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); 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 = dp->dccps_swl; 80 u64 lawl = dp->dccps_awl; 81 82 /* 83 * Step 5: Prepare sequence numbers for Sync 84 * If P.type == Sync or P.type == SyncAck, 85 * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL, 86 * / * P is valid, so update sequence number variables 87 * accordingly. After this update, P will pass the tests 88 * in Step 6. A SyncAck is generated if necessary in 89 * Step 15 * / 90 * Update S.GSR, S.SWL, S.SWH 91 * Otherwise, 92 * Drop packet and return 93 */ 94 if (dh->dccph_type == DCCP_PKT_SYNC || 95 dh->dccph_type == DCCP_PKT_SYNCACK) { 96 if (between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, 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 * Step 6: Check sequence numbers 103 * Let LSWL = S.SWL and LAWL = S.AWL 104 * If P.type == CloseReq or P.type == Close or P.type == Reset, 105 * LSWL := S.GSR + 1, LAWL := S.GAR 106 * If LSWL <= P.seqno <= S.SWH 107 * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH), 108 * Update S.GSR, S.SWL, S.SWH 109 * If P.type != Sync, 110 * Update S.GAR 111 * Otherwise, 112 * Send Sync packet acknowledging P.seqno 113 * Drop packet and return 114 */ 115 } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ || 116 dh->dccph_type == DCCP_PKT_CLOSE || 117 dh->dccph_type == DCCP_PKT_RESET) { 118 lswl = dp->dccps_gsr; 119 dccp_inc_seqno(&lswl); 120 lawl = dp->dccps_gar; 121 } 122 123 if (between48(DCCP_SKB_CB(skb)->dccpd_seq, lswl, dp->dccps_swh) && 124 (DCCP_SKB_CB(skb)->dccpd_ack_seq == DCCP_PKT_WITHOUT_ACK_SEQ || 125 between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, lawl, dp->dccps_awh))) { 126 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq); 127 128 if (dh->dccph_type != DCCP_PKT_SYNC && 129 DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ) 130 dp->dccps_gar = DCCP_SKB_CB(skb)->dccpd_ack_seq; 131 } else { 132 dccp_pr_debug("Step 6 failed, sending SYNC...\n"); 133 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq); 134 return -1; 135 } 136 137 return 0; 138 } 139 140 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb, 141 const struct dccp_hdr *dh, const unsigned len) 142 { 143 struct dccp_sock *dp = dccp_sk(sk); 144 145 if (dccp_check_seqno(sk, skb)) 146 goto discard; 147 148 if (dccp_parse_options(sk, skb)) 149 goto discard; 150 151 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ) 152 dccp_event_ack_recv(sk, skb); 153 154 /* 155 * FIXME: check ECN to see if we should use 156 * DCCP_ACKPKTS_STATE_ECN_MARKED 157 */ 158 if (dp->dccps_options.dccpo_send_ack_vector) { 159 struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts; 160 161 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts, 162 DCCP_SKB_CB(skb)->dccpd_seq, 163 DCCP_ACKPKTS_STATE_RECEIVED)) { 164 LIMIT_NETDEBUG(pr_info("DCCP: acknowledgeable packets buffer full!\n")); 165 ap->dccpap_ack_seqno = DCCP_MAX_SEQNO + 1; 166 inet_csk_schedule_ack(sk); 167 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, TCP_DELACK_MIN, TCP_RTO_MAX); 168 goto discard; 169 } 170 171 /* 172 * FIXME: this activation is probably wrong, have to study more 173 * TCP delack machinery and how it fits into DCCP draft, but 174 * for now it kinda "works" 8) 175 */ 176 if (!inet_csk_ack_scheduled(sk)) { 177 inet_csk_schedule_ack(sk); 178 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 5 * HZ, TCP_RTO_MAX); 179 } 180 } 181 182 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb); 183 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb); 184 185 switch (dccp_hdr(skb)->dccph_type) { 186 case DCCP_PKT_DATAACK: 187 case DCCP_PKT_DATA: 188 /* 189 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED option 190 * if it is. 191 */ 192 __skb_pull(skb, dh->dccph_doff * 4); 193 __skb_queue_tail(&sk->sk_receive_queue, skb); 194 skb_set_owner_r(skb, sk); 195 sk->sk_data_ready(sk, 0); 196 return 0; 197 case DCCP_PKT_ACK: 198 goto discard; 199 case DCCP_PKT_RESET: 200 /* 201 * Step 9: Process Reset 202 * If P.type == Reset, 203 * Tear down connection 204 * S.state := TIMEWAIT 205 * Set TIMEWAIT timer 206 * Drop packet and return 207 */ 208 dccp_fin(sk, skb); 209 dccp_time_wait(sk, DCCP_TIME_WAIT, 0); 210 return 0; 211 case DCCP_PKT_CLOSEREQ: 212 dccp_rcv_closereq(sk, skb); 213 goto discard; 214 case DCCP_PKT_CLOSE: 215 dccp_rcv_close(sk, skb); 216 return 0; 217 case DCCP_PKT_REQUEST: 218 /* Step 7 219 * or (S.is_server and P.type == Response) 220 * or (S.is_client and P.type == Request) 221 * or (S.state >= OPEN and P.type == Request 222 * and P.seqno >= S.OSR) 223 * or (S.state >= OPEN and P.type == Response 224 * and P.seqno >= S.OSR) 225 * or (S.state == RESPOND and P.type == Data), 226 * Send Sync packet acknowledging P.seqno 227 * Drop packet and return 228 */ 229 if (dp->dccps_role != DCCP_ROLE_LISTEN) 230 goto send_sync; 231 goto check_seq; 232 case DCCP_PKT_RESPONSE: 233 if (dp->dccps_role != DCCP_ROLE_CLIENT) 234 goto send_sync; 235 check_seq: 236 if (!before48(DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_osr)) { 237 send_sync: 238 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq); 239 } 240 break; 241 } 242 243 DCCP_INC_STATS_BH(DCCP_MIB_INERRS); 244 discard: 245 __kfree_skb(skb); 246 return 0; 247 } 248 249 static int dccp_rcv_request_sent_state_process(struct sock *sk, 250 struct sk_buff *skb, 251 const struct dccp_hdr *dh, 252 const unsigned len) 253 { 254 /* 255 * Step 4: Prepare sequence numbers in REQUEST 256 * If S.state == REQUEST, 257 * If (P.type == Response or P.type == Reset) 258 * and S.AWL <= P.ackno <= S.AWH, 259 * / * Set sequence number variables corresponding to the 260 * other endpoint, so P will pass the tests in Step 6 * / 261 * Set S.GSR, S.ISR, S.SWL, S.SWH 262 * / * Response processing continues in Step 10; Reset 263 * processing continues in Step 9 * / 264 */ 265 if (dh->dccph_type == DCCP_PKT_RESPONSE) { 266 const struct inet_connection_sock *icsk = inet_csk(sk); 267 struct dccp_sock *dp = dccp_sk(sk); 268 269 /* Stop the REQUEST timer */ 270 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); 271 BUG_TRAP(sk->sk_send_head != NULL); 272 __kfree_skb(sk->sk_send_head); 273 sk->sk_send_head = NULL; 274 275 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, dp->dccps_awl, dp->dccps_awh)) { 276 dccp_pr_debug("invalid ackno: S.AWL=%llu, P.ackno=%llu, S.AWH=%llu \n", 277 dp->dccps_awl, DCCP_SKB_CB(skb)->dccpd_ack_seq, dp->dccps_awh); 278 goto out_invalid_packet; 279 } 280 281 dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq; 282 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq); 283 284 if (ccid_hc_rx_init(dp->dccps_hc_rx_ccid, sk) != 0 || 285 ccid_hc_tx_init(dp->dccps_hc_tx_ccid, sk) != 0) { 286 ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk); 287 ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk); 288 /* FIXME: send appropriate RESET code */ 289 goto out_invalid_packet; 290 } 291 292 dccp_sync_mss(sk, dp->dccps_pmtu_cookie); 293 294 /* 295 * Step 10: Process REQUEST state (second part) 296 * If S.state == REQUEST, 297 * / * If we get here, P is a valid Response from the server (see 298 * Step 4), and we should move to PARTOPEN state. PARTOPEN 299 * means send an Ack, don't send Data packets, retransmit 300 * Acks periodically, and always include any Init Cookie from 301 * the Response * / 302 * S.state := PARTOPEN 303 * Set PARTOPEN timer 304 * Continue with S.state == PARTOPEN 305 * / * Step 12 will send the Ack completing the three-way 306 * handshake * / 307 */ 308 dccp_set_state(sk, DCCP_PARTOPEN); 309 310 /* Make sure socket is routed, for correct metrics. */ 311 inet_sk_rebuild_header(sk); 312 313 if (!sock_flag(sk, SOCK_DEAD)) { 314 sk->sk_state_change(sk); 315 sk_wake_async(sk, 0, POLL_OUT); 316 } 317 318 if (sk->sk_write_pending || icsk->icsk_ack.pingpong || 319 icsk->icsk_accept_queue.rskq_defer_accept) { 320 /* Save one ACK. Data will be ready after 321 * several ticks, if write_pending is set. 322 * 323 * It may be deleted, but with this feature tcpdumps 324 * look so _wonderfully_ clever, that I was not able 325 * to stand against the temptation 8) --ANK 326 */ 327 /* 328 * OK, in DCCP we can as well do a similar trick, its 329 * even in the draft, but there is no need for us to 330 * schedule an ack here, as dccp_sendmsg does this for 331 * us, also stated in the draft. -acme 332 */ 333 __kfree_skb(skb); 334 return 0; 335 } 336 dccp_send_ack(sk); 337 return -1; 338 } 339 340 out_invalid_packet: 341 return 1; /* dccp_v4_do_rcv will send a reset, but... 342 FIXME: the reset code should be DCCP_RESET_CODE_PACKET_ERROR */ 343 } 344 345 static int dccp_rcv_respond_partopen_state_process(struct sock *sk, 346 struct sk_buff *skb, 347 const struct dccp_hdr *dh, 348 const unsigned len) 349 { 350 int queued = 0; 351 352 switch (dh->dccph_type) { 353 case DCCP_PKT_RESET: 354 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 355 break; 356 case DCCP_PKT_DATAACK: 357 case DCCP_PKT_ACK: 358 /* 359 * FIXME: we should be reseting the PARTOPEN (DELACK) timer here, 360 * but only if we haven't used the DELACK timer for something else, 361 * like sending a delayed ack for a TIMESTAMP echo, etc, for now 362 * were not clearing it, sending an extra ACK when there is nothing 363 * else to do in DELACK is not a big deal after all. 364 */ 365 366 /* Stop the PARTOPEN timer */ 367 if (sk->sk_state == DCCP_PARTOPEN) 368 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); 369 370 dccp_sk(sk)->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq; 371 dccp_set_state(sk, DCCP_OPEN); 372 373 if (dh->dccph_type == DCCP_PKT_DATAACK) { 374 dccp_rcv_established(sk, skb, dh, len); 375 queued = 1; /* packet was queued (by dccp_rcv_established) */ 376 } 377 break; 378 } 379 380 return queued; 381 } 382 383 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 384 struct dccp_hdr *dh, unsigned len) 385 { 386 struct dccp_sock *dp = dccp_sk(sk); 387 const int old_state = sk->sk_state; 388 int queued = 0; 389 390 if (sk->sk_state != DCCP_LISTEN && sk->sk_state != DCCP_REQUESTING) { 391 if (dccp_check_seqno(sk, skb)) 392 goto discard; 393 394 /* 395 * Step 8: Process options and mark acknowledgeable 396 */ 397 if (dccp_parse_options(sk, skb)) 398 goto discard; 399 400 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ) 401 dccp_event_ack_recv(sk, skb); 402 403 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb); 404 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb); 405 406 /* 407 * FIXME: check ECN to see if we should use 408 * DCCP_ACKPKTS_STATE_ECN_MARKED 409 */ 410 if (dp->dccps_options.dccpo_send_ack_vector) { 411 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts, 412 DCCP_SKB_CB(skb)->dccpd_seq, 413 DCCP_ACKPKTS_STATE_RECEIVED)) 414 goto discard; 415 /* 416 * FIXME: this activation is probably wrong, have to study more 417 * TCP delack machinery and how it fits into DCCP draft, but 418 * for now it kinda "works" 8) 419 */ 420 if (dp->dccps_hc_rx_ackpkts->dccpap_ack_seqno == DCCP_MAX_SEQNO + 1 && 421 !inet_csk_ack_scheduled(sk)) { 422 inet_csk_schedule_ack(sk); 423 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, TCP_DELACK_MIN, TCP_RTO_MAX); 424 } 425 } 426 } 427 428 /* 429 * Step 9: Process Reset 430 * If P.type == Reset, 431 * Tear down connection 432 * S.state := TIMEWAIT 433 * Set TIMEWAIT timer 434 * Drop packet and return 435 */ 436 if (dh->dccph_type == DCCP_PKT_RESET) { 437 /* Queue the equivalent of TCP fin so that dccp_recvmsg exits the loop */ 438 dccp_fin(sk, skb); 439 dccp_time_wait(sk, DCCP_TIME_WAIT, 0); 440 return 0; 441 /* 442 * Step 7: Check for unexpected packet types 443 * If (S.is_server and P.type == CloseReq) 444 * or (S.is_server and P.type == Response) 445 * or (S.is_client and P.type == Request) 446 * or (S.state == RESPOND and P.type == Data), 447 * Send Sync packet acknowledging P.seqno 448 * Drop packet and return 449 */ 450 } else if ((dp->dccps_role != DCCP_ROLE_CLIENT && 451 (dh->dccph_type == DCCP_PKT_RESPONSE || dh->dccph_type == DCCP_PKT_CLOSEREQ)) || 452 (dp->dccps_role == DCCP_ROLE_CLIENT && 453 dh->dccph_type == DCCP_PKT_REQUEST) || 454 (sk->sk_state == DCCP_RESPOND && dh->dccph_type == DCCP_PKT_DATA)) { 455 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq); 456 goto discard; 457 } 458 459 switch (sk->sk_state) { 460 case DCCP_CLOSED: 461 return 1; 462 463 case DCCP_LISTEN: 464 if (dh->dccph_type == DCCP_PKT_ACK || 465 dh->dccph_type == DCCP_PKT_DATAACK) 466 return 1; 467 468 if (dh->dccph_type == DCCP_PKT_RESET) 469 goto discard; 470 471 if (dh->dccph_type == DCCP_PKT_REQUEST) { 472 if (dccp_v4_conn_request(sk, skb) < 0) 473 return 1; 474 475 /* FIXME: do congestion control initialization */ 476 goto discard; 477 } 478 goto discard; 479 480 case DCCP_REQUESTING: 481 /* FIXME: do congestion control initialization */ 482 483 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len); 484 if (queued >= 0) 485 return queued; 486 487 __kfree_skb(skb); 488 return 0; 489 490 case DCCP_RESPOND: 491 case DCCP_PARTOPEN: 492 queued = dccp_rcv_respond_partopen_state_process(sk, skb, dh, len); 493 break; 494 } 495 496 if (dh->dccph_type == DCCP_PKT_ACK || dh->dccph_type == DCCP_PKT_DATAACK) { 497 switch (old_state) { 498 case DCCP_PARTOPEN: 499 sk->sk_state_change(sk); 500 sk_wake_async(sk, 0, POLL_OUT); 501 break; 502 } 503 } 504 505 if (!queued) { 506 discard: 507 __kfree_skb(skb); 508 } 509 return 0; 510 } 511