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/gfp.h>
  20 #include <linux/bio.h>
  21 #include <linux/fs.h>
  22 #include <linux/buffer_head.h>
  23 #include <linux/blkdev.h>
  24 #include <linux/highmem.h>
  25 #include <linux/prefetch.h>
  26 #include <linux/mpage.h>
  27 #include <linux/writeback.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/pagevec.h>
  30 #include <linux/cleancache.h>
  31
  32 /*
  33  * I/O completion handler for multipage BIOs.
  34  *
  35  * The mpage code never puts partial pages into a BIO (except for end-of-file).
  36  * If a page does not map to a contiguous run of blocks then it simply falls
  37  * back to block_read_full_page().
  38  *
  39  * Why is this?  If a page's completion depends on a number of different BIOs
  40  * which can complete in any order (or at the same time) then determining the
  41  * status of that page is hard.  See end_buffer_async_read() for the details.
  42  * There is no point in duplicating all that complexity.
  43  */
  44 static void mpage_end_io(struct bio *bio, int err)
  45 {
  46         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  47         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
  48
  49         do {
  50                 struct page *page = bvec->bv_page;
  51
  52                 if (--bvec >= bio->bi_io_vec)
  53                         prefetchw(&bvec->bv_page->flags);
  54                 if (bio_data_dir(bio) == READ) {
  55                         if (uptodate) {
  56                                 SetPageUptodate(page);
  57                         } else {
  58                                 ClearPageUptodate(page);
  59                                 SetPageError(page);
  60                         }
  61                         unlock_page(page);
  62                 } else { /* bio_data_dir(bio) == WRITE */
  63                         if (!uptodate) {
  64                                 SetPageError(page);
  65                                 if (page->mapping)
  66                                         set_bit(AS_EIO, &page->mapping->flags);
  67                         }
  68                         end_page_writeback(page);
  69                 }
  70         } while (bvec >= bio->bi_io_vec);
  71         bio_put(bio);
  72 }
  73
  74 static struct bio *mpage_bio_submit(int rw, struct bio *bio)
  75 {
  76         bio->bi_end_io = mpage_end_io;
  77         submit_bio(rw, bio);
  78         return NULL;
  79 }
  80
  81 static struct bio *
  82 mpage_alloc(struct block_device *bdev,
  83                 sector_t first_sector, int nr_vecs,
  84                 gfp_t gfp_flags)
  85 {
  86         struct bio *bio;
  87
  88         bio = bio_alloc(gfp_flags, nr_vecs);
  89
  90         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
  91                 while (!bio && (nr_vecs /= 2))
  92                         bio = bio_alloc(gfp_flags, nr_vecs);
  93         }
  94
  95         if (bio) {
  96                 bio->bi_bdev = bdev;
  97                 bio->bi_sector = first_sector;
  98         }
  99         return bio;
 100 }
 101
 102 /*
 103  * support function for mpage_readpages.  The fs supplied get_block might
 104  * return an up to date buffer.  This is used to map that buffer into
 105  * the page, which allows readpage to avoid triggering a duplicate call
 106  * to get_block.
 107  *
 108  * The idea is to avoid adding buffers to pages that don't already have
 109  * them.  So when the buffer is up to date and the page size == block size,
 110  * this marks the page up to date instead of adding new buffers.
 111  */
 112 static void
 113 map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
 114 {
 115         struct inode *inode = page->mapping->host;
 116         struct buffer_head *page_bh, *head;
 117         int block = 0;
 118
 119         if (!page_has_buffers(page)) {
 120                 /*
 121                  * don't make any buffers if there is only one buffer on
 122                  * the page and the page just needs to be set up to date
 123                  */
 124                 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
 125                     buffer_uptodate(bh)) {
 126                         SetPageUptodate(page);
 127                         return;
 128                 }
 129                 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
 130         }
 131         head = page_buffers(page);
 132         page_bh = head;
 133         do {
 134                 if (block == page_block) {
 135                         page_bh->b_state = bh->b_state;
 136                         page_bh->b_bdev = bh->b_bdev;
 137                         page_bh->b_blocknr = bh->b_blocknr;
 138                         break;
 139                 }
 140                 page_bh = page_bh->b_this_page;
 141                 block++;
 142         } while (page_bh != head);
 143 }
 144
 145 /*
 146  * This is the worker routine which does all the work of mapping the disk
 147  * blocks and constructs largest possible bios, submits them for IO if the
 148  * blocks are not contiguous on the disk.
 149  *
 150  * We pass a buffer_head back and forth and use its buffer_mapped() flag to
 151  * represent the validity of its disk mapping and to decide when to do the next
 152  * get_block() call.
 153  */
 154 static struct bio *
 155 do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
 156                 sector_t *last_block_in_bio, struct buffer_head *map_bh,
 157                 unsigned long *first_logical_block, get_block_t get_block)
 158 {
 159         struct inode *inode = page->mapping->host;
 160         const unsigned blkbits = inode->i_blkbits;
 161         const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
 162         const unsigned blocksize = 1 << blkbits;
 163         sector_t block_in_file;
 164         sector_t last_block;
 165         sector_t last_block_in_file;
 166         sector_t blocks[MAX_BUF_PER_PAGE];
 167         unsigned page_block;
 168         unsigned first_hole = blocks_per_page;
 169         struct block_device *bdev = NULL;
 170         int length;
 171         int fully_mapped = 1;
 172         unsigned nblocks;
 173         unsigned relative_block;
 174
 175         if (page_has_buffers(page))
 176                 goto confused;
 177
 178         block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 179         last_block = block_in_file + nr_pages * blocks_per_page;
 180         last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
 181         if (last_block > last_block_in_file)
 182                 last_block = last_block_in_file;
 183         page_block = 0;
 184
 185         /*
 186          * Map blocks using the result from the previous get_blocks call first.
 187          */
 188         nblocks = map_bh->b_size >> blkbits;
 189         if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
 190                         block_in_file < (*first_logical_block + nblocks)) {
 191                 unsigned map_offset = block_in_file - *first_logical_block;
 192                 unsigned last = nblocks - map_offset;
 193
 194                 for (relative_block = 0; ; relative_block++) {
 195                         if (relative_block == last) {
 196                                 clear_buffer_mapped(map_bh);
 197                                 break;
 198                         }
 199                         if (page_block == blocks_per_page)
 200                                 break;
 201                         blocks[page_block] = map_bh->b_blocknr + map_offset +
 202                                                 relative_block;
 203                         page_block++;
 204                         block_in_file++;
 205                 }
 206                 bdev = map_bh->b_bdev;
 207         }
 208
 209         /*
 210          * Then do more get_blocks calls until we are done with this page.
 211          */
 212         map_bh->b_page = page;
 213         while (page_block < blocks_per_page) {
 214                 map_bh->b_state = 0;
 215                 map_bh->b_size = 0;
 216
 217                 if (block_in_file < last_block) {
 218                         map_bh->b_size = (last_block-block_in_file) << blkbits;
 219                         if (get_block(inode, block_in_file, map_bh, 0))
 220                                 goto confused;
 221                         *first_logical_block = block_in_file;
 222                 }
 223
 224                 if (!buffer_mapped(map_bh)) {
 225                         fully_mapped = 0;
 226                         if (first_hole == blocks_per_page)
 227                                 first_hole = page_block;
 228                         page_block++;
 229                         block_in_file++;
 230                         continue;
 231                 }
 232
 233                 /* some filesystems will copy data into the page during
 234                  * the get_block call, in which case we don't want to
 235                  * read it again.  map_buffer_to_page copies the data
 236                  * we just collected from get_block into the page's buffers
 237                  * so readpage doesn't have to repeat the get_block call
 238                  */
 239                 if (buffer_uptodate(map_bh)) {
 240                         map_buffer_to_page(page, map_bh, page_block);
 241                         goto confused;
 242                 }
 243
 244                 if (first_hole != blocks_per_page)
 245                         goto confused;          /* hole -> non-hole */
 246
 247                 /* Contiguous blocks? */
 248                 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
 249                         goto confused;
 250                 nblocks = map_bh->b_size >> blkbits;
 251                 for (relative_block = 0; ; relative_block++) {
 252                         if (relative_block == nblocks) {
 253                                 clear_buffer_mapped(map_bh);
 254                                 break;
 255                         } else if (page_block == blocks_per_page)
 256                                 break;
 257                         blocks[page_block] = map_bh->b_blocknr+relative_block;
 258                         page_block++;
 259                         block_in_file++;
 260                 }
 261                 bdev = map_bh->b_bdev;
 262         }
 263
 264         if (first_hole != blocks_per_page) {
 265                 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
 266                 if (first_hole == 0) {
 267                         SetPageUptodate(page);
 268                         unlock_page(page);
 269                         goto out;
 270                 }
 271         } else if (fully_mapped) {
 272                 SetPageMappedToDisk(page);
 273         }
 274
 275         if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
 276             cleancache_get_page(page) == 0) {
 277                 SetPageUptodate(page);
 278                 goto confused;
 279         }
 280
 281         /*
 282          * This page will go to BIO.  Do we need to send this BIO off first?
 283          */
 284         if (bio && (*last_block_in_bio != blocks[0] - 1))
 285                 bio = mpage_bio_submit(READ, bio);
 286
 287 alloc_new:
 288         if (bio == NULL) {
 289                 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
 290                                 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
 291                                 GFP_KERNEL);
 292                 if (bio == NULL)
 293                         goto confused;
 294         }
 295
 296         length = first_hole << blkbits;
 297         if (bio_add_page(bio, page, length, 0) < length) {
 298                 bio = mpage_bio_submit(READ, bio);
 299                 goto alloc_new;
 300         }
 301
 302         relative_block = block_in_file - *first_logical_block;
 303         nblocks = map_bh->b_size >> blkbits;
 304         if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
 305             (first_hole != blocks_per_page))
 306                 bio = mpage_bio_submit(READ, bio);
 307         else
 308                 *last_block_in_bio = blocks[blocks_per_page - 1];
 309 out:
 310         return bio;
 311
 312 confused:
 313         if (bio)
 314                 bio = mpage_bio_submit(READ, bio);
 315         if (!PageUptodate(page))
 316                 block_read_full_page(page, get_block);
 317         else
 318                 unlock_page(page);
 319         goto out;
 320 }
 321
 322 /**
 323  * mpage_readpages - populate an address space with some pages & start reads against them
 324  * @mapping: the address_space
 325  * @pages: The address of a list_head which contains the target pages.  These
 326  *   pages have their ->index populated and are otherwise uninitialised.
 327  *   The page at @pages->prev has the lowest file offset, and reads should be
 328  *   issued in @pages->prev to @pages->next order.
 329  * @nr_pages: The number of pages at *@pages
 330  * @get_block: The filesystem's block mapper function.
 331  *
 332  * This function walks the pages and the blocks within each page, building and
 333  * emitting large BIOs.
 334  *
 335  * If anything unusual happens, such as:
 336  *
 337  * - encountering a page which has buffers
 338  * - encountering a page which has a non-hole after a hole
 339  * - encountering a page with non-contiguous blocks
 340  *
 341  * then this code just gives up and calls the buffer_head-based read function.
 342  * It does handle a page which has holes at the end - that is a common case:
 343  * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
 344  *
 345  * BH_Boundary explanation:
 346  *
 347  * There is a problem.  The mpage read code assembles several pages, gets all
 348  * their disk mappings, and then submits them all.  That's fine, but obtaining
 349  * the disk mappings may require I/O.  Reads of indirect blocks, for example.
 350  *
 351  * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
 352  * submitted in the following order:
 353  *      12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
 354  *
 355  * because the indirect block has to be read to get the mappings of blocks
 356  * 13,14,15,16.  Obviously, this impacts performance.
 357  *
 358  * So what we do it to allow the filesystem's get_block() function to set
 359  * BH_Boundary when it maps block 11.  BH_Boundary says: mapping of the block
 360  * after this one will require I/O against a block which is probably close to
 361  * this one.  So you should push what I/O you have currently accumulated.
 362  *
 363  * This all causes the disk requests to be issued in the correct order.
 364  */
 365 int
 366 mpage_readpages(struct address_space *mapping, struct list_head *pages,
 367                                 unsigned nr_pages, get_block_t get_block)
 368 {
 369         struct bio *bio = NULL;
 370         unsigned page_idx;
 371         sector_t last_block_in_bio = 0;
 372         struct buffer_head map_bh;
 373         unsigned long first_logical_block = 0;
 374
 375         map_bh.b_state = 0;
 376         map_bh.b_size = 0;
 377         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
 378                 struct page *page = list_entry(pages->prev, struct page, lru);
 379
 380                 prefetchw(&page->flags);
 381                 list_del(&page->lru);
 382                 if (!add_to_page_cache_lru(page, mapping,
 383                                         page->index, GFP_KERNEL)) {
 384                         bio = do_mpage_readpage(bio, page,
 385                                         nr_pages - page_idx,
 386                                         &last_block_in_bio, &map_bh,
 387                                         &first_logical_block,
 388                                         get_block);
 389                 }
 390                 page_cache_release(page);
 391         }
 392         BUG_ON(!list_empty(pages));
 393         if (bio)
 394                 mpage_bio_submit(READ, bio);
 395         return 0;
 396 }
 397 EXPORT_SYMBOL(mpage_readpages);
 398
 399 /*
 400  * This isn't called much at all
 401  */
 402 int mpage_readpage(struct page *page, get_block_t get_block)
 403 {
 404         struct bio *bio = NULL;
 405         sector_t last_block_in_bio = 0;
 406         struct buffer_head map_bh;
 407         unsigned long first_logical_block = 0;
 408
 409         map_bh.b_state = 0;
 410         map_bh.b_size = 0;
 411         bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
 412                         &map_bh, &first_logical_block, get_block);
 413         if (bio)
 414                 mpage_bio_submit(READ, bio);
 415         return 0;
 416 }
 417 EXPORT_SYMBOL(mpage_readpage);
 418
 419 /*
 420  * Writing is not so simple.
 421  *
 422  * If the page has buffers then they will be used for obtaining the disk
 423  * mapping.  We only support pages which are fully mapped-and-dirty, with a
 424  * special case for pages which are unmapped at the end: end-of-file.
 425  *
 426  * If the page has no buffers (preferred) then the page is mapped here.
 427  *
 428  * If all blocks are found to be contiguous then the page can go into the
 429  * BIO.  Otherwise fall back to the mapping's writepage().
 430  *
 431  * FIXME: This code wants an estimate of how many pages are still to be
 432  * written, so it can intelligently allocate a suitably-sized BIO.  For now,
 433  * just allocate full-size (16-page) BIOs.
 434  */
 435
 436 struct mpage_data {
 437         struct bio *bio;
 438         sector_t last_block_in_bio;
 439         get_block_t *get_block;
 440         unsigned use_writepage;
 441 };
 442
 443 static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
 444                       void *data)
 445 {
 446         struct mpage_data *mpd = data;
 447         struct bio *bio = mpd->bio;
 448         struct address_space *mapping = page->mapping;
 449         struct inode *inode = page->mapping->host;
 450         const unsigned blkbits = inode->i_blkbits;
 451         unsigned long end_index;
 452         const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
 453         sector_t last_block;
 454         sector_t block_in_file;
 455         sector_t blocks[MAX_BUF_PER_PAGE];
 456         unsigned page_block;
 457         unsigned first_unmapped = blocks_per_page;
 458         struct block_device *bdev = NULL;
 459         int boundary = 0;
 460         sector_t boundary_block = 0;
 461         struct block_device *boundary_bdev = NULL;
 462         int length;
 463         struct buffer_head map_bh;
 464         loff_t i_size = i_size_read(inode);
 465         int ret = 0;
 466
 467         if (page_has_buffers(page)) {
 468                 struct buffer_head *head = page_buffers(page);
 469                 struct buffer_head *bh = head;
 470
 471                 /* If they're all mapped and dirty, do it */
 472                 page_block = 0;
 473                 do {
 474                         BUG_ON(buffer_locked(bh));
 475                         if (!buffer_mapped(bh)) {
 476                                 /*
 477                                  * unmapped dirty buffers are created by
 478                                  * __set_page_dirty_buffers -> mmapped data
 479                                  */
 480                                 if (buffer_dirty(bh))
 481                                         goto confused;
 482                                 if (first_unmapped == blocks_per_page)
 483                                         first_unmapped = page_block;
 484                                 continue;
 485                         }
 486
 487                         if (first_unmapped != blocks_per_page)
 488                                 goto confused;  /* hole -> non-hole */
 489
 490                         if (!buffer_dirty(bh) || !buffer_uptodate(bh))
 491                                 goto confused;
 492                         if (page_block) {
 493                                 if (bh->b_blocknr != blocks[page_block-1] + 1)
 494                                         goto confused;
 495                         }
 496                         blocks[page_block++] = bh->b_blocknr;
 497                         boundary = buffer_boundary(bh);
 498                         if (boundary) {
 499                                 boundary_block = bh->b_blocknr;
 500                                 boundary_bdev = bh->b_bdev;
 501                         }
 502                         bdev = bh->b_bdev;
 503                 } while ((bh = bh->b_this_page) != head);
 504
 505                 if (first_unmapped)
 506                         goto page_is_mapped;
 507
 508                 /*
 509                  * Page has buffers, but they are all unmapped. The page was
 510                  * created by pagein or read over a hole which was handled by
 511                  * block_read_full_page().  If this address_space is also
 512                  * using mpage_readpages then this can rarely happen.
 513                  */
 514                 goto confused;
 515         }
 516
 517         /*
 518          * The page has no buffers: map it to disk
 519          */
 520         BUG_ON(!PageUptodate(page));
 521         block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 522         last_block = (i_size - 1) >> blkbits;
 523         map_bh.b_page = page;
 524         for (page_block = 0; page_block < blocks_per_page; ) {
 525
 526                 map_bh.b_state = 0;
 527                 map_bh.b_size = 1 << blkbits;
 528                 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
 529                         goto confused;
 530                 if (buffer_new(&map_bh))
 531                         unmap_underlying_metadata(map_bh.b_bdev,
 532                                                 map_bh.b_blocknr);
 533                 if (buffer_boundary(&map_bh)) {
 534                         boundary_block = map_bh.b_blocknr;
 535                         boundary_bdev = map_bh.b_bdev;
 536                 }
 537                 if (page_block) {
 538                         if (map_bh.b_blocknr != blocks[page_block-1] + 1)
 539                                 goto confused;
 540                 }
 541                 blocks[page_block++] = map_bh.b_blocknr;
 542                 boundary = buffer_boundary(&map_bh);
 543                 bdev = map_bh.b_bdev;
 544                 if (block_in_file == last_block)
 545                         break;
 546                 block_in_file++;
 547         }
 548         BUG_ON(page_block == 0);
 549
 550         first_unmapped = page_block;
 551
 552 page_is_mapped:
 553         end_index = i_size >> PAGE_CACHE_SHIFT;
 554         if (page->index >= end_index) {
 555                 /*
 556                  * The page straddles i_size.  It must be zeroed out on each
 557                  * and every writepage invocation because it may be mmapped.
 558                  * "A file is mapped in multiples of the page size.  For a file
 559                  * that is not a multiple of the page size, the remaining memory
 560                  * is zeroed when mapped, and writes to that region are not
 561                  * written out to the file."
 562                  */
 563                 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
 564
 565                 if (page->index > end_index || !offset)
 566                         goto confused;
 567                 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
 568         }
 569
 570         /*
 571          * This page will go to BIO.  Do we need to send this BIO off first?
 572          */
 573         if (bio && mpd->last_block_in_bio != blocks[0] - 1)
 574                 bio = mpage_bio_submit(WRITE, bio);
 575
 576 alloc_new:
 577         if (bio == NULL) {
 578                 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
 579                                 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
 580                 if (bio == NULL)
 581                         goto confused;
 582         }
 583
 584         /*
 585          * Must try to add the page before marking the buffer clean or
 586          * the confused fail path above (OOM) will be very confused when
 587          * it finds all bh marked clean (i.e. it will not write anything)
 588          */
 589         length = first_unmapped << blkbits;
 590         if (bio_add_page(bio, page, length, 0) < length) {
 591                 bio = mpage_bio_submit(WRITE, bio);
 592                 goto alloc_new;
 593         }
 594
 595         /*
 596          * OK, we have our BIO, so we can now mark the buffers clean.  Make
 597          * sure to only clean buffers which we know we'll be writing.
 598          */
 599         if (page_has_buffers(page)) {
 600                 struct buffer_head *head = page_buffers(page);
 601                 struct buffer_head *bh = head;
 602                 unsigned buffer_counter = 0;
 603
 604                 do {
 605                         if (buffer_counter++ == first_unmapped)
 606                                 break;
 607                         clear_buffer_dirty(bh);
 608                         bh = bh->b_this_page;
 609                 } while (bh != head);
 610
 611                 /*
 612                  * we cannot drop the bh if the page is not uptodate
 613                  * or a concurrent readpage would fail to serialize with the bh
 614                  * and it would read from disk before we reach the platter.
 615                  */
 616                 if (buffer_heads_over_limit && PageUptodate(page))
 617                         try_to_free_buffers(page);
 618         }
 619
 620         BUG_ON(PageWriteback(page));
 621         set_page_writeback(page);
 622         unlock_page(page);
 623         if (boundary || (first_unmapped != blocks_per_page)) {
 624                 bio = mpage_bio_submit(WRITE, bio);
 625                 if (boundary_block) {
 626                         write_boundary_block(boundary_bdev,
 627                                         boundary_block, 1 << blkbits);
 628                 }
 629         } else {
 630                 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
 631         }
 632         goto out;
 633
 634 confused:
 635         if (bio)
 636                 bio = mpage_bio_submit(WRITE, bio);
 637
 638         if (mpd->use_writepage) {
 639                 ret = mapping->a_ops->writepage(page, wbc);
 640         } else {
 641                 ret = -EAGAIN;
 642                 goto out;
 643         }
 644         /*
 645          * The caller has a ref on the inode, so *mapping is stable
 646          */
 647         mapping_set_error(mapping, ret);
 648 out:
 649         mpd->bio = bio;
 650         return ret;
 651 }
 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         struct blk_plug plug;
 677         int ret;
 678
 679         blk_start_plug(&plug);
 680
 681         if (!get_block)
 682                 ret = generic_writepages(mapping, wbc);
 683         else {
 684                 struct mpage_data mpd = {
 685                         .bio = NULL,
 686                         .last_block_in_bio = 0,
 687                         .get_block = get_block,
 688                         .use_writepage = 1,
 689                 };
 690
 691                 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
 692                 if (mpd.bio)
 693                         mpage_bio_submit(WRITE, mpd.bio);
 694         }
 695         blk_finish_plug(&plug);
 696         return ret;
 697 }
 698 EXPORT_SYMBOL(mpage_writepages);
 699
 700 int mpage_writepage(struct page *page, get_block_t get_block,
 701         struct writeback_control *wbc)
 702 {
 703         struct mpage_data mpd = {
 704                 .bio = NULL,
 705                 .last_block_in_bio = 0,
 706                 .get_block = get_block,
 707                 .use_writepage = 0,
 708         };
 709         int ret = __mpage_writepage(page, wbc, &mpd);
 710         if (mpd.bio)
 711                 mpage_bio_submit(WRITE, mpd.bio);
 712         return ret;
 713 }
 714 EXPORT_SYMBOL(mpage_writepage);