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    akpm@zip.com.au
  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 int mpage_end_io_read(struct bio *bio, unsigned int bytes_done, 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         if (bio->bi_size)
  48                 return 1;
  49
  50         do {
  51                 struct page *page = bvec->bv_page;
  52
  53                 if (--bvec >= bio->bi_io_vec)
  54                         prefetchw(&bvec->bv_page->flags);
  55
  56                 if (uptodate) {
  57                         SetPageUptodate(page);
  58                 } else {
  59                         ClearPageUptodate(page);
  60                         SetPageError(page);
  61                 }
  62                 unlock_page(page);
  63         } while (bvec >= bio->bi_io_vec);
  64         bio_put(bio);
  65         return 0;
  66 }
  67
  68 static int mpage_end_io_write(struct bio *bio, unsigned int bytes_done, int err)
  69 {
  70         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  71         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
  72
  73         if (bio->bi_size)
  74                 return 1;
  75
  76         do {
  77                 struct page *page = bvec->bv_page;
  78
  79                 if (--bvec >= bio->bi_io_vec)
  80                         prefetchw(&bvec->bv_page->flags);
  81
  82                 if (!uptodate){
  83                         SetPageError(page);
  84 #if 0   // mask by Victor Yu. 02-12-2007
  85                         if (page->mapping)
  86                                 set_bit(AS_EIO, &page->mapping->flags);
  87 #else
  88                         if (page->u.xx.mapping)
  89                                 set_bit(AS_EIO, &page->u.xx.mapping->flags);
  90 #endif
  91                 }
  92                 end_page_writeback(page);
  93         } while (bvec >= bio->bi_io_vec);
  94         bio_put(bio);
  95         return 0;
  96 }
  97
  98 static struct bio *mpage_bio_submit(int rw, struct bio *bio)
  99 {
 100         bio->bi_end_io = mpage_end_io_read;
 101         if (rw == WRITE)
 102                 bio->bi_end_io = mpage_end_io_write;
 103         submit_bio(rw, bio);
 104         return NULL;
 105 }
 106
 107 static struct bio *
 108 mpage_alloc(struct block_device *bdev,
 109                 sector_t first_sector, int nr_vecs,
 110                 gfp_t gfp_flags)
 111 {
 112         struct bio *bio;
 113
 114         bio = bio_alloc(gfp_flags, nr_vecs);
 115
 116         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
 117                 while (!bio && (nr_vecs /= 2))
 118                         bio = bio_alloc(gfp_flags, nr_vecs);
 119         }
 120
 121         if (bio) {
 122                 bio->bi_bdev = bdev;
 123                 bio->bi_sector = first_sector;
 124         }
 125         return bio;
 126 }
 127
 128 /*
 129  * support function for mpage_readpages.  The fs supplied get_block might
 130  * return an up to date buffer.  This is used to map that buffer into
 131  * the page, which allows readpage to avoid triggering a duplicate call
 132  * to get_block.
 133  *
 134  * The idea is to avoid adding buffers to pages that don't already have
 135  * them.  So when the buffer is up to date and the page size == block size,
 136  * this marks the page up to date instead of adding new buffers.
 137  */
 138 static void
 139 map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
 140 {
 141 #if 0   // mask by Victor Yu. 02-12-2007
 142         struct inode *inode = page->mapping->host;
 143 #else
 144         struct inode *inode = page->u.xx.mapping->host;
 145 #endif
 146         struct buffer_head *page_bh, *head;
 147         int block = 0;
 148
 149         if (!page_has_buffers(page)) {
 150                 /*
 151                  * don't make any buffers if there is only one buffer on
 152                  * the page and the page just needs to be set up to date
 153                  */
 154                 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
 155                     buffer_uptodate(bh)) {
 156                         SetPageUptodate(page);
 157                         return;
 158                 }
 159                 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
 160         }
 161         head = page_buffers(page);
 162         page_bh = head;
 163         do {
 164                 if (block == page_block) {
 165                         page_bh->b_state = bh->b_state;
 166                         page_bh->b_bdev = bh->b_bdev;
 167                         page_bh->b_blocknr = bh->b_blocknr;
 168                         break;
 169                 }
 170                 page_bh = page_bh->b_this_page;
 171                 block++;
 172         } while (page_bh != head);
 173 }
 174
 175 /*
 176  * This is the worker routine which does all the work of mapping the disk
 177  * blocks and constructs largest possible bios, submits them for IO if the
 178  * blocks are not contiguous on the disk.
 179  *
 180  * We pass a buffer_head back and forth and use its buffer_mapped() flag to
 181  * represent the validity of its disk mapping and to decide when to do the next
 182  * get_block() call.
 183  */
 184 static struct bio *
 185 do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
 186                 sector_t *last_block_in_bio, struct buffer_head *map_bh,
 187                 unsigned long *first_logical_block, get_block_t get_block)
 188 {
 189 #if 0   // mask by Victor Yu. 02-12-2007
 190         struct inode *inode = page->mapping->host;
 191 #else
 192         struct inode *inode = page->u.xx.mapping->host;
 193 #endif
 194         const unsigned blkbits = inode->i_blkbits;
 195         const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
 196         const unsigned blocksize = 1 << blkbits;
 197         sector_t block_in_file;
 198         sector_t last_block;
 199         sector_t last_block_in_file;
 200         sector_t blocks[MAX_BUF_PER_PAGE];
 201         unsigned page_block;
 202         unsigned first_hole = blocks_per_page;
 203         struct block_device *bdev = NULL;
 204         int length;
 205         int fully_mapped = 1;
 206         unsigned nblocks;
 207         unsigned relative_block;
 208
 209         if (page_has_buffers(page))
 210                 goto confused;
 211
 212         block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 213         last_block = block_in_file + nr_pages * blocks_per_page;
 214         last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
 215         if (last_block > last_block_in_file)
 216                 last_block = last_block_in_file;
 217         page_block = 0;
 218
 219         /*
 220          * Map blocks using the result from the previous get_blocks call first.
 221          */
 222         nblocks = map_bh->b_size >> blkbits;
 223         if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
 224                         block_in_file < (*first_logical_block + nblocks)) {
 225                 unsigned map_offset = block_in_file - *first_logical_block;
 226                 unsigned last = nblocks - map_offset;
 227
 228                 for (relative_block = 0; ; relative_block++) {
 229                         if (relative_block == last) {
 230                                 clear_buffer_mapped(map_bh);
 231                                 break;
 232                         }
 233                         if (page_block == blocks_per_page)
 234                                 break;
 235                         blocks[page_block] = map_bh->b_blocknr + map_offset +
 236                                                 relative_block;
 237                         page_block++;
 238                         block_in_file++;
 239                 }
 240                 bdev = map_bh->b_bdev;
 241         }
 242
 243         /*
 244          * Then do more get_blocks calls until we are done with this page.
 245          */
 246         map_bh->b_page = page;
 247         while (page_block < blocks_per_page) {
 248                 map_bh->b_state = 0;
 249                 map_bh->b_size = 0;
 250
 251                 if (block_in_file < last_block) {
 252                         map_bh->b_size = (last_block-block_in_file) << blkbits;
 253                         if (get_block(inode, block_in_file, map_bh, 0))
 254                                 goto confused;
 255                         *first_logical_block = block_in_file;
 256                 }
 257
 258                 if (!buffer_mapped(map_bh)) {
 259                         fully_mapped = 0;
 260                         if (first_hole == blocks_per_page)
 261                                 first_hole = page_block;
 262                         page_block++;
 263                         block_in_file++;
 264                         clear_buffer_mapped(map_bh);
 265                         continue;
 266                 }
 267
 268                 /* some filesystems will copy data into the page during
 269                  * the get_block call, in which case we don't want to
 270                  * read it again.  map_buffer_to_page copies the data
 271                  * we just collected from get_block into the page's buffers
 272                  * so readpage doesn't have to repeat the get_block call
 273                  */
 274                 if (buffer_uptodate(map_bh)) {
 275                         map_buffer_to_page(page, map_bh, page_block);
 276                         goto confused;
 277                 }
 278
 279                 if (first_hole != blocks_per_page)
 280                         goto confused;          /* hole -> non-hole */
 281
 282                 /* Contiguous blocks? */
 283                 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
 284                         goto confused;
 285                 nblocks = map_bh->b_size >> blkbits;
 286                 for (relative_block = 0; ; relative_block++) {
 287                         if (relative_block == nblocks) {
 288                                 clear_buffer_mapped(map_bh);
 289                                 break;
 290                         } else if (page_block == blocks_per_page)
 291                                 break;
 292                         blocks[page_block] = map_bh->b_blocknr+relative_block;
 293                         page_block++;
 294                         block_in_file++;
 295                 }
 296                 bdev = map_bh->b_bdev;
 297         }
 298
 299         if (first_hole != blocks_per_page) {
 300                 char *kaddr = kmap_atomic(page, KM_USER0);
 301                 memset(kaddr + (first_hole << blkbits), 0,
 302                                 PAGE_CACHE_SIZE - (first_hole << blkbits));
 303                 flush_dcache_page(page);
 304                 kunmap_atomic(kaddr, KM_USER0);
 305                 if (first_hole == 0) {
 306                         SetPageUptodate(page);
 307                         unlock_page(page);
 308                         goto out;
 309                 }
 310         } else if (fully_mapped) {
 311                 SetPageMappedToDisk(page);
 312         }
 313
 314         /*
 315          * This page will go to BIO.  Do we need to send this BIO off first?
 316          */
 317         if (bio && (*last_block_in_bio != blocks[0] - 1))
 318                 bio = mpage_bio_submit(READ, bio);
 319
 320 alloc_new:
 321         if (bio == NULL) {
 322                 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
 323                                 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
 324                                 GFP_KERNEL);
 325                 if (bio == NULL)
 326                         goto confused;
 327         }
 328
 329         length = first_hole << blkbits;
 330         if (bio_add_page(bio, page, length, 0) < length) {
 331                 bio = mpage_bio_submit(READ, bio);
 332                 goto alloc_new;
 333         }
 334
 335         if (buffer_boundary(map_bh) || (first_hole != blocks_per_page))
 336                 bio = mpage_bio_submit(READ, bio);
 337         else
 338                 *last_block_in_bio = blocks[blocks_per_page - 1];
 339 out:
 340         return bio;
 341
 342 confused:
 343         if (bio)
 344                 bio = mpage_bio_submit(READ, bio);
 345         if (!PageUptodate(page))
 346                 block_read_full_page(page, get_block);
 347         else
 348                 unlock_page(page);
 349         goto out;
 350 }
 351
 352 /**
 353  * mpage_readpages - populate an address space with some pages, and
 354  *                       start reads against them.
 355  *
 356  * @mapping: the address_space
 357  * @pages: The address of a list_head which contains the target pages.  These
 358  *   pages have their ->index populated and are otherwise uninitialised.
 359  *
 360  *   The page at @pages->prev has the lowest file offset, and reads should be
 361  *   issued in @pages->prev to @pages->next order.
 362  *
 363  * @nr_pages: The number of pages at *@pages
 364  * @get_block: The filesystem's block mapper function.
 365  *
 366  * This function walks the pages and the blocks within each page, building and
 367  * emitting large BIOs.
 368  *
 369  * If anything unusual happens, such as:
 370  *
 371  * - encountering a page which has buffers
 372  * - encountering a page which has a non-hole after a hole
 373  * - encountering a page with non-contiguous blocks
 374  *
 375  * then this code just gives up and calls the buffer_head-based read function.
 376  * It does handle a page which has holes at the end - that is a common case:
 377  * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
 378  *
 379  * BH_Boundary explanation:
 380  *
 381  * There is a problem.  The mpage read code assembles several pages, gets all
 382  * their disk mappings, and then submits them all.  That's fine, but obtaining
 383  * the disk mappings may require I/O.  Reads of indirect blocks, for example.
 384  *
 385  * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
 386  * submitted in the following order:
 387  *      12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
 388  * because the indirect block has to be read to get the mappings of blocks
 389  * 13,14,15,16.  Obviously, this impacts performance.
 390  *
 391  * So what we do it to allow the filesystem's get_block() function to set
 392  * BH_Boundary when it maps block 11.  BH_Boundary says: mapping of the block
 393  * after this one will require I/O against a block which is probably close to
 394  * this one.  So you should push what I/O you have currently accumulated.
 395  *
 396  * This all causes the disk requests to be issued in the correct order.
 397  */
 398 int
 399 mpage_readpages(struct address_space *mapping, struct list_head *pages,
 400                                 unsigned nr_pages, get_block_t get_block)
 401 {
 402         struct bio *bio = NULL;
 403         unsigned page_idx;
 404         sector_t last_block_in_bio = 0;
 405         struct pagevec lru_pvec;
 406         struct buffer_head map_bh;
 407         unsigned long first_logical_block = 0;
 408
 409         clear_buffer_mapped(&map_bh);
 410         pagevec_init(&lru_pvec, 0);
 411         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
 412                 struct page *page = list_entry(pages->prev, struct page, lru);
 413
 414                 prefetchw(&page->flags);
 415                 list_del(&page->lru);
 416                 if (!add_to_page_cache(page, mapping,
 417                                         page->index, GFP_KERNEL)) {
 418                         bio = do_mpage_readpage(bio, page,
 419                                         nr_pages - page_idx,
 420                                         &last_block_in_bio, &map_bh,
 421                                         &first_logical_block,
 422                                         get_block);
 423                         if (!pagevec_add(&lru_pvec, page))
 424                                 __pagevec_lru_add(&lru_pvec);
 425                 } else {
 426                         page_cache_release(page);
 427                 }
 428         }
 429         pagevec_lru_add(&lru_pvec);
 430         BUG_ON(!list_empty(pages));
 431         if (bio)
 432                 mpage_bio_submit(READ, bio);
 433         return 0;
 434 }
 435 EXPORT_SYMBOL(mpage_readpages);
 436
 437 /*
 438  * This isn't called much at all
 439  */
 440 int mpage_readpage(struct page *page, get_block_t get_block)
 441 {
 442         struct bio *bio = NULL;
 443         sector_t last_block_in_bio = 0;
 444         struct buffer_head map_bh;
 445         unsigned long first_logical_block = 0;
 446
 447         clear_buffer_mapped(&map_bh);
 448         bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
 449                         &map_bh, &first_logical_block, get_block);
 450         if (bio)
 451                 mpage_bio_submit(READ, bio);
 452         return 0;
 453 }
 454 EXPORT_SYMBOL(mpage_readpage);
 455
 456 /*
 457  * Writing is not so simple.
 458  *
 459  * If the page has buffers then they will be used for obtaining the disk
 460  * mapping.  We only support pages which are fully mapped-and-dirty, with a
 461  * special case for pages which are unmapped at the end: end-of-file.
 462  *
 463  * If the page has no buffers (preferred) then the page is mapped here.
 464  *
 465  * If all blocks are found to be contiguous then the page can go into the
 466  * BIO.  Otherwise fall back to the mapping's writepage().
 467  *
 468  * FIXME: This code wants an estimate of how many pages are still to be
 469  * written, so it can intelligently allocate a suitably-sized BIO.  For now,
 470  * just allocate full-size (16-page) BIOs.
 471  */
 472 static struct bio *
 473 __mpage_writepage(struct bio *bio, struct page *page, get_block_t get_block,
 474         sector_t *last_block_in_bio, int *ret, struct writeback_control *wbc,
 475         writepage_t writepage_fn)
 476 {
 477 #if 0   // mask by Victor Yu. 02-12-2007
 478         struct address_space *mapping = page->mapping;
 479         struct inode *inode = page->mapping->host;
 480 #else
 481         struct address_space *mapping = page->u.xx.mapping;
 482         struct inode *inode = page->u.xx.mapping->host;
 483 #endif
 484         const unsigned blkbits = inode->i_blkbits;
 485         unsigned long end_index;
 486         const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
 487         sector_t last_block;
 488         sector_t block_in_file;
 489         sector_t blocks[MAX_BUF_PER_PAGE];
 490         unsigned page_block;
 491         unsigned first_unmapped = blocks_per_page;
 492         struct block_device *bdev = NULL;
 493         int boundary = 0;
 494         sector_t boundary_block = 0;
 495         struct block_device *boundary_bdev = NULL;
 496         int length;
 497         struct buffer_head map_bh;
 498         loff_t i_size = i_size_read(inode);
 499
 500         if (page_has_buffers(page)) {
 501                 struct buffer_head *head = page_buffers(page);
 502                 struct buffer_head *bh = head;
 503
 504                 /* If they're all mapped and dirty, do it */
 505                 page_block = 0;
 506                 do {
 507                         BUG_ON(buffer_locked(bh));
 508                         if (!buffer_mapped(bh)) {
 509                                 /*
 510                                  * unmapped dirty buffers are created by
 511                                  * __set_page_dirty_buffers -> mmapped data
 512                                  */
 513                                 if (buffer_dirty(bh))
 514                                         goto confused;
 515                                 if (first_unmapped == blocks_per_page)
 516                                         first_unmapped = page_block;
 517                                 continue;
 518                         }
 519
 520                         if (first_unmapped != blocks_per_page)
 521                                 goto confused;  /* hole -> non-hole */
 522
 523                         if (!buffer_dirty(bh) || !buffer_uptodate(bh))
 524                                 goto confused;
 525                         if (page_block) {
 526                                 if (bh->b_blocknr != blocks[page_block-1] + 1)
 527                                         goto confused;
 528                         }
 529                         blocks[page_block++] = bh->b_blocknr;
 530                         boundary = buffer_boundary(bh);
 531                         if (boundary) {
 532                                 boundary_block = bh->b_blocknr;
 533                                 boundary_bdev = bh->b_bdev;
 534                         }
 535                         bdev = bh->b_bdev;
 536                 } while ((bh = bh->b_this_page) != head);
 537
 538                 if (first_unmapped)
 539                         goto page_is_mapped;
 540
 541                 /*
 542                  * Page has buffers, but they are all unmapped. The page was
 543                  * created by pagein or read over a hole which was handled by
 544                  * block_read_full_page().  If this address_space is also
 545                  * using mpage_readpages then this can rarely happen.
 546                  */
 547                 goto confused;
 548         }
 549
 550         /*
 551          * The page has no buffers: map it to disk
 552          */
 553         BUG_ON(!PageUptodate(page));
 554         block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 555         last_block = (i_size - 1) >> blkbits;
 556         map_bh.b_page = page;
 557         for (page_block = 0; page_block < blocks_per_page; ) {
 558
 559                 map_bh.b_state = 0;
 560                 map_bh.b_size = 1 << blkbits;
 561                 if (get_block(inode, block_in_file, &map_bh, 1))
 562                         goto confused;
 563                 if (buffer_new(&map_bh))
 564                         unmap_underlying_metadata(map_bh.b_bdev,
 565                                                 map_bh.b_blocknr);
 566                 if (buffer_boundary(&map_bh)) {
 567                         boundary_block = map_bh.b_blocknr;
 568                         boundary_bdev = map_bh.b_bdev;
 569                 }
 570                 if (page_block) {
 571                         if (map_bh.b_blocknr != blocks[page_block-1] + 1)
 572                                 goto confused;
 573                 }
 574                 blocks[page_block++] = map_bh.b_blocknr;
 575                 boundary = buffer_boundary(&map_bh);
 576                 bdev = map_bh.b_bdev;
 577                 if (block_in_file == last_block)
 578                         break;
 579                 block_in_file++;
 580         }
 581         BUG_ON(page_block == 0);
 582
 583         first_unmapped = page_block;
 584
 585 page_is_mapped:
 586         end_index = i_size >> PAGE_CACHE_SHIFT;
 587         if (page->index >= end_index) {
 588                 /*
 589                  * The page straddles i_size.  It must be zeroed out on each
 590                  * and every writepage invokation because it may be mmapped.
 591                  * "A file is mapped in multiples of the page size.  For a file
 592                  * that is not a multiple of the page size, the remaining memory
 593                  * is zeroed when mapped, and writes to that region are not
 594                  * written out to the file."
 595                  */
 596                 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
 597                 char *kaddr;
 598
 599                 if (page->index > end_index || !offset)
 600                         goto confused;
 601                 kaddr = kmap_atomic(page, KM_USER0);
 602                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
 603                 flush_dcache_page(page);
 604                 kunmap_atomic(kaddr, KM_USER0);
 605         }
 606
 607         /*
 608          * This page will go to BIO.  Do we need to send this BIO off first?
 609          */
 610         if (bio && *last_block_in_bio != blocks[0] - 1)
 611                 bio = mpage_bio_submit(WRITE, bio);
 612
 613 alloc_new:
 614         if (bio == NULL) {
 615                 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
 616                                 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
 617                 if (bio == NULL)
 618                         goto confused;
 619         }
 620
 621         /*
 622          * Must try to add the page before marking the buffer clean or
 623          * the confused fail path above (OOM) will be very confused when
 624          * it finds all bh marked clean (i.e. it will not write anything)
 625          */
 626         length = first_unmapped << blkbits;
 627         if (bio_add_page(bio, page, length, 0) < length) {
 628                 bio = mpage_bio_submit(WRITE, bio);
 629                 goto alloc_new;
 630         }
 631
 632         /*
 633          * OK, we have our BIO, so we can now mark the buffers clean.  Make
 634          * sure to only clean buffers which we know we'll be writing.
 635          */
 636         if (page_has_buffers(page)) {
 637                 struct buffer_head *head = page_buffers(page);
 638                 struct buffer_head *bh = head;
 639                 unsigned buffer_counter = 0;
 640
 641                 do {
 642                         if (buffer_counter++ == first_unmapped)
 643                                 break;
 644                         clear_buffer_dirty(bh);
 645                         bh = bh->b_this_page;
 646                 } while (bh != head);
 647
 648                 /*
 649                  * we cannot drop the bh if the page is not uptodate
 650                  * or a concurrent readpage would fail to serialize with the bh
 651                  * and it would read from disk before we reach the platter.
 652                  */
 653                 if (buffer_heads_over_limit && PageUptodate(page))
 654                         try_to_free_buffers(page);
 655         }
 656
 657         BUG_ON(PageWriteback(page));
 658         set_page_writeback(page);
 659         unlock_page(page);
 660         if (boundary || (first_unmapped != blocks_per_page)) {
 661                 bio = mpage_bio_submit(WRITE, bio);
 662                 if (boundary_block) {
 663                         write_boundary_block(boundary_bdev,
 664                                         boundary_block, 1 << blkbits);
 665                 }
 666         } else {
 667                 *last_block_in_bio = blocks[blocks_per_page - 1];
 668         }
 669         goto out;
 670
 671 confused:
 672         if (bio)
 673                 bio = mpage_bio_submit(WRITE, bio);
 674
 675         if (writepage_fn) {
 676                 *ret = (*writepage_fn)(page, wbc);
 677         } else {
 678                 *ret = -EAGAIN;
 679                 goto out;
 680         }
 681         /*
 682          * The caller has a ref on the inode, so *mapping is stable
 683          */
 684         if (*ret) {
 685                 if (*ret == -ENOSPC)
 686                         set_bit(AS_ENOSPC, &mapping->flags);
 687                 else
 688                         set_bit(AS_EIO, &mapping->flags);
 689         }
 690 out:
 691         return bio;
 692 }
 693
 694 /**
 695  * mpage_writepages - walk the list of dirty pages of the given
 696  * address space and writepage() all of them.
 697  *
 698  * @mapping: address space structure to write
 699  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
 700  * @get_block: the filesystem's block mapper function.
 701  *             If this is NULL then use a_ops->writepage.  Otherwise, go
 702  *             direct-to-BIO.
 703  *
 704  * This is a library function, which implements the writepages()
 705  * address_space_operation.
 706  *
 707  * If a page is already under I/O, generic_writepages() skips it, even
 708  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
 709  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
 710  * and msync() need to guarantee that all the data which was dirty at the time
 711  * the call was made get new I/O started against them.  If wbc->sync_mode is
 712  * WB_SYNC_ALL then we were called for data integrity and we must wait for
 713  * existing IO to complete.
 714  *
 715  * If you fix this you should check generic_writepages() also!
 716  */
 717 int
 718 mpage_writepages(struct address_space *mapping,
 719                 struct writeback_control *wbc, get_block_t get_block)
 720 {
 721         struct backing_dev_info *bdi = mapping->backing_dev_info;
 722         struct bio *bio = NULL;
 723         sector_t last_block_in_bio = 0;
 724         int ret = 0;
 725         int done = 0;
 726         int (*writepage)(struct page *page, struct writeback_control *wbc);
 727         struct pagevec pvec;
 728         int nr_pages;
 729         pgoff_t index;
 730         pgoff_t end;            /* Inclusive */
 731         int scanned = 0;
 732         int range_whole = 0;
 733
 734         if (wbc->nonblocking && bdi_write_congested(bdi)) {
 735                 wbc->encountered_congestion = 1;
 736                 return 0;
 737         }
 738
 739         writepage = NULL;
 740         if (get_block == NULL)
 741                 writepage = mapping->a_ops->writepage;
 742
 743         pagevec_init(&pvec, 0);
 744         if (wbc->range_cyclic) {
 745                 index = mapping->writeback_index; /* Start from prev offset */
 746                 end = -1;
 747         } else {
 748                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
 749                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
 750                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 751                         range_whole = 1;
 752                 scanned = 1;
 753         }
 754 retry:
 755         while (!done && (index <= end) &&
 756                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 757                         PAGECACHE_TAG_DIRTY,
 758                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
 759                 unsigned i;
 760
 761                 scanned = 1;
 762                 for (i = 0; i < nr_pages; i++) {
 763                         struct page *page = pvec.pages[i];
 764
 765                         /*
 766                          * At this point we hold neither mapping->tree_lock nor
 767                          * lock on the page itself: the page may be truncated or
 768                          * invalidated (changing page->mapping to NULL), or even
 769                          * swizzled back from swapper_space to tmpfs file
 770                          * mapping
 771                          */
 772
 773                         lock_page(page);
 774
 775 #if 0   // mask by Victor Yu. 02-12-2007
 776                         if (unlikely(page->mapping != mapping)) {
 777 #else
 778                         if (unlikely(page->u.xx.mapping != mapping)) {
 779 #endif
 780                                 unlock_page(page);
 781                                 continue;
 782                         }
 783
 784                         if (!wbc->range_cyclic && page->index > end) {
 785                                 done = 1;
 786                                 unlock_page(page);
 787                                 continue;
 788                         }
 789
 790                         if (wbc->sync_mode != WB_SYNC_NONE)
 791                                 wait_on_page_writeback(page);
 792
 793                         if (PageWriteback(page) ||
 794                                         !clear_page_dirty_for_io(page)) {
 795                                 unlock_page(page);
 796                                 continue;
 797                         }
 798
 799                         if (writepage) {
 800                                 ret = (*writepage)(page, wbc);
 801                                 if (ret) {
 802                                         if (ret == -ENOSPC)
 803                                                 set_bit(AS_ENOSPC,
 804                                                         &mapping->flags);
 805                                         else
 806                                                 set_bit(AS_EIO,
 807                                                         &mapping->flags);
 808                                 }
 809                         } else {
 810 #if 0   // mask by Victor Yu. 02-12-2007
 811                                 bio = __mpage_writepage(bio, page, get_block,
 812                                                 &last_block_in_bio, &ret, wbc,
 813                                                 page->mapping->a_ops->writepage);
 814 #else
 815                                 bio = __mpage_writepage(bio, page, get_block,
 816                                                 &last_block_in_bio, &ret, wbc,
 817                                                 page->u.xx.mapping->a_ops->writepage);
 818 #endif
 819                         }
 820                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE))
 821                                 unlock_page(page);
 822                         if (ret || (--(wbc->nr_to_write) <= 0))
 823                                 done = 1;
 824                         if (wbc->nonblocking && bdi_write_congested(bdi)) {
 825                                 wbc->encountered_congestion = 1;
 826                                 done = 1;
 827                         }
 828                 }
 829                 pagevec_release(&pvec);
 830                 cond_resched();
 831         }
 832         if (!scanned && !done) {
 833                 /*
 834                  * We hit the last page and there is more work to be done: wrap
 835                  * back to the start of the file
 836                  */
 837                 scanned = 1;
 838                 index = 0;
 839                 goto retry;
 840         }
 841         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
 842                 mapping->writeback_index = index;
 843         if (bio)
 844                 mpage_bio_submit(WRITE, bio);
 845         return ret;
 846 }
 847 EXPORT_SYMBOL(mpage_writepages);
 848
 849 int mpage_writepage(struct page *page, get_block_t get_block,
 850         struct writeback_control *wbc)
 851 {
 852         int ret = 0;
 853         struct bio *bio;
 854         sector_t last_block_in_bio = 0;
 855
 856         bio = __mpage_writepage(NULL, page, get_block,
 857                         &last_block_in_bio, &ret, wbc, NULL);
 858         if (bio)
 859                 mpage_bio_submit(WRITE, bio);
 860
 861         return ret;
 862 }
 863 EXPORT_SYMBOL(mpage_writepage);