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, "Checksum bad for group %u",
 101                                         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                         ext4_group_first_block_no(sb, ngroups - 1);
 134         } else {
 135                 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
 136         }
 137
 138         free_blocks = group_blocks - bit_max;
 139
 140         if (bh) {
 141                 ext4_fsblk_t start, tmp;
 142                 int flex_bg = 0;
 143
 144                 for (bit = 0; bit < bit_max; bit++)
 145                         ext4_set_bit(bit, bh->b_data);
 146
 147                 start = ext4_group_first_block_no(sb, block_group);
 148
 149                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
 150                                               EXT4_FEATURE_INCOMPAT_FLEX_BG))
 151                         flex_bg = 1;
 152
 153                 /* Set bits for block and inode bitmaps, and inode table */
 154                 tmp = ext4_block_bitmap(sb, gdp);
 155                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 156                         ext4_set_bit(tmp - start, bh->b_data);
 157
 158                 tmp = ext4_inode_bitmap(sb, gdp);
 159                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 160                         ext4_set_bit(tmp - start, bh->b_data);
 161
 162                 tmp = ext4_inode_table(sb, gdp);
 163                 for (; tmp < ext4_inode_table(sb, gdp) +
 164                                 sbi->s_itb_per_group; tmp++) {
 165                         if (!flex_bg ||
 166                                 ext4_block_in_group(sb, tmp, block_group))
 167                                 ext4_set_bit(tmp - start, bh->b_data);
 168                 }
 169                 /*
 170                  * Also if the number of blocks within the group is
 171                  * less than the blocksize * 8 ( which is the size
 172                  * of bitmap ), set rest of the block bitmap to 1
 173                  */
 174                 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
 175         }
 176         return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
 177 }
 178
 179
 180 /*
 181  * The free blocks are managed by bitmaps.  A file system contains several
 182  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 183  * block for inodes, N blocks for the inode table and data blocks.
 184  *
 185  * The file system contains group descriptors which are located after the
 186  * super block.  Each descriptor contains the number of the bitmap block and
 187  * the free blocks count in the block.  The descriptors are loaded in memory
 188  * when a file system is mounted (see ext4_fill_super).
 189  */
 190
 191 /**
 192  * ext4_get_group_desc() -- load group descriptor from disk
 193  * @sb:                 super block
 194  * @block_group:        given block group
 195  * @bh:                 pointer to the buffer head to store the block
 196  *                      group descriptor
 197  */
 198 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 199                                              ext4_group_t block_group,
 200                                              struct buffer_head **bh)
 201 {
 202         unsigned int group_desc;
 203         unsigned int offset;
 204         ext4_group_t ngroups = ext4_get_groups_count(sb);
 205         struct ext4_group_desc *desc;
 206         struct ext4_sb_info *sbi = EXT4_SB(sb);
 207
 208         if (block_group >= ngroups) {
 209                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
 210                            " groups_count = %u", block_group, ngroups);
 211
 212                 return NULL;
 213         }
 214
 215         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 216         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 217         if (!sbi->s_group_desc[group_desc]) {
 218                 ext4_error(sb, "Group descriptor not loaded - "
 219                            "block_group = %u, group_desc = %u, desc = %u",
 220                            block_group, group_desc, offset);
 221                 return NULL;
 222         }
 223
 224         desc = (struct ext4_group_desc *)(
 225                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
 226                 offset * EXT4_DESC_SIZE(sb));
 227         if (bh)
 228                 *bh = sbi->s_group_desc[group_desc];
 229         return desc;
 230 }
 231
 232 static int ext4_valid_block_bitmap(struct super_block *sb,
 233                                         struct ext4_group_desc *desc,
 234                                         unsigned int block_group,
 235                                         struct buffer_head *bh)
 236 {
 237         ext4_grpblk_t offset;
 238         ext4_grpblk_t next_zero_bit;
 239         ext4_fsblk_t bitmap_blk;
 240         ext4_fsblk_t group_first_block;
 241
 242         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
 243                 /* with FLEX_BG, the inode/block bitmaps and itable
 244                  * blocks may not be in the group at all
 245                  * so the bitmap validation will be skipped for those groups
 246                  * or it has to also read the block group where the bitmaps
 247                  * are located to verify they are set.
 248                  */
 249                 return 1;
 250         }
 251         group_first_block = ext4_group_first_block_no(sb, block_group);
 252
 253         /* check whether block bitmap block number is set */
 254         bitmap_blk = ext4_block_bitmap(sb, desc);
 255         offset = bitmap_blk - group_first_block;
 256         if (!ext4_test_bit(offset, bh->b_data))
 257                 /* bad block bitmap */
 258                 goto err_out;
 259
 260         /* check whether the inode bitmap block number is set */
 261         bitmap_blk = ext4_inode_bitmap(sb, desc);
 262         offset = bitmap_blk - group_first_block;
 263         if (!ext4_test_bit(offset, bh->b_data))
 264                 /* bad block bitmap */
 265                 goto err_out;
 266
 267         /* check whether the inode table block number is set */
 268         bitmap_blk = ext4_inode_table(sb, desc);
 269         offset = bitmap_blk - group_first_block;
 270         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 271                                 offset + EXT4_SB(sb)->s_itb_per_group,
 272                                 offset);
 273         if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
 274                 /* good bitmap for inode tables */
 275                 return 1;
 276
 277 err_out:
 278         ext4_error(sb, "Invalid block bitmap - block_group = %d, block = %llu",
 279                         block_group, bitmap_blk);
 280         return 0;
 281 }
 282 /**
 283  * ext4_read_block_bitmap()
 284  * @sb:                 super block
 285  * @block_group:        given block group
 286  *
 287  * Read the bitmap for a given block_group,and validate the
 288  * bits for block/inode/inode tables are set in the bitmaps
 289  *
 290  * Return buffer_head on success or NULL in case of failure.
 291  */
 292 struct buffer_head *
 293 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 294 {
 295         struct ext4_group_desc *desc;
 296         struct buffer_head *bh = NULL;
 297         ext4_fsblk_t bitmap_blk;
 298
 299         desc = ext4_get_group_desc(sb, block_group, NULL);
 300         if (!desc)
 301                 return NULL;
 302         bitmap_blk = ext4_block_bitmap(sb, desc);
 303         bh = sb_getblk(sb, bitmap_blk);
 304         if (unlikely(!bh)) {
 305                 ext4_error(sb, "Cannot read block bitmap - "
 306                             "block_group = %u, block_bitmap = %llu",
 307                             block_group, bitmap_blk);
 308                 return NULL;
 309         }
 310
 311         if (bitmap_uptodate(bh))
 312                 return bh;
 313
 314         lock_buffer(bh);
 315         if (bitmap_uptodate(bh)) {
 316                 unlock_buffer(bh);
 317                 return bh;
 318         }
 319         ext4_lock_group(sb, block_group);
 320         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 321                 ext4_init_block_bitmap(sb, bh, block_group, desc);
 322                 set_bitmap_uptodate(bh);
 323                 set_buffer_uptodate(bh);
 324                 ext4_unlock_group(sb, block_group);
 325                 unlock_buffer(bh);
 326                 return bh;
 327         }
 328         ext4_unlock_group(sb, block_group);
 329         if (buffer_uptodate(bh)) {
 330                 /*
 331                  * if not uninit if bh is uptodate,
 332                  * bitmap is also uptodate
 333                  */
 334                 set_bitmap_uptodate(bh);
 335                 unlock_buffer(bh);
 336                 return bh;
 337         }
 338         /*
 339          * submit the buffer_head for read. We can
 340          * safely mark the bitmap as uptodate now.
 341          * We do it here so the bitmap uptodate bit
 342          * get set with buffer lock held.
 343          */
 344         set_bitmap_uptodate(bh);
 345         if (bh_submit_read(bh) < 0) {
 346                 put_bh(bh);
 347                 ext4_error(sb, "Cannot read block bitmap - "
 348                             "block_group = %u, block_bitmap = %llu",
 349                             block_group, bitmap_blk);
 350                 return NULL;
 351         }
 352         ext4_valid_block_bitmap(sb, desc, block_group, bh);
 353         /*
 354          * file system mounted not to panic on error,
 355          * continue with corrupt bitmap
 356          */
 357         return bh;
 358 }
 359
 360 /**
 361  * ext4_add_groupblocks() -- Add given blocks to an existing group
 362  * @handle:                     handle to this transaction
 363  * @sb:                         super block
 364  * @block:                      start physcial block to add to the block group
 365  * @count:                      number of blocks to free
 366  *
 367  * This marks the blocks as free in the bitmap. We ask the
 368  * mballoc to reload the buddy after this by setting group
 369  * EXT4_GROUP_INFO_NEED_INIT_BIT flag
 370  */
 371 void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
 372                          ext4_fsblk_t block, unsigned long count)
 373 {
 374         struct buffer_head *bitmap_bh = NULL;
 375         struct buffer_head *gd_bh;
 376         ext4_group_t block_group;
 377         ext4_grpblk_t bit;
 378         unsigned int i;
 379         struct ext4_group_desc *desc;
 380         struct ext4_super_block *es;
 381         struct ext4_sb_info *sbi;
 382         int err = 0, ret, blk_free_count;
 383         ext4_grpblk_t blocks_freed;
 384         struct ext4_group_info *grp;
 385
 386         sbi = EXT4_SB(sb);
 387         es = sbi->s_es;
 388         ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
 389
 390         ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
 391         grp = ext4_get_group_info(sb, block_group);
 392         /*
 393          * Check to see if we are freeing blocks across a group
 394          * boundary.
 395          */
 396         if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
 397                 goto error_return;
 398         }
 399         bitmap_bh = ext4_read_block_bitmap(sb, block_group);
 400         if (!bitmap_bh)
 401                 goto error_return;
 402         desc = ext4_get_group_desc(sb, block_group, &gd_bh);
 403         if (!desc)
 404                 goto error_return;
 405
 406         if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
 407             in_range(ext4_inode_bitmap(sb, desc), block, count) ||
 408             in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
 409             in_range(block + count - 1, ext4_inode_table(sb, desc),
 410                      sbi->s_itb_per_group)) {
 411                 ext4_error(sb, "Adding blocks in system zones - "
 412                            "Block = %llu, count = %lu",
 413                            block, count);
 414                 goto error_return;
 415         }
 416
 417         /*
 418          * We are about to add blocks to the bitmap,
 419          * so we need undo access.
 420          */
 421         BUFFER_TRACE(bitmap_bh, "getting undo access");
 422         err = ext4_journal_get_undo_access(handle, bitmap_bh);
 423         if (err)
 424                 goto error_return;
 425
 426         /*
 427          * We are about to modify some metadata.  Call the journal APIs
 428          * to unshare ->b_data if a currently-committing transaction is
 429          * using it
 430          */
 431         BUFFER_TRACE(gd_bh, "get_write_access");
 432         err = ext4_journal_get_write_access(handle, gd_bh);
 433         if (err)
 434                 goto error_return;
 435         /*
 436          * make sure we don't allow a parallel init on other groups in the
 437          * same buddy cache
 438          */
 439         down_write(&grp->alloc_sem);
 440         for (i = 0, blocks_freed = 0; i < count; i++) {
 441                 BUFFER_TRACE(bitmap_bh, "clear bit");
 442                 if (!ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
 443                                                 bit + i, bitmap_bh->b_data)) {
 444                         ext4_error(sb, "bit already cleared for block %llu",
 445                                    (ext4_fsblk_t)(block + i));
 446                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
 447                 } else {
 448                         blocks_freed++;
 449                 }
 450         }
 451         ext4_lock_group(sb, block_group);
 452         blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
 453         ext4_free_blks_set(sb, desc, blk_free_count);
 454         desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
 455         ext4_unlock_group(sb, block_group);
 456         percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
 457
 458         if (sbi->s_log_groups_per_flex) {
 459                 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
 460                 atomic_add(blocks_freed,
 461                            &sbi->s_flex_groups[flex_group].free_blocks);
 462         }
 463         /*
 464          * request to reload the buddy with the
 465          * new bitmap information
 466          */
 467         set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
 468         grp->bb_free += blocks_freed;
 469         up_write(&grp->alloc_sem);
 470
 471         /* We dirtied the bitmap block */
 472         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 473         err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 474
 475         /* And the group descriptor block */
 476         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
 477         ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
 478         if (!err)
 479                 err = ret;
 480         sb->s_dirt = 1;
 481
 482 error_return:
 483         brelse(bitmap_bh);
 484         ext4_std_error(sb, err);
 485         return;
 486 }
 487
 488 /**
 489  * ext4_has_free_blocks()
 490  * @sbi:        in-core super block structure.
 491  * @nblocks:    number of needed blocks
 492  *
 493  * Check if filesystem has nblocks free & available for allocation.
 494  * On success return 1, return 0 on failure.
 495  */
 496 int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
 497 {
 498         s64 free_blocks, dirty_blocks, root_blocks;
 499         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
 500         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
 501
 502         free_blocks  = percpu_counter_read_positive(fbc);
 503         dirty_blocks = percpu_counter_read_positive(dbc);
 504         root_blocks = ext4_r_blocks_count(sbi->s_es);
 505
 506         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
 507                                                 EXT4_FREEBLOCKS_WATERMARK) {
 508                 free_blocks  = percpu_counter_sum_positive(fbc);
 509                 dirty_blocks = percpu_counter_sum_positive(dbc);
 510                 if (dirty_blocks < 0) {
 511                         printk(KERN_CRIT "Dirty block accounting "
 512                                         "went wrong %lld\n",
 513                                         (long long)dirty_blocks);
 514                 }
 515         }
 516         /* Check whether we have space after
 517          * accounting for current dirty blocks & root reserved blocks.
 518          */
 519         if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
 520                 return 1;
 521
 522         /* Hm, nope.  Are (enough) root reserved blocks available? */
 523         if (sbi->s_resuid == current_fsuid() ||
 524             ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
 525             capable(CAP_SYS_RESOURCE)) {
 526                 if (free_blocks >= (nblocks + dirty_blocks))
 527                         return 1;
 528         }
 529
 530         return 0;
 531 }
 532
 533 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
 534                                                 s64 nblocks)
 535 {
 536         if (ext4_has_free_blocks(sbi, nblocks)) {
 537                 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
 538                 return 0;
 539         } else
 540                 return -ENOSPC;
 541 }
 542
 543 /**
 544  * ext4_should_retry_alloc()
 545  * @sb:                 super block
 546  * @retries             number of attemps has been made
 547  *
 548  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 549  * it is profitable to retry the operation, this function will wait
 550  * for the current or commiting transaction to complete, and then
 551  * return TRUE.
 552  *
 553  * if the total number of retries exceed three times, return FALSE.
 554  */
 555 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 556 {
 557         if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
 558             (*retries)++ > 3 ||
 559             !EXT4_SB(sb)->s_journal)
 560                 return 0;
 561
 562         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 563
 564         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 565 }
 566
 567 /*
 568  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 569  *
 570  * @handle:             handle to this transaction
 571  * @inode:              file inode
 572  * @goal:               given target block(filesystem wide)
 573  * @count:              pointer to total number of blocks needed
 574  * @errp:               error code
 575  *
 576  * Return 1st allocated block number on success, *count stores total account
 577  * error stores in errp pointer
 578  */
 579 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 580                 ext4_fsblk_t goal, unsigned long *count, int *errp)
 581 {
 582         struct ext4_allocation_request ar;
 583         ext4_fsblk_t ret;
 584
 585         memset(&ar, 0, sizeof(ar));
 586         /* Fill with neighbour allocated blocks */
 587         ar.inode = inode;
 588         ar.goal = goal;
 589         ar.len = count ? *count : 1;
 590
 591         ret = ext4_mb_new_blocks(handle, &ar, errp);
 592         if (count)
 593                 *count = ar.len;
 594         /*
 595          * Account for the allocated meta blocks.  We will never
 596          * fail EDQUOT for metdata, but we do account for it.
 597          */
 598         if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
 599                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 600                 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
 601                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 602                 dquot_alloc_block_nofail(inode, ar.len);
 603         }
 604         return ret;
 605 }
 606
 607 /**
 608  * ext4_count_free_blocks() -- count filesystem free blocks
 609  * @sb:         superblock
 610  *
 611  * Adds up the number of free blocks from each block group.
 612  */
 613 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
 614 {
 615         ext4_fsblk_t desc_count;
 616         struct ext4_group_desc *gdp;
 617         ext4_group_t i;
 618         ext4_group_t ngroups = ext4_get_groups_count(sb);
 619 #ifdef EXT4FS_DEBUG
 620         struct ext4_super_block *es;
 621         ext4_fsblk_t bitmap_count;
 622         unsigned int x;
 623         struct buffer_head *bitmap_bh = NULL;
 624
 625         es = EXT4_SB(sb)->s_es;
 626         desc_count = 0;
 627         bitmap_count = 0;
 628         gdp = NULL;
 629
 630         for (i = 0; i < ngroups; i++) {
 631                 gdp = ext4_get_group_desc(sb, i, NULL);
 632                 if (!gdp)
 633                         continue;
 634                 desc_count += ext4_free_blks_count(sb, gdp);
 635                 brelse(bitmap_bh);
 636                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 637                 if (bitmap_bh == NULL)
 638                         continue;
 639
 640                 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
 641                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 642                         i, ext4_free_blks_count(sb, gdp), x);
 643                 bitmap_count += x;
 644         }
 645         brelse(bitmap_bh);
 646         printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
 647                 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
 648                desc_count, bitmap_count);
 649         return bitmap_count;
 650 #else
 651         desc_count = 0;
 652         for (i = 0; i < ngroups; i++) {
 653                 gdp = ext4_get_group_desc(sb, i, NULL);
 654                 if (!gdp)
 655                         continue;
 656                 desc_count += ext4_free_blks_count(sb, gdp);
 657         }
 658
 659         return desc_count;
 660 #endif
 661 }
 662
 663 static inline int test_root(ext4_group_t a, int b)
 664 {
 665         int num = b;
 666
 667         while (a > num)
 668                 num *= b;
 669         return num == a;
 670 }
 671
 672 static int ext4_group_sparse(ext4_group_t group)
 673 {
 674         if (group <= 1)
 675                 return 1;
 676         if (!(group & 1))
 677                 return 0;
 678         return (test_root(group, 7) || test_root(group, 5) ||
 679                 test_root(group, 3));
 680 }
 681
 682 /**
 683  *      ext4_bg_has_super - number of blocks used by the superblock in group
 684  *      @sb: superblock for filesystem
 685  *      @group: group number to check
 686  *
 687  *      Return the number of blocks used by the superblock (primary or backup)
 688  *      in this group.  Currently this will be only 0 or 1.
 689  */
 690 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 691 {
 692         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
 693                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
 694                         !ext4_group_sparse(group))
 695                 return 0;
 696         return 1;
 697 }
 698
 699 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 700                                         ext4_group_t group)
 701 {
 702         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 703         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 704         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 705
 706         if (group == first || group == first + 1 || group == last)
 707                 return 1;
 708         return 0;
 709 }
 710
 711 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 712                                         ext4_group_t group)
 713 {
 714         if (!ext4_bg_has_super(sb, group))
 715                 return 0;
 716
 717         if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
 718                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 719         else
 720                 return EXT4_SB(sb)->s_gdb_count;
 721 }
 722
 723 /**
 724  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 725  *      @sb: superblock for filesystem
 726  *      @group: group number to check
 727  *
 728  *      Return the number of blocks used by the group descriptor table
 729  *      (primary or backup) in this group.  In the future there may be a
 730  *      different number of descriptor blocks in each group.
 731  */
 732 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 733 {
 734         unsigned long first_meta_bg =
 735                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 736         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 737
 738         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
 739                         metagroup < first_meta_bg)
 740                 return ext4_bg_num_gdb_nometa(sb, group);
 741
 742         return ext4_bg_num_gdb_meta(sb,group);
 743
 744 }
 745