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 <linux/mm.h>
  41 #include <linux/interrupt.h>
  42 #include <linux/errno.h>
  43 #include <linux/sched.h>
  44 #include <linux/inet.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/rtnetlink.h>
  47 #include <linux/poll.h>
  48 #include <linux/highmem.h>
  49 #include <linux/spinlock.h>
  50 #include <linux/slab.h>
  51 #include <linux/pagemap.h>
  52
  53 #include <net/protocol.h>
  54 #include <linux/skbuff.h>
  55
  56 #include <net/checksum.h>
  57 #include <net/sock.h>
  58 #include <net/tcp_states.h>
  59 #include <trace/events/skb.h>
  60 #include <net/busy_poll.h>
  61
  62 /*
  63  *      Is a socket 'connection oriented' ?
  64  */
  65 static inline int connection_based(struct sock *sk)
  66 {
  67         return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
  68 }
  69
  70 static int receiver_wake_function(wait_queue_t *wait, unsigned int mode, int sync,
  71                                   void *key)
  72 {
  73         unsigned long bits = (unsigned long)key;
  74
  75         /*
  76          * Avoid a wakeup if event not interesting for us
  77          */
  78         if (bits && !(bits & (POLLIN | POLLERR)))
  79                 return 0;
  80         return autoremove_wake_function(wait, mode, sync, key);
  81 }
  82 /*
  83  * Wait for the last received packet to be different from skb
  84  */
  85 static int wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
  86                                  const struct sk_buff *skb)
  87 {
  88         int error;
  89         DEFINE_WAIT_FUNC(wait, receiver_wake_function);
  90
  91         prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
  92
  93         /* Socket errors? */
  94         error = sock_error(sk);
  95         if (error)
  96                 goto out_err;
  97
  98         if (sk->sk_receive_queue.prev != skb)
  99                 goto out;
 100
 101         /* Socket shut down? */
 102         if (sk->sk_shutdown & RCV_SHUTDOWN)
 103                 goto out_noerr;
 104
 105         /* Sequenced packets can come disconnected.
 106          * If so we report the problem
 107          */
 108         error = -ENOTCONN;
 109         if (connection_based(sk) &&
 110             !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
 111                 goto out_err;
 112
 113         /* handle signals */
 114         if (signal_pending(current))
 115                 goto interrupted;
 116
 117         error = 0;
 118         *timeo_p = schedule_timeout(*timeo_p);
 119 out:
 120         finish_wait(sk_sleep(sk), &wait);
 121         return error;
 122 interrupted:
 123         error = sock_intr_errno(*timeo_p);
 124 out_err:
 125         *err = error;
 126         goto out;
 127 out_noerr:
 128         *err = 0;
 129         error = 1;
 130         goto out;
 131 }
 132
 133 /**
 134  *      __skb_recv_datagram - Receive a datagram skbuff
 135  *      @sk: socket
 136  *      @flags: MSG_ flags
 137  *      @peeked: returns non-zero if this packet has been seen before
 138  *      @off: an offset in bytes to peek skb from. Returns an offset
 139  *            within an skb where data actually starts
 140  *      @err: error code returned
 141  *
 142  *      Get a datagram skbuff, understands the peeking, nonblocking wakeups
 143  *      and possible races. This replaces identical code in packet, raw and
 144  *      udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
 145  *      the long standing peek and read race for datagram sockets. If you
 146  *      alter this routine remember it must be re-entrant.
 147  *
 148  *      This function will lock the socket if a skb is returned, so the caller
 149  *      needs to unlock the socket in that case (usually by calling
 150  *      skb_free_datagram)
 151  *
 152  *      * It does not lock socket since today. This function is
 153  *      * free of race conditions. This measure should/can improve
 154  *      * significantly datagram socket latencies at high loads,
 155  *      * when data copying to user space takes lots of time.
 156  *      * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
 157  *      *  8) Great win.)
 158  *      *                                           --ANK (980729)
 159  *
 160  *      The order of the tests when we find no data waiting are specified
 161  *      quite explicitly by POSIX 1003.1g, don't change them without having
 162  *      the standard around please.
 163  */
 164 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags,
 165                                     int *peeked, int *off, int *err)
 166 {
 167         struct sk_buff *skb, *last;
 168         long timeo;
 169         /*
 170          * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
 171          */
 172         int error = sock_error(sk);
 173
 174         if (error)
 175                 goto no_packet;
 176
 177         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
 178
 179         do {
 180                 /* Again only user level code calls this function, so nothing
 181                  * interrupt level will suddenly eat the receive_queue.
 182                  *
 183                  * Look at current nfs client by the way...
 184                  * However, this function was correct in any case. 8)
 185                  */
 186                 unsigned long cpu_flags;
 187                 struct sk_buff_head *queue = &sk->sk_receive_queue;
 188                 int _off = *off;
 189
 190                 last = (struct sk_buff *)queue;
 191                 spin_lock_irqsave(&queue->lock, cpu_flags);
 192                 skb_queue_walk(queue, skb) {
 193                         last = skb;
 194                         *peeked = skb->peeked;
 195                         if (flags & MSG_PEEK) {
 196                                 if (_off >= skb->len && (skb->len || _off ||
 197                                                          skb->peeked)) {
 198                                         _off -= skb->len;
 199                                         continue;
 200                                 }
 201                                 skb->peeked = 1;
 202                                 atomic_inc(&skb->users);
 203                         } else
 204                                 __skb_unlink(skb, queue);
 205
 206                         spin_unlock_irqrestore(&queue->lock, cpu_flags);
 207                         *off = _off;
 208                         return skb;
 209                 }
 210                 spin_unlock_irqrestore(&queue->lock, cpu_flags);
 211
 212                 if (sk_can_busy_loop(sk) &&
 213                     sk_busy_loop(sk, flags & MSG_DONTWAIT))
 214                         continue;
 215
 216                 /* User doesn't want to wait */
 217                 error = -EAGAIN;
 218                 if (!timeo)
 219                         goto no_packet;
 220
 221         } while (!wait_for_more_packets(sk, err, &timeo, last));
 222
 223         return NULL;
 224
 225 no_packet:
 226         *err = error;
 227         return NULL;
 228 }
 229 EXPORT_SYMBOL(__skb_recv_datagram);
 230
 231 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
 232                                   int noblock, int *err)
 233 {
 234         int peeked, off = 0;
 235
 236         return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
 237                                    &peeked, &off, err);
 238 }
 239 EXPORT_SYMBOL(skb_recv_datagram);
 240
 241 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
 242 {
 243         consume_skb(skb);
 244         sk_mem_reclaim_partial(sk);
 245 }
 246 EXPORT_SYMBOL(skb_free_datagram);
 247
 248 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb)
 249 {
 250         bool slow;
 251
 252         if (likely(atomic_read(&skb->users) == 1))
 253                 smp_rmb();
 254         else if (likely(!atomic_dec_and_test(&skb->users)))
 255                 return;
 256
 257         slow = lock_sock_fast(sk);
 258         skb_orphan(skb);
 259         sk_mem_reclaim_partial(sk);
 260         unlock_sock_fast(sk, slow);
 261
 262         /* skb is now orphaned, can be freed outside of locked section */
 263         __kfree_skb(skb);
 264 }
 265 EXPORT_SYMBOL(skb_free_datagram_locked);
 266
 267 /**
 268  *      skb_kill_datagram - Free a datagram skbuff forcibly
 269  *      @sk: socket
 270  *      @skb: datagram skbuff
 271  *      @flags: MSG_ flags
 272  *
 273  *      This function frees a datagram skbuff that was received by
 274  *      skb_recv_datagram.  The flags argument must match the one
 275  *      used for skb_recv_datagram.
 276  *
 277  *      If the MSG_PEEK flag is set, and the packet is still on the
 278  *      receive queue of the socket, it will be taken off the queue
 279  *      before it is freed.
 280  *
 281  *      This function currently only disables BH when acquiring the
 282  *      sk_receive_queue lock.  Therefore it must not be used in a
 283  *      context where that lock is acquired in an IRQ context.
 284  *
 285  *      It returns 0 if the packet was removed by us.
 286  */
 287
 288 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
 289 {
 290         int err = 0;
 291
 292         if (flags & MSG_PEEK) {
 293                 err = -ENOENT;
 294                 spin_lock_bh(&sk->sk_receive_queue.lock);
 295                 if (skb == skb_peek(&sk->sk_receive_queue)) {
 296                         __skb_unlink(skb, &sk->sk_receive_queue);
 297                         atomic_dec(&skb->users);
 298                         err = 0;
 299                 }
 300                 spin_unlock_bh(&sk->sk_receive_queue.lock);
 301         }
 302
 303         kfree_skb(skb);
 304         atomic_inc(&sk->sk_drops);
 305         sk_mem_reclaim_partial(sk);
 306
 307         return err;
 308 }
 309 EXPORT_SYMBOL(skb_kill_datagram);
 310
 311 /**
 312  *      skb_copy_datagram_iovec - Copy a datagram to an iovec.
 313  *      @skb: buffer to copy
 314  *      @offset: offset in the buffer to start copying from
 315  *      @to: io vector to copy to
 316  *      @len: amount of data to copy from buffer to iovec
 317  *
 318  *      Note: the iovec is modified during the copy.
 319  */
 320 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
 321                             struct iovec *to, int len)
 322 {
 323         int start = skb_headlen(skb);
 324         int i, copy = start - offset;
 325         struct sk_buff *frag_iter;
 326
 327         trace_skb_copy_datagram_iovec(skb, len);
 328
 329         /* Copy header. */
 330         if (copy > 0) {
 331                 if (copy > len)
 332                         copy = len;
 333                 if (memcpy_toiovec(to, skb->data + offset, copy))
 334                         goto fault;
 335                 if ((len -= copy) == 0)
 336                         return 0;
 337                 offset += copy;
 338         }
 339
 340         /* Copy paged appendix. Hmm... why does this look so complicated? */
 341         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 342                 int end;
 343                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 344
 345                 WARN_ON(start > offset + len);
 346
 347                 end = start + skb_frag_size(frag);
 348                 if ((copy = end - offset) > 0) {
 349                         int err;
 350                         u8  *vaddr;
 351                         struct page *page = skb_frag_page(frag);
 352
 353                         if (copy > len)
 354                                 copy = len;
 355                         vaddr = kmap(page);
 356                         err = memcpy_toiovec(to, vaddr + frag->page_offset +
 357                                              offset - start, copy);
 358                         kunmap(page);
 359                         if (err)
 360                                 goto fault;
 361                         if (!(len -= copy))
 362                                 return 0;
 363                         offset += copy;
 364                 }
 365                 start = end;
 366         }
 367
 368         skb_walk_frags(skb, frag_iter) {
 369                 int end;
 370
 371                 WARN_ON(start > offset + len);
 372
 373                 end = start + frag_iter->len;
 374                 if ((copy = end - offset) > 0) {
 375                         if (copy > len)
 376                                 copy = len;
 377                         if (skb_copy_datagram_iovec(frag_iter,
 378                                                     offset - start,
 379                                                     to, copy))
 380                                 goto fault;
 381                         if ((len -= copy) == 0)
 382                                 return 0;
 383                         offset += copy;
 384                 }
 385                 start = end;
 386         }
 387         if (!len)
 388                 return 0;
 389
 390 fault:
 391         return -EFAULT;
 392 }
 393 EXPORT_SYMBOL(skb_copy_datagram_iovec);
 394
 395 /**
 396  *      skb_copy_datagram_const_iovec - Copy a datagram to an iovec.
 397  *      @skb: buffer to copy
 398  *      @offset: offset in the buffer to start copying from
 399  *      @to: io vector to copy to
 400  *      @to_offset: offset in the io vector to start copying to
 401  *      @len: amount of data to copy from buffer to iovec
 402  *
 403  *      Returns 0 or -EFAULT.
 404  *      Note: the iovec is not modified during the copy.
 405  */
 406 int skb_copy_datagram_const_iovec(const struct sk_buff *skb, int offset,
 407                                   const struct iovec *to, int to_offset,
 408                                   int len)
 409 {
 410         int start = skb_headlen(skb);
 411         int i, copy = start - offset;
 412         struct sk_buff *frag_iter;
 413
 414         /* Copy header. */
 415         if (copy > 0) {
 416                 if (copy > len)
 417                         copy = len;
 418                 if (memcpy_toiovecend(to, skb->data + offset, to_offset, copy))
 419                         goto fault;
 420                 if ((len -= copy) == 0)
 421                         return 0;
 422                 offset += copy;
 423                 to_offset += copy;
 424         }
 425
 426         /* Copy paged appendix. Hmm... why does this look so complicated? */
 427         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 428                 int end;
 429                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 430
 431                 WARN_ON(start > offset + len);
 432
 433                 end = start + skb_frag_size(frag);
 434                 if ((copy = end - offset) > 0) {
 435                         int err;
 436                         u8  *vaddr;
 437                         struct page *page = skb_frag_page(frag);
 438
 439                         if (copy > len)
 440                                 copy = len;
 441                         vaddr = kmap(page);
 442                         err = memcpy_toiovecend(to, vaddr + frag->page_offset +
 443                                                 offset - start, to_offset, copy);
 444                         kunmap(page);
 445                         if (err)
 446                                 goto fault;
 447                         if (!(len -= copy))
 448                                 return 0;
 449                         offset += copy;
 450                         to_offset += copy;
 451                 }
 452                 start = end;
 453         }
 454
 455         skb_walk_frags(skb, frag_iter) {
 456                 int end;
 457
 458                 WARN_ON(start > offset + len);
 459
 460                 end = start + frag_iter->len;
 461                 if ((copy = end - offset) > 0) {
 462                         if (copy > len)
 463                                 copy = len;
 464                         if (skb_copy_datagram_const_iovec(frag_iter,
 465                                                           offset - start,
 466                                                           to, to_offset,
 467                                                           copy))
 468                                 goto fault;
 469                         if ((len -= copy) == 0)
 470                                 return 0;
 471                         offset += copy;
 472                         to_offset += copy;
 473                 }
 474                 start = end;
 475         }
 476         if (!len)
 477                 return 0;
 478
 479 fault:
 480         return -EFAULT;
 481 }
 482 EXPORT_SYMBOL(skb_copy_datagram_const_iovec);
 483
 484 /**
 485  *      skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
 486  *      @skb: buffer to copy
 487  *      @offset: offset in the buffer to start copying to
 488  *      @from: io vector to copy to
 489  *      @from_offset: offset in the io vector to start copying from
 490  *      @len: amount of data to copy to buffer from iovec
 491  *
 492  *      Returns 0 or -EFAULT.
 493  *      Note: the iovec is not modified during the copy.
 494  */
 495 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
 496                                  const struct iovec *from, int from_offset,
 497                                  int len)
 498 {
 499         int start = skb_headlen(skb);
 500         int i, copy = start - offset;
 501         struct sk_buff *frag_iter;
 502
 503         /* Copy header. */
 504         if (copy > 0) {
 505                 if (copy > len)
 506                         copy = len;
 507                 if (memcpy_fromiovecend(skb->data + offset, from, from_offset,
 508                                         copy))
 509                         goto fault;
 510                 if ((len -= copy) == 0)
 511                         return 0;
 512                 offset += copy;
 513                 from_offset += copy;
 514         }
 515
 516         /* Copy paged appendix. Hmm... why does this look so complicated? */
 517         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 518                 int end;
 519                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 520
 521                 WARN_ON(start > offset + len);
 522
 523                 end = start + skb_frag_size(frag);
 524                 if ((copy = end - offset) > 0) {
 525                         int err;
 526                         u8  *vaddr;
 527                         struct page *page = skb_frag_page(frag);
 528
 529                         if (copy > len)
 530                                 copy = len;
 531                         vaddr = kmap(page);
 532                         err = memcpy_fromiovecend(vaddr + frag->page_offset +
 533                                                   offset - start,
 534                                                   from, from_offset, copy);
 535                         kunmap(page);
 536                         if (err)
 537                                 goto fault;
 538
 539                         if (!(len -= copy))
 540                                 return 0;
 541                         offset += copy;
 542                         from_offset += copy;
 543                 }
 544                 start = end;
 545         }
 546
 547         skb_walk_frags(skb, frag_iter) {
 548                 int end;
 549
 550                 WARN_ON(start > offset + len);
 551
 552                 end = start + frag_iter->len;
 553                 if ((copy = end - offset) > 0) {
 554                         if (copy > len)
 555                                 copy = len;
 556                         if (skb_copy_datagram_from_iovec(frag_iter,
 557                                                          offset - start,
 558                                                          from,
 559                                                          from_offset,
 560                                                          copy))
 561                                 goto fault;
 562                         if ((len -= copy) == 0)
 563                                 return 0;
 564                         offset += copy;
 565                         from_offset += copy;
 566                 }
 567                 start = end;
 568         }
 569         if (!len)
 570                 return 0;
 571
 572 fault:
 573         return -EFAULT;
 574 }
 575 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
 576
 577 /**
 578  *      zerocopy_sg_from_iovec - Build a zerocopy datagram from an iovec
 579  *      @skb: buffer to copy
 580  *      @from: io vector to copy to
 581  *      @offset: offset in the io vector to start copying from
 582  *      @count: amount of vectors to copy to buffer from
 583  *
 584  *      The function will first copy up to headlen, and then pin the userspace
 585  *      pages and build frags through them.
 586  *
 587  *      Returns 0, -EFAULT or -EMSGSIZE.
 588  *      Note: the iovec is not modified during the copy
 589  */
 590 int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 591                                   int offset, size_t count)
 592 {
 593         int len = iov_length(from, count) - offset;
 594         int copy = skb_headlen(skb);
 595         int size, offset1 = 0;
 596         int i = 0;
 597
 598         /* Skip over from offset */
 599         while (count && (offset >= from->iov_len)) {
 600                 offset -= from->iov_len;
 601                 ++from;
 602                 --count;
 603         }
 604
 605         /* copy up to skb headlen */
 606         while (count && (copy > 0)) {
 607                 size = min_t(unsigned int, copy, from->iov_len - offset);
 608                 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 609                                    size))
 610                         return -EFAULT;
 611                 if (copy > size) {
 612                         ++from;
 613                         --count;
 614                         offset = 0;
 615                 } else
 616                         offset += size;
 617                 copy -= size;
 618                 offset1 += size;
 619         }
 620
 621         if (len == offset1)
 622                 return 0;
 623
 624         while (count--) {
 625                 struct page *page[MAX_SKB_FRAGS];
 626                 int num_pages;
 627                 unsigned long base;
 628                 unsigned long truesize;
 629
 630                 len = from->iov_len - offset;
 631                 if (!len) {
 632                         offset = 0;
 633                         ++from;
 634                         continue;
 635                 }
 636                 base = (unsigned long)from->iov_base + offset;
 637                 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 638                 if (i + size > MAX_SKB_FRAGS)
 639                         return -EMSGSIZE;
 640                 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 641                 if (num_pages != size) {
 642                         release_pages(&page[i], num_pages, 0);
 643                         return -EFAULT;
 644                 }
 645                 truesize = size * PAGE_SIZE;
 646                 skb->data_len += len;
 647                 skb->len += len;
 648                 skb->truesize += truesize;
 649                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 650                 while (len) {
 651                         int off = base & ~PAGE_MASK;
 652                         int size = min_t(int, len, PAGE_SIZE - off);
 653                         skb_fill_page_desc(skb, i, page[i], off, size);
 654                         base += size;
 655                         len -= size;
 656                         i++;
 657                 }
 658                 offset = 0;
 659                 ++from;
 660         }
 661         return 0;
 662 }
 663 EXPORT_SYMBOL(zerocopy_sg_from_iovec);
 664
 665 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
 666                                       u8 __user *to, int len,
 667                                       __wsum *csump)
 668 {
 669         int start = skb_headlen(skb);
 670         int i, copy = start - offset;
 671         struct sk_buff *frag_iter;
 672         int pos = 0;
 673
 674         /* Copy header. */
 675         if (copy > 0) {
 676                 int err = 0;
 677                 if (copy > len)
 678                         copy = len;
 679                 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
 680                                                *csump, &err);
 681                 if (err)
 682                         goto fault;
 683                 if ((len -= copy) == 0)
 684                         return 0;
 685                 offset += copy;
 686                 to += copy;
 687                 pos = copy;
 688         }
 689
 690         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 691                 int end;
 692                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 693
 694                 WARN_ON(start > offset + len);
 695
 696                 end = start + skb_frag_size(frag);
 697                 if ((copy = end - offset) > 0) {
 698                         __wsum csum2;
 699                         int err = 0;
 700                         u8  *vaddr;
 701                         struct page *page = skb_frag_page(frag);
 702
 703                         if (copy > len)
 704                                 copy = len;
 705                         vaddr = kmap(page);
 706                         csum2 = csum_and_copy_to_user(vaddr +
 707                                                         frag->page_offset +
 708                                                         offset - start,
 709                                                       to, copy, 0, &err);
 710                         kunmap(page);
 711                         if (err)
 712                                 goto fault;
 713                         *csump = csum_block_add(*csump, csum2, pos);
 714                         if (!(len -= copy))
 715                                 return 0;
 716                         offset += copy;
 717                         to += copy;
 718                         pos += copy;
 719                 }
 720                 start = end;
 721         }
 722
 723         skb_walk_frags(skb, frag_iter) {
 724                 int end;
 725
 726                 WARN_ON(start > offset + len);
 727
 728                 end = start + frag_iter->len;
 729                 if ((copy = end - offset) > 0) {
 730                         __wsum csum2 = 0;
 731                         if (copy > len)
 732                                 copy = len;
 733                         if (skb_copy_and_csum_datagram(frag_iter,
 734                                                        offset - start,
 735                                                        to, copy,
 736                                                        &csum2))
 737                                 goto fault;
 738                         *csump = csum_block_add(*csump, csum2, pos);
 739                         if ((len -= copy) == 0)
 740                                 return 0;
 741                         offset += copy;
 742                         to += copy;
 743                         pos += copy;
 744                 }
 745                 start = end;
 746         }
 747         if (!len)
 748                 return 0;
 749
 750 fault:
 751         return -EFAULT;
 752 }
 753
 754 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
 755 {
 756         __sum16 sum;
 757
 758         sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
 759         if (likely(!sum)) {
 760                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
 761                         netdev_rx_csum_fault(skb->dev);
 762                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 763         }
 764         return sum;
 765 }
 766 EXPORT_SYMBOL(__skb_checksum_complete_head);
 767
 768 __sum16 __skb_checksum_complete(struct sk_buff *skb)
 769 {
 770         return __skb_checksum_complete_head(skb, skb->len);
 771 }
 772 EXPORT_SYMBOL(__skb_checksum_complete);
 773
 774 /**
 775  *      skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
 776  *      @skb: skbuff
 777  *      @hlen: hardware length
 778  *      @iov: io vector
 779  *
 780  *      Caller _must_ check that skb will fit to this iovec.
 781  *
 782  *      Returns: 0       - success.
 783  *               -EINVAL - checksum failure.
 784  *               -EFAULT - fault during copy. Beware, in this case iovec
 785  *                         can be modified!
 786  */
 787 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
 788                                      int hlen, struct iovec *iov)
 789 {
 790         __wsum csum;
 791         int chunk = skb->len - hlen;
 792
 793         if (!chunk)
 794                 return 0;
 795
 796         /* Skip filled elements.
 797          * Pretty silly, look at memcpy_toiovec, though 8)
 798          */
 799         while (!iov->iov_len)
 800                 iov++;
 801
 802         if (iov->iov_len < chunk) {
 803                 if (__skb_checksum_complete(skb))
 804                         goto csum_error;
 805                 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
 806                         goto fault;
 807         } else {
 808                 csum = csum_partial(skb->data, hlen, skb->csum);
 809                 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
 810                                                chunk, &csum))
 811                         goto fault;
 812                 if (csum_fold(csum))
 813                         goto csum_error;
 814                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
 815                         netdev_rx_csum_fault(skb->dev);
 816                 iov->iov_len -= chunk;
 817                 iov->iov_base += chunk;
 818         }
 819         return 0;
 820 csum_error:
 821         return -EINVAL;
 822 fault:
 823         return -EFAULT;
 824 }
 825 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
 826
 827 /**
 828  *      datagram_poll - generic datagram poll
 829  *      @file: file struct
 830  *      @sock: socket
 831  *      @wait: poll table
 832  *
 833  *      Datagram poll: Again totally generic. This also handles
 834  *      sequenced packet sockets providing the socket receive queue
 835  *      is only ever holding data ready to receive.
 836  *
 837  *      Note: when you _don't_ use this routine for this protocol,
 838  *      and you use a different write policy from sock_writeable()
 839  *      then please supply your own write_space callback.
 840  */
 841 unsigned int datagram_poll(struct file *file, struct socket *sock,
 842                            poll_table *wait)
 843 {
 844         struct sock *sk = sock->sk;
 845         unsigned int mask;
 846
 847         sock_poll_wait(file, sk_sleep(sk), wait);
 848         mask = 0;
 849
 850         /* exceptional events? */
 851         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
 852                 mask |= POLLERR |
 853                         (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
 854
 855         if (sk->sk_shutdown & RCV_SHUTDOWN)
 856                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
 857         if (sk->sk_shutdown == SHUTDOWN_MASK)
 858                 mask |= POLLHUP;
 859
 860         /* readable? */
 861         if (!skb_queue_empty(&sk->sk_receive_queue))
 862                 mask |= POLLIN | POLLRDNORM;
 863
 864         /* Connection-based need to check for termination and startup */
 865         if (connection_based(sk)) {
 866                 if (sk->sk_state == TCP_CLOSE)
 867                         mask |= POLLHUP;
 868                 /* connection hasn't started yet? */
 869                 if (sk->sk_state == TCP_SYN_SENT)
 870                         return mask;
 871         }
 872
 873         /* writable? */
 874         if (sock_writeable(sk))
 875                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 876         else
 877                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 878
 879         return mask;
 880 }
 881 EXPORT_SYMBOL(datagram_poll);