| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * net/dccp/input.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/skbuff.h> |
| 11 | #include <linux/slab.h> |
| 12 | |
| 13 | #include <net/sock.h> |
| 14 | |
| 15 | #include "ackvec.h" |
| 16 | #include "ccid.h" |
| 17 | #include "dccp.h" |
| 18 | |
| 19 | /* rate-limit for syncs in reply to sequence-invalid packets; RFC 4340, 7.5.4 */ |
| 20 | int sysctl_dccp_sync_ratelimit __read_mostly = HZ / 8; |
| 21 | |
| 22 | static void dccp_enqueue_skb(struct sock *sk, struct sk_buff *skb) |
| 23 | { |
| 24 | __skb_pull(skb, dccp_hdr(skb)->dccph_doff * 4); |
| 25 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
| 26 | skb_set_owner_r(skb, sk); |
| 27 | sk->sk_data_ready(sk); |
| 28 | } |
| 29 | |
| 30 | static void dccp_fin(struct sock *sk, struct sk_buff *skb) |
| 31 | { |
| 32 | /* |
| 33 | * On receiving Close/CloseReq, both RD/WR shutdown are performed. |
| 34 | * RFC 4340, 8.3 says that we MAY send further Data/DataAcks after |
| 35 | * receiving the closing segment, but there is no guarantee that such |
| 36 | * data will be processed at all. |
| 37 | */ |
| 38 | sk->sk_shutdown = SHUTDOWN_MASK; |
| 39 | sock_set_flag(sk, SOCK_DONE); |
| 40 | dccp_enqueue_skb(sk, skb); |
| 41 | } |
| 42 | |
| 43 | static int dccp_rcv_close(struct sock *sk, struct sk_buff *skb) |
| 44 | { |
| 45 | int queued = 0; |
| 46 | |
| 47 | switch (sk->sk_state) { |
| 48 | /* |
| 49 | * We ignore Close when received in one of the following states: |
| 50 | * - CLOSED (may be a late or duplicate packet) |
| 51 | * - PASSIVE_CLOSEREQ (the peer has sent a CloseReq earlier) |
| 52 | * - RESPOND (already handled by dccp_check_req) |
| 53 | */ |
| 54 | case DCCP_CLOSING: |
| 55 | /* |
| 56 | * Simultaneous-close: receiving a Close after sending one. This |
| 57 | * can happen if both client and server perform active-close and |
| 58 | * will result in an endless ping-pong of crossing and retrans- |
| 59 | * mitted Close packets, which only terminates when one of the |
| 60 | * nodes times out (min. 64 seconds). Quicker convergence can be |
| 61 | * achieved when one of the nodes acts as tie-breaker. |
| 62 | * This is ok as both ends are done with data transfer and each |
| 63 | * end is just waiting for the other to acknowledge termination. |
| 64 | */ |
| 65 | if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT) |
| 66 | break; |
| 67 | fallthrough; |
| 68 | case DCCP_REQUESTING: |
| 69 | case DCCP_ACTIVE_CLOSEREQ: |
| 70 | dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED); |
| 71 | dccp_done(sk); |
| 72 | break; |
| 73 | case DCCP_OPEN: |
| 74 | case DCCP_PARTOPEN: |
| 75 | /* Give waiting application a chance to read pending data */ |
| 76 | queued = 1; |
| 77 | dccp_fin(sk, skb); |
| 78 | dccp_set_state(sk, DCCP_PASSIVE_CLOSE); |
| 79 | fallthrough; |
| 80 | case DCCP_PASSIVE_CLOSE: |
| 81 | /* |
| 82 | * Retransmitted Close: we have already enqueued the first one. |
| 83 | */ |
| 84 | sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP); |
| 85 | } |
| 86 | return queued; |
| 87 | } |
| 88 | |
| 89 | static int dccp_rcv_closereq(struct sock *sk, struct sk_buff *skb) |
| 90 | { |
| 91 | int queued = 0; |
| 92 | |
| 93 | /* |
| 94 | * Step 7: Check for unexpected packet types |
| 95 | * If (S.is_server and P.type == CloseReq) |
| 96 | * Send Sync packet acknowledging P.seqno |
| 97 | * Drop packet and return |
| 98 | */ |
| 99 | if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT) { |
| 100 | dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC); |
| 101 | return queued; |
| 102 | } |
| 103 | |
| 104 | /* Step 13: process relevant Client states < CLOSEREQ */ |
| 105 | switch (sk->sk_state) { |
| 106 | case DCCP_REQUESTING: |
| 107 | dccp_send_close(sk, 0); |
| 108 | dccp_set_state(sk, DCCP_CLOSING); |
| 109 | break; |
| 110 | case DCCP_OPEN: |
| 111 | case DCCP_PARTOPEN: |
| 112 | /* Give waiting application a chance to read pending data */ |
| 113 | queued = 1; |
| 114 | dccp_fin(sk, skb); |
| 115 | dccp_set_state(sk, DCCP_PASSIVE_CLOSEREQ); |
| 116 | fallthrough; |
| 117 | case DCCP_PASSIVE_CLOSEREQ: |
| 118 | sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP); |
| 119 | } |
| 120 | return queued; |
| 121 | } |
| 122 | |
| 123 | static u16 dccp_reset_code_convert(const u8 code) |
| 124 | { |
| 125 | static const u16 error_code[] = { |
| 126 | [DCCP_RESET_CODE_CLOSED] = 0, /* normal termination */ |
| 127 | [DCCP_RESET_CODE_UNSPECIFIED] = 0, /* nothing known */ |
| 128 | [DCCP_RESET_CODE_ABORTED] = ECONNRESET, |
| 129 | |
| 130 | [DCCP_RESET_CODE_NO_CONNECTION] = ECONNREFUSED, |
| 131 | [DCCP_RESET_CODE_CONNECTION_REFUSED] = ECONNREFUSED, |
| 132 | [DCCP_RESET_CODE_TOO_BUSY] = EUSERS, |
| 133 | [DCCP_RESET_CODE_AGGRESSION_PENALTY] = EDQUOT, |
| 134 | |
| 135 | [DCCP_RESET_CODE_PACKET_ERROR] = ENOMSG, |
| 136 | [DCCP_RESET_CODE_BAD_INIT_COOKIE] = EBADR, |
| 137 | [DCCP_RESET_CODE_BAD_SERVICE_CODE] = EBADRQC, |
| 138 | [DCCP_RESET_CODE_OPTION_ERROR] = EILSEQ, |
| 139 | [DCCP_RESET_CODE_MANDATORY_ERROR] = EOPNOTSUPP, |
| 140 | }; |
| 141 | |
| 142 | return code >= DCCP_MAX_RESET_CODES ? 0 : error_code[code]; |
| 143 | } |
| 144 | |
| 145 | static void dccp_rcv_reset(struct sock *sk, struct sk_buff *skb) |
| 146 | { |
| 147 | u16 err = dccp_reset_code_convert(dccp_hdr_reset(skb)->dccph_reset_code); |
| 148 | |
| 149 | sk->sk_err = err; |
| 150 | |
| 151 | /* Queue the equivalent of TCP fin so that dccp_recvmsg exits the loop */ |
| 152 | dccp_fin(sk, skb); |
| 153 | |
| 154 | if (err && !sock_flag(sk, SOCK_DEAD)) |
| 155 | sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); |
| 156 | dccp_time_wait(sk, DCCP_TIME_WAIT, 0); |
| 157 | } |
| 158 | |
| 159 | static void dccp_handle_ackvec_processing(struct sock *sk, struct sk_buff *skb) |
| 160 | { |
| 161 | struct dccp_ackvec *av = dccp_sk(sk)->dccps_hc_rx_ackvec; |
| 162 | |
| 163 | if (av == NULL) |
| 164 | return; |
| 165 | if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ) |
| 166 | dccp_ackvec_clear_state(av, DCCP_SKB_CB(skb)->dccpd_ack_seq); |
| 167 | dccp_ackvec_input(av, skb); |
| 168 | } |
| 169 | |
| 170 | static void dccp_deliver_input_to_ccids(struct sock *sk, struct sk_buff *skb) |
| 171 | { |
| 172 | const struct dccp_sock *dp = dccp_sk(sk); |
| 173 | |
| 174 | /* Don't deliver to RX CCID when node has shut down read end. */ |
| 175 | if (!(sk->sk_shutdown & RCV_SHUTDOWN)) |
| 176 | ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb); |
| 177 | /* |
| 178 | * Until the TX queue has been drained, we can not honour SHUT_WR, since |
| 179 | * we need received feedback as input to adjust congestion control. |
| 180 | */ |
| 181 | if (sk->sk_write_queue.qlen > 0 || !(sk->sk_shutdown & SEND_SHUTDOWN)) |
| 182 | ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb); |
| 183 | } |
| 184 | |
| 185 | static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb) |
| 186 | { |
| 187 | const struct dccp_hdr *dh = dccp_hdr(skb); |
| 188 | struct dccp_sock *dp = dccp_sk(sk); |
| 189 | u64 lswl, lawl, seqno = DCCP_SKB_CB(skb)->dccpd_seq, |
| 190 | ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq; |
| 191 | |
| 192 | /* |
| 193 | * Step 5: Prepare sequence numbers for Sync |
| 194 | * If P.type == Sync or P.type == SyncAck, |
| 195 | * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL, |
| 196 | * / * P is valid, so update sequence number variables |
| 197 | * accordingly. After this update, P will pass the tests |
| 198 | * in Step 6. A SyncAck is generated if necessary in |
| 199 | * Step 15 * / |
| 200 | * Update S.GSR, S.SWL, S.SWH |
| 201 | * Otherwise, |
| 202 | * Drop packet and return |
| 203 | */ |
| 204 | if (dh->dccph_type == DCCP_PKT_SYNC || |
| 205 | dh->dccph_type == DCCP_PKT_SYNCACK) { |
| 206 | if (between48(ackno, dp->dccps_awl, dp->dccps_awh) && |
| 207 | dccp_delta_seqno(dp->dccps_swl, seqno) >= 0) |
| 208 | dccp_update_gsr(sk, seqno); |
| 209 | else |
| 210 | return -1; |
| 211 | } |
| 212 | |
| 213 | /* |
| 214 | * Step 6: Check sequence numbers |
| 215 | * Let LSWL = S.SWL and LAWL = S.AWL |
| 216 | * If P.type == CloseReq or P.type == Close or P.type == Reset, |
| 217 | * LSWL := S.GSR + 1, LAWL := S.GAR |
| 218 | * If LSWL <= P.seqno <= S.SWH |
| 219 | * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH), |
| 220 | * Update S.GSR, S.SWL, S.SWH |
| 221 | * If P.type != Sync, |
| 222 | * Update S.GAR |
| 223 | */ |
| 224 | lswl = dp->dccps_swl; |
| 225 | lawl = dp->dccps_awl; |
| 226 | |
| 227 | if (dh->dccph_type == DCCP_PKT_CLOSEREQ || |
| 228 | dh->dccph_type == DCCP_PKT_CLOSE || |
| 229 | dh->dccph_type == DCCP_PKT_RESET) { |
| 230 | lswl = ADD48(dp->dccps_gsr, 1); |
| 231 | lawl = dp->dccps_gar; |
| 232 | } |
| 233 | |
| 234 | if (between48(seqno, lswl, dp->dccps_swh) && |
| 235 | (ackno == DCCP_PKT_WITHOUT_ACK_SEQ || |
| 236 | between48(ackno, lawl, dp->dccps_awh))) { |
| 237 | dccp_update_gsr(sk, seqno); |
| 238 | |
| 239 | if (dh->dccph_type != DCCP_PKT_SYNC && |
| 240 | ackno != DCCP_PKT_WITHOUT_ACK_SEQ && |
| 241 | after48(ackno, dp->dccps_gar)) |
| 242 | dp->dccps_gar = ackno; |
| 243 | } else { |
| 244 | unsigned long now = jiffies; |
| 245 | /* |
| 246 | * Step 6: Check sequence numbers |
| 247 | * Otherwise, |
| 248 | * If P.type == Reset, |
| 249 | * Send Sync packet acknowledging S.GSR |
| 250 | * Otherwise, |
| 251 | * Send Sync packet acknowledging P.seqno |
| 252 | * Drop packet and return |
| 253 | * |
| 254 | * These Syncs are rate-limited as per RFC 4340, 7.5.4: |
| 255 | * at most 1 / (dccp_sync_rate_limit * HZ) Syncs per second. |
| 256 | */ |
| 257 | if (time_before(now, (dp->dccps_rate_last + |
| 258 | sysctl_dccp_sync_ratelimit))) |
| 259 | return -1; |
| 260 | |
| 261 | DCCP_WARN("Step 6 failed for %s packet, " |
| 262 | "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and " |
| 263 | "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), " |
| 264 | "sending SYNC...\n", dccp_packet_name(dh->dccph_type), |
| 265 | (unsigned long long) lswl, (unsigned long long) seqno, |
| 266 | (unsigned long long) dp->dccps_swh, |
| 267 | (ackno == DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist" |
| 268 | : "exists", |
| 269 | (unsigned long long) lawl, (unsigned long long) ackno, |
| 270 | (unsigned long long) dp->dccps_awh); |
| 271 | |
| 272 | dp->dccps_rate_last = now; |
| 273 | |
| 274 | if (dh->dccph_type == DCCP_PKT_RESET) |
| 275 | seqno = dp->dccps_gsr; |
| 276 | dccp_send_sync(sk, seqno, DCCP_PKT_SYNC); |
| 277 | return -1; |
| 278 | } |
| 279 | |
| 280 | return 0; |
| 281 | } |
| 282 | |
| 283 | static int __dccp_rcv_established(struct sock *sk, struct sk_buff *skb, |
| 284 | const struct dccp_hdr *dh, const unsigned int len) |
| 285 | { |
| 286 | struct dccp_sock *dp = dccp_sk(sk); |
| 287 | |
| 288 | switch (dccp_hdr(skb)->dccph_type) { |
| 289 | case DCCP_PKT_DATAACK: |
| 290 | case DCCP_PKT_DATA: |
| 291 | /* |
| 292 | * FIXME: schedule DATA_DROPPED (RFC 4340, 11.7.2) if and when |
| 293 | * - sk_shutdown == RCV_SHUTDOWN, use Code 1, "Not Listening" |
| 294 | * - sk_receive_queue is full, use Code 2, "Receive Buffer" |
| 295 | */ |
| 296 | dccp_enqueue_skb(sk, skb); |
| 297 | return 0; |
| 298 | case DCCP_PKT_ACK: |
| 299 | goto discard; |
| 300 | case DCCP_PKT_RESET: |
| 301 | /* |
| 302 | * Step 9: Process Reset |
| 303 | * If P.type == Reset, |
| 304 | * Tear down connection |
| 305 | * S.state := TIMEWAIT |
| 306 | * Set TIMEWAIT timer |
| 307 | * Drop packet and return |
| 308 | */ |
| 309 | dccp_rcv_reset(sk, skb); |
| 310 | return 0; |
| 311 | case DCCP_PKT_CLOSEREQ: |
| 312 | if (dccp_rcv_closereq(sk, skb)) |
| 313 | return 0; |
| 314 | goto discard; |
| 315 | case DCCP_PKT_CLOSE: |
| 316 | if (dccp_rcv_close(sk, skb)) |
| 317 | return 0; |
| 318 | goto discard; |
| 319 | case DCCP_PKT_REQUEST: |
| 320 | /* Step 7 |
| 321 | * or (S.is_server and P.type == Response) |
| 322 | * or (S.is_client and P.type == Request) |
| 323 | * or (S.state >= OPEN and P.type == Request |
| 324 | * and P.seqno >= S.OSR) |
| 325 | * or (S.state >= OPEN and P.type == Response |
| 326 | * and P.seqno >= S.OSR) |
| 327 | * or (S.state == RESPOND and P.type == Data), |
| 328 | * Send Sync packet acknowledging P.seqno |
| 329 | * Drop packet and return |
| 330 | */ |
| 331 | if (dp->dccps_role != DCCP_ROLE_LISTEN) |
| 332 | goto send_sync; |
| 333 | goto check_seq; |
| 334 | case DCCP_PKT_RESPONSE: |
| 335 | if (dp->dccps_role != DCCP_ROLE_CLIENT) |
| 336 | goto send_sync; |
| 337 | check_seq: |
| 338 | if (dccp_delta_seqno(dp->dccps_osr, |
| 339 | DCCP_SKB_CB(skb)->dccpd_seq) >= 0) { |
| 340 | send_sync: |
| 341 | dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, |
| 342 | DCCP_PKT_SYNC); |
| 343 | } |
| 344 | break; |
| 345 | case DCCP_PKT_SYNC: |
| 346 | dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, |
| 347 | DCCP_PKT_SYNCACK); |
| 348 | /* |
| 349 | * From RFC 4340, sec. 5.7 |
| 350 | * |
| 351 | * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets |
| 352 | * MAY have non-zero-length application data areas, whose |
| 353 | * contents receivers MUST ignore. |
| 354 | */ |
| 355 | goto discard; |
| 356 | } |
| 357 | |
| 358 | DCCP_INC_STATS(DCCP_MIB_INERRS); |
| 359 | discard: |
| 360 | __kfree_skb(skb); |
| 361 | return 0; |
| 362 | } |
| 363 | |
| 364 | int dccp_rcv_established(struct sock *sk, struct sk_buff *skb, |
| 365 | const struct dccp_hdr *dh, const unsigned int len) |
| 366 | { |
| 367 | if (dccp_check_seqno(sk, skb)) |
| 368 | goto discard; |
| 369 | |
| 370 | if (dccp_parse_options(sk, NULL, skb)) |
| 371 | return 1; |
| 372 | |
| 373 | dccp_handle_ackvec_processing(sk, skb); |
| 374 | dccp_deliver_input_to_ccids(sk, skb); |
| 375 | |
| 376 | return __dccp_rcv_established(sk, skb, dh, len); |
| 377 | discard: |
| 378 | __kfree_skb(skb); |
| 379 | return 0; |
| 380 | } |
| 381 | |
| 382 | EXPORT_SYMBOL_GPL(dccp_rcv_established); |
| 383 | |
| 384 | static int dccp_rcv_request_sent_state_process(struct sock *sk, |
| 385 | struct sk_buff *skb, |
| 386 | const struct dccp_hdr *dh, |
| 387 | const unsigned int len) |
| 388 | { |
| 389 | /* |
| 390 | * Step 4: Prepare sequence numbers in REQUEST |
| 391 | * If S.state == REQUEST, |
| 392 | * If (P.type == Response or P.type == Reset) |
| 393 | * and S.AWL <= P.ackno <= S.AWH, |
| 394 | * / * Set sequence number variables corresponding to the |
| 395 | * other endpoint, so P will pass the tests in Step 6 * / |
| 396 | * Set S.GSR, S.ISR, S.SWL, S.SWH |
| 397 | * / * Response processing continues in Step 10; Reset |
| 398 | * processing continues in Step 9 * / |
| 399 | */ |
| 400 | if (dh->dccph_type == DCCP_PKT_RESPONSE) { |
| 401 | const struct inet_connection_sock *icsk = inet_csk(sk); |
| 402 | struct dccp_sock *dp = dccp_sk(sk); |
| 403 | long tstamp = dccp_timestamp(); |
| 404 | |
| 405 | if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, |
| 406 | dp->dccps_awl, dp->dccps_awh)) { |
| 407 | dccp_pr_debug("invalid ackno: S.AWL=%llu, " |
| 408 | "P.ackno=%llu, S.AWH=%llu\n", |
| 409 | (unsigned long long)dp->dccps_awl, |
| 410 | (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq, |
| 411 | (unsigned long long)dp->dccps_awh); |
| 412 | goto out_invalid_packet; |
| 413 | } |
| 414 | |
| 415 | /* |
| 416 | * If option processing (Step 8) failed, return 1 here so that |
| 417 | * dccp_v4_do_rcv() sends a Reset. The Reset code depends on |
| 418 | * the option type and is set in dccp_parse_options(). |
| 419 | */ |
| 420 | if (dccp_parse_options(sk, NULL, skb)) |
| 421 | return 1; |
| 422 | |
| 423 | /* Obtain usec RTT sample from SYN exchange (used by TFRC). */ |
| 424 | if (likely(dp->dccps_options_received.dccpor_timestamp_echo)) |
| 425 | dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * (tstamp - |
| 426 | dp->dccps_options_received.dccpor_timestamp_echo)); |
| 427 | |
| 428 | /* Stop the REQUEST timer */ |
| 429 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); |
| 430 | WARN_ON(sk->sk_send_head == NULL); |
| 431 | kfree_skb(sk->sk_send_head); |
| 432 | sk->sk_send_head = NULL; |
| 433 | |
| 434 | /* |
| 435 | * Set ISR, GSR from packet. ISS was set in dccp_v{4,6}_connect |
| 436 | * and GSS in dccp_transmit_skb(). Setting AWL/AWH and SWL/SWH |
| 437 | * is done as part of activating the feature values below, since |
| 438 | * these settings depend on the local/remote Sequence Window |
| 439 | * features, which were undefined or not confirmed until now. |
| 440 | */ |
| 441 | dp->dccps_gsr = dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq; |
| 442 | |
| 443 | dccp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
| 444 | |
| 445 | /* |
| 446 | * Step 10: Process REQUEST state (second part) |
| 447 | * If S.state == REQUEST, |
| 448 | * / * If we get here, P is a valid Response from the |
| 449 | * server (see Step 4), and we should move to |
| 450 | * PARTOPEN state. PARTOPEN means send an Ack, |
| 451 | * don't send Data packets, retransmit Acks |
| 452 | * periodically, and always include any Init Cookie |
| 453 | * from the Response * / |
| 454 | * S.state := PARTOPEN |
| 455 | * Set PARTOPEN timer |
| 456 | * Continue with S.state == PARTOPEN |
| 457 | * / * Step 12 will send the Ack completing the |
| 458 | * three-way handshake * / |
| 459 | */ |
| 460 | dccp_set_state(sk, DCCP_PARTOPEN); |
| 461 | |
| 462 | /* |
| 463 | * If feature negotiation was successful, activate features now; |
| 464 | * an activation failure means that this host could not activate |
| 465 | * one ore more features (e.g. insufficient memory), which would |
| 466 | * leave at least one feature in an undefined state. |
| 467 | */ |
| 468 | if (dccp_feat_activate_values(sk, &dp->dccps_featneg)) |
| 469 | goto unable_to_proceed; |
| 470 | |
| 471 | /* Make sure socket is routed, for correct metrics. */ |
| 472 | icsk->icsk_af_ops->rebuild_header(sk); |
| 473 | |
| 474 | if (!sock_flag(sk, SOCK_DEAD)) { |
| 475 | sk->sk_state_change(sk); |
| 476 | sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT); |
| 477 | } |
| 478 | |
| 479 | if (sk->sk_write_pending || inet_csk_in_pingpong_mode(sk) || |
| 480 | icsk->icsk_accept_queue.rskq_defer_accept) { |
| 481 | /* Save one ACK. Data will be ready after |
| 482 | * several ticks, if write_pending is set. |
| 483 | * |
| 484 | * It may be deleted, but with this feature tcpdumps |
| 485 | * look so _wonderfully_ clever, that I was not able |
| 486 | * to stand against the temptation 8) --ANK |
| 487 | */ |
| 488 | /* |
| 489 | * OK, in DCCP we can as well do a similar trick, its |
| 490 | * even in the draft, but there is no need for us to |
| 491 | * schedule an ack here, as dccp_sendmsg does this for |
| 492 | * us, also stated in the draft. -acme |
| 493 | */ |
| 494 | __kfree_skb(skb); |
| 495 | return 0; |
| 496 | } |
| 497 | dccp_send_ack(sk); |
| 498 | return -1; |
| 499 | } |
| 500 | |
| 501 | out_invalid_packet: |
| 502 | /* dccp_v4_do_rcv will send a reset */ |
| 503 | DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR; |
| 504 | return 1; |
| 505 | |
| 506 | unable_to_proceed: |
| 507 | DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_ABORTED; |
| 508 | /* |
| 509 | * We mark this socket as no longer usable, so that the loop in |
| 510 | * dccp_sendmsg() terminates and the application gets notified. |
| 511 | */ |
| 512 | dccp_set_state(sk, DCCP_CLOSED); |
| 513 | sk->sk_err = ECOMM; |
| 514 | return 1; |
| 515 | } |
| 516 | |
| 517 | static int dccp_rcv_respond_partopen_state_process(struct sock *sk, |
| 518 | struct sk_buff *skb, |
| 519 | const struct dccp_hdr *dh, |
| 520 | const unsigned int len) |
| 521 | { |
| 522 | struct dccp_sock *dp = dccp_sk(sk); |
| 523 | u32 sample = dp->dccps_options_received.dccpor_timestamp_echo; |
| 524 | int queued = 0; |
| 525 | |
| 526 | switch (dh->dccph_type) { |
| 527 | case DCCP_PKT_RESET: |
| 528 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); |
| 529 | break; |
| 530 | case DCCP_PKT_DATA: |
| 531 | if (sk->sk_state == DCCP_RESPOND) |
| 532 | break; |
| 533 | fallthrough; |
| 534 | case DCCP_PKT_DATAACK: |
| 535 | case DCCP_PKT_ACK: |
| 536 | /* |
| 537 | * FIXME: we should be resetting the PARTOPEN (DELACK) timer |
| 538 | * here but only if we haven't used the DELACK timer for |
| 539 | * something else, like sending a delayed ack for a TIMESTAMP |
| 540 | * echo, etc, for now were not clearing it, sending an extra |
| 541 | * ACK when there is nothing else to do in DELACK is not a big |
| 542 | * deal after all. |
| 543 | */ |
| 544 | |
| 545 | /* Stop the PARTOPEN timer */ |
| 546 | if (sk->sk_state == DCCP_PARTOPEN) |
| 547 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK); |
| 548 | |
| 549 | /* Obtain usec RTT sample from SYN exchange (used by TFRC). */ |
| 550 | if (likely(sample)) { |
| 551 | long delta = dccp_timestamp() - sample; |
| 552 | |
| 553 | dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * delta); |
| 554 | } |
| 555 | |
| 556 | dp->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq; |
| 557 | dccp_set_state(sk, DCCP_OPEN); |
| 558 | |
| 559 | if (dh->dccph_type == DCCP_PKT_DATAACK || |
| 560 | dh->dccph_type == DCCP_PKT_DATA) { |
| 561 | __dccp_rcv_established(sk, skb, dh, len); |
| 562 | queued = 1; /* packet was queued |
| 563 | (by __dccp_rcv_established) */ |
| 564 | } |
| 565 | break; |
| 566 | } |
| 567 | |
| 568 | return queued; |
| 569 | } |
| 570 | |
| 571 | int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb, |
| 572 | struct dccp_hdr *dh, unsigned int len) |
| 573 | { |
| 574 | struct dccp_sock *dp = dccp_sk(sk); |
| 575 | struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); |
| 576 | const int old_state = sk->sk_state; |
| 577 | bool acceptable; |
| 578 | int queued = 0; |
| 579 | |
| 580 | /* |
| 581 | * Step 3: Process LISTEN state |
| 582 | * |
| 583 | * If S.state == LISTEN, |
| 584 | * If P.type == Request or P contains a valid Init Cookie option, |
| 585 | * (* Must scan the packet's options to check for Init |
| 586 | * Cookies. Only Init Cookies are processed here, |
| 587 | * however; other options are processed in Step 8. This |
| 588 | * scan need only be performed if the endpoint uses Init |
| 589 | * Cookies *) |
| 590 | * (* Generate a new socket and switch to that socket *) |
| 591 | * Set S := new socket for this port pair |
| 592 | * S.state = RESPOND |
| 593 | * Choose S.ISS (initial seqno) or set from Init Cookies |
| 594 | * Initialize S.GAR := S.ISS |
| 595 | * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init |
| 596 | * Cookies Continue with S.state == RESPOND |
| 597 | * (* A Response packet will be generated in Step 11 *) |
| 598 | * Otherwise, |
| 599 | * Generate Reset(No Connection) unless P.type == Reset |
| 600 | * Drop packet and return |
| 601 | */ |
| 602 | if (sk->sk_state == DCCP_LISTEN) { |
| 603 | if (dh->dccph_type == DCCP_PKT_REQUEST) { |
| 604 | /* It is possible that we process SYN packets from backlog, |
| 605 | * so we need to make sure to disable BH and RCU right there. |
| 606 | */ |
| 607 | rcu_read_lock(); |
| 608 | local_bh_disable(); |
| 609 | acceptable = inet_csk(sk)->icsk_af_ops->conn_request(sk, skb) >= 0; |
| 610 | local_bh_enable(); |
| 611 | rcu_read_unlock(); |
| 612 | if (!acceptable) |
| 613 | return 1; |
| 614 | consume_skb(skb); |
| 615 | return 0; |
| 616 | } |
| 617 | if (dh->dccph_type == DCCP_PKT_RESET) |
| 618 | goto discard; |
| 619 | |
| 620 | /* Caller (dccp_v4_do_rcv) will send Reset */ |
| 621 | dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION; |
| 622 | return 1; |
| 623 | } else if (sk->sk_state == DCCP_CLOSED) { |
| 624 | dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION; |
| 625 | return 1; |
| 626 | } |
| 627 | |
| 628 | /* Step 6: Check sequence numbers (omitted in LISTEN/REQUEST state) */ |
| 629 | if (sk->sk_state != DCCP_REQUESTING && dccp_check_seqno(sk, skb)) |
| 630 | goto discard; |
| 631 | |
| 632 | /* |
| 633 | * Step 7: Check for unexpected packet types |
| 634 | * If (S.is_server and P.type == Response) |
| 635 | * or (S.is_client and P.type == Request) |
| 636 | * or (S.state == RESPOND and P.type == Data), |
| 637 | * Send Sync packet acknowledging P.seqno |
| 638 | * Drop packet and return |
| 639 | */ |
| 640 | if ((dp->dccps_role != DCCP_ROLE_CLIENT && |
| 641 | dh->dccph_type == DCCP_PKT_RESPONSE) || |
| 642 | (dp->dccps_role == DCCP_ROLE_CLIENT && |
| 643 | dh->dccph_type == DCCP_PKT_REQUEST) || |
| 644 | (sk->sk_state == DCCP_RESPOND && dh->dccph_type == DCCP_PKT_DATA)) { |
| 645 | dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC); |
| 646 | goto discard; |
| 647 | } |
| 648 | |
| 649 | /* Step 8: Process options */ |
| 650 | if (dccp_parse_options(sk, NULL, skb)) |
| 651 | return 1; |
| 652 | |
| 653 | /* |
| 654 | * Step 9: Process Reset |
| 655 | * If P.type == Reset, |
| 656 | * Tear down connection |
| 657 | * S.state := TIMEWAIT |
| 658 | * Set TIMEWAIT timer |
| 659 | * Drop packet and return |
| 660 | */ |
| 661 | if (dh->dccph_type == DCCP_PKT_RESET) { |
| 662 | dccp_rcv_reset(sk, skb); |
| 663 | return 0; |
| 664 | } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) { /* Step 13 */ |
| 665 | if (dccp_rcv_closereq(sk, skb)) |
| 666 | return 0; |
| 667 | goto discard; |
| 668 | } else if (dh->dccph_type == DCCP_PKT_CLOSE) { /* Step 14 */ |
| 669 | if (dccp_rcv_close(sk, skb)) |
| 670 | return 0; |
| 671 | goto discard; |
| 672 | } |
| 673 | |
| 674 | switch (sk->sk_state) { |
| 675 | case DCCP_REQUESTING: |
| 676 | queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len); |
| 677 | if (queued >= 0) |
| 678 | return queued; |
| 679 | |
| 680 | __kfree_skb(skb); |
| 681 | return 0; |
| 682 | |
| 683 | case DCCP_PARTOPEN: |
| 684 | /* Step 8: if using Ack Vectors, mark packet acknowledgeable */ |
| 685 | dccp_handle_ackvec_processing(sk, skb); |
| 686 | dccp_deliver_input_to_ccids(sk, skb); |
| 687 | fallthrough; |
| 688 | case DCCP_RESPOND: |
| 689 | queued = dccp_rcv_respond_partopen_state_process(sk, skb, |
| 690 | dh, len); |
| 691 | break; |
| 692 | } |
| 693 | |
| 694 | if (dh->dccph_type == DCCP_PKT_ACK || |
| 695 | dh->dccph_type == DCCP_PKT_DATAACK) { |
| 696 | switch (old_state) { |
| 697 | case DCCP_PARTOPEN: |
| 698 | sk->sk_state_change(sk); |
| 699 | sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT); |
| 700 | break; |
| 701 | } |
| 702 | } else if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) { |
| 703 | dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK); |
| 704 | goto discard; |
| 705 | } |
| 706 | |
| 707 | if (!queued) { |
| 708 | discard: |
| 709 | __kfree_skb(skb); |
| 710 | } |
| 711 | return 0; |
| 712 | } |
| 713 | |
| 714 | EXPORT_SYMBOL_GPL(dccp_rcv_state_process); |
| 715 | |
| 716 | /** |
| 717 | * dccp_sample_rtt - Validate and finalise computation of RTT sample |
| 718 | * @sk: socket structure |
| 719 | * @delta: number of microseconds between packet and acknowledgment |
| 720 | * |
| 721 | * The routine is kept generic to work in different contexts. It should be |
| 722 | * called immediately when the ACK used for the RTT sample arrives. |
| 723 | */ |
| 724 | u32 dccp_sample_rtt(struct sock *sk, long delta) |
| 725 | { |
| 726 | /* dccpor_elapsed_time is either zeroed out or set and > 0 */ |
| 727 | delta -= dccp_sk(sk)->dccps_options_received.dccpor_elapsed_time * 10; |
| 728 | |
| 729 | if (unlikely(delta <= 0)) { |
| 730 | DCCP_WARN("unusable RTT sample %ld, using min\n", delta); |
| 731 | return DCCP_SANE_RTT_MIN; |
| 732 | } |
| 733 | if (unlikely(delta > DCCP_SANE_RTT_MAX)) { |
| 734 | DCCP_WARN("RTT sample %ld too large, using max\n", delta); |
| 735 | return DCCP_SANE_RTT_MAX; |
| 736 | } |
| 737 | |
| 738 | return delta; |
| 739 | } |