fs/ext4/balloc.c

   1 /*
   2  *  linux/fs/ext4/balloc.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
  10  *  Big-endian to little-endian byte-swapping/bitmaps by
  11  *        David S. Miller (davem@caip.rutgers.edu), 1995
  12  */
  13
  14 #include <linux/time.h>
  15 #include <linux/capability.h>
  16 #include <linux/fs.h>
  17 #include <linux/jbd2.h>
  18 #include <linux/quotaops.h>
  19 #include <linux/buffer_head.h>
  20 #include "ext4.h"
  21 #include "ext4_jbd2.h"
  22 #include "mballoc.h"
  23
  24 #include <trace/events/ext4.h>
  25
  26 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
  27                                             ext4_group_t block_group);
  28 /*
  29  * balloc.c contains the blocks allocation and deallocation routines
  30  */
  31
  32 /*
  33  * Calculate block group number for a given block number
  34  */
  35 ext4_group_t ext4_get_group_number(struct super_block *sb,
  36                                    ext4_fsblk_t block)
  37 {
  38         ext4_group_t group;
  39
  40         if (test_opt2(sb, STD_GROUP_SIZE))
  41                 group = (block -
  42                          le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
  43                         (EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
  44         else
  45                 ext4_get_group_no_and_offset(sb, block, &group, NULL);
  46         return group;
  47 }
  48
  49 /*
  50  * Calculate the block group number and offset into the block/cluster
  51  * allocation bitmap, given a block number
  52  */
  53 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
  54                 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
  55 {
  56         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
  57         ext4_grpblk_t offset;
  58
  59         blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
  60         offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
  61                 EXT4_SB(sb)->s_cluster_bits;
  62         if (offsetp)
  63                 *offsetp = offset;
  64         if (blockgrpp)
  65                 *blockgrpp = blocknr;
  66
  67 }
  68
  69 /*
  70  * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
  71  * and 0 otherwise.
  72  */
  73 static inline int ext4_block_in_group(struct super_block *sb,
  74                                       ext4_fsblk_t block,
  75                                       ext4_group_t block_group)
  76 {
  77         ext4_group_t actual_group;
  78
  79         actual_group = ext4_get_group_number(sb, block);
  80         return (actual_group == block_group) ? 1 : 0;
  81 }
  82
  83 /* Return the number of clusters used for file system metadata; this
  84  * represents the overhead needed by the file system.
  85  */
  86 unsigned ext4_num_overhead_clusters(struct super_block *sb,
  87                                     ext4_group_t block_group,
  88                                     struct ext4_group_desc *gdp)
  89 {
  90         unsigned num_clusters;
  91         int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
  92         ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
  93         ext4_fsblk_t itbl_blk;
  94         struct ext4_sb_info *sbi = EXT4_SB(sb);
  95
  96         /* This is the number of clusters used by the superblock,
  97          * block group descriptors, and reserved block group
  98          * descriptor blocks */
  99         num_clusters = ext4_num_base_meta_clusters(sb, block_group);
 100
 101         /*
 102          * For the allocation bitmaps and inode table, we first need
 103          * to check to see if the block is in the block group.  If it
 104          * is, then check to see if the cluster is already accounted
 105          * for in the clusters used for the base metadata cluster, or
 106          * if we can increment the base metadata cluster to include
 107          * that block.  Otherwise, we will have to track the cluster
 108          * used for the allocation bitmap or inode table explicitly.
 109          * Normally all of these blocks are contiguous, so the special
 110          * case handling shouldn't be necessary except for *very*
 111          * unusual file system layouts.
 112          */
 113         if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
 114                 block_cluster = EXT4_B2C(sbi,
 115                                          ext4_block_bitmap(sb, gdp) - start);
 116                 if (block_cluster < num_clusters)
 117                         block_cluster = -1;
 118                 else if (block_cluster == num_clusters) {
 119                         num_clusters++;
 120                         block_cluster = -1;
 121                 }
 122         }
 123
 124         if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
 125                 inode_cluster = EXT4_B2C(sbi,
 126                                          ext4_inode_bitmap(sb, gdp) - start);
 127                 if (inode_cluster < num_clusters)
 128                         inode_cluster = -1;
 129                 else if (inode_cluster == num_clusters) {
 130                         num_clusters++;
 131                         inode_cluster = -1;
 132                 }
 133         }
 134
 135         itbl_blk = ext4_inode_table(sb, gdp);
 136         for (i = 0; i < sbi->s_itb_per_group; i++) {
 137                 if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
 138                         c = EXT4_B2C(sbi, itbl_blk + i - start);
 139                         if ((c < num_clusters) || (c == inode_cluster) ||
 140                             (c == block_cluster) || (c == itbl_cluster))
 141                                 continue;
 142                         if (c == num_clusters) {
 143                                 num_clusters++;
 144                                 continue;
 145                         }
 146                         num_clusters++;
 147                         itbl_cluster = c;
 148                 }
 149         }
 150
 151         if (block_cluster != -1)
 152                 num_clusters++;
 153         if (inode_cluster != -1)
 154                 num_clusters++;
 155
 156         return num_clusters;
 157 }
 158
 159 static unsigned int num_clusters_in_group(struct super_block *sb,
 160                                           ext4_group_t block_group)
 161 {
 162         unsigned int blocks;
 163
 164         if (block_group == ext4_get_groups_count(sb) - 1) {
 165                 /*
 166                  * Even though mke2fs always initializes the first and
 167                  * last group, just in case some other tool was used,
 168                  * we need to make sure we calculate the right free
 169                  * blocks.
 170                  */
 171                 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
 172                         ext4_group_first_block_no(sb, block_group);
 173         } else
 174                 blocks = EXT4_BLOCKS_PER_GROUP(sb);
 175         return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
 176 }
 177
 178 /* Initializes an uninitialized block bitmap */
 179 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
 180                             ext4_group_t block_group,
 181                             struct ext4_group_desc *gdp)
 182 {
 183         unsigned int bit, bit_max;
 184         struct ext4_sb_info *sbi = EXT4_SB(sb);
 185         ext4_fsblk_t start, tmp;
 186         int flex_bg = 0;
 187
 188         J_ASSERT_BH(bh, buffer_locked(bh));
 189
 190         /* If checksum is bad mark all blocks used to prevent allocation
 191          * essentially implementing a per-group read-only flag. */
 192         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
 193                 ext4_error(sb, "Checksum bad for group %u", block_group);
 194                 ext4_free_group_clusters_set(sb, gdp, 0);
 195                 ext4_free_inodes_set(sb, gdp, 0);
 196                 ext4_itable_unused_set(sb, gdp, 0);
 197                 memset(bh->b_data, 0xff, sb->s_blocksize);
 198                 ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
 199                 return;
 200         }
 201         memset(bh->b_data, 0, sb->s_blocksize);
 202
 203         bit_max = ext4_num_base_meta_clusters(sb, block_group);
 204         for (bit = 0; bit < bit_max; bit++)
 205                 ext4_set_bit(bit, bh->b_data);
 206
 207         start = ext4_group_first_block_no(sb, block_group);
 208
 209         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
 210                 flex_bg = 1;
 211
 212         /* Set bits for block and inode bitmaps, and inode table */
 213         tmp = ext4_block_bitmap(sb, gdp);
 214         if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 215                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 216
 217         tmp = ext4_inode_bitmap(sb, gdp);
 218         if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 219                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 220
 221         tmp = ext4_inode_table(sb, gdp);
 222         for (; tmp < ext4_inode_table(sb, gdp) +
 223                      sbi->s_itb_per_group; tmp++) {
 224                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 225                         ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 226         }
 227
 228         /*
 229          * Also if the number of blocks within the group is less than
 230          * the blocksize * 8 ( which is the size of bitmap ), set rest
 231          * of the block bitmap to 1
 232          */
 233         ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
 234                              sb->s_blocksize * 8, bh->b_data);
 235         ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
 236         ext4_group_desc_csum_set(sb, block_group, gdp);
 237 }
 238
 239 /* Return the number of free blocks in a block group.  It is used when
 240  * the block bitmap is uninitialized, so we can't just count the bits
 241  * in the bitmap. */
 242 unsigned ext4_free_clusters_after_init(struct super_block *sb,
 243                                        ext4_group_t block_group,
 244                                        struct ext4_group_desc *gdp)
 245 {
 246         return num_clusters_in_group(sb, block_group) -
 247                 ext4_num_overhead_clusters(sb, block_group, gdp);
 248 }
 249
 250 /*
 251  * The free blocks are managed by bitmaps.  A file system contains several
 252  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 253  * block for inodes, N blocks for the inode table and data blocks.
 254  *
 255  * The file system contains group descriptors which are located after the
 256  * super block.  Each descriptor contains the number of the bitmap block and
 257  * the free blocks count in the block.  The descriptors are loaded in memory
 258  * when a file system is mounted (see ext4_fill_super).
 259  */
 260
 261 /**
 262  * ext4_get_group_desc() -- load group descriptor from disk
 263  * @sb:                 super block
 264  * @block_group:        given block group
 265  * @bh:                 pointer to the buffer head to store the block
 266  *                      group descriptor
 267  */
 268 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 269                                              ext4_group_t block_group,
 270                                              struct buffer_head **bh)
 271 {
 272         unsigned int group_desc;
 273         unsigned int offset;
 274         ext4_group_t ngroups = ext4_get_groups_count(sb);
 275         struct ext4_group_desc *desc;
 276         struct ext4_sb_info *sbi = EXT4_SB(sb);
 277
 278         if (block_group >= ngroups) {
 279                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
 280                            " groups_count = %u", block_group, ngroups);
 281
 282                 return NULL;
 283         }
 284
 285         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 286         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 287         if (!sbi->s_group_desc[group_desc]) {
 288                 ext4_error(sb, "Group descriptor not loaded - "
 289                            "block_group = %u, group_desc = %u, desc = %u",
 290                            block_group, group_desc, offset);
 291                 return NULL;
 292         }
 293
 294         desc = (struct ext4_group_desc *)(
 295                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
 296                 offset * EXT4_DESC_SIZE(sb));
 297         if (bh)
 298                 *bh = sbi->s_group_desc[group_desc];
 299         return desc;
 300 }
 301
 302 /*
 303  * Return the block number which was discovered to be invalid, or 0 if
 304  * the block bitmap is valid.
 305  */
 306 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 307                                             struct ext4_group_desc *desc,
 308                                             unsigned int block_group,
 309                                             struct buffer_head *bh)
 310 {
 311         ext4_grpblk_t offset;
 312         ext4_grpblk_t next_zero_bit;
 313         ext4_fsblk_t blk;
 314         ext4_fsblk_t group_first_block;
 315
 316         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
 317                 /* with FLEX_BG, the inode/block bitmaps and itable
 318                  * blocks may not be in the group at all
 319                  * so the bitmap validation will be skipped for those groups
 320                  * or it has to also read the block group where the bitmaps
 321                  * are located to verify they are set.
 322                  */
 323                 return 0;
 324         }
 325         group_first_block = ext4_group_first_block_no(sb, block_group);
 326
 327         /* check whether block bitmap block number is set */
 328         blk = ext4_block_bitmap(sb, desc);
 329         offset = blk - group_first_block;
 330         if (!ext4_test_bit(offset, bh->b_data))
 331                 /* bad block bitmap */
 332                 return blk;
 333
 334         /* check whether the inode bitmap block number is set */
 335         blk = ext4_inode_bitmap(sb, desc);
 336         offset = blk - group_first_block;
 337         if (!ext4_test_bit(offset, bh->b_data))
 338                 /* bad block bitmap */
 339                 return blk;
 340
 341         /* check whether the inode table block number is set */
 342         blk = ext4_inode_table(sb, desc);
 343         offset = blk - group_first_block;
 344         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 345                                 offset + EXT4_SB(sb)->s_itb_per_group,
 346                                 offset);
 347         if (next_zero_bit < offset + EXT4_SB(sb)->s_itb_per_group)
 348                 /* bad bitmap for inode tables */
 349                 return blk;
 350         return 0;
 351 }
 352
 353 void ext4_validate_block_bitmap(struct super_block *sb,
 354                                struct ext4_group_desc *desc,
 355                                unsigned int block_group,
 356                                struct buffer_head *bh)
 357 {
 358         ext4_fsblk_t    blk;
 359
 360         if (buffer_verified(bh))
 361                 return;
 362
 363         ext4_lock_group(sb, block_group);
 364         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
 365         if (unlikely(blk != 0)) {
 366                 ext4_unlock_group(sb, block_group);
 367                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
 368                            block_group, blk);
 369                 return;
 370         }
 371         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 372                         desc, bh))) {
 373                 ext4_unlock_group(sb, block_group);
 374                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 375                 return;
 376         }
 377         set_buffer_verified(bh);
 378         ext4_unlock_group(sb, block_group);
 379 }
 380
 381 /**
 382  * ext4_read_block_bitmap_nowait()
 383  * @sb:                 super block
 384  * @block_group:        given block group
 385  *
 386  * Read the bitmap for a given block_group,and validate the
 387  * bits for block/inode/inode tables are set in the bitmaps
 388  *
 389  * Return buffer_head on success or NULL in case of failure.
 390  */
 391 struct buffer_head *
 392 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 393 {
 394         struct ext4_group_desc *desc;
 395         struct buffer_head *bh;
 396         ext4_fsblk_t bitmap_blk;
 397
 398         desc = ext4_get_group_desc(sb, block_group, NULL);
 399         if (!desc)
 400                 return NULL;
 401         bitmap_blk = ext4_block_bitmap(sb, desc);
 402         bh = sb_getblk(sb, bitmap_blk);
 403         if (unlikely(!bh)) {
 404                 ext4_error(sb, "Cannot get buffer for block bitmap - "
 405                            "block_group = %u, block_bitmap = %llu",
 406                            block_group, bitmap_blk);
 407                 return NULL;
 408         }
 409
 410         if (bitmap_uptodate(bh))
 411                 goto verify;
 412
 413         lock_buffer(bh);
 414         if (bitmap_uptodate(bh)) {
 415                 unlock_buffer(bh);
 416                 goto verify;
 417         }
 418         ext4_lock_group(sb, block_group);
 419         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 420                 ext4_init_block_bitmap(sb, bh, block_group, desc);
 421                 set_bitmap_uptodate(bh);
 422                 set_buffer_uptodate(bh);
 423                 ext4_unlock_group(sb, block_group);
 424                 unlock_buffer(bh);
 425                 return bh;
 426         }
 427         ext4_unlock_group(sb, block_group);
 428         if (buffer_uptodate(bh)) {
 429                 /*
 430                  * if not uninit if bh is uptodate,
 431                  * bitmap is also uptodate
 432                  */
 433                 set_bitmap_uptodate(bh);
 434                 unlock_buffer(bh);
 435                 goto verify;
 436         }
 437         /*
 438          * submit the buffer_head for reading
 439          */
 440         set_buffer_new(bh);
 441         trace_ext4_read_block_bitmap_load(sb, block_group);
 442         bh->b_end_io = ext4_end_bitmap_read;
 443         get_bh(bh);
 444         submit_bh(READ | REQ_META | REQ_PRIO, bh);
 445         return bh;
 446 verify:
 447         ext4_validate_block_bitmap(sb, desc, block_group, bh);
 448         return bh;
 449 }
 450
 451 /* Returns 0 on success, 1 on error */
 452 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
 453                            struct buffer_head *bh)
 454 {
 455         struct ext4_group_desc *desc;
 456
 457         if (!buffer_new(bh))
 458                 return 0;
 459         desc = ext4_get_group_desc(sb, block_group, NULL);
 460         if (!desc)
 461                 return 1;
 462         wait_on_buffer(bh);
 463         if (!buffer_uptodate(bh)) {
 464                 ext4_error(sb, "Cannot read block bitmap - "
 465                            "block_group = %u, block_bitmap = %llu",
 466                            block_group, (unsigned long long) bh->b_blocknr);
 467                 return 1;
 468         }
 469         clear_buffer_new(bh);
 470         /* Panic or remount fs read-only if block bitmap is invalid */
 471         ext4_validate_block_bitmap(sb, desc, block_group, bh);
 472         return 0;
 473 }
 474
 475 struct buffer_head *
 476 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 477 {
 478         struct buffer_head *bh;
 479
 480         bh = ext4_read_block_bitmap_nowait(sb, block_group);
 481         if (!bh)
 482                 return NULL;
 483         if (ext4_wait_block_bitmap(sb, block_group, bh)) {
 484                 put_bh(bh);
 485                 return NULL;
 486         }
 487         return bh;
 488 }
 489
 490 /**
 491  * ext4_has_free_clusters()
 492  * @sbi:        in-core super block structure.
 493  * @nclusters:  number of needed blocks
 494  * @flags:      flags from ext4_mb_new_blocks()
 495  *
 496  * Check if filesystem has nclusters free & available for allocation.
 497  * On success return 1, return 0 on failure.
 498  */
 499 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 500                                   s64 nclusters, unsigned int flags)
 501 {
 502         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 503         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 504         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 505
 506         free_clusters  = percpu_counter_read_positive(fcc);
 507         dirty_clusters = percpu_counter_read_positive(dcc);
 508         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 509
 510         /*
 511          * r_blocks_count should always be multiple of the cluster ratio so
 512          * we are safe to do a plane bit shift only.
 513          */
 514         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 515               resv_clusters;
 516
 517         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 518                                         EXT4_FREECLUSTERS_WATERMARK) {
 519                 free_clusters  = percpu_counter_sum_positive(fcc);
 520                 dirty_clusters = percpu_counter_sum_positive(dcc);
 521         }
 522         /* Check whether we have space after accounting for current
 523          * dirty clusters & root reserved clusters.
 524          */
 525         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 526                 return 1;
 527
 528         /* Hm, nope.  Are (enough) root reserved clusters available? */
 529         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 530             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 531             capable(CAP_SYS_RESOURCE) ||
 532             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 533
 534                 if (free_clusters >= (nclusters + dirty_clusters +
 535                                       resv_clusters))
 536                         return 1;
 537         }
 538         /* No free blocks. Let's see if we can dip into reserved pool */
 539         if (flags & EXT4_MB_USE_RESERVED) {
 540                 if (free_clusters >= (nclusters + dirty_clusters))
 541                         return 1;
 542         }
 543
 544         return 0;
 545 }
 546
 547 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 548                              s64 nclusters, unsigned int flags)
 549 {
 550         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 551                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 552                 return 0;
 553         } else
 554                 return -ENOSPC;
 555 }
 556
 557 /**
 558  * ext4_should_retry_alloc()
 559  * @sb:                 super block
 560  * @retries             number of attemps has been made
 561  *
 562  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 563  * it is profitable to retry the operation, this function will wait
 564  * for the current or committing transaction to complete, and then
 565  * return TRUE.
 566  *
 567  * if the total number of retries exceed three times, return FALSE.
 568  */
 569 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 570 {
 571         if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
 572             (*retries)++ > 3 ||
 573             !EXT4_SB(sb)->s_journal)
 574                 return 0;
 575
 576         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 577
 578         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 579 }
 580
 581 /*
 582  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 583  *
 584  * @handle:             handle to this transaction
 585  * @inode:              file inode
 586  * @goal:               given target block(filesystem wide)
 587  * @count:              pointer to total number of clusters needed
 588  * @errp:               error code
 589  *
 590  * Return 1st allocated block number on success, *count stores total account
 591  * error stores in errp pointer
 592  */
 593 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 594                                   ext4_fsblk_t goal, unsigned int flags,
 595                                   unsigned long *count, int *errp)
 596 {
 597         struct ext4_allocation_request ar;
 598         ext4_fsblk_t ret;
 599
 600         memset(&ar, 0, sizeof(ar));
 601         /* Fill with neighbour allocated blocks */
 602         ar.inode = inode;
 603         ar.goal = goal;
 604         ar.len = count ? *count : 1;
 605         ar.flags = flags;
 606
 607         ret = ext4_mb_new_blocks(handle, &ar, errp);
 608         if (count)
 609                 *count = ar.len;
 610         /*
 611          * Account for the allocated meta blocks.  We will never
 612          * fail EDQUOT for metdata, but we do account for it.
 613          */
 614         if (!(*errp) &&
 615             ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
 616                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 617                 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
 618                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 619                 dquot_alloc_block_nofail(inode,
 620                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 621         }
 622         return ret;
 623 }
 624
 625 /**
 626  * ext4_count_free_clusters() -- count filesystem free clusters
 627  * @sb:         superblock
 628  *
 629  * Adds up the number of free clusters from each block group.
 630  */
 631 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 632 {
 633         ext4_fsblk_t desc_count;
 634         struct ext4_group_desc *gdp;
 635         ext4_group_t i;
 636         ext4_group_t ngroups = ext4_get_groups_count(sb);
 637 #ifdef EXT4FS_DEBUG
 638         struct ext4_super_block *es;
 639         ext4_fsblk_t bitmap_count;
 640         unsigned int x;
 641         struct buffer_head *bitmap_bh = NULL;
 642
 643         es = EXT4_SB(sb)->s_es;
 644         desc_count = 0;
 645         bitmap_count = 0;
 646         gdp = NULL;
 647
 648         for (i = 0; i < ngroups; i++) {
 649                 gdp = ext4_get_group_desc(sb, i, NULL);
 650                 if (!gdp)
 651                         continue;
 652                 desc_count += ext4_free_group_clusters(sb, gdp);
 653                 brelse(bitmap_bh);
 654                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 655                 if (bitmap_bh == NULL)
 656                         continue;
 657
 658                 x = ext4_count_free(bitmap_bh->b_data,
 659                                     EXT4_BLOCKS_PER_GROUP(sb) / 8);
 660                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 661                         i, ext4_free_group_clusters(sb, gdp), x);
 662                 bitmap_count += x;
 663         }
 664         brelse(bitmap_bh);
 665         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 666                ", computed = %llu, %llu\n",
 667                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 668                desc_count, bitmap_count);
 669         return bitmap_count;
 670 #else
 671         desc_count = 0;
 672         for (i = 0; i < ngroups; i++) {
 673                 gdp = ext4_get_group_desc(sb, i, NULL);
 674                 if (!gdp)
 675                         continue;
 676                 desc_count += ext4_free_group_clusters(sb, gdp);
 677         }
 678
 679         return desc_count;
 680 #endif
 681 }
 682
 683 static inline int test_root(ext4_group_t a, int b)
 684 {
 685         int num = b;
 686
 687         while (a > num)
 688                 num *= b;
 689         return num == a;
 690 }
 691
 692 static int ext4_group_sparse(ext4_group_t group)
 693 {
 694         if (group <= 1)
 695                 return 1;
 696         if (!(group & 1))
 697                 return 0;
 698         return (test_root(group, 7) || test_root(group, 5) ||
 699                 test_root(group, 3));
 700 }
 701
 702 /**
 703  *      ext4_bg_has_super - number of blocks used by the superblock in group
 704  *      @sb: superblock for filesystem
 705  *      @group: group number to check
 706  *
 707  *      Return the number of blocks used by the superblock (primary or backup)
 708  *      in this group.  Currently this will be only 0 or 1.
 709  */
 710 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 711 {
 712         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
 713                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
 714                         !ext4_group_sparse(group))
 715                 return 0;
 716         return 1;
 717 }
 718
 719 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 720                                         ext4_group_t group)
 721 {
 722         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 723         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 724         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 725
 726         if (group == first || group == first + 1 || group == last)
 727                 return 1;
 728         return 0;
 729 }
 730
 731 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 732                                         ext4_group_t group)
 733 {
 734         if (!ext4_bg_has_super(sb, group))
 735                 return 0;
 736
 737         if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
 738                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 739         else
 740                 return EXT4_SB(sb)->s_gdb_count;
 741 }
 742
 743 /**
 744  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 745  *      @sb: superblock for filesystem
 746  *      @group: group number to check
 747  *
 748  *      Return the number of blocks used by the group descriptor table
 749  *      (primary or backup) in this group.  In the future there may be a
 750  *      different number of descriptor blocks in each group.
 751  */
 752 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 753 {
 754         unsigned long first_meta_bg =
 755                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 756         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 757
 758         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
 759                         metagroup < first_meta_bg)
 760                 return ext4_bg_num_gdb_nometa(sb, group);
 761
 762         return ext4_bg_num_gdb_meta(sb,group);
 763
 764 }
 765
 766 /*
 767  * This function returns the number of file system metadata clusters at
 768  * the beginning of a block group, including the reserved gdt blocks.
 769  */
 770 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
 771                                      ext4_group_t block_group)
 772 {
 773         struct ext4_sb_info *sbi = EXT4_SB(sb);
 774         unsigned num;
 775
 776         /* Check for superblock and gdt backups in this group */
 777         num = ext4_bg_has_super(sb, block_group);
 778
 779         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
 780             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 781                           sbi->s_desc_per_block) {
 782                 if (num) {
 783                         num += ext4_bg_num_gdb(sb, block_group);
 784                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 785                 }
 786         } else { /* For META_BG_BLOCK_GROUPS */
 787                 num += ext4_bg_num_gdb(sb, block_group);
 788         }
 789         return EXT4_NUM_B2C(sbi, num);
 790 }
 791 /**
 792  *      ext4_inode_to_goal_block - return a hint for block allocation
 793  *      @inode: inode for block allocation
 794  *
 795  *      Return the ideal location to start allocating blocks for a
 796  *      newly created inode.
 797  */
 798 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 799 {
 800         struct ext4_inode_info *ei = EXT4_I(inode);
 801         ext4_group_t block_group;
 802         ext4_grpblk_t colour;
 803         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 804         ext4_fsblk_t bg_start;
 805         ext4_fsblk_t last_block;
 806
 807         block_group = ei->i_block_group;
 808         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 809                 /*
 810                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 811                  * block groups per flexgroup, reserve the first block
 812                  * group for directories and special files.  Regular
 813                  * files will start at the second block group.  This
 814                  * tends to speed up directory access and improves
 815                  * fsck times.
 816                  */
 817                 block_group &= ~(flex_size-1);
 818                 if (S_ISREG(inode->i_mode))
 819                         block_group++;
 820         }
 821         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 822         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 823
 824         /*
 825          * If we are doing delayed allocation, we don't need take
 826          * colour into account.
 827          */
 828         if (test_opt(inode->i_sb, DELALLOC))
 829                 return bg_start;
 830
 831         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 832                 colour = (current->pid % 16) *
 833                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 834         else
 835                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 836         return bg_start + colour;
 837 }
 838