Arnaldo Carvalho de Melo | 7c65787 | 2005-08-09 20:14:34 -0700 | [diff] [blame] | 1 | /* |
| 2 | * net/dccp/proto.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 modify it |
| 8 | * under the terms of the GNU General Public License version 2 as |
| 9 | * published by the Free Software Foundation. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/config.h> |
| 13 | #include <linux/dccp.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <linux/types.h> |
| 16 | #include <linux/sched.h> |
| 17 | #include <linux/kernel.h> |
| 18 | #include <linux/skbuff.h> |
| 19 | #include <linux/netdevice.h> |
| 20 | #include <linux/in.h> |
| 21 | #include <linux/if_arp.h> |
| 22 | #include <linux/init.h> |
| 23 | #include <linux/random.h> |
| 24 | #include <net/checksum.h> |
| 25 | |
| 26 | #include <net/inet_common.h> |
| 27 | #include <net/ip.h> |
| 28 | #include <net/protocol.h> |
| 29 | #include <net/sock.h> |
| 30 | #include <net/xfrm.h> |
| 31 | |
| 32 | #include <asm/semaphore.h> |
| 33 | #include <linux/spinlock.h> |
| 34 | #include <linux/timer.h> |
| 35 | #include <linux/delay.h> |
| 36 | #include <linux/poll.h> |
| 37 | #include <linux/dccp.h> |
| 38 | |
| 39 | #include "ccid.h" |
| 40 | #include "dccp.h" |
| 41 | |
| 42 | DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics); |
| 43 | |
| 44 | atomic_t dccp_orphan_count = ATOMIC_INIT(0); |
| 45 | |
| 46 | static struct net_protocol dccp_protocol = { |
| 47 | .handler = dccp_v4_rcv, |
| 48 | .err_handler = dccp_v4_err, |
| 49 | }; |
| 50 | |
| 51 | const char *dccp_packet_name(const int type) |
| 52 | { |
| 53 | static const char *dccp_packet_names[] = { |
| 54 | [DCCP_PKT_REQUEST] = "REQUEST", |
| 55 | [DCCP_PKT_RESPONSE] = "RESPONSE", |
| 56 | [DCCP_PKT_DATA] = "DATA", |
| 57 | [DCCP_PKT_ACK] = "ACK", |
| 58 | [DCCP_PKT_DATAACK] = "DATAACK", |
| 59 | [DCCP_PKT_CLOSEREQ] = "CLOSEREQ", |
| 60 | [DCCP_PKT_CLOSE] = "CLOSE", |
| 61 | [DCCP_PKT_RESET] = "RESET", |
| 62 | [DCCP_PKT_SYNC] = "SYNC", |
| 63 | [DCCP_PKT_SYNCACK] = "SYNCACK", |
| 64 | }; |
| 65 | |
| 66 | if (type >= DCCP_NR_PKT_TYPES) |
| 67 | return "INVALID"; |
| 68 | else |
| 69 | return dccp_packet_names[type]; |
| 70 | } |
| 71 | |
| 72 | EXPORT_SYMBOL_GPL(dccp_packet_name); |
| 73 | |
| 74 | const char *dccp_state_name(const int state) |
| 75 | { |
| 76 | static char *dccp_state_names[] = { |
| 77 | [DCCP_OPEN] = "OPEN", |
| 78 | [DCCP_REQUESTING] = "REQUESTING", |
| 79 | [DCCP_PARTOPEN] = "PARTOPEN", |
| 80 | [DCCP_LISTEN] = "LISTEN", |
| 81 | [DCCP_RESPOND] = "RESPOND", |
| 82 | [DCCP_CLOSING] = "CLOSING", |
| 83 | [DCCP_TIME_WAIT] = "TIME_WAIT", |
| 84 | [DCCP_CLOSED] = "CLOSED", |
| 85 | }; |
| 86 | |
| 87 | if (state >= DCCP_MAX_STATES) |
| 88 | return "INVALID STATE!"; |
| 89 | else |
| 90 | return dccp_state_names[state]; |
| 91 | } |
| 92 | |
| 93 | EXPORT_SYMBOL_GPL(dccp_state_name); |
| 94 | |
| 95 | static inline int dccp_listen_start(struct sock *sk) |
| 96 | { |
| 97 | dccp_sk(sk)->dccps_role = DCCP_ROLE_LISTEN; |
| 98 | return inet_csk_listen_start(sk, TCP_SYNQ_HSIZE); |
| 99 | } |
| 100 | |
| 101 | int dccp_disconnect(struct sock *sk, int flags) |
| 102 | { |
| 103 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 104 | struct inet_sock *inet = inet_sk(sk); |
| 105 | int err = 0; |
| 106 | const int old_state = sk->sk_state; |
| 107 | |
| 108 | if (old_state != DCCP_CLOSED) |
| 109 | dccp_set_state(sk, DCCP_CLOSED); |
| 110 | |
| 111 | /* ABORT function of RFC793 */ |
| 112 | if (old_state == DCCP_LISTEN) { |
| 113 | inet_csk_listen_stop(sk); |
| 114 | /* FIXME: do the active reset thing */ |
| 115 | } else if (old_state == DCCP_REQUESTING) |
| 116 | sk->sk_err = ECONNRESET; |
| 117 | |
| 118 | dccp_clear_xmit_timers(sk); |
| 119 | __skb_queue_purge(&sk->sk_receive_queue); |
| 120 | if (sk->sk_send_head != NULL) { |
| 121 | __kfree_skb(sk->sk_send_head); |
| 122 | sk->sk_send_head = NULL; |
| 123 | } |
| 124 | |
| 125 | inet->dport = 0; |
| 126 | |
| 127 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) |
| 128 | inet_reset_saddr(sk); |
| 129 | |
| 130 | sk->sk_shutdown = 0; |
| 131 | sock_reset_flag(sk, SOCK_DONE); |
| 132 | |
| 133 | icsk->icsk_backoff = 0; |
| 134 | inet_csk_delack_init(sk); |
| 135 | __sk_dst_reset(sk); |
| 136 | |
| 137 | BUG_TRAP(!inet->num || icsk->icsk_bind_hash); |
| 138 | |
| 139 | sk->sk_error_report(sk); |
| 140 | return err; |
| 141 | } |
| 142 | |
| 143 | int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg) |
| 144 | { |
| 145 | dccp_pr_debug("entry\n"); |
| 146 | return -ENOIOCTLCMD; |
| 147 | } |
| 148 | |
| 149 | int dccp_setsockopt(struct sock *sk, int level, int optname, |
| 150 | char *optval, int optlen) |
| 151 | { |
| 152 | dccp_pr_debug("entry\n"); |
| 153 | |
| 154 | if (level != SOL_DCCP) |
| 155 | return ip_setsockopt(sk, level, optname, optval, optlen); |
| 156 | |
| 157 | return -EOPNOTSUPP; |
| 158 | } |
| 159 | |
| 160 | int dccp_getsockopt(struct sock *sk, int level, int optname, |
| 161 | char *optval, int *optlen) |
| 162 | { |
| 163 | dccp_pr_debug("entry\n"); |
| 164 | |
| 165 | if (level != SOL_DCCP) |
| 166 | return ip_getsockopt(sk, level, optname, optval, optlen); |
| 167 | |
| 168 | return -EOPNOTSUPP; |
| 169 | } |
| 170 | |
| 171 | int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, |
| 172 | size_t len) |
| 173 | { |
| 174 | const struct dccp_sock *dp = dccp_sk(sk); |
| 175 | const int flags = msg->msg_flags; |
| 176 | const int noblock = flags & MSG_DONTWAIT; |
| 177 | struct sk_buff *skb; |
| 178 | int rc, size; |
| 179 | long timeo; |
| 180 | |
| 181 | if (len > dp->dccps_mss_cache) |
| 182 | return -EMSGSIZE; |
| 183 | |
| 184 | lock_sock(sk); |
| 185 | |
| 186 | timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT); |
| 187 | |
| 188 | /* |
| 189 | * We have to use sk_stream_wait_connect here to set sk_write_pending, |
| 190 | * so that the trick in dccp_rcv_request_sent_state_process. |
| 191 | */ |
| 192 | /* Wait for a connection to finish. */ |
| 193 | if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN | DCCPF_CLOSING)) |
| 194 | if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0) |
| 195 | goto out_err; |
| 196 | |
| 197 | size = sk->sk_prot->max_header + len; |
| 198 | release_sock(sk); |
| 199 | skb = sock_alloc_send_skb(sk, size, noblock, &rc); |
| 200 | lock_sock(sk); |
| 201 | |
| 202 | if (skb == NULL) |
| 203 | goto out_release; |
| 204 | |
| 205 | skb_reserve(skb, sk->sk_prot->max_header); |
| 206 | rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len); |
| 207 | if (rc == 0) { |
| 208 | struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); |
| 209 | const struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts; |
| 210 | long delay; |
| 211 | |
| 212 | /* |
| 213 | * XXX: This is just to match the Waikato tree CA interaction |
| 214 | * points, after the CCID3 code is stable and I have a better |
| 215 | * understanding of behaviour I'll change this to look more like |
| 216 | * TCP. |
| 217 | */ |
| 218 | while (1) { |
| 219 | rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, |
| 220 | skb, len, &delay); |
| 221 | if (rc == 0) |
| 222 | break; |
| 223 | if (rc != -EAGAIN) |
| 224 | goto out_discard; |
| 225 | if (delay > timeo) |
| 226 | goto out_discard; |
| 227 | release_sock(sk); |
| 228 | delay = schedule_timeout(delay); |
| 229 | lock_sock(sk); |
| 230 | timeo -= delay; |
| 231 | if (signal_pending(current)) |
| 232 | goto out_interrupted; |
| 233 | rc = -EPIPE; |
| 234 | if (!(sk->sk_state == DCCP_PARTOPEN || sk->sk_state == DCCP_OPEN)) |
| 235 | goto out_discard; |
| 236 | } |
| 237 | |
| 238 | if (sk->sk_state == DCCP_PARTOPEN) { |
| 239 | /* See 8.1.5. Handshake Completion */ |
| 240 | inet_csk_schedule_ack(sk); |
| 241 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, inet_csk(sk)->icsk_rto, TCP_RTO_MAX); |
| 242 | dcb->dccpd_type = DCCP_PKT_DATAACK; |
| 243 | /* FIXME: we really should have a dccps_ack_pending or use icsk */ |
| 244 | } else if (inet_csk_ack_scheduled(sk) || |
| 245 | (dp->dccps_options.dccpo_send_ack_vector && |
| 246 | ap->dccpap_buf_ackno != DCCP_MAX_SEQNO + 1 && |
| 247 | ap->dccpap_ack_seqno == DCCP_MAX_SEQNO + 1)) |
| 248 | dcb->dccpd_type = DCCP_PKT_DATAACK; |
| 249 | else |
| 250 | dcb->dccpd_type = DCCP_PKT_DATA; |
| 251 | dccp_transmit_skb(sk, skb); |
| 252 | ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len); |
| 253 | } else { |
| 254 | out_discard: |
| 255 | kfree_skb(skb); |
| 256 | } |
| 257 | out_release: |
| 258 | release_sock(sk); |
| 259 | return rc ? : len; |
| 260 | out_err: |
| 261 | rc = sk_stream_error(sk, flags, rc); |
| 262 | goto out_release; |
| 263 | out_interrupted: |
| 264 | rc = sock_intr_errno(timeo); |
| 265 | goto out_discard; |
| 266 | } |
| 267 | |
| 268 | EXPORT_SYMBOL(dccp_sendmsg); |
| 269 | |
| 270 | int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, |
| 271 | size_t len, int nonblock, int flags, int *addr_len) |
| 272 | { |
| 273 | const struct dccp_hdr *dh; |
| 274 | int copied = 0; |
| 275 | unsigned long used; |
| 276 | int err; |
| 277 | int target; /* Read at least this many bytes */ |
| 278 | long timeo; |
| 279 | |
| 280 | lock_sock(sk); |
| 281 | |
| 282 | err = -ENOTCONN; |
| 283 | if (sk->sk_state == DCCP_LISTEN) |
| 284 | goto out; |
| 285 | |
| 286 | timeo = sock_rcvtimeo(sk, nonblock); |
| 287 | |
| 288 | /* Urgent data needs to be handled specially. */ |
| 289 | if (flags & MSG_OOB) |
| 290 | goto recv_urg; |
| 291 | |
| 292 | /* FIXME */ |
| 293 | #if 0 |
| 294 | seq = &tp->copied_seq; |
| 295 | if (flags & MSG_PEEK) { |
| 296 | peek_seq = tp->copied_seq; |
| 297 | seq = &peek_seq; |
| 298 | } |
| 299 | #endif |
| 300 | |
| 301 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); |
| 302 | |
| 303 | do { |
| 304 | struct sk_buff *skb; |
| 305 | u32 offset; |
| 306 | |
| 307 | /* FIXME */ |
| 308 | #if 0 |
| 309 | /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */ |
| 310 | if (tp->urg_data && tp->urg_seq == *seq) { |
| 311 | if (copied) |
| 312 | break; |
| 313 | if (signal_pending(current)) { |
| 314 | copied = timeo ? sock_intr_errno(timeo) : -EAGAIN; |
| 315 | break; |
| 316 | } |
| 317 | } |
| 318 | #endif |
| 319 | |
| 320 | /* Next get a buffer. */ |
| 321 | |
| 322 | skb = skb_peek(&sk->sk_receive_queue); |
| 323 | do { |
| 324 | if (!skb) |
| 325 | break; |
| 326 | |
| 327 | offset = 0; |
| 328 | dh = dccp_hdr(skb); |
| 329 | |
| 330 | if (dh->dccph_type == DCCP_PKT_DATA || |
| 331 | dh->dccph_type == DCCP_PKT_DATAACK) |
| 332 | goto found_ok_skb; |
| 333 | |
| 334 | if (dh->dccph_type == DCCP_PKT_RESET || |
| 335 | dh->dccph_type == DCCP_PKT_CLOSE) { |
| 336 | dccp_pr_debug("found fin ok!\n"); |
| 337 | goto found_fin_ok; |
| 338 | } |
| 339 | dccp_pr_debug("packet_type=%s\n", dccp_packet_name(dh->dccph_type)); |
| 340 | BUG_TRAP(flags & MSG_PEEK); |
| 341 | skb = skb->next; |
| 342 | } while (skb != (struct sk_buff *)&sk->sk_receive_queue); |
| 343 | |
| 344 | /* Well, if we have backlog, try to process it now yet. */ |
| 345 | if (copied >= target && !sk->sk_backlog.tail) |
| 346 | break; |
| 347 | |
| 348 | if (copied) { |
| 349 | if (sk->sk_err || |
| 350 | sk->sk_state == DCCP_CLOSED || |
| 351 | (sk->sk_shutdown & RCV_SHUTDOWN) || |
| 352 | !timeo || |
| 353 | signal_pending(current) || |
| 354 | (flags & MSG_PEEK)) |
| 355 | break; |
| 356 | } else { |
| 357 | if (sock_flag(sk, SOCK_DONE)) |
| 358 | break; |
| 359 | |
| 360 | if (sk->sk_err) { |
| 361 | copied = sock_error(sk); |
| 362 | break; |
| 363 | } |
| 364 | |
| 365 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
| 366 | break; |
| 367 | |
| 368 | if (sk->sk_state == DCCP_CLOSED) { |
| 369 | if (!sock_flag(sk, SOCK_DONE)) { |
| 370 | /* This occurs when user tries to read |
| 371 | * from never connected socket. |
| 372 | */ |
| 373 | copied = -ENOTCONN; |
| 374 | break; |
| 375 | } |
| 376 | break; |
| 377 | } |
| 378 | |
| 379 | if (!timeo) { |
| 380 | copied = -EAGAIN; |
| 381 | break; |
| 382 | } |
| 383 | |
| 384 | if (signal_pending(current)) { |
| 385 | copied = sock_intr_errno(timeo); |
| 386 | break; |
| 387 | } |
| 388 | } |
| 389 | |
| 390 | /* FIXME: cleanup_rbuf(sk, copied); */ |
| 391 | |
| 392 | if (copied >= target) { |
| 393 | /* Do not sleep, just process backlog. */ |
| 394 | release_sock(sk); |
| 395 | lock_sock(sk); |
| 396 | } else |
| 397 | sk_wait_data(sk, &timeo); |
| 398 | |
| 399 | continue; |
| 400 | |
| 401 | found_ok_skb: |
| 402 | /* Ok so how much can we use? */ |
| 403 | used = skb->len - offset; |
| 404 | if (len < used) |
| 405 | used = len; |
| 406 | |
| 407 | if (!(flags & MSG_TRUNC)) { |
| 408 | err = skb_copy_datagram_iovec(skb, offset, |
| 409 | msg->msg_iov, used); |
| 410 | if (err) { |
| 411 | /* Exception. Bailout! */ |
| 412 | if (!copied) |
| 413 | copied = -EFAULT; |
| 414 | break; |
| 415 | } |
| 416 | } |
| 417 | |
| 418 | copied += used; |
| 419 | len -= used; |
| 420 | |
| 421 | /* FIXME: tcp_rcv_space_adjust(sk); */ |
| 422 | |
| 423 | //skip_copy: |
| 424 | if (used + offset < skb->len) |
| 425 | continue; |
| 426 | |
| 427 | if (!(flags & MSG_PEEK)) |
| 428 | sk_eat_skb(sk, skb); |
| 429 | continue; |
| 430 | found_fin_ok: |
| 431 | if (!(flags & MSG_PEEK)) |
| 432 | sk_eat_skb(sk, skb); |
| 433 | break; |
| 434 | |
| 435 | } while (len > 0); |
| 436 | |
| 437 | /* According to UNIX98, msg_name/msg_namelen are ignored |
| 438 | * on connected socket. I was just happy when found this 8) --ANK |
| 439 | */ |
| 440 | |
| 441 | /* Clean up data we have read: This will do ACK frames. */ |
| 442 | /* FIXME: cleanup_rbuf(sk, copied); */ |
| 443 | |
| 444 | release_sock(sk); |
| 445 | return copied; |
| 446 | |
| 447 | out: |
| 448 | release_sock(sk); |
| 449 | return err; |
| 450 | |
| 451 | recv_urg: |
| 452 | /* FIXME: err = tcp_recv_urg(sk, timeo, msg, len, flags, addr_len); */ |
| 453 | goto out; |
| 454 | } |
| 455 | |
| 456 | static int inet_dccp_listen(struct socket *sock, int backlog) |
| 457 | { |
| 458 | struct sock *sk = sock->sk; |
| 459 | unsigned char old_state; |
| 460 | int err; |
| 461 | |
| 462 | lock_sock(sk); |
| 463 | |
| 464 | err = -EINVAL; |
| 465 | if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP) |
| 466 | goto out; |
| 467 | |
| 468 | old_state = sk->sk_state; |
| 469 | if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN))) |
| 470 | goto out; |
| 471 | |
| 472 | /* Really, if the socket is already in listen state |
| 473 | * we can only allow the backlog to be adjusted. |
| 474 | */ |
| 475 | if (old_state != DCCP_LISTEN) { |
| 476 | /* |
| 477 | * FIXME: here it probably should be sk->sk_prot->listen_start |
| 478 | * see tcp_listen_start |
| 479 | */ |
| 480 | err = dccp_listen_start(sk); |
| 481 | if (err) |
| 482 | goto out; |
| 483 | } |
| 484 | sk->sk_max_ack_backlog = backlog; |
| 485 | err = 0; |
| 486 | |
| 487 | out: |
| 488 | release_sock(sk); |
| 489 | return err; |
| 490 | } |
| 491 | |
| 492 | static const unsigned char dccp_new_state[] = { |
| 493 | /* current state: new state: action: */ |
| 494 | [0] = DCCP_CLOSED, |
| 495 | [DCCP_OPEN] = DCCP_CLOSING | DCCP_ACTION_FIN, |
| 496 | [DCCP_REQUESTING] = DCCP_CLOSED, |
| 497 | [DCCP_PARTOPEN] = DCCP_CLOSING | DCCP_ACTION_FIN, |
| 498 | [DCCP_LISTEN] = DCCP_CLOSED, |
| 499 | [DCCP_RESPOND] = DCCP_CLOSED, |
| 500 | [DCCP_CLOSING] = DCCP_CLOSED, |
| 501 | [DCCP_TIME_WAIT] = DCCP_CLOSED, |
| 502 | [DCCP_CLOSED] = DCCP_CLOSED, |
| 503 | }; |
| 504 | |
| 505 | static int dccp_close_state(struct sock *sk) |
| 506 | { |
| 507 | const int next = dccp_new_state[sk->sk_state]; |
| 508 | const int ns = next & DCCP_STATE_MASK; |
| 509 | |
| 510 | if (ns != sk->sk_state) |
| 511 | dccp_set_state(sk, ns); |
| 512 | |
| 513 | return next & DCCP_ACTION_FIN; |
| 514 | } |
| 515 | |
| 516 | void dccp_close(struct sock *sk, long timeout) |
| 517 | { |
| 518 | struct sk_buff *skb; |
| 519 | |
| 520 | lock_sock(sk); |
| 521 | |
| 522 | sk->sk_shutdown = SHUTDOWN_MASK; |
| 523 | |
| 524 | if (sk->sk_state == DCCP_LISTEN) { |
| 525 | dccp_set_state(sk, DCCP_CLOSED); |
| 526 | |
| 527 | /* Special case. */ |
| 528 | inet_csk_listen_stop(sk); |
| 529 | |
| 530 | goto adjudge_to_death; |
| 531 | } |
| 532 | |
| 533 | /* |
| 534 | * We need to flush the recv. buffs. We do this only on the |
| 535 | * descriptor close, not protocol-sourced closes, because the |
| 536 | *reader process may not have drained the data yet! |
| 537 | */ |
| 538 | /* FIXME: check for unread data */ |
| 539 | while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) { |
| 540 | __kfree_skb(skb); |
| 541 | } |
| 542 | |
| 543 | if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) { |
| 544 | /* Check zero linger _after_ checking for unread data. */ |
| 545 | sk->sk_prot->disconnect(sk, 0); |
| 546 | } else if (dccp_close_state(sk)) { |
| 547 | dccp_send_close(sk); |
| 548 | } |
| 549 | |
| 550 | sk_stream_wait_close(sk, timeout); |
| 551 | |
| 552 | adjudge_to_death: |
| 553 | release_sock(sk); |
| 554 | /* |
| 555 | * Now socket is owned by kernel and we acquire BH lock |
| 556 | * to finish close. No need to check for user refs. |
| 557 | */ |
| 558 | local_bh_disable(); |
| 559 | bh_lock_sock(sk); |
| 560 | BUG_TRAP(!sock_owned_by_user(sk)); |
| 561 | |
| 562 | sock_hold(sk); |
| 563 | sock_orphan(sk); |
| 564 | |
| 565 | if (sk->sk_state != DCCP_CLOSED) |
| 566 | dccp_set_state(sk, DCCP_CLOSED); |
| 567 | |
| 568 | atomic_inc(&dccp_orphan_count); |
| 569 | if (sk->sk_state == DCCP_CLOSED) |
| 570 | inet_csk_destroy_sock(sk); |
| 571 | |
| 572 | /* Otherwise, socket is reprieved until protocol close. */ |
| 573 | |
| 574 | bh_unlock_sock(sk); |
| 575 | local_bh_enable(); |
| 576 | sock_put(sk); |
| 577 | } |
| 578 | |
| 579 | void dccp_shutdown(struct sock *sk, int how) |
| 580 | { |
| 581 | dccp_pr_debug("entry\n"); |
| 582 | } |
| 583 | |
| 584 | struct proto_ops inet_dccp_ops = { |
| 585 | .family = PF_INET, |
| 586 | .owner = THIS_MODULE, |
| 587 | .release = inet_release, |
| 588 | .bind = inet_bind, |
| 589 | .connect = inet_stream_connect, |
| 590 | .socketpair = sock_no_socketpair, |
| 591 | .accept = inet_accept, |
| 592 | .getname = inet_getname, |
| 593 | .poll = sock_no_poll, |
| 594 | .ioctl = inet_ioctl, |
| 595 | .listen = inet_dccp_listen, /* FIXME: work on inet_listen to rename it to sock_common_listen */ |
| 596 | .shutdown = inet_shutdown, |
| 597 | .setsockopt = sock_common_setsockopt, |
| 598 | .getsockopt = sock_common_getsockopt, |
| 599 | .sendmsg = inet_sendmsg, |
| 600 | .recvmsg = sock_common_recvmsg, |
| 601 | .mmap = sock_no_mmap, |
| 602 | .sendpage = sock_no_sendpage, |
| 603 | }; |
| 604 | |
| 605 | extern struct net_proto_family inet_family_ops; |
| 606 | |
| 607 | static struct inet_protosw dccp_v4_protosw = { |
| 608 | .type = SOCK_DCCP, |
| 609 | .protocol = IPPROTO_DCCP, |
| 610 | .prot = &dccp_v4_prot, |
| 611 | .ops = &inet_dccp_ops, |
| 612 | .capability = -1, |
| 613 | .no_check = 0, |
| 614 | .flags = 0, |
| 615 | }; |
| 616 | |
| 617 | /* |
| 618 | * This is the global socket data structure used for responding to |
| 619 | * the Out-of-the-blue (OOTB) packets. A control sock will be created |
| 620 | * for this socket at the initialization time. |
| 621 | */ |
| 622 | struct socket *dccp_ctl_socket; |
| 623 | |
| 624 | static char dccp_ctl_socket_err_msg[] __initdata = |
| 625 | KERN_ERR "DCCP: Failed to create the control socket.\n"; |
| 626 | |
| 627 | static int __init dccp_ctl_sock_init(void) |
| 628 | { |
| 629 | int rc = sock_create_kern(PF_INET, SOCK_DCCP, IPPROTO_DCCP, |
| 630 | &dccp_ctl_socket); |
| 631 | if (rc < 0) |
| 632 | printk(dccp_ctl_socket_err_msg); |
| 633 | else { |
| 634 | dccp_ctl_socket->sk->sk_allocation = GFP_ATOMIC; |
| 635 | inet_sk(dccp_ctl_socket->sk)->uc_ttl = -1; |
| 636 | |
| 637 | /* Unhash it so that IP input processing does not even |
| 638 | * see it, we do not wish this socket to see incoming |
| 639 | * packets. |
| 640 | */ |
| 641 | dccp_ctl_socket->sk->sk_prot->unhash(dccp_ctl_socket->sk); |
| 642 | } |
| 643 | |
| 644 | return rc; |
| 645 | } |
| 646 | |
| 647 | static void __exit dccp_ctl_sock_exit(void) |
| 648 | { |
| 649 | if (dccp_ctl_socket != NULL) |
| 650 | sock_release(dccp_ctl_socket); |
| 651 | } |
| 652 | |
| 653 | static int __init init_dccp_v4_mibs(void) |
| 654 | { |
| 655 | int rc = -ENOMEM; |
| 656 | |
| 657 | dccp_statistics[0] = alloc_percpu(struct dccp_mib); |
| 658 | if (dccp_statistics[0] == NULL) |
| 659 | goto out; |
| 660 | |
| 661 | dccp_statistics[1] = alloc_percpu(struct dccp_mib); |
| 662 | if (dccp_statistics[1] == NULL) |
| 663 | goto out_free_one; |
| 664 | |
| 665 | rc = 0; |
| 666 | out: |
| 667 | return rc; |
| 668 | out_free_one: |
| 669 | free_percpu(dccp_statistics[0]); |
| 670 | dccp_statistics[0] = NULL; |
| 671 | goto out; |
| 672 | |
| 673 | } |
| 674 | |
| 675 | static int thash_entries; |
| 676 | module_param(thash_entries, int, 0444); |
| 677 | MODULE_PARM_DESC(thash_entries, "Number of ehash buckets"); |
| 678 | |
| 679 | int dccp_debug; |
| 680 | module_param(dccp_debug, int, 0444); |
| 681 | MODULE_PARM_DESC(dccp_debug, "Enable debug messages"); |
| 682 | |
| 683 | static int __init dccp_init(void) |
| 684 | { |
| 685 | unsigned long goal; |
| 686 | int ehash_order, bhash_order, i; |
| 687 | int rc = proto_register(&dccp_v4_prot, 1); |
| 688 | |
| 689 | if (rc) |
| 690 | goto out; |
| 691 | |
| 692 | dccp_hashinfo.bind_bucket_cachep = kmem_cache_create("dccp_bind_bucket", |
| 693 | sizeof(struct inet_bind_bucket), |
| 694 | 0, SLAB_HWCACHE_ALIGN, |
| 695 | NULL, NULL); |
| 696 | if (!dccp_hashinfo.bind_bucket_cachep) |
| 697 | goto out_proto_unregister; |
| 698 | |
| 699 | /* |
| 700 | * Size and allocate the main established and bind bucket |
| 701 | * hash tables. |
| 702 | * |
| 703 | * The methodology is similar to that of the buffer cache. |
| 704 | */ |
| 705 | if (num_physpages >= (128 * 1024)) |
| 706 | goal = num_physpages >> (21 - PAGE_SHIFT); |
| 707 | else |
| 708 | goal = num_physpages >> (23 - PAGE_SHIFT); |
| 709 | |
| 710 | if (thash_entries) |
| 711 | goal = (thash_entries * sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT; |
| 712 | for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++) |
| 713 | ; |
| 714 | do { |
| 715 | dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE / |
| 716 | sizeof(struct inet_ehash_bucket); |
| 717 | dccp_hashinfo.ehash_size >>= 1; |
| 718 | while (dccp_hashinfo.ehash_size & (dccp_hashinfo.ehash_size - 1)) |
| 719 | dccp_hashinfo.ehash_size--; |
| 720 | dccp_hashinfo.ehash = (struct inet_ehash_bucket *) |
| 721 | __get_free_pages(GFP_ATOMIC, ehash_order); |
| 722 | } while (!dccp_hashinfo.ehash && --ehash_order > 0); |
| 723 | |
| 724 | if (!dccp_hashinfo.ehash) { |
| 725 | printk(KERN_CRIT "Failed to allocate DCCP " |
| 726 | "established hash table\n"); |
| 727 | goto out_free_bind_bucket_cachep; |
| 728 | } |
| 729 | |
| 730 | for (i = 0; i < (dccp_hashinfo.ehash_size << 1); i++) { |
| 731 | rwlock_init(&dccp_hashinfo.ehash[i].lock); |
| 732 | INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].chain); |
| 733 | } |
| 734 | |
| 735 | bhash_order = ehash_order; |
| 736 | |
| 737 | do { |
| 738 | dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE / |
| 739 | sizeof(struct inet_bind_hashbucket); |
| 740 | if ((dccp_hashinfo.bhash_size > (64 * 1024)) && bhash_order > 0) |
| 741 | continue; |
| 742 | dccp_hashinfo.bhash = (struct inet_bind_hashbucket *) |
| 743 | __get_free_pages(GFP_ATOMIC, bhash_order); |
| 744 | } while (!dccp_hashinfo.bhash && --bhash_order >= 0); |
| 745 | |
| 746 | if (!dccp_hashinfo.bhash) { |
| 747 | printk(KERN_CRIT "Failed to allocate DCCP bind hash table\n"); |
| 748 | goto out_free_dccp_ehash; |
| 749 | } |
| 750 | |
| 751 | for (i = 0; i < dccp_hashinfo.bhash_size; i++) { |
| 752 | spin_lock_init(&dccp_hashinfo.bhash[i].lock); |
| 753 | INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain); |
| 754 | } |
| 755 | |
| 756 | if (init_dccp_v4_mibs()) |
| 757 | goto out_free_dccp_bhash; |
| 758 | |
| 759 | rc = -EAGAIN; |
| 760 | if (inet_add_protocol(&dccp_protocol, IPPROTO_DCCP)) |
| 761 | goto out_free_dccp_v4_mibs; |
| 762 | |
| 763 | inet_register_protosw(&dccp_v4_protosw); |
| 764 | |
| 765 | rc = dccp_ctl_sock_init(); |
| 766 | if (rc) |
| 767 | goto out_unregister_protosw; |
| 768 | out: |
| 769 | return rc; |
| 770 | out_unregister_protosw: |
| 771 | inet_unregister_protosw(&dccp_v4_protosw); |
| 772 | inet_del_protocol(&dccp_protocol, IPPROTO_DCCP); |
| 773 | out_free_dccp_v4_mibs: |
| 774 | free_percpu(dccp_statistics[0]); |
| 775 | free_percpu(dccp_statistics[1]); |
| 776 | dccp_statistics[0] = dccp_statistics[1] = NULL; |
| 777 | out_free_dccp_bhash: |
| 778 | free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order); |
| 779 | dccp_hashinfo.bhash = NULL; |
| 780 | out_free_dccp_ehash: |
| 781 | free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order); |
| 782 | dccp_hashinfo.ehash = NULL; |
| 783 | out_free_bind_bucket_cachep: |
| 784 | kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); |
| 785 | dccp_hashinfo.bind_bucket_cachep = NULL; |
| 786 | out_proto_unregister: |
| 787 | proto_unregister(&dccp_v4_prot); |
| 788 | goto out; |
| 789 | } |
| 790 | |
| 791 | static const char dccp_del_proto_err_msg[] __exitdata = |
| 792 | KERN_ERR "can't remove dccp net_protocol\n"; |
| 793 | |
| 794 | static void __exit dccp_fini(void) |
| 795 | { |
| 796 | dccp_ctl_sock_exit(); |
| 797 | |
| 798 | inet_unregister_protosw(&dccp_v4_protosw); |
| 799 | |
| 800 | if (inet_del_protocol(&dccp_protocol, IPPROTO_DCCP) < 0) |
| 801 | printk(dccp_del_proto_err_msg); |
| 802 | |
| 803 | /* Free the control endpoint. */ |
| 804 | sock_release(dccp_ctl_socket); |
| 805 | |
| 806 | proto_unregister(&dccp_v4_prot); |
| 807 | |
| 808 | kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep); |
| 809 | } |
| 810 | |
| 811 | module_init(dccp_init); |
| 812 | module_exit(dccp_fini); |
| 813 | |
Arnaldo Carvalho de Melo | bb97d31 | 2005-08-09 20:19:14 -0700 | [diff] [blame^] | 814 | /* |
| 815 | * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33) |
| 816 | * values directly, Also cover the case where the protocol is not specified, |
| 817 | * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP |
| 818 | */ |
| 819 | MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6"); |
| 820 | MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6"); |
Arnaldo Carvalho de Melo | 7c65787 | 2005-08-09 20:14:34 -0700 | [diff] [blame] | 821 | MODULE_LICENSE("GPL"); |
| 822 | MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>"); |
| 823 | MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol"); |