SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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_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         timeo = sock_sndtimeo(sk, noblock);
 186
 187         /*
 188          * We have to use sk_stream_wait_connect here to set sk_write_pending,
 189          * so that the trick in dccp_rcv_request_sent_state_process.
 190          */
 191         /* Wait for a connection to finish. */
 192         if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN | DCCPF_CLOSING))
 193                 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
 194                         goto out_release;
 195
 196         size = sk->sk_prot->max_header + len;
 197         release_sock(sk);
 198         skb = sock_alloc_send_skb(sk, size, noblock, &rc);
 199         lock_sock(sk);
 200         if (skb == NULL)
 201                 goto out_release;
 202
 203         skb_reserve(skb, sk->sk_prot->max_header);
 204         rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
 205         if (rc != 0)
 206                 goto out_discard;
 207
 208         rc = dccp_write_xmit(sk, skb, len);
 209 out_release:
 210         release_sock(sk);
 211         return rc ? : len;
 212 out_discard:
 213         kfree_skb(skb);
 214         goto out_release;
 215 }
 216
 217 EXPORT_SYMBOL(dccp_sendmsg);
 218
 219 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 220                  size_t len, int nonblock, int flags, int *addr_len)
 221 {
 222         const struct dccp_hdr *dh;
 223         int copied = 0;
 224         unsigned long used;
 225         int err;
 226         int target;             /* Read at least this many bytes */
 227         long timeo;
 228
 229         lock_sock(sk);
 230
 231         err = -ENOTCONN;
 232         if (sk->sk_state == DCCP_LISTEN)
 233                 goto out;
 234
 235         timeo = sock_rcvtimeo(sk, nonblock);
 236
 237         /* Urgent data needs to be handled specially. */
 238         if (flags & MSG_OOB)
 239                 goto recv_urg;
 240
 241         /* FIXME */
 242 #if 0
 243         seq = &tp->copied_seq;
 244         if (flags & MSG_PEEK) {
 245                 peek_seq = tp->copied_seq;
 246                 seq = &peek_seq;
 247         }
 248 #endif
 249
 250         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
 251
 252         do {
 253                 struct sk_buff *skb;
 254                 u32 offset;
 255
 256         /* FIXME */
 257 #if 0
 258                 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
 259                 if (tp->urg_data && tp->urg_seq == *seq) {
 260                         if (copied)
 261                                 break;
 262                         if (signal_pending(current)) {
 263                                 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
 264                                 break;
 265                         }
 266                 }
 267 #endif
 268
 269                 /* Next get a buffer. */
 270
 271                 skb = skb_peek(&sk->sk_receive_queue);
 272                 do {
 273                         if (!skb)
 274                                 break;
 275
 276                         offset = 0;
 277                         dh = dccp_hdr(skb);
 278
 279                         if (dh->dccph_type == DCCP_PKT_DATA ||
 280                             dh->dccph_type == DCCP_PKT_DATAACK)
 281                                 goto found_ok_skb;
 282
 283                         if (dh->dccph_type == DCCP_PKT_RESET ||
 284                             dh->dccph_type == DCCP_PKT_CLOSE) {
 285                                 dccp_pr_debug("found fin ok!\n");
 286                                 goto found_fin_ok;
 287                         }
 288                         dccp_pr_debug("packet_type=%s\n", dccp_packet_name(dh->dccph_type));
 289                         BUG_TRAP(flags & MSG_PEEK);
 290                         skb = skb->next;
 291                 } while (skb != (struct sk_buff *)&sk->sk_receive_queue);
 292
 293                 /* Well, if we have backlog, try to process it now yet. */
 294                 if (copied >= target && !sk->sk_backlog.tail)
 295                         break;
 296
 297                 if (copied) {
 298                         if (sk->sk_err ||
 299                             sk->sk_state == DCCP_CLOSED ||
 300                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
 301                             !timeo ||
 302                             signal_pending(current) ||
 303                             (flags & MSG_PEEK))
 304                                 break;
 305                 } else {
 306                         if (sock_flag(sk, SOCK_DONE))
 307                                 break;
 308
 309                         if (sk->sk_err) {
 310                                 copied = sock_error(sk);
 311                                 break;
 312                         }
 313
 314                         if (sk->sk_shutdown & RCV_SHUTDOWN)
 315                                 break;
 316
 317                         if (sk->sk_state == DCCP_CLOSED) {
 318                                 if (!sock_flag(sk, SOCK_DONE)) {
 319                                         /* This occurs when user tries to read
 320                                          * from never connected socket.
 321                                          */
 322                                         copied = -ENOTCONN;
 323                                         break;
 324                                 }
 325                                 break;
 326                         }
 327
 328                         if (!timeo) {
 329                                 copied = -EAGAIN;
 330                                 break;
 331                         }
 332
 333                         if (signal_pending(current)) {
 334                                 copied = sock_intr_errno(timeo);
 335                                 break;
 336                         }
 337                 }
 338
 339                 /* FIXME: cleanup_rbuf(sk, copied); */
 340
 341                 if (copied >= target) {
 342                         /* Do not sleep, just process backlog. */
 343                         release_sock(sk);
 344                         lock_sock(sk);
 345                 } else
 346                         sk_wait_data(sk, &timeo);
 347
 348                 continue;
 349
 350         found_ok_skb:
 351                 /* Ok so how much can we use? */
 352                 used = skb->len - offset;
 353                 if (len < used)
 354                         used = len;
 355
 356                 if (!(flags & MSG_TRUNC)) {
 357                         err = skb_copy_datagram_iovec(skb, offset,
 358                                                       msg->msg_iov, used);
 359                         if (err) {
 360                                 /* Exception. Bailout! */
 361                                 if (!copied)
 362                                         copied = -EFAULT;
 363                                 break;
 364                         }
 365                 }
 366
 367                 copied += used;
 368                 len -= used;
 369
 370                 /* FIXME: tcp_rcv_space_adjust(sk); */
 371
 372 //skip_copy:
 373                 if (used + offset < skb->len)
 374                         continue;
 375
 376                 if (!(flags & MSG_PEEK))
 377                         sk_eat_skb(sk, skb);
 378                 continue;
 379         found_fin_ok:
 380                 if (!(flags & MSG_PEEK))
 381                         sk_eat_skb(sk, skb);
 382                 break;
 383
 384         } while (len > 0);
 385
 386         /* According to UNIX98, msg_name/msg_namelen are ignored
 387          * on connected socket. I was just happy when found this 8) --ANK
 388          */
 389
 390         /* Clean up data we have read: This will do ACK frames. */
 391         /* FIXME: cleanup_rbuf(sk, copied); */
 392
 393         release_sock(sk);
 394         return copied;
 395
 396 out:
 397         release_sock(sk);
 398         return err;
 399
 400 recv_urg:
 401         /* FIXME: err = tcp_recv_urg(sk, timeo, msg, len, flags, addr_len); */
 402         goto out;
 403 }
 404
 405 static int inet_dccp_listen(struct socket *sock, int backlog)
 406 {
 407         struct sock *sk = sock->sk;
 408         unsigned char old_state;
 409         int err;
 410
 411         lock_sock(sk);
 412
 413         err = -EINVAL;
 414         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
 415                 goto out;
 416
 417         old_state = sk->sk_state;
 418         if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
 419                 goto out;
 420
 421         /* Really, if the socket is already in listen state
 422          * we can only allow the backlog to be adjusted.
 423          */
 424         if (old_state != DCCP_LISTEN) {
 425                 /*
 426                  * FIXME: here it probably should be sk->sk_prot->listen_start
 427                  * see tcp_listen_start
 428                  */
 429                 err = dccp_listen_start(sk);
 430                 if (err)
 431                         goto out;
 432         }
 433         sk->sk_max_ack_backlog = backlog;
 434         err = 0;
 435
 436 out:
 437         release_sock(sk);
 438         return err;
 439 }
 440
 441 static const unsigned char dccp_new_state[] = {
 442         /* current state:        new state:      action:        */
 443         [0]                     = DCCP_CLOSED,
 444         [DCCP_OPEN]             = DCCP_CLOSING | DCCP_ACTION_FIN,
 445         [DCCP_REQUESTING]       = DCCP_CLOSED,
 446         [DCCP_PARTOPEN] = DCCP_CLOSING | DCCP_ACTION_FIN,
 447         [DCCP_LISTEN]           = DCCP_CLOSED,
 448         [DCCP_RESPOND]  = DCCP_CLOSED,
 449         [DCCP_CLOSING]  = DCCP_CLOSED,
 450         [DCCP_TIME_WAIT]        = DCCP_CLOSED,
 451         [DCCP_CLOSED]   = DCCP_CLOSED,
 452 };
 453
 454 static int dccp_close_state(struct sock *sk)
 455 {
 456         const int next = dccp_new_state[sk->sk_state];
 457         const int ns = next & DCCP_STATE_MASK;
 458
 459         if (ns != sk->sk_state)
 460                 dccp_set_state(sk, ns);
 461
 462         return next & DCCP_ACTION_FIN;
 463 }
 464
 465 void dccp_close(struct sock *sk, long timeout)
 466 {
 467         struct sk_buff *skb;
 468
 469         lock_sock(sk);
 470
 471         sk->sk_shutdown = SHUTDOWN_MASK;
 472
 473         if (sk->sk_state == DCCP_LISTEN) {
 474                 dccp_set_state(sk, DCCP_CLOSED);
 475
 476                 /* Special case. */
 477                 inet_csk_listen_stop(sk);
 478
 479                 goto adjudge_to_death;
 480         }
 481
 482         /*
 483          * We need to flush the recv. buffs.  We do this only on the
 484          * descriptor close, not protocol-sourced closes, because the
 485           *reader process may not have drained the data yet!
 486          */
 487         /* FIXME: check for unread data */
 488         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 489                 __kfree_skb(skb);
 490         }
 491
 492         if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
 493                 /* Check zero linger _after_ checking for unread data. */
 494                 sk->sk_prot->disconnect(sk, 0);
 495         } else if (dccp_close_state(sk)) {
 496                 dccp_send_close(sk);
 497         }
 498
 499         sk_stream_wait_close(sk, timeout);
 500
 501 adjudge_to_death:
 502         release_sock(sk);
 503         /*
 504          * Now socket is owned by kernel and we acquire BH lock
 505          * to finish close. No need to check for user refs.
 506          */
 507         local_bh_disable();
 508         bh_lock_sock(sk);
 509         BUG_TRAP(!sock_owned_by_user(sk));
 510
 511         sock_hold(sk);
 512         sock_orphan(sk);
 513
 514         if (sk->sk_state != DCCP_CLOSED)
 515                 dccp_set_state(sk, DCCP_CLOSED);
 516
 517         atomic_inc(&dccp_orphan_count);
 518         if (sk->sk_state == DCCP_CLOSED)
 519                 inet_csk_destroy_sock(sk);
 520
 521         /* Otherwise, socket is reprieved until protocol close. */
 522
 523         bh_unlock_sock(sk);
 524         local_bh_enable();
 525         sock_put(sk);
 526 }
 527
 528 void dccp_shutdown(struct sock *sk, int how)
 529 {
 530         dccp_pr_debug("entry\n");
 531 }
 532
 533 struct proto_ops inet_dccp_ops = {
 534         .family         = PF_INET,
 535         .owner          = THIS_MODULE,
 536         .release        = inet_release,
 537         .bind           = inet_bind,
 538         .connect        = inet_stream_connect,
 539         .socketpair     = sock_no_socketpair,
 540         .accept         = inet_accept,
 541         .getname        = inet_getname,
 542         .poll           = sock_no_poll,
 543         .ioctl          = inet_ioctl,
 544         .listen         = inet_dccp_listen, /* FIXME: work on inet_listen to rename it to sock_common_listen */
 545         .shutdown       = inet_shutdown,
 546         .setsockopt     = sock_common_setsockopt,
 547         .getsockopt     = sock_common_getsockopt,
 548         .sendmsg        = inet_sendmsg,
 549         .recvmsg        = sock_common_recvmsg,
 550         .mmap           = sock_no_mmap,
 551         .sendpage       = sock_no_sendpage,
 552 };
 553
 554 extern struct net_proto_family inet_family_ops;
 555
 556 static struct inet_protosw dccp_v4_protosw = {
 557         .type           = SOCK_DCCP,
 558         .protocol       = IPPROTO_DCCP,
 559         .prot           = &dccp_v4_prot,
 560         .ops            = &inet_dccp_ops,
 561         .capability     = -1,
 562         .no_check       = 0,
 563         .flags          = 0,
 564 };
 565
 566 /*
 567  * This is the global socket data structure used for responding to
 568  * the Out-of-the-blue (OOTB) packets. A control sock will be created
 569  * for this socket at the initialization time.
 570  */
 571 struct socket *dccp_ctl_socket;
 572
 573 static char dccp_ctl_socket_err_msg[] __initdata =
 574         KERN_ERR "DCCP: Failed to create the control socket.\n";
 575
 576 static int __init dccp_ctl_sock_init(void)
 577 {
 578         int rc = sock_create_kern(PF_INET, SOCK_DCCP, IPPROTO_DCCP,
 579                                   &dccp_ctl_socket);
 580         if (rc < 0)
 581                 printk(dccp_ctl_socket_err_msg);
 582         else {
 583                 dccp_ctl_socket->sk->sk_allocation = GFP_ATOMIC;
 584                 inet_sk(dccp_ctl_socket->sk)->uc_ttl = -1;
 585
 586                 /* Unhash it so that IP input processing does not even
 587                  * see it, we do not wish this socket to see incoming
 588                  * packets.
 589                  */
 590                 dccp_ctl_socket->sk->sk_prot->unhash(dccp_ctl_socket->sk);
 591         }
 592
 593         return rc;
 594 }
 595
 596 static void __exit dccp_ctl_sock_exit(void)
 597 {
 598         if (dccp_ctl_socket != NULL)
 599                 sock_release(dccp_ctl_socket);
 600 }
 601
 602 static int __init init_dccp_v4_mibs(void)
 603 {
 604         int rc = -ENOMEM;
 605
 606         dccp_statistics[0] = alloc_percpu(struct dccp_mib);
 607         if (dccp_statistics[0] == NULL)
 608                 goto out;
 609
 610         dccp_statistics[1] = alloc_percpu(struct dccp_mib);
 611         if (dccp_statistics[1] == NULL)
 612                 goto out_free_one;
 613
 614         rc = 0;
 615 out:
 616         return rc;
 617 out_free_one:
 618         free_percpu(dccp_statistics[0]);
 619         dccp_statistics[0] = NULL;
 620         goto out;
 621
 622 }
 623
 624 static int thash_entries;
 625 module_param(thash_entries, int, 0444);
 626 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
 627
 628 int dccp_debug;
 629 module_param(dccp_debug, int, 0444);
 630 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
 631
 632 static int __init dccp_init(void)
 633 {
 634         unsigned long goal;
 635         int ehash_order, bhash_order, i;
 636         int rc = proto_register(&dccp_v4_prot, 1);
 637
 638         if (rc)
 639                 goto out;
 640
 641         dccp_hashinfo.bind_bucket_cachep = kmem_cache_create("dccp_bind_bucket",
 642                                                sizeof(struct inet_bind_bucket),
 643                                                0, SLAB_HWCACHE_ALIGN,
 644                                                NULL, NULL);
 645         if (!dccp_hashinfo.bind_bucket_cachep)
 646                 goto out_proto_unregister;
 647
 648         /*
 649          * Size and allocate the main established and bind bucket
 650          * hash tables.
 651          *
 652          * The methodology is similar to that of the buffer cache.
 653          */
 654         if (num_physpages >= (128 * 1024))
 655                 goal = num_physpages >> (21 - PAGE_SHIFT);
 656         else
 657                 goal = num_physpages >> (23 - PAGE_SHIFT);
 658
 659         if (thash_entries)
 660                 goal = (thash_entries * sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
 661         for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
 662                 ;
 663         do {
 664                 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
 665                                         sizeof(struct inet_ehash_bucket);
 666                 dccp_hashinfo.ehash_size >>= 1;
 667                 while (dccp_hashinfo.ehash_size & (dccp_hashinfo.ehash_size - 1))
 668                         dccp_hashinfo.ehash_size--;
 669                 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
 670                         __get_free_pages(GFP_ATOMIC, ehash_order);
 671         } while (!dccp_hashinfo.ehash && --ehash_order > 0);
 672
 673         if (!dccp_hashinfo.ehash) {
 674                 printk(KERN_CRIT "Failed to allocate DCCP "
 675                                  "established hash table\n");
 676                 goto out_free_bind_bucket_cachep;
 677         }
 678
 679         for (i = 0; i < (dccp_hashinfo.ehash_size << 1); i++) {
 680                 rwlock_init(&dccp_hashinfo.ehash[i].lock);
 681                 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].chain);
 682         }
 683
 684         bhash_order = ehash_order;
 685
 686         do {
 687                 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
 688                                         sizeof(struct inet_bind_hashbucket);
 689                 if ((dccp_hashinfo.bhash_size > (64 * 1024)) && bhash_order > 0)
 690                         continue;
 691                 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
 692                         __get_free_pages(GFP_ATOMIC, bhash_order);
 693         } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
 694
 695         if (!dccp_hashinfo.bhash) {
 696                 printk(KERN_CRIT "Failed to allocate DCCP bind hash table\n");
 697                 goto out_free_dccp_ehash;
 698         }
 699
 700         for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
 701                 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
 702                 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
 703         }
 704
 705         if (init_dccp_v4_mibs())
 706                 goto out_free_dccp_bhash;
 707
 708         rc = -EAGAIN;
 709         if (inet_add_protocol(&dccp_protocol, IPPROTO_DCCP))
 710                 goto out_free_dccp_v4_mibs;
 711
 712         inet_register_protosw(&dccp_v4_protosw);
 713
 714         rc = dccp_ctl_sock_init();
 715         if (rc)
 716                 goto out_unregister_protosw;
 717 out:
 718         return rc;
 719 out_unregister_protosw:
 720         inet_unregister_protosw(&dccp_v4_protosw);
 721         inet_del_protocol(&dccp_protocol, IPPROTO_DCCP);
 722 out_free_dccp_v4_mibs:
 723         free_percpu(dccp_statistics[0]);
 724         free_percpu(dccp_statistics[1]);
 725         dccp_statistics[0] = dccp_statistics[1] = NULL;
 726 out_free_dccp_bhash:
 727         free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
 728         dccp_hashinfo.bhash = NULL;
 729 out_free_dccp_ehash:
 730         free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
 731         dccp_hashinfo.ehash = NULL;
 732 out_free_bind_bucket_cachep:
 733         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
 734         dccp_hashinfo.bind_bucket_cachep = NULL;
 735 out_proto_unregister:
 736         proto_unregister(&dccp_v4_prot);
 737         goto out;
 738 }
 739
 740 static const char dccp_del_proto_err_msg[] __exitdata =
 741         KERN_ERR "can't remove dccp net_protocol\n";
 742
 743 static void __exit dccp_fini(void)
 744 {
 745         dccp_ctl_sock_exit();
 746
 747         inet_unregister_protosw(&dccp_v4_protosw);
 748
 749         if (inet_del_protocol(&dccp_protocol, IPPROTO_DCCP) < 0)
 750                 printk(dccp_del_proto_err_msg);
 751
 752         /* Free the control endpoint.  */
 753         sock_release(dccp_ctl_socket);
 754
 755         proto_unregister(&dccp_v4_prot);
 756
 757         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
 758 }
 759
 760 module_init(dccp_init);
 761 module_exit(dccp_fini);
 762
 763 /*
 764  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
 765  * values directly, Also cover the case where the protocol is not specified,
 766  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
 767  */
 768 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-33-type-6");
 769 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
 770 MODULE_LICENSE("GPL");
 771 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
 772 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");