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         if (buffer_uptodate(bh) &&
 324             !(desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)))
 325                 return bh;
 326
 327         lock_buffer(bh);
 328         spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
 329         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 330                 ext4_init_block_bitmap(sb, bh, block_group, desc);
 331                 set_buffer_uptodate(bh);
 332                 unlock_buffer(bh);
 333                 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 334                 return bh;
 335         }
 336         spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 337         if (bh_submit_read(bh) < 0) {
 338                 put_bh(bh);
 339                 ext4_error(sb, __func__,
 340                             "Cannot read block bitmap - "
 341                             "block_group = %u, block_bitmap = %llu",
 342                             block_group, bitmap_blk);
 343                 return NULL;
 344         }
 345         ext4_valid_block_bitmap(sb, desc, block_group, bh);
 346         /*
 347          * file system mounted not to panic on error,
 348          * continue with corrupt bitmap
 349          */
 350         return bh;
 351 }
 352
 353 /**
 354  * ext4_add_groupblocks() -- Add given blocks to an existing group
 355  * @handle:                     handle to this transaction
 356  * @sb:                         super block
 357  * @block:                      start physcial block to add to the block group
 358  * @count:                      number of blocks to free
 359  *
 360  * This marks the blocks as free in the bitmap. We ask the
 361  * mballoc to reload the buddy after this by setting group
 362  * EXT4_GROUP_INFO_NEED_INIT_BIT flag
 363  */
 364 void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
 365                          ext4_fsblk_t block, unsigned long count)
 366 {
 367         struct buffer_head *bitmap_bh = NULL;
 368         struct buffer_head *gd_bh;
 369         ext4_group_t block_group;
 370         ext4_grpblk_t bit;
 371         unsigned int i;
 372         struct ext4_group_desc *desc;
 373         struct ext4_super_block *es;
 374         struct ext4_sb_info *sbi;
 375         int err = 0, ret, blk_free_count;
 376         ext4_grpblk_t blocks_freed;
 377         struct ext4_group_info *grp;
 378
 379         sbi = EXT4_SB(sb);
 380         es = sbi->s_es;
 381         ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
 382
 383         ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
 384         grp = ext4_get_group_info(sb, block_group);
 385         /*
 386          * Check to see if we are freeing blocks across a group
 387          * boundary.
 388          */
 389         if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
 390                 goto error_return;
 391         }
 392         bitmap_bh = ext4_read_block_bitmap(sb, block_group);
 393         if (!bitmap_bh)
 394                 goto error_return;
 395         desc = ext4_get_group_desc(sb, block_group, &gd_bh);
 396         if (!desc)
 397                 goto error_return;
 398
 399         if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
 400             in_range(ext4_inode_bitmap(sb, desc), block, count) ||
 401             in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
 402             in_range(block + count - 1, ext4_inode_table(sb, desc),
 403                      sbi->s_itb_per_group)) {
 404                 ext4_error(sb, __func__,
 405                            "Adding blocks in system zones - "
 406                            "Block = %llu, count = %lu",
 407                            block, count);
 408                 goto error_return;
 409         }
 410
 411         /*
 412          * We are about to add blocks to the bitmap,
 413          * so we need undo access.
 414          */
 415         BUFFER_TRACE(bitmap_bh, "getting undo access");
 416         err = ext4_journal_get_undo_access(handle, bitmap_bh);
 417         if (err)
 418                 goto error_return;
 419
 420         /*
 421          * We are about to modify some metadata.  Call the journal APIs
 422          * to unshare ->b_data if a currently-committing transaction is
 423          * using it
 424          */
 425         BUFFER_TRACE(gd_bh, "get_write_access");
 426         err = ext4_journal_get_write_access(handle, gd_bh);
 427         if (err)
 428                 goto error_return;
 429         /*
 430          * make sure we don't allow a parallel init on other groups in the
 431          * same buddy cache
 432          */
 433         down_write(&grp->alloc_sem);
 434         for (i = 0, blocks_freed = 0; i < count; i++) {
 435                 BUFFER_TRACE(bitmap_bh, "clear bit");
 436                 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
 437                                                 bit + i, bitmap_bh->b_data)) {
 438                         ext4_error(sb, __func__,
 439                                    "bit already cleared for block %llu",
 440                                    (ext4_fsblk_t)(block + i));
 441                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
 442                 } else {
 443                         blocks_freed++;
 444                 }
 445         }
 446         spin_lock(sb_bgl_lock(sbi, block_group));
 447         blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
 448         ext4_free_blks_set(sb, desc, blk_free_count);
 449         desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
 450         spin_unlock(sb_bgl_lock(sbi, block_group));
 451         percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
 452
 453         if (sbi->s_log_groups_per_flex) {
 454                 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
 455                 spin_lock(sb_bgl_lock(sbi, flex_group));
 456                 sbi->s_flex_groups[flex_group].free_blocks += blocks_freed;
 457                 spin_unlock(sb_bgl_lock(sbi, flex_group));
 458         }
 459         /*
 460          * request to reload the buddy with the
 461          * new bitmap information
 462          */
 463         set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
 464         ext4_mb_update_group_info(grp, blocks_freed);
 465         up_write(&grp->alloc_sem);
 466
 467         /* We dirtied the bitmap block */
 468         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 469         err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 470
 471         /* And the group descriptor block */
 472         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
 473         ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
 474         if (!err)
 475                 err = ret;
 476         sb->s_dirt = 1;
 477
 478 error_return:
 479         brelse(bitmap_bh);
 480         ext4_std_error(sb, err);
 481         return;
 482 }
 483
 484 /**
 485  * ext4_free_blocks() -- Free given blocks and update quota
 486  * @handle:             handle for this transaction
 487  * @inode:              inode
 488  * @block:              start physical block to free
 489  * @count:              number of blocks to count
 490  * @metadata:           Are these metadata blocks
 491  */
 492 void ext4_free_blocks(handle_t *handle, struct inode *inode,
 493                         ext4_fsblk_t block, unsigned long count,
 494                         int metadata)
 495 {
 496         struct super_block *sb;
 497         unsigned long dquot_freed_blocks;
 498
 499         /* this isn't the right place to decide whether block is metadata
 500          * inode.c/extents.c knows better, but for safety ... */
 501         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
 502                 metadata = 1;
 503
 504         /* We need to make sure we don't reuse
 505          * block released untill the transaction commit.
 506          * writeback mode have weak data consistency so
 507          * don't force data as metadata when freeing block
 508          * for writeback mode.
 509          */
 510         if (metadata == 0 && !ext4_should_writeback_data(inode))
 511                 metadata = 1;
 512
 513         sb = inode->i_sb;
 514
 515         ext4_mb_free_blocks(handle, inode, block, count,
 516                             metadata, &dquot_freed_blocks);
 517         if (dquot_freed_blocks)
 518                 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
 519         return;
 520 }
 521
 522 /**
 523  * ext4_has_free_blocks()
 524  * @sbi:        in-core super block structure.
 525  * @nblocks:    number of needed blocks
 526  *
 527  * Check if filesystem has nblocks free & available for allocation.
 528  * On success return 1, return 0 on failure.
 529  */
 530 int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
 531 {
 532         s64 free_blocks, dirty_blocks, root_blocks;
 533         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
 534         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
 535
 536         free_blocks  = percpu_counter_read_positive(fbc);
 537         dirty_blocks = percpu_counter_read_positive(dbc);
 538         root_blocks = ext4_r_blocks_count(sbi->s_es);
 539
 540         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
 541                                                 EXT4_FREEBLOCKS_WATERMARK) {
 542                 free_blocks  = percpu_counter_sum_positive(fbc);
 543                 dirty_blocks = percpu_counter_sum_positive(dbc);
 544                 if (dirty_blocks < 0) {
 545                         printk(KERN_CRIT "Dirty block accounting "
 546                                         "went wrong %lld\n",
 547                                         (long long)dirty_blocks);
 548                 }
 549         }
 550         /* Check whether we have space after
 551          * accounting for current dirty blocks & root reserved blocks.
 552          */
 553         if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
 554                 return 1;
 555
 556         /* Hm, nope.  Are (enough) root reserved blocks available? */
 557         if (sbi->s_resuid == current_fsuid() ||
 558             ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
 559             capable(CAP_SYS_RESOURCE)) {
 560                 if (free_blocks >= (nblocks + dirty_blocks))
 561                         return 1;
 562         }
 563
 564         return 0;
 565 }
 566
 567 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
 568                                                 s64 nblocks)
 569 {
 570         if (ext4_has_free_blocks(sbi, nblocks)) {
 571                 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
 572                 return 0;
 573         } else
 574                 return -ENOSPC;
 575 }
 576
 577 /**
 578  * ext4_should_retry_alloc()
 579  * @sb:                 super block
 580  * @retries             number of attemps has been made
 581  *
 582  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 583  * it is profitable to retry the operation, this function will wait
 584  * for the current or commiting transaction to complete, and then
 585  * return TRUE.
 586  *
 587  * if the total number of retries exceed three times, return FALSE.
 588  */
 589 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 590 {
 591         if (!ext4_has_free_blocks(EXT4_SB(sb), 1) || (*retries)++ > 3)
 592                 return 0;
 593
 594         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 595
 596         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 597 }
 598
 599 /*
 600  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 601  *
 602  * @handle:             handle to this transaction
 603  * @inode:              file inode
 604  * @goal:               given target block(filesystem wide)
 605  * @count:              pointer to total number of blocks needed
 606  * @errp:               error code
 607  *
 608  * Return 1st allocated block number on success, *count stores total account
 609  * error stores in errp pointer
 610  */
 611 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 612                 ext4_fsblk_t goal, unsigned long *count, int *errp)
 613 {
 614         struct ext4_allocation_request ar;
 615         ext4_fsblk_t ret;
 616
 617         memset(&ar, 0, sizeof(ar));
 618         /* Fill with neighbour allocated blocks */
 619         ar.inode = inode;
 620         ar.goal = goal;
 621         ar.len = count ? *count : 1;
 622
 623         ret = ext4_mb_new_blocks(handle, &ar, errp);
 624         if (count)
 625                 *count = ar.len;
 626
 627         /*
 628          * Account for the allocated meta blocks
 629          */
 630         if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
 631                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 632                 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
 633                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 634         }
 635         return ret;
 636 }
 637
 638 /**
 639  * ext4_count_free_blocks() -- count filesystem free blocks
 640  * @sb:         superblock
 641  *
 642  * Adds up the number of free blocks from each block group.
 643  */
 644 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
 645 {
 646         ext4_fsblk_t desc_count;
 647         struct ext4_group_desc *gdp;
 648         ext4_group_t i;
 649         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
 650 #ifdef EXT4FS_DEBUG
 651         struct ext4_super_block *es;
 652         ext4_fsblk_t bitmap_count;
 653         unsigned int x;
 654         struct buffer_head *bitmap_bh = NULL;
 655
 656         es = EXT4_SB(sb)->s_es;
 657         desc_count = 0;
 658         bitmap_count = 0;
 659         gdp = NULL;
 660
 661         smp_rmb();
 662         for (i = 0; i < ngroups; i++) {
 663                 gdp = ext4_get_group_desc(sb, i, NULL);
 664                 if (!gdp)
 665                         continue;
 666                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
 667                 brelse(bitmap_bh);
 668                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 669                 if (bitmap_bh == NULL)
 670                         continue;
 671
 672                 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
 673                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %u\n",
 674                         i, le16_to_cpu(gdp->bg_free_blocks_count), x);
 675                 bitmap_count += x;
 676         }
 677         brelse(bitmap_bh);
 678         printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
 679                 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
 680                desc_count, bitmap_count);
 681         return bitmap_count;
 682 #else
 683         desc_count = 0;
 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         }
 691
 692         return desc_count;
 693 #endif
 694 }
 695
 696 static inline int test_root(ext4_group_t a, int b)
 697 {
 698         int num = b;
 699
 700         while (a > num)
 701                 num *= b;
 702         return num == a;
 703 }
 704
 705 static int ext4_group_sparse(ext4_group_t group)
 706 {
 707         if (group <= 1)
 708                 return 1;
 709         if (!(group & 1))
 710                 return 0;
 711         return (test_root(group, 7) || test_root(group, 5) ||
 712                 test_root(group, 3));
 713 }
 714
 715 /**
 716  *      ext4_bg_has_super - number of blocks used by the superblock in group
 717  *      @sb: superblock for filesystem
 718  *      @group: group number to check
 719  *
 720  *      Return the number of blocks used by the superblock (primary or backup)
 721  *      in this group.  Currently this will be only 0 or 1.
 722  */
 723 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 724 {
 725         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
 726                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
 727                         !ext4_group_sparse(group))
 728                 return 0;
 729         return 1;
 730 }
 731
 732 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 733                                         ext4_group_t group)
 734 {
 735         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 736         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 737         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 738
 739         if (group == first || group == first + 1 || group == last)
 740                 return 1;
 741         return 0;
 742 }
 743
 744 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 745                                         ext4_group_t group)
 746 {
 747         return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
 748 }
 749
 750 /**
 751  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 752  *      @sb: superblock for filesystem
 753  *      @group: group number to check
 754  *
 755  *      Return the number of blocks used by the group descriptor table
 756  *      (primary or backup) in this group.  In the future there may be a
 757  *      different number of descriptor blocks in each group.
 758  */
 759 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 760 {
 761         unsigned long first_meta_bg =
 762                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 763         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 764
 765         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
 766                         metagroup < first_meta_bg)
 767                 return ext4_bg_num_gdb_nometa(sb, group);
 768
 769         return ext4_bg_num_gdb_meta(sb,group);
 770
 771 }
 772