone's label. 1599 * Otherwise, it must be in the label range defined by tnrh. 1600 * This is ensured by tsol_receive_local. 1601 * 1602 * Note that we don't check tsol_receive_local for 1603 * the connected case. 1604 */ 1605 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) && 1606 !tsol_receive_local(mp, &ipha->ipha_dst, IPV4_VERSION, 1607 ira, connp)) { 1608 DTRACE_PROBE3(tx__ip__log__info__classify__tcp, 1609 char *, "connp(1) could not receive mp(2)", 1610 conn_t *, connp, mblk_t *, mp); 1611 connp = NULL; 1612 } 1613 1614 if (connp != NULL) { 1615 /* Have a listener at least */ 1616 CONN_INC_REF(connp); 1617 mutex_exit(&bind_connfp->connf_lock); 1618 return (connp); 1619 } 1620 1621 mutex_exit(&bind_connfp->connf_lock); 1622 break; 1623 1624 case IPPROTO_UDP: 1625 lport = up[1]; 1626 fport = up[0]; 1627 connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)]; 1628 mutex_enter(&connfp->connf_lock); 1629 for (connp = connfp->connf_head; connp != NULL; 1630 connp = connp->conn_next) { 1631 if (IPCL_UDP_MATCH(connp, lport, ipha->ipha_dst, 1632 fport, ipha->ipha_src) && 1633 (connp->conn_zoneid == zoneid || 1634 connp->conn_allzones || 1635 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) && 1636 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE)))) 1637 break; 1638 } 1639 1640 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) && 1641 !tsol_receive_local(mp, &ipha->ipha_dst, IPV4_VERSION, 1642 ira, connp)) { 1643 DTRACE_PROBE3(tx__ip__log__info__classify__udp, 1644 char *, "connp(1) could not receive mp(2)", 1645 conn_t *, connp, mblk_t *, mp); 1646 connp = NULL; 1647 } 1648 1649 if (connp != NULL) { 1650 CONN_INC_REF(connp); 1651 mutex_exit(&connfp->connf_lock); 1652 return (connp); 1653 } 1654 1655 /* 1656 * We shouldn't come here for multicast/broadcast packets 1657 */ 1658 mutex_exit(&connfp->connf_lock); 1659 1660 break; 1661 1662 case IPPROTO_ENCAP: 1663 case IPPROTO_IPV6: 1664 return (ipcl_iptun_classify_v4(&ipha->ipha_src, 1665 &ipha->ipha_dst, ipst)); 1666 } 1667 1668 return (NULL); 1669 } 1670 1671 conn_t * 1672 ipcl_classify_v6(mblk_t *mp, uint8_t protocol, uint_t hdr_len, 1673 ip_recv_attr_t *ira, ip_stack_t *ipst) 1674 { 1675 ip6_t *ip6h; 1676 connf_t *connfp, *bind_connfp; 1677 uint16_t lport; 1678 uint16_t fport; 1679 tcpha_t *tcpha; 1680 uint32_t ports; 1681 conn_t *connp; 1682 uint16_t *up; 1683 zoneid_t zoneid = ira->ira_zoneid; 1684 1685 ip6h = (ip6_t *)mp->b_rptr; 1686 1687 switch (protocol) { 1688 case IPPROTO_TCP: 1689 tcpha = (tcpha_t *)&mp->b_rptr[hdr_len]; 1690 up = &tcpha->tha_lport; 1691 ports = *(uint32_t *)up; 1692 1693 connfp = 1694 &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_src, 1695 ports, ipst)]; 1696 mutex_enter(&connfp->connf_lock); 1697 for (connp = connfp->connf_head; connp != NULL; 1698 connp = connp->conn_next) { 1699 if (IPCL_CONN_MATCH_V6(connp, protocol, 1700 ip6h->ip6_src, ip6h->ip6_dst, ports) && 1701 (connp->conn_zoneid == zoneid || 1702 connp->conn_allzones || 1703 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) && 1704 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) && 1705 (ira->ira_flags & IRAF_TX_SHARED_ADDR)))) 1706 break; 1707 } 1708 1709 if (connp != NULL) { 1710 /* 1711 * We have a fully-bound TCP connection. 1712 * 1713 * For labeled systems, there's no need to check the 1714 * label here. It's known to be good as we checked 1715 * before allowing the connection to become bound. 1716 */ 1717 CONN_INC_REF(connp); 1718 mutex_exit(&connfp->connf_lock); 1719 return (connp); 1720 } 1721 1722 mutex_exit(&connfp->connf_lock); 1723 1724 lport = up[1]; 1725 bind_connfp = 1726 &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)]; 1727 mutex_enter(&bind_connfp->connf_lock); 1728 for (connp = bind_connfp->connf_head; connp != NULL; 1729 connp = connp->conn_next) { 1730 if (IPCL_BIND_MATCH_V6(connp, protocol, 1731 ip6h->ip6_dst, lport) && 1732 (connp->conn_zoneid == zoneid || 1733 connp->conn_allzones || 1734 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) && 1735 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) && 1736 (ira->ira_flags & IRAF_TX_SHARED_ADDR)))) 1737 break; 1738 } 1739 1740 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) && 1741 !tsol_receive_local(mp, &ip6h->ip6_dst, IPV6_VERSION, 1742 ira, connp)) { 1743 DTRACE_PROBE3(tx__ip__log__info__classify__tcp6, 1744 char *, "connp(1) could not receive mp(2)", 1745 conn_t *, connp, mblk_t *, mp); 1746 connp = NULL; 1747 } 1748 1749 if (connp != NULL) { 1750 /* Have a listner at least */ 1751 CONN_INC_REF(connp); 1752 mutex_exit(&bind_connfp->connf_lock); 1753 return (connp); 1754 } 1755 1756 mutex_exit(&bind_connfp->connf_lock); 1757 break; 1758 1759 case IPPROTO_UDP: 1760 up = (uint16_t *)&mp->b_rptr[hdr_len]; 1761 lport = up[1]; 1762 fport = up[0]; 1763 connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)]; 1764 mutex_enter(&connfp->connf_lock); 1765 for (connp = connfp->connf_head; connp != NULL; 1766 connp = connp->conn_next) { 1767 if (IPCL_UDP_MATCH_V6(connp, lport, ip6h->ip6_dst, 1768 fport, ip6h->ip6_src) && 1769 (connp->conn_zoneid == zoneid || 1770 connp->conn_allzones || 1771 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) && 1772 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) && 1773 (ira->ira_flags & IRAF_TX_SHARED_ADDR)))) 1774 break; 1775 } 1776 1777 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) && 1778 !tsol_receive_local(mp, &ip6h->ip6_dst, IPV6_VERSION, 1779 ira, connp)) { 1780 DTRACE_PROBE3(tx__ip__log__info__classify__udp6, 1781 char *, "connp(1) could not receive mp(2)", 1782 conn_t *, connp, mblk_t *, mp); 1783 connp = NULL; 1784 } 1785 1786 if (connp != NULL) { 1787 CONN_INC_REF(connp); 1788 mutex_exit(&connfp->connf_lock); 1789 return (connp); 1790 } 1791 1792 /* 1793 * We shouldn't come here for multicast/broadcast packets 1794 */ 1795 mutex_exit(&connfp->connf_lock); 1796 break; 1797 case IPPROTO_ENCAP: 1798 case IPPROTO_IPV6: 1799 return (ipcl_iptun_classify_v6(&ip6h->ip6_src, 1800 &ip6h->ip6_dst, ipst)); 1801 } 1802 1803 return (NULL); 1804 } 1805 1806 /* 1807 * wrapper around ipcl_classify_(v4,v6) routines. 1808 */ 1809 conn_t * 1810 ipcl_classify(mblk_t *mp, ip_recv_attr_t *ira, ip_stack_t *ipst) 1811 { 1812 if (ira->ira_flags & IRAF_IS_IPV4) { 1813 return (ipcl_classify_v4(mp, ira->ira_protocol, 1814 ira->ira_ip_hdr_length, ira, ipst)); 1815 } else { 1816 return (ipcl_classify_v6(mp, ira->ira_protocol, 1817 ira->ira_ip_hdr_length, ira, ipst)); 1818 } 1819 } 1820 1821 /* 1822 * Only used to classify SCTP RAW sockets 1823 */ 1824 conn_t * 1825 ipcl_classify_raw(mblk_t *mp, uint8_t protocol, uint32_t ports, 1826 ipha_t *ipha, ip6_t *ip6h, ip_recv_attr_t *ira, ip_stack_t *ipst) 1827 { 1828 connf_t *connfp; 1829 conn_t *connp; 1830 in_port_t lport; 1831 int ipversion; 1832 const void *dst; 1833 zoneid_t zoneid = ira->ira_zoneid; 1834 1835 lport = ((uint16_t *)&ports)[1]; 1836 if (ira->ira_flags & IRAF_IS_IPV4) { 1837 dst = (const void *)&ipha->ipha_dst; 1838 ipversion = IPV4_VERSION; 1839 } else { 1840 dst = (const void *)&ip6h->ip6_dst; 1841 ipversion = IPV6_VERSION; 1842 } 1843 1844 connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(ntohs(lport), ipst)]; 1845 mutex_enter(&connfp->connf_lock); 1846 for (connp = connfp->connf_head; connp != NULL; 1847 connp = connp->conn_next) { 1848 /* We don't allow v4 fallback for v6 raw socket. */ 1849 if (ipversion != connp->conn_ipversion) 1850 continue; 1851 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) && 1852 !IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) { 1853 if (ipversion == IPV4_VERSION) { 1854 if (!IPCL_CONN_MATCH(connp, protocol, 1855 ipha->ipha_src, ipha->ipha_dst, ports)) 1856 continue; 1857 } else { 1858 if (!IPCL_CONN_MATCH_V6(connp, protocol, 1859 ip6h->ip6_src, ip6h->ip6_dst, ports)) 1860 continue; 1861 } 1862 } else { 1863 if (ipversion == IPV4_VERSION) { 1864 if (!IPCL_BIND_MATCH(connp, protocol, 1865 ipha->ipha_dst, lport)) 1866 continue; 1867 } else { 1868 if (!IPCL_BIND_MATCH_V6(connp, protocol, 1869 ip6h->ip6_dst, lport)) 1870 continue; 1871 } 1872 } 1873 1874 if (connp->conn_zoneid == zoneid || 1875 connp->conn_allzones || 1876 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) && 1877 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) && 1878 (ira->ira_flags & IRAF_TX_SHARED_ADDR))) 1879 break; 1880 } 1881 1882 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) && 1883 !tsol_receive_local(mp, dst, ipversion, ira, connp)) { 1884 DTRACE_PROBE3(tx__ip__log__info__classify__rawip, 1885 char *, "connp(1) could not receive mp(2)", 1886 conn_t *, connp, mblk_t *, mp); 1887 connp = NULL; 1888 } 1889 1890 if (connp != NULL) 1891 goto found; 1892 mutex_exit(&connfp->connf_lock); 1893 1894 /* Try to look for a wildcard SCTP RAW socket match. */ 1895 connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(0, ipst)]; 1896 mutex_enter(&connfp->connf_lock); 1897 for (connp = connfp->connf_head; connp != NULL; 1898 connp = connp->conn_next) { 1899 /* We don't allow v4 fallback for v6 raw socket. */ 1900 if (ipversion != connp->conn_ipversion) 1901 continue; 1902 if (!IPCL_ZONE_MATCH(connp, zoneid)) 1903 continue; 1904 1905 if (ipversion == IPV4_VERSION) { 1906 if (IPCL_RAW_MATCH(connp, protocol, ipha->ipha_dst)) 1907 break; 1908 } else { 1909 if (IPCL_RAW_MATCH_V6(connp, protocol, ip6h->ip6_dst)) { 1910 break; 1911 } 1912 } 1913 } 1914 1915 if (connp != NULL) 1916 goto found; 1917 1918 mutex_exit(&connfp->connf_lock); 1919 return (NULL); 1920 1921 found: 1922 ASSERT(connp != NULL); 1923 CONN_INC_REF(connp); 1924 mutex_exit(&connfp->connf_lock); 1925 return (connp); 1926 } 1927 1928 /* ARGSUSED */ 1929 static int 1930 tcp_conn_constructor(void *buf, void *cdrarg, int kmflags) 1931 { 1932 itc_t *itc = (itc_t *)buf; 1933 conn_t *connp = &itc->itc_conn; 1934 tcp_t *tcp = (tcp_t *)&itc[1]; 1935 1936 bzero(connp, sizeof (conn_t)); 1937 bzero(tcp, sizeof (tcp_t)); 1938 1939 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL); 1940 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL); 1941 cv_init(&connp->conn_sq_cv, NULL, CV_DEFAULT, NULL); 1942 tcp->tcp_timercache = tcp_timermp_alloc(kmflags); 1943 if (tcp->tcp_timercache == NULL) 1944 return (ENOMEM); 1945 connp->conn_tcp = tcp; 1946 connp->conn_flags = IPCL_TCPCONN; 1947 connp->conn_proto = IPPROTO_TCP; 1948 tcp->tcp_connp = connp; 1949 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL); 1950 1951 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags); 1952 if (connp->conn_ixa == NULL) { 1953 tcp_timermp_free(tcp); 1954 return (ENOMEM); 1955 } 1956 connp->conn_ixa->ixa_refcnt = 1; 1957 connp->conn_ixa->ixa_protocol = connp->conn_proto; 1958 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp); 1959 return (0); 1960 } 1961 1962 /* ARGSUSED */ 1963 static void 1964 tcp_conn_destructor(void *buf, void *cdrarg) 1965 { 1966 itc_t *itc = (itc_t *)buf; 1967 conn_t *connp = &itc->itc_conn; 1968 tcp_t *tcp = (tcp_t *)&itc[1]; 1969 1970 ASSERT(connp->conn_flags & IPCL_TCPCONN); 1971 ASSERT(tcp->tcp_connp == connp); 1972 ASSERT(connp->conn_tcp == tcp); 1973 tcp_timermp_free(tcp); 1974 mutex_destroy(&connp->conn_lock); 1975 cv_destroy(&connp->conn_cv); 1976 cv_destroy(&connp->conn_sq_cv); 1977 rw_destroy(&connp->conn_ilg_lock); 1978 1979 /* Can be NULL if constructor failed */ 1980 if (connp->conn_ixa != NULL) { 1981 ASSERT(connp->conn_ixa->ixa_refcnt == 1); 1982 ASSERT(connp->conn_ixa->ixa_ire == NULL); 1983 ASSERT(connp->conn_ixa->ixa_nce == NULL); 1984 ixa_refrele(connp->conn_ixa); 1985 } 1986 } 1987 1988 /* ARGSUSED */ 1989 static int 1990 ip_conn_constructor(void *buf, void *cdrarg, int kmflags) 1991 { 1992 itc_t *itc = (itc_t *)buf; 1993 conn_t *connp = &itc->itc_conn; 1994 1995 bzero(connp, sizeof (conn_t)); 1996 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL); 1997 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL); 1998 connp->conn_flags = IPCL_IPCCONN; 1999 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL); 2000 2001 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags); 2002 if (connp->conn_ixa == NULL) 2003 return (ENOMEM); 2004 connp->conn_ixa->ixa_refcnt = 1; 2005 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp); 2006 return (0); 2007 } 2008 2009 /* ARGSUSED */ 2010 static void 2011 ip_conn_destructor(void *buf, void *cdrarg) 2012 { 2013 itc_t *itc = (itc_t *)buf; 2014 conn_t *connp = &itc->itc_conn; 2015 2016 ASSERT(connp->conn_flags & IPCL_IPCCONN); 2017 ASSERT(connp->conn_priv == NULL); 2018 mutex_destroy(&connp->conn_lock); 2019 cv_destroy(&connp->conn_cv); 2020 rw_destroy(&connp->conn_ilg_lock); 2021 2022 /* Can be NULL if constructor failed */ 2023 if (connp->conn_ixa != NULL) { 2024 ASSERT(connp->conn_ixa->ixa_refcnt == 1); 2025 ASSERT(connp->conn_ixa->ixa_ire == NULL); 2026 ASSERT(connp->conn_ixa->ixa_nce == NULL); 2027 ixa_refrele(connp->conn_ixa); 2028 } 2029 } 2030 2031 /* ARGSUSED */ 2032 static int 2033 udp_conn_constructor(void *buf, void *cdrarg, int kmflags) 2034 { 2035 itc_t *itc = (itc_t *)buf; 2036 conn_t *connp = &itc->itc_conn; 2037 udp_t *udp = (udp_t *)&itc[1]; 2038 2039 bzero(connp, sizeof (conn_t)); 2040 bzero(udp, sizeof (udp_t)); 2041 2042 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL); 2043 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL); 2044 connp->conn_udp = udp; 2045 connp->conn_flags = IPCL_UDPCONN; 2046 connp->conn_proto = IPPROTO_UDP; 2047 udp->udp_connp = connp; 2048 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL); 2049 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags); 2050 if (connp->conn_ixa == NULL) 2051 return (ENOMEM); 2052 connp->conn_ixa->ixa_refcnt = 1; 2053 connp->conn_ixa->ixa_protocol = connp->conn_proto; 2054 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp); 2055 return (0); 2056 } 2057 2058 /* ARGSUSED */ 2059 static void 2060 udp_conn_destructor(void *buf, void *cdrarg) 2061 { 2062 itc_t *itc = (itc_t *)buf; 2063 conn_t *connp = &itc->itc_conn; 2064 udp_t *udp = (udp_t *)&itc[1]; 2065 2066 ASSERT(connp->conn_flags & IPCL_UDPCONN); 2067 ASSERT(udp->udp_connp == connp); 2068 ASSERT(connp->conn_udp == udp); 2069 mutex_destroy(&connp->conn_lock); 2070 cv_destroy(&connp->conn_cv); 2071 rw_destroy(&connp->conn_ilg_lock); 2072 2073 /* Can be NULL if constructor failed */ 2074 if (connp->conn_ixa != NULL) { 2075 ASSERT(connp->conn_ixa->ixa_refcnt == 1); 2076 ASSERT(connp->conn_ixa->ixa_ire == NULL); 2077 ASSERT(connp->conn_ixa->ixa_nce == NULL); 2078 ixa_refrele(connp->conn_ixa); 2079 } 2080 } 2081 2082 /* ARGSUSED */ 2083 static int 2084 rawip_conn_constructor(void *buf, void *cdrarg, int kmflags) 2085 { 2086 itc_t *itc = (itc_t *)buf; 2087 conn_t *connp = &itc->itc_conn; 2088 icmp_t *icmp = (icmp_t *)&itc[1]; 2089 2090 bzero(connp, sizeof (conn_t)); 2091 bzero(icmp, sizeof (icmp_t)); 2092 2093 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL); 2094 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL); 2095 connp->conn_icmp = icmp; 2096 connp->conn_flags = IPCL_RAWIPCONN; 2097 connp->conn_proto = IPPROTO_ICMP; 2098 icmp->icmp_connp = connp; 2099 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL); 2100 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags); 2101 if (connp->conn_ixa == NULL) 2102 return (ENOMEM); 2103 connp->conn_ixa->ixa_refcnt = 1; 2104 connp->conn_ixa->ixa_protocol = connp->conn_proto; 2105 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp); 2106 return (0); 2107 } 2108 2109 /* ARGSUSED */ 2110 static void 2111 rawip_conn_destructor(void *buf, void *cdrarg) 2112 { 2113 itc_t *itc = (itc_t *)buf; 2114 conn_t *connp = &itc->itc_conn; 2115 icmp_t *icmp = (icmp_t *)&itc[1]; 2116 2117 ASSERT(connp->conn_flags & IPCL_RAWIPCONN); 2118 ASSERT(icmp->icmp_connp == connp); 2119 ASSERT(connp->conn_icmp == icmp); 2120 mutex_destroy(&connp->conn_lock); 2121 cv_destroy(&connp->conn_cv); 2122 rw_destroy(&connp->conn_ilg_lock); 2123 2124 /* Can be NULL if constructor failed */ 2125 if (connp->conn_ixa != NULL) { 2126 ASSERT(connp->conn_ixa->ixa_refcnt == 1); 2127 ASSERT(connp->conn_ixa->ixa_ire == NULL); 2128 ASSERT(connp->conn_ixa->ixa_nce == NULL); 2129 ixa_refrele(connp->conn_ixa); 2130 } 2131 } 2132 2133 /* ARGSUSED */ 2134 static int 2135 rts_conn_constructor(void *buf, void *cdrarg, int kmflags) 2136 { 2137 itc_t *itc = (itc_t *)buf; 2138 conn_t *connp = &itc->itc_conn; 2139 rts_t *rts = (rts_t *)&itc[1]; 2140 2141 bzero(connp, sizeof (conn_t)); 2142 bzero(rts, sizeof (rts_t)); 2143 2144 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL); 2145 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL); 2146 connp->conn_rts = rts; 2147 connp->conn_flags = IPCL_RTSCONN; 2148 rts->rts_connp = connp; 2149 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL); 2150 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags); 2151 if (connp->conn_ixa == NULL) 2152 return (ENOMEM); 2153 connp->conn_ixa->ixa_refcnt = 1; 2154 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp); 2155 return (0); 2156 } 2157 2158 /* ARGSUSED */ 2159 static void 2160 rts_conn_destructor(void *buf, void *cdrarg) 2161 { 2162 itc_t *itc = (itc_t *)buf; 2163 conn_t *connp = &itc->itc_conn; 2164 rts_t *rts = (rts_t *)&itc[1]; 2165 2166 ASSERT(connp->conn_flags & IPCL_RTSCONN); 2167 ASSERT(rts->rts_connp == connp); 2168 ASSERT(connp->conn_rts == rts); 2169 mutex_destroy(&connp->conn_lock); 2170 cv_destroy(&connp->conn_cv); 2171 rw_destroy(&connp->conn_ilg_lock); 2172 2173 /* Can be NULL if constructor failed */ 2174 if (connp->conn_ixa != NULL) { 2175 ASSERT(connp->conn_ixa->ixa_refcnt == 1); 2176 ASSERT(connp->conn_ixa->ixa_ire == NULL); 2177 ASSERT(connp->conn_ixa->ixa_nce == NULL); 2178 ixa_refrele(connp->conn_ixa); 2179 } 2180 } 2181 2182 /* 2183 * Called as part of ipcl_conn_destroy to assert and clear any pointers 2184 * in the conn_t. 2185 * 2186 * Below we list all the pointers in the conn_t as a documentation aid. 2187 * The ones that we can not ASSERT to be NULL are #ifdef'ed out. 2188 * If you add any pointers to the conn_t please add an ASSERT here 2189 * and #ifdef it out if it can't be actually asserted to be NULL. 2190 * In any case, we bzero most of the conn_t at the end of the function. 2191 */ 2192 void 2193 ipcl_conn_cleanup(conn_t *connp) 2194 { 2195 ip_xmit_attr_t *ixa; 2196 2197 ASSERT(connp->conn_latch == NULL); 2198 ASSERT(connp->conn_latch_in_policy == NULL); 2199 ASSERT(connp->conn_latch_in_action == NULL); 2200 #ifdef notdef 2201 ASSERT(connp->conn_rq == NULL); 2202 ASSERT(connp->conn_wq == NULL); 2203 #endif 2204 ASSERT(connp->conn_cred == NULL); 2205 ASSERT(connp->conn_g_fanout == NULL); 2206 ASSERT(connp->conn_g_next == NULL); 2207 ASSERT(connp->conn_g_prev == NULL); 2208 ASSERT(connp->conn_policy == NULL); 2209 ASSERT(connp->conn_fanout == NULL); 2210 ASSERT(connp->conn_next == NULL); 2211 ASSERT(connp->conn_prev == NULL); 2212 ASSERT(connp->conn_oper_pending_ill == NULL); 2213 ASSERT(connp->conn_ilg == NULL); 2214 ASSERT(connp->conn_drain_next == NULL); 2215 ASSERT(connp->conn_drain_prev == NULL); 2216 #ifdef notdef 2217 /* conn_idl is not cleared when removed from idl list */ 2218 ASSERT(connp->conn_idl == NULL); 2219 #endif 2220 ASSERT(connp->conn_ipsec_opt_mp == NULL); 2221 #ifdef notdef 2222 /* conn_netstack is cleared by the caller; needed by ixa_cleanup */ 2223 ASSERT(connp->conn_netstack == NULL); 2224 #endif 2225 2226 ASSERT(connp->conn_helper_info == NULL); 2227 ASSERT(connp->conn_ixa != NULL); 2228 ixa = connp->conn_ixa; 2229 ASSERT(ixa->ixa_refcnt == 1); 2230 /* Need to preserve ixa_protocol */ 2231 ixa_cleanup(ixa); 2232 ixa->ixa_flags = 0; 2233 2234 /* Clear out the conn_t fields that are not preserved */ 2235 bzero(&connp->conn_start_clr, 2236 sizeof (conn_t) - 2237 ((uchar_t *)&connp->conn_start_clr - (uchar_t *)connp)); 2238 } 2239 2240 /* 2241 * All conns are inserted in a global multi-list for the benefit of 2242 * walkers. The walk is guaranteed to walk all open conns at the time 2243 * of the start of the walk exactly once. This property is needed to 2244 * achieve some cleanups during unplumb of interfaces. This is achieved 2245 * as follows. 2246 * 2247 * ipcl_conn_create and ipcl_conn_destroy are the only functions that 2248 * call the insert and delete functions below at creation and deletion 2249 * time respectively. The conn never moves or changes its position in this 2250 * multi-list during its lifetime. CONN_CONDEMNED ensures that the refcnt 2251 * won't increase due to walkers, once the conn deletion has started. Note 2252 * that we can't remove the conn from the global list and then wait for 2253 * the refcnt to drop to zero, since walkers would then see a truncated 2254 * list. CONN_INCIPIENT ensures that walkers don't start looking at 2255 * conns until ip_open is ready to make them globally visible. 2256 * The global round robin multi-list locks are held only to get the 2257 * next member/insertion/deletion and contention should be negligible 2258 * if the multi-list is much greater than the number of cpus. 2259 */ 2260 void 2261 ipcl_globalhash_insert(conn_t *connp) 2262 { 2263 int index; 2264 struct connf_s *connfp; 2265 ip_stack_t *ipst = connp->conn_netstack->netstack_ip; 2266 2267 /* 2268 * No need for atomic here. Approximate even distribution 2269 * in the global lists is sufficient. 2270 */ 2271 ipst->ips_conn_g_index++; 2272 index = ipst->ips_conn_g_index & (CONN_G_HASH_SIZE - 1); 2273 2274 connp->conn_g_prev = NULL; 2275 /* 2276 * Mark as INCIPIENT, so that walkers will ignore this 2277 * for now, till ip_open is ready to make it visible globally. 2278 */ 2279 connp->conn_state_flags |= CONN_INCIPIENT; 2280 2281 connfp = &ipst->ips_ipcl_globalhash_fanout[index]; 2282 /* Insert at the head of the list */ 2283 mutex_enter(&connfp->connf_lock); 2284 connp->conn_g_next = connfp->connf_head; 2285 if (connp->conn_g_next != NULL) 2286 connp->conn_g_next->conn_g_prev = connp; 2287 connfp->connf_head = connp; 2288 2289 /* The fanout bucket this conn points to */ 2290 connp->conn_g_fanout = connfp; 2291 2292 mutex_exit(&connfp->connf_lock); 2293 } 2294 2295 void 2296 ipcl_globalhash_remove(conn_t *connp) 2297 { 2298 struct connf_s *connfp; 2299 2300 /* 2301 * We were never inserted in the global multi list. 2302 * IPCL_NONE variety is never inserted in the global multilist 2303 * since it is presumed to not need any cleanup and is transient. 2304 */ 2305 if (connp->conn_g_fanout == NULL) 2306 return; 2307 2308 connfp = connp->conn_g_fanout; 2309 mutex_enter(&connfp->connf_lock); 2310 if (connp->conn_g_prev != NULL) 2311 connp->conn_g_prev->conn_g_next = connp->conn_g_next; 2312 else 2313 connfp->connf_head = connp->conn_g_next; 2314 if (connp->conn_g_next != NULL) 2315 connp->conn_g_next->conn_g_prev = connp->conn_g_prev; 2316 mutex_exit(&connfp->connf_lock); 2317 2318 /* Better to stumble on a null pointer than to corrupt memory */ 2319 connp->conn_g_next = NULL; 2320 connp->conn_g_prev = NULL; 2321 connp->conn_g_fanout = NULL; 2322 } 2323 2324 /* 2325 * Walk the list of all conn_t's in the system, calling the function provided 2326 * With the specified argument for each. 2327 * Applies to both IPv4 and IPv6. 2328 * 2329 * CONNs may hold pointers to ills (conn_dhcpinit_ill and 2330 * conn_oper_pending_ill). To guard against stale pointers 2331 * ipcl_walk() is called to cleanup the conn_t's, typically when an interface is 2332 * unplumbed or removed. New conn_t's that are created while we are walking 2333 * may be missed by this walk, because they are not necessarily inserted 2334 * at the tail of the list. They are new conn_t's and thus don't have any 2335 * stale pointers. The CONN_CLOSING flag ensures that no new reference 2336 * is created to the struct that is going away. 2337 */ 2338 void 2339 ipcl_walk(pfv_t func, void *arg, ip_stack_t *ipst) 2340 { 2341 int i; 2342 conn_t *connp; 2343 conn_t *prev_connp; 2344 2345 for (i = 0; i < CONN_G_HASH_SIZE; i++) { 2346 mutex_enter(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock); 2347 prev_connp = NULL; 2348 connp = ipst->ips_ipcl_globalhash_fanout[i].connf_head; 2349 while (connp != NULL) { 2350 mutex_enter(&connp->conn_lock); 2351 if (connp->conn_state_flags & 2352 (CONN_CONDEMNED | CONN_INCIPIENT)) { 2353 mutex_exit(&connp->conn_lock); 2354 connp = connp->conn_g_next; 2355 continue; 2356 } 2357 CONN_INC_REF_LOCKED(connp); 2358 mutex_exit(&connp->conn_lock); 2359 mutex_exit( 2360 &ipst->ips_ipcl_globalhash_fanout[i].connf_lock); 2361 (*func)(connp, arg); 2362 if (prev_connp != NULL) 2363 CONN_DEC_REF(prev_connp); 2364 mutex_enter( 2365 &ipst->ips_ipcl_globalhash_fanout[i].connf_lock); 2366 prev_connp = connp; 2367 connp = connp->conn_g_next; 2368 } 2369 mutex_exit(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock); 2370 if (prev_connp != NULL) 2371 CONN_DEC_REF(prev_connp); 2372 } 2373 } 2374 2375 /* 2376 * Search for a peer TCP/IPv4 loopback conn by doing a reverse lookup on 2377 * the {src, dst, lport, fport} quadruplet. Returns with conn reference 2378 * held; caller must call CONN_DEC_REF. Only checks for connected entries 2379 * (peer tcp in ESTABLISHED state). 2380 */ 2381 conn_t * 2382 ipcl_conn_tcp_lookup_reversed_ipv4(conn_t *connp, ipha_t *ipha, tcpha_t *tcpha, 2383 ip_stack_t *ipst) 2384 { 2385 uint32_t ports; 2386 uint16_t *pports = (uint16_t *)&ports; 2387 connf_t *connfp; 2388 conn_t *tconnp; 2389 boolean_t zone_chk; 2390 2391 /* 2392 * If either the source of destination address is loopback, then 2393 * both endpoints must be in the same Zone. Otherwise, both of 2394 * the addresses are system-wide unique (tcp is in ESTABLISHED 2395 * state) and the endpoints may reside in different Zones. 2396 */ 2397 zone_chk = (ipha->ipha_src == htonl(INADDR_LOOPBACK) || 2398 ipha->ipha_dst == htonl(INADDR_LOOPBACK)); 2399 2400 pports[0] = tcpha->tha_fport; 2401 pports[1] = tcpha->tha_lport; 2402 2403 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_dst, 2404 ports, ipst)]; 2405 2406 mutex_enter(&connfp->connf_lock); 2407 for (tconnp = connfp->connf_head; tconnp != NULL; 2408 tconnp = tconnp->conn_next) { 2409 2410 if (IPCL_CONN_MATCH(tconnp, IPPROTO_TCP, 2411 ipha->ipha_dst, ipha->ipha_src, ports) && 2412 tconnp->conn_tcp->tcp_state == TCPS_ESTABLISHED && 2413 (!zone_chk || tconnp->conn_zoneid == connp->conn_zoneid)) { 2414 2415 ASSERT(tconnp != connp); 2416 CONN_INC_REF(tconnp); 2417 mutex_exit(&connfp->connf_lock); 2418 return (tconnp); 2419 } 2420 } 2421 mutex_exit(&connfp->connf_lock); 2422 return (NULL); 2423 } 2424 2425 /* 2426 * Search for a peer TCP/IPv6 loopback conn by doing a reverse lookup on 2427 * the {src, dst, lport, fport} quadruplet. Returns with conn reference 2428 * held; caller must call CONN_DEC_REF. Only checks for connected entries 2429 * (peer tcp in ESTABLISHED state). 2430 */ 2431 conn_t * 2432 ipcl_conn_tcp_lookup_reversed_ipv6(conn_t *connp, ip6_t *ip6h, tcpha_t *tcpha, 2433 ip_stack_t *ipst) 2434 { 2435 uint32_t ports; 2436 uint16_t *pports = (uint16_t *)&ports; 2437 connf_t *connfp; 2438 conn_t *tconnp; 2439 boolean_t zone_chk; 2440 2441 /* 2442 * If either the source of destination address is loopback, then 2443 * both endpoints must be in the same Zone. Otherwise, both of 2444 * the addresses are system-wide unique (tcp is in ESTABLISHED 2445 * state) and the endpoints may reside in different Zones. We 2446 * don't do Zone check for link local address(es) because the 2447 * current Zone implementation treats each link local address as 2448 * being unique per system node, i.e. they belong to global Zone. 2449 */ 2450 zone_chk = (IN6_IS_ADDR_LOOPBACK(&ip6h->ip6_src) || 2451 IN6_IS_ADDR_LOOPBACK(&ip6h->ip6_dst)); 2452 2453 pports[0] = tcpha->tha_fport; 2454 pports[1] = tcpha->tha_lport; 2455 2456 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_dst, 2457 ports, ipst)]; 2458 2459 mutex_enter(&connfp->connf_lock); 2460 for (tconnp = connfp->connf_head; tconnp != NULL; 2461 tconnp = tconnp->conn_next) { 2462 2463 /* We skip conn_bound_if check here as this is loopback tcp */ 2464 if (IPCL_CONN_MATCH_V6(tconnp, IPPROTO_TCP, 2465 ip6h->ip6_dst, ip6h->ip6_src, ports) && 2466 tconnp->conn_tcp->tcp_state == TCPS_ESTABLISHED && 2467 (!zone_chk || tconnp->conn_zoneid == connp->conn_zoneid)) { 2468 2469 ASSERT(tconnp != connp); 2470 CONN_INC_REF(tconnp); 2471 mutex_exit(&connfp->connf_lock); 2472 return (tconnp); 2473 } 2474 } 2475 mutex_exit(&connfp->connf_lock); 2476 return (NULL); 2477 } 2478 2479 /* 2480 * Find an exact {src, dst, lport, fport} match for a bounced datagram. 2481 * Returns with conn reference held. Caller must call CONN_DEC_REF. 2482 * Only checks for connected entries i.e. no INADDR_ANY checks. 2483 */ 2484 conn_t * 2485 ipcl_tcp_lookup_reversed_ipv4(ipha_t *ipha, tcpha_t *tcpha, int min_state, 2486 ip_stack_t *ipst) 2487 { 2488 uint32_t ports; 2489 uint16_t *pports; 2490 connf_t *connfp; 2491 conn_t *tconnp; 2492 2493 pports = (uint16_t *)&ports; 2494 pports[0] = tcpha->tha_fport; 2495 pports[1] = tcpha->tha_lport; 2496 2497 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_dst, 2498 ports, ipst)]; 2499 2500 mutex_enter(&connfp->connf_lock); 2501 for (tconnp = connfp->connf_head; tconnp != NULL; 2502 tconnp = tconnp->conn_next) { 2503 2504 if (IPCL_CONN_MATCH(tconnp, IPPROTO_TCP, 2505 ipha->ipha_dst, ipha->ipha_src, ports) && 2506 tconnp->conn_tcp->tcp_state >= min_state) { 2507 2508 CONN_INC_REF(tconnp); 2509 mutex_exit(&connfp->connf_lock); 2510 return (tconnp); 2511 } 2512 } 2513 mutex_exit(&connfp->connf_lock); 2514 return (NULL); 2515 } 2516 2517 /* 2518 * Find an exact {src, dst, lport, fport} match for a bounced datagram. 2519 * Returns with conn reference held. Caller must call CONN_DEC_REF. 2520 * Only checks for connected entries i.e. no INADDR_ANY checks. 2521 * Match on ifindex in addition to addresses. 2522 */ 2523 conn_t * 2524 ipcl_tcp_lookup_reversed_ipv6(ip6_t *ip6h, tcpha_t *tcpha, int min_state, 2525 uint_t ifindex, ip_stack_t *ipst) 2526 { 2527 tcp_t *tcp; 2528 uint32_t ports; 2529 uint16_t *pports; 2530 connf_t *connfp; 2531 conn_t *tconnp; 2532 2533 pports = (uint16_t *)&ports; 2534 pports[0] = tcpha->tha_fport; 2535 pports[1] = tcpha->tha_lport; 2536 2537 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_dst, 2538 ports, ipst)]; 2539 2540 mutex_enter(&connfp->connf_lock); 2541 for (tconnp = connfp->connf_head; tconnp != NULL; 2542 tconnp = tconnp->conn_next) { 2543 2544 tcp = tconnp->conn_tcp; 2545 if (IPCL_CONN_MATCH_V6(tconnp, IPPROTO_TCP, 2546 ip6h->ip6_dst, ip6h->ip6_src, ports) && 2547 tcp->tcp_state >= min_state && 2548 (tconnp->conn_bound_if == 0 || 2549 tconnp->conn_bound_if == ifindex)) { 2550 2551 CONN_INC_REF(tconnp); 2552 mutex_exit(&connfp->connf_lock); 2553 return (tconnp); 2554 } 2555 } 2556 mutex_exit(&connfp->connf_lock); 2557 return (NULL); 2558 } 2559 2560 /* 2561 * Finds a TCP/IPv4 listening connection; called by tcp_disconnect to locate 2562 * a listener when changing state. 2563 */ 2564 conn_t * 2565 ipcl_lookup_listener_v4(uint16_t lport, ipaddr_t laddr, zoneid_t zoneid, 2566 ip_stack_t *ipst) 2567 { 2568 connf_t *bind_connfp; 2569 conn_t *connp; 2570 tcp_t *tcp; 2571 2572 /* 2573 * Avoid false matches for packets sent to an IP destination of 2574 * all zeros. 2575 */ 2576 if (laddr == 0) 2577 return (NULL); 2578 2579 ASSERT(zoneid != ALL_ZONES); 2580 2581 bind_connfp = &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)]; 2582 mutex_enter(&bind_connfp->connf_lock); 2583 for (connp = bind_connfp->connf_head; connp != NULL; 2584 connp = connp->conn_next) { 2585 tcp = connp->conn_tcp; 2586 if (IPCL_BIND_MATCH(connp, IPPROTO_TCP, laddr, lport) && 2587 IPCL_ZONE_MATCH(connp, zoneid) && 2588 (tcp->tcp_listener == NULL)) { 2589 CONN_INC_REF(connp); 2590 mutex_exit(&bind_connfp->connf_lock); 2591 return (connp); 2592 } 2593 } 2594 mutex_exit(&bind_connfp->connf_lock); 2595 return (NULL); 2596 } 2597 2598 /* 2599 * Finds a TCP/IPv6 listening connection; called by tcp_disconnect to locate 2600 * a listener when changing state. 2601 */ 2602 conn_t * 2603 ipcl_lookup_listener_v6(uint16_t lport, in6_addr_t *laddr, uint_t ifindex, 2604 zoneid_t zoneid, ip_stack_t *ipst) 2605 { 2606 connf_t *bind_connfp; 2607 conn_t *connp = NULL; 2608 tcp_t *tcp; 2609 2610 /* 2611 * Avoid false matches for packets sent to an IP destination of 2612 * all zeros. 2613 */ 2614 if (IN6_IS_ADDR_UNSPECIFIED(laddr)) 2615 return (NULL); 2616 2617 ASSERT(zoneid != ALL_ZONES); 2618 2619 bind_connfp = &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)]; 2620 mutex_enter(&bind_connfp->connf_lock); 2621 for (connp = bind_connfp->connf_head; connp != NULL; 2622 connp = connp->conn_next) { 2623 tcp = connp->conn_tcp; 2624 if (IPCL_BIND_MATCH_V6(connp, IPPROTO_TCP, *laddr, lport) && 2625 IPCL_ZONE_MATCH(connp, zoneid) && 2626 (connp->conn_bound_if == 0 || 2627 connp->conn_bound_if == ifindex) && 2628 tcp->tcp_listener == NULL) { 2629 CONN_INC_REF(connp); 2630 mutex_exit(&bind_connfp->connf_lock); 2631 return (connp); 2632 } 2633 } 2634 mutex_exit(&bind_connfp->connf_lock); 2635 return (NULL); 2636 } 2637 2638 /* 2639 * ipcl_get_next_conn 2640 * get the next entry in the conn global list 2641 * and put a reference on the next_conn. 2642 * decrement the reference on the current conn. 2643 * 2644 * This is an iterator based walker function that also provides for 2645 * some selection by the caller. It walks through the conn_hash bucket 2646 * searching for the next valid connp in the list, and selects connections 2647 * that are neither closed nor condemned. It also REFHOLDS the conn 2648 * thus ensuring that the conn exists when the caller uses the conn. 2649 */ 2650 conn_t * 2651 ipcl_get_next_conn(connf_t *connfp, conn_t *connp, uint32_t conn_flags) 2652 { 2653 conn_t *next_connp; 2654 2655 if (connfp == NULL) 2656 return (NULL); 2657 2658 mutex_enter(&connfp->connf_lock); 2659 2660 next_connp = (connp == NULL) ? 2661 connfp->connf_head : connp->conn_g_next; 2662 2663 while (next_connp != NULL) { 2664 mutex_enter(&next_connp->conn_lock); 2665 if (!(next_connp->conn_flags & conn_flags) || 2666 (next_connp->conn_state_flags & 2667 (CONN_CONDEMNED | CONN_INCIPIENT))) { 2668 /* 2669 * This conn has been condemned or 2670 * is closing, or the flags don't match 2671 */ 2672 mutex_exit(&next_connp->conn_lock); 2673 next_connp = next_connp->conn_g_next; 2674 continue; 2675 } 2676 CONN_INC_REF_LOCKED(next_connp); 2677 mutex_exit(&next_connp->conn_lock); 2678 break; 2679 } 2680 2681 mutex_exit(&connfp->connf_lock); 2682 2683 if (connp != NULL) 2684 CONN_DEC_REF(connp); 2685 2686 return (next_connp); 2687 } 2688 2689 #ifdef CONN_DEBUG 2690 /* 2691 * Trace of the last NBUF refhold/refrele 2692 */ 2693 int 2694 conn_trace_ref(conn_t *connp) 2695 { 2696 int last; 2697 conn_trace_t *ctb; 2698 2699 ASSERT(MUTEX_HELD(&connp->conn_lock)); 2700 last = connp->conn_trace_last; 2701 last++; 2702 if (last == CONN_TRACE_MAX) 2703 last = 0; 2704 2705 ctb = &connp->conn_trace_buf[last]; 2706 ctb->ctb_depth = getpcstack(ctb->ctb_stack, CONN_STACK_DEPTH); 2707 connp->conn_trace_last = last; 2708 return (1); 2709 } 2710 2711 int 2712 conn_untrace_ref(conn_t *connp) 2713 { 2714 int last; 2715 conn_trace_t *ctb; 2716 2717 ASSERT(MUTEX_HELD(&connp->conn_lock)); 2718 last = connp->conn_trace_last; 2719 last++; 2720 if (last == CONN_TRACE_MAX) 2721 last = 0; 2722 2723 ctb = &connp->conn_trace_buf[last]; 2724 ctb->ctb_depth = getpcstack(ctb->ctb_stack, CONN_STACK_DEPTH); 2725 connp->conn_trace_last = last; 2726 return (1); 2727 } 2728 #endif 2729