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