fs/ext4/balloc.c

   1 /*
   2  *  linux/fs/ext4/balloc.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
  10  *  Big-endian to little-endian byte-swapping/bitmaps by
  11  *        David S. Miller (davem@caip.rutgers.edu), 1995
  12  */
  13
  14 #include <linux/time.h>
  15 #include <linux/capability.h>
  16 #include <linux/fs.h>
  17 #include <linux/jbd2.h>
  18 #include <linux/quotaops.h>
  19 #include <linux/buffer_head.h>
  20 #include "ext4.h"
  21 #include "ext4_jbd2.h"
  22 #include "mballoc.h"
  23
  24 /*
  25  * balloc.c contains the blocks allocation and deallocation routines
  26  */
  27
  28 /*
  29  * Calculate the block group number and offset, given a block number
  30  */
  31 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
  32                 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
  33 {
  34         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
  35         ext4_grpblk_t offset;
  36
  37         blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
  38         offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
  39         if (offsetp)
  40                 *offsetp = offset;
  41         if (blockgrpp)
  42                 *blockgrpp = blocknr;
  43
  44 }
  45
  46 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
  47                         ext4_group_t block_group)
  48 {
  49         ext4_group_t actual_group;
  50         ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
  51         if (actual_group == block_group)
  52                 return 1;
  53         return 0;
  54 }
  55
  56 static int ext4_group_used_meta_blocks(struct super_block *sb,
  57                                        ext4_group_t block_group,
  58                                        struct ext4_group_desc *gdp)
  59 {
  60         ext4_fsblk_t tmp;
  61         struct ext4_sb_info *sbi = EXT4_SB(sb);
  62         /* block bitmap, inode bitmap, and inode table blocks */
  63         int used_blocks = sbi->s_itb_per_group + 2;
  64
  65         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
  66                 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
  67                                         block_group))
  68                         used_blocks--;
  69
  70                 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
  71                                         block_group))
  72                         used_blocks--;
  73
  74                 tmp = ext4_inode_table(sb, gdp);
  75                 for (; tmp < ext4_inode_table(sb, gdp) +
  76                                 sbi->s_itb_per_group; tmp++) {
  77                         if (!ext4_block_in_group(sb, tmp, block_group))
  78                                 used_blocks -= 1;
  79                 }
  80         }
  81         return used_blocks;
  82 }
  83
  84 /* Initializes an uninitialized block bitmap if given, and returns the
  85  * number of blocks free in the group. */
  86 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
  87                  ext4_group_t block_group, struct ext4_group_desc *gdp)
  88 {
  89         int bit, bit_max;
  90         ext4_group_t ngroups = ext4_get_groups_count(sb);
  91         unsigned free_blocks, group_blocks;
  92         struct ext4_sb_info *sbi = EXT4_SB(sb);
  93
  94         if (bh) {
  95                 J_ASSERT_BH(bh, buffer_locked(bh));
  96
  97                 /* If checksum is bad mark all blocks used to prevent allocation
  98                  * essentially implementing a per-group read-only flag. */
  99                 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
 100                         ext4_error(sb, __func__,
 101                                   "Checksum bad for group %u", block_group);
 102                         ext4_free_blks_set(sb, gdp, 0);
 103                         ext4_free_inodes_set(sb, gdp, 0);
 104                         ext4_itable_unused_set(sb, gdp, 0);
 105                         memset(bh->b_data, 0xff, sb->s_blocksize);
 106                         return 0;
 107                 }
 108                 memset(bh->b_data, 0, sb->s_blocksize);
 109         }
 110
 111         /* Check for superblock and gdt backups in this group */
 112         bit_max = ext4_bg_has_super(sb, block_group);
 113
 114         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
 115             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 116                           sbi->s_desc_per_block) {
 117                 if (bit_max) {
 118                         bit_max += ext4_bg_num_gdb(sb, block_group);
 119                         bit_max +=
 120                                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 121                 }
 122         } else { /* For META_BG_BLOCK_GROUPS */
 123                 bit_max += ext4_bg_num_gdb(sb, block_group);
 124         }
 125
 126         if (block_group == ngroups - 1) {
 127                 /*
 128                  * Even though mke2fs always initialize first and last group
 129                  * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
 130                  * to make sure we calculate the right free blocks
 131                  */
 132                 group_blocks = ext4_blocks_count(sbi->s_es) -
 133                         le32_to_cpu(sbi->s_es->s_first_data_block) -
 134                         (EXT4_BLOCKS_PER_GROUP(sb) * (ngroups - 1));
 135         } else {
 136                 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
 137         }
 138
 139         free_blocks = group_blocks - bit_max;
 140
 141         if (bh) {
 142                 ext4_fsblk_t start, tmp;
 143                 int flex_bg = 0;
 144
 145                 for (bit = 0; bit < bit_max; bit++)
 146                         ext4_set_bit(bit, bh->b_data);
 147
 148                 start = ext4_group_first_block_no(sb, block_group);
 149
 150                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
 151                                               EXT4_FEATURE_INCOMPAT_FLEX_BG))
 152                         flex_bg = 1;
 153
 154                 /* Set bits for block and inode bitmaps, and inode table */
 155                 tmp = ext4_block_bitmap(sb, gdp);
 156                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 157                         ext4_set_bit(tmp - start, bh->b_data);
 158
 159                 tmp = ext4_inode_bitmap(sb, gdp);
 160                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 161                         ext4_set_bit(tmp - start, bh->b_data);
 162
 163                 tmp = ext4_inode_table(sb, gdp);
 164                 for (; tmp < ext4_inode_table(sb, gdp) +
 165                                 sbi->s_itb_per_group; tmp++) {
 166                         if (!flex_bg ||
 167                                 ext4_block_in_group(sb, tmp, block_group))
 168                                 ext4_set_bit(tmp - start, bh->b_data);
 169                 }
 170                 /*
 171                  * Also if the number of blocks within the group is
 172                  * less than the blocksize * 8 ( which is the size
 173                  * of bitmap ), set rest of the block bitmap to 1
 174                  */
 175                 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
 176         }
 177         return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
 178 }
 179
 180
 181 /*
 182  * The free blocks are managed by bitmaps.  A file system contains several
 183  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 184  * block for inodes, N blocks for the inode table and data blocks.
 185  *
 186  * The file system contains group descriptors which are located after the
 187  * super block.  Each descriptor contains the number of the bitmap block and
 188  * the free blocks count in the block.  The descriptors are loaded in memory
 189  * when a file system is mounted (see ext4_fill_super).
 190  */
 191
 192 /**
 193  * ext4_get_group_desc() -- load group descriptor from disk
 194  * @sb:                 super block
 195  * @block_group:        given block group
 196  * @bh:                 pointer to the buffer head to store the block
 197  *                      group descriptor
 198  */
 199 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 200                                              ext4_group_t block_group,
 201                                              struct buffer_head **bh)
 202 {
 203         unsigned int group_desc;
 204         unsigned int offset;
 205         ext4_group_t ngroups = ext4_get_groups_count(sb);
 206         struct ext4_group_desc *desc;
 207         struct ext4_sb_info *sbi = EXT4_SB(sb);
 208
 209         if (block_group >= ngroups) {
 210                 ext4_error(sb, "ext4_get_group_desc",
 211                            "block_group >= groups_count - "
 212                            "block_group = %u, groups_count = %u",
 213                            block_group, ngroups);
 214
 215                 return NULL;
 216         }
 217
 218         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 219         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 220         if (!sbi->s_group_desc[group_desc]) {
 221                 ext4_error(sb, "ext4_get_group_desc",
 222                            "Group descriptor not loaded - "
 223                            "block_group = %u, group_desc = %u, desc = %u",
 224                            block_group, group_desc, offset);
 225                 return NULL;
 226         }
 227
 228         desc = (struct ext4_group_desc *)(
 229                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
 230                 offset * EXT4_DESC_SIZE(sb));
 231         if (bh)
 232                 *bh = sbi->s_group_desc[group_desc];
 233         return desc;
 234 }
 235
 236 static int ext4_valid_block_bitmap(struct super_block *sb,
 237                                         struct ext4_group_desc *desc,
 238                                         unsigned int block_group,
 239                                         struct buffer_head *bh)
 240 {
 241         ext4_grpblk_t offset;
 242         ext4_grpblk_t next_zero_bit;
 243         ext4_fsblk_t bitmap_blk;
 244         ext4_fsblk_t group_first_block;
 245
 246         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
 247                 /* with FLEX_BG, the inode/block bitmaps and itable
 248                  * blocks may not be in the group at all
 249                  * so the bitmap validation will be skipped for those groups
 250                  * or it has to also read the block group where the bitmaps
 251                  * are located to verify they are set.
 252                  */
 253                 return 1;
 254         }
 255         group_first_block = ext4_group_first_block_no(sb, block_group);
 256
 257         /* check whether block bitmap block number is set */
 258         bitmap_blk = ext4_block_bitmap(sb, desc);
 259         offset = bitmap_blk - group_first_block;
 260         if (!ext4_test_bit(offset, bh->b_data))
 261                 /* bad block bitmap */
 262                 goto err_out;
 263
 264         /* check whether the inode bitmap block number is set */
 265         bitmap_blk = ext4_inode_bitmap(sb, desc);
 266         offset = bitmap_blk - group_first_block;
 267         if (!ext4_test_bit(offset, bh->b_data))
 268                 /* bad block bitmap */
 269                 goto err_out;
 270
 271         /* check whether the inode table block number is set */
 272         bitmap_blk = ext4_inode_table(sb, desc);
 273         offset = bitmap_blk - group_first_block;
 274         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 275                                 offset + EXT4_SB(sb)->s_itb_per_group,
 276                                 offset);
 277         if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
 278                 /* good bitmap for inode tables */
 279                 return 1;
 280
 281 err_out:
 282         ext4_error(sb, __func__,
 283                         "Invalid block bitmap - "
 284                         "block_group = %d, block = %llu",
 285                         block_group, bitmap_blk);
 286         return 0;
 287 }
 288 /**
 289  * ext4_read_block_bitmap()
 290  * @sb:                 super block
 291  * @block_group:        given block group
 292  *
 293  * Read the bitmap for a given block_group,and validate the
 294  * bits for block/inode/inode tables are set in the bitmaps
 295  *
 296  * Return buffer_head on success or NULL in case of failure.
 297  */
 298 struct buffer_head *
 299 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 300 {
 301         struct ext4_group_desc *desc;
 302         struct buffer_head *bh = NULL;
 303         ext4_fsblk_t bitmap_blk;
 304
 305         desc = ext4_get_group_desc(sb, block_group, NULL);
 306         if (!desc)
 307                 return NULL;
 308         bitmap_blk = ext4_block_bitmap(sb, desc);
 309         bh = sb_getblk(sb, bitmap_blk);
 310         if (unlikely(!bh)) {
 311                 ext4_error(sb, __func__,
 312                             "Cannot read block bitmap - "
 313                             "block_group = %u, block_bitmap = %llu",
 314                             block_group, bitmap_blk);
 315                 return NULL;
 316         }
 317
 318         if (bitmap_uptodate(bh))
 319                 return bh;
 320
 321         lock_buffer(bh);
 322         if (bitmap_uptodate(bh)) {
 323                 unlock_buffer(bh);
 324                 return bh;
 325         }
 326         ext4_lock_group(sb, block_group);
 327         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 328                 ext4_init_block_bitmap(sb, bh, block_group, desc);
 329                 set_bitmap_uptodate(bh);
 330                 set_buffer_uptodate(bh);
 331                 ext4_unlock_group(sb, block_group);
 332                 unlock_buffer(bh);
 333                 return bh;
 334         }
 335         ext4_unlock_group(sb, block_group);
 336         if (buffer_uptodate(bh)) {
 337                 /*
 338                  * if not uninit if bh is uptodate,
 339                  * bitmap is also uptodate
 340                  */
 341                 set_bitmap_uptodate(bh);
 342                 unlock_buffer(bh);
 343                 return bh;
 344         }
 345         /*
 346          * submit the buffer_head for read. We can
 347          * safely mark the bitmap as uptodate now.
 348          * We do it here so the bitmap uptodate bit
 349          * get set with buffer lock held.
 350          */
 351         set_bitmap_uptodate(bh);
 352         if (bh_submit_read(bh) < 0) {
 353                 put_bh(bh);
 354                 ext4_error(sb, __func__,
 355                             "Cannot read block bitmap - "
 356                             "block_group = %u, block_bitmap = %llu",
 357                             block_group, bitmap_blk);
 358                 return NULL;
 359         }
 360         ext4_valid_block_bitmap(sb, desc, block_group, bh);
 361         /*
 362          * file system mounted not to panic on error,
 363          * continue with corrupt bitmap
 364          */
 365         return bh;
 366 }
 367
 368 /**
 369  * ext4_add_groupblocks() -- Add given blocks to an existing group
 370  * @handle:                     handle to this transaction
 371  * @sb:                         super block
 372  * @block:                      start physcial block to add to the block group
 373  * @count:                      number of blocks to free
 374  *
 375  * This marks the blocks as free in the bitmap. We ask the
 376  * mballoc to reload the buddy after this by setting group
 377  * EXT4_GROUP_INFO_NEED_INIT_BIT flag
 378  */
 379 void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
 380                          ext4_fsblk_t block, unsigned long count)
 381 {
 382         struct buffer_head *bitmap_bh = NULL;
 383         struct buffer_head *gd_bh;
 384         ext4_group_t block_group;
 385         ext4_grpblk_t bit;
 386         unsigned int i;
 387         struct ext4_group_desc *desc;
 388         struct ext4_super_block *es;
 389         struct ext4_sb_info *sbi;
 390         int err = 0, ret, blk_free_count;
 391         ext4_grpblk_t blocks_freed;
 392         struct ext4_group_info *grp;
 393
 394         sbi = EXT4_SB(sb);
 395         es = sbi->s_es;
 396         ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
 397
 398         ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
 399         grp = ext4_get_group_info(sb, block_group);
 400         /*
 401          * Check to see if we are freeing blocks across a group
 402          * boundary.
 403          */
 404         if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
 405                 goto error_return;
 406         }
 407         bitmap_bh = ext4_read_block_bitmap(sb, block_group);
 408         if (!bitmap_bh)
 409                 goto error_return;
 410         desc = ext4_get_group_desc(sb, block_group, &gd_bh);
 411         if (!desc)
 412                 goto error_return;
 413
 414         if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
 415             in_range(ext4_inode_bitmap(sb, desc), block, count) ||
 416             in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
 417             in_range(block + count - 1, ext4_inode_table(sb, desc),
 418                      sbi->s_itb_per_group)) {
 419                 ext4_error(sb, __func__,
 420                            "Adding blocks in system zones - "
 421                            "Block = %llu, count = %lu",
 422                            block, count);
 423                 goto error_return;
 424         }
 425
 426         /*
 427          * We are about to add blocks to the bitmap,
 428          * so we need undo access.
 429          */
 430         BUFFER_TRACE(bitmap_bh, "getting undo access");
 431         err = ext4_journal_get_undo_access(handle, bitmap_bh);
 432         if (err)
 433                 goto error_return;
 434
 435         /*
 436          * We are about to modify some metadata.  Call the journal APIs
 437          * to unshare ->b_data if a currently-committing transaction is
 438          * using it
 439          */
 440         BUFFER_TRACE(gd_bh, "get_write_access");
 441         err = ext4_journal_get_write_access(handle, gd_bh);
 442         if (err)
 443                 goto error_return;
 444         /*
 445          * make sure we don't allow a parallel init on other groups in the
 446          * same buddy cache
 447          */
 448         down_write(&grp->alloc_sem);
 449         for (i = 0, blocks_freed = 0; i < count; i++) {
 450                 BUFFER_TRACE(bitmap_bh, "clear bit");
 451                 if (!ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
 452                                                 bit + i, bitmap_bh->b_data)) {
 453                         ext4_error(sb, __func__,
 454                                    "bit already cleared for block %llu",
 455                                    (ext4_fsblk_t)(block + i));
 456                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
 457                 } else {
 458                         blocks_freed++;
 459                 }
 460         }
 461         ext4_lock_group(sb, block_group);
 462         blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
 463         ext4_free_blks_set(sb, desc, blk_free_count);
 464         desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
 465         ext4_unlock_group(sb, block_group);
 466         percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
 467
 468         if (sbi->s_log_groups_per_flex) {
 469                 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
 470                 atomic_add(blocks_freed,
 471                            &sbi->s_flex_groups[flex_group].free_blocks);
 472         }
 473         /*
 474          * request to reload the buddy with the
 475          * new bitmap information
 476          */
 477         set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
 478         grp->bb_free += blocks_freed;
 479         up_write(&grp->alloc_sem);
 480
 481         /* We dirtied the bitmap block */
 482         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 483         err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 484
 485         /* And the group descriptor block */
 486         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
 487         ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
 488         if (!err)
 489                 err = ret;
 490         sb->s_dirt = 1;
 491
 492 error_return:
 493         brelse(bitmap_bh);
 494         ext4_std_error(sb, err);
 495         return;
 496 }
 497
 498 /**
 499  * ext4_free_blocks() -- Free given blocks and update quota
 500  * @handle:             handle for this transaction
 501  * @inode:              inode
 502  * @block:              start physical block to free
 503  * @count:              number of blocks to count
 504  * @metadata:           Are these metadata blocks
 505  */
 506 void ext4_free_blocks(handle_t *handle, struct inode *inode,
 507                         ext4_fsblk_t block, unsigned long count,
 508                         int metadata)
 509 {
 510         struct super_block *sb;
 511         unsigned long dquot_freed_blocks;
 512
 513         /* this isn't the right place to decide whether block is metadata
 514          * inode.c/extents.c knows better, but for safety ... */
 515         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
 516                 metadata = 1;
 517
 518         /* We need to make sure we don't reuse
 519          * block released untill the transaction commit.
 520          * writeback mode have weak data consistency so
 521          * don't force data as metadata when freeing block
 522          * for writeback mode.
 523          */
 524         if (metadata == 0 && !ext4_should_writeback_data(inode))
 525                 metadata = 1;
 526
 527         sb = inode->i_sb;
 528
 529         ext4_mb_free_blocks(handle, inode, block, count,
 530                             metadata, &dquot_freed_blocks);
 531         if (dquot_freed_blocks)
 532                 vfs_dq_free_block(inode, dquot_freed_blocks);
 533         return;
 534 }
 535
 536 /**
 537  * ext4_has_free_blocks()
 538  * @sbi:        in-core super block structure.
 539  * @nblocks:    number of needed blocks
 540  *
 541  * Check if filesystem has nblocks free & available for allocation.
 542  * On success return 1, return 0 on failure.
 543  */
 544 int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
 545 {
 546         s64 free_blocks, dirty_blocks, root_blocks;
 547         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
 548         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
 549
 550         free_blocks  = percpu_counter_read_positive(fbc);
 551         dirty_blocks = percpu_counter_read_positive(dbc);
 552         root_blocks = ext4_r_blocks_count(sbi->s_es);
 553
 554         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
 555                                                 EXT4_FREEBLOCKS_WATERMARK) {
 556                 free_blocks  = percpu_counter_sum_positive(fbc);
 557                 dirty_blocks = percpu_counter_sum_positive(dbc);
 558                 if (dirty_blocks < 0) {
 559                         printk(KERN_CRIT "Dirty block accounting "
 560                                         "went wrong %lld\n",
 561                                         (long long)dirty_blocks);
 562                 }
 563         }
 564         /* Check whether we have space after
 565          * accounting for current dirty blocks & root reserved blocks.
 566          */
 567         if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
 568                 return 1;
 569
 570         /* Hm, nope.  Are (enough) root reserved blocks available? */
 571         if (sbi->s_resuid == current_fsuid() ||
 572             ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
 573             capable(CAP_SYS_RESOURCE)) {
 574                 if (free_blocks >= (nblocks + dirty_blocks))
 575                         return 1;
 576         }
 577
 578         return 0;
 579 }
 580
 581 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
 582                                                 s64 nblocks)
 583 {
 584         if (ext4_has_free_blocks(sbi, nblocks)) {
 585                 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
 586                 return 0;
 587         } else
 588                 return -ENOSPC;
 589 }
 590
 591 /**
 592  * ext4_should_retry_alloc()
 593  * @sb:                 super block
 594  * @retries             number of attemps has been made
 595  *
 596  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 597  * it is profitable to retry the operation, this function will wait
 598  * for the current or commiting transaction to complete, and then
 599  * return TRUE.
 600  *
 601  * if the total number of retries exceed three times, return FALSE.
 602  */
 603 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 604 {
 605         if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
 606             (*retries)++ > 3 ||
 607             !EXT4_SB(sb)->s_journal)
 608                 return 0;
 609
 610         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 611
 612         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 613 }
 614
 615 /*
 616  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 617  *
 618  * @handle:             handle to this transaction
 619  * @inode:              file inode
 620  * @goal:               given target block(filesystem wide)
 621  * @count:              pointer to total number of blocks needed
 622  * @errp:               error code
 623  *
 624  * Return 1st allocated block number on success, *count stores total account
 625  * error stores in errp pointer
 626  */
 627 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 628                 ext4_fsblk_t goal, unsigned long *count, int *errp)
 629 {
 630         struct ext4_allocation_request ar;
 631         ext4_fsblk_t ret;
 632
 633         memset(&ar, 0, sizeof(ar));
 634         /* Fill with neighbour allocated blocks */
 635         ar.inode = inode;
 636         ar.goal = goal;
 637         ar.len = count ? *count : 1;
 638
 639         ret = ext4_mb_new_blocks(handle, &ar, errp);
 640         if (count)
 641                 *count = ar.len;
 642
 643         /*
 644          * Account for the allocated meta blocks
 645          */
 646         if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
 647                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 648                 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
 649                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 650         }
 651         return ret;
 652 }
 653
 654 /**
 655  * ext4_count_free_blocks() -- count filesystem free blocks
 656  * @sb:         superblock
 657  *
 658  * Adds up the number of free blocks from each block group.
 659  */
 660 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
 661 {
 662         ext4_fsblk_t desc_count;
 663         struct ext4_group_desc *gdp;
 664         ext4_group_t i;
 665         ext4_group_t ngroups = ext4_get_groups_count(sb);
 666 #ifdef EXT4FS_DEBUG
 667         struct ext4_super_block *es;
 668         ext4_fsblk_t bitmap_count;
 669         unsigned int x;
 670         struct buffer_head *bitmap_bh = NULL;
 671
 672         es = EXT4_SB(sb)->s_es;
 673         desc_count = 0;
 674         bitmap_count = 0;
 675         gdp = NULL;
 676
 677         for (i = 0; i < ngroups; i++) {
 678                 gdp = ext4_get_group_desc(sb, i, NULL);
 679                 if (!gdp)
 680                         continue;
 681                 desc_count += ext4_free_blks_count(sb, gdp);
 682                 brelse(bitmap_bh);
 683                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 684                 if (bitmap_bh == NULL)
 685                         continue;
 686
 687                 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
 688                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 689                         i, ext4_free_blks_count(sb, gdp), x);
 690                 bitmap_count += x;
 691         }
 692         brelse(bitmap_bh);
 693         printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
 694                 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
 695                desc_count, bitmap_count);
 696         return bitmap_count;
 697 #else
 698         desc_count = 0;
 699         for (i = 0; i < ngroups; i++) {
 700                 gdp = ext4_get_group_desc(sb, i, NULL);
 701                 if (!gdp)
 702                         continue;
 703                 desc_count += ext4_free_blks_count(sb, gdp);
 704         }
 705
 706         return desc_count;
 707 #endif
 708 }
 709
 710 static inline int test_root(ext4_group_t a, int b)
 711 {
 712         int num = b;
 713
 714         while (a > num)
 715                 num *= b;
 716         return num == a;
 717 }
 718
 719 static int ext4_group_sparse(ext4_group_t group)
 720 {
 721         if (group <= 1)
 722                 return 1;
 723         if (!(group & 1))
 724                 return 0;
 725         return (test_root(group, 7) || test_root(group, 5) ||
 726                 test_root(group, 3));
 727 }
 728
 729 /**
 730  *      ext4_bg_has_super - number of blocks used by the superblock in group
 731  *      @sb: superblock for filesystem
 732  *      @group: group number to check
 733  *
 734  *      Return the number of blocks used by the superblock (primary or backup)
 735  *      in this group.  Currently this will be only 0 or 1.
 736  */
 737 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 738 {
 739         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
 740                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
 741                         !ext4_group_sparse(group))
 742                 return 0;
 743         return 1;
 744 }
 745
 746 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 747                                         ext4_group_t group)
 748 {
 749         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 750         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 751         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 752
 753         if (group == first || group == first + 1 || group == last)
 754                 return 1;
 755         return 0;
 756 }
 757
 758 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 759                                         ext4_group_t group)
 760 {
 761         if (!ext4_bg_has_super(sb, group))
 762                 return 0;
 763
 764         if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
 765                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 766         else
 767                 return EXT4_SB(sb)->s_gdb_count;
 768 }
 769
 770 /**
 771  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 772  *      @sb: superblock for filesystem
 773  *      @group: group number to check
 774  *
 775  *      Return the number of blocks used by the group descriptor table
 776  *      (primary or backup) in this group.  In the future there may be a
 777  *      different number of descriptor blocks in each group.
 778  */
 779 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 780 {
 781         unsigned long first_meta_bg =
 782                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 783         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 784
 785         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
 786                         metagroup < first_meta_bg)
 787                 return ext4_bg_num_gdb_nometa(sb, group);
 788
 789         return ext4_bg_num_gdb_meta(sb,group);
 790
 791 }
 792