fs/splice.c

   1 /*
   2  * "splice": joining two ropes together by interweaving their strands.
   3  *
   4  * This is the "extended pipe" functionality, where a pipe is used as
   5  * an arbitrary in-memory buffer. Think of a pipe as a small kernel
   6  * buffer that you can use to transfer data from one end to the other.
   7  *
   8  * The traditional unix read/write is extended with a "splice()" operation
   9  * that transfers data buffers to or from a pipe buffer.
  10  *
  11  * Named by Larry McVoy, original implementation from Linus, extended by
  12  * Jens to support splicing to files, network, direct splicing, etc and
  13  * fixing lots of bugs.
  14  *
  15  * Copyright (C) 2005-2006 Jens Axboe <axboe@suse.de>
  16  * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
  17  * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
  18  *
  19  */
  20 #include <linux/fs.h>
  21 #include <linux/file.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/pipe_fs_i.h>
  24 #include <linux/mm_inline.h>
  25 #include <linux/swap.h>
  26 #include <linux/writeback.h>
  27 #include <linux/buffer_head.h>
  28 #include <linux/module.h>
  29 #include <linux/syscalls.h>
  30
  31 /*
  32  * Passed to the actors
  33  */
  34 struct splice_desc {
  35         unsigned int len, total_len;    /* current and remaining length */
  36         unsigned int flags;             /* splice flags */
  37         struct file *file;              /* file to read/write */
  38         loff_t pos;                     /* file position */
  39 };
  40
  41 /*
  42  * Attempt to steal a page from a pipe buffer. This should perhaps go into
  43  * a vm helper function, it's already simplified quite a bit by the
  44  * addition of remove_mapping(). If success is returned, the caller may
  45  * attempt to reuse this page for another destination.
  46  */
  47 static int page_cache_pipe_buf_steal(struct pipe_inode_info *info,
  48                                      struct pipe_buffer *buf)
  49 {
  50         struct page *page = buf->page;
  51         struct address_space *mapping = page_mapping(page);
  52
  53         lock_page(page);
  54
  55         WARN_ON(!PageUptodate(page));
  56
  57         /*
  58          * At least for ext2 with nobh option, we need to wait on writeback
  59          * completing on this page, since we'll remove it from the pagecache.
  60          * Otherwise truncate wont wait on the page, allowing the disk
  61          * blocks to be reused by someone else before we actually wrote our
  62          * data to them. fs corruption ensues.
  63          */
  64         wait_on_page_writeback(page);
  65
  66         if (PagePrivate(page))
  67                 try_to_release_page(page, mapping_gfp_mask(mapping));
  68
  69         if (!remove_mapping(mapping, page)) {
  70                 unlock_page(page);
  71                 return 1;
  72         }
  73
  74         buf->flags |= PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU;
  75         return 0;
  76 }
  77
  78 static void page_cache_pipe_buf_release(struct pipe_inode_info *info,
  79                                         struct pipe_buffer *buf)
  80 {
  81         page_cache_release(buf->page);
  82         buf->page = NULL;
  83         buf->flags &= ~(PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU);
  84 }
  85
  86 static void *page_cache_pipe_buf_map(struct file *file,
  87                                      struct pipe_inode_info *info,
  88                                      struct pipe_buffer *buf)
  89 {
  90         struct page *page = buf->page;
  91         int err;
  92
  93         if (!PageUptodate(page)) {
  94                 lock_page(page);
  95
  96                 /*
  97                  * Page got truncated/unhashed. This will cause a 0-byte
  98                  * splice, if this is the first page.
  99                  */
 100                 if (!page->mapping) {
 101                         err = -ENODATA;
 102                         goto error;
 103                 }
 104
 105                 /*
 106                  * Uh oh, read-error from disk.
 107                  */
 108                 if (!PageUptodate(page)) {
 109                         err = -EIO;
 110                         goto error;
 111                 }
 112
 113                 /*
 114                  * Page is ok afterall, fall through to mapping.
 115                  */
 116                 unlock_page(page);
 117         }
 118
 119         return kmap(page);
 120 error:
 121         unlock_page(page);
 122         return ERR_PTR(err);
 123 }
 124
 125 static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info,
 126                                       struct pipe_buffer *buf)
 127 {
 128         kunmap(buf->page);
 129 }
 130
 131 static void page_cache_pipe_buf_get(struct pipe_inode_info *info,
 132                                     struct pipe_buffer *buf)
 133 {
 134         page_cache_get(buf->page);
 135 }
 136
 137 static struct pipe_buf_operations page_cache_pipe_buf_ops = {
 138         .can_merge = 0,
 139         .map = page_cache_pipe_buf_map,
 140         .unmap = page_cache_pipe_buf_unmap,
 141         .release = page_cache_pipe_buf_release,
 142         .steal = page_cache_pipe_buf_steal,
 143         .get = page_cache_pipe_buf_get,
 144 };
 145
 146 /*
 147  * Pipe output worker. This sets up our pipe format with the page cache
 148  * pipe buffer operations. Otherwise very similar to the regular pipe_writev().
 149  */
 150 static ssize_t move_to_pipe(struct pipe_inode_info *pipe, struct page **pages,
 151                             int nr_pages, unsigned long len,
 152                             unsigned int offset, unsigned int flags)
 153 {
 154         int ret, do_wakeup, i;
 155
 156         ret = 0;
 157         do_wakeup = 0;
 158         i = 0;
 159
 160         if (pipe->inode)
 161                 mutex_lock(&pipe->inode->i_mutex);
 162
 163         for (;;) {
 164                 if (!pipe->readers) {
 165                         send_sig(SIGPIPE, current, 0);
 166                         if (!ret)
 167                                 ret = -EPIPE;
 168                         break;
 169                 }
 170
 171                 if (pipe->nrbufs < PIPE_BUFFERS) {
 172                         int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
 173                         struct pipe_buffer *buf = pipe->bufs + newbuf;
 174                         struct page *page = pages[i++];
 175                         unsigned long this_len;
 176
 177                         this_len = PAGE_CACHE_SIZE - offset;
 178                         if (this_len > len)
 179                                 this_len = len;
 180
 181                         buf->page = page;
 182                         buf->offset = offset;
 183                         buf->len = this_len;
 184                         buf->ops = &page_cache_pipe_buf_ops;
 185                         pipe->nrbufs++;
 186                         if (pipe->inode)
 187                                 do_wakeup = 1;
 188
 189                         ret += this_len;
 190                         len -= this_len;
 191                         offset = 0;
 192                         if (!--nr_pages)
 193                                 break;
 194                         if (!len)
 195                                 break;
 196                         if (pipe->nrbufs < PIPE_BUFFERS)
 197                                 continue;
 198
 199                         break;
 200                 }
 201
 202                 if (flags & SPLICE_F_NONBLOCK) {
 203                         if (!ret)
 204                                 ret = -EAGAIN;
 205                         break;
 206                 }
 207
 208                 if (signal_pending(current)) {
 209                         if (!ret)
 210                                 ret = -ERESTARTSYS;
 211                         break;
 212                 }
 213
 214                 if (do_wakeup) {
 215                         smp_mb();
 216                         if (waitqueue_active(&pipe->wait))
 217                                 wake_up_interruptible_sync(&pipe->wait);
 218                         kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 219                         do_wakeup = 0;
 220                 }
 221
 222                 pipe->waiting_writers++;
 223                 pipe_wait(pipe);
 224                 pipe->waiting_writers--;
 225         }
 226
 227         if (pipe->inode)
 228                 mutex_unlock(&pipe->inode->i_mutex);
 229
 230         if (do_wakeup) {
 231                 smp_mb();
 232                 if (waitqueue_active(&pipe->wait))
 233                         wake_up_interruptible(&pipe->wait);
 234                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 235         }
 236
 237         while (i < nr_pages)
 238                 page_cache_release(pages[i++]);
 239
 240         return ret;
 241 }
 242
 243 static int
 244 __generic_file_splice_read(struct file *in, loff_t *ppos,
 245                            struct pipe_inode_info *pipe, size_t len,
 246                            unsigned int flags)
 247 {
 248         struct address_space *mapping = in->f_mapping;
 249         unsigned int loff, offset, nr_pages;
 250         struct page *pages[PIPE_BUFFERS];
 251         struct page *page;
 252         pgoff_t index, end_index;
 253         loff_t isize;
 254         size_t bytes;
 255         int i, error;
 256
 257         index = *ppos >> PAGE_CACHE_SHIFT;
 258         loff = offset = *ppos & ~PAGE_CACHE_MASK;
 259         nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 260
 261         if (nr_pages > PIPE_BUFFERS)
 262                 nr_pages = PIPE_BUFFERS;
 263
 264         /*
 265          * Initiate read-ahead on this page range. however, don't call into
 266          * read-ahead if this is a non-zero offset (we are likely doing small
 267          * chunk splice and the page is already there) for a single page.
 268          */
 269         if (!offset || nr_pages > 1)
 270                 do_page_cache_readahead(mapping, in, index, nr_pages);
 271
 272         /*
 273          * Now fill in the holes:
 274          */
 275         error = 0;
 276         bytes = 0;
 277         for (i = 0; i < nr_pages; i++, index++) {
 278 find_page:
 279                 /*
 280                  * lookup the page for this index
 281                  */
 282                 page = find_get_page(mapping, index);
 283                 if (!page) {
 284                         /*
 285                          * page didn't exist, allocate one
 286                          */
 287                         page = page_cache_alloc_cold(mapping);
 288                         if (!page)
 289                                 break;
 290
 291                         error = add_to_page_cache_lru(page, mapping, index,
 292                                                 mapping_gfp_mask(mapping));
 293                         if (unlikely(error)) {
 294                                 page_cache_release(page);
 295                                 break;
 296                         }
 297
 298                         goto readpage;
 299                 }
 300
 301                 /*
 302                  * If the page isn't uptodate, we may need to start io on it
 303                  */
 304                 if (!PageUptodate(page)) {
 305                         /*
 306                          * If in nonblock mode then dont block on waiting
 307                          * for an in-flight io page
 308                          */
 309                         if (flags & SPLICE_F_NONBLOCK)
 310                                 break;
 311
 312                         lock_page(page);
 313
 314                         /*
 315                          * page was truncated, stop here. if this isn't the
 316                          * first page, we'll just complete what we already
 317                          * added
 318                          */
 319                         if (!page->mapping) {
 320                                 unlock_page(page);
 321                                 page_cache_release(page);
 322                                 break;
 323                         }
 324                         /*
 325                          * page was already under io and is now done, great
 326                          */
 327                         if (PageUptodate(page)) {
 328                                 unlock_page(page);
 329                                 goto fill_it;
 330                         }
 331
 332 readpage:
 333                         /*
 334                          * need to read in the page
 335                          */
 336                         error = mapping->a_ops->readpage(in, page);
 337
 338                         if (unlikely(error)) {
 339                                 page_cache_release(page);
 340                                 if (error == AOP_TRUNCATED_PAGE)
 341                                         goto find_page;
 342                                 break;
 343                         }
 344
 345                         /*
 346                          * i_size must be checked after ->readpage().
 347                          */
 348                         isize = i_size_read(mapping->host);
 349                         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
 350                         if (unlikely(!isize || index > end_index)) {
 351                                 page_cache_release(page);
 352                                 break;
 353                         }
 354
 355                         /*
 356                          * if this is the last page, see if we need to shrink
 357                          * the length and stop
 358                          */
 359                         if (end_index == index) {
 360                                 loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK);
 361                                 if (bytes + loff > isize) {
 362                                         page_cache_release(page);
 363                                         break;
 364                                 }
 365                                 /*
 366                                  * force quit after adding this page
 367                                  */
 368                                 nr_pages = i;
 369                         }
 370                 }
 371 fill_it:
 372                 pages[i] = page;
 373                 bytes += PAGE_CACHE_SIZE - loff;
 374                 loff = 0;
 375         }
 376
 377         if (i)
 378                 return move_to_pipe(pipe, pages, i, bytes, offset, flags);
 379
 380         return error;
 381 }
 382
 383 /**
 384  * generic_file_splice_read - splice data from file to a pipe
 385  * @in:         file to splice from
 386  * @pipe:       pipe to splice to
 387  * @len:        number of bytes to splice
 388  * @flags:      splice modifier flags
 389  *
 390  * Will read pages from given file and fill them into a pipe.
 391  */
 392 ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
 393                                  struct pipe_inode_info *pipe, size_t len,
 394                                  unsigned int flags)
 395 {
 396         ssize_t spliced;
 397         int ret;
 398
 399         ret = 0;
 400         spliced = 0;
 401
 402         while (len) {
 403                 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
 404
 405                 if (ret < 0)
 406                         break;
 407                 else if (!ret) {
 408                         if (spliced)
 409                                 break;
 410                         if (flags & SPLICE_F_NONBLOCK) {
 411                                 ret = -EAGAIN;
 412                                 break;
 413                         }
 414                 }
 415
 416                 *ppos += ret;
 417                 len -= ret;
 418                 spliced += ret;
 419         }
 420
 421         if (spliced)
 422                 return spliced;
 423
 424         return ret;
 425 }
 426
 427 EXPORT_SYMBOL(generic_file_splice_read);
 428
 429 /*
 430  * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
 431  * using sendpage().
 432  */
 433 static int pipe_to_sendpage(struct pipe_inode_info *info,
 434                             struct pipe_buffer *buf, struct splice_desc *sd)
 435 {
 436         struct file *file = sd->file;
 437         loff_t pos = sd->pos;
 438         unsigned int offset;
 439         ssize_t ret;
 440         void *ptr;
 441         int more;
 442
 443         /*
 444          * Sub-optimal, but we are limited by the pipe ->map. We don't
 445          * need a kmap'ed buffer here, we just want to make sure we
 446          * have the page pinned if the pipe page originates from the
 447          * page cache.
 448          */
 449         ptr = buf->ops->map(file, info, buf);
 450         if (IS_ERR(ptr))
 451                 return PTR_ERR(ptr);
 452
 453         offset = pos & ~PAGE_CACHE_MASK;
 454         more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
 455
 456         ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more);
 457
 458         buf->ops->unmap(info, buf);
 459         if (ret == sd->len)
 460                 return 0;
 461
 462         return -EIO;
 463 }
 464
 465 /*
 466  * This is a little more tricky than the file -> pipe splicing. There are
 467  * basically three cases:
 468  *
 469  *      - Destination page already exists in the address space and there
 470  *        are users of it. For that case we have no other option that
 471  *        copying the data. Tough luck.
 472  *      - Destination page already exists in the address space, but there
 473  *        are no users of it. Make sure it's uptodate, then drop it. Fall
 474  *        through to last case.
 475  *      - Destination page does not exist, we can add the pipe page to
 476  *        the page cache and avoid the copy.
 477  *
 478  * If asked to move pages to the output file (SPLICE_F_MOVE is set in
 479  * sd->flags), we attempt to migrate pages from the pipe to the output
 480  * file address space page cache. This is possible if no one else has
 481  * the pipe page referenced outside of the pipe and page cache. If
 482  * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
 483  * a new page in the output file page cache and fill/dirty that.
 484  */
 485 static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf,
 486                         struct splice_desc *sd)
 487 {
 488         struct file *file = sd->file;
 489         struct address_space *mapping = file->f_mapping;
 490         gfp_t gfp_mask = mapping_gfp_mask(mapping);
 491         unsigned int offset;
 492         struct page *page;
 493         pgoff_t index;
 494         char *src;
 495         int ret;
 496
 497         /*
 498          * make sure the data in this buffer is uptodate
 499          */
 500         src = buf->ops->map(file, info, buf);
 501         if (IS_ERR(src))
 502                 return PTR_ERR(src);
 503
 504         index = sd->pos >> PAGE_CACHE_SHIFT;
 505         offset = sd->pos & ~PAGE_CACHE_MASK;
 506
 507         /*
 508          * Reuse buf page, if SPLICE_F_MOVE is set.
 509          */
 510         if (sd->flags & SPLICE_F_MOVE) {
 511                 /*
 512                  * If steal succeeds, buf->page is now pruned from the vm
 513                  * side (LRU and page cache) and we can reuse it. The page
 514                  * will also be looked on successful return.
 515                  */
 516                 if (buf->ops->steal(info, buf))
 517                         goto find_page;
 518
 519                 page = buf->page;
 520                 if (add_to_page_cache(page, mapping, index, gfp_mask))
 521                         goto find_page;
 522
 523                 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
 524                         lru_cache_add(page);
 525         } else {
 526 find_page:
 527                 page = find_lock_page(mapping, index);
 528                 if (!page) {
 529                         ret = -ENOMEM;
 530                         page = page_cache_alloc_cold(mapping);
 531                         if (unlikely(!page))
 532                                 goto out_nomem;
 533
 534                         /*
 535                          * This will also lock the page
 536                          */
 537                         ret = add_to_page_cache_lru(page, mapping, index,
 538                                                     gfp_mask);
 539                         if (unlikely(ret))
 540                                 goto out;
 541                 }
 542
 543                 /*
 544                  * We get here with the page locked. If the page is also
 545                  * uptodate, we don't need to do more. If it isn't, we
 546                  * may need to bring it in if we are not going to overwrite
 547                  * the full page.
 548                  */
 549                 if (!PageUptodate(page)) {
 550                         if (sd->len < PAGE_CACHE_SIZE) {
 551                                 ret = mapping->a_ops->readpage(file, page);
 552                                 if (unlikely(ret))
 553                                         goto out;
 554
 555                                 lock_page(page);
 556
 557                                 if (!PageUptodate(page)) {
 558                                         /*
 559                                          * Page got invalidated, repeat.
 560                                          */
 561                                         if (!page->mapping) {
 562                                                 unlock_page(page);
 563                                                 page_cache_release(page);
 564                                                 goto find_page;
 565                                         }
 566                                         ret = -EIO;
 567                                         goto out;
 568                                 }
 569                         } else
 570                                 SetPageUptodate(page);
 571                 }
 572         }
 573
 574         ret = mapping->a_ops->prepare_write(file, page, 0, sd->len);
 575         if (ret == AOP_TRUNCATED_PAGE) {
 576                 page_cache_release(page);
 577                 goto find_page;
 578         } else if (ret)
 579                 goto out;
 580
 581         if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) {
 582                 char *dst = kmap_atomic(page, KM_USER0);
 583
 584                 memcpy(dst + offset, src + buf->offset, sd->len);
 585                 flush_dcache_page(page);
 586                 kunmap_atomic(dst, KM_USER0);
 587         }
 588
 589         ret = mapping->a_ops->commit_write(file, page, 0, sd->len);
 590         if (ret == AOP_TRUNCATED_PAGE) {
 591                 page_cache_release(page);
 592                 goto find_page;
 593         } else if (ret)
 594                 goto out;
 595
 596         mark_page_accessed(page);
 597         balance_dirty_pages_ratelimited(mapping);
 598 out:
 599         if (!(buf->flags & PIPE_BUF_FLAG_STOLEN))
 600                 page_cache_release(page);
 601
 602         unlock_page(page);
 603 out_nomem:
 604         buf->ops->unmap(info, buf);
 605         return ret;
 606 }
 607
 608 typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *,
 609                            struct splice_desc *);
 610
 611 /*
 612  * Pipe input worker. Most of this logic works like a regular pipe, the
 613  * key here is the 'actor' worker passed in that actually moves the data
 614  * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
 615  */
 616 static ssize_t move_from_pipe(struct pipe_inode_info *pipe, struct file *out,
 617                               loff_t *ppos, size_t len, unsigned int flags,
 618                               splice_actor *actor)
 619 {
 620         int ret, do_wakeup, err;
 621         struct splice_desc sd;
 622
 623         ret = 0;
 624         do_wakeup = 0;
 625
 626         sd.total_len = len;
 627         sd.flags = flags;
 628         sd.file = out;
 629         sd.pos = *ppos;
 630
 631         if (pipe->inode)
 632                 mutex_lock(&pipe->inode->i_mutex);
 633
 634         for (;;) {
 635                 if (pipe->nrbufs) {
 636                         struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
 637                         struct pipe_buf_operations *ops = buf->ops;
 638
 639                         sd.len = buf->len;
 640                         if (sd.len > sd.total_len)
 641                                 sd.len = sd.total_len;
 642
 643                         err = actor(pipe, buf, &sd);
 644                         if (err) {
 645                                 if (!ret && err != -ENODATA)
 646                                         ret = err;
 647
 648                                 break;
 649                         }
 650
 651                         ret += sd.len;
 652                         buf->offset += sd.len;
 653                         buf->len -= sd.len;
 654
 655                         if (!buf->len) {
 656                                 buf->ops = NULL;
 657                                 ops->release(pipe, buf);
 658                                 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
 659                                 pipe->nrbufs--;
 660                                 if (pipe->inode)
 661                                         do_wakeup = 1;
 662                         }
 663
 664                         sd.pos += sd.len;
 665                         sd.total_len -= sd.len;
 666                         if (!sd.total_len)
 667                                 break;
 668                 }
 669
 670                 if (pipe->nrbufs)
 671                         continue;
 672                 if (!pipe->writers)
 673                         break;
 674                 if (!pipe->waiting_writers) {
 675                         if (ret)
 676                                 break;
 677                 }
 678
 679                 if (flags & SPLICE_F_NONBLOCK) {
 680                         if (!ret)
 681                                 ret = -EAGAIN;
 682                         break;
 683                 }
 684
 685                 if (signal_pending(current)) {
 686                         if (!ret)
 687                                 ret = -ERESTARTSYS;
 688                         break;
 689                 }
 690
 691                 if (do_wakeup) {
 692                         smp_mb();
 693                         if (waitqueue_active(&pipe->wait))
 694                                 wake_up_interruptible_sync(&pipe->wait);
 695                         kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 696                         do_wakeup = 0;
 697                 }
 698
 699                 pipe_wait(pipe);
 700         }
 701
 702         if (pipe->inode)
 703                 mutex_unlock(&pipe->inode->i_mutex);
 704
 705         if (do_wakeup) {
 706                 smp_mb();
 707                 if (waitqueue_active(&pipe->wait))
 708                         wake_up_interruptible(&pipe->wait);
 709                 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 710         }
 711
 712         return ret;
 713 }
 714
 715 /**
 716  * generic_file_splice_write - splice data from a pipe to a file
 717  * @pipe:       pipe info
 718  * @out:        file to write to
 719  * @len:        number of bytes to splice
 720  * @flags:      splice modifier flags
 721  *
 722  * Will either move or copy pages (determined by @flags options) from
 723  * the given pipe inode to the given file.
 724  *
 725  */
 726 ssize_t
 727 generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
 728                           loff_t *ppos, size_t len, unsigned int flags)
 729 {
 730         struct address_space *mapping = out->f_mapping;
 731         ssize_t ret;
 732
 733         ret = move_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
 734         if (ret > 0) {
 735                 struct inode *inode = mapping->host;
 736
 737                 *ppos += ret;
 738
 739                 /*
 740                  * If file or inode is SYNC and we actually wrote some data,
 741                  * sync it.
 742                  */
 743                 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
 744                         int err;
 745
 746                         mutex_lock(&inode->i_mutex);
 747                         err = generic_osync_inode(inode, mapping,
 748                                                   OSYNC_METADATA|OSYNC_DATA);
 749                         mutex_unlock(&inode->i_mutex);
 750
 751                         if (err)
 752                                 ret = err;
 753                 }
 754         }
 755
 756         return ret;
 757 }
 758
 759 EXPORT_SYMBOL(generic_file_splice_write);
 760
 761 /**
 762  * generic_splice_sendpage - splice data from a pipe to a socket
 763  * @inode:      pipe inode
 764  * @out:        socket to write to
 765  * @len:        number of bytes to splice
 766  * @flags:      splice modifier flags
 767  *
 768  * Will send @len bytes from the pipe to a network socket. No data copying
 769  * is involved.
 770  *
 771  */
 772 ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
 773                                 loff_t *ppos, size_t len, unsigned int flags)
 774 {
 775         return move_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
 776 }
 777
 778 EXPORT_SYMBOL(generic_splice_sendpage);
 779
 780 /*
 781  * Attempt to initiate a splice from pipe to file.
 782  */
 783 static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 784                            loff_t *ppos, size_t len, unsigned int flags)
 785 {
 786         int ret;
 787
 788         if (unlikely(!out->f_op || !out->f_op->splice_write))
 789                 return -EINVAL;
 790
 791         if (unlikely(!(out->f_mode & FMODE_WRITE)))
 792                 return -EBADF;
 793
 794         ret = rw_verify_area(WRITE, out, ppos, len);
 795         if (unlikely(ret < 0))
 796                 return ret;
 797
 798         return out->f_op->splice_write(pipe, out, ppos, len, flags);
 799 }
 800
 801 /*
 802  * Attempt to initiate a splice from a file to a pipe.
 803  */
 804 static long do_splice_to(struct file *in, loff_t *ppos,
 805                          struct pipe_inode_info *pipe, size_t len,
 806                          unsigned int flags)
 807 {
 808         loff_t isize, left;
 809         int ret;
 810
 811         if (unlikely(!in->f_op || !in->f_op->splice_read))
 812                 return -EINVAL;
 813
 814         if (unlikely(!(in->f_mode & FMODE_READ)))
 815                 return -EBADF;
 816
 817         ret = rw_verify_area(READ, in, ppos, len);
 818         if (unlikely(ret < 0))
 819                 return ret;
 820
 821         isize = i_size_read(in->f_mapping->host);
 822         if (unlikely(*ppos >= isize))
 823                 return 0;
 824
 825         left = isize - *ppos;
 826         if (unlikely(left < len))
 827                 len = left;
 828
 829         return in->f_op->splice_read(in, ppos, pipe, len, flags);
 830 }
 831
 832 long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
 833                       size_t len, unsigned int flags)
 834 {
 835         struct pipe_inode_info *pipe;
 836         long ret, bytes;
 837         loff_t out_off;
 838         umode_t i_mode;
 839         int i;
 840
 841         /*
 842          * We require the input being a regular file, as we don't want to
 843          * randomly drop data for eg socket -> socket splicing. Use the
 844          * piped splicing for that!
 845          */
 846         i_mode = in->f_dentry->d_inode->i_mode;
 847         if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
 848                 return -EINVAL;
 849
 850         /*
 851          * neither in nor out is a pipe, setup an internal pipe attached to
 852          * 'out' and transfer the wanted data from 'in' to 'out' through that
 853          */
 854         pipe = current->splice_pipe;
 855         if (unlikely(!pipe)) {
 856                 pipe = alloc_pipe_info(NULL);
 857                 if (!pipe)
 858                         return -ENOMEM;
 859
 860                 /*
 861                  * We don't have an immediate reader, but we'll read the stuff
 862                  * out of the pipe right after the move_to_pipe(). So set
 863                  * PIPE_READERS appropriately.
 864                  */
 865                 pipe->readers = 1;
 866
 867                 current->splice_pipe = pipe;
 868         }
 869
 870         /*
 871          * Do the splice.
 872          */
 873         ret = 0;
 874         bytes = 0;
 875         out_off = 0;
 876
 877         while (len) {
 878                 size_t read_len, max_read_len;
 879
 880                 /*
 881                  * Do at most PIPE_BUFFERS pages worth of transfer:
 882                  */
 883                 max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE));
 884
 885                 ret = do_splice_to(in, ppos, pipe, max_read_len, flags);
 886                 if (unlikely(ret < 0))
 887                         goto out_release;
 888
 889                 read_len = ret;
 890
 891                 /*
 892                  * NOTE: nonblocking mode only applies to the input. We
 893                  * must not do the output in nonblocking mode as then we
 894                  * could get stuck data in the internal pipe:
 895                  */
 896                 ret = do_splice_from(pipe, out, &out_off, read_len,
 897                                      flags & ~SPLICE_F_NONBLOCK);
 898                 if (unlikely(ret < 0))
 899                         goto out_release;
 900
 901                 bytes += ret;
 902                 len -= ret;
 903
 904                 /*
 905                  * In nonblocking mode, if we got back a short read then
 906                  * that was due to either an IO error or due to the
 907                  * pagecache entry not being there. In the IO error case
 908                  * the _next_ splice attempt will produce a clean IO error
 909                  * return value (not a short read), so in both cases it's
 910                  * correct to break out of the loop here:
 911                  */
 912                 if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len))
 913                         break;
 914         }
 915
 916         pipe->nrbufs = pipe->curbuf = 0;
 917
 918         return bytes;
 919
 920 out_release:
 921         /*
 922          * If we did an incomplete transfer we must release
 923          * the pipe buffers in question:
 924          */
 925         for (i = 0; i < PIPE_BUFFERS; i++) {
 926                 struct pipe_buffer *buf = pipe->bufs + i;
 927
 928                 if (buf->ops) {
 929                         buf->ops->release(pipe, buf);
 930                         buf->ops = NULL;
 931                 }
 932         }
 933         pipe->nrbufs = pipe->curbuf = 0;
 934
 935         /*
 936          * If we transferred some data, return the number of bytes:
 937          */
 938         if (bytes > 0)
 939                 return bytes;
 940
 941         return ret;
 942 }
 943
 944 EXPORT_SYMBOL(do_splice_direct);
 945
 946 /*
 947  * Determine where to splice to/from.
 948  */
 949 static long do_splice(struct file *in, loff_t __user *off_in,
 950                       struct file *out, loff_t __user *off_out,
 951                       size_t len, unsigned int flags)
 952 {
 953         struct pipe_inode_info *pipe;
 954         loff_t offset, *off;
 955         long ret;
 956
 957         pipe = in->f_dentry->d_inode->i_pipe;
 958         if (pipe) {
 959                 if (off_in)
 960                         return -ESPIPE;
 961                 if (off_out) {
 962                         if (out->f_op->llseek == no_llseek)
 963                                 return -EINVAL;
 964                         if (copy_from_user(&offset, off_out, sizeof(loff_t)))
 965                                 return -EFAULT;
 966                         off = &offset;
 967                 } else
 968                         off = &out->f_pos;
 969
 970                 ret = do_splice_from(pipe, out, off, len, flags);
 971
 972                 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
 973                         ret = -EFAULT;
 974
 975                 return ret;
 976         }
 977
 978         pipe = out->f_dentry->d_inode->i_pipe;
 979         if (pipe) {
 980                 if (off_out)
 981                         return -ESPIPE;
 982                 if (off_in) {
 983                         if (in->f_op->llseek == no_llseek)
 984                                 return -EINVAL;
 985                         if (copy_from_user(&offset, off_in, sizeof(loff_t)))
 986                                 return -EFAULT;
 987                         off = &offset;
 988                 } else
 989                         off = &in->f_pos;
 990
 991                 ret = do_splice_to(in, off, pipe, len, flags);
 992
 993                 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
 994                         ret = -EFAULT;
 995
 996                 return ret;
 997         }
 998
 999         return -EINVAL;
1000 }
1001
1002 asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1003                            int fd_out, loff_t __user *off_out,
1004                            size_t len, unsigned int flags)
1005 {
1006         long error;
1007         struct file *in, *out;
1008         int fput_in, fput_out;
1009
1010         if (unlikely(!len))
1011                 return 0;
1012
1013         error = -EBADF;
1014         in = fget_light(fd_in, &fput_in);
1015         if (in) {
1016                 if (in->f_mode & FMODE_READ) {
1017                         out = fget_light(fd_out, &fput_out);
1018                         if (out) {
1019                                 if (out->f_mode & FMODE_WRITE)
1020                                         error = do_splice(in, off_in,
1021                                                           out, off_out,
1022                                                           len, flags);
1023                                 fput_light(out, fput_out);
1024                         }
1025                 }
1026
1027                 fput_light(in, fput_in);
1028         }
1029
1030         return error;
1031 }
1032
1033 /*
1034  * Link contents of ipipe to opipe.
1035  */
1036 static int link_pipe(struct pipe_inode_info *ipipe,
1037                      struct pipe_inode_info *opipe,
1038                      size_t len, unsigned int flags)
1039 {
1040         struct pipe_buffer *ibuf, *obuf;
1041         int ret, do_wakeup, i, ipipe_first;
1042
1043         ret = do_wakeup = ipipe_first = 0;
1044
1045         /*
1046          * Potential ABBA deadlock, work around it by ordering lock
1047          * grabbing by inode address. Otherwise two different processes
1048          * could deadlock (one doing tee from A -> B, the other from B -> A).
1049          */
1050         if (ipipe->inode < opipe->inode) {
1051                 ipipe_first = 1;
1052                 mutex_lock(&ipipe->inode->i_mutex);
1053                 mutex_lock(&opipe->inode->i_mutex);
1054         } else {
1055                 mutex_lock(&opipe->inode->i_mutex);
1056                 mutex_lock(&ipipe->inode->i_mutex);
1057         }
1058
1059         for (i = 0;; i++) {
1060                 if (!opipe->readers) {
1061                         send_sig(SIGPIPE, current, 0);
1062                         if (!ret)
1063                                 ret = -EPIPE;
1064                         break;
1065                 }
1066                 if (ipipe->nrbufs - i) {
1067                         ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1068
1069                         /*
1070                          * If we have room, fill this buffer
1071                          */
1072                         if (opipe->nrbufs < PIPE_BUFFERS) {
1073                                 int nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1074
1075                                 /*
1076                                  * Get a reference to this pipe buffer,
1077                                  * so we can copy the contents over.
1078                                  */
1079                                 ibuf->ops->get(ipipe, ibuf);
1080
1081                                 obuf = opipe->bufs + nbuf;
1082                                 *obuf = *ibuf;
1083
1084                                 if (obuf->len > len)
1085                                         obuf->len = len;
1086
1087                                 opipe->nrbufs++;
1088                                 do_wakeup = 1;
1089                                 ret += obuf->len;
1090                                 len -= obuf->len;
1091
1092                                 if (!len)
1093                                         break;
1094                                 if (opipe->nrbufs < PIPE_BUFFERS)
1095                                         continue;
1096                         }
1097
1098                         /*
1099                          * We have input available, but no output room.
1100                          * If we already copied data, return that. If we
1101                          * need to drop the opipe lock, it must be ordered
1102                          * last to avoid deadlocks.
1103                          */
1104                         if ((flags & SPLICE_F_NONBLOCK) || !ipipe_first) {
1105                                 if (!ret)
1106                                         ret = -EAGAIN;
1107                                 break;
1108                         }
1109                         if (signal_pending(current)) {
1110                                 if (!ret)
1111                                         ret = -ERESTARTSYS;
1112                                 break;
1113                         }
1114                         if (do_wakeup) {
1115                                 smp_mb();
1116                                 if (waitqueue_active(&opipe->wait))
1117                                         wake_up_interruptible(&opipe->wait);
1118                                 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1119                                 do_wakeup = 0;
1120                         }
1121
1122                         opipe->waiting_writers++;
1123                         pipe_wait(opipe);
1124                         opipe->waiting_writers--;
1125                         continue;
1126                 }
1127
1128                 /*
1129                  * No input buffers, do the usual checks for available
1130                  * writers and blocking and wait if necessary
1131                  */
1132                 if (!ipipe->writers)
1133                         break;
1134                 if (!ipipe->waiting_writers) {
1135                         if (ret)
1136                                 break;
1137                 }
1138                 /*
1139                  * pipe_wait() drops the ipipe mutex. To avoid deadlocks
1140                  * with another process, we can only safely do that if
1141                  * the ipipe lock is ordered last.
1142                  */
1143                 if ((flags & SPLICE_F_NONBLOCK) || ipipe_first) {
1144                         if (!ret)
1145                                 ret = -EAGAIN;
1146                         break;
1147                 }
1148                 if (signal_pending(current)) {
1149                         if (!ret)
1150                                 ret = -ERESTARTSYS;
1151                         break;
1152                 }
1153
1154                 if (waitqueue_active(&ipipe->wait))
1155                         wake_up_interruptible_sync(&ipipe->wait);
1156                 kill_fasync(&ipipe->fasync_writers, SIGIO, POLL_OUT);
1157
1158                 pipe_wait(ipipe);
1159         }
1160
1161         mutex_unlock(&ipipe->inode->i_mutex);
1162         mutex_unlock(&opipe->inode->i_mutex);
1163
1164         if (do_wakeup) {
1165                 smp_mb();
1166                 if (waitqueue_active(&opipe->wait))
1167                         wake_up_interruptible(&opipe->wait);
1168                 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1169         }
1170
1171         return ret;
1172 }
1173
1174 /*
1175  * This is a tee(1) implementation that works on pipes. It doesn't copy
1176  * any data, it simply references the 'in' pages on the 'out' pipe.
1177  * The 'flags' used are the SPLICE_F_* variants, currently the only
1178  * applicable one is SPLICE_F_NONBLOCK.
1179  */
1180 static long do_tee(struct file *in, struct file *out, size_t len,
1181                    unsigned int flags)
1182 {
1183         struct pipe_inode_info *ipipe = in->f_dentry->d_inode->i_pipe;
1184         struct pipe_inode_info *opipe = out->f_dentry->d_inode->i_pipe;
1185
1186         /*
1187          * Link ipipe to the two output pipes, consuming as we go along.
1188          */
1189         if (ipipe && opipe)
1190                 return link_pipe(ipipe, opipe, len, flags);
1191
1192         return -EINVAL;
1193 }
1194
1195 asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1196 {
1197         struct file *in;
1198         int error, fput_in;
1199
1200         if (unlikely(!len))
1201                 return 0;
1202
1203         error = -EBADF;
1204         in = fget_light(fdin, &fput_in);
1205         if (in) {
1206                 if (in->f_mode & FMODE_READ) {
1207                         int fput_out;
1208                         struct file *out = fget_light(fdout, &fput_out);
1209
1210                         if (out) {
1211                                 if (out->f_mode & FMODE_WRITE)
1212                                         error = do_tee(in, out, len, flags);
1213                                 fput_light(out, fput_out);
1214                         }
1215                 }
1216                 fput_light(in, fput_in);
1217         }
1218
1219         return error;
1220 }