net/core/datagram.c

   1 /*
   2  *      SUCS NET3:
   3  *
   4  *      Generic datagram handling routines. These are generic for all
   5  *      protocols. Possibly a generic IP version on top of these would
   6  *      make sense. Not tonight however 8-).
   7  *      This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and
   8  *      NetROM layer all have identical poll code and mostly
   9  *      identical recvmsg() code. So we share it here. The poll was
  10  *      shared before but buried in udp.c so I moved it.
  11  *
  12  *      Authors:        Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old
  13  *                                                   udp.c code)
  14  *
  15  *      Fixes:
  16  *              Alan Cox        :       NULL return from skb_peek_copy()
  17  *                                      understood
  18  *              Alan Cox        :       Rewrote skb_read_datagram to avoid the
  19  *                                      skb_peek_copy stuff.
  20  *              Alan Cox        :       Added support for SOCK_SEQPACKET.
  21  *                                      IPX can no longer use the SO_TYPE hack
  22  *                                      but AX.25 now works right, and SPX is
  23  *                                      feasible.
  24  *              Alan Cox        :       Fixed write poll of non IP protocol
  25  *                                      crash.
  26  *              Florian  La Roche:      Changed for my new skbuff handling.
  27  *              Darryl Miles    :       Fixed non-blocking SOCK_SEQPACKET.
  28  *              Linus Torvalds  :       BSD semantic fixes.
  29  *              Alan Cox        :       Datagram iovec handling
  30  *              Darryl Miles    :       Fixed non-blocking SOCK_STREAM.
  31  *              Alan Cox        :       POSIXisms
  32  *              Pete Wyckoff    :       Unconnected accept() fix.
  33  *
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/types.h>
  38 #include <linux/kernel.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/system.h>
  41 #include <linux/mm.h>
  42 #include <linux/interrupt.h>
  43 #include <linux/errno.h>
  44 #include <linux/sched.h>
  45 #include <linux/inet.h>
  46 #include <linux/netdevice.h>
  47 #include <linux/rtnetlink.h>
  48 #include <linux/poll.h>
  49 #include <linux/highmem.h>
  50 #include <linux/spinlock.h>
  51
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54
  55 #include <net/checksum.h>
  56 #include <net/sock.h>
  57 #include <net/tcp_states.h>
  58
  59 /*
  60  *      Is a socket 'connection oriented' ?
  61  */
  62 static inline int connection_based(struct sock *sk)
  63 {
  64         return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
  65 }
  66
  67 static int receiver_wake_function(wait_queue_t *wait, unsigned mode, int sync,
  68                                   void *key)
  69 {
  70         unsigned long bits = (unsigned long)key;
  71
  72         /*
  73          * Avoid a wakeup if event not interesting for us
  74          */
  75         if (bits && !(bits & (POLLIN | POLLERR)))
  76                 return 0;
  77         return autoremove_wake_function(wait, mode, sync, key);
  78 }
  79 /*
  80  * Wait for a packet..
  81  */
  82 static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
  83 {
  84         int error;
  85         DEFINE_WAIT_FUNC(wait, receiver_wake_function);
  86
  87         prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
  88
  89         /* Socket errors? */
  90         error = sock_error(sk);
  91         if (error)
  92                 goto out_err;
  93
  94         if (!skb_queue_empty(&sk->sk_receive_queue))
  95                 goto out;
  96
  97         /* Socket shut down? */
  98         if (sk->sk_shutdown & RCV_SHUTDOWN)
  99                 goto out_noerr;
 100
 101         /* Sequenced packets can come disconnected.
 102          * If so we report the problem
 103          */
 104         error = -ENOTCONN;
 105         if (connection_based(sk) &&
 106             !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
 107                 goto out_err;
 108
 109         /* handle signals */
 110         if (signal_pending(current))
 111                 goto interrupted;
 112
 113         error = 0;
 114         *timeo_p = schedule_timeout(*timeo_p);
 115 out:
 116         finish_wait(sk->sk_sleep, &wait);
 117         return error;
 118 interrupted:
 119         error = sock_intr_errno(*timeo_p);
 120 out_err:
 121         *err = error;
 122         goto out;
 123 out_noerr:
 124         *err = 0;
 125         error = 1;
 126         goto out;
 127 }
 128
 129 /**
 130  *      __skb_recv_datagram - Receive a datagram skbuff
 131  *      @sk: socket
 132  *      @flags: MSG_ flags
 133  *      @peeked: returns non-zero if this packet has been seen before
 134  *      @err: error code returned
 135  *
 136  *      Get a datagram skbuff, understands the peeking, nonblocking wakeups
 137  *      and possible races. This replaces identical code in packet, raw and
 138  *      udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
 139  *      the long standing peek and read race for datagram sockets. If you
 140  *      alter this routine remember it must be re-entrant.
 141  *
 142  *      This function will lock the socket if a skb is returned, so the caller
 143  *      needs to unlock the socket in that case (usually by calling
 144  *      skb_free_datagram)
 145  *
 146  *      * It does not lock socket since today. This function is
 147  *      * free of race conditions. This measure should/can improve
 148  *      * significantly datagram socket latencies at high loads,
 149  *      * when data copying to user space takes lots of time.
 150  *      * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
 151  *      *  8) Great win.)
 152  *      *                                           --ANK (980729)
 153  *
 154  *      The order of the tests when we find no data waiting are specified
 155  *      quite explicitly by POSIX 1003.1g, don't change them without having
 156  *      the standard around please.
 157  */
 158 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
 159                                     int *peeked, int *err)
 160 {
 161         struct sk_buff *skb;
 162         long timeo;
 163         /*
 164          * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
 165          */
 166         int error = sock_error(sk);
 167
 168         if (error)
 169                 goto no_packet;
 170
 171         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
 172
 173         do {
 174                 /* Again only user level code calls this function, so nothing
 175                  * interrupt level will suddenly eat the receive_queue.
 176                  *
 177                  * Look at current nfs client by the way...
 178                  * However, this function was corrent in any case. 8)
 179                  */
 180                 unsigned long cpu_flags;
 181
 182                 spin_lock_irqsave(&sk->sk_receive_queue.lock, cpu_flags);
 183                 skb = skb_peek(&sk->sk_receive_queue);
 184                 if (skb) {
 185                         *peeked = skb->peeked;
 186                         if (flags & MSG_PEEK) {
 187                                 skb->peeked = 1;
 188                                 atomic_inc(&skb->users);
 189                         } else
 190                                 __skb_unlink(skb, &sk->sk_receive_queue);
 191                 }
 192                 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, cpu_flags);
 193
 194                 if (skb)
 195                         return skb;
 196
 197                 /* User doesn't want to wait */
 198                 error = -EAGAIN;
 199                 if (!timeo)
 200                         goto no_packet;
 201
 202         } while (!wait_for_packet(sk, err, &timeo));
 203
 204         return NULL;
 205
 206 no_packet:
 207         *err = error;
 208         return NULL;
 209 }
 210 EXPORT_SYMBOL(__skb_recv_datagram);
 211
 212 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
 213                                   int noblock, int *err)
 214 {
 215         int peeked;
 216
 217         return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
 218                                    &peeked, err);
 219 }
 220
 221 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
 222 {
 223         consume_skb(skb);
 224         sk_mem_reclaim_partial(sk);
 225 }
 226
 227 /**
 228  *      skb_kill_datagram - Free a datagram skbuff forcibly
 229  *      @sk: socket
 230  *      @skb: datagram skbuff
 231  *      @flags: MSG_ flags
 232  *
 233  *      This function frees a datagram skbuff that was received by
 234  *      skb_recv_datagram.  The flags argument must match the one
 235  *      used for skb_recv_datagram.
 236  *
 237  *      If the MSG_PEEK flag is set, and the packet is still on the
 238  *      receive queue of the socket, it will be taken off the queue
 239  *      before it is freed.
 240  *
 241  *      This function currently only disables BH when acquiring the
 242  *      sk_receive_queue lock.  Therefore it must not be used in a
 243  *      context where that lock is acquired in an IRQ context.
 244  *
 245  *      It returns 0 if the packet was removed by us.
 246  */
 247
 248 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
 249 {
 250         int err = 0;
 251
 252         if (flags & MSG_PEEK) {
 253                 err = -ENOENT;
 254                 spin_lock_bh(&sk->sk_receive_queue.lock);
 255                 if (skb == skb_peek(&sk->sk_receive_queue)) {
 256                         __skb_unlink(skb, &sk->sk_receive_queue);
 257                         atomic_dec(&skb->users);
 258                         err = 0;
 259                 }
 260                 spin_unlock_bh(&sk->sk_receive_queue.lock);
 261         }
 262
 263         kfree_skb(skb);
 264         sk_mem_reclaim_partial(sk);
 265
 266         return err;
 267 }
 268
 269 EXPORT_SYMBOL(skb_kill_datagram);
 270
 271 /**
 272  *      skb_copy_datagram_iovec - Copy a datagram to an iovec.
 273  *      @skb: buffer to copy
 274  *      @offset: offset in the buffer to start copying from
 275  *      @to: io vector to copy to
 276  *      @len: amount of data to copy from buffer to iovec
 277  *
 278  *      Note: the iovec is modified during the copy.
 279  */
 280 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
 281                             struct iovec *to, int len)
 282 {
 283         int start = skb_headlen(skb);
 284         int i, copy = start - offset;
 285         struct sk_buff *frag_iter;
 286
 287         /* Copy header. */
 288         if (copy > 0) {
 289                 if (copy > len)
 290                         copy = len;
 291                 if (memcpy_toiovec(to, skb->data + offset, copy))
 292                         goto fault;
 293                 if ((len -= copy) == 0)
 294                         return 0;
 295                 offset += copy;
 296         }
 297
 298         /* Copy paged appendix. Hmm... why does this look so complicated? */
 299         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 300                 int end;
 301
 302                 WARN_ON(start > offset + len);
 303
 304                 end = start + skb_shinfo(skb)->frags[i].size;
 305                 if ((copy = end - offset) > 0) {
 306                         int err;
 307                         u8  *vaddr;
 308                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 309                         struct page *page = frag->page;
 310
 311                         if (copy > len)
 312                                 copy = len;
 313                         vaddr = kmap(page);
 314                         err = memcpy_toiovec(to, vaddr + frag->page_offset +
 315                                              offset - start, copy);
 316                         kunmap(page);
 317                         if (err)
 318                                 goto fault;
 319                         if (!(len -= copy))
 320                                 return 0;
 321                         offset += copy;
 322                 }
 323                 start = end;
 324         }
 325
 326         skb_walk_frags(skb, frag_iter) {
 327                 int end;
 328
 329                 WARN_ON(start > offset + len);
 330
 331                 end = start + frag_iter->len;
 332                 if ((copy = end - offset) > 0) {
 333                         if (copy > len)
 334                                 copy = len;
 335                         if (skb_copy_datagram_iovec(frag_iter,
 336                                                     offset - start,
 337                                                     to, copy))
 338                                 goto fault;
 339                         if ((len -= copy) == 0)
 340                                 return 0;
 341                         offset += copy;
 342                 }
 343                 start = end;
 344         }
 345         if (!len)
 346                 return 0;
 347
 348 fault:
 349         return -EFAULT;
 350 }
 351
 352 /**
 353  *      skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
 354  *      @skb: buffer to copy
 355  *      @offset: offset in the buffer to start copying from
 356  *      @to: io vector to copy to
 357  *      @to_offset: offset in the io vector to start copying to
 358  *      @len: amount of data to copy from buffer to iovec
 359  *
 360  *      Returns 0 or -EFAULT.
 361  *      Note: the iovec is not modified during the copy.
 362  */
 363 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
 364                                   const struct iovec *to, int to_offset,
 365                                   int len)
 366 {
 367         int start = skb_headlen(skb);
 368         int i, copy = start - offset;
 369         struct sk_buff *frag_iter;
 370
 371         /* Copy header. */
 372         if (copy > 0) {
 373                 if (copy > len)
 374                         copy = len;
 375                 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
 376                         goto fault;
 377                 if ((len -= copy) == 0)
 378                         return 0;
 379                 offset += copy;
 380                 to_offset += copy;
 381         }
 382
 383         /* Copy paged appendix. Hmm... why does this look so complicated? */
 384         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 385                 int end;
 386
 387                 WARN_ON(start > offset + len);
 388
 389                 end = start + skb_shinfo(skb)->frags[i].size;
 390                 if ((copy = end - offset) > 0) {
 391                         int err;
 392                         u8  *vaddr;
 393                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 394                         struct page *page = frag->page;
 395
 396                         if (copy > len)
 397                                 copy = len;
 398                         vaddr = kmap(page);
 399                         err = memcpy_toiovecend(to, vaddr + frag->page_offset +
 400                                                 offset - start, to_offset, copy);
 401                         kunmap(page);
 402                         if (err)
 403                                 goto fault;
 404                         if (!(len -= copy))
 405                                 return 0;
 406                         offset += copy;
 407                         to_offset += copy;
 408                 }
 409                 start = end;
 410         }
 411
 412         skb_walk_frags(skb, frag_iter) {
 413                 int end;
 414
 415                 WARN_ON(start > offset + len);
 416
 417                 end = start + frag_iter->len;
 418                 if ((copy = end - offset) > 0) {
 419                         if (copy > len)
 420                                 copy = len;
 421                         if (skb_copy_datagram_const_iovec(frag_iter,
 422                                                           offset - start,
 423                                                           to, to_offset,
 424                                                           copy))
 425                                 goto fault;
 426                         if ((len -= copy) == 0)
 427                                 return 0;
 428                         offset += copy;
 429                         to_offset += copy;
 430                 }
 431                 start = end;
 432         }
 433         if (!len)
 434                 return 0;
 435
 436 fault:
 437         return -EFAULT;
 438 }
 439 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
 440
 441 /**
 442  *      skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
 443  *      @skb: buffer to copy
 444  *      @offset: offset in the buffer to start copying to
 445  *      @from: io vector to copy to
 446  *      @from_offset: offset in the io vector to start copying from
 447  *      @len: amount of data to copy to buffer from iovec
 448  *
 449  *      Returns 0 or -EFAULT.
 450  *      Note: the iovec is not modified during the copy.
 451  */
 452 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
 453                                  const struct iovec *from, int from_offset,
 454                                  int len)
 455 {
 456         int start = skb_headlen(skb);
 457         int i, copy = start - offset;
 458         struct sk_buff *frag_iter;
 459
 460         /* Copy header. */
 461         if (copy > 0) {
 462                 if (copy > len)
 463                         copy = len;
 464                 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
 465                                         copy))
 466                         goto fault;
 467                 if ((len -= copy) == 0)
 468                         return 0;
 469                 offset += copy;
 470                 from_offset += copy;
 471         }
 472
 473         /* Copy paged appendix. Hmm... why does this look so complicated? */
 474         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 475                 int end;
 476
 477                 WARN_ON(start > offset + len);
 478
 479                 end = start + skb_shinfo(skb)->frags[i].size;
 480                 if ((copy = end - offset) > 0) {
 481                         int err;
 482                         u8  *vaddr;
 483                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 484                         struct page *page = frag->page;
 485
 486                         if (copy > len)
 487                                 copy = len;
 488                         vaddr = kmap(page);
 489                         err = memcpy_fromiovecend(vaddr + frag->page_offset +
 490                                                   offset - start,
 491                                                   from, from_offset, copy);
 492                         kunmap(page);
 493                         if (err)
 494                                 goto fault;
 495
 496                         if (!(len -= copy))
 497                                 return 0;
 498                         offset += copy;
 499                         from_offset += copy;
 500                 }
 501                 start = end;
 502         }
 503
 504         skb_walk_frags(skb, frag_iter) {
 505                 int end;
 506
 507                 WARN_ON(start > offset + len);
 508
 509                 end = start + frag_iter->len;
 510                 if ((copy = end - offset) > 0) {
 511                         if (copy > len)
 512                                 copy = len;
 513                         if (skb_copy_datagram_from_iovec(frag_iter,
 514                                                          offset - start,
 515                                                          from,
 516                                                          from_offset,
 517                                                          copy))
 518                                 goto fault;
 519                         if ((len -= copy) == 0)
 520                                 return 0;
 521                         offset += copy;
 522                         from_offset += copy;
 523                 }
 524                 start = end;
 525         }
 526         if (!len)
 527                 return 0;
 528
 529 fault:
 530         return -EFAULT;
 531 }
 532 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
 533
 534 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
 535                                       u8 __user *to, int len,
 536                                       __wsum *csump)
 537 {
 538         int start = skb_headlen(skb);
 539         int i, copy = start - offset;
 540         struct sk_buff *frag_iter;
 541         int pos = 0;
 542
 543         /* Copy header. */
 544         if (copy > 0) {
 545                 int err = 0;
 546                 if (copy > len)
 547                         copy = len;
 548                 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
 549                                                *csump, &err);
 550                 if (err)
 551                         goto fault;
 552                 if ((len -= copy) == 0)
 553                         return 0;
 554                 offset += copy;
 555                 to += copy;
 556                 pos = copy;
 557         }
 558
 559         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 560                 int end;
 561
 562                 WARN_ON(start > offset + len);
 563
 564                 end = start + skb_shinfo(skb)->frags[i].size;
 565                 if ((copy = end - offset) > 0) {
 566                         __wsum csum2;
 567                         int err = 0;
 568                         u8  *vaddr;
 569                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 570                         struct page *page = frag->page;
 571
 572                         if (copy > len)
 573                                 copy = len;
 574                         vaddr = kmap(page);
 575                         csum2 = csum_and_copy_to_user(vaddr +
 576                                                         frag->page_offset +
 577                                                         offset - start,
 578                                                       to, copy, 0, &err);
 579                         kunmap(page);
 580                         if (err)
 581                                 goto fault;
 582                         *csump = csum_block_add(*csump, csum2, pos);
 583                         if (!(len -= copy))
 584                                 return 0;
 585                         offset += copy;
 586                         to += copy;
 587                         pos += copy;
 588                 }
 589                 start = end;
 590         }
 591
 592         skb_walk_frags(skb, frag_iter) {
 593                 int end;
 594
 595                 WARN_ON(start > offset + len);
 596
 597                 end = start + frag_iter->len;
 598                 if ((copy = end - offset) > 0) {
 599                         __wsum csum2 = 0;
 600                         if (copy > len)
 601                                 copy = len;
 602                         if (skb_copy_and_csum_datagram(frag_iter,
 603                                                        offset - start,
 604                                                        to, copy,
 605                                                        &csum2))
 606                                 goto fault;
 607                         *csump = csum_block_add(*csump, csum2, pos);
 608                         if ((len -= copy) == 0)
 609                                 return 0;
 610                         offset += copy;
 611                         to += copy;
 612                         pos += copy;
 613                 }
 614                 start = end;
 615         }
 616         if (!len)
 617                 return 0;
 618
 619 fault:
 620         return -EFAULT;
 621 }
 622
 623 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
 624 {
 625         __sum16 sum;
 626
 627         sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
 628         if (likely(!sum)) {
 629                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
 630                         netdev_rx_csum_fault(skb->dev);
 631                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 632         }
 633         return sum;
 634 }
 635 EXPORT_SYMBOL(__skb_checksum_complete_head);
 636
 637 __sum16 __skb_checksum_complete(struct sk_buff *skb)
 638 {
 639         return __skb_checksum_complete_head(skb, skb->len);
 640 }
 641 EXPORT_SYMBOL(__skb_checksum_complete);
 642
 643 /**
 644  *      skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
 645  *      @skb: skbuff
 646  *      @hlen: hardware length
 647  *      @iov: io vector
 648  *
 649  *      Caller _must_ check that skb will fit to this iovec.
 650  *
 651  *      Returns: 0       - success.
 652  *               -EINVAL - checksum failure.
 653  *               -EFAULT - fault during copy. Beware, in this case iovec
 654  *                         can be modified!
 655  */
 656 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
 657                                      int hlen, struct iovec *iov)
 658 {
 659         __wsum csum;
 660         int chunk = skb->len - hlen;
 661
 662         if (!chunk)
 663                 return 0;
 664
 665         /* Skip filled elements.
 666          * Pretty silly, look at memcpy_toiovec, though 8)
 667          */
 668         while (!iov->iov_len)
 669                 iov++;
 670
 671         if (iov->iov_len < chunk) {
 672                 if (__skb_checksum_complete(skb))
 673                         goto csum_error;
 674                 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
 675                         goto fault;
 676         } else {
 677                 csum = csum_partial(skb->data, hlen, skb->csum);
 678                 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
 679                                                chunk, &csum))
 680                         goto fault;
 681                 if (csum_fold(csum))
 682                         goto csum_error;
 683                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
 684                         netdev_rx_csum_fault(skb->dev);
 685                 iov->iov_len -= chunk;
 686                 iov->iov_base += chunk;
 687         }
 688         return 0;
 689 csum_error:
 690         return -EINVAL;
 691 fault:
 692         return -EFAULT;
 693 }
 694
 695 /**
 696  *      datagram_poll - generic datagram poll
 697  *      @file: file struct
 698  *      @sock: socket
 699  *      @wait: poll table
 700  *
 701  *      Datagram poll: Again totally generic. This also handles
 702  *      sequenced packet sockets providing the socket receive queue
 703  *      is only ever holding data ready to receive.
 704  *
 705  *      Note: when you _don't_ use this routine for this protocol,
 706  *      and you use a different write policy from sock_writeable()
 707  *      then please supply your own write_space callback.
 708  */
 709 unsigned int datagram_poll(struct file *file, struct socket *sock,
 710                            poll_table *wait)
 711 {
 712         struct sock *sk = sock->sk;
 713         unsigned int mask;
 714
 715         sock_poll_wait(file, sk->sk_sleep, wait);
 716         mask = 0;
 717
 718         /* exceptional events? */
 719         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
 720                 mask |= POLLERR;
 721         if (sk->sk_shutdown & RCV_SHUTDOWN)
 722                 mask |= POLLRDHUP;
 723         if (sk->sk_shutdown == SHUTDOWN_MASK)
 724                 mask |= POLLHUP;
 725
 726         /* readable? */
 727         if (!skb_queue_empty(&sk->sk_receive_queue) ||
 728             (sk->sk_shutdown & RCV_SHUTDOWN))
 729                 mask |= POLLIN | POLLRDNORM;
 730
 731         /* Connection-based need to check for termination and startup */
 732         if (connection_based(sk)) {
 733                 if (sk->sk_state == TCP_CLOSE)
 734                         mask |= POLLHUP;
 735                 /* connection hasn't started yet? */
 736                 if (sk->sk_state == TCP_SYN_SENT)
 737                         return mask;
 738         }
 739
 740         /* writable? */
 741         if (sock_writeable(sk))
 742                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 743         else
 744                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 745
 746         return mask;
 747 }
 748
 749 EXPORT_SYMBOL(datagram_poll);
 750 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
 751 EXPORT_SYMBOL(skb_copy_datagram_iovec);
 752 EXPORT_SYMBOL(skb_free_datagram);
 753 EXPORT_SYMBOL(skb_recv_datagram);