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(KERN_WARNING "DCCP: Step 6 failed for %s packet, " 140 "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and " 141 "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), " 142 "sending SYNC...\n", 143 dccp_packet_name(dh->dccph_type), 144 lswl, DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_swh, 145 (DCCP_SKB_CB(skb)->dccpd_ack_seq == 146 DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist" : "exists", 147 lawl, DCCP_SKB_CB(skb)->dccpd_ack_seq, dp->dccps_awh); 148 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC); 149 return -1; 150 } 151 152 return 0; 153 } 154 155 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb, 156 const struct dccp_hdr *dh, const unsigned len) 157 { 158 struct dccp_sock *dp = dccp_sk(sk); 159 160 if (dccp_check_seqno(sk, skb)) 161 goto discard; 162 163 if (dccp_parse_options(sk, skb)) 164 goto discard; 165 166 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ) 167 dccp_event_ack_recv(sk, skb); 168 169 /* 170 * FIXME: check ECN to see if we should use 171 * DCCP_ACKPKTS_STATE_ECN_MARKED 172 */ 173 if (dp->dccps_options.dccpo_send_ack_vector) { 174 struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts; 175 176 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts, 177 DCCP_SKB_CB(skb)->dccpd_seq, 178 DCCP_ACKPKTS_STATE_RECEIVED)) { 179 LIMIT_NETDEBUG(KERN_WARNING "DCCP: acknowledgeable " 180 "packets buffer full!\n"); 181 ap->dccpap_ack_seqno = DCCP_MAX_SEQNO + 1; 182 inet_csk_schedule_ack(sk); 183 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 184 TCP_DELACK_MIN, 185 DCCP_RTO_MAX); 186 goto discard; 187 } 188 189 /* 190 * FIXME: this activation is probably wrong, have to study more 191 * TCP delack machinery and how it fits into DCCP draft, but 192 * for now it kinda "works" 8) 193 */ 194 if (!inet_csk_ack_scheduled(sk)) { 195 inet_csk_schedule_ack(sk); 196 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 5 * HZ, 197 DCCP_RTO_MAX); 198 } 199 } 200 201 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb); 202 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb); 203 204 switch (dccp_hdr(skb)->dccph_type) { 205 case DCCP_PKT_DATAACK: 206 case DCCP_PKT_DATA: 207 /* 208 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED 209 * option if it is. 210 */ 211 __skb_pull(skb, dh->dccph_doff * 4); 212 __skb_queue_tail(&sk->sk_receive_queue, skb); 213 skb_set_owner_r(skb, sk); 214 sk->sk_data_ready(sk, 0); 215 return 0; 216 case DCCP_PKT_ACK: 217 goto discard; 218 case DCCP_PKT_RESET: 219 /* 220 * Step 9: Process Reset 221 * If P.type == Reset, 222 * Tear down connection 223 * S.state := TIMEWAIT 224 * Set TIMEWAIT timer 225 * Drop packet and return 226 */ 227 dccp_fin(sk, skb); 228 dccp_time_wait(sk, DCCP_TIME_WAIT, 0); 229 return 0; 230 case DCCP_PKT_CLOSEREQ: 231 dccp_rcv_closereq(sk, skb); 232 goto discard; 233 case DCCP_PKT_CLOSE: 234 dccp_rcv_close(sk, skb); 235 return 0; 236 case DCCP_PKT_REQUEST: 237 /* Step 7 238 * or (S.is_server and P.type == Response) 239 * or (S.is_client and P.type == Request) 240 * or (S.state >= OPEN and P.type == Request 241 * and P.seqno >= S.OSR) 242 * or (S.state >= OPEN and P.type == Response 243 * and P.seqno >= S.OSR) 244 * or (S.state == RESPOND and P.type == Data), 245 * Send Sync packet acknowledging P.seqno 246 * Drop packet and return 247 */ 248 if (dp->dccps_role != DCCP_ROLE_LISTEN) 249 goto send_sync; 250 goto check_seq; 251 case DCCP_PKT_RESPONSE: 252 if (dp->dccps_role != DCCP_ROLE_CLIENT) 253 goto send_sync; 254 check_seq: 255 if (!before48(DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_osr)) { 256 send_sync: 257 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, 258 DCCP_PKT_SYNC); 259 } 260 break; 261 case DCCP_PKT_SYNC: 262 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, 263 DCCP_PKT_SYNCACK); 264 /* 265 * From the draft: 266 * 267 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets 268 * MAY have non-zero-length application data areas, whose 269 * contents * receivers MUST ignore. 270 */ 271 goto discard; 272 } 273 274 DCCP_INC_STATS_BH(DCCP_MIB_INERRS); 275 discard: 276 __kfree_skb(skb); 277 return 0; 278 } 279 280 static int dccp_rcv_request_sent_state_process(struct sock *sk, 281 struct sk_buff *skb, 282 const struct dccp_hdr *dh, 283 const unsigned len) 284 { 285 /* 286 * Step 4: Prepare sequence numbers in REQUEST 287 * If S.state == REQUEST, 288 * If (P.type == Response or P.type == Reset) 289 * and S.AWL <= P.ackno <= S.AWH, 290 * / * Set sequence number variables corresponding to the 291 * other endpoint, so P will pass the tests in Step 6 * / 292 * Set S.GSR, S.ISR, S.SWL, S.SWH 293 * / * Response processing continues in Step 10; Reset 294 * processing continues in Step 9 * / 295 */ 296 if (dh->dccph_type == DCCP_PKT_RESPONSE) { 297 const struct inet_connection_sock *icsk = inet_csk(sk); 298 struct dccp_sock *dp = dccp_sk(sk); 299 300 /* Stop the REQUEST timer */ 301 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); 302 BUG_TRAP(sk->sk_send_head != NULL); 303 __kfree_skb(sk->sk_send_head); 304 sk->sk_send_head = NULL; 305 306 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, 307 dp->dccps_awl, dp->dccps_awh)) { 308 dccp_pr_debug("invalid ackno: S.AWL=%llu, " 309 "P.ackno=%llu, S.AWH=%llu \n", 310 (unsigned long long)dp->dccps_awl, 311 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq, 312 (unsigned long long)dp->dccps_awh); 313 goto out_invalid_packet; 314 } 315 316 dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq; 317 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq); 318 319 if (ccid_hc_rx_init(dp->dccps_hc_rx_ccid, sk) != 0 || 320 ccid_hc_tx_init(dp->dccps_hc_tx_ccid, sk) != 0) { 321 ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk); 322 ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk); 323 /* FIXME: send appropriate RESET code */ 324 goto out_invalid_packet; 325 } 326 327 dccp_sync_mss(sk, dp->dccps_pmtu_cookie); 328 329 /* 330 * Step 10: Process REQUEST state (second part) 331 * If S.state == REQUEST, 332 * / * If we get here, P is a valid Response from the 333 * server (see Step 4), and we should move to 334 * PARTOPEN state. PARTOPEN means send an Ack, 335 * don't send Data packets, retransmit Acks 336 * periodically, and always include any Init Cookie 337 * from the Response * / 338 * S.state := PARTOPEN 339 * Set PARTOPEN timer 340 * Continue with S.state == PARTOPEN 341 * / * Step 12 will send the Ack completing the 342 * three-way handshake * / 343 */ 344 dccp_set_state(sk, DCCP_PARTOPEN); 345 346 /* Make sure socket is routed, for correct metrics. */ 347 inet_sk_rebuild_header(sk); 348 349 if (!sock_flag(sk, SOCK_DEAD)) { 350 sk->sk_state_change(sk); 351 sk_wake_async(sk, 0, POLL_OUT); 352 } 353 354 if (sk->sk_write_pending || icsk->icsk_ack.pingpong || 355 icsk->icsk_accept_queue.rskq_defer_accept) { 356 /* Save one ACK. Data will be ready after 357 * several ticks, if write_pending is set. 358 * 359 * It may be deleted, but with this feature tcpdumps 360 * look so _wonderfully_ clever, that I was not able 361 * to stand against the temptation 8) --ANK 362 */ 363 /* 364 * OK, in DCCP we can as well do a similar trick, its 365 * even in the draft, but there is no need for us to 366 * schedule an ack here, as dccp_sendmsg does this for 367 * us, also stated in the draft. -acme 368 */ 369 __kfree_skb(skb); 370 return 0; 371 } 372 dccp_send_ack(sk); 373 return -1; 374 } 375 376 out_invalid_packet: 377 return 1; /* dccp_v4_do_rcv will send a reset, but... 378 FIXME: the reset code should be 379 DCCP_RESET_CODE_PACKET_ERROR */ 380 } 381 382 static int dccp_rcv_respond_partopen_state_process(struct sock *sk, 383 struct sk_buff *skb, 384 const struct dccp_hdr *dh, 385 const unsigned len) 386 { 387 int queued = 0; 388 389 switch (dh->dccph_type) { 390 case DCCP_PKT_RESET: 391 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 392 break; 393 case DCCP_PKT_DATAACK: 394 case DCCP_PKT_ACK: 395 /* 396 * FIXME: we should be reseting the PARTOPEN (DELACK) timer 397 * here but only if we haven't used the DELACK timer for 398 * something else, like sending a delayed ack for a TIMESTAMP 399 * echo, etc, for now were not clearing it, sending an extra 400 * ACK when there is nothing else to do in DELACK is not a big 401 * deal after all. 402 */ 403 404 /* Stop the PARTOPEN timer */ 405 if (sk->sk_state == DCCP_PARTOPEN) 406 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 407 408 dccp_sk(sk)->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq; 409 dccp_set_state(sk, DCCP_OPEN); 410 411 if (dh->dccph_type == DCCP_PKT_DATAACK) { 412 dccp_rcv_established(sk, skb, dh, len); 413 queued = 1; /* packet was queued 414 (by dccp_rcv_established) */ 415 } 416 break; 417 } 418 419 return queued; 420 } 421 422 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 423 struct dccp_hdr *dh, unsigned len) 424 { 425 struct dccp_sock *dp = dccp_sk(sk); 426 const int old_state = sk->sk_state; 427 int queued = 0; 428 429 /* 430 * Step 3: Process LISTEN state 431 * (Continuing from dccp_v4_do_rcv and dccp_v6_do_rcv) 432 * 433 * If S.state == LISTEN, 434 * If P.type == Request or P contains a valid Init Cookie 435 * option, 436 * * Must scan the packet's options to check for an Init 437 * Cookie. Only the Init Cookie is processed here, 438 * however; other options are processed in Step 8. This 439 * scan need only be performed if the endpoint uses Init 440 * Cookies * 441 * * Generate a new socket and switch to that socket * 442 * Set S := new socket for this port pair 443 * S.state = RESPOND 444 * Choose S.ISS (initial seqno) or set from Init Cookie 445 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie 446 * Continue with S.state == RESPOND 447 * * A Response packet will be generated in Step 11 * 448 * Otherwise, 449 * Generate Reset(No Connection) unless P.type == Reset 450 * Drop packet and return 451 * 452 * NOTE: the check for the packet types is done in 453 * dccp_rcv_state_process 454 */ 455 if (sk->sk_state == DCCP_LISTEN) { 456 if (dh->dccph_type == DCCP_PKT_REQUEST) { 457 if (dccp_v4_conn_request(sk, skb) < 0) 458 return 1; 459 460 /* FIXME: do congestion control initialization */ 461 goto discard; 462 } 463 if (dh->dccph_type == DCCP_PKT_RESET) 464 goto discard; 465 466 /* Caller (dccp_v4_do_rcv) will send Reset(No Connection)*/ 467 return 1; 468 } 469 470 if (sk->sk_state != DCCP_REQUESTING) { 471 if (dccp_check_seqno(sk, skb)) 472 goto discard; 473 474 /* 475 * Step 8: Process options and mark acknowledgeable 476 */ 477 if (dccp_parse_options(sk, skb)) 478 goto discard; 479 480 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != 481 DCCP_PKT_WITHOUT_ACK_SEQ) 482 dccp_event_ack_recv(sk, skb); 483 484 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb); 485 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb); 486 487 /* 488 * FIXME: check ECN to see if we should use 489 * DCCP_ACKPKTS_STATE_ECN_MARKED 490 */ 491 if (dp->dccps_options.dccpo_send_ack_vector) { 492 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts, 493 DCCP_SKB_CB(skb)->dccpd_seq, 494 DCCP_ACKPKTS_STATE_RECEIVED)) 495 goto discard; 496 /* 497 * FIXME: this activation is probably wrong, have to 498 * study more TCP delack machinery and how it fits into 499 * DCCP draft, but for now it kinda "works" 8) 500 */ 501 if ((dp->dccps_hc_rx_ackpkts->dccpap_ack_seqno == 502 DCCP_MAX_SEQNO + 1) && 503 !inet_csk_ack_scheduled(sk)) { 504 inet_csk_schedule_ack(sk); 505 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 506 TCP_DELACK_MIN, 507 DCCP_RTO_MAX); 508 } 509 } 510 } 511 512 /* 513 * Step 9: Process Reset 514 * If P.type == Reset, 515 * Tear down connection 516 * S.state := TIMEWAIT 517 * Set TIMEWAIT timer 518 * Drop packet and return 519 */ 520 if (dh->dccph_type == DCCP_PKT_RESET) { 521 /* 522 * Queue the equivalent of TCP fin so that dccp_recvmsg 523 * exits the loop 524 */ 525 dccp_fin(sk, skb); 526 dccp_time_wait(sk, DCCP_TIME_WAIT, 0); 527 return 0; 528 /* 529 * Step 7: Check for unexpected packet types 530 * If (S.is_server and P.type == CloseReq) 531 * or (S.is_server and P.type == Response) 532 * or (S.is_client and P.type == Request) 533 * or (S.state == RESPOND and P.type == Data), 534 * Send Sync packet acknowledging P.seqno 535 * Drop packet and return 536 */ 537 } else if ((dp->dccps_role != DCCP_ROLE_CLIENT && 538 (dh->dccph_type == DCCP_PKT_RESPONSE || 539 dh->dccph_type == DCCP_PKT_CLOSEREQ)) || 540 (dp->dccps_role == DCCP_ROLE_CLIENT && 541 dh->dccph_type == DCCP_PKT_REQUEST) || 542 (sk->sk_state == DCCP_RESPOND && 543 dh->dccph_type == DCCP_PKT_DATA)) { 544 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, 545 DCCP_PKT_SYNC); 546 goto discard; 547 } 548 549 switch (sk->sk_state) { 550 case DCCP_CLOSED: 551 return 1; 552 553 case DCCP_REQUESTING: 554 /* FIXME: do congestion control initialization */ 555 556 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len); 557 if (queued >= 0) 558 return queued; 559 560 __kfree_skb(skb); 561 return 0; 562 563 case DCCP_RESPOND: 564 case DCCP_PARTOPEN: 565 queued = dccp_rcv_respond_partopen_state_process(sk, skb, 566 dh, len); 567 break; 568 } 569 570 if (dh->dccph_type == DCCP_PKT_ACK || 571 dh->dccph_type == DCCP_PKT_DATAACK) { 572 switch (old_state) { 573 case DCCP_PARTOPEN: 574 sk->sk_state_change(sk); 575 sk_wake_async(sk, 0, POLL_OUT); 576 break; 577 } 578 } 579 580 if (!queued) { 581 discard: 582 __kfree_skb(skb); 583 } 584 return 0; 585 } 586