1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * net/dccp/proto.c 4 * 5 * An implementation of the DCCP protocol 6 * Arnaldo Carvalho de Melo <acme@conectiva.com.br> 7 */ 8 9 #include <linux/dccp.h> 10 #include <linux/module.h> 11 #include <linux/types.h> 12 #include <linux/sched.h> 13 #include <linux/kernel.h> 14 #include <linux/skbuff.h> 15 #include <linux/netdevice.h> 16 #include <linux/in.h> 17 #include <linux/if_arp.h> 18 #include <linux/init.h> 19 #include <linux/random.h> 20 #include <linux/slab.h> 21 #include <net/checksum.h> 22 23 #include <net/inet_sock.h> 24 #include <net/inet_common.h> 25 #include <net/sock.h> 26 #include <net/xfrm.h> 27 28 #include <asm/ioctls.h> 29 #include <linux/spinlock.h> 30 #include <linux/timer.h> 31 #include <linux/delay.h> 32 #include <linux/poll.h> 33 34 #include "ccid.h" 35 #include "dccp.h" 36 #include "feat.h" 37 38 #define CREATE_TRACE_POINTS 39 #include "trace.h" 40 41 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly; 42 43 EXPORT_SYMBOL_GPL(dccp_statistics); 44 45 struct percpu_counter dccp_orphan_count; 46 EXPORT_SYMBOL_GPL(dccp_orphan_count); 47 48 struct inet_hashinfo dccp_hashinfo; 49 EXPORT_SYMBOL_GPL(dccp_hashinfo); 50 51 /* the maximum queue length for tx in packets. 0 is no limit */ 52 int sysctl_dccp_tx_qlen __read_mostly = 5; 53 54 #ifdef CONFIG_IP_DCCP_DEBUG 55 static const char *dccp_state_name(const int state) 56 { 57 static const char *const dccp_state_names[] = { 58 [DCCP_OPEN] = "OPEN", 59 [DCCP_REQUESTING] = "REQUESTING", 60 [DCCP_PARTOPEN] = "PARTOPEN", 61 [DCCP_LISTEN] = "LISTEN", 62 [DCCP_RESPOND] = "RESPOND", 63 [DCCP_CLOSING] = "CLOSING", 64 [DCCP_ACTIVE_CLOSEREQ] = "CLOSEREQ", 65 [DCCP_PASSIVE_CLOSE] = "PASSIVE_CLOSE", 66 [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ", 67 [DCCP_TIME_WAIT] = "TIME_WAIT", 68 [DCCP_CLOSED] = "CLOSED", 69 }; 70 71 if (state >= DCCP_MAX_STATES) 72 return "INVALID STATE!"; 73 else 74 return dccp_state_names[state]; 75 } 76 #endif 77 78 void dccp_set_state(struct sock *sk, const int state) 79 { 80 const int oldstate = sk->sk_state; 81 82 dccp_pr_debug("%s(%p) %s --> %s\n", dccp_role(sk), sk, 83 dccp_state_name(oldstate), dccp_state_name(state)); 84 WARN_ON(state == oldstate); 85 86 switch (state) { 87 case DCCP_OPEN: 88 if (oldstate != DCCP_OPEN) 89 DCCP_INC_STATS(DCCP_MIB_CURRESTAB); 90 /* Client retransmits all Confirm options until entering OPEN */ 91 if (oldstate == DCCP_PARTOPEN) 92 dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg); 93 break; 94 95 case DCCP_CLOSED: 96 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ || 97 oldstate == DCCP_CLOSING) 98 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS); 99 100 sk->sk_prot->unhash(sk); 101 if (inet_csk(sk)->icsk_bind_hash != NULL && 102 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) 103 inet_put_port(sk); 104 /* fall through */ 105 default: 106 if (oldstate == DCCP_OPEN) 107 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB); 108 } 109 110 /* Change state AFTER socket is unhashed to avoid closed 111 * socket sitting in hash tables. 112 */ 113 inet_sk_set_state(sk, state); 114 } 115 116 EXPORT_SYMBOL_GPL(dccp_set_state); 117 118 static void dccp_finish_passive_close(struct sock *sk) 119 { 120 switch (sk->sk_state) { 121 case DCCP_PASSIVE_CLOSE: 122 /* Node (client or server) has received Close packet. */ 123 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED); 124 dccp_set_state(sk, DCCP_CLOSED); 125 break; 126 case DCCP_PASSIVE_CLOSEREQ: 127 /* 128 * Client received CloseReq. We set the `active' flag so that 129 * dccp_send_close() retransmits the Close as per RFC 4340, 8.3. 130 */ 131 dccp_send_close(sk, 1); 132 dccp_set_state(sk, DCCP_CLOSING); 133 } 134 } 135 136 void dccp_done(struct sock *sk) 137 { 138 dccp_set_state(sk, DCCP_CLOSED); 139 dccp_clear_xmit_timers(sk); 140 141 sk->sk_shutdown = SHUTDOWN_MASK; 142 143 if (!sock_flag(sk, SOCK_DEAD)) 144 sk->sk_state_change(sk); 145 else 146 inet_csk_destroy_sock(sk); 147 } 148 149 EXPORT_SYMBOL_GPL(dccp_done); 150 151 const char *dccp_packet_name(const int type) 152 { 153 static const char *const dccp_packet_names[] = { 154 [DCCP_PKT_REQUEST] = "REQUEST", 155 [DCCP_PKT_RESPONSE] = "RESPONSE", 156 [DCCP_PKT_DATA] = "DATA", 157 [DCCP_PKT_ACK] = "ACK", 158 [DCCP_PKT_DATAACK] = "DATAACK", 159 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ", 160 [DCCP_PKT_CLOSE] = "CLOSE", 161 [DCCP_PKT_RESET] = "RESET", 162 [DCCP_PKT_SYNC] = "SYNC", 163 [DCCP_PKT_SYNCACK] = "SYNCACK", 164 }; 165 166 if (type >= DCCP_NR_PKT_TYPES) 167 return "INVALID"; 168 else 169 return dccp_packet_names[type]; 170 } 171 172 EXPORT_SYMBOL_GPL(dccp_packet_name); 173 174 static void dccp_sk_destruct(struct sock *sk) 175 { 176 struct dccp_sock *dp = dccp_sk(sk); 177 178 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk); 179 dp->dccps_hc_tx_ccid = NULL; 180 inet_sock_destruct(sk); 181 } 182 183 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized) 184 { 185 struct dccp_sock *dp = dccp_sk(sk); 186 struct inet_connection_sock *icsk = inet_csk(sk); 187 188 icsk->icsk_rto = DCCP_TIMEOUT_INIT; 189 icsk->icsk_syn_retries = sysctl_dccp_request_retries; 190 sk->sk_state = DCCP_CLOSED; 191 sk->sk_write_space = dccp_write_space; 192 sk->sk_destruct = dccp_sk_destruct; 193 icsk->icsk_sync_mss = dccp_sync_mss; 194 dp->dccps_mss_cache = 536; 195 dp->dccps_rate_last = jiffies; 196 dp->dccps_role = DCCP_ROLE_UNDEFINED; 197 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT; 198 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen; 199 200 dccp_init_xmit_timers(sk); 201 202 INIT_LIST_HEAD(&dp->dccps_featneg); 203 /* control socket doesn't need feat nego */ 204 if (likely(ctl_sock_initialized)) 205 return dccp_feat_init(sk); 206 return 0; 207 } 208 209 EXPORT_SYMBOL_GPL(dccp_init_sock); 210 211 void dccp_destroy_sock(struct sock *sk) 212 { 213 struct dccp_sock *dp = dccp_sk(sk); 214 215 __skb_queue_purge(&sk->sk_write_queue); 216 if (sk->sk_send_head != NULL) { 217 kfree_skb(sk->sk_send_head); 218 sk->sk_send_head = NULL; 219 } 220 221 /* Clean up a referenced DCCP bind bucket. */ 222 if (inet_csk(sk)->icsk_bind_hash != NULL) 223 inet_put_port(sk); 224 225 kfree(dp->dccps_service_list); 226 dp->dccps_service_list = NULL; 227 228 if (dp->dccps_hc_rx_ackvec != NULL) { 229 dccp_ackvec_free(dp->dccps_hc_rx_ackvec); 230 dp->dccps_hc_rx_ackvec = NULL; 231 } 232 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk); 233 dp->dccps_hc_rx_ccid = NULL; 234 235 /* clean up feature negotiation state */ 236 dccp_feat_list_purge(&dp->dccps_featneg); 237 } 238 239 EXPORT_SYMBOL_GPL(dccp_destroy_sock); 240 241 static inline int dccp_listen_start(struct sock *sk, int backlog) 242 { 243 struct dccp_sock *dp = dccp_sk(sk); 244 245 dp->dccps_role = DCCP_ROLE_LISTEN; 246 /* do not start to listen if feature negotiation setup fails */ 247 if (dccp_feat_finalise_settings(dp)) 248 return -EPROTO; 249 return inet_csk_listen_start(sk, backlog); 250 } 251 252 static inline int dccp_need_reset(int state) 253 { 254 return state != DCCP_CLOSED && state != DCCP_LISTEN && 255 state != DCCP_REQUESTING; 256 } 257 258 int dccp_disconnect(struct sock *sk, int flags) 259 { 260 struct inet_connection_sock *icsk = inet_csk(sk); 261 struct inet_sock *inet = inet_sk(sk); 262 struct dccp_sock *dp = dccp_sk(sk); 263 const int old_state = sk->sk_state; 264 265 if (old_state != DCCP_CLOSED) 266 dccp_set_state(sk, DCCP_CLOSED); 267 268 /* 269 * This corresponds to the ABORT function of RFC793, sec. 3.8 270 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted". 271 */ 272 if (old_state == DCCP_LISTEN) { 273 inet_csk_listen_stop(sk); 274 } else if (dccp_need_reset(old_state)) { 275 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED); 276 sk->sk_err = ECONNRESET; 277 } else if (old_state == DCCP_REQUESTING) 278 sk->sk_err = ECONNRESET; 279 280 dccp_clear_xmit_timers(sk); 281 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk); 282 dp->dccps_hc_rx_ccid = NULL; 283 284 __skb_queue_purge(&sk->sk_receive_queue); 285 __skb_queue_purge(&sk->sk_write_queue); 286 if (sk->sk_send_head != NULL) { 287 __kfree_skb(sk->sk_send_head); 288 sk->sk_send_head = NULL; 289 } 290 291 inet->inet_dport = 0; 292 293 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 294 inet_reset_saddr(sk); 295 296 sk->sk_shutdown = 0; 297 sock_reset_flag(sk, SOCK_DONE); 298 299 icsk->icsk_backoff = 0; 300 inet_csk_delack_init(sk); 301 __sk_dst_reset(sk); 302 303 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); 304 305 sk->sk_error_report(sk); 306 return 0; 307 } 308 309 EXPORT_SYMBOL_GPL(dccp_disconnect); 310 311 /* 312 * Wait for a DCCP event. 313 * 314 * Note that we don't need to lock the socket, as the upper poll layers 315 * take care of normal races (between the test and the event) and we don't 316 * go look at any of the socket buffers directly. 317 */ 318 __poll_t dccp_poll(struct file *file, struct socket *sock, 319 poll_table *wait) 320 { 321 __poll_t mask; 322 struct sock *sk = sock->sk; 323 324 sock_poll_wait(file, sock, wait); 325 if (sk->sk_state == DCCP_LISTEN) 326 return inet_csk_listen_poll(sk); 327 328 /* Socket is not locked. We are protected from async events 329 by poll logic and correct handling of state changes 330 made by another threads is impossible in any case. 331 */ 332 333 mask = 0; 334 if (sk->sk_err) 335 mask = EPOLLERR; 336 337 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED) 338 mask |= EPOLLHUP; 339 if (sk->sk_shutdown & RCV_SHUTDOWN) 340 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 341 342 /* Connected? */ 343 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) { 344 if (atomic_read(&sk->sk_rmem_alloc) > 0) 345 mask |= EPOLLIN | EPOLLRDNORM; 346 347 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 348 if (sk_stream_is_writeable(sk)) { 349 mask |= EPOLLOUT | EPOLLWRNORM; 350 } else { /* send SIGIO later */ 351 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 352 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 353 354 /* Race breaker. If space is freed after 355 * wspace test but before the flags are set, 356 * IO signal will be lost. 357 */ 358 if (sk_stream_is_writeable(sk)) 359 mask |= EPOLLOUT | EPOLLWRNORM; 360 } 361 } 362 } 363 return mask; 364 } 365 366 EXPORT_SYMBOL_GPL(dccp_poll); 367 368 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg) 369 { 370 int rc = -ENOTCONN; 371 372 lock_sock(sk); 373 374 if (sk->sk_state == DCCP_LISTEN) 375 goto out; 376 377 switch (cmd) { 378 case SIOCINQ: { 379 struct sk_buff *skb; 380 unsigned long amount = 0; 381 382 skb = skb_peek(&sk->sk_receive_queue); 383 if (skb != NULL) { 384 /* 385 * We will only return the amount of this packet since 386 * that is all that will be read. 387 */ 388 amount = skb->len; 389 } 390 rc = put_user(amount, (int __user *)arg); 391 } 392 break; 393 default: 394 rc = -ENOIOCTLCMD; 395 break; 396 } 397 out: 398 release_sock(sk); 399 return rc; 400 } 401 402 EXPORT_SYMBOL_GPL(dccp_ioctl); 403 404 static int dccp_setsockopt_service(struct sock *sk, const __be32 service, 405 char __user *optval, unsigned int optlen) 406 { 407 struct dccp_sock *dp = dccp_sk(sk); 408 struct dccp_service_list *sl = NULL; 409 410 if (service == DCCP_SERVICE_INVALID_VALUE || 411 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32)) 412 return -EINVAL; 413 414 if (optlen > sizeof(service)) { 415 sl = kmalloc(optlen, GFP_KERNEL); 416 if (sl == NULL) 417 return -ENOMEM; 418 419 sl->dccpsl_nr = optlen / sizeof(u32) - 1; 420 if (copy_from_user(sl->dccpsl_list, 421 optval + sizeof(service), 422 optlen - sizeof(service)) || 423 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) { 424 kfree(sl); 425 return -EFAULT; 426 } 427 } 428 429 lock_sock(sk); 430 dp->dccps_service = service; 431 432 kfree(dp->dccps_service_list); 433 434 dp->dccps_service_list = sl; 435 release_sock(sk); 436 return 0; 437 } 438 439 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx) 440 { 441 u8 *list, len; 442 int i, rc; 443 444 if (cscov < 0 || cscov > 15) 445 return -EINVAL; 446 /* 447 * Populate a list of permissible values, in the range cscov...15. This 448 * is necessary since feature negotiation of single values only works if 449 * both sides incidentally choose the same value. Since the list starts 450 * lowest-value first, negotiation will pick the smallest shared value. 451 */ 452 if (cscov == 0) 453 return 0; 454 len = 16 - cscov; 455 456 list = kmalloc(len, GFP_KERNEL); 457 if (list == NULL) 458 return -ENOBUFS; 459 460 for (i = 0; i < len; i++) 461 list[i] = cscov++; 462 463 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len); 464 465 if (rc == 0) { 466 if (rx) 467 dccp_sk(sk)->dccps_pcrlen = cscov; 468 else 469 dccp_sk(sk)->dccps_pcslen = cscov; 470 } 471 kfree(list); 472 return rc; 473 } 474 475 static int dccp_setsockopt_ccid(struct sock *sk, int type, 476 char __user *optval, unsigned int optlen) 477 { 478 u8 *val; 479 int rc = 0; 480 481 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS) 482 return -EINVAL; 483 484 val = memdup_user(optval, optlen); 485 if (IS_ERR(val)) 486 return PTR_ERR(val); 487 488 lock_sock(sk); 489 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID) 490 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen); 491 492 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID)) 493 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen); 494 release_sock(sk); 495 496 kfree(val); 497 return rc; 498 } 499 500 static int do_dccp_setsockopt(struct sock *sk, int level, int optname, 501 char __user *optval, unsigned int optlen) 502 { 503 struct dccp_sock *dp = dccp_sk(sk); 504 int val, err = 0; 505 506 switch (optname) { 507 case DCCP_SOCKOPT_PACKET_SIZE: 508 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n"); 509 return 0; 510 case DCCP_SOCKOPT_CHANGE_L: 511 case DCCP_SOCKOPT_CHANGE_R: 512 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n"); 513 return 0; 514 case DCCP_SOCKOPT_CCID: 515 case DCCP_SOCKOPT_RX_CCID: 516 case DCCP_SOCKOPT_TX_CCID: 517 return dccp_setsockopt_ccid(sk, optname, optval, optlen); 518 } 519 520 if (optlen < (int)sizeof(int)) 521 return -EINVAL; 522 523 if (get_user(val, (int __user *)optval)) 524 return -EFAULT; 525 526 if (optname == DCCP_SOCKOPT_SERVICE) 527 return dccp_setsockopt_service(sk, val, optval, optlen); 528 529 lock_sock(sk); 530 switch (optname) { 531 case DCCP_SOCKOPT_SERVER_TIMEWAIT: 532 if (dp->dccps_role != DCCP_ROLE_SERVER) 533 err = -EOPNOTSUPP; 534 else 535 dp->dccps_server_timewait = (val != 0); 536 break; 537 case DCCP_SOCKOPT_SEND_CSCOV: 538 err = dccp_setsockopt_cscov(sk, val, false); 539 break; 540 case DCCP_SOCKOPT_RECV_CSCOV: 541 err = dccp_setsockopt_cscov(sk, val, true); 542 break; 543 case DCCP_SOCKOPT_QPOLICY_ID: 544 if (sk->sk_state != DCCP_CLOSED) 545 err = -EISCONN; 546 else if (val < 0 || val >= DCCPQ_POLICY_MAX) 547 err = -EINVAL; 548 else 549 dp->dccps_qpolicy = val; 550 break; 551 case DCCP_SOCKOPT_QPOLICY_TXQLEN: 552 if (val < 0) 553 err = -EINVAL; 554 else 555 dp->dccps_tx_qlen = val; 556 break; 557 default: 558 err = -ENOPROTOOPT; 559 break; 560 } 561 release_sock(sk); 562 563 return err; 564 } 565 566 int dccp_setsockopt(struct sock *sk, int level, int optname, 567 char __user *optval, unsigned int optlen) 568 { 569 if (level != SOL_DCCP) 570 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level, 571 optname, optval, 572 optlen); 573 return do_dccp_setsockopt(sk, level, optname, optval, optlen); 574 } 575 576 EXPORT_SYMBOL_GPL(dccp_setsockopt); 577 578 #ifdef CONFIG_COMPAT 579 int compat_dccp_setsockopt(struct sock *sk, int level, int optname, 580 char __user *optval, unsigned int optlen) 581 { 582 if (level != SOL_DCCP) 583 return inet_csk_compat_setsockopt(sk, level, optname, 584 optval, optlen); 585 return do_dccp_setsockopt(sk, level, optname, optval, optlen); 586 } 587 588 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt); 589 #endif 590 591 static int dccp_getsockopt_service(struct sock *sk, int len, 592 __be32 __user *optval, 593 int __user *optlen) 594 { 595 const struct dccp_sock *dp = dccp_sk(sk); 596 const struct dccp_service_list *sl; 597 int err = -ENOENT, slen = 0, total_len = sizeof(u32); 598 599 lock_sock(sk); 600 if ((sl = dp->dccps_service_list) != NULL) { 601 slen = sl->dccpsl_nr * sizeof(u32); 602 total_len += slen; 603 } 604 605 err = -EINVAL; 606 if (total_len > len) 607 goto out; 608 609 err = 0; 610 if (put_user(total_len, optlen) || 611 put_user(dp->dccps_service, optval) || 612 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen))) 613 err = -EFAULT; 614 out: 615 release_sock(sk); 616 return err; 617 } 618 619 static int do_dccp_getsockopt(struct sock *sk, int level, int optname, 620 char __user *optval, int __user *optlen) 621 { 622 struct dccp_sock *dp; 623 int val, len; 624 625 if (get_user(len, optlen)) 626 return -EFAULT; 627 628 if (len < (int)sizeof(int)) 629 return -EINVAL; 630 631 dp = dccp_sk(sk); 632 633 switch (optname) { 634 case DCCP_SOCKOPT_PACKET_SIZE: 635 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n"); 636 return 0; 637 case DCCP_SOCKOPT_SERVICE: 638 return dccp_getsockopt_service(sk, len, 639 (__be32 __user *)optval, optlen); 640 case DCCP_SOCKOPT_GET_CUR_MPS: 641 val = dp->dccps_mss_cache; 642 break; 643 case DCCP_SOCKOPT_AVAILABLE_CCIDS: 644 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen); 645 case DCCP_SOCKOPT_TX_CCID: 646 val = ccid_get_current_tx_ccid(dp); 647 if (val < 0) 648 return -ENOPROTOOPT; 649 break; 650 case DCCP_SOCKOPT_RX_CCID: 651 val = ccid_get_current_rx_ccid(dp); 652 if (val < 0) 653 return -ENOPROTOOPT; 654 break; 655 case DCCP_SOCKOPT_SERVER_TIMEWAIT: 656 val = dp->dccps_server_timewait; 657 break; 658 case DCCP_SOCKOPT_SEND_CSCOV: 659 val = dp->dccps_pcslen; 660 break; 661 case DCCP_SOCKOPT_RECV_CSCOV: 662 val = dp->dccps_pcrlen; 663 break; 664 case DCCP_SOCKOPT_QPOLICY_ID: 665 val = dp->dccps_qpolicy; 666 break; 667 case DCCP_SOCKOPT_QPOLICY_TXQLEN: 668 val = dp->dccps_tx_qlen; 669 break; 670 case 128 ... 191: 671 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname, 672 len, (u32 __user *)optval, optlen); 673 case 192 ... 255: 674 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname, 675 len, (u32 __user *)optval, optlen); 676 default: 677 return -ENOPROTOOPT; 678 } 679 680 len = sizeof(val); 681 if (put_user(len, optlen) || copy_to_user(optval, &val, len)) 682 return -EFAULT; 683 684 return 0; 685 } 686 687 int dccp_getsockopt(struct sock *sk, int level, int optname, 688 char __user *optval, int __user *optlen) 689 { 690 if (level != SOL_DCCP) 691 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level, 692 optname, optval, 693 optlen); 694 return do_dccp_getsockopt(sk, level, optname, optval, optlen); 695 } 696 697 EXPORT_SYMBOL_GPL(dccp_getsockopt); 698 699 #ifdef CONFIG_COMPAT 700 int compat_dccp_getsockopt(struct sock *sk, int level, int optname, 701 char __user *optval, int __user *optlen) 702 { 703 if (level != SOL_DCCP) 704 return inet_csk_compat_getsockopt(sk, level, optname, 705 optval, optlen); 706 return do_dccp_getsockopt(sk, level, optname, optval, optlen); 707 } 708 709 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt); 710 #endif 711 712 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb) 713 { 714 struct cmsghdr *cmsg; 715 716 /* 717 * Assign an (opaque) qpolicy priority value to skb->priority. 718 * 719 * We are overloading this skb field for use with the qpolicy subystem. 720 * The skb->priority is normally used for the SO_PRIORITY option, which 721 * is initialised from sk_priority. Since the assignment of sk_priority 722 * to skb->priority happens later (on layer 3), we overload this field 723 * for use with queueing priorities as long as the skb is on layer 4. 724 * The default priority value (if nothing is set) is 0. 725 */ 726 skb->priority = 0; 727 728 for_each_cmsghdr(cmsg, msg) { 729 if (!CMSG_OK(msg, cmsg)) 730 return -EINVAL; 731 732 if (cmsg->cmsg_level != SOL_DCCP) 733 continue; 734 735 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX && 736 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type)) 737 return -EINVAL; 738 739 switch (cmsg->cmsg_type) { 740 case DCCP_SCM_PRIORITY: 741 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32))) 742 return -EINVAL; 743 skb->priority = *(__u32 *)CMSG_DATA(cmsg); 744 break; 745 default: 746 return -EINVAL; 747 } 748 } 749 return 0; 750 } 751 752 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 753 { 754 const struct dccp_sock *dp = dccp_sk(sk); 755 const int flags = msg->msg_flags; 756 const int noblock = flags & MSG_DONTWAIT; 757 struct sk_buff *skb; 758 int rc, size; 759 long timeo; 760 761 trace_dccp_probe(sk, len); 762 763 if (len > dp->dccps_mss_cache) 764 return -EMSGSIZE; 765 766 lock_sock(sk); 767 768 if (dccp_qpolicy_full(sk)) { 769 rc = -EAGAIN; 770 goto out_release; 771 } 772 773 timeo = sock_sndtimeo(sk, noblock); 774 775 /* 776 * We have to use sk_stream_wait_connect here to set sk_write_pending, 777 * so that the trick in dccp_rcv_request_sent_state_process. 778 */ 779 /* Wait for a connection to finish. */ 780 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN)) 781 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0) 782 goto out_release; 783 784 size = sk->sk_prot->max_header + len; 785 release_sock(sk); 786 skb = sock_alloc_send_skb(sk, size, noblock, &rc); 787 lock_sock(sk); 788 if (skb == NULL) 789 goto out_release; 790 791 if (sk->sk_state == DCCP_CLOSED) { 792 rc = -ENOTCONN; 793 goto out_discard; 794 } 795 796 skb_reserve(skb, sk->sk_prot->max_header); 797 rc = memcpy_from_msg(skb_put(skb, len), msg, len); 798 if (rc != 0) 799 goto out_discard; 800 801 rc = dccp_msghdr_parse(msg, skb); 802 if (rc != 0) 803 goto out_discard; 804 805 dccp_qpolicy_push(sk, skb); 806 /* 807 * The xmit_timer is set if the TX CCID is rate-based and will expire 808 * when congestion control permits to release further packets into the 809 * network. Window-based CCIDs do not use this timer. 810 */ 811 if (!timer_pending(&dp->dccps_xmit_timer)) 812 dccp_write_xmit(sk); 813 out_release: 814 release_sock(sk); 815 return rc ? : len; 816 out_discard: 817 kfree_skb(skb); 818 goto out_release; 819 } 820 821 EXPORT_SYMBOL_GPL(dccp_sendmsg); 822 823 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 824 int flags, int *addr_len) 825 { 826 const struct dccp_hdr *dh; 827 long timeo; 828 829 lock_sock(sk); 830 831 if (sk->sk_state == DCCP_LISTEN) { 832 len = -ENOTCONN; 833 goto out; 834 } 835 836 timeo = sock_rcvtimeo(sk, nonblock); 837 838 do { 839 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); 840 841 if (skb == NULL) 842 goto verify_sock_status; 843 844 dh = dccp_hdr(skb); 845 846 switch (dh->dccph_type) { 847 case DCCP_PKT_DATA: 848 case DCCP_PKT_DATAACK: 849 goto found_ok_skb; 850 851 case DCCP_PKT_CLOSE: 852 case DCCP_PKT_CLOSEREQ: 853 if (!(flags & MSG_PEEK)) 854 dccp_finish_passive_close(sk); 855 /* fall through */ 856 case DCCP_PKT_RESET: 857 dccp_pr_debug("found fin (%s) ok!\n", 858 dccp_packet_name(dh->dccph_type)); 859 len = 0; 860 goto found_fin_ok; 861 default: 862 dccp_pr_debug("packet_type=%s\n", 863 dccp_packet_name(dh->dccph_type)); 864 sk_eat_skb(sk, skb); 865 } 866 verify_sock_status: 867 if (sock_flag(sk, SOCK_DONE)) { 868 len = 0; 869 break; 870 } 871 872 if (sk->sk_err) { 873 len = sock_error(sk); 874 break; 875 } 876 877 if (sk->sk_shutdown & RCV_SHUTDOWN) { 878 len = 0; 879 break; 880 } 881 882 if (sk->sk_state == DCCP_CLOSED) { 883 if (!sock_flag(sk, SOCK_DONE)) { 884 /* This occurs when user tries to read 885 * from never connected socket. 886 */ 887 len = -ENOTCONN; 888 break; 889 } 890 len = 0; 891 break; 892 } 893 894 if (!timeo) { 895 len = -EAGAIN; 896 break; 897 } 898 899 if (signal_pending(current)) { 900 len = sock_intr_errno(timeo); 901 break; 902 } 903 904 sk_wait_data(sk, &timeo, NULL); 905 continue; 906 found_ok_skb: 907 if (len > skb->len) 908 len = skb->len; 909 else if (len < skb->len) 910 msg->msg_flags |= MSG_TRUNC; 911 912 if (skb_copy_datagram_msg(skb, 0, msg, len)) { 913 /* Exception. Bailout! */ 914 len = -EFAULT; 915 break; 916 } 917 if (flags & MSG_TRUNC) 918 len = skb->len; 919 found_fin_ok: 920 if (!(flags & MSG_PEEK)) 921 sk_eat_skb(sk, skb); 922 break; 923 } while (1); 924 out: 925 release_sock(sk); 926 return len; 927 } 928 929 EXPORT_SYMBOL_GPL(dccp_recvmsg); 930 931 int inet_dccp_listen(struct socket *sock, int backlog) 932 { 933 struct sock *sk = sock->sk; 934 unsigned char old_state; 935 int err; 936 937 lock_sock(sk); 938 939 err = -EINVAL; 940 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP) 941 goto out; 942 943 old_state = sk->sk_state; 944 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN))) 945 goto out; 946 947 sk->sk_max_ack_backlog = backlog; 948 /* Really, if the socket is already in listen state 949 * we can only allow the backlog to be adjusted. 950 */ 951 if (old_state != DCCP_LISTEN) { 952 /* 953 * FIXME: here it probably should be sk->sk_prot->listen_start 954 * see tcp_listen_start 955 */ 956 err = dccp_listen_start(sk, backlog); 957 if (err) 958 goto out; 959 } 960 err = 0; 961 962 out: 963 release_sock(sk); 964 return err; 965 } 966 967 EXPORT_SYMBOL_GPL(inet_dccp_listen); 968 969 static void dccp_terminate_connection(struct sock *sk) 970 { 971 u8 next_state = DCCP_CLOSED; 972 973 switch (sk->sk_state) { 974 case DCCP_PASSIVE_CLOSE: 975 case DCCP_PASSIVE_CLOSEREQ: 976 dccp_finish_passive_close(sk); 977 break; 978 case DCCP_PARTOPEN: 979 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk); 980 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 981 /* fall through */ 982 case DCCP_OPEN: 983 dccp_send_close(sk, 1); 984 985 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER && 986 !dccp_sk(sk)->dccps_server_timewait) 987 next_state = DCCP_ACTIVE_CLOSEREQ; 988 else 989 next_state = DCCP_CLOSING; 990 /* fall through */ 991 default: 992 dccp_set_state(sk, next_state); 993 } 994 } 995 996 void dccp_close(struct sock *sk, long timeout) 997 { 998 struct dccp_sock *dp = dccp_sk(sk); 999 struct sk_buff *skb; 1000 u32 data_was_unread = 0; 1001 int state; 1002 1003 lock_sock(sk); 1004 1005 sk->sk_shutdown = SHUTDOWN_MASK; 1006 1007 if (sk->sk_state == DCCP_LISTEN) { 1008 dccp_set_state(sk, DCCP_CLOSED); 1009 1010 /* Special case. */ 1011 inet_csk_listen_stop(sk); 1012 1013 goto adjudge_to_death; 1014 } 1015 1016 sk_stop_timer(sk, &dp->dccps_xmit_timer); 1017 1018 /* 1019 * We need to flush the recv. buffs. We do this only on the 1020 * descriptor close, not protocol-sourced closes, because the 1021 *reader process may not have drained the data yet! 1022 */ 1023 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { 1024 data_was_unread += skb->len; 1025 __kfree_skb(skb); 1026 } 1027 1028 /* If socket has been already reset kill it. */ 1029 if (sk->sk_state == DCCP_CLOSED) 1030 goto adjudge_to_death; 1031 1032 if (data_was_unread) { 1033 /* Unread data was tossed, send an appropriate Reset Code */ 1034 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread); 1035 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED); 1036 dccp_set_state(sk, DCCP_CLOSED); 1037 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { 1038 /* Check zero linger _after_ checking for unread data. */ 1039 sk->sk_prot->disconnect(sk, 0); 1040 } else if (sk->sk_state != DCCP_CLOSED) { 1041 /* 1042 * Normal connection termination. May need to wait if there are 1043 * still packets in the TX queue that are delayed by the CCID. 1044 */ 1045 dccp_flush_write_queue(sk, &timeout); 1046 dccp_terminate_connection(sk); 1047 } 1048 1049 /* 1050 * Flush write queue. This may be necessary in several cases: 1051 * - we have been closed by the peer but still have application data; 1052 * - abortive termination (unread data or zero linger time), 1053 * - normal termination but queue could not be flushed within time limit 1054 */ 1055 __skb_queue_purge(&sk->sk_write_queue); 1056 1057 sk_stream_wait_close(sk, timeout); 1058 1059 adjudge_to_death: 1060 state = sk->sk_state; 1061 sock_hold(sk); 1062 sock_orphan(sk); 1063 1064 /* 1065 * It is the last release_sock in its life. It will remove backlog. 1066 */ 1067 release_sock(sk); 1068 /* 1069 * Now socket is owned by kernel and we acquire BH lock 1070 * to finish close. No need to check for user refs. 1071 */ 1072 local_bh_disable(); 1073 bh_lock_sock(sk); 1074 WARN_ON(sock_owned_by_user(sk)); 1075 1076 percpu_counter_inc(sk->sk_prot->orphan_count); 1077 1078 /* Have we already been destroyed by a softirq or backlog? */ 1079 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED) 1080 goto out; 1081 1082 if (sk->sk_state == DCCP_CLOSED) 1083 inet_csk_destroy_sock(sk); 1084 1085 /* Otherwise, socket is reprieved until protocol close. */ 1086 1087 out: 1088 bh_unlock_sock(sk); 1089 local_bh_enable(); 1090 sock_put(sk); 1091 } 1092 1093 EXPORT_SYMBOL_GPL(dccp_close); 1094 1095 void dccp_shutdown(struct sock *sk, int how) 1096 { 1097 dccp_pr_debug("called shutdown(%x)\n", how); 1098 } 1099 1100 EXPORT_SYMBOL_GPL(dccp_shutdown); 1101 1102 static inline int __init dccp_mib_init(void) 1103 { 1104 dccp_statistics = alloc_percpu(struct dccp_mib); 1105 if (!dccp_statistics) 1106 return -ENOMEM; 1107 return 0; 1108 } 1109 1110 static inline void dccp_mib_exit(void) 1111 { 1112 free_percpu(dccp_statistics); 1113 } 1114 1115 static int thash_entries; 1116 module_param(thash_entries, int, 0444); 1117 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets"); 1118 1119 #ifdef CONFIG_IP_DCCP_DEBUG 1120 bool dccp_debug; 1121 module_param(dccp_debug, bool, 0644); 1122 MODULE_PARM_DESC(dccp_debug, "Enable debug messages"); 1123 1124 EXPORT_SYMBOL_GPL(dccp_debug); 1125 #endif 1126 1127 static int __init dccp_init(void) 1128 { 1129 unsigned long goal; 1130 unsigned long nr_pages = totalram_pages(); 1131 int ehash_order, bhash_order, i; 1132 int rc; 1133 1134 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) > 1135 FIELD_SIZEOF(struct sk_buff, cb)); 1136 rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL); 1137 if (rc) 1138 goto out_fail; 1139 inet_hashinfo_init(&dccp_hashinfo); 1140 rc = inet_hashinfo2_init_mod(&dccp_hashinfo); 1141 if (rc) 1142 goto out_fail; 1143 rc = -ENOBUFS; 1144 dccp_hashinfo.bind_bucket_cachep = 1145 kmem_cache_create("dccp_bind_bucket", 1146 sizeof(struct inet_bind_bucket), 0, 1147 SLAB_HWCACHE_ALIGN, NULL); 1148 if (!dccp_hashinfo.bind_bucket_cachep) 1149 goto out_free_percpu; 1150 1151 /* 1152 * Size and allocate the main established and bind bucket 1153 * hash tables. 1154 * 1155 * The methodology is similar to that of the buffer cache. 1156 */ 1157 if (nr_pages >= (128 * 1024)) 1158 goal = nr_pages >> (21 - PAGE_SHIFT); 1159 else 1160 goal = nr_pages >> (23 - PAGE_SHIFT); 1161 1162 if (thash_entries) 1163 goal = (thash_entries * 1164 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT; 1165 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++) 1166 ; 1167 do { 1168 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE / 1169 sizeof(struct inet_ehash_bucket); 1170 1171 while (hash_size & (hash_size - 1)) 1172 hash_size--; 1173 dccp_hashinfo.ehash_mask = hash_size - 1; 1174 dccp_hashinfo.ehash = (struct inet_ehash_bucket *) 1175 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order); 1176 } while (!dccp_hashinfo.ehash && --ehash_order > 0); 1177 1178 if (!dccp_hashinfo.ehash) { 1179 DCCP_CRIT("Failed to allocate DCCP established hash table"); 1180 goto out_free_bind_bucket_cachep; 1181 } 1182 1183 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++) 1184 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i); 1185 1186 if (inet_ehash_locks_alloc(&dccp_hashinfo)) 1187 goto out_free_dccp_ehash; 1188 1189 bhash_order = ehash_order; 1190 1191 do { 1192 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE / 1193 sizeof(struct inet_bind_hashbucket); 1194 if ((dccp_hashinfo.bhash_size > (64 * 1024)) && 1195 bhash_order > 0) 1196 continue; 1197 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *) 1198 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order); 1199 } while (!dccp_hashinfo.bhash && --bhash_order >= 0); 1200 1201 if (!dccp_hashinfo.bhash) { 1202 DCCP_CRIT("Failed to allocate DCCP bind hash table"); 1203 goto out_free_dccp_locks; 1204 } 1205 1206 for (i = 0; i < dccp_hashinfo.bhash_size; i++) { 1207 spin_lock_init(&dccp_hashinfo.bhash[i].lock); 1208 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain); 1209 } 1210 1211 rc = dccp_mib_init(); 1212 if (rc) 1213 goto out_free_dccp_bhash; 1214 1215 rc = dccp_ackvec_init(); 1216 if (rc) 1217 goto out_free_dccp_mib; 1218 1219 rc = dccp_sysctl_init(); 1220 if (rc) 1221 goto out_ackvec_exit; 1222 1223 rc = ccid_initialize_builtins(); 1224 if (rc) 1225 goto out_sysctl_exit; 1226 1227 dccp_timestamping_init(); 1228 1229 return 0; 1230 1231 out_sysctl_exit: 1232 dccp_sysctl_exit(); 1233 out_ackvec_exit: 1234 dccp_ackvec_exit(); 1235 out_free_dccp_mib: 1236 dccp_mib_exit(); 1237 out_free_dccp_bhash: 1238 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order); 1239 out_free_dccp_locks: 1240 inet_ehash_locks_free(&dccp_hashinfo); 1241 out_free_dccp_ehash: 1242 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order); 1243 out_free_bind_bucket_cachep: 1244 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); 1245 out_free_percpu: 1246 percpu_counter_destroy(&dccp_orphan_count); 1247 out_fail: 1248 dccp_hashinfo.bhash = NULL; 1249 dccp_hashinfo.ehash = NULL; 1250 dccp_hashinfo.bind_bucket_cachep = NULL; 1251 return rc; 1252 } 1253 1254 static void __exit dccp_fini(void) 1255 { 1256 ccid_cleanup_builtins(); 1257 dccp_mib_exit(); 1258 free_pages((unsigned long)dccp_hashinfo.bhash, 1259 get_order(dccp_hashinfo.bhash_size * 1260 sizeof(struct inet_bind_hashbucket))); 1261 free_pages((unsigned long)dccp_hashinfo.ehash, 1262 get_order((dccp_hashinfo.ehash_mask + 1) * 1263 sizeof(struct inet_ehash_bucket))); 1264 inet_ehash_locks_free(&dccp_hashinfo); 1265 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); 1266 dccp_ackvec_exit(); 1267 dccp_sysctl_exit(); 1268 percpu_counter_destroy(&dccp_orphan_count); 1269 } 1270 1271 module_init(dccp_init); 1272 module_exit(dccp_fini); 1273 1274 MODULE_LICENSE("GPL"); 1275 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>"); 1276 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol"); 1277