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 static int dccp_getsockopt_service(struct sock *sk, int len, 579 __be32 __user *optval, 580 int __user *optlen) 581 { 582 const struct dccp_sock *dp = dccp_sk(sk); 583 const struct dccp_service_list *sl; 584 int err = -ENOENT, slen = 0, total_len = sizeof(u32); 585 586 lock_sock(sk); 587 if ((sl = dp->dccps_service_list) != NULL) { 588 slen = sl->dccpsl_nr * sizeof(u32); 589 total_len += slen; 590 } 591 592 err = -EINVAL; 593 if (total_len > len) 594 goto out; 595 596 err = 0; 597 if (put_user(total_len, optlen) || 598 put_user(dp->dccps_service, optval) || 599 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen))) 600 err = -EFAULT; 601 out: 602 release_sock(sk); 603 return err; 604 } 605 606 static int do_dccp_getsockopt(struct sock *sk, int level, int optname, 607 char __user *optval, int __user *optlen) 608 { 609 struct dccp_sock *dp; 610 int val, len; 611 612 if (get_user(len, optlen)) 613 return -EFAULT; 614 615 if (len < (int)sizeof(int)) 616 return -EINVAL; 617 618 dp = dccp_sk(sk); 619 620 switch (optname) { 621 case DCCP_SOCKOPT_PACKET_SIZE: 622 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n"); 623 return 0; 624 case DCCP_SOCKOPT_SERVICE: 625 return dccp_getsockopt_service(sk, len, 626 (__be32 __user *)optval, optlen); 627 case DCCP_SOCKOPT_GET_CUR_MPS: 628 val = dp->dccps_mss_cache; 629 break; 630 case DCCP_SOCKOPT_AVAILABLE_CCIDS: 631 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen); 632 case DCCP_SOCKOPT_TX_CCID: 633 val = ccid_get_current_tx_ccid(dp); 634 if (val < 0) 635 return -ENOPROTOOPT; 636 break; 637 case DCCP_SOCKOPT_RX_CCID: 638 val = ccid_get_current_rx_ccid(dp); 639 if (val < 0) 640 return -ENOPROTOOPT; 641 break; 642 case DCCP_SOCKOPT_SERVER_TIMEWAIT: 643 val = dp->dccps_server_timewait; 644 break; 645 case DCCP_SOCKOPT_SEND_CSCOV: 646 val = dp->dccps_pcslen; 647 break; 648 case DCCP_SOCKOPT_RECV_CSCOV: 649 val = dp->dccps_pcrlen; 650 break; 651 case DCCP_SOCKOPT_QPOLICY_ID: 652 val = dp->dccps_qpolicy; 653 break; 654 case DCCP_SOCKOPT_QPOLICY_TXQLEN: 655 val = dp->dccps_tx_qlen; 656 break; 657 case 128 ... 191: 658 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname, 659 len, (u32 __user *)optval, optlen); 660 case 192 ... 255: 661 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname, 662 len, (u32 __user *)optval, optlen); 663 default: 664 return -ENOPROTOOPT; 665 } 666 667 len = sizeof(val); 668 if (put_user(len, optlen) || copy_to_user(optval, &val, len)) 669 return -EFAULT; 670 671 return 0; 672 } 673 674 int dccp_getsockopt(struct sock *sk, int level, int optname, 675 char __user *optval, int __user *optlen) 676 { 677 if (level != SOL_DCCP) 678 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level, 679 optname, optval, 680 optlen); 681 return do_dccp_getsockopt(sk, level, optname, optval, optlen); 682 } 683 684 EXPORT_SYMBOL_GPL(dccp_getsockopt); 685 686 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb) 687 { 688 struct cmsghdr *cmsg; 689 690 /* 691 * Assign an (opaque) qpolicy priority value to skb->priority. 692 * 693 * We are overloading this skb field for use with the qpolicy subystem. 694 * The skb->priority is normally used for the SO_PRIORITY option, which 695 * is initialised from sk_priority. Since the assignment of sk_priority 696 * to skb->priority happens later (on layer 3), we overload this field 697 * for use with queueing priorities as long as the skb is on layer 4. 698 * The default priority value (if nothing is set) is 0. 699 */ 700 skb->priority = 0; 701 702 for_each_cmsghdr(cmsg, msg) { 703 if (!CMSG_OK(msg, cmsg)) 704 return -EINVAL; 705 706 if (cmsg->cmsg_level != SOL_DCCP) 707 continue; 708 709 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX && 710 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type)) 711 return -EINVAL; 712 713 switch (cmsg->cmsg_type) { 714 case DCCP_SCM_PRIORITY: 715 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32))) 716 return -EINVAL; 717 skb->priority = *(__u32 *)CMSG_DATA(cmsg); 718 break; 719 default: 720 return -EINVAL; 721 } 722 } 723 return 0; 724 } 725 726 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 727 { 728 const struct dccp_sock *dp = dccp_sk(sk); 729 const int flags = msg->msg_flags; 730 const int noblock = flags & MSG_DONTWAIT; 731 struct sk_buff *skb; 732 int rc, size; 733 long timeo; 734 735 trace_dccp_probe(sk, len); 736 737 if (len > dp->dccps_mss_cache) 738 return -EMSGSIZE; 739 740 lock_sock(sk); 741 742 if (dccp_qpolicy_full(sk)) { 743 rc = -EAGAIN; 744 goto out_release; 745 } 746 747 timeo = sock_sndtimeo(sk, noblock); 748 749 /* 750 * We have to use sk_stream_wait_connect here to set sk_write_pending, 751 * so that the trick in dccp_rcv_request_sent_state_process. 752 */ 753 /* Wait for a connection to finish. */ 754 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN)) 755 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0) 756 goto out_release; 757 758 size = sk->sk_prot->max_header + len; 759 release_sock(sk); 760 skb = sock_alloc_send_skb(sk, size, noblock, &rc); 761 lock_sock(sk); 762 if (skb == NULL) 763 goto out_release; 764 765 if (sk->sk_state == DCCP_CLOSED) { 766 rc = -ENOTCONN; 767 goto out_discard; 768 } 769 770 skb_reserve(skb, sk->sk_prot->max_header); 771 rc = memcpy_from_msg(skb_put(skb, len), msg, len); 772 if (rc != 0) 773 goto out_discard; 774 775 rc = dccp_msghdr_parse(msg, skb); 776 if (rc != 0) 777 goto out_discard; 778 779 dccp_qpolicy_push(sk, skb); 780 /* 781 * The xmit_timer is set if the TX CCID is rate-based and will expire 782 * when congestion control permits to release further packets into the 783 * network. Window-based CCIDs do not use this timer. 784 */ 785 if (!timer_pending(&dp->dccps_xmit_timer)) 786 dccp_write_xmit(sk); 787 out_release: 788 release_sock(sk); 789 return rc ? : len; 790 out_discard: 791 kfree_skb(skb); 792 goto out_release; 793 } 794 795 EXPORT_SYMBOL_GPL(dccp_sendmsg); 796 797 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 798 int flags, int *addr_len) 799 { 800 const struct dccp_hdr *dh; 801 long timeo; 802 803 lock_sock(sk); 804 805 if (sk->sk_state == DCCP_LISTEN) { 806 len = -ENOTCONN; 807 goto out; 808 } 809 810 timeo = sock_rcvtimeo(sk, nonblock); 811 812 do { 813 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); 814 815 if (skb == NULL) 816 goto verify_sock_status; 817 818 dh = dccp_hdr(skb); 819 820 switch (dh->dccph_type) { 821 case DCCP_PKT_DATA: 822 case DCCP_PKT_DATAACK: 823 goto found_ok_skb; 824 825 case DCCP_PKT_CLOSE: 826 case DCCP_PKT_CLOSEREQ: 827 if (!(flags & MSG_PEEK)) 828 dccp_finish_passive_close(sk); 829 /* fall through */ 830 case DCCP_PKT_RESET: 831 dccp_pr_debug("found fin (%s) ok!\n", 832 dccp_packet_name(dh->dccph_type)); 833 len = 0; 834 goto found_fin_ok; 835 default: 836 dccp_pr_debug("packet_type=%s\n", 837 dccp_packet_name(dh->dccph_type)); 838 sk_eat_skb(sk, skb); 839 } 840 verify_sock_status: 841 if (sock_flag(sk, SOCK_DONE)) { 842 len = 0; 843 break; 844 } 845 846 if (sk->sk_err) { 847 len = sock_error(sk); 848 break; 849 } 850 851 if (sk->sk_shutdown & RCV_SHUTDOWN) { 852 len = 0; 853 break; 854 } 855 856 if (sk->sk_state == DCCP_CLOSED) { 857 if (!sock_flag(sk, SOCK_DONE)) { 858 /* This occurs when user tries to read 859 * from never connected socket. 860 */ 861 len = -ENOTCONN; 862 break; 863 } 864 len = 0; 865 break; 866 } 867 868 if (!timeo) { 869 len = -EAGAIN; 870 break; 871 } 872 873 if (signal_pending(current)) { 874 len = sock_intr_errno(timeo); 875 break; 876 } 877 878 sk_wait_data(sk, &timeo, NULL); 879 continue; 880 found_ok_skb: 881 if (len > skb->len) 882 len = skb->len; 883 else if (len < skb->len) 884 msg->msg_flags |= MSG_TRUNC; 885 886 if (skb_copy_datagram_msg(skb, 0, msg, len)) { 887 /* Exception. Bailout! */ 888 len = -EFAULT; 889 break; 890 } 891 if (flags & MSG_TRUNC) 892 len = skb->len; 893 found_fin_ok: 894 if (!(flags & MSG_PEEK)) 895 sk_eat_skb(sk, skb); 896 break; 897 } while (1); 898 out: 899 release_sock(sk); 900 return len; 901 } 902 903 EXPORT_SYMBOL_GPL(dccp_recvmsg); 904 905 int inet_dccp_listen(struct socket *sock, int backlog) 906 { 907 struct sock *sk = sock->sk; 908 unsigned char old_state; 909 int err; 910 911 lock_sock(sk); 912 913 err = -EINVAL; 914 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP) 915 goto out; 916 917 old_state = sk->sk_state; 918 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN))) 919 goto out; 920 921 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 922 /* Really, if the socket is already in listen state 923 * we can only allow the backlog to be adjusted. 924 */ 925 if (old_state != DCCP_LISTEN) { 926 /* 927 * FIXME: here it probably should be sk->sk_prot->listen_start 928 * see tcp_listen_start 929 */ 930 err = dccp_listen_start(sk, backlog); 931 if (err) 932 goto out; 933 } 934 err = 0; 935 936 out: 937 release_sock(sk); 938 return err; 939 } 940 941 EXPORT_SYMBOL_GPL(inet_dccp_listen); 942 943 static void dccp_terminate_connection(struct sock *sk) 944 { 945 u8 next_state = DCCP_CLOSED; 946 947 switch (sk->sk_state) { 948 case DCCP_PASSIVE_CLOSE: 949 case DCCP_PASSIVE_CLOSEREQ: 950 dccp_finish_passive_close(sk); 951 break; 952 case DCCP_PARTOPEN: 953 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk); 954 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 955 /* fall through */ 956 case DCCP_OPEN: 957 dccp_send_close(sk, 1); 958 959 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER && 960 !dccp_sk(sk)->dccps_server_timewait) 961 next_state = DCCP_ACTIVE_CLOSEREQ; 962 else 963 next_state = DCCP_CLOSING; 964 /* fall through */ 965 default: 966 dccp_set_state(sk, next_state); 967 } 968 } 969 970 void dccp_close(struct sock *sk, long timeout) 971 { 972 struct dccp_sock *dp = dccp_sk(sk); 973 struct sk_buff *skb; 974 u32 data_was_unread = 0; 975 int state; 976 977 lock_sock(sk); 978 979 sk->sk_shutdown = SHUTDOWN_MASK; 980 981 if (sk->sk_state == DCCP_LISTEN) { 982 dccp_set_state(sk, DCCP_CLOSED); 983 984 /* Special case. */ 985 inet_csk_listen_stop(sk); 986 987 goto adjudge_to_death; 988 } 989 990 sk_stop_timer(sk, &dp->dccps_xmit_timer); 991 992 /* 993 * We need to flush the recv. buffs. We do this only on the 994 * descriptor close, not protocol-sourced closes, because the 995 *reader process may not have drained the data yet! 996 */ 997 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { 998 data_was_unread += skb->len; 999 __kfree_skb(skb); 1000 } 1001 1002 /* If socket has been already reset kill it. */ 1003 if (sk->sk_state == DCCP_CLOSED) 1004 goto adjudge_to_death; 1005 1006 if (data_was_unread) { 1007 /* Unread data was tossed, send an appropriate Reset Code */ 1008 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread); 1009 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED); 1010 dccp_set_state(sk, DCCP_CLOSED); 1011 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { 1012 /* Check zero linger _after_ checking for unread data. */ 1013 sk->sk_prot->disconnect(sk, 0); 1014 } else if (sk->sk_state != DCCP_CLOSED) { 1015 /* 1016 * Normal connection termination. May need to wait if there are 1017 * still packets in the TX queue that are delayed by the CCID. 1018 */ 1019 dccp_flush_write_queue(sk, &timeout); 1020 dccp_terminate_connection(sk); 1021 } 1022 1023 /* 1024 * Flush write queue. This may be necessary in several cases: 1025 * - we have been closed by the peer but still have application data; 1026 * - abortive termination (unread data or zero linger time), 1027 * - normal termination but queue could not be flushed within time limit 1028 */ 1029 __skb_queue_purge(&sk->sk_write_queue); 1030 1031 sk_stream_wait_close(sk, timeout); 1032 1033 adjudge_to_death: 1034 state = sk->sk_state; 1035 sock_hold(sk); 1036 sock_orphan(sk); 1037 1038 /* 1039 * It is the last release_sock in its life. It will remove backlog. 1040 */ 1041 release_sock(sk); 1042 /* 1043 * Now socket is owned by kernel and we acquire BH lock 1044 * to finish close. No need to check for user refs. 1045 */ 1046 local_bh_disable(); 1047 bh_lock_sock(sk); 1048 WARN_ON(sock_owned_by_user(sk)); 1049 1050 percpu_counter_inc(sk->sk_prot->orphan_count); 1051 1052 /* Have we already been destroyed by a softirq or backlog? */ 1053 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED) 1054 goto out; 1055 1056 if (sk->sk_state == DCCP_CLOSED) 1057 inet_csk_destroy_sock(sk); 1058 1059 /* Otherwise, socket is reprieved until protocol close. */ 1060 1061 out: 1062 bh_unlock_sock(sk); 1063 local_bh_enable(); 1064 sock_put(sk); 1065 } 1066 1067 EXPORT_SYMBOL_GPL(dccp_close); 1068 1069 void dccp_shutdown(struct sock *sk, int how) 1070 { 1071 dccp_pr_debug("called shutdown(%x)\n", how); 1072 } 1073 1074 EXPORT_SYMBOL_GPL(dccp_shutdown); 1075 1076 static inline int __init dccp_mib_init(void) 1077 { 1078 dccp_statistics = alloc_percpu(struct dccp_mib); 1079 if (!dccp_statistics) 1080 return -ENOMEM; 1081 return 0; 1082 } 1083 1084 static inline void dccp_mib_exit(void) 1085 { 1086 free_percpu(dccp_statistics); 1087 } 1088 1089 static int thash_entries; 1090 module_param(thash_entries, int, 0444); 1091 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets"); 1092 1093 #ifdef CONFIG_IP_DCCP_DEBUG 1094 bool dccp_debug; 1095 module_param(dccp_debug, bool, 0644); 1096 MODULE_PARM_DESC(dccp_debug, "Enable debug messages"); 1097 1098 EXPORT_SYMBOL_GPL(dccp_debug); 1099 #endif 1100 1101 static int __init dccp_init(void) 1102 { 1103 unsigned long goal; 1104 unsigned long nr_pages = totalram_pages(); 1105 int ehash_order, bhash_order, i; 1106 int rc; 1107 1108 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) > 1109 sizeof_field(struct sk_buff, cb)); 1110 rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL); 1111 if (rc) 1112 goto out_fail; 1113 inet_hashinfo_init(&dccp_hashinfo); 1114 rc = inet_hashinfo2_init_mod(&dccp_hashinfo); 1115 if (rc) 1116 goto out_free_percpu; 1117 rc = -ENOBUFS; 1118 dccp_hashinfo.bind_bucket_cachep = 1119 kmem_cache_create("dccp_bind_bucket", 1120 sizeof(struct inet_bind_bucket), 0, 1121 SLAB_HWCACHE_ALIGN, NULL); 1122 if (!dccp_hashinfo.bind_bucket_cachep) 1123 goto out_free_hashinfo2; 1124 1125 /* 1126 * Size and allocate the main established and bind bucket 1127 * hash tables. 1128 * 1129 * The methodology is similar to that of the buffer cache. 1130 */ 1131 if (nr_pages >= (128 * 1024)) 1132 goal = nr_pages >> (21 - PAGE_SHIFT); 1133 else 1134 goal = nr_pages >> (23 - PAGE_SHIFT); 1135 1136 if (thash_entries) 1137 goal = (thash_entries * 1138 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT; 1139 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++) 1140 ; 1141 do { 1142 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE / 1143 sizeof(struct inet_ehash_bucket); 1144 1145 while (hash_size & (hash_size - 1)) 1146 hash_size--; 1147 dccp_hashinfo.ehash_mask = hash_size - 1; 1148 dccp_hashinfo.ehash = (struct inet_ehash_bucket *) 1149 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order); 1150 } while (!dccp_hashinfo.ehash && --ehash_order > 0); 1151 1152 if (!dccp_hashinfo.ehash) { 1153 DCCP_CRIT("Failed to allocate DCCP established hash table"); 1154 goto out_free_bind_bucket_cachep; 1155 } 1156 1157 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++) 1158 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i); 1159 1160 if (inet_ehash_locks_alloc(&dccp_hashinfo)) 1161 goto out_free_dccp_ehash; 1162 1163 bhash_order = ehash_order; 1164 1165 do { 1166 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE / 1167 sizeof(struct inet_bind_hashbucket); 1168 if ((dccp_hashinfo.bhash_size > (64 * 1024)) && 1169 bhash_order > 0) 1170 continue; 1171 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *) 1172 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order); 1173 } while (!dccp_hashinfo.bhash && --bhash_order >= 0); 1174 1175 if (!dccp_hashinfo.bhash) { 1176 DCCP_CRIT("Failed to allocate DCCP bind hash table"); 1177 goto out_free_dccp_locks; 1178 } 1179 1180 for (i = 0; i < dccp_hashinfo.bhash_size; i++) { 1181 spin_lock_init(&dccp_hashinfo.bhash[i].lock); 1182 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain); 1183 } 1184 1185 rc = dccp_mib_init(); 1186 if (rc) 1187 goto out_free_dccp_bhash; 1188 1189 rc = dccp_ackvec_init(); 1190 if (rc) 1191 goto out_free_dccp_mib; 1192 1193 rc = dccp_sysctl_init(); 1194 if (rc) 1195 goto out_ackvec_exit; 1196 1197 rc = ccid_initialize_builtins(); 1198 if (rc) 1199 goto out_sysctl_exit; 1200 1201 dccp_timestamping_init(); 1202 1203 return 0; 1204 1205 out_sysctl_exit: 1206 dccp_sysctl_exit(); 1207 out_ackvec_exit: 1208 dccp_ackvec_exit(); 1209 out_free_dccp_mib: 1210 dccp_mib_exit(); 1211 out_free_dccp_bhash: 1212 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order); 1213 out_free_dccp_locks: 1214 inet_ehash_locks_free(&dccp_hashinfo); 1215 out_free_dccp_ehash: 1216 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order); 1217 out_free_bind_bucket_cachep: 1218 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); 1219 out_free_hashinfo2: 1220 inet_hashinfo2_free_mod(&dccp_hashinfo); 1221 out_free_percpu: 1222 percpu_counter_destroy(&dccp_orphan_count); 1223 out_fail: 1224 dccp_hashinfo.bhash = NULL; 1225 dccp_hashinfo.ehash = NULL; 1226 dccp_hashinfo.bind_bucket_cachep = NULL; 1227 return rc; 1228 } 1229 1230 static void __exit dccp_fini(void) 1231 { 1232 ccid_cleanup_builtins(); 1233 dccp_mib_exit(); 1234 free_pages((unsigned long)dccp_hashinfo.bhash, 1235 get_order(dccp_hashinfo.bhash_size * 1236 sizeof(struct inet_bind_hashbucket))); 1237 free_pages((unsigned long)dccp_hashinfo.ehash, 1238 get_order((dccp_hashinfo.ehash_mask + 1) * 1239 sizeof(struct inet_ehash_bucket))); 1240 inet_ehash_locks_free(&dccp_hashinfo); 1241 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); 1242 dccp_ackvec_exit(); 1243 dccp_sysctl_exit(); 1244 inet_hashinfo2_free_mod(&dccp_hashinfo); 1245 percpu_counter_destroy(&dccp_orphan_count); 1246 } 1247 1248 module_init(dccp_init); 1249 module_exit(dccp_fini); 1250 1251 MODULE_LICENSE("GPL"); 1252 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>"); 1253 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol"); 1254