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 DEFINE_PER_CPU(unsigned int, dccp_orphan_count); 46 EXPORT_PER_CPU_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 fallthrough; 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 void dccp_destruct_common(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 } 181 EXPORT_SYMBOL_GPL(dccp_destruct_common); 182 183 static void dccp_sk_destruct(struct sock *sk) 184 { 185 dccp_destruct_common(sk); 186 inet_sock_destruct(sk); 187 } 188 189 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized) 190 { 191 struct dccp_sock *dp = dccp_sk(sk); 192 struct inet_connection_sock *icsk = inet_csk(sk); 193 194 pr_warn_once("DCCP is deprecated and scheduled to be removed in 2025, " 195 "please contact the netdev mailing list\n"); 196 197 icsk->icsk_rto = DCCP_TIMEOUT_INIT; 198 icsk->icsk_syn_retries = sysctl_dccp_request_retries; 199 sk->sk_state = DCCP_CLOSED; 200 sk->sk_write_space = dccp_write_space; 201 sk->sk_destruct = dccp_sk_destruct; 202 icsk->icsk_sync_mss = dccp_sync_mss; 203 dp->dccps_mss_cache = 536; 204 dp->dccps_rate_last = jiffies; 205 dp->dccps_role = DCCP_ROLE_UNDEFINED; 206 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT; 207 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen; 208 209 dccp_init_xmit_timers(sk); 210 211 INIT_LIST_HEAD(&dp->dccps_featneg); 212 /* control socket doesn't need feat nego */ 213 if (likely(ctl_sock_initialized)) 214 return dccp_feat_init(sk); 215 return 0; 216 } 217 218 EXPORT_SYMBOL_GPL(dccp_init_sock); 219 220 void dccp_destroy_sock(struct sock *sk) 221 { 222 struct dccp_sock *dp = dccp_sk(sk); 223 224 __skb_queue_purge(&sk->sk_write_queue); 225 if (sk->sk_send_head != NULL) { 226 kfree_skb(sk->sk_send_head); 227 sk->sk_send_head = NULL; 228 } 229 230 /* Clean up a referenced DCCP bind bucket. */ 231 if (inet_csk(sk)->icsk_bind_hash != NULL) 232 inet_put_port(sk); 233 234 kfree(dp->dccps_service_list); 235 dp->dccps_service_list = NULL; 236 237 if (dp->dccps_hc_rx_ackvec != NULL) { 238 dccp_ackvec_free(dp->dccps_hc_rx_ackvec); 239 dp->dccps_hc_rx_ackvec = NULL; 240 } 241 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk); 242 dp->dccps_hc_rx_ccid = NULL; 243 244 /* clean up feature negotiation state */ 245 dccp_feat_list_purge(&dp->dccps_featneg); 246 } 247 248 EXPORT_SYMBOL_GPL(dccp_destroy_sock); 249 250 static inline int dccp_need_reset(int state) 251 { 252 return state != DCCP_CLOSED && state != DCCP_LISTEN && 253 state != DCCP_REQUESTING; 254 } 255 256 int dccp_disconnect(struct sock *sk, int flags) 257 { 258 struct inet_connection_sock *icsk = inet_csk(sk); 259 struct inet_sock *inet = inet_sk(sk); 260 struct dccp_sock *dp = dccp_sk(sk); 261 const int old_state = sk->sk_state; 262 263 if (old_state != DCCP_CLOSED) 264 dccp_set_state(sk, DCCP_CLOSED); 265 266 /* 267 * This corresponds to the ABORT function of RFC793, sec. 3.8 268 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted". 269 */ 270 if (old_state == DCCP_LISTEN) { 271 inet_csk_listen_stop(sk); 272 } else if (dccp_need_reset(old_state)) { 273 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED); 274 sk->sk_err = ECONNRESET; 275 } else if (old_state == DCCP_REQUESTING) 276 sk->sk_err = ECONNRESET; 277 278 dccp_clear_xmit_timers(sk); 279 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk); 280 dp->dccps_hc_rx_ccid = NULL; 281 282 __skb_queue_purge(&sk->sk_receive_queue); 283 __skb_queue_purge(&sk->sk_write_queue); 284 if (sk->sk_send_head != NULL) { 285 __kfree_skb(sk->sk_send_head); 286 sk->sk_send_head = NULL; 287 } 288 289 inet->inet_dport = 0; 290 291 inet_bhash2_reset_saddr(sk); 292 293 sk->sk_shutdown = 0; 294 sock_reset_flag(sk, SOCK_DONE); 295 296 icsk->icsk_backoff = 0; 297 inet_csk_delack_init(sk); 298 __sk_dst_reset(sk); 299 300 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash); 301 302 sk_error_report(sk); 303 return 0; 304 } 305 306 EXPORT_SYMBOL_GPL(dccp_disconnect); 307 308 /* 309 * Wait for a DCCP event. 310 * 311 * Note that we don't need to lock the socket, as the upper poll layers 312 * take care of normal races (between the test and the event) and we don't 313 * go look at any of the socket buffers directly. 314 */ 315 __poll_t dccp_poll(struct file *file, struct socket *sock, 316 poll_table *wait) 317 { 318 __poll_t mask; 319 struct sock *sk = sock->sk; 320 321 sock_poll_wait(file, sock, wait); 322 if (sk->sk_state == DCCP_LISTEN) 323 return inet_csk_listen_poll(sk); 324 325 /* Socket is not locked. We are protected from async events 326 by poll logic and correct handling of state changes 327 made by another threads is impossible in any case. 328 */ 329 330 mask = 0; 331 if (sk->sk_err) 332 mask = EPOLLERR; 333 334 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED) 335 mask |= EPOLLHUP; 336 if (sk->sk_shutdown & RCV_SHUTDOWN) 337 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 338 339 /* Connected? */ 340 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) { 341 if (atomic_read(&sk->sk_rmem_alloc) > 0) 342 mask |= EPOLLIN | EPOLLRDNORM; 343 344 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 345 if (sk_stream_is_writeable(sk)) { 346 mask |= EPOLLOUT | EPOLLWRNORM; 347 } else { /* send SIGIO later */ 348 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 349 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 350 351 /* Race breaker. If space is freed after 352 * wspace test but before the flags are set, 353 * IO signal will be lost. 354 */ 355 if (sk_stream_is_writeable(sk)) 356 mask |= EPOLLOUT | EPOLLWRNORM; 357 } 358 } 359 } 360 return mask; 361 } 362 363 EXPORT_SYMBOL_GPL(dccp_poll); 364 365 int dccp_ioctl(struct sock *sk, int cmd, int *karg) 366 { 367 int rc = -ENOTCONN; 368 369 lock_sock(sk); 370 371 if (sk->sk_state == DCCP_LISTEN) 372 goto out; 373 374 switch (cmd) { 375 case SIOCOUTQ: { 376 *karg = sk_wmem_alloc_get(sk); 377 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and 378 * always 0, comparably to UDP. 379 */ 380 381 rc = 0; 382 } 383 break; 384 case SIOCINQ: { 385 struct sk_buff *skb; 386 *karg = 0; 387 388 skb = skb_peek(&sk->sk_receive_queue); 389 if (skb != NULL) { 390 /* 391 * We will only return the amount of this packet since 392 * that is all that will be read. 393 */ 394 *karg = skb->len; 395 } 396 rc = 0; 397 } 398 break; 399 default: 400 rc = -ENOIOCTLCMD; 401 break; 402 } 403 out: 404 release_sock(sk); 405 return rc; 406 } 407 408 EXPORT_SYMBOL_GPL(dccp_ioctl); 409 410 static int dccp_setsockopt_service(struct sock *sk, const __be32 service, 411 sockptr_t optval, unsigned int optlen) 412 { 413 struct dccp_sock *dp = dccp_sk(sk); 414 struct dccp_service_list *sl = NULL; 415 416 if (service == DCCP_SERVICE_INVALID_VALUE || 417 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32)) 418 return -EINVAL; 419 420 if (optlen > sizeof(service)) { 421 sl = kmalloc(optlen, GFP_KERNEL); 422 if (sl == NULL) 423 return -ENOMEM; 424 425 sl->dccpsl_nr = optlen / sizeof(u32) - 1; 426 if (copy_from_sockptr_offset(sl->dccpsl_list, optval, 427 sizeof(service), optlen - sizeof(service)) || 428 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) { 429 kfree(sl); 430 return -EFAULT; 431 } 432 } 433 434 lock_sock(sk); 435 dp->dccps_service = service; 436 437 kfree(dp->dccps_service_list); 438 439 dp->dccps_service_list = sl; 440 release_sock(sk); 441 return 0; 442 } 443 444 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx) 445 { 446 u8 *list, len; 447 int i, rc; 448 449 if (cscov < 0 || cscov > 15) 450 return -EINVAL; 451 /* 452 * Populate a list of permissible values, in the range cscov...15. This 453 * is necessary since feature negotiation of single values only works if 454 * both sides incidentally choose the same value. Since the list starts 455 * lowest-value first, negotiation will pick the smallest shared value. 456 */ 457 if (cscov == 0) 458 return 0; 459 len = 16 - cscov; 460 461 list = kmalloc(len, GFP_KERNEL); 462 if (list == NULL) 463 return -ENOBUFS; 464 465 for (i = 0; i < len; i++) 466 list[i] = cscov++; 467 468 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len); 469 470 if (rc == 0) { 471 if (rx) 472 dccp_sk(sk)->dccps_pcrlen = cscov; 473 else 474 dccp_sk(sk)->dccps_pcslen = cscov; 475 } 476 kfree(list); 477 return rc; 478 } 479 480 static int dccp_setsockopt_ccid(struct sock *sk, int type, 481 sockptr_t optval, unsigned int optlen) 482 { 483 u8 *val; 484 int rc = 0; 485 486 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS) 487 return -EINVAL; 488 489 val = memdup_sockptr(optval, optlen); 490 if (IS_ERR(val)) 491 return PTR_ERR(val); 492 493 lock_sock(sk); 494 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID) 495 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen); 496 497 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID)) 498 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen); 499 release_sock(sk); 500 501 kfree(val); 502 return rc; 503 } 504 505 static int do_dccp_setsockopt(struct sock *sk, int level, int optname, 506 sockptr_t optval, unsigned int optlen) 507 { 508 struct dccp_sock *dp = dccp_sk(sk); 509 int val, err = 0; 510 511 switch (optname) { 512 case DCCP_SOCKOPT_PACKET_SIZE: 513 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n"); 514 return 0; 515 case DCCP_SOCKOPT_CHANGE_L: 516 case DCCP_SOCKOPT_CHANGE_R: 517 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n"); 518 return 0; 519 case DCCP_SOCKOPT_CCID: 520 case DCCP_SOCKOPT_RX_CCID: 521 case DCCP_SOCKOPT_TX_CCID: 522 return dccp_setsockopt_ccid(sk, optname, optval, optlen); 523 } 524 525 if (optlen < (int)sizeof(int)) 526 return -EINVAL; 527 528 if (copy_from_sockptr(&val, optval, sizeof(int))) 529 return -EFAULT; 530 531 if (optname == DCCP_SOCKOPT_SERVICE) 532 return dccp_setsockopt_service(sk, val, optval, optlen); 533 534 lock_sock(sk); 535 switch (optname) { 536 case DCCP_SOCKOPT_SERVER_TIMEWAIT: 537 if (dp->dccps_role != DCCP_ROLE_SERVER) 538 err = -EOPNOTSUPP; 539 else 540 dp->dccps_server_timewait = (val != 0); 541 break; 542 case DCCP_SOCKOPT_SEND_CSCOV: 543 err = dccp_setsockopt_cscov(sk, val, false); 544 break; 545 case DCCP_SOCKOPT_RECV_CSCOV: 546 err = dccp_setsockopt_cscov(sk, val, true); 547 break; 548 case DCCP_SOCKOPT_QPOLICY_ID: 549 if (sk->sk_state != DCCP_CLOSED) 550 err = -EISCONN; 551 else if (val < 0 || val >= DCCPQ_POLICY_MAX) 552 err = -EINVAL; 553 else 554 dp->dccps_qpolicy = val; 555 break; 556 case DCCP_SOCKOPT_QPOLICY_TXQLEN: 557 if (val < 0) 558 err = -EINVAL; 559 else 560 dp->dccps_tx_qlen = val; 561 break; 562 default: 563 err = -ENOPROTOOPT; 564 break; 565 } 566 release_sock(sk); 567 568 return err; 569 } 570 571 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 572 unsigned int optlen) 573 { 574 if (level != SOL_DCCP) 575 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level, 576 optname, optval, 577 optlen); 578 return do_dccp_setsockopt(sk, level, optname, optval, optlen); 579 } 580 581 EXPORT_SYMBOL_GPL(dccp_setsockopt); 582 583 static int dccp_getsockopt_service(struct sock *sk, int len, 584 __be32 __user *optval, 585 int __user *optlen) 586 { 587 const struct dccp_sock *dp = dccp_sk(sk); 588 const struct dccp_service_list *sl; 589 int err = -ENOENT, slen = 0, total_len = sizeof(u32); 590 591 lock_sock(sk); 592 if ((sl = dp->dccps_service_list) != NULL) { 593 slen = sl->dccpsl_nr * sizeof(u32); 594 total_len += slen; 595 } 596 597 err = -EINVAL; 598 if (total_len > len) 599 goto out; 600 601 err = 0; 602 if (put_user(total_len, optlen) || 603 put_user(dp->dccps_service, optval) || 604 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen))) 605 err = -EFAULT; 606 out: 607 release_sock(sk); 608 return err; 609 } 610 611 static int do_dccp_getsockopt(struct sock *sk, int level, int optname, 612 char __user *optval, int __user *optlen) 613 { 614 struct dccp_sock *dp; 615 int val, len; 616 617 if (get_user(len, optlen)) 618 return -EFAULT; 619 620 if (len < (int)sizeof(int)) 621 return -EINVAL; 622 623 dp = dccp_sk(sk); 624 625 switch (optname) { 626 case DCCP_SOCKOPT_PACKET_SIZE: 627 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n"); 628 return 0; 629 case DCCP_SOCKOPT_SERVICE: 630 return dccp_getsockopt_service(sk, len, 631 (__be32 __user *)optval, optlen); 632 case DCCP_SOCKOPT_GET_CUR_MPS: 633 val = dp->dccps_mss_cache; 634 break; 635 case DCCP_SOCKOPT_AVAILABLE_CCIDS: 636 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen); 637 case DCCP_SOCKOPT_TX_CCID: 638 val = ccid_get_current_tx_ccid(dp); 639 if (val < 0) 640 return -ENOPROTOOPT; 641 break; 642 case DCCP_SOCKOPT_RX_CCID: 643 val = ccid_get_current_rx_ccid(dp); 644 if (val < 0) 645 return -ENOPROTOOPT; 646 break; 647 case DCCP_SOCKOPT_SERVER_TIMEWAIT: 648 val = dp->dccps_server_timewait; 649 break; 650 case DCCP_SOCKOPT_SEND_CSCOV: 651 val = dp->dccps_pcslen; 652 break; 653 case DCCP_SOCKOPT_RECV_CSCOV: 654 val = dp->dccps_pcrlen; 655 break; 656 case DCCP_SOCKOPT_QPOLICY_ID: 657 val = dp->dccps_qpolicy; 658 break; 659 case DCCP_SOCKOPT_QPOLICY_TXQLEN: 660 val = dp->dccps_tx_qlen; 661 break; 662 case 128 ... 191: 663 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname, 664 len, (u32 __user *)optval, optlen); 665 case 192 ... 255: 666 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname, 667 len, (u32 __user *)optval, optlen); 668 default: 669 return -ENOPROTOOPT; 670 } 671 672 len = sizeof(val); 673 if (put_user(len, optlen) || copy_to_user(optval, &val, len)) 674 return -EFAULT; 675 676 return 0; 677 } 678 679 int dccp_getsockopt(struct sock *sk, int level, int optname, 680 char __user *optval, int __user *optlen) 681 { 682 if (level != SOL_DCCP) 683 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level, 684 optname, optval, 685 optlen); 686 return do_dccp_getsockopt(sk, level, optname, optval, optlen); 687 } 688 689 EXPORT_SYMBOL_GPL(dccp_getsockopt); 690 691 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb) 692 { 693 struct cmsghdr *cmsg; 694 695 /* 696 * Assign an (opaque) qpolicy priority value to skb->priority. 697 * 698 * We are overloading this skb field for use with the qpolicy subystem. 699 * The skb->priority is normally used for the SO_PRIORITY option, which 700 * is initialised from sk_priority. Since the assignment of sk_priority 701 * to skb->priority happens later (on layer 3), we overload this field 702 * for use with queueing priorities as long as the skb is on layer 4. 703 * The default priority value (if nothing is set) is 0. 704 */ 705 skb->priority = 0; 706 707 for_each_cmsghdr(cmsg, msg) { 708 if (!CMSG_OK(msg, cmsg)) 709 return -EINVAL; 710 711 if (cmsg->cmsg_level != SOL_DCCP) 712 continue; 713 714 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX && 715 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type)) 716 return -EINVAL; 717 718 switch (cmsg->cmsg_type) { 719 case DCCP_SCM_PRIORITY: 720 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32))) 721 return -EINVAL; 722 skb->priority = *(__u32 *)CMSG_DATA(cmsg); 723 break; 724 default: 725 return -EINVAL; 726 } 727 } 728 return 0; 729 } 730 731 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 732 { 733 const struct dccp_sock *dp = dccp_sk(sk); 734 const int flags = msg->msg_flags; 735 const int noblock = flags & MSG_DONTWAIT; 736 struct sk_buff *skb; 737 int rc, size; 738 long timeo; 739 740 trace_dccp_probe(sk, len); 741 742 if (len > dp->dccps_mss_cache) 743 return -EMSGSIZE; 744 745 lock_sock(sk); 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 (dccp_qpolicy_full(sk)) { 766 rc = -EAGAIN; 767 goto out_discard; 768 } 769 770 if (sk->sk_state == DCCP_CLOSED) { 771 rc = -ENOTCONN; 772 goto out_discard; 773 } 774 775 skb_reserve(skb, sk->sk_prot->max_header); 776 rc = memcpy_from_msg(skb_put(skb, len), msg, len); 777 if (rc != 0) 778 goto out_discard; 779 780 rc = dccp_msghdr_parse(msg, skb); 781 if (rc != 0) 782 goto out_discard; 783 784 dccp_qpolicy_push(sk, skb); 785 /* 786 * The xmit_timer is set if the TX CCID is rate-based and will expire 787 * when congestion control permits to release further packets into the 788 * network. Window-based CCIDs do not use this timer. 789 */ 790 if (!timer_pending(&dp->dccps_xmit_timer)) 791 dccp_write_xmit(sk); 792 out_release: 793 release_sock(sk); 794 return rc ? : len; 795 out_discard: 796 kfree_skb(skb); 797 goto out_release; 798 } 799 800 EXPORT_SYMBOL_GPL(dccp_sendmsg); 801 802 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags, 803 int *addr_len) 804 { 805 const struct dccp_hdr *dh; 806 long timeo; 807 808 lock_sock(sk); 809 810 if (sk->sk_state == DCCP_LISTEN) { 811 len = -ENOTCONN; 812 goto out; 813 } 814 815 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 816 817 do { 818 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); 819 820 if (skb == NULL) 821 goto verify_sock_status; 822 823 dh = dccp_hdr(skb); 824 825 switch (dh->dccph_type) { 826 case DCCP_PKT_DATA: 827 case DCCP_PKT_DATAACK: 828 goto found_ok_skb; 829 830 case DCCP_PKT_CLOSE: 831 case DCCP_PKT_CLOSEREQ: 832 if (!(flags & MSG_PEEK)) 833 dccp_finish_passive_close(sk); 834 fallthrough; 835 case DCCP_PKT_RESET: 836 dccp_pr_debug("found fin (%s) ok!\n", 837 dccp_packet_name(dh->dccph_type)); 838 len = 0; 839 goto found_fin_ok; 840 default: 841 dccp_pr_debug("packet_type=%s\n", 842 dccp_packet_name(dh->dccph_type)); 843 sk_eat_skb(sk, skb); 844 } 845 verify_sock_status: 846 if (sock_flag(sk, SOCK_DONE)) { 847 len = 0; 848 break; 849 } 850 851 if (sk->sk_err) { 852 len = sock_error(sk); 853 break; 854 } 855 856 if (sk->sk_shutdown & RCV_SHUTDOWN) { 857 len = 0; 858 break; 859 } 860 861 if (sk->sk_state == DCCP_CLOSED) { 862 if (!sock_flag(sk, SOCK_DONE)) { 863 /* This occurs when user tries to read 864 * from never connected socket. 865 */ 866 len = -ENOTCONN; 867 break; 868 } 869 len = 0; 870 break; 871 } 872 873 if (!timeo) { 874 len = -EAGAIN; 875 break; 876 } 877 878 if (signal_pending(current)) { 879 len = sock_intr_errno(timeo); 880 break; 881 } 882 883 sk_wait_data(sk, &timeo, NULL); 884 continue; 885 found_ok_skb: 886 if (len > skb->len) 887 len = skb->len; 888 else if (len < skb->len) 889 msg->msg_flags |= MSG_TRUNC; 890 891 if (skb_copy_datagram_msg(skb, 0, msg, len)) { 892 /* Exception. Bailout! */ 893 len = -EFAULT; 894 break; 895 } 896 if (flags & MSG_TRUNC) 897 len = skb->len; 898 found_fin_ok: 899 if (!(flags & MSG_PEEK)) 900 sk_eat_skb(sk, skb); 901 break; 902 } while (1); 903 out: 904 release_sock(sk); 905 return len; 906 } 907 908 EXPORT_SYMBOL_GPL(dccp_recvmsg); 909 910 int inet_dccp_listen(struct socket *sock, int backlog) 911 { 912 struct sock *sk = sock->sk; 913 unsigned char old_state; 914 int err; 915 916 lock_sock(sk); 917 918 err = -EINVAL; 919 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP) 920 goto out; 921 922 old_state = sk->sk_state; 923 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN))) 924 goto out; 925 926 WRITE_ONCE(sk->sk_max_ack_backlog, backlog); 927 /* Really, if the socket is already in listen state 928 * we can only allow the backlog to be adjusted. 929 */ 930 if (old_state != DCCP_LISTEN) { 931 struct dccp_sock *dp = dccp_sk(sk); 932 933 dp->dccps_role = DCCP_ROLE_LISTEN; 934 935 /* do not start to listen if feature negotiation setup fails */ 936 if (dccp_feat_finalise_settings(dp)) { 937 err = -EPROTO; 938 goto out; 939 } 940 941 err = inet_csk_listen_start(sk); 942 if (err) 943 goto out; 944 } 945 err = 0; 946 947 out: 948 release_sock(sk); 949 return err; 950 } 951 952 EXPORT_SYMBOL_GPL(inet_dccp_listen); 953 954 static void dccp_terminate_connection(struct sock *sk) 955 { 956 u8 next_state = DCCP_CLOSED; 957 958 switch (sk->sk_state) { 959 case DCCP_PASSIVE_CLOSE: 960 case DCCP_PASSIVE_CLOSEREQ: 961 dccp_finish_passive_close(sk); 962 break; 963 case DCCP_PARTOPEN: 964 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk); 965 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 966 fallthrough; 967 case DCCP_OPEN: 968 dccp_send_close(sk, 1); 969 970 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER && 971 !dccp_sk(sk)->dccps_server_timewait) 972 next_state = DCCP_ACTIVE_CLOSEREQ; 973 else 974 next_state = DCCP_CLOSING; 975 fallthrough; 976 default: 977 dccp_set_state(sk, next_state); 978 } 979 } 980 981 void dccp_close(struct sock *sk, long timeout) 982 { 983 struct dccp_sock *dp = dccp_sk(sk); 984 struct sk_buff *skb; 985 u32 data_was_unread = 0; 986 int state; 987 988 lock_sock(sk); 989 990 sk->sk_shutdown = SHUTDOWN_MASK; 991 992 if (sk->sk_state == DCCP_LISTEN) { 993 dccp_set_state(sk, DCCP_CLOSED); 994 995 /* Special case. */ 996 inet_csk_listen_stop(sk); 997 998 goto adjudge_to_death; 999 } 1000 1001 sk_stop_timer(sk, &dp->dccps_xmit_timer); 1002 1003 /* 1004 * We need to flush the recv. buffs. We do this only on the 1005 * descriptor close, not protocol-sourced closes, because the 1006 *reader process may not have drained the data yet! 1007 */ 1008 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { 1009 data_was_unread += skb->len; 1010 __kfree_skb(skb); 1011 } 1012 1013 /* If socket has been already reset kill it. */ 1014 if (sk->sk_state == DCCP_CLOSED) 1015 goto adjudge_to_death; 1016 1017 if (data_was_unread) { 1018 /* Unread data was tossed, send an appropriate Reset Code */ 1019 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread); 1020 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED); 1021 dccp_set_state(sk, DCCP_CLOSED); 1022 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { 1023 /* Check zero linger _after_ checking for unread data. */ 1024 sk->sk_prot->disconnect(sk, 0); 1025 } else if (sk->sk_state != DCCP_CLOSED) { 1026 /* 1027 * Normal connection termination. May need to wait if there are 1028 * still packets in the TX queue that are delayed by the CCID. 1029 */ 1030 dccp_flush_write_queue(sk, &timeout); 1031 dccp_terminate_connection(sk); 1032 } 1033 1034 /* 1035 * Flush write queue. This may be necessary in several cases: 1036 * - we have been closed by the peer but still have application data; 1037 * - abortive termination (unread data or zero linger time), 1038 * - normal termination but queue could not be flushed within time limit 1039 */ 1040 __skb_queue_purge(&sk->sk_write_queue); 1041 1042 sk_stream_wait_close(sk, timeout); 1043 1044 adjudge_to_death: 1045 state = sk->sk_state; 1046 sock_hold(sk); 1047 sock_orphan(sk); 1048 1049 /* 1050 * It is the last release_sock in its life. It will remove backlog. 1051 */ 1052 release_sock(sk); 1053 /* 1054 * Now socket is owned by kernel and we acquire BH lock 1055 * to finish close. No need to check for user refs. 1056 */ 1057 local_bh_disable(); 1058 bh_lock_sock(sk); 1059 WARN_ON(sock_owned_by_user(sk)); 1060 1061 this_cpu_inc(dccp_orphan_count); 1062 1063 /* Have we already been destroyed by a softirq or backlog? */ 1064 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED) 1065 goto out; 1066 1067 if (sk->sk_state == DCCP_CLOSED) 1068 inet_csk_destroy_sock(sk); 1069 1070 /* Otherwise, socket is reprieved until protocol close. */ 1071 1072 out: 1073 bh_unlock_sock(sk); 1074 local_bh_enable(); 1075 sock_put(sk); 1076 } 1077 1078 EXPORT_SYMBOL_GPL(dccp_close); 1079 1080 void dccp_shutdown(struct sock *sk, int how) 1081 { 1082 dccp_pr_debug("called shutdown(%x)\n", how); 1083 } 1084 1085 EXPORT_SYMBOL_GPL(dccp_shutdown); 1086 1087 static inline int __init dccp_mib_init(void) 1088 { 1089 dccp_statistics = alloc_percpu(struct dccp_mib); 1090 if (!dccp_statistics) 1091 return -ENOMEM; 1092 return 0; 1093 } 1094 1095 static inline void dccp_mib_exit(void) 1096 { 1097 free_percpu(dccp_statistics); 1098 } 1099 1100 static int thash_entries; 1101 module_param(thash_entries, int, 0444); 1102 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets"); 1103 1104 #ifdef CONFIG_IP_DCCP_DEBUG 1105 bool dccp_debug; 1106 module_param(dccp_debug, bool, 0644); 1107 MODULE_PARM_DESC(dccp_debug, "Enable debug messages"); 1108 1109 EXPORT_SYMBOL_GPL(dccp_debug); 1110 #endif 1111 1112 static int __init dccp_init(void) 1113 { 1114 unsigned long goal; 1115 unsigned long nr_pages = totalram_pages(); 1116 int ehash_order, bhash_order, i; 1117 int rc; 1118 1119 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) > 1120 sizeof_field(struct sk_buff, cb)); 1121 rc = inet_hashinfo2_init_mod(&dccp_hashinfo); 1122 if (rc) 1123 goto out_fail; 1124 rc = -ENOBUFS; 1125 dccp_hashinfo.bind_bucket_cachep = 1126 kmem_cache_create("dccp_bind_bucket", 1127 sizeof(struct inet_bind_bucket), 0, 1128 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL); 1129 if (!dccp_hashinfo.bind_bucket_cachep) 1130 goto out_free_hashinfo2; 1131 dccp_hashinfo.bind2_bucket_cachep = 1132 kmem_cache_create("dccp_bind2_bucket", 1133 sizeof(struct inet_bind2_bucket), 0, 1134 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL); 1135 if (!dccp_hashinfo.bind2_bucket_cachep) 1136 goto out_free_bind_bucket_cachep; 1137 1138 /* 1139 * Size and allocate the main established and bind bucket 1140 * hash tables. 1141 * 1142 * The methodology is similar to that of the buffer cache. 1143 */ 1144 if (nr_pages >= (128 * 1024)) 1145 goal = nr_pages >> (21 - PAGE_SHIFT); 1146 else 1147 goal = nr_pages >> (23 - PAGE_SHIFT); 1148 1149 if (thash_entries) 1150 goal = (thash_entries * 1151 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT; 1152 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++) 1153 ; 1154 do { 1155 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE / 1156 sizeof(struct inet_ehash_bucket); 1157 1158 while (hash_size & (hash_size - 1)) 1159 hash_size--; 1160 dccp_hashinfo.ehash_mask = hash_size - 1; 1161 dccp_hashinfo.ehash = (struct inet_ehash_bucket *) 1162 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order); 1163 } while (!dccp_hashinfo.ehash && --ehash_order > 0); 1164 1165 if (!dccp_hashinfo.ehash) { 1166 DCCP_CRIT("Failed to allocate DCCP established hash table"); 1167 goto out_free_bind2_bucket_cachep; 1168 } 1169 1170 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++) 1171 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i); 1172 1173 if (inet_ehash_locks_alloc(&dccp_hashinfo)) 1174 goto out_free_dccp_ehash; 1175 1176 bhash_order = ehash_order; 1177 1178 do { 1179 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE / 1180 sizeof(struct inet_bind_hashbucket); 1181 if ((dccp_hashinfo.bhash_size > (64 * 1024)) && 1182 bhash_order > 0) 1183 continue; 1184 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *) 1185 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order); 1186 } while (!dccp_hashinfo.bhash && --bhash_order >= 0); 1187 1188 if (!dccp_hashinfo.bhash) { 1189 DCCP_CRIT("Failed to allocate DCCP bind hash table"); 1190 goto out_free_dccp_locks; 1191 } 1192 1193 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *) 1194 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order); 1195 1196 if (!dccp_hashinfo.bhash2) { 1197 DCCP_CRIT("Failed to allocate DCCP bind2 hash table"); 1198 goto out_free_dccp_bhash; 1199 } 1200 1201 for (i = 0; i < dccp_hashinfo.bhash_size; i++) { 1202 spin_lock_init(&dccp_hashinfo.bhash[i].lock); 1203 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain); 1204 spin_lock_init(&dccp_hashinfo.bhash2[i].lock); 1205 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain); 1206 } 1207 1208 dccp_hashinfo.pernet = false; 1209 1210 rc = dccp_mib_init(); 1211 if (rc) 1212 goto out_free_dccp_bhash2; 1213 1214 rc = dccp_ackvec_init(); 1215 if (rc) 1216 goto out_free_dccp_mib; 1217 1218 rc = dccp_sysctl_init(); 1219 if (rc) 1220 goto out_ackvec_exit; 1221 1222 rc = ccid_initialize_builtins(); 1223 if (rc) 1224 goto out_sysctl_exit; 1225 1226 dccp_timestamping_init(); 1227 1228 return 0; 1229 1230 out_sysctl_exit: 1231 dccp_sysctl_exit(); 1232 out_ackvec_exit: 1233 dccp_ackvec_exit(); 1234 out_free_dccp_mib: 1235 dccp_mib_exit(); 1236 out_free_dccp_bhash2: 1237 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order); 1238 out_free_dccp_bhash: 1239 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order); 1240 out_free_dccp_locks: 1241 inet_ehash_locks_free(&dccp_hashinfo); 1242 out_free_dccp_ehash: 1243 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order); 1244 out_free_bind2_bucket_cachep: 1245 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep); 1246 out_free_bind_bucket_cachep: 1247 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); 1248 out_free_hashinfo2: 1249 inet_hashinfo2_free_mod(&dccp_hashinfo); 1250 out_fail: 1251 dccp_hashinfo.bhash = NULL; 1252 dccp_hashinfo.bhash2 = NULL; 1253 dccp_hashinfo.ehash = NULL; 1254 dccp_hashinfo.bind_bucket_cachep = NULL; 1255 dccp_hashinfo.bind2_bucket_cachep = NULL; 1256 return rc; 1257 } 1258 1259 static void __exit dccp_fini(void) 1260 { 1261 int bhash_order = get_order(dccp_hashinfo.bhash_size * 1262 sizeof(struct inet_bind_hashbucket)); 1263 1264 ccid_cleanup_builtins(); 1265 dccp_mib_exit(); 1266 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order); 1267 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order); 1268 free_pages((unsigned long)dccp_hashinfo.ehash, 1269 get_order((dccp_hashinfo.ehash_mask + 1) * 1270 sizeof(struct inet_ehash_bucket))); 1271 inet_ehash_locks_free(&dccp_hashinfo); 1272 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); 1273 dccp_ackvec_exit(); 1274 dccp_sysctl_exit(); 1275 inet_hashinfo2_free_mod(&dccp_hashinfo); 1276 } 1277 1278 module_init(dccp_init); 1279 module_exit(dccp_fini); 1280 1281 MODULE_LICENSE("GPL"); 1282 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>"); 1283 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol"); 1284