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 "group.h"
  23 #include "mballoc.h"
  24
  25 /*
  26  * balloc.c contains the blocks allocation and deallocation routines
  27  */
  28
  29 /*
  30  * Calculate the block group number and offset, given a block number
  31  */
  32 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
  33                 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
  34 {
  35         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
  36         ext4_grpblk_t offset;
  37
  38         blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
  39         offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
  40         if (offsetp)
  41                 *offsetp = offset;
  42         if (blockgrpp)
  43                 *blockgrpp = blocknr;
  44
  45 }
  46
  47 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
  48                         ext4_group_t block_group)
  49 {
  50         ext4_group_t actual_group;
  51         ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
  52         if (actual_group == block_group)
  53                 return 1;
  54         return 0;
  55 }
  56
  57 static int ext4_group_used_meta_blocks(struct super_block *sb,
  58                                 ext4_group_t block_group)
  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                 struct ext4_group_desc *gdp;
  67                 struct buffer_head *bh;
  68
  69                 gdp = ext4_get_group_desc(sb, block_group, &bh);
  70                 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
  71                                         block_group))
  72                         used_blocks--;
  73
  74                 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
  75                                         block_group))
  76                         used_blocks--;
  77
  78                 tmp = ext4_inode_table(sb, gdp);
  79                 for (; tmp < ext4_inode_table(sb, gdp) +
  80                                 sbi->s_itb_per_group; tmp++) {
  81                         if (!ext4_block_in_group(sb, tmp, block_group))
  82                                 used_blocks -= 1;
  83                 }
  84         }
  85         return used_blocks;
  86 }
  87
  88 /* Initializes an uninitialized block bitmap if given, and returns the
  89  * number of blocks free in the group. */
  90 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
  91                  ext4_group_t block_group, struct ext4_group_desc *gdp)
  92 {
  93         int bit, bit_max;
  94         unsigned free_blocks, group_blocks;
  95         struct ext4_sb_info *sbi = EXT4_SB(sb);
  96
  97         if (bh) {
  98                 J_ASSERT_BH(bh, buffer_locked(bh));
  99
 100                 /* If checksum is bad mark all blocks used to prevent allocation
 101                  * essentially implementing a per-group read-only flag. */
 102                 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
 103                         ext4_error(sb, __func__,
 104                                   "Checksum bad for group %u", block_group);
 105                         ext4_free_blks_set(sb, gdp, 0);
 106                         ext4_free_inodes_set(sb, gdp, 0);
 107                         ext4_itable_unused_set(sb, gdp, 0);
 108                         memset(bh->b_data, 0xff, sb->s_blocksize);
 109                         return 0;
 110                 }
 111                 memset(bh->b_data, 0, sb->s_blocksize);
 112         }
 113
 114         /* Check for superblock and gdt backups in this group */
 115         bit_max = ext4_bg_has_super(sb, block_group);
 116
 117         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
 118             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 119                           sbi->s_desc_per_block) {
 120                 if (bit_max) {
 121                         bit_max += ext4_bg_num_gdb(sb, block_group);
 122                         bit_max +=
 123                                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 124                 }
 125         } else { /* For META_BG_BLOCK_GROUPS */
 126                 bit_max += ext4_bg_num_gdb(sb, block_group);
 127         }
 128
 129         if (block_group == sbi->s_groups_count - 1) {
 130                 /*
 131                  * Even though mke2fs always initialize first and last group
 132                  * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
 133                  * to make sure we calculate the right free blocks
 134                  */
 135                 group_blocks = ext4_blocks_count(sbi->s_es) -
 136                         le32_to_cpu(sbi->s_es->s_first_data_block) -
 137                         (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count - 1));
 138         } else {
 139                 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
 140         }
 141
 142         free_blocks = group_blocks - bit_max;
 143
 144         if (bh) {
 145                 ext4_fsblk_t start, tmp;
 146                 int flex_bg = 0;
 147
 148                 for (bit = 0; bit < bit_max; bit++)
 149                         ext4_set_bit(bit, bh->b_data);
 150
 151                 start = ext4_group_first_block_no(sb, block_group);
 152
 153                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
 154                                               EXT4_FEATURE_INCOMPAT_FLEX_BG))
 155                         flex_bg = 1;
 156
 157                 /* Set bits for block and inode bitmaps, and inode table */
 158                 tmp = ext4_block_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_bitmap(sb, gdp);
 163                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 164                         ext4_set_bit(tmp - start, bh->b_data);
 165
 166                 tmp = ext4_inode_table(sb, gdp);
 167                 for (; tmp < ext4_inode_table(sb, gdp) +
 168                                 sbi->s_itb_per_group; tmp++) {
 169                         if (!flex_bg ||
 170                                 ext4_block_in_group(sb, tmp, block_group))
 171                                 ext4_set_bit(tmp - start, bh->b_data);
 172                 }
 173                 /*
 174                  * Also if the number of blocks within the group is
 175                  * less than the blocksize * 8 ( which is the size
 176                  * of bitmap ), set rest of the block bitmap to 1
 177                  */
 178                 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
 179         }
 180         return free_blocks - ext4_group_used_meta_blocks(sb, block_group);
 181 }
 182
 183
 184 /*
 185  * The free blocks are managed by bitmaps.  A file system contains several
 186  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 187  * block for inodes, N blocks for the inode table and data blocks.
 188  *
 189  * The file system contains group descriptors which are located after the
 190  * super block.  Each descriptor contains the number of the bitmap block and
 191  * the free blocks count in the block.  The descriptors are loaded in memory
 192  * when a file system is mounted (see ext4_fill_super).
 193  */
 194
 195
 196 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
 197
 198 /**
 199  * ext4_get_group_desc() -- load group descriptor from disk
 200  * @sb:                 super block
 201  * @block_group:        given block group
 202  * @bh:                 pointer to the buffer head to store the block
 203  *                      group descriptor
 204  */
 205 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 206                                              ext4_group_t block_group,
 207                                              struct buffer_head **bh)
 208 {
 209         unsigned int group_desc;
 210         unsigned int offset;
 211         struct ext4_group_desc *desc;
 212         struct ext4_sb_info *sbi = EXT4_SB(sb);
 213
 214         if (block_group >= sbi->s_groups_count) {
 215                 ext4_error(sb, "ext4_get_group_desc",
 216                            "block_group >= groups_count - "
 217                            "block_group = %u, groups_count = %u",
 218                            block_group, sbi->s_groups_count);
 219
 220                 return NULL;
 221         }
 222         smp_rmb();
 223
 224         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 225         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 226         if (!sbi->s_group_desc[group_desc]) {
 227                 ext4_error(sb, "ext4_get_group_desc",
 228                            "Group descriptor not loaded - "
 229                            "block_group = %u, group_desc = %u, desc = %u",
 230                            block_group, group_desc, offset);
 231                 return NULL;
 232         }
 233
 234         desc = (struct ext4_group_desc *)(
 235                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
 236                 offset * EXT4_DESC_SIZE(sb));
 237         if (bh)
 238                 *bh = sbi->s_group_desc[group_desc];
 239         return desc;
 240 }
 241
 242 static int ext4_valid_block_bitmap(struct super_block *sb,
 243                                         struct ext4_group_desc *desc,
 244                                         unsigned int block_group,
 245                                         struct buffer_head *bh)
 246 {
 247         ext4_grpblk_t offset;
 248         ext4_grpblk_t next_zero_bit;
 249         ext4_fsblk_t bitmap_blk;
 250         ext4_fsblk_t group_first_block;
 251
 252         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
 253                 /* with FLEX_BG, the inode/block bitmaps and itable
 254                  * blocks may not be in the group at all
 255                  * so the bitmap validation will be skipped for those groups
 256                  * or it has to also read the block group where the bitmaps
 257                  * are located to verify they are set.
 258                  */
 259                 return 1;
 260         }
 261         group_first_block = ext4_group_first_block_no(sb, block_group);
 262
 263         /* check whether block bitmap block number is set */
 264         bitmap_blk = ext4_block_bitmap(sb, desc);
 265         offset = bitmap_blk - group_first_block;
 266         if (!ext4_test_bit(offset, bh->b_data))
 267                 /* bad block bitmap */
 268                 goto err_out;
 269
 270         /* check whether the inode bitmap block number is set */
 271         bitmap_blk = ext4_inode_bitmap(sb, desc);
 272         offset = bitmap_blk - group_first_block;
 273         if (!ext4_test_bit(offset, bh->b_data))
 274                 /* bad block bitmap */
 275                 goto err_out;
 276
 277         /* check whether the inode table block number is set */
 278         bitmap_blk = ext4_inode_table(sb, desc);
 279         offset = bitmap_blk - group_first_block;
 280         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 281                                 offset + EXT4_SB(sb)->s_itb_per_group,
 282                                 offset);
 283         if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
 284                 /* good bitmap for inode tables */
 285                 return 1;
 286
 287 err_out:
 288         ext4_error(sb, __func__,
 289                         "Invalid block bitmap - "
 290                         "block_group = %d, block = %llu",
 291                         block_group, bitmap_blk);
 292         return 0;
 293 }
 294 /**
 295  * ext4_read_block_bitmap()
 296  * @sb:                 super block
 297  * @block_group:        given block group
 298  *
 299  * Read the bitmap for a given block_group,and validate the
 300  * bits for block/inode/inode tables are set in the bitmaps
 301  *
 302  * Return buffer_head on success or NULL in case of failure.
 303  */
 304 struct buffer_head *
 305 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 306 {
 307         struct ext4_group_desc *desc;
 308         struct buffer_head *bh = NULL;
 309         ext4_fsblk_t bitmap_blk;
 310
 311         desc = ext4_get_group_desc(sb, block_group, NULL);
 312         if (!desc)
 313                 return NULL;
 314         bitmap_blk = ext4_block_bitmap(sb, desc);
 315         bh = sb_getblk(sb, bitmap_blk);
 316         if (unlikely(!bh)) {
 317                 ext4_error(sb, __func__,
 318                             "Cannot read block bitmap - "
 319                             "block_group = %u, block_bitmap = %llu",
 320                             block_group, bitmap_blk);
 321                 return NULL;
 322         }
 323
 324         if (bitmap_uptodate(bh))
 325                 return bh;
 326
 327         lock_buffer(bh);
 328         if (bitmap_uptodate(bh)) {
 329                 unlock_buffer(bh);
 330                 return bh;
 331         }
 332         spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
 333         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 334                 ext4_init_block_bitmap(sb, bh, block_group, desc);
 335                 set_bitmap_uptodate(bh);
 336                 set_buffer_uptodate(bh);
 337                 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 338                 unlock_buffer(bh);
 339                 return bh;
 340         }
 341         spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 342         if (buffer_uptodate(bh)) {
 343                 /*
 344                  * if not uninit if bh is uptodate,
 345                  * bitmap is also uptodate
 346                  */
 347                 set_bitmap_uptodate(bh);
 348                 unlock_buffer(bh);
 349                 return bh;
 350         }
 351         /*
 352          * submit the buffer_head for read. We can
 353          * safely mark the bitmap as uptodate now.
 354          * We do it here so the bitmap uptodate bit
 355          * get set with buffer lock held.
 356          */
 357         set_bitmap_uptodate(bh);
 358         if (bh_submit_read(bh) < 0) {
 359                 put_bh(bh);
 360                 ext4_error(sb, __func__,
 361                             "Cannot read block bitmap - "
 362                             "block_group = %u, block_bitmap = %llu",
 363                             block_group, bitmap_blk);
 364                 return NULL;
 365         }
 366         ext4_valid_block_bitmap(sb, desc, block_group, bh);
 367         /*
 368          * file system mounted not to panic on error,
 369          * continue with corrupt bitmap
 370          */
 371         return bh;
 372 }
 373
 374 /**
 375  * ext4_add_groupblocks() -- Add given blocks to an existing group
 376  * @handle:                     handle to this transaction
 377  * @sb:                         super block
 378  * @block:                      start physcial block to add to the block group
 379  * @count:                      number of blocks to free
 380  *
 381  * This marks the blocks as free in the bitmap. We ask the
 382  * mballoc to reload the buddy after this by setting group
 383  * EXT4_GROUP_INFO_NEED_INIT_BIT flag
 384  */
 385 void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
 386                          ext4_fsblk_t block, unsigned long count)
 387 {
 388         struct buffer_head *bitmap_bh = NULL;
 389         struct buffer_head *gd_bh;
 390         ext4_group_t block_group;
 391         ext4_grpblk_t bit;
 392         unsigned int i;
 393         struct ext4_group_desc *desc;
 394         struct ext4_super_block *es;
 395         struct ext4_sb_info *sbi;
 396         int err = 0, ret, blk_free_count;
 397         ext4_grpblk_t blocks_freed;
 398         struct ext4_group_info *grp;
 399
 400         sbi = EXT4_SB(sb);
 401         es = sbi->s_es;
 402         ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
 403
 404         ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
 405         grp = ext4_get_group_info(sb, block_group);
 406         /*
 407          * Check to see if we are freeing blocks across a group
 408          * boundary.
 409          */
 410         if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
 411                 goto error_return;
 412         }
 413         bitmap_bh = ext4_read_block_bitmap(sb, block_group);
 414         if (!bitmap_bh)
 415                 goto error_return;
 416         desc = ext4_get_group_desc(sb, block_group, &gd_bh);
 417         if (!desc)
 418                 goto error_return;
 419
 420         if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
 421             in_range(ext4_inode_bitmap(sb, desc), block, count) ||
 422             in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
 423             in_range(block + count - 1, ext4_inode_table(sb, desc),
 424                      sbi->s_itb_per_group)) {
 425                 ext4_error(sb, __func__,
 426                            "Adding blocks in system zones - "
 427                            "Block = %llu, count = %lu",
 428                            block, count);
 429                 goto error_return;
 430         }
 431
 432         /*
 433          * We are about to add blocks to the bitmap,
 434          * so we need undo access.
 435          */
 436         BUFFER_TRACE(bitmap_bh, "getting undo access");
 437         err = ext4_journal_get_undo_access(handle, bitmap_bh);
 438         if (err)
 439                 goto error_return;
 440
 441         /*
 442          * We are about to modify some metadata.  Call the journal APIs
 443          * to unshare ->b_data if a currently-committing transaction is
 444          * using it
 445          */
 446         BUFFER_TRACE(gd_bh, "get_write_access");
 447         err = ext4_journal_get_write_access(handle, gd_bh);
 448         if (err)
 449                 goto error_return;
 450         /*
 451          * make sure we don't allow a parallel init on other groups in the
 452          * same buddy cache
 453          */
 454         down_write(&grp->alloc_sem);
 455         for (i = 0, blocks_freed = 0; i < count; i++) {
 456                 BUFFER_TRACE(bitmap_bh, "clear bit");
 457                 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
 458                                                 bit + i, bitmap_bh->b_data)) {
 459                         ext4_error(sb, __func__,
 460                                    "bit already cleared for block %llu",
 461                                    (ext4_fsblk_t)(block + i));
 462                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
 463                 } else {
 464                         blocks_freed++;
 465                 }
 466         }
 467         spin_lock(sb_bgl_lock(sbi, block_group));
 468         blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
 469         ext4_free_blks_set(sb, desc, blk_free_count);
 470         desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
 471         spin_unlock(sb_bgl_lock(sbi, block_group));
 472         percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
 473
 474         if (sbi->s_log_groups_per_flex) {
 475                 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
 476                 spin_lock(sb_bgl_lock(sbi, flex_group));
 477                 sbi->s_flex_groups[flex_group].free_blocks += blocks_freed;
 478                 spin_unlock(sb_bgl_lock(sbi, flex_group));
 479         }
 480         /*
 481          * request to reload the buddy with the
 482          * new bitmap information
 483          */
 484         set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
 485         ext4_mb_update_group_info(grp, blocks_freed);
 486         up_write(&grp->alloc_sem);
 487
 488         /* We dirtied the bitmap block */
 489         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 490         err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 491
 492         /* And the group descriptor block */
 493         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
 494         ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
 495         if (!err)
 496                 err = ret;
 497         sb->s_dirt = 1;
 498
 499 error_return:
 500         brelse(bitmap_bh);
 501         ext4_std_error(sb, err);
 502         return;
 503 }
 504
 505 /**
 506  * ext4_free_blocks() -- Free given blocks and update quota
 507  * @handle:             handle for this transaction
 508  * @inode:              inode
 509  * @block:              start physical block to free
 510  * @count:              number of blocks to count
 511  * @metadata:           Are these metadata blocks
 512  */
 513 void ext4_free_blocks(handle_t *handle, struct inode *inode,
 514                         ext4_fsblk_t block, unsigned long count,
 515                         int metadata)
 516 {
 517         struct super_block *sb;
 518         unsigned long dquot_freed_blocks;
 519
 520         /* this isn't the right place to decide whether block is metadata
 521          * inode.c/extents.c knows better, but for safety ... */
 522         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
 523                 metadata = 1;
 524
 525         /* We need to make sure we don't reuse
 526          * block released untill the transaction commit.
 527          * writeback mode have weak data consistency so
 528          * don't force data as metadata when freeing block
 529          * for writeback mode.
 530          */
 531         if (metadata == 0 && !ext4_should_writeback_data(inode))
 532                 metadata = 1;
 533
 534         sb = inode->i_sb;
 535
 536         ext4_mb_free_blocks(handle, inode, block, count,
 537                             metadata, &dquot_freed_blocks);
 538         if (dquot_freed_blocks)
 539                 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
 540         return;
 541 }
 542
 543 /**
 544  * ext4_has_free_blocks()
 545  * @sbi:        in-core super block structure.
 546  * @nblocks:    number of needed blocks
 547  *
 548  * Check if filesystem has nblocks free & available for allocation.
 549  * On success return 1, return 0 on failure.
 550  */
 551 int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
 552 {
 553         s64 free_blocks, dirty_blocks, root_blocks;
 554         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
 555         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
 556
 557         free_blocks  = percpu_counter_read_positive(fbc);
 558         dirty_blocks = percpu_counter_read_positive(dbc);
 559         root_blocks = ext4_r_blocks_count(sbi->s_es);
 560
 561         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
 562                                                 EXT4_FREEBLOCKS_WATERMARK) {
 563                 free_blocks  = percpu_counter_sum_positive(fbc);
 564                 dirty_blocks = percpu_counter_sum_positive(dbc);
 565                 if (dirty_blocks < 0) {
 566                         printk(KERN_CRIT "Dirty block accounting "
 567                                         "went wrong %lld\n",
 568                                         (long long)dirty_blocks);
 569                 }
 570         }
 571         /* Check whether we have space after
 572          * accounting for current dirty blocks & root reserved blocks.
 573          */
 574         if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
 575                 return 1;
 576
 577         /* Hm, nope.  Are (enough) root reserved blocks available? */
 578         if (sbi->s_resuid == current_fsuid() ||
 579             ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
 580             capable(CAP_SYS_RESOURCE)) {
 581                 if (free_blocks >= (nblocks + dirty_blocks))
 582                         return 1;
 583         }
 584
 585         return 0;
 586 }
 587
 588 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
 589                                                 s64 nblocks)
 590 {
 591         if (ext4_has_free_blocks(sbi, nblocks)) {
 592                 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
 593                 return 0;
 594         } else
 595                 return -ENOSPC;
 596 }
 597
 598 /**
 599  * ext4_should_retry_alloc()
 600  * @sb:                 super block
 601  * @retries             number of attemps has been made
 602  *
 603  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 604  * it is profitable to retry the operation, this function will wait
 605  * for the current or commiting transaction to complete, and then
 606  * return TRUE.
 607  *
 608  * if the total number of retries exceed three times, return FALSE.
 609  */
 610 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 611 {
 612         if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
 613             (*retries)++ > 3 ||
 614             !EXT4_SB(sb)->s_journal)
 615                 return 0;
 616
 617         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 618
 619         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 620 }
 621
 622 /*
 623  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 624  *
 625  * @handle:             handle to this transaction
 626  * @inode:              file inode
 627  * @goal:               given target block(filesystem wide)
 628  * @count:              pointer to total number of blocks needed
 629  * @errp:               error code
 630  *
 631  * Return 1st allocated block number on success, *count stores total account
 632  * error stores in errp pointer
 633  */
 634 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 635                 ext4_fsblk_t goal, unsigned long *count, int *errp)
 636 {
 637         struct ext4_allocation_request ar;
 638         ext4_fsblk_t ret;
 639
 640         memset(&ar, 0, sizeof(ar));
 641         /* Fill with neighbour allocated blocks */
 642         ar.inode = inode;
 643         ar.goal = goal;
 644         ar.len = count ? *count : 1;
 645
 646         ret = ext4_mb_new_blocks(handle, &ar, errp);
 647         if (count)
 648                 *count = ar.len;
 649
 650         /*
 651          * Account for the allocated meta blocks
 652          */
 653         if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
 654                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 655                 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
 656                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 657         }
 658         return ret;
 659 }
 660
 661 /**
 662  * ext4_count_free_blocks() -- count filesystem free blocks
 663  * @sb:         superblock
 664  *
 665  * Adds up the number of free blocks from each block group.
 666  */
 667 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
 668 {
 669         ext4_fsblk_t desc_count;
 670         struct ext4_group_desc *gdp;
 671         ext4_group_t i;
 672         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
 673 #ifdef EXT4FS_DEBUG
 674         struct ext4_super_block *es;
 675         ext4_fsblk_t bitmap_count;
 676         unsigned int x;
 677         struct buffer_head *bitmap_bh = NULL;
 678
 679         es = EXT4_SB(sb)->s_es;
 680         desc_count = 0;
 681         bitmap_count = 0;
 682         gdp = NULL;
 683
 684         smp_rmb();
 685         for (i = 0; i < ngroups; i++) {
 686                 gdp = ext4_get_group_desc(sb, i, NULL);
 687                 if (!gdp)
 688                         continue;
 689                 desc_count += ext4_free_blks_count(sb, gdp);
 690                 brelse(bitmap_bh);
 691                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 692                 if (bitmap_bh == NULL)
 693                         continue;
 694
 695                 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
 696                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 697                         i, ext4_free_blks_count(sb, gdp), x);
 698                 bitmap_count += x;
 699         }
 700         brelse(bitmap_bh);
 701         printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
 702                 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
 703                desc_count, bitmap_count);
 704         return bitmap_count;
 705 #else
 706         desc_count = 0;
 707         smp_rmb();
 708         for (i = 0; i < ngroups; i++) {
 709                 gdp = ext4_get_group_desc(sb, i, NULL);
 710                 if (!gdp)
 711                         continue;
 712                 desc_count += ext4_free_blks_count(sb, gdp);
 713         }
 714
 715         return desc_count;
 716 #endif
 717 }
 718
 719 static inline int test_root(ext4_group_t a, int b)
 720 {
 721         int num = b;
 722
 723         while (a > num)
 724                 num *= b;
 725         return num == a;
 726 }
 727
 728 static int ext4_group_sparse(ext4_group_t group)
 729 {
 730         if (group <= 1)
 731                 return 1;
 732         if (!(group & 1))
 733                 return 0;
 734         return (test_root(group, 7) || test_root(group, 5) ||
 735                 test_root(group, 3));
 736 }
 737
 738 /**
 739  *      ext4_bg_has_super - number of blocks used by the superblock in group
 740  *      @sb: superblock for filesystem
 741  *      @group: group number to check
 742  *
 743  *      Return the number of blocks used by the superblock (primary or backup)
 744  *      in this group.  Currently this will be only 0 or 1.
 745  */
 746 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 747 {
 748         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
 749                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
 750                         !ext4_group_sparse(group))
 751                 return 0;
 752         return 1;
 753 }
 754
 755 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 756                                         ext4_group_t group)
 757 {
 758         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 759         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 760         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 761
 762         if (group == first || group == first + 1 || group == last)
 763                 return 1;
 764         return 0;
 765 }
 766
 767 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 768                                         ext4_group_t group)
 769 {
 770         return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
 771 }
 772
 773 /**
 774  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 775  *      @sb: superblock for filesystem
 776  *      @group: group number to check
 777  *
 778  *      Return the number of blocks used by the group descriptor table
 779  *      (primary or backup) in this group.  In the future there may be a
 780  *      different number of descriptor blocks in each group.
 781  */
 782 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 783 {
 784         unsigned long first_meta_bg =
 785                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 786         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 787
 788         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
 789                         metagroup < first_meta_bg)
 790                 return ext4_bg_num_gdb_nometa(sb, group);
 791
 792         return ext4_bg_num_gdb_meta(sb,group);
 793
 794 }
 795