net/dccp/proto.c

   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_sock.h>
  27 #include <net/sock.h>
  28 #include <net/xfrm.h>
  29
  30 #include <asm/semaphore.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/timer.h>
  33 #include <linux/delay.h>
  34 #include <linux/poll.h>
  35
  36 #include "ccid.h"
  37 #include "dccp.h"
  38 #include "feat.h"
  39
  40 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
  41
  42 EXPORT_SYMBOL_GPL(dccp_statistics);
  43
  44 atomic_t dccp_orphan_count = ATOMIC_INIT(0);
  45
  46 EXPORT_SYMBOL_GPL(dccp_orphan_count);
  47
  48 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  49         .lhash_lock     = RW_LOCK_UNLOCKED,
  50         .lhash_users    = ATOMIC_INIT(0),
  51         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  52 };
  53
  54 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  55
  56 void dccp_set_state(struct sock *sk, const int state)
  57 {
  58         const int oldstate = sk->sk_state;
  59
  60         dccp_pr_debug("%s(%p) %-10.10s -> %s\n",
  61                       dccp_role(sk), sk,
  62                       dccp_state_name(oldstate), dccp_state_name(state));
  63         WARN_ON(state == oldstate);
  64
  65         switch (state) {
  66         case DCCP_OPEN:
  67                 if (oldstate != DCCP_OPEN)
  68                         DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
  69                 break;
  70
  71         case DCCP_CLOSED:
  72                 if (oldstate == DCCP_CLOSING || oldstate == DCCP_OPEN)
  73                         DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
  74
  75                 sk->sk_prot->unhash(sk);
  76                 if (inet_csk(sk)->icsk_bind_hash != NULL &&
  77                     !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
  78                         inet_put_port(&dccp_hashinfo, sk);
  79                 /* fall through */
  80         default:
  81                 if (oldstate == DCCP_OPEN)
  82                         DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
  83         }
  84
  85         /* Change state AFTER socket is unhashed to avoid closed
  86          * socket sitting in hash tables.
  87          */
  88         sk->sk_state = state;
  89 }
  90
  91 EXPORT_SYMBOL_GPL(dccp_set_state);
  92
  93 void dccp_done(struct sock *sk)
  94 {
  95         dccp_set_state(sk, DCCP_CLOSED);
  96         dccp_clear_xmit_timers(sk);
  97
  98         sk->sk_shutdown = SHUTDOWN_MASK;
  99
 100         if (!sock_flag(sk, SOCK_DEAD))
 101                 sk->sk_state_change(sk);
 102         else
 103                 inet_csk_destroy_sock(sk);
 104 }
 105
 106 EXPORT_SYMBOL_GPL(dccp_done);
 107
 108 const char *dccp_packet_name(const int type)
 109 {
 110         static const char *dccp_packet_names[] = {
 111                 [DCCP_PKT_REQUEST]  = "REQUEST",
 112                 [DCCP_PKT_RESPONSE] = "RESPONSE",
 113                 [DCCP_PKT_DATA]     = "DATA",
 114                 [DCCP_PKT_ACK]      = "ACK",
 115                 [DCCP_PKT_DATAACK]  = "DATAACK",
 116                 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
 117                 [DCCP_PKT_CLOSE]    = "CLOSE",
 118                 [DCCP_PKT_RESET]    = "RESET",
 119                 [DCCP_PKT_SYNC]     = "SYNC",
 120                 [DCCP_PKT_SYNCACK]  = "SYNCACK",
 121         };
 122
 123         if (type >= DCCP_NR_PKT_TYPES)
 124                 return "INVALID";
 125         else
 126                 return dccp_packet_names[type];
 127 }
 128
 129 EXPORT_SYMBOL_GPL(dccp_packet_name);
 130
 131 const char *dccp_state_name(const int state)
 132 {
 133         static char *dccp_state_names[] = {
 134         [DCCP_OPEN]       = "OPEN",
 135         [DCCP_REQUESTING] = "REQUESTING",
 136         [DCCP_PARTOPEN]   = "PARTOPEN",
 137         [DCCP_LISTEN]     = "LISTEN",
 138         [DCCP_RESPOND]    = "RESPOND",
 139         [DCCP_CLOSING]    = "CLOSING",
 140         [DCCP_TIME_WAIT]  = "TIME_WAIT",
 141         [DCCP_CLOSED]     = "CLOSED",
 142         };
 143
 144         if (state >= DCCP_MAX_STATES)
 145                 return "INVALID STATE!";
 146         else
 147                 return dccp_state_names[state];
 148 }
 149
 150 EXPORT_SYMBOL_GPL(dccp_state_name);
 151
 152 void dccp_hash(struct sock *sk)
 153 {
 154         inet_hash(&dccp_hashinfo, sk);
 155 }
 156
 157 EXPORT_SYMBOL_GPL(dccp_hash);
 158
 159 void dccp_unhash(struct sock *sk)
 160 {
 161         inet_unhash(&dccp_hashinfo, sk);
 162 }
 163
 164 EXPORT_SYMBOL_GPL(dccp_unhash);
 165
 166 int dccp_init_sock(struct sock *sk)
 167 {
 168         struct dccp_sock *dp = dccp_sk(sk);
 169         struct inet_connection_sock *icsk = inet_csk(sk);
 170         static int dccp_ctl_socket_init = 1;
 171
 172         dccp_options_init(&dp->dccps_options);
 173         do_gettimeofday(&dp->dccps_epoch);
 174
 175         /*
 176          * FIXME: We're hardcoding the CCID, and doing this at this point makes
 177          * the listening (master) sock get CCID control blocks, which is not
 178          * necessary, but for now, to not mess with the test userspace apps,
 179          * lets leave it here, later the real solution is to do this in a
 180          * setsockopt(CCIDs-I-want/accept). -acme
 181          */
 182         if (likely(!dccp_ctl_socket_init)) {
 183                 int rc = dccp_feat_init(sk);
 184
 185                 if (rc)
 186                         return rc;
 187
 188                 if (dp->dccps_options.dccpo_send_ack_vector) {
 189                         dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(GFP_KERNEL);
 190                         if (dp->dccps_hc_rx_ackvec == NULL)
 191                                 return -ENOMEM;
 192                 }
 193                 dp->dccps_hc_rx_ccid =
 194                                 ccid_hc_rx_new(dp->dccps_options.dccpo_rx_ccid,
 195                                                sk, GFP_KERNEL);
 196                 dp->dccps_hc_tx_ccid =
 197                                 ccid_hc_tx_new(dp->dccps_options.dccpo_tx_ccid,
 198                                                sk, GFP_KERNEL);
 199                 if (unlikely(dp->dccps_hc_rx_ccid == NULL ||
 200                              dp->dccps_hc_tx_ccid == NULL)) {
 201                         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
 202                         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
 203                         if (dp->dccps_options.dccpo_send_ack_vector) {
 204                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
 205                                 dp->dccps_hc_rx_ackvec = NULL;
 206                         }
 207                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
 208                         return -ENOMEM;
 209                 }
 210         } else {
 211                 /* control socket doesn't need feat nego */
 212                 INIT_LIST_HEAD(&dp->dccps_options.dccpo_pending);
 213                 INIT_LIST_HEAD(&dp->dccps_options.dccpo_conf);
 214                 dccp_ctl_socket_init = 0;
 215         }
 216
 217         dccp_init_xmit_timers(sk);
 218         icsk->icsk_rto          = DCCP_TIMEOUT_INIT;
 219         sk->sk_state            = DCCP_CLOSED;
 220         sk->sk_write_space      = dccp_write_space;
 221         icsk->icsk_sync_mss     = dccp_sync_mss;
 222         dp->dccps_mss_cache     = 536;
 223         dp->dccps_role          = DCCP_ROLE_UNDEFINED;
 224         dp->dccps_service       = DCCP_SERVICE_INVALID_VALUE;
 225         dp->dccps_l_ack_ratio   = dp->dccps_r_ack_ratio = 1;
 226
 227         return 0;
 228 }
 229
 230 EXPORT_SYMBOL_GPL(dccp_init_sock);
 231
 232 int dccp_destroy_sock(struct sock *sk)
 233 {
 234         struct dccp_sock *dp = dccp_sk(sk);
 235
 236         /*
 237          * DCCP doesn't use sk_write_queue, just sk_send_head
 238          * for retransmissions
 239          */
 240         if (sk->sk_send_head != NULL) {
 241                 kfree_skb(sk->sk_send_head);
 242                 sk->sk_send_head = NULL;
 243         }
 244
 245         /* Clean up a referenced DCCP bind bucket. */
 246         if (inet_csk(sk)->icsk_bind_hash != NULL)
 247                 inet_put_port(&dccp_hashinfo, sk);
 248
 249         kfree(dp->dccps_service_list);
 250         dp->dccps_service_list = NULL;
 251
 252         if (dp->dccps_options.dccpo_send_ack_vector) {
 253                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
 254                 dp->dccps_hc_rx_ackvec = NULL;
 255         }
 256         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
 257         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
 258         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
 259
 260         /* clean up feature negotiation state */
 261         dccp_feat_clean(sk);
 262
 263         return 0;
 264 }
 265
 266 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
 267
 268 static inline int dccp_listen_start(struct sock *sk)
 269 {
 270         struct dccp_sock *dp = dccp_sk(sk);
 271
 272         dp->dccps_role = DCCP_ROLE_LISTEN;
 273         /*
 274          * Apps need to use setsockopt(DCCP_SOCKOPT_SERVICE)
 275          * before calling listen()
 276          */
 277         if (dccp_service_not_initialized(sk))
 278                 return -EPROTO;
 279         return inet_csk_listen_start(sk, TCP_SYNQ_HSIZE);
 280 }
 281
 282 int dccp_disconnect(struct sock *sk, int flags)
 283 {
 284         struct inet_connection_sock *icsk = inet_csk(sk);
 285         struct inet_sock *inet = inet_sk(sk);
 286         int err = 0;
 287         const int old_state = sk->sk_state;
 288
 289         if (old_state != DCCP_CLOSED)
 290                 dccp_set_state(sk, DCCP_CLOSED);
 291
 292         /* ABORT function of RFC793 */
 293         if (old_state == DCCP_LISTEN) {
 294                 inet_csk_listen_stop(sk);
 295         /* FIXME: do the active reset thing */
 296         } else if (old_state == DCCP_REQUESTING)
 297                 sk->sk_err = ECONNRESET;
 298
 299         dccp_clear_xmit_timers(sk);
 300         __skb_queue_purge(&sk->sk_receive_queue);
 301         if (sk->sk_send_head != NULL) {
 302                 __kfree_skb(sk->sk_send_head);
 303                 sk->sk_send_head = NULL;
 304         }
 305
 306         inet->dport = 0;
 307
 308         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
 309                 inet_reset_saddr(sk);
 310
 311         sk->sk_shutdown = 0;
 312         sock_reset_flag(sk, SOCK_DONE);
 313
 314         icsk->icsk_backoff = 0;
 315         inet_csk_delack_init(sk);
 316         __sk_dst_reset(sk);
 317
 318         BUG_TRAP(!inet->num || icsk->icsk_bind_hash);
 319
 320         sk->sk_error_report(sk);
 321         return err;
 322 }
 323
 324 EXPORT_SYMBOL_GPL(dccp_disconnect);
 325
 326 /*
 327  *      Wait for a DCCP event.
 328  *
 329  *      Note that we don't need to lock the socket, as the upper poll layers
 330  *      take care of normal races (between the test and the event) and we don't
 331  *      go look at any of the socket buffers directly.
 332  */
 333 unsigned int dccp_poll(struct file *file, struct socket *sock,
 334                        poll_table *wait)
 335 {
 336         unsigned int mask;
 337         struct sock *sk = sock->sk;
 338
 339         poll_wait(file, sk->sk_sleep, wait);
 340         if (sk->sk_state == DCCP_LISTEN)
 341                 return inet_csk_listen_poll(sk);
 342
 343         /* Socket is not locked. We are protected from async events
 344            by poll logic and correct handling of state changes
 345            made by another threads is impossible in any case.
 346          */
 347
 348         mask = 0;
 349         if (sk->sk_err)
 350                 mask = POLLERR;
 351
 352         if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
 353                 mask |= POLLHUP;
 354         if (sk->sk_shutdown & RCV_SHUTDOWN)
 355                 mask |= POLLIN | POLLRDNORM;
 356
 357         /* Connected? */
 358         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
 359                 if (atomic_read(&sk->sk_rmem_alloc) > 0)
 360                         mask |= POLLIN | POLLRDNORM;
 361
 362                 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 363                         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 364                                 mask |= POLLOUT | POLLWRNORM;
 365                         } else {  /* send SIGIO later */
 366                                 set_bit(SOCK_ASYNC_NOSPACE,
 367                                         &sk->sk_socket->flags);
 368                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 369
 370                                 /* Race breaker. If space is freed after
 371                                  * wspace test but before the flags are set,
 372                                  * IO signal will be lost.
 373                                  */
 374                                 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
 375                                         mask |= POLLOUT | POLLWRNORM;
 376                         }
 377                 }
 378         }
 379         return mask;
 380 }
 381
 382 EXPORT_SYMBOL_GPL(dccp_poll);
 383
 384 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
 385 {
 386         dccp_pr_debug("entry\n");
 387         return -ENOIOCTLCMD;
 388 }
 389
 390 EXPORT_SYMBOL_GPL(dccp_ioctl);
 391
 392 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
 393                                    char __user *optval, int optlen)
 394 {
 395         struct dccp_sock *dp = dccp_sk(sk);
 396         struct dccp_service_list *sl = NULL;
 397
 398         if (service == DCCP_SERVICE_INVALID_VALUE ||
 399             optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
 400                 return -EINVAL;
 401
 402         if (optlen > sizeof(service)) {
 403                 sl = kmalloc(optlen, GFP_KERNEL);
 404                 if (sl == NULL)
 405                         return -ENOMEM;
 406
 407                 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
 408                 if (copy_from_user(sl->dccpsl_list,
 409                                    optval + sizeof(service),
 410                                    optlen - sizeof(service)) ||
 411                     dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
 412                         kfree(sl);
 413                         return -EFAULT;
 414                 }
 415         }
 416
 417         lock_sock(sk);
 418         dp->dccps_service = service;
 419
 420         kfree(dp->dccps_service_list);
 421
 422         dp->dccps_service_list = sl;
 423         release_sock(sk);
 424         return 0;
 425 }
 426
 427 /* byte 1 is feature.  the rest is the preference list */
 428 static int dccp_setsockopt_change(struct sock *sk, int type,
 429                                   struct dccp_so_feat __user *optval)
 430 {
 431         struct dccp_so_feat opt;
 432         u8 *val;
 433         int rc;
 434
 435         if (copy_from_user(&opt, optval, sizeof(opt)))
 436                 return -EFAULT;
 437
 438         val = kmalloc(opt.dccpsf_len, GFP_KERNEL);
 439         if (!val)
 440                 return -ENOMEM;
 441
 442         if (copy_from_user(val, opt.dccpsf_val, opt.dccpsf_len)) {
 443                 rc = -EFAULT;
 444                 goto out_free_val;
 445         }
 446
 447         rc = dccp_feat_change(sk, type, opt.dccpsf_feat, val, opt.dccpsf_len,
 448                               GFP_KERNEL);
 449         if (rc)
 450                 goto out_free_val;
 451
 452 out:
 453         return rc;
 454
 455 out_free_val:
 456         kfree(val);
 457         goto out;
 458 }
 459
 460 int dccp_setsockopt(struct sock *sk, int level, int optname,
 461                     char __user *optval, int optlen)
 462 {
 463         struct dccp_sock *dp;
 464         int err;
 465         int val;
 466
 467         if (level != SOL_DCCP)
 468                 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
 469                                                              optname, optval,
 470                                                              optlen);
 471
 472         if (optlen < sizeof(int))
 473                 return -EINVAL;
 474
 475         if (get_user(val, (int __user *)optval))
 476                 return -EFAULT;
 477
 478         if (optname == DCCP_SOCKOPT_SERVICE)
 479                 return dccp_setsockopt_service(sk, val, optval, optlen);
 480
 481         lock_sock(sk);
 482         dp = dccp_sk(sk);
 483         err = 0;
 484
 485         switch (optname) {
 486         case DCCP_SOCKOPT_PACKET_SIZE:
 487                 dp->dccps_packet_size = val;
 488                 break;
 489
 490         case DCCP_SOCKOPT_CHANGE_L:
 491                 if (optlen != sizeof(struct dccp_so_feat))
 492                         err = -EINVAL;
 493                 else
 494                         err = dccp_setsockopt_change(sk, DCCPO_CHANGE_L,
 495                                                      (struct dccp_so_feat *)
 496                                                      optval);
 497                 break;
 498
 499         case DCCP_SOCKOPT_CHANGE_R:
 500                 if (optlen != sizeof(struct dccp_so_feat))
 501                         err = -EINVAL;
 502                 else
 503                         err = dccp_setsockopt_change(sk, DCCPO_CHANGE_R,
 504                                                      (struct dccp_so_feat *)
 505                                                      optval);
 506                 break;
 507
 508         default:
 509                 err = -ENOPROTOOPT;
 510                 break;
 511         }
 512
 513         release_sock(sk);
 514         return err;
 515 }
 516
 517 EXPORT_SYMBOL_GPL(dccp_setsockopt);
 518
 519 static int dccp_getsockopt_service(struct sock *sk, int len,
 520                                    __be32 __user *optval,
 521                                    int __user *optlen)
 522 {
 523         const struct dccp_sock *dp = dccp_sk(sk);
 524         const struct dccp_service_list *sl;
 525         int err = -ENOENT, slen = 0, total_len = sizeof(u32);
 526
 527         lock_sock(sk);
 528         if (dccp_service_not_initialized(sk))
 529                 goto out;
 530
 531         if ((sl = dp->dccps_service_list) != NULL) {
 532                 slen = sl->dccpsl_nr * sizeof(u32);
 533                 total_len += slen;
 534         }
 535
 536         err = -EINVAL;
 537         if (total_len > len)
 538                 goto out;
 539
 540         err = 0;
 541         if (put_user(total_len, optlen) ||
 542             put_user(dp->dccps_service, optval) ||
 543             (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
 544                 err = -EFAULT;
 545 out:
 546         release_sock(sk);
 547         return err;
 548 }
 549
 550 int dccp_getsockopt(struct sock *sk, int level, int optname,
 551                     char __user *optval, int __user *optlen)
 552 {
 553         struct dccp_sock *dp;
 554         int val, len;
 555
 556         if (level != SOL_DCCP)
 557                 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
 558                                                              optname, optval,
 559                                                              optlen);
 560         if (get_user(len, optlen))
 561                 return -EFAULT;
 562
 563         if (len < sizeof(int))
 564                 return -EINVAL;
 565
 566         dp = dccp_sk(sk);
 567
 568         switch (optname) {
 569         case DCCP_SOCKOPT_PACKET_SIZE:
 570                 val = dp->dccps_packet_size;
 571                 len = sizeof(dp->dccps_packet_size);
 572                 break;
 573         case DCCP_SOCKOPT_SERVICE:
 574                 return dccp_getsockopt_service(sk, len,
 575                                                (__be32 __user *)optval, optlen);
 576         case 128 ... 191:
 577                 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
 578                                              len, (u32 __user *)optval, optlen);
 579         case 192 ... 255:
 580                 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
 581                                              len, (u32 __user *)optval, optlen);
 582         default:
 583                 return -ENOPROTOOPT;
 584         }
 585
 586         if (put_user(len, optlen) || copy_to_user(optval, &val, len))
 587                 return -EFAULT;
 588
 589         return 0;
 590 }
 591
 592 EXPORT_SYMBOL_GPL(dccp_getsockopt);
 593
 594 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 595                  size_t len)
 596 {
 597         const struct dccp_sock *dp = dccp_sk(sk);
 598         const int flags = msg->msg_flags;
 599         const int noblock = flags & MSG_DONTWAIT;
 600         struct sk_buff *skb;
 601         int rc, size;
 602         long timeo;
 603
 604         if (len > dp->dccps_mss_cache)
 605                 return -EMSGSIZE;
 606
 607         lock_sock(sk);
 608         timeo = sock_sndtimeo(sk, noblock);
 609
 610         /*
 611          * We have to use sk_stream_wait_connect here to set sk_write_pending,
 612          * so that the trick in dccp_rcv_request_sent_state_process.
 613          */
 614         /* Wait for a connection to finish. */
 615         if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN | DCCPF_CLOSING))
 616                 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
 617                         goto out_release;
 618
 619         size = sk->sk_prot->max_header + len;
 620         release_sock(sk);
 621         skb = sock_alloc_send_skb(sk, size, noblock, &rc);
 622         lock_sock(sk);
 623         if (skb == NULL)
 624                 goto out_release;
 625
 626         skb_reserve(skb, sk->sk_prot->max_header);
 627         rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
 628         if (rc != 0)
 629                 goto out_discard;
 630
 631         rc = dccp_write_xmit(sk, skb, &timeo);
 632         /*
 633          * XXX we don't use sk_write_queue, so just discard the packet.
 634          *     Current plan however is to _use_ sk_write_queue with
 635          *     an algorith similar to tcp_sendmsg, where the main difference
 636          *     is that in DCCP we have to respect packet boundaries, so
 637          *     no coalescing of skbs.
 638          *
 639          *     This bug was _quickly_ found & fixed by just looking at an OSTRA
 640          *     generated callgraph 8) -acme
 641          */
 642 out_release:
 643         release_sock(sk);
 644         return rc ? : len;
 645 out_discard:
 646         kfree_skb(skb);
 647         goto out_release;
 648 }
 649
 650 EXPORT_SYMBOL_GPL(dccp_sendmsg);
 651
 652 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 653                  size_t len, int nonblock, int flags, int *addr_len)
 654 {
 655         const struct dccp_hdr *dh;
 656         long timeo;
 657
 658         lock_sock(sk);
 659
 660         if (sk->sk_state == DCCP_LISTEN) {
 661                 len = -ENOTCONN;
 662                 goto out;
 663         }
 664
 665         timeo = sock_rcvtimeo(sk, nonblock);
 666
 667         do {
 668                 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
 669
 670                 if (skb == NULL)
 671                         goto verify_sock_status;
 672
 673                 dh = dccp_hdr(skb);
 674
 675                 if (dh->dccph_type == DCCP_PKT_DATA ||
 676                     dh->dccph_type == DCCP_PKT_DATAACK)
 677                         goto found_ok_skb;
 678
 679                 if (dh->dccph_type == DCCP_PKT_RESET ||
 680                     dh->dccph_type == DCCP_PKT_CLOSE) {
 681                         dccp_pr_debug("found fin ok!\n");
 682                         len = 0;
 683                         goto found_fin_ok;
 684                 }
 685                 dccp_pr_debug("packet_type=%s\n",
 686                               dccp_packet_name(dh->dccph_type));
 687                 sk_eat_skb(sk, skb);
 688 verify_sock_status:
 689                 if (sock_flag(sk, SOCK_DONE)) {
 690                         len = 0;
 691                         break;
 692                 }
 693
 694                 if (sk->sk_err) {
 695                         len = sock_error(sk);
 696                         break;
 697                 }
 698
 699                 if (sk->sk_shutdown & RCV_SHUTDOWN) {
 700                         len = 0;
 701                         break;
 702                 }
 703
 704                 if (sk->sk_state == DCCP_CLOSED) {
 705                         if (!sock_flag(sk, SOCK_DONE)) {
 706                                 /* This occurs when user tries to read
 707                                  * from never connected socket.
 708                                  */
 709                                 len = -ENOTCONN;
 710                                 break;
 711                         }
 712                         len = 0;
 713                         break;
 714                 }
 715
 716                 if (!timeo) {
 717                         len = -EAGAIN;
 718                         break;
 719                 }
 720
 721                 if (signal_pending(current)) {
 722                         len = sock_intr_errno(timeo);
 723                         break;
 724                 }
 725
 726                 sk_wait_data(sk, &timeo);
 727                 continue;
 728         found_ok_skb:
 729                 if (len > skb->len)
 730                         len = skb->len;
 731                 else if (len < skb->len)
 732                         msg->msg_flags |= MSG_TRUNC;
 733
 734                 if (skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len)) {
 735                         /* Exception. Bailout! */
 736                         len = -EFAULT;
 737                         break;
 738                 }
 739         found_fin_ok:
 740                 if (!(flags & MSG_PEEK))
 741                         sk_eat_skb(sk, skb);
 742                 break;
 743         } while (1);
 744 out:
 745         release_sock(sk);
 746         return len;
 747 }
 748
 749 EXPORT_SYMBOL_GPL(dccp_recvmsg);
 750
 751 int inet_dccp_listen(struct socket *sock, int backlog)
 752 {
 753         struct sock *sk = sock->sk;
 754         unsigned char old_state;
 755         int err;
 756
 757         lock_sock(sk);
 758
 759         err = -EINVAL;
 760         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
 761                 goto out;
 762
 763         old_state = sk->sk_state;
 764         if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
 765                 goto out;
 766
 767         /* Really, if the socket is already in listen state
 768          * we can only allow the backlog to be adjusted.
 769          */
 770         if (old_state != DCCP_LISTEN) {
 771                 /*
 772                  * FIXME: here it probably should be sk->sk_prot->listen_start
 773                  * see tcp_listen_start
 774                  */
 775                 err = dccp_listen_start(sk);
 776                 if (err)
 777                         goto out;
 778         }
 779         sk->sk_max_ack_backlog = backlog;
 780         err = 0;
 781
 782 out:
 783         release_sock(sk);
 784         return err;
 785 }
 786
 787 EXPORT_SYMBOL_GPL(inet_dccp_listen);
 788
 789 static const unsigned char dccp_new_state[] = {
 790         /* current state:   new state:      action:     */
 791         [0]               = DCCP_CLOSED,
 792         [DCCP_OPEN]       = DCCP_CLOSING | DCCP_ACTION_FIN,
 793         [DCCP_REQUESTING] = DCCP_CLOSED,
 794         [DCCP_PARTOPEN]   = DCCP_CLOSING | DCCP_ACTION_FIN,
 795         [DCCP_LISTEN]     = DCCP_CLOSED,
 796         [DCCP_RESPOND]    = DCCP_CLOSED,
 797         [DCCP_CLOSING]    = DCCP_CLOSED,
 798         [DCCP_TIME_WAIT]  = DCCP_CLOSED,
 799         [DCCP_CLOSED]     = DCCP_CLOSED,
 800 };
 801
 802 static int dccp_close_state(struct sock *sk)
 803 {
 804         const int next = dccp_new_state[sk->sk_state];
 805         const int ns = next & DCCP_STATE_MASK;
 806
 807         if (ns != sk->sk_state)
 808                 dccp_set_state(sk, ns);
 809
 810         return next & DCCP_ACTION_FIN;
 811 }
 812
 813 void dccp_close(struct sock *sk, long timeout)
 814 {
 815         struct sk_buff *skb;
 816
 817         lock_sock(sk);
 818
 819         sk->sk_shutdown = SHUTDOWN_MASK;
 820
 821         if (sk->sk_state == DCCP_LISTEN) {
 822                 dccp_set_state(sk, DCCP_CLOSED);
 823
 824                 /* Special case. */
 825                 inet_csk_listen_stop(sk);
 826
 827                 goto adjudge_to_death;
 828         }
 829
 830         /*
 831          * We need to flush the recv. buffs.  We do this only on the
 832          * descriptor close, not protocol-sourced closes, because the
 833           *reader process may not have drained the data yet!
 834          */
 835         /* FIXME: check for unread data */
 836         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 837                 __kfree_skb(skb);
 838         }
 839
 840         if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
 841                 /* Check zero linger _after_ checking for unread data. */
 842                 sk->sk_prot->disconnect(sk, 0);
 843         } else if (dccp_close_state(sk)) {
 844                 dccp_send_close(sk, 1);
 845         }
 846
 847         sk_stream_wait_close(sk, timeout);
 848
 849 adjudge_to_death:
 850         /*
 851          * It is the last release_sock in its life. It will remove backlog.
 852          */
 853         release_sock(sk);
 854         /*
 855          * Now socket is owned by kernel and we acquire BH lock
 856          * to finish close. No need to check for user refs.
 857          */
 858         local_bh_disable();
 859         bh_lock_sock(sk);
 860         BUG_TRAP(!sock_owned_by_user(sk));
 861
 862         sock_hold(sk);
 863         sock_orphan(sk);
 864
 865         /*
 866          * The last release_sock may have processed the CLOSE or RESET
 867          * packet moving sock to CLOSED state, if not we have to fire
 868          * the CLOSE/CLOSEREQ retransmission timer, see "8.3. Termination"
 869          * in draft-ietf-dccp-spec-11. -acme
 870          */
 871         if (sk->sk_state == DCCP_CLOSING) {
 872                 /* FIXME: should start at 2 * RTT */
 873                 /* Timer for repeating the CLOSE/CLOSEREQ until an answer. */
 874                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 875                                           inet_csk(sk)->icsk_rto,
 876                                           DCCP_RTO_MAX);
 877 #if 0
 878                 /* Yeah, we should use sk->sk_prot->orphan_count, etc */
 879                 dccp_set_state(sk, DCCP_CLOSED);
 880 #endif
 881         }
 882
 883         atomic_inc(sk->sk_prot->orphan_count);
 884         if (sk->sk_state == DCCP_CLOSED)
 885                 inet_csk_destroy_sock(sk);
 886
 887         /* Otherwise, socket is reprieved until protocol close. */
 888
 889         bh_unlock_sock(sk);
 890         local_bh_enable();
 891         sock_put(sk);
 892 }
 893
 894 EXPORT_SYMBOL_GPL(dccp_close);
 895
 896 void dccp_shutdown(struct sock *sk, int how)
 897 {
 898         dccp_pr_debug("entry\n");
 899 }
 900
 901 EXPORT_SYMBOL_GPL(dccp_shutdown);
 902
 903 static int __init dccp_mib_init(void)
 904 {
 905         int rc = -ENOMEM;
 906
 907         dccp_statistics[0] = alloc_percpu(struct dccp_mib);
 908         if (dccp_statistics[0] == NULL)
 909                 goto out;
 910
 911         dccp_statistics[1] = alloc_percpu(struct dccp_mib);
 912         if (dccp_statistics[1] == NULL)
 913                 goto out_free_one;
 914
 915         rc = 0;
 916 out:
 917         return rc;
 918 out_free_one:
 919         free_percpu(dccp_statistics[0]);
 920         dccp_statistics[0] = NULL;
 921         goto out;
 922
 923 }
 924
 925 static void dccp_mib_exit(void)
 926 {
 927         free_percpu(dccp_statistics[0]);
 928         free_percpu(dccp_statistics[1]);
 929         dccp_statistics[0] = dccp_statistics[1] = NULL;
 930 }
 931
 932 static int thash_entries;
 933 module_param(thash_entries, int, 0444);
 934 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
 935
 936 #ifdef CONFIG_IP_DCCP_DEBUG
 937 int dccp_debug;
 938 module_param(dccp_debug, int, 0444);
 939 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
 940
 941 EXPORT_SYMBOL_GPL(dccp_debug);
 942 #endif
 943
 944 static int __init dccp_init(void)
 945 {
 946         unsigned long goal;
 947         int ehash_order, bhash_order, i;
 948         int rc = -ENOBUFS;
 949
 950         dccp_hashinfo.bind_bucket_cachep =
 951                 kmem_cache_create("dccp_bind_bucket",
 952                                   sizeof(struct inet_bind_bucket), 0,
 953                                   SLAB_HWCACHE_ALIGN, NULL, NULL);
 954         if (!dccp_hashinfo.bind_bucket_cachep)
 955                 goto out;
 956
 957         /*
 958          * Size and allocate the main established and bind bucket
 959          * hash tables.
 960          *
 961          * The methodology is similar to that of the buffer cache.
 962          */
 963         if (num_physpages >= (128 * 1024))
 964                 goal = num_physpages >> (21 - PAGE_SHIFT);
 965         else
 966                 goal = num_physpages >> (23 - PAGE_SHIFT);
 967
 968         if (thash_entries)
 969                 goal = (thash_entries *
 970                         sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
 971         for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
 972                 ;
 973         do {
 974                 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
 975                                         sizeof(struct inet_ehash_bucket);
 976                 dccp_hashinfo.ehash_size >>= 1;
 977                 while (dccp_hashinfo.ehash_size &
 978                        (dccp_hashinfo.ehash_size - 1))
 979                         dccp_hashinfo.ehash_size--;
 980                 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
 981                         __get_free_pages(GFP_ATOMIC, ehash_order);
 982         } while (!dccp_hashinfo.ehash && --ehash_order > 0);
 983
 984         if (!dccp_hashinfo.ehash) {
 985                 printk(KERN_CRIT "Failed to allocate DCCP "
 986                                  "established hash table\n");
 987                 goto out_free_bind_bucket_cachep;
 988         }
 989
 990         for (i = 0; i < (dccp_hashinfo.ehash_size << 1); i++) {
 991                 rwlock_init(&dccp_hashinfo.ehash[i].lock);
 992                 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].chain);
 993         }
 994
 995         bhash_order = ehash_order;
 996
 997         do {
 998                 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
 999                                         sizeof(struct inet_bind_hashbucket);
1000                 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1001                     bhash_order > 0)
1002                         continue;
1003                 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1004                         __get_free_pages(GFP_ATOMIC, bhash_order);
1005         } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1006
1007         if (!dccp_hashinfo.bhash) {
1008                 printk(KERN_CRIT "Failed to allocate DCCP bind hash table\n");
1009                 goto out_free_dccp_ehash;
1010         }
1011
1012         for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1013                 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1014                 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1015         }
1016
1017         rc = dccp_mib_init();
1018         if (rc)
1019                 goto out_free_dccp_bhash;
1020
1021         rc = dccp_ackvec_init();
1022         if (rc)
1023                 goto out_free_dccp_mib;
1024
1025         rc = dccp_sysctl_init();
1026         if (rc)
1027                 goto out_ackvec_exit;
1028 out:
1029         return rc;
1030 out_ackvec_exit:
1031         dccp_ackvec_exit();
1032 out_free_dccp_mib:
1033         dccp_mib_exit();
1034 out_free_dccp_bhash:
1035         free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1036         dccp_hashinfo.bhash = NULL;
1037 out_free_dccp_ehash:
1038         free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1039         dccp_hashinfo.ehash = NULL;
1040 out_free_bind_bucket_cachep:
1041         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1042         dccp_hashinfo.bind_bucket_cachep = NULL;
1043         goto out;
1044 }
1045
1046 static void __exit dccp_fini(void)
1047 {
1048         dccp_mib_exit();
1049         free_pages((unsigned long)dccp_hashinfo.bhash,
1050                    get_order(dccp_hashinfo.bhash_size *
1051                              sizeof(struct inet_bind_hashbucket)));
1052         free_pages((unsigned long)dccp_hashinfo.ehash,
1053                    get_order(dccp_hashinfo.ehash_size *
1054                              sizeof(struct inet_ehash_bucket)));
1055         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1056         dccp_ackvec_exit();
1057         dccp_sysctl_exit();
1058 }
1059
1060 module_init(dccp_init);
1061 module_exit(dccp_fini);
1062
1063 MODULE_LICENSE("GPL");
1064 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1065 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");