fs/mpage.c

   1 /*
   2  * fs/mpage.c
   3  *
   4  * Copyright (C) 2002, Linus Torvalds.
   5  *
   6  * Contains functions related to preparing and submitting BIOs which contain
   7  * multiple pagecache pages.
   8  *
   9  * 15May2002    Andrew Morton
  10  *              Initial version
  11  * 27Jun2002    axboe@suse.de
  12  *              use bio_add_page() to build bio's just the right size
  13  */
  14
  15 #include <linux/kernel.h>
  16 #include <linux/module.h>
  17 #include <linux/mm.h>
  18 #include <linux/kdev_t.h>
  19 #include <linux/bio.h>
  20 #include <linux/fs.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/highmem.h>
  24 #include <linux/prefetch.h>
  25 #include <linux/mpage.h>
  26 #include <linux/writeback.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/pagevec.h>
  29
  30 /*
  31  * I/O completion handler for multipage BIOs.
  32  *
  33  * The mpage code never puts partial pages into a BIO (except for end-of-file).
  34  * If a page does not map to a contiguous run of blocks then it simply falls
  35  * back to block_read_full_page().
  36  *
  37  * Why is this?  If a page's completion depends on a number of different BIOs
  38  * which can complete in any order (or at the same time) then determining the
  39  * status of that page is hard.  See end_buffer_async_read() for the details.
  40  * There is no point in duplicating all that complexity.
  41  */
  42 static void mpage_end_io_read(struct bio *bio, int err)
  43 {
  44         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  45         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
  46
  47         do {
  48                 struct page *page = bvec->bv_page;
  49
  50                 if (--bvec >= bio->bi_io_vec)
  51                         prefetchw(&bvec->bv_page->flags);
  52
  53                 if (uptodate) {
  54                         SetPageUptodate(page);
  55                 } else {
  56                         ClearPageUptodate(page);
  57                         SetPageError(page);
  58                 }
  59                 unlock_page(page);
  60         } while (bvec >= bio->bi_io_vec);
  61         bio_put(bio);
  62 }
  63
  64 static void mpage_end_io_write(struct bio *bio, int err)
  65 {
  66         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  67         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
  68
  69         do {
  70                 struct page *page = bvec->bv_page;
  71
  72                 if (--bvec >= bio->bi_io_vec)
  73                         prefetchw(&bvec->bv_page->flags);
  74
  75                 if (!uptodate){
  76                         SetPageError(page);
  77                         if (page->mapping)
  78                                 set_bit(AS_EIO, &page->mapping->flags);
  79                 }
  80                 end_page_writeback(page);
  81         } while (bvec >= bio->bi_io_vec);
  82         bio_put(bio);
  83 }
  84
  85 struct bio *mpage_bio_submit(int rw, struct bio *bio)
  86 {
  87         bio->bi_end_io = mpage_end_io_read;
  88         if (rw == WRITE)
  89                 bio->bi_end_io = mpage_end_io_write;
  90         submit_bio(rw, bio);
  91         return NULL;
  92 }
  93 EXPORT_SYMBOL(mpage_bio_submit);
  94
  95 static struct bio *
  96 mpage_alloc(struct block_device *bdev,
  97                 sector_t first_sector, int nr_vecs,
  98                 gfp_t gfp_flags)
  99 {
 100         struct bio *bio;
 101
 102         bio = bio_alloc(gfp_flags, nr_vecs);
 103
 104         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
 105                 while (!bio && (nr_vecs /= 2))
 106                         bio = bio_alloc(gfp_flags, nr_vecs);
 107         }
 108
 109         if (bio) {
 110                 bio->bi_bdev = bdev;
 111                 bio->bi_sector = first_sector;
 112         }
 113         return bio;
 114 }
 115
 116 /*
 117  * support function for mpage_readpages.  The fs supplied get_block might
 118  * return an up to date buffer.  This is used to map that buffer into
 119  * the page, which allows readpage to avoid triggering a duplicate call
 120  * to get_block.
 121  *
 122  * The idea is to avoid adding buffers to pages that don't already have
 123  * them.  So when the buffer is up to date and the page size == block size,
 124  * this marks the page up to date instead of adding new buffers.
 125  */
 126 static void
 127 map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
 128 {
 129         struct inode *inode = page->mapping->host;
 130         struct buffer_head *page_bh, *head;
 131         int block = 0;
 132
 133         if (!page_has_buffers(page)) {
 134                 /*
 135                  * don't make any buffers if there is only one buffer on
 136                  * the page and the page just needs to be set up to date
 137                  */
 138                 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
 139                     buffer_uptodate(bh)) {
 140                         SetPageUptodate(page);
 141                         return;
 142                 }
 143                 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
 144         }
 145         head = page_buffers(page);
 146         page_bh = head;
 147         do {
 148                 if (block == page_block) {
 149                         page_bh->b_state = bh->b_state;
 150                         page_bh->b_bdev = bh->b_bdev;
 151                         page_bh->b_blocknr = bh->b_blocknr;
 152                         break;
 153                 }
 154                 page_bh = page_bh->b_this_page;
 155                 block++;
 156         } while (page_bh != head);
 157 }
 158
 159 /*
 160  * This is the worker routine which does all the work of mapping the disk
 161  * blocks and constructs largest possible bios, submits them for IO if the
 162  * blocks are not contiguous on the disk.
 163  *
 164  * We pass a buffer_head back and forth and use its buffer_mapped() flag to
 165  * represent the validity of its disk mapping and to decide when to do the next
 166  * get_block() call.
 167  */
 168 static struct bio *
 169 do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
 170                 sector_t *last_block_in_bio, struct buffer_head *map_bh,
 171                 unsigned long *first_logical_block, get_block_t get_block)
 172 {
 173         struct inode *inode = page->mapping->host;
 174         const unsigned blkbits = inode->i_blkbits;
 175         const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
 176         const unsigned blocksize = 1 << blkbits;
 177         sector_t block_in_file;
 178         sector_t last_block;
 179         sector_t last_block_in_file;
 180         sector_t blocks[MAX_BUF_PER_PAGE];
 181         unsigned page_block;
 182         unsigned first_hole = blocks_per_page;
 183         struct block_device *bdev = NULL;
 184         int length;
 185         int fully_mapped = 1;
 186         unsigned nblocks;
 187         unsigned relative_block;
 188
 189         if (page_has_buffers(page))
 190                 goto confused;
 191
 192         block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 193         last_block = block_in_file + nr_pages * blocks_per_page;
 194         last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
 195         if (last_block > last_block_in_file)
 196                 last_block = last_block_in_file;
 197         page_block = 0;
 198
 199         /*
 200          * Map blocks using the result from the previous get_blocks call first.
 201          */
 202         nblocks = map_bh->b_size >> blkbits;
 203         if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
 204                         block_in_file < (*first_logical_block + nblocks)) {
 205                 unsigned map_offset = block_in_file - *first_logical_block;
 206                 unsigned last = nblocks - map_offset;
 207
 208                 for (relative_block = 0; ; relative_block++) {
 209                         if (relative_block == last) {
 210                                 clear_buffer_mapped(map_bh);
 211                                 break;
 212                         }
 213                         if (page_block == blocks_per_page)
 214                                 break;
 215                         blocks[page_block] = map_bh->b_blocknr + map_offset +
 216                                                 relative_block;
 217                         page_block++;
 218                         block_in_file++;
 219                 }
 220                 bdev = map_bh->b_bdev;
 221         }
 222
 223         /*
 224          * Then do more get_blocks calls until we are done with this page.
 225          */
 226         map_bh->b_page = page;
 227         while (page_block < blocks_per_page) {
 228                 map_bh->b_state = 0;
 229                 map_bh->b_size = 0;
 230
 231                 if (block_in_file < last_block) {
 232                         map_bh->b_size = (last_block-block_in_file) << blkbits;
 233                         if (get_block(inode, block_in_file, map_bh, 0))
 234                                 goto confused;
 235                         *first_logical_block = block_in_file;
 236                 }
 237
 238                 if (!buffer_mapped(map_bh)) {
 239                         fully_mapped = 0;
 240                         if (first_hole == blocks_per_page)
 241                                 first_hole = page_block;
 242                         page_block++;
 243                         block_in_file++;
 244                         continue;
 245                 }
 246
 247                 /* some filesystems will copy data into the page during
 248                  * the get_block call, in which case we don't want to
 249                  * read it again.  map_buffer_to_page copies the data
 250                  * we just collected from get_block into the page's buffers
 251                  * so readpage doesn't have to repeat the get_block call
 252                  */
 253                 if (buffer_uptodate(map_bh)) {
 254                         map_buffer_to_page(page, map_bh, page_block);
 255                         goto confused;
 256                 }
 257
 258                 if (first_hole != blocks_per_page)
 259                         goto confused;          /* hole -> non-hole */
 260
 261                 /* Contiguous blocks? */
 262                 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
 263                         goto confused;
 264                 nblocks = map_bh->b_size >> blkbits;
 265                 for (relative_block = 0; ; relative_block++) {
 266                         if (relative_block == nblocks) {
 267                                 clear_buffer_mapped(map_bh);
 268                                 break;
 269                         } else if (page_block == blocks_per_page)
 270                                 break;
 271                         blocks[page_block] = map_bh->b_blocknr+relative_block;
 272                         page_block++;
 273                         block_in_file++;
 274                 }
 275                 bdev = map_bh->b_bdev;
 276         }
 277
 278         if (first_hole != blocks_per_page) {
 279                 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
 280                 if (first_hole == 0) {
 281                         SetPageUptodate(page);
 282                         unlock_page(page);
 283                         goto out;
 284                 }
 285         } else if (fully_mapped) {
 286                 SetPageMappedToDisk(page);
 287         }
 288
 289         /*
 290          * This page will go to BIO.  Do we need to send this BIO off first?
 291          */
 292         if (bio && (*last_block_in_bio != blocks[0] - 1))
 293                 bio = mpage_bio_submit(READ, bio);
 294
 295 alloc_new:
 296         if (bio == NULL) {
 297                 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
 298                                 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
 299                                 GFP_KERNEL);
 300                 if (bio == NULL)
 301                         goto confused;
 302         }
 303
 304         length = first_hole << blkbits;
 305         if (bio_add_page(bio, page, length, 0) < length) {
 306                 bio = mpage_bio_submit(READ, bio);
 307                 goto alloc_new;
 308         }
 309
 310         relative_block = block_in_file - *first_logical_block;
 311         nblocks = map_bh->b_size >> blkbits;
 312         if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
 313             (first_hole != blocks_per_page))
 314                 bio = mpage_bio_submit(READ, bio);
 315         else
 316                 *last_block_in_bio = blocks[blocks_per_page - 1];
 317 out:
 318         return bio;
 319
 320 confused:
 321         if (bio)
 322                 bio = mpage_bio_submit(READ, bio);
 323         if (!PageUptodate(page))
 324                 block_read_full_page(page, get_block);
 325         else
 326                 unlock_page(page);
 327         goto out;
 328 }
 329
 330 /**
 331  * mpage_readpages - populate an address space with some pages & start reads against them
 332  * @mapping: the address_space
 333  * @pages: The address of a list_head which contains the target pages.  These
 334  *   pages have their ->index populated and are otherwise uninitialised.
 335  *   The page at @pages->prev has the lowest file offset, and reads should be
 336  *   issued in @pages->prev to @pages->next order.
 337  * @nr_pages: The number of pages at *@pages
 338  * @get_block: The filesystem's block mapper function.
 339  *
 340  * This function walks the pages and the blocks within each page, building and
 341  * emitting large BIOs.
 342  *
 343  * If anything unusual happens, such as:
 344  *
 345  * - encountering a page which has buffers
 346  * - encountering a page which has a non-hole after a hole
 347  * - encountering a page with non-contiguous blocks
 348  *
 349  * then this code just gives up and calls the buffer_head-based read function.
 350  * It does handle a page which has holes at the end - that is a common case:
 351  * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
 352  *
 353  * BH_Boundary explanation:
 354  *
 355  * There is a problem.  The mpage read code assembles several pages, gets all
 356  * their disk mappings, and then submits them all.  That's fine, but obtaining
 357  * the disk mappings may require I/O.  Reads of indirect blocks, for example.
 358  *
 359  * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
 360  * submitted in the following order:
 361  *      12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
 362  *
 363  * because the indirect block has to be read to get the mappings of blocks
 364  * 13,14,15,16.  Obviously, this impacts performance.
 365  *
 366  * So what we do it to allow the filesystem's get_block() function to set
 367  * BH_Boundary when it maps block 11.  BH_Boundary says: mapping of the block
 368  * after this one will require I/O against a block which is probably close to
 369  * this one.  So you should push what I/O you have currently accumulated.
 370  *
 371  * This all causes the disk requests to be issued in the correct order.
 372  */
 373 int
 374 mpage_readpages(struct address_space *mapping, struct list_head *pages,
 375                                 unsigned nr_pages, get_block_t get_block)
 376 {
 377         struct bio *bio = NULL;
 378         unsigned page_idx;
 379         sector_t last_block_in_bio = 0;
 380         struct buffer_head map_bh;
 381         unsigned long first_logical_block = 0;
 382
 383         clear_buffer_mapped(&map_bh);
 384         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
 385                 struct page *page = list_entry(pages->prev, struct page, lru);
 386
 387                 prefetchw(&page->flags);
 388                 list_del(&page->lru);
 389                 if (!add_to_page_cache_lru(page, mapping,
 390                                         page->index, GFP_KERNEL)) {
 391                         bio = do_mpage_readpage(bio, page,
 392                                         nr_pages - page_idx,
 393                                         &last_block_in_bio, &map_bh,
 394                                         &first_logical_block,
 395                                         get_block);
 396                 }
 397                 page_cache_release(page);
 398         }
 399         BUG_ON(!list_empty(pages));
 400         if (bio)
 401                 mpage_bio_submit(READ, bio);
 402         return 0;
 403 }
 404 EXPORT_SYMBOL(mpage_readpages);
 405
 406 /*
 407  * This isn't called much at all
 408  */
 409 int mpage_readpage(struct page *page, get_block_t get_block)
 410 {
 411         struct bio *bio = NULL;
 412         sector_t last_block_in_bio = 0;
 413         struct buffer_head map_bh;
 414         unsigned long first_logical_block = 0;
 415
 416         clear_buffer_mapped(&map_bh);
 417         bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
 418                         &map_bh, &first_logical_block, get_block);
 419         if (bio)
 420                 mpage_bio_submit(READ, bio);
 421         return 0;
 422 }
 423 EXPORT_SYMBOL(mpage_readpage);
 424
 425 /*
 426  * Writing is not so simple.
 427  *
 428  * If the page has buffers then they will be used for obtaining the disk
 429  * mapping.  We only support pages which are fully mapped-and-dirty, with a
 430  * special case for pages which are unmapped at the end: end-of-file.
 431  *
 432  * If the page has no buffers (preferred) then the page is mapped here.
 433  *
 434  * If all blocks are found to be contiguous then the page can go into the
 435  * BIO.  Otherwise fall back to the mapping's writepage().
 436  *
 437  * FIXME: This code wants an estimate of how many pages are still to be
 438  * written, so it can intelligently allocate a suitably-sized BIO.  For now,
 439  * just allocate full-size (16-page) BIOs.
 440  */
 441
 442 int __mpage_writepage(struct page *page, struct writeback_control *wbc,
 443                       void *data)
 444 {
 445         struct mpage_data *mpd = data;
 446         struct bio *bio = mpd->bio;
 447         struct address_space *mapping = page->mapping;
 448         struct inode *inode = page->mapping->host;
 449         const unsigned blkbits = inode->i_blkbits;
 450         unsigned long end_index;
 451         const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
 452         sector_t last_block;
 453         sector_t block_in_file;
 454         sector_t blocks[MAX_BUF_PER_PAGE];
 455         unsigned page_block;
 456         unsigned first_unmapped = blocks_per_page;
 457         struct block_device *bdev = NULL;
 458         int boundary = 0;
 459         sector_t boundary_block = 0;
 460         struct block_device *boundary_bdev = NULL;
 461         int length;
 462         struct buffer_head map_bh;
 463         loff_t i_size = i_size_read(inode);
 464         int ret = 0;
 465
 466         if (page_has_buffers(page)) {
 467                 struct buffer_head *head = page_buffers(page);
 468                 struct buffer_head *bh = head;
 469
 470                 /* If they're all mapped and dirty, do it */
 471                 page_block = 0;
 472                 do {
 473                         BUG_ON(buffer_locked(bh));
 474                         if (!buffer_mapped(bh)) {
 475                                 /*
 476                                  * unmapped dirty buffers are created by
 477                                  * __set_page_dirty_buffers -> mmapped data
 478                                  */
 479                                 if (buffer_dirty(bh))
 480                                         goto confused;
 481                                 if (first_unmapped == blocks_per_page)
 482                                         first_unmapped = page_block;
 483                                 continue;
 484                         }
 485
 486                         if (first_unmapped != blocks_per_page)
 487                                 goto confused;  /* hole -> non-hole */
 488
 489                         if (!buffer_dirty(bh) || !buffer_uptodate(bh))
 490                                 goto confused;
 491                         if (page_block) {
 492                                 if (bh->b_blocknr != blocks[page_block-1] + 1)
 493                                         goto confused;
 494                         }
 495                         blocks[page_block++] = bh->b_blocknr;
 496                         boundary = buffer_boundary(bh);
 497                         if (boundary) {
 498                                 boundary_block = bh->b_blocknr;
 499                                 boundary_bdev = bh->b_bdev;
 500                         }
 501                         bdev = bh->b_bdev;
 502                 } while ((bh = bh->b_this_page) != head);
 503
 504                 if (first_unmapped)
 505                         goto page_is_mapped;
 506
 507                 /*
 508                  * Page has buffers, but they are all unmapped. The page was
 509                  * created by pagein or read over a hole which was handled by
 510                  * block_read_full_page().  If this address_space is also
 511                  * using mpage_readpages then this can rarely happen.
 512                  */
 513                 goto confused;
 514         }
 515
 516         /*
 517          * The page has no buffers: map it to disk
 518          */
 519         BUG_ON(!PageUptodate(page));
 520         block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 521         last_block = (i_size - 1) >> blkbits;
 522         map_bh.b_page = page;
 523         for (page_block = 0; page_block < blocks_per_page; ) {
 524
 525                 map_bh.b_state = 0;
 526                 map_bh.b_size = 1 << blkbits;
 527                 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
 528                         goto confused;
 529                 if (buffer_new(&map_bh))
 530                         unmap_underlying_metadata(map_bh.b_bdev,
 531                                                 map_bh.b_blocknr);
 532                 if (buffer_boundary(&map_bh)) {
 533                         boundary_block = map_bh.b_blocknr;
 534                         boundary_bdev = map_bh.b_bdev;
 535                 }
 536                 if (page_block) {
 537                         if (map_bh.b_blocknr != blocks[page_block-1] + 1)
 538                                 goto confused;
 539                 }
 540                 blocks[page_block++] = map_bh.b_blocknr;
 541                 boundary = buffer_boundary(&map_bh);
 542                 bdev = map_bh.b_bdev;
 543                 if (block_in_file == last_block)
 544                         break;
 545                 block_in_file++;
 546         }
 547         BUG_ON(page_block == 0);
 548
 549         first_unmapped = page_block;
 550
 551 page_is_mapped:
 552         end_index = i_size >> PAGE_CACHE_SHIFT;
 553         if (page->index >= end_index) {
 554                 /*
 555                  * The page straddles i_size.  It must be zeroed out on each
 556                  * and every writepage invokation because it may be mmapped.
 557                  * "A file is mapped in multiples of the page size.  For a file
 558                  * that is not a multiple of the page size, the remaining memory
 559                  * is zeroed when mapped, and writes to that region are not
 560                  * written out to the file."
 561                  */
 562                 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
 563
 564                 if (page->index > end_index || !offset)
 565                         goto confused;
 566                 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
 567         }
 568
 569         /*
 570          * This page will go to BIO.  Do we need to send this BIO off first?
 571          */
 572         if (bio && mpd->last_block_in_bio != blocks[0] - 1)
 573                 bio = mpage_bio_submit(WRITE, bio);
 574
 575 alloc_new:
 576         if (bio == NULL) {
 577                 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
 578                                 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
 579                 if (bio == NULL)
 580                         goto confused;
 581         }
 582
 583         /*
 584          * Must try to add the page before marking the buffer clean or
 585          * the confused fail path above (OOM) will be very confused when
 586          * it finds all bh marked clean (i.e. it will not write anything)
 587          */
 588         length = first_unmapped << blkbits;
 589         if (bio_add_page(bio, page, length, 0) < length) {
 590                 bio = mpage_bio_submit(WRITE, bio);
 591                 goto alloc_new;
 592         }
 593
 594         /*
 595          * OK, we have our BIO, so we can now mark the buffers clean.  Make
 596          * sure to only clean buffers which we know we'll be writing.
 597          */
 598         if (page_has_buffers(page)) {
 599                 struct buffer_head *head = page_buffers(page);
 600                 struct buffer_head *bh = head;
 601                 unsigned buffer_counter = 0;
 602
 603                 do {
 604                         if (buffer_counter++ == first_unmapped)
 605                                 break;
 606                         clear_buffer_dirty(bh);
 607                         bh = bh->b_this_page;
 608                 } while (bh != head);
 609
 610                 /*
 611                  * we cannot drop the bh if the page is not uptodate
 612                  * or a concurrent readpage would fail to serialize with the bh
 613                  * and it would read from disk before we reach the platter.
 614                  */
 615                 if (buffer_heads_over_limit && PageUptodate(page))
 616                         try_to_free_buffers(page);
 617         }
 618
 619         BUG_ON(PageWriteback(page));
 620         set_page_writeback(page);
 621         unlock_page(page);
 622         if (boundary || (first_unmapped != blocks_per_page)) {
 623                 bio = mpage_bio_submit(WRITE, bio);
 624                 if (boundary_block) {
 625                         write_boundary_block(boundary_bdev,
 626                                         boundary_block, 1 << blkbits);
 627                 }
 628         } else {
 629                 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
 630         }
 631         goto out;
 632
 633 confused:
 634         if (bio)
 635                 bio = mpage_bio_submit(WRITE, bio);
 636
 637         if (mpd->use_writepage) {
 638                 ret = mapping->a_ops->writepage(page, wbc);
 639         } else {
 640                 ret = -EAGAIN;
 641                 goto out;
 642         }
 643         /*
 644          * The caller has a ref on the inode, so *mapping is stable
 645          */
 646         mapping_set_error(mapping, ret);
 647 out:
 648         mpd->bio = bio;
 649         return ret;
 650 }
 651 EXPORT_SYMBOL(__mpage_writepage);
 652
 653 /**
 654  * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
 655  * @mapping: address space structure to write
 656  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
 657  * @get_block: the filesystem's block mapper function.
 658  *             If this is NULL then use a_ops->writepage.  Otherwise, go
 659  *             direct-to-BIO.
 660  *
 661  * This is a library function, which implements the writepages()
 662  * address_space_operation.
 663  *
 664  * If a page is already under I/O, generic_writepages() skips it, even
 665  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
 666  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
 667  * and msync() need to guarantee that all the data which was dirty at the time
 668  * the call was made get new I/O started against them.  If wbc->sync_mode is
 669  * WB_SYNC_ALL then we were called for data integrity and we must wait for
 670  * existing IO to complete.
 671  */
 672 int
 673 mpage_writepages(struct address_space *mapping,
 674                 struct writeback_control *wbc, get_block_t get_block)
 675 {
 676         int ret;
 677
 678         if (!get_block)
 679                 ret = generic_writepages(mapping, wbc);
 680         else {
 681                 struct mpage_data mpd = {
 682                         .bio = NULL,
 683                         .last_block_in_bio = 0,
 684                         .get_block = get_block,
 685                         .use_writepage = 1,
 686                 };
 687
 688                 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
 689                 if (mpd.bio)
 690                         mpage_bio_submit(WRITE, mpd.bio);
 691         }
 692         return ret;
 693 }
 694 EXPORT_SYMBOL(mpage_writepages);
 695
 696 int mpage_writepage(struct page *page, get_block_t get_block,
 697         struct writeback_control *wbc)
 698 {
 699         struct mpage_data mpd = {
 700                 .bio = NULL,
 701                 .last_block_in_bio = 0,
 702                 .get_block = get_block,
 703                 .use_writepage = 0,
 704         };
 705         int ret = __mpage_writepage(page, wbc, &mpd);
 706         if (mpd.bio)
 707                 mpage_bio_submit(WRITE, mpd.bio);
 708         return ret;
 709 }
 710 EXPORT_SYMBOL(mpage_writepage);