1 /* 2 * net/dccp/input.c 3 * 4 * An implementation of the DCCP protocol 5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br> 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License 9 * as published by the Free Software Foundation; either version 10 * 2 of the License, or (at your option) any later version. 11 */ 12 13 #include <linux/config.h> 14 #include <linux/dccp.h> 15 #include <linux/skbuff.h> 16 17 #include <net/sock.h> 18 19 #include "ccid.h" 20 #include "dccp.h" 21 22 static void dccp_fin(struct sock *sk, struct sk_buff *skb) 23 { 24 sk->sk_shutdown |= RCV_SHUTDOWN; 25 sock_set_flag(sk, SOCK_DONE); 26 __skb_pull(skb, dccp_hdr(skb)->dccph_doff * 4); 27 __skb_queue_tail(&sk->sk_receive_queue, skb); 28 skb_set_owner_r(skb, sk); 29 sk->sk_data_ready(sk, 0); 30 } 31 32 static void dccp_rcv_close(struct sock *sk, struct sk_buff *skb) 33 { 34 switch (sk->sk_state) { 35 case DCCP_PARTOPEN: 36 case DCCP_OPEN: 37 dccp_v4_send_reset(sk, DCCP_RESET_CODE_CLOSED); 38 dccp_fin(sk, skb); 39 dccp_set_state(sk, DCCP_CLOSED); 40 break; 41 } 42 } 43 44 static void dccp_rcv_closereq(struct sock *sk, struct sk_buff *skb) 45 { 46 /* 47 * Step 7: Check for unexpected packet types 48 * If (S.is_server and P.type == CloseReq) 49 * Send Sync packet acknowledging P.seqno 50 * Drop packet and return 51 */ 52 if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT) { 53 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC); 54 return; 55 } 56 57 switch (sk->sk_state) { 58 case DCCP_PARTOPEN: 59 case DCCP_OPEN: 60 dccp_set_state(sk, DCCP_CLOSING); 61 dccp_send_close(sk); 62 break; 63 } 64 } 65 66 static inline void dccp_event_ack_recv(struct sock *sk, struct sk_buff *skb) 67 { 68 struct dccp_sock *dp = dccp_sk(sk); 69 70 if (dp->dccps_options.dccpo_send_ack_vector) 71 dccp_ackpkts_check_rcv_ackno(dp->dccps_hc_rx_ackpkts, sk, 72 DCCP_SKB_CB(skb)->dccpd_ack_seq); 73 } 74 75 static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb) 76 { 77 const struct dccp_hdr *dh = dccp_hdr(skb); 78 struct dccp_sock *dp = dccp_sk(sk); 79 u64 lswl, lawl; 80 81 /* 82 * Step 5: Prepare sequence numbers for Sync 83 * If P.type == Sync or P.type == SyncAck, 84 * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL, 85 * / * P is valid, so update sequence number variables 86 * accordingly. After this update, P will pass the tests 87 * in Step 6. A SyncAck is generated if necessary in 88 * Step 15 * / 89 * Update S.GSR, S.SWL, S.SWH 90 * Otherwise, 91 * Drop packet and return 92 */ 93 if (dh->dccph_type == DCCP_PKT_SYNC || 94 dh->dccph_type == DCCP_PKT_SYNCACK) { 95 if (between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, 96 dp->dccps_awl, dp->dccps_awh) && 97 !before48(DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_swl)) 98 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq); 99 else 100 return -1; 101 } 102 103 /* 104 * Step 6: Check sequence numbers 105 * Let LSWL = S.SWL and LAWL = S.AWL 106 * If P.type == CloseReq or P.type == Close or P.type == Reset, 107 * LSWL := S.GSR + 1, LAWL := S.GAR 108 * If LSWL <= P.seqno <= S.SWH 109 * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH), 110 * Update S.GSR, S.SWL, S.SWH 111 * If P.type != Sync, 112 * Update S.GAR 113 * Otherwise, 114 * Send Sync packet acknowledging P.seqno 115 * Drop packet and return 116 */ 117 lswl = dp->dccps_swl; 118 lawl = dp->dccps_awl; 119 120 if (dh->dccph_type == DCCP_PKT_CLOSEREQ || 121 dh->dccph_type == DCCP_PKT_CLOSE || 122 dh->dccph_type == DCCP_PKT_RESET) { 123 lswl = dp->dccps_gsr; 124 dccp_inc_seqno(&lswl); 125 lawl = dp->dccps_gar; 126 } 127 128 if (between48(DCCP_SKB_CB(skb)->dccpd_seq, lswl, dp->dccps_swh) && 129 (DCCP_SKB_CB(skb)->dccpd_ack_seq == DCCP_PKT_WITHOUT_ACK_SEQ || 130 between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, 131 lawl, dp->dccps_awh))) { 132 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq); 133 134 if (dh->dccph_type != DCCP_PKT_SYNC && 135 (DCCP_SKB_CB(skb)->dccpd_ack_seq != 136 DCCP_PKT_WITHOUT_ACK_SEQ)) 137 dp->dccps_gar = DCCP_SKB_CB(skb)->dccpd_ack_seq; 138 } else { 139 LIMIT_NETDEBUG("Step 6 failed, sending SYNC...\n"); 140 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC); 141 return -1; 142 } 143 144 return 0; 145 } 146 147 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb, 148 const struct dccp_hdr *dh, const unsigned len) 149 { 150 struct dccp_sock *dp = dccp_sk(sk); 151 152 if (dccp_check_seqno(sk, skb)) 153 goto discard; 154 155 if (dccp_parse_options(sk, skb)) 156 goto discard; 157 158 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ) 159 dccp_event_ack_recv(sk, skb); 160 161 /* 162 * FIXME: check ECN to see if we should use 163 * DCCP_ACKPKTS_STATE_ECN_MARKED 164 */ 165 if (dp->dccps_options.dccpo_send_ack_vector) { 166 struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts; 167 168 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts, 169 DCCP_SKB_CB(skb)->dccpd_seq, 170 DCCP_ACKPKTS_STATE_RECEIVED)) { 171 LIMIT_NETDEBUG(KERN_INFO "DCCP: acknowledgeable " 172 "packets buffer full!\n"); 173 ap->dccpap_ack_seqno = DCCP_MAX_SEQNO + 1; 174 inet_csk_schedule_ack(sk); 175 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 176 TCP_DELACK_MIN, 177 DCCP_RTO_MAX); 178 goto discard; 179 } 180 181 /* 182 * FIXME: this activation is probably wrong, have to study more 183 * TCP delack machinery and how it fits into DCCP draft, but 184 * for now it kinda "works" 8) 185 */ 186 if (!inet_csk_ack_scheduled(sk)) { 187 inet_csk_schedule_ack(sk); 188 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 5 * HZ, 189 DCCP_RTO_MAX); 190 } 191 } 192 193 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb); 194 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb); 195 196 switch (dccp_hdr(skb)->dccph_type) { 197 case DCCP_PKT_DATAACK: 198 case DCCP_PKT_DATA: 199 /* 200 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED 201 * option if it is. 202 */ 203 __skb_pull(skb, dh->dccph_doff * 4); 204 __skb_queue_tail(&sk->sk_receive_queue, skb); 205 skb_set_owner_r(skb, sk); 206 sk->sk_data_ready(sk, 0); 207 return 0; 208 case DCCP_PKT_ACK: 209 goto discard; 210 case DCCP_PKT_RESET: 211 /* 212 * Step 9: Process Reset 213 * If P.type == Reset, 214 * Tear down connection 215 * S.state := TIMEWAIT 216 * Set TIMEWAIT timer 217 * Drop packet and return 218 */ 219 dccp_fin(sk, skb); 220 dccp_time_wait(sk, DCCP_TIME_WAIT, 0); 221 return 0; 222 case DCCP_PKT_CLOSEREQ: 223 dccp_rcv_closereq(sk, skb); 224 goto discard; 225 case DCCP_PKT_CLOSE: 226 dccp_rcv_close(sk, skb); 227 return 0; 228 case DCCP_PKT_REQUEST: 229 /* Step 7 230 * or (S.is_server and P.type == Response) 231 * or (S.is_client and P.type == Request) 232 * or (S.state >= OPEN and P.type == Request 233 * and P.seqno >= S.OSR) 234 * or (S.state >= OPEN and P.type == Response 235 * and P.seqno >= S.OSR) 236 * or (S.state == RESPOND and P.type == Data), 237 * Send Sync packet acknowledging P.seqno 238 * Drop packet and return 239 */ 240 if (dp->dccps_role != DCCP_ROLE_LISTEN) 241 goto send_sync; 242 goto check_seq; 243 case DCCP_PKT_RESPONSE: 244 if (dp->dccps_role != DCCP_ROLE_CLIENT) 245 goto send_sync; 246 check_seq: 247 if (!before48(DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_osr)) { 248 send_sync: 249 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, 250 DCCP_PKT_SYNC); 251 } 252 break; 253 case DCCP_PKT_SYNC: 254 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, 255 DCCP_PKT_SYNCACK); 256 /* 257 * From the draft: 258 * 259 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets 260 * MAY have non-zero-length application data areas, whose 261 * contents * receivers MUST ignore. 262 */ 263 goto discard; 264 } 265 266 DCCP_INC_STATS_BH(DCCP_MIB_INERRS); 267 discard: 268 __kfree_skb(skb); 269 return 0; 270 } 271 272 static int dccp_rcv_request_sent_state_process(struct sock *sk, 273 struct sk_buff *skb, 274 const struct dccp_hdr *dh, 275 const unsigned len) 276 { 277 /* 278 * Step 4: Prepare sequence numbers in REQUEST 279 * If S.state == REQUEST, 280 * If (P.type == Response or P.type == Reset) 281 * and S.AWL <= P.ackno <= S.AWH, 282 * / * Set sequence number variables corresponding to the 283 * other endpoint, so P will pass the tests in Step 6 * / 284 * Set S.GSR, S.ISR, S.SWL, S.SWH 285 * / * Response processing continues in Step 10; Reset 286 * processing continues in Step 9 * / 287 */ 288 if (dh->dccph_type == DCCP_PKT_RESPONSE) { 289 const struct inet_connection_sock *icsk = inet_csk(sk); 290 struct dccp_sock *dp = dccp_sk(sk); 291 292 /* Stop the REQUEST timer */ 293 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); 294 BUG_TRAP(sk->sk_send_head != NULL); 295 __kfree_skb(sk->sk_send_head); 296 sk->sk_send_head = NULL; 297 298 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, 299 dp->dccps_awl, dp->dccps_awh)) { 300 dccp_pr_debug("invalid ackno: S.AWL=%llu, " 301 "P.ackno=%llu, S.AWH=%llu \n", 302 (unsigned long long)dp->dccps_awl, 303 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq, 304 (unsigned long long)dp->dccps_awh); 305 goto out_invalid_packet; 306 } 307 308 dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq; 309 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq); 310 311 if (ccid_hc_rx_init(dp->dccps_hc_rx_ccid, sk) != 0 || 312 ccid_hc_tx_init(dp->dccps_hc_tx_ccid, sk) != 0) { 313 ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk); 314 ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk); 315 /* FIXME: send appropriate RESET code */ 316 goto out_invalid_packet; 317 } 318 319 dccp_sync_mss(sk, dp->dccps_pmtu_cookie); 320 321 /* 322 * Step 10: Process REQUEST state (second part) 323 * If S.state == REQUEST, 324 * / * If we get here, P is a valid Response from the 325 * server (see Step 4), and we should move to 326 * PARTOPEN state. PARTOPEN means send an Ack, 327 * don't send Data packets, retransmit Acks 328 * periodically, and always include any Init Cookie 329 * from the Response * / 330 * S.state := PARTOPEN 331 * Set PARTOPEN timer 332 * Continue with S.state == PARTOPEN 333 * / * Step 12 will send the Ack completing the 334 * three-way handshake * / 335 */ 336 dccp_set_state(sk, DCCP_PARTOPEN); 337 338 /* Make sure socket is routed, for correct metrics. */ 339 inet_sk_rebuild_header(sk); 340 341 if (!sock_flag(sk, SOCK_DEAD)) { 342 sk->sk_state_change(sk); 343 sk_wake_async(sk, 0, POLL_OUT); 344 } 345 346 if (sk->sk_write_pending || icsk->icsk_ack.pingpong || 347 icsk->icsk_accept_queue.rskq_defer_accept) { 348 /* Save one ACK. Data will be ready after 349 * several ticks, if write_pending is set. 350 * 351 * It may be deleted, but with this feature tcpdumps 352 * look so _wonderfully_ clever, that I was not able 353 * to stand against the temptation 8) --ANK 354 */ 355 /* 356 * OK, in DCCP we can as well do a similar trick, its 357 * even in the draft, but there is no need for us to 358 * schedule an ack here, as dccp_sendmsg does this for 359 * us, also stated in the draft. -acme 360 */ 361 __kfree_skb(skb); 362 return 0; 363 } 364 dccp_send_ack(sk); 365 return -1; 366 } 367 368 out_invalid_packet: 369 return 1; /* dccp_v4_do_rcv will send a reset, but... 370 FIXME: the reset code should be 371 DCCP_RESET_CODE_PACKET_ERROR */ 372 } 373 374 static int dccp_rcv_respond_partopen_state_process(struct sock *sk, 375 struct sk_buff *skb, 376 const struct dccp_hdr *dh, 377 const unsigned len) 378 { 379 int queued = 0; 380 381 switch (dh->dccph_type) { 382 case DCCP_PKT_RESET: 383 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 384 break; 385 case DCCP_PKT_DATAACK: 386 case DCCP_PKT_ACK: 387 /* 388 * FIXME: we should be reseting the PARTOPEN (DELACK) timer 389 * here but only if we haven't used the DELACK timer for 390 * something else, like sending a delayed ack for a TIMESTAMP 391 * echo, etc, for now were not clearing it, sending an extra 392 * ACK when there is nothing else to do in DELACK is not a big 393 * deal after all. 394 */ 395 396 /* Stop the PARTOPEN timer */ 397 if (sk->sk_state == DCCP_PARTOPEN) 398 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 399 400 dccp_sk(sk)->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq; 401 dccp_set_state(sk, DCCP_OPEN); 402 403 if (dh->dccph_type == DCCP_PKT_DATAACK) { 404 dccp_rcv_established(sk, skb, dh, len); 405 queued = 1; /* packet was queued 406 (by dccp_rcv_established) */ 407 } 408 break; 409 } 410 411 return queued; 412 } 413 414 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 415 struct dccp_hdr *dh, unsigned len) 416 { 417 struct dccp_sock *dp = dccp_sk(sk); 418 const int old_state = sk->sk_state; 419 int queued = 0; 420 421 /* 422 * Step 3: Process LISTEN state 423 * (Continuing from dccp_v4_do_rcv and dccp_v6_do_rcv) 424 * 425 * If S.state == LISTEN, 426 * If P.type == Request or P contains a valid Init Cookie 427 * option, 428 * * Must scan the packet's options to check for an Init 429 * Cookie. Only the Init Cookie is processed here, 430 * however; other options are processed in Step 8. This 431 * scan need only be performed if the endpoint uses Init 432 * Cookies * 433 * * Generate a new socket and switch to that socket * 434 * Set S := new socket for this port pair 435 * S.state = RESPOND 436 * Choose S.ISS (initial seqno) or set from Init Cookie 437 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie 438 * Continue with S.state == RESPOND 439 * * A Response packet will be generated in Step 11 * 440 * Otherwise, 441 * Generate Reset(No Connection) unless P.type == Reset 442 * Drop packet and return 443 * 444 * NOTE: the check for the packet types is done in 445 * dccp_rcv_state_process 446 */ 447 if (sk->sk_state == DCCP_LISTEN) { 448 if (dh->dccph_type == DCCP_PKT_REQUEST) { 449 if (dccp_v4_conn_request(sk, skb) < 0) 450 return 1; 451 452 /* FIXME: do congestion control initialization */ 453 goto discard; 454 } 455 if (dh->dccph_type == DCCP_PKT_RESET) 456 goto discard; 457 458 /* Caller (dccp_v4_do_rcv) will send Reset(No Connection)*/ 459 return 1; 460 } 461 462 if (sk->sk_state != DCCP_REQUESTING) { 463 if (dccp_check_seqno(sk, skb)) 464 goto discard; 465 466 /* 467 * Step 8: Process options and mark acknowledgeable 468 */ 469 if (dccp_parse_options(sk, skb)) 470 goto discard; 471 472 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != 473 DCCP_PKT_WITHOUT_ACK_SEQ) 474 dccp_event_ack_recv(sk, skb); 475 476 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb); 477 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb); 478 479 /* 480 * FIXME: check ECN to see if we should use 481 * DCCP_ACKPKTS_STATE_ECN_MARKED 482 */ 483 if (dp->dccps_options.dccpo_send_ack_vector) { 484 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts, 485 DCCP_SKB_CB(skb)->dccpd_seq, 486 DCCP_ACKPKTS_STATE_RECEIVED)) 487 goto discard; 488 /* 489 * FIXME: this activation is probably wrong, have to 490 * study more TCP delack machinery and how it fits into 491 * DCCP draft, but for now it kinda "works" 8) 492 */ 493 if ((dp->dccps_hc_rx_ackpkts->dccpap_ack_seqno == 494 DCCP_MAX_SEQNO + 1) && 495 !inet_csk_ack_scheduled(sk)) { 496 inet_csk_schedule_ack(sk); 497 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 498 TCP_DELACK_MIN, 499 DCCP_RTO_MAX); 500 } 501 } 502 } 503 504 /* 505 * Step 9: Process Reset 506 * If P.type == Reset, 507 * Tear down connection 508 * S.state := TIMEWAIT 509 * Set TIMEWAIT timer 510 * Drop packet and return 511 */ 512 if (dh->dccph_type == DCCP_PKT_RESET) { 513 /* 514 * Queue the equivalent of TCP fin so that dccp_recvmsg 515 * exits the loop 516 */ 517 dccp_fin(sk, skb); 518 dccp_time_wait(sk, DCCP_TIME_WAIT, 0); 519 return 0; 520 /* 521 * Step 7: Check for unexpected packet types 522 * If (S.is_server and P.type == CloseReq) 523 * or (S.is_server and P.type == Response) 524 * or (S.is_client and P.type == Request) 525 * or (S.state == RESPOND and P.type == Data), 526 * Send Sync packet acknowledging P.seqno 527 * Drop packet and return 528 */ 529 } else if ((dp->dccps_role != DCCP_ROLE_CLIENT && 530 (dh->dccph_type == DCCP_PKT_RESPONSE || 531 dh->dccph_type == DCCP_PKT_CLOSEREQ)) || 532 (dp->dccps_role == DCCP_ROLE_CLIENT && 533 dh->dccph_type == DCCP_PKT_REQUEST) || 534 (sk->sk_state == DCCP_RESPOND && 535 dh->dccph_type == DCCP_PKT_DATA)) { 536 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, 537 DCCP_PKT_SYNC); 538 goto discard; 539 } 540 541 switch (sk->sk_state) { 542 case DCCP_CLOSED: 543 return 1; 544 545 case DCCP_REQUESTING: 546 /* FIXME: do congestion control initialization */ 547 548 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len); 549 if (queued >= 0) 550 return queued; 551 552 __kfree_skb(skb); 553 return 0; 554 555 case DCCP_RESPOND: 556 case DCCP_PARTOPEN: 557 queued = dccp_rcv_respond_partopen_state_process(sk, skb, 558 dh, len); 559 break; 560 } 561 562 if (dh->dccph_type == DCCP_PKT_ACK || 563 dh->dccph_type == DCCP_PKT_DATAACK) { 564 switch (old_state) { 565 case DCCP_PARTOPEN: 566 sk->sk_state_change(sk); 567 sk_wake_async(sk, 0, POLL_OUT); 568 break; 569 } 570 } 571 572 if (!queued) { 573 discard: 574 __kfree_skb(skb); 575 } 576 return 0; 577 } 578