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
  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 {
  59         ext4_fsblk_t tmp;
  60         struct ext4_sb_info *sbi = EXT4_SB(sb);
  61         /* block bitmap, inode bitmap, and inode table blocks */
  62         int used_blocks = sbi->s_itb_per_group + 2;
  63
  64         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
  65                 struct ext4_group_desc *gdp;
  66                 struct buffer_head *bh;
  67
  68                 gdp = ext4_get_group_desc(sb, block_group, &bh);
  69                 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
  70                                         block_group))
  71                         used_blocks--;
  72
  73                 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
  74                                         block_group))
  75                         used_blocks--;
  76
  77                 tmp = ext4_inode_table(sb, gdp);
  78                 for (; tmp < ext4_inode_table(sb, gdp) +
  79                                 sbi->s_itb_per_group; tmp++) {
  80                         if (!ext4_block_in_group(sb, tmp, block_group))
  81                                 used_blocks -= 1;
  82                 }
  83         }
  84         return used_blocks;
  85 }
  86
  87 /* Initializes an uninitialized block bitmap if given, and returns the
  88  * number of blocks free in the group. */
  89 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
  90                  ext4_group_t block_group, struct ext4_group_desc *gdp)
  91 {
  92         int bit, bit_max;
  93         unsigned free_blocks, group_blocks;
  94         struct ext4_sb_info *sbi = EXT4_SB(sb);
  95
  96         if (bh) {
  97                 J_ASSERT_BH(bh, buffer_locked(bh));
  98
  99                 /* If checksum is bad mark all blocks used to prevent allocation
 100                  * essentially implementing a per-group read-only flag. */
 101                 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
 102                         ext4_error(sb, __func__,
 103                                   "Checksum bad for group %lu\n", block_group);
 104                         gdp->bg_free_blocks_count = 0;
 105                         gdp->bg_free_inodes_count = 0;
 106                         gdp->bg_itable_unused = 0;
 107                         memset(bh->b_data, 0xff, sb->s_blocksize);
 108                         return 0;
 109                 }
 110                 memset(bh->b_data, 0, sb->s_blocksize);
 111         }
 112
 113         /* Check for superblock and gdt backups in this group */
 114         bit_max = ext4_bg_has_super(sb, block_group);
 115
 116         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
 117             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 118                           sbi->s_desc_per_block) {
 119                 if (bit_max) {
 120                         bit_max += ext4_bg_num_gdb(sb, block_group);
 121                         bit_max +=
 122                                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 123                 }
 124         } else { /* For META_BG_BLOCK_GROUPS */
 125                 bit_max += ext4_bg_num_gdb(sb, block_group);
 126         }
 127
 128         if (block_group == sbi->s_groups_count - 1) {
 129                 /*
 130                  * Even though mke2fs always initialize first and last group
 131                  * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
 132                  * to make sure we calculate the right free blocks
 133                  */
 134                 group_blocks = ext4_blocks_count(sbi->s_es) -
 135                         le32_to_cpu(sbi->s_es->s_first_data_block) -
 136                         (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count - 1));
 137         } else {
 138                 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
 139         }
 140
 141         free_blocks = group_blocks - bit_max;
 142
 143         if (bh) {
 144                 ext4_fsblk_t start, tmp;
 145                 int flex_bg = 0;
 146
 147                 for (bit = 0; bit < bit_max; bit++)
 148                         ext4_set_bit(bit, bh->b_data);
 149
 150                 start = ext4_group_first_block_no(sb, block_group);
 151
 152                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
 153                                               EXT4_FEATURE_INCOMPAT_FLEX_BG))
 154                         flex_bg = 1;
 155
 156                 /* Set bits for block and inode bitmaps, and inode table */
 157                 tmp = ext4_block_bitmap(sb, gdp);
 158                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 159                         ext4_set_bit(tmp - start, bh->b_data);
 160
 161                 tmp = ext4_inode_bitmap(sb, gdp);
 162                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 163                         ext4_set_bit(tmp - start, bh->b_data);
 164
 165                 tmp = ext4_inode_table(sb, gdp);
 166                 for (; tmp < ext4_inode_table(sb, gdp) +
 167                                 sbi->s_itb_per_group; tmp++) {
 168                         if (!flex_bg ||
 169                                 ext4_block_in_group(sb, tmp, block_group))
 170                                 ext4_set_bit(tmp - start, bh->b_data);
 171                 }
 172                 /*
 173                  * Also if the number of blocks within the group is
 174                  * less than the blocksize * 8 ( which is the size
 175                  * of bitmap ), set rest of the block bitmap to 1
 176                  */
 177                 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
 178         }
 179         return free_blocks - ext4_group_used_meta_blocks(sb, block_group);
 180 }
 181
 182
 183 /*
 184  * The free blocks are managed by bitmaps.  A file system contains several
 185  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 186  * block for inodes, N blocks for the inode table and data blocks.
 187  *
 188  * The file system contains group descriptors which are located after the
 189  * super block.  Each descriptor contains the number of the bitmap block and
 190  * the free blocks count in the block.  The descriptors are loaded in memory
 191  * when a file system is mounted (see ext4_fill_super).
 192  */
 193
 194
 195 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
 196
 197 /**
 198  * ext4_get_group_desc() -- load group descriptor from disk
 199  * @sb:                 super block
 200  * @block_group:        given block group
 201  * @bh:                 pointer to the buffer head to store the block
 202  *                      group descriptor
 203  */
 204 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 205                                              ext4_group_t block_group,
 206                                              struct buffer_head **bh)
 207 {
 208         unsigned long group_desc;
 209         unsigned long offset;
 210         struct ext4_group_desc *desc;
 211         struct ext4_sb_info *sbi = EXT4_SB(sb);
 212
 213         if (block_group >= sbi->s_groups_count) {
 214                 ext4_error(sb, "ext4_get_group_desc",
 215                            "block_group >= groups_count - "
 216                            "block_group = %lu, groups_count = %lu",
 217                            block_group, sbi->s_groups_count);
 218
 219                 return NULL;
 220         }
 221         smp_rmb();
 222
 223         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 224         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 225         if (!sbi->s_group_desc[group_desc]) {
 226                 ext4_error(sb, "ext4_get_group_desc",
 227                            "Group descriptor not loaded - "
 228                            "block_group = %lu, group_desc = %lu, desc = %lu",
 229                            block_group, group_desc, offset);
 230                 return NULL;
 231         }
 232
 233         desc = (struct ext4_group_desc *)(
 234                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
 235                 offset * EXT4_DESC_SIZE(sb));
 236         if (bh)
 237                 *bh = sbi->s_group_desc[group_desc];
 238         return desc;
 239 }
 240
 241 static int ext4_valid_block_bitmap(struct super_block *sb,
 242                                         struct ext4_group_desc *desc,
 243                                         unsigned int block_group,
 244                                         struct buffer_head *bh)
 245 {
 246         ext4_grpblk_t offset;
 247         ext4_grpblk_t next_zero_bit;
 248         ext4_fsblk_t bitmap_blk;
 249         ext4_fsblk_t group_first_block;
 250
 251         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
 252                 /* with FLEX_BG, the inode/block bitmaps and itable
 253                  * blocks may not be in the group at all
 254                  * so the bitmap validation will be skipped for those groups
 255                  * or it has to also read the block group where the bitmaps
 256                  * are located to verify they are set.
 257                  */
 258                 return 1;
 259         }
 260         group_first_block = ext4_group_first_block_no(sb, block_group);
 261
 262         /* check whether block bitmap block number is set */
 263         bitmap_blk = ext4_block_bitmap(sb, desc);
 264         offset = bitmap_blk - group_first_block;
 265         if (!ext4_test_bit(offset, bh->b_data))
 266                 /* bad block bitmap */
 267                 goto err_out;
 268
 269         /* check whether the inode bitmap block number is set */
 270         bitmap_blk = ext4_inode_bitmap(sb, desc);
 271         offset = bitmap_blk - group_first_block;
 272         if (!ext4_test_bit(offset, bh->b_data))
 273                 /* bad block bitmap */
 274                 goto err_out;
 275
 276         /* check whether the inode table block number is set */
 277         bitmap_blk = ext4_inode_table(sb, desc);
 278         offset = bitmap_blk - group_first_block;
 279         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 280                                 offset + EXT4_SB(sb)->s_itb_per_group,
 281                                 offset);
 282         if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
 283                 /* good bitmap for inode tables */
 284                 return 1;
 285
 286 err_out:
 287         ext4_error(sb, __func__,
 288                         "Invalid block bitmap - "
 289                         "block_group = %d, block = %llu",
 290                         block_group, bitmap_blk);
 291         return 0;
 292 }
 293 /**
 294  * ext4_read_block_bitmap()
 295  * @sb:                 super block
 296  * @block_group:        given block group
 297  *
 298  * Read the bitmap for a given block_group,and validate the
 299  * bits for block/inode/inode tables are set in the bitmaps
 300  *
 301  * Return buffer_head on success or NULL in case of failure.
 302  */
 303 struct buffer_head *
 304 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 305 {
 306         struct ext4_group_desc *desc;
 307         struct buffer_head *bh = NULL;
 308         ext4_fsblk_t bitmap_blk;
 309
 310         desc = ext4_get_group_desc(sb, block_group, NULL);
 311         if (!desc)
 312                 return NULL;
 313         bitmap_blk = ext4_block_bitmap(sb, desc);
 314         bh = sb_getblk(sb, bitmap_blk);
 315         if (unlikely(!bh)) {
 316                 ext4_error(sb, __func__,
 317                             "Cannot read block bitmap - "
 318                             "block_group = %lu, block_bitmap = %llu",
 319                             block_group, bitmap_blk);
 320                 return NULL;
 321         }
 322         if (bh_uptodate_or_lock(bh))
 323                 return bh;
 324
 325         spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
 326         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 327                 ext4_init_block_bitmap(sb, bh, block_group, desc);
 328                 set_buffer_uptodate(bh);
 329                 unlock_buffer(bh);
 330                 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 331                 return bh;
 332         }
 333         spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 334         if (bh_submit_read(bh) < 0) {
 335                 put_bh(bh);
 336                 ext4_error(sb, __func__,
 337                             "Cannot read block bitmap - "
 338                             "block_group = %lu, block_bitmap = %llu",
 339                             block_group, bitmap_blk);
 340                 return NULL;
 341         }
 342         ext4_valid_block_bitmap(sb, desc, block_group, bh);
 343         /*
 344          * file system mounted not to panic on error,
 345          * continue with corrupt bitmap
 346          */
 347         return bh;
 348 }
 349
 350 /**
 351  * ext4_free_blocks_sb() -- Free given blocks and update quota
 352  * @handle:                     handle to this transaction
 353  * @sb:                         super block
 354  * @block:                      start physcial block to free
 355  * @count:                      number of blocks to free
 356  * @pdquot_freed_blocks:        pointer to quota
 357  *
 358  * XXX This function is only used by the on-line resizing code, which
 359  * should probably be fixed up to call the mballoc variant.  There
 360  * this needs to be cleaned up later; in fact, I'm not convinced this
 361  * is 100% correct in the face of the mballoc code.  The online resizing
 362  * code needs to be fixed up to more tightly (and correctly) interlock
 363  * with the mballoc code.
 364  */
 365 void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb,
 366                          ext4_fsblk_t block, unsigned long count,
 367                          unsigned long *pdquot_freed_blocks)
 368 {
 369         struct buffer_head *bitmap_bh = NULL;
 370         struct buffer_head *gd_bh;
 371         ext4_group_t block_group;
 372         ext4_grpblk_t bit;
 373         unsigned long i;
 374         unsigned long overflow;
 375         struct ext4_group_desc *desc;
 376         struct ext4_super_block *es;
 377         struct ext4_sb_info *sbi;
 378         int err = 0, ret;
 379         ext4_grpblk_t group_freed;
 380
 381         *pdquot_freed_blocks = 0;
 382         sbi = EXT4_SB(sb);
 383         es = sbi->s_es;
 384         if (block < le32_to_cpu(es->s_first_data_block) ||
 385             block + count < block ||
 386             block + count > ext4_blocks_count(es)) {
 387                 ext4_error(sb, "ext4_free_blocks",
 388                            "Freeing blocks not in datazone - "
 389                            "block = %llu, count = %lu", block, count);
 390                 goto error_return;
 391         }
 392
 393         ext4_debug("freeing block(s) %llu-%llu\n", block, block + count - 1);
 394
 395 do_more:
 396         overflow = 0;
 397         ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
 398         /*
 399          * Check to see if we are freeing blocks across a group
 400          * boundary.
 401          */
 402         if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
 403                 overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb);
 404                 count -= overflow;
 405         }
 406         brelse(bitmap_bh);
 407         bitmap_bh = ext4_read_block_bitmap(sb, block_group);
 408         if (!bitmap_bh)
 409                 goto error_return;
 410         desc = ext4_get_group_desc(sb, block_group, &gd_bh);
 411         if (!desc)
 412                 goto error_return;
 413
 414         if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
 415             in_range(ext4_inode_bitmap(sb, desc), block, count) ||
 416             in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
 417             in_range(block + count - 1, ext4_inode_table(sb, desc),
 418                      sbi->s_itb_per_group)) {
 419                 ext4_error(sb, "ext4_free_blocks",
 420                            "Freeing blocks in system zones - "
 421                            "Block = %llu, count = %lu",
 422                            block, count);
 423                 goto error_return;
 424         }
 425
 426         /*
 427          * We are about to start releasing blocks in the bitmap,
 428          * so we need undo access.
 429          */
 430         /* @@@ check errors */
 431         BUFFER_TRACE(bitmap_bh, "getting undo access");
 432         err = ext4_journal_get_undo_access(handle, bitmap_bh);
 433         if (err)
 434                 goto error_return;
 435
 436         /*
 437          * We are about to modify some metadata.  Call the journal APIs
 438          * to unshare ->b_data if a currently-committing transaction is
 439          * using it
 440          */
 441         BUFFER_TRACE(gd_bh, "get_write_access");
 442         err = ext4_journal_get_write_access(handle, gd_bh);
 443         if (err)
 444                 goto error_return;
 445
 446         jbd_lock_bh_state(bitmap_bh);
 447
 448         for (i = 0, group_freed = 0; i < count; i++) {
 449                 /*
 450                  * An HJ special.  This is expensive...
 451                  */
 452 #ifdef CONFIG_JBD2_DEBUG
 453                 jbd_unlock_bh_state(bitmap_bh);
 454                 {
 455                         struct buffer_head *debug_bh;
 456                         debug_bh = sb_find_get_block(sb, block + i);
 457                         if (debug_bh) {
 458                                 BUFFER_TRACE(debug_bh, "Deleted!");
 459                                 if (!bh2jh(bitmap_bh)->b_committed_data)
 460                                         BUFFER_TRACE(debug_bh,
 461                                                 "No commited data in bitmap");
 462                                 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
 463                                 __brelse(debug_bh);
 464                         }
 465                 }
 466                 jbd_lock_bh_state(bitmap_bh);
 467 #endif
 468                 if (need_resched()) {
 469                         jbd_unlock_bh_state(bitmap_bh);
 470                         cond_resched();
 471                         jbd_lock_bh_state(bitmap_bh);
 472                 }
 473                 /* @@@ This prevents newly-allocated data from being
 474                  * freed and then reallocated within the same
 475                  * transaction.
 476                  *
 477                  * Ideally we would want to allow that to happen, but to
 478                  * do so requires making jbd2_journal_forget() capable of
 479                  * revoking the queued write of a data block, which
 480                  * implies blocking on the journal lock.  *forget()
 481                  * cannot block due to truncate races.
 482                  *
 483                  * Eventually we can fix this by making jbd2_journal_forget()
 484                  * return a status indicating whether or not it was able
 485                  * to revoke the buffer.  On successful revoke, it is
 486                  * safe not to set the allocation bit in the committed
 487                  * bitmap, because we know that there is no outstanding
 488                  * activity on the buffer any more and so it is safe to
 489                  * reallocate it.
 490                  */
 491                 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
 492                 J_ASSERT_BH(bitmap_bh,
 493                                 bh2jh(bitmap_bh)->b_committed_data != NULL);
 494                 ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
 495                                 bh2jh(bitmap_bh)->b_committed_data);
 496
 497                 /*
 498                  * We clear the bit in the bitmap after setting the committed
 499                  * data bit, because this is the reverse order to that which
 500                  * the allocator uses.
 501                  */
 502                 BUFFER_TRACE(bitmap_bh, "clear bit");
 503                 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
 504                                                 bit + i, bitmap_bh->b_data)) {
 505                         jbd_unlock_bh_state(bitmap_bh);
 506                         ext4_error(sb, __func__,
 507                                    "bit already cleared for block %llu",
 508                                    (ext4_fsblk_t)(block + i));
 509                         jbd_lock_bh_state(bitmap_bh);
 510                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
 511                 } else {
 512                         group_freed++;
 513                 }
 514         }
 515         jbd_unlock_bh_state(bitmap_bh);
 516
 517         spin_lock(sb_bgl_lock(sbi, block_group));
 518         le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
 519         desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
 520         spin_unlock(sb_bgl_lock(sbi, block_group));
 521         percpu_counter_add(&sbi->s_freeblocks_counter, count);
 522
 523         if (sbi->s_log_groups_per_flex) {
 524                 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
 525                 spin_lock(sb_bgl_lock(sbi, flex_group));
 526                 sbi->s_flex_groups[flex_group].free_blocks += count;
 527                 spin_unlock(sb_bgl_lock(sbi, flex_group));
 528         }
 529
 530         /* We dirtied the bitmap block */
 531         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 532         err = ext4_journal_dirty_metadata(handle, bitmap_bh);
 533
 534         /* And the group descriptor block */
 535         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
 536         ret = ext4_journal_dirty_metadata(handle, gd_bh);
 537         if (!err) err = ret;
 538         *pdquot_freed_blocks += group_freed;
 539
 540         if (overflow && !err) {
 541                 block += count;
 542                 count = overflow;
 543                 goto do_more;
 544         }
 545         sb->s_dirt = 1;
 546 error_return:
 547         brelse(bitmap_bh);
 548         ext4_std_error(sb, err);
 549         return;
 550 }
 551
 552 /**
 553  * ext4_free_blocks() -- Free given blocks and update quota
 554  * @handle:             handle for this transaction
 555  * @inode:              inode
 556  * @block:              start physical block to free
 557  * @count:              number of blocks to count
 558  * @metadata:           Are these metadata blocks
 559  */
 560 void ext4_free_blocks(handle_t *handle, struct inode *inode,
 561                         ext4_fsblk_t block, unsigned long count,
 562                         int metadata)
 563 {
 564         struct super_block *sb;
 565         unsigned long dquot_freed_blocks;
 566
 567         /* this isn't the right place to decide whether block is metadata
 568          * inode.c/extents.c knows better, but for safety ... */
 569         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
 570                         ext4_should_journal_data(inode))
 571                 metadata = 1;
 572
 573         sb = inode->i_sb;
 574
 575         ext4_mb_free_blocks(handle, inode, block, count,
 576                             metadata, &dquot_freed_blocks);
 577         if (dquot_freed_blocks)
 578                 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
 579         return;
 580 }
 581
 582 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
 583                                                 s64 nblocks)
 584 {
 585         s64 free_blocks, dirty_blocks;
 586         s64 root_blocks = 0;
 587         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
 588         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
 589
 590         free_blocks  = percpu_counter_read_positive(fbc);
 591         dirty_blocks = percpu_counter_read_positive(dbc);
 592
 593         if (!capable(CAP_SYS_RESOURCE) &&
 594                 sbi->s_resuid != current->fsuid &&
 595                 (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
 596                 root_blocks = ext4_r_blocks_count(sbi->s_es);
 597
 598         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
 599                                                 EXT4_FREEBLOCKS_WATERMARK) {
 600                 free_blocks  = percpu_counter_sum(fbc);
 601                 dirty_blocks = percpu_counter_sum(dbc);
 602                 if (dirty_blocks < 0) {
 603                         printk(KERN_CRIT "Dirty block accounting "
 604                                         "went wrong %lld\n",
 605                                         dirty_blocks);
 606                 }
 607         }
 608         /* Check whether we have space after
 609          * accounting for current dirty blocks
 610          */
 611         if (free_blocks < ((root_blocks + nblocks) + dirty_blocks))
 612                 /* we don't have free space */
 613                 return -ENOSPC;
 614
 615         /* Add the blocks to nblocks */
 616         percpu_counter_add(dbc, nblocks);
 617         return 0;
 618 }
 619
 620 /**
 621  * ext4_has_free_blocks()
 622  * @sbi:        in-core super block structure.
 623  * @nblocks:    number of neeed blocks
 624  *
 625  * Check if filesystem has free blocks available for allocation.
 626  * Return the number of blocks avaible for allocation for this request
 627  * On success, return nblocks
 628  */
 629 ext4_fsblk_t ext4_has_free_blocks(struct ext4_sb_info *sbi,
 630                                                 s64 nblocks)
 631 {
 632         s64 free_blocks, dirty_blocks;
 633         s64 root_blocks = 0;
 634         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
 635         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
 636
 637         free_blocks  = percpu_counter_read_positive(fbc);
 638         dirty_blocks = percpu_counter_read_positive(dbc);
 639
 640         if (!capable(CAP_SYS_RESOURCE) &&
 641                 sbi->s_resuid != current->fsuid &&
 642                 (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
 643                 root_blocks = ext4_r_blocks_count(sbi->s_es);
 644
 645         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
 646                                                 EXT4_FREEBLOCKS_WATERMARK) {
 647                 free_blocks  = percpu_counter_sum(fbc);
 648                 dirty_blocks = percpu_counter_sum(dbc);
 649         }
 650         if (free_blocks <= (root_blocks + dirty_blocks))
 651                 /* we don't have free space */
 652                 return 0;
 653
 654         if (free_blocks - (root_blocks + dirty_blocks) < nblocks)
 655                 return free_blocks - (root_blocks + dirty_blocks);
 656         return nblocks;
 657 }
 658
 659
 660 /**
 661  * ext4_should_retry_alloc()
 662  * @sb:                 super block
 663  * @retries             number of attemps has been made
 664  *
 665  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 666  * it is profitable to retry the operation, this function will wait
 667  * for the current or commiting transaction to complete, and then
 668  * return TRUE.
 669  *
 670  * if the total number of retries exceed three times, return FALSE.
 671  */
 672 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 673 {
 674         if (!ext4_has_free_blocks(EXT4_SB(sb), 1) || (*retries)++ > 3)
 675                 return 0;
 676
 677         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 678
 679         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 680 }
 681
 682 #define EXT4_META_BLOCK 0x1
 683
 684 static ext4_fsblk_t do_blk_alloc(handle_t *handle, struct inode *inode,
 685                                 ext4_lblk_t iblock, ext4_fsblk_t goal,
 686                                 unsigned long *count, int *errp, int flags)
 687 {
 688         struct ext4_allocation_request ar;
 689         ext4_fsblk_t ret;
 690
 691         memset(&ar, 0, sizeof(ar));
 692         /* Fill with neighbour allocated blocks */
 693
 694         ar.inode = inode;
 695         ar.goal = goal;
 696         ar.len = *count;
 697         ar.logical = iblock;
 698
 699         if (S_ISREG(inode->i_mode) && !(flags & EXT4_META_BLOCK))
 700                 /* enable in-core preallocation for data block allocation */
 701                 ar.flags = EXT4_MB_HINT_DATA;
 702         else
 703                 /* disable in-core preallocation for non-regular files */
 704                 ar.flags = 0;
 705
 706         ret = ext4_mb_new_blocks(handle, &ar, errp);
 707         *count = ar.len;
 708         return ret;
 709 }
 710
 711 /*
 712  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 713  *
 714  * @handle:             handle to this transaction
 715  * @inode:              file inode
 716  * @goal:               given target block(filesystem wide)
 717  * @count:              total number of blocks need
 718  * @errp:               error code
 719  *
 720  * Return 1st allocated block numberon success, *count stores total account
 721  * error stores in errp pointer
 722  */
 723 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 724                 ext4_fsblk_t goal, unsigned long *count, int *errp)
 725 {
 726         ext4_fsblk_t ret;
 727         ret = do_blk_alloc(handle, inode, 0, goal,
 728                                 count, errp, EXT4_META_BLOCK);
 729         /*
 730          * Account for the allocated meta blocks
 731          */
 732         if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
 733                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 734                 EXT4_I(inode)->i_allocated_meta_blocks += *count;
 735                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 736         }
 737         return ret;
 738 }
 739
 740 /*
 741  * ext4_new_meta_block() -- allocate block for meta data (indexing) blocks
 742  *
 743  * @handle:             handle to this transaction
 744  * @inode:              file inode
 745  * @goal:               given target block(filesystem wide)
 746  * @errp:               error code
 747  *
 748  * Return allocated block number on success
 749  */
 750 ext4_fsblk_t ext4_new_meta_block(handle_t *handle, struct inode *inode,
 751                 ext4_fsblk_t goal, int *errp)
 752 {
 753         unsigned long count = 1;
 754         return ext4_new_meta_blocks(handle, inode, goal, &count, errp);
 755 }
 756
 757 /*
 758  * ext4_new_blocks() -- allocate data blocks
 759  *
 760  * @handle:             handle to this transaction
 761  * @inode:              file inode
 762  * @goal:               given target block(filesystem wide)
 763  * @count:              total number of blocks need
 764  * @errp:               error code
 765  *
 766  * Return 1st allocated block numberon success, *count stores total account
 767  * error stores in errp pointer
 768  */
 769
 770 ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode,
 771                                 ext4_lblk_t iblock, ext4_fsblk_t goal,
 772                                 unsigned long *count, int *errp)
 773 {
 774         return do_blk_alloc(handle, inode, iblock, goal, count, errp, 0);
 775 }
 776
 777 /**
 778  * ext4_count_free_blocks() -- count filesystem free blocks
 779  * @sb:         superblock
 780  *
 781  * Adds up the number of free blocks from each block group.
 782  */
 783 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
 784 {
 785         ext4_fsblk_t desc_count;
 786         struct ext4_group_desc *gdp;
 787         ext4_group_t i;
 788         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
 789 #ifdef EXT4FS_DEBUG
 790         struct ext4_super_block *es;
 791         ext4_fsblk_t bitmap_count;
 792         unsigned long x;
 793         struct buffer_head *bitmap_bh = NULL;
 794
 795         es = EXT4_SB(sb)->s_es;
 796         desc_count = 0;
 797         bitmap_count = 0;
 798         gdp = NULL;
 799
 800         smp_rmb();
 801         for (i = 0; i < ngroups; i++) {
 802                 gdp = ext4_get_group_desc(sb, i, NULL);
 803                 if (!gdp)
 804                         continue;
 805                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
 806                 brelse(bitmap_bh);
 807                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 808                 if (bitmap_bh == NULL)
 809                         continue;
 810
 811                 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
 812                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
 813                         i, le16_to_cpu(gdp->bg_free_blocks_count), x);
 814                 bitmap_count += x;
 815         }
 816         brelse(bitmap_bh);
 817         printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
 818                 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
 819                desc_count, bitmap_count);
 820         return bitmap_count;
 821 #else
 822         desc_count = 0;
 823         smp_rmb();
 824         for (i = 0; i < ngroups; i++) {
 825                 gdp = ext4_get_group_desc(sb, i, NULL);
 826                 if (!gdp)
 827                         continue;
 828                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
 829         }
 830
 831         return desc_count;
 832 #endif
 833 }
 834
 835 static inline int test_root(ext4_group_t a, int b)
 836 {
 837         int num = b;
 838
 839         while (a > num)
 840                 num *= b;
 841         return num == a;
 842 }
 843
 844 static int ext4_group_sparse(ext4_group_t group)
 845 {
 846         if (group <= 1)
 847                 return 1;
 848         if (!(group & 1))
 849                 return 0;
 850         return (test_root(group, 7) || test_root(group, 5) ||
 851                 test_root(group, 3));
 852 }
 853
 854 /**
 855  *      ext4_bg_has_super - number of blocks used by the superblock in group
 856  *      @sb: superblock for filesystem
 857  *      @group: group number to check
 858  *
 859  *      Return the number of blocks used by the superblock (primary or backup)
 860  *      in this group.  Currently this will be only 0 or 1.
 861  */
 862 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 863 {
 864         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
 865                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
 866                         !ext4_group_sparse(group))
 867                 return 0;
 868         return 1;
 869 }
 870
 871 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 872                                         ext4_group_t group)
 873 {
 874         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 875         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 876         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 877
 878         if (group == first || group == first + 1 || group == last)
 879                 return 1;
 880         return 0;
 881 }
 882
 883 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 884                                         ext4_group_t group)
 885 {
 886         return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
 887 }
 888
 889 /**
 890  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 891  *      @sb: superblock for filesystem
 892  *      @group: group number to check
 893  *
 894  *      Return the number of blocks used by the group descriptor table
 895  *      (primary or backup) in this group.  In the future there may be a
 896  *      different number of descriptor blocks in each group.
 897  */
 898 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 899 {
 900         unsigned long first_meta_bg =
 901                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 902         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 903
 904         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
 905                         metagroup < first_meta_bg)
 906                 return ext4_bg_num_gdb_nometa(sb, group);
 907
 908         return ext4_bg_num_gdb_meta(sb,group);
 909
 910 }
 911