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 static 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 static void ext4_init_block_bitmap(struct super_block *sb,
 180                                    struct buffer_head *bh,
 181                                    ext4_group_t block_group,
 182                                    struct ext4_group_desc *gdp)
 183 {
 184         unsigned int bit, bit_max;
 185         struct ext4_sb_info *sbi = EXT4_SB(sb);
 186         ext4_fsblk_t start, tmp;
 187         int flex_bg = 0;
 188         struct ext4_group_info *grp;
 189
 190         J_ASSERT_BH(bh, buffer_locked(bh));
 191
 192         /* If checksum is bad mark all blocks used to prevent allocation
 193          * essentially implementing a per-group read-only flag. */
 194         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
 195                 ext4_error(sb, "Checksum bad for group %u", block_group);
 196                 grp = ext4_get_group_info(sb, block_group);
 197                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 198                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 199                                            grp->bb_free);
 200                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
 201                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 202                         int count;
 203                         count = ext4_free_inodes_count(sb, gdp);
 204                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 205                                            count);
 206                 }
 207                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 208                 return;
 209         }
 210         memset(bh->b_data, 0, sb->s_blocksize);
 211
 212         bit_max = ext4_num_base_meta_clusters(sb, block_group);
 213         for (bit = 0; bit < bit_max; bit++)
 214                 ext4_set_bit(bit, bh->b_data);
 215
 216         start = ext4_group_first_block_no(sb, block_group);
 217
 218         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
 219                 flex_bg = 1;
 220
 221         /* Set bits for block and inode bitmaps, and inode table */
 222         tmp = ext4_block_bitmap(sb, gdp);
 223         if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 224                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 225
 226         tmp = ext4_inode_bitmap(sb, gdp);
 227         if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 228                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 229
 230         tmp = ext4_inode_table(sb, gdp);
 231         for (; tmp < ext4_inode_table(sb, gdp) +
 232                      sbi->s_itb_per_group; tmp++) {
 233                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
 234                         ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 235         }
 236
 237         /*
 238          * Also if the number of blocks within the group is less than
 239          * the blocksize * 8 ( which is the size of bitmap ), set rest
 240          * of the block bitmap to 1
 241          */
 242         ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
 243                              sb->s_blocksize * 8, bh->b_data);
 244         ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
 245         ext4_group_desc_csum_set(sb, block_group, gdp);
 246 }
 247
 248 /* Return the number of free blocks in a block group.  It is used when
 249  * the block bitmap is uninitialized, so we can't just count the bits
 250  * in the bitmap. */
 251 unsigned ext4_free_clusters_after_init(struct super_block *sb,
 252                                        ext4_group_t block_group,
 253                                        struct ext4_group_desc *gdp)
 254 {
 255         return num_clusters_in_group(sb, block_group) -
 256                 ext4_num_overhead_clusters(sb, block_group, gdp);
 257 }
 258
 259 /*
 260  * The free blocks are managed by bitmaps.  A file system contains several
 261  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 262  * block for inodes, N blocks for the inode table and data blocks.
 263  *
 264  * The file system contains group descriptors which are located after the
 265  * super block.  Each descriptor contains the number of the bitmap block and
 266  * the free blocks count in the block.  The descriptors are loaded in memory
 267  * when a file system is mounted (see ext4_fill_super).
 268  */
 269
 270 /**
 271  * ext4_get_group_desc() -- load group descriptor from disk
 272  * @sb:                 super block
 273  * @block_group:        given block group
 274  * @bh:                 pointer to the buffer head to store the block
 275  *                      group descriptor
 276  */
 277 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 278                                              ext4_group_t block_group,
 279                                              struct buffer_head **bh)
 280 {
 281         unsigned int group_desc;
 282         unsigned int offset;
 283         ext4_group_t ngroups = ext4_get_groups_count(sb);
 284         struct ext4_group_desc *desc;
 285         struct ext4_sb_info *sbi = EXT4_SB(sb);
 286
 287         if (block_group >= ngroups) {
 288                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
 289                            " groups_count = %u", block_group, ngroups);
 290
 291                 return NULL;
 292         }
 293
 294         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 295         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 296         if (!sbi->s_group_desc[group_desc]) {
 297                 ext4_error(sb, "Group descriptor not loaded - "
 298                            "block_group = %u, group_desc = %u, desc = %u",
 299                            block_group, group_desc, offset);
 300                 return NULL;
 301         }
 302
 303         desc = (struct ext4_group_desc *)(
 304                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
 305                 offset * EXT4_DESC_SIZE(sb));
 306         if (bh)
 307                 *bh = sbi->s_group_desc[group_desc];
 308         return desc;
 309 }
 310
 311 /*
 312  * Return the block number which was discovered to be invalid, or 0 if
 313  * the block bitmap is valid.
 314  */
 315 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 316                                             struct ext4_group_desc *desc,
 317                                             ext4_group_t block_group,
 318                                             struct buffer_head *bh)
 319 {
 320         struct ext4_sb_info *sbi = EXT4_SB(sb);
 321         ext4_grpblk_t offset;
 322         ext4_grpblk_t next_zero_bit;
 323         ext4_fsblk_t blk;
 324         ext4_fsblk_t group_first_block;
 325
 326         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
 327                 /* with FLEX_BG, the inode/block bitmaps and itable
 328                  * blocks may not be in the group at all
 329                  * so the bitmap validation will be skipped for those groups
 330                  * or it has to also read the block group where the bitmaps
 331                  * are located to verify they are set.
 332                  */
 333                 return 0;
 334         }
 335         group_first_block = ext4_group_first_block_no(sb, block_group);
 336
 337         /* check whether block bitmap block number is set */
 338         blk = ext4_block_bitmap(sb, desc);
 339         offset = blk - group_first_block;
 340         if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 341                 /* bad block bitmap */
 342                 return blk;
 343
 344         /* check whether the inode bitmap block number is set */
 345         blk = ext4_inode_bitmap(sb, desc);
 346         offset = blk - group_first_block;
 347         if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 348                 /* bad block bitmap */
 349                 return blk;
 350
 351         /* check whether the inode table block number is set */
 352         blk = ext4_inode_table(sb, desc);
 353         offset = blk - group_first_block;
 354         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 355                         EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
 356                         EXT4_B2C(sbi, offset));
 357         if (next_zero_bit <
 358             EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
 359                 /* bad bitmap for inode tables */
 360                 return blk;
 361         return 0;
 362 }
 363
 364 static void ext4_validate_block_bitmap(struct super_block *sb,
 365                                        struct ext4_group_desc *desc,
 366                                        ext4_group_t block_group,
 367                                        struct buffer_head *bh)
 368 {
 369         ext4_fsblk_t    blk;
 370         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
 371         struct ext4_sb_info *sbi = EXT4_SB(sb);
 372
 373         if (buffer_verified(bh))
 374                 return;
 375
 376         ext4_lock_group(sb, block_group);
 377         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
 378         if (unlikely(blk != 0)) {
 379                 ext4_unlock_group(sb, block_group);
 380                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
 381                            block_group, blk);
 382                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 383                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 384                                            grp->bb_free);
 385                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
 386                 return;
 387         }
 388         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 389                         desc, bh))) {
 390                 ext4_unlock_group(sb, block_group);
 391                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 392                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 393                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 394                                            grp->bb_free);
 395                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
 396                 return;
 397         }
 398         set_buffer_verified(bh);
 399         ext4_unlock_group(sb, block_group);
 400 }
 401
 402 /**
 403  * ext4_read_block_bitmap_nowait()
 404  * @sb:                 super block
 405  * @block_group:        given block group
 406  *
 407  * Read the bitmap for a given block_group,and validate the
 408  * bits for block/inode/inode tables are set in the bitmaps
 409  *
 410  * Return buffer_head on success or NULL in case of failure.
 411  */
 412 struct buffer_head *
 413 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 414 {
 415         struct ext4_group_desc *desc;
 416         struct buffer_head *bh;
 417         ext4_fsblk_t bitmap_blk;
 418
 419         desc = ext4_get_group_desc(sb, block_group, NULL);
 420         if (!desc)
 421                 return NULL;
 422         bitmap_blk = ext4_block_bitmap(sb, desc);
 423         bh = sb_getblk(sb, bitmap_blk);
 424         if (unlikely(!bh)) {
 425                 ext4_error(sb, "Cannot get buffer for block bitmap - "
 426                            "block_group = %u, block_bitmap = %llu",
 427                            block_group, bitmap_blk);
 428                 return NULL;
 429         }
 430
 431         if (bitmap_uptodate(bh))
 432                 goto verify;
 433
 434         lock_buffer(bh);
 435         if (bitmap_uptodate(bh)) {
 436                 unlock_buffer(bh);
 437                 goto verify;
 438         }
 439         ext4_lock_group(sb, block_group);
 440         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 441                 ext4_init_block_bitmap(sb, bh, block_group, desc);
 442                 set_bitmap_uptodate(bh);
 443                 set_buffer_uptodate(bh);
 444                 ext4_unlock_group(sb, block_group);
 445                 unlock_buffer(bh);
 446                 return bh;
 447         }
 448         ext4_unlock_group(sb, block_group);
 449         if (buffer_uptodate(bh)) {
 450                 /*
 451                  * if not uninit if bh is uptodate,
 452                  * bitmap is also uptodate
 453                  */
 454                 set_bitmap_uptodate(bh);
 455                 unlock_buffer(bh);
 456                 goto verify;
 457         }
 458         /*
 459          * submit the buffer_head for reading
 460          */
 461         set_buffer_new(bh);
 462         trace_ext4_read_block_bitmap_load(sb, block_group);
 463         bh->b_end_io = ext4_end_bitmap_read;
 464         get_bh(bh);
 465         submit_bh(READ | REQ_META | REQ_PRIO, bh);
 466         return bh;
 467 verify:
 468         ext4_validate_block_bitmap(sb, desc, block_group, bh);
 469         if (buffer_verified(bh))
 470                 return bh;
 471         put_bh(bh);
 472         return NULL;
 473 }
 474
 475 /* Returns 0 on success, 1 on error */
 476 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
 477                            struct buffer_head *bh)
 478 {
 479         struct ext4_group_desc *desc;
 480
 481         if (!buffer_new(bh))
 482                 return 0;
 483         desc = ext4_get_group_desc(sb, block_group, NULL);
 484         if (!desc)
 485                 return 1;
 486         wait_on_buffer(bh);
 487         if (!buffer_uptodate(bh)) {
 488                 ext4_error(sb, "Cannot read block bitmap - "
 489                            "block_group = %u, block_bitmap = %llu",
 490                            block_group, (unsigned long long) bh->b_blocknr);
 491                 return 1;
 492         }
 493         clear_buffer_new(bh);
 494         /* Panic or remount fs read-only if block bitmap is invalid */
 495         ext4_validate_block_bitmap(sb, desc, block_group, bh);
 496         /* ...but check for error just in case errors=continue. */
 497         return !buffer_verified(bh);
 498 }
 499
 500 struct buffer_head *
 501 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 502 {
 503         struct buffer_head *bh;
 504
 505         bh = ext4_read_block_bitmap_nowait(sb, block_group);
 506         if (!bh)
 507                 return NULL;
 508         if (ext4_wait_block_bitmap(sb, block_group, bh)) {
 509                 put_bh(bh);
 510                 return NULL;
 511         }
 512         return bh;
 513 }
 514
 515 /**
 516  * ext4_has_free_clusters()
 517  * @sbi:        in-core super block structure.
 518  * @nclusters:  number of needed blocks
 519  * @flags:      flags from ext4_mb_new_blocks()
 520  *
 521  * Check if filesystem has nclusters free & available for allocation.
 522  * On success return 1, return 0 on failure.
 523  */
 524 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 525                                   s64 nclusters, unsigned int flags)
 526 {
 527         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 528         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 529         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 530
 531         free_clusters  = percpu_counter_read_positive(fcc);
 532         dirty_clusters = percpu_counter_read_positive(dcc);
 533         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 534
 535         /*
 536          * r_blocks_count should always be multiple of the cluster ratio so
 537          * we are safe to do a plane bit shift only.
 538          */
 539         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 540               resv_clusters;
 541
 542         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 543                                         EXT4_FREECLUSTERS_WATERMARK) {
 544                 free_clusters  = percpu_counter_sum_positive(fcc);
 545                 dirty_clusters = percpu_counter_sum_positive(dcc);
 546         }
 547         /* Check whether we have space after accounting for current
 548          * dirty clusters & root reserved clusters.
 549          */
 550         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 551                 return 1;
 552
 553         /* Hm, nope.  Are (enough) root reserved clusters available? */
 554         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 555             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 556             capable(CAP_SYS_RESOURCE) ||
 557             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 558
 559                 if (free_clusters >= (nclusters + dirty_clusters +
 560                                       resv_clusters))
 561                         return 1;
 562         }
 563         /* No free blocks. Let's see if we can dip into reserved pool */
 564         if (flags & EXT4_MB_USE_RESERVED) {
 565                 if (free_clusters >= (nclusters + dirty_clusters))
 566                         return 1;
 567         }
 568
 569         return 0;
 570 }
 571
 572 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 573                              s64 nclusters, unsigned int flags)
 574 {
 575         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 576                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 577                 return 0;
 578         } else
 579                 return -ENOSPC;
 580 }
 581
 582 /**
 583  * ext4_should_retry_alloc()
 584  * @sb:                 super block
 585  * @retries             number of attemps has been made
 586  *
 587  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 588  * it is profitable to retry the operation, this function will wait
 589  * for the current or committing transaction to complete, and then
 590  * return TRUE.
 591  *
 592  * if the total number of retries exceed three times, return FALSE.
 593  */
 594 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 595 {
 596         if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
 597             (*retries)++ > 3 ||
 598             !EXT4_SB(sb)->s_journal)
 599                 return 0;
 600
 601         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 602
 603         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 604 }
 605
 606 /*
 607  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 608  *
 609  * @handle:             handle to this transaction
 610  * @inode:              file inode
 611  * @goal:               given target block(filesystem wide)
 612  * @count:              pointer to total number of clusters needed
 613  * @errp:               error code
 614  *
 615  * Return 1st allocated block number on success, *count stores total account
 616  * error stores in errp pointer
 617  */
 618 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 619                                   ext4_fsblk_t goal, unsigned int flags,
 620                                   unsigned long *count, int *errp)
 621 {
 622         struct ext4_allocation_request ar;
 623         ext4_fsblk_t ret;
 624
 625         memset(&ar, 0, sizeof(ar));
 626         /* Fill with neighbour allocated blocks */
 627         ar.inode = inode;
 628         ar.goal = goal;
 629         ar.len = count ? *count : 1;
 630         ar.flags = flags;
 631
 632         ret = ext4_mb_new_blocks(handle, &ar, errp);
 633         if (count)
 634                 *count = ar.len;
 635         /*
 636          * Account for the allocated meta blocks.  We will never
 637          * fail EDQUOT for metdata, but we do account for it.
 638          */
 639         if (!(*errp) &&
 640             ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
 641                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 642                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 643                 dquot_alloc_block_nofail(inode,
 644                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 645         }
 646         return ret;
 647 }
 648
 649 /**
 650  * ext4_count_free_clusters() -- count filesystem free clusters
 651  * @sb:         superblock
 652  *
 653  * Adds up the number of free clusters from each block group.
 654  */
 655 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 656 {
 657         ext4_fsblk_t desc_count;
 658         struct ext4_group_desc *gdp;
 659         ext4_group_t i;
 660         ext4_group_t ngroups = ext4_get_groups_count(sb);
 661         struct ext4_group_info *grp;
 662 #ifdef EXT4FS_DEBUG
 663         struct ext4_super_block *es;
 664         ext4_fsblk_t bitmap_count;
 665         unsigned int x;
 666         struct buffer_head *bitmap_bh = NULL;
 667
 668         es = EXT4_SB(sb)->s_es;
 669         desc_count = 0;
 670         bitmap_count = 0;
 671         gdp = NULL;
 672
 673         for (i = 0; i < ngroups; i++) {
 674                 gdp = ext4_get_group_desc(sb, i, NULL);
 675                 if (!gdp)
 676                         continue;
 677                 grp = NULL;
 678                 if (EXT4_SB(sb)->s_group_info)
 679                         grp = ext4_get_group_info(sb, i);
 680                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 681                         desc_count += ext4_free_group_clusters(sb, gdp);
 682                 brelse(bitmap_bh);
 683                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 684                 if (bitmap_bh == NULL)
 685                         continue;
 686
 687                 x = ext4_count_free(bitmap_bh->b_data,
 688                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
 689                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 690                         i, ext4_free_group_clusters(sb, gdp), x);
 691                 bitmap_count += x;
 692         }
 693         brelse(bitmap_bh);
 694         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 695                ", computed = %llu, %llu\n",
 696                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 697                desc_count, bitmap_count);
 698         return bitmap_count;
 699 #else
 700         desc_count = 0;
 701         for (i = 0; i < ngroups; i++) {
 702                 gdp = ext4_get_group_desc(sb, i, NULL);
 703                 if (!gdp)
 704                         continue;
 705                 grp = NULL;
 706                 if (EXT4_SB(sb)->s_group_info)
 707                         grp = ext4_get_group_info(sb, i);
 708                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 709                         desc_count += ext4_free_group_clusters(sb, gdp);
 710         }
 711
 712         return desc_count;
 713 #endif
 714 }
 715
 716 static inline int test_root(ext4_group_t a, int b)
 717 {
 718         while (1) {
 719                 if (a < b)
 720                         return 0;
 721                 if (a == b)
 722                         return 1;
 723                 if ((a % b) != 0)
 724                         return 0;
 725                 a = a / b;
 726         }
 727 }
 728
 729 /**
 730  *      ext4_bg_has_super - number of blocks used by the superblock in group
 731  *      @sb: superblock for filesystem
 732  *      @group: group number to check
 733  *
 734  *      Return the number of blocks used by the superblock (primary or backup)
 735  *      in this group.  Currently this will be only 0 or 1.
 736  */
 737 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 738 {
 739         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 740
 741         if (group == 0)
 742                 return 1;
 743         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_SPARSE_SUPER2)) {
 744                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
 745                     group == le32_to_cpu(es->s_backup_bgs[1]))
 746                         return 1;
 747                 return 0;
 748         }
 749         if ((group <= 1) || !EXT4_HAS_RO_COMPAT_FEATURE(sb,
 750                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER))
 751                 return 1;
 752         if (!(group & 1))
 753                 return 0;
 754         if (test_root(group, 3) || (test_root(group, 5)) ||
 755             test_root(group, 7))
 756                 return 1;
 757
 758         return 0;
 759 }
 760
 761 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 762                                         ext4_group_t group)
 763 {
 764         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 765         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 766         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 767
 768         if (group == first || group == first + 1 || group == last)
 769                 return 1;
 770         return 0;
 771 }
 772
 773 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 774                                         ext4_group_t group)
 775 {
 776         if (!ext4_bg_has_super(sb, group))
 777                 return 0;
 778
 779         if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
 780                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 781         else
 782                 return EXT4_SB(sb)->s_gdb_count;
 783 }
 784
 785 /**
 786  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 787  *      @sb: superblock for filesystem
 788  *      @group: group number to check
 789  *
 790  *      Return the number of blocks used by the group descriptor table
 791  *      (primary or backup) in this group.  In the future there may be a
 792  *      different number of descriptor blocks in each group.
 793  */
 794 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 795 {
 796         unsigned long first_meta_bg =
 797                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 798         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 799
 800         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
 801                         metagroup < first_meta_bg)
 802                 return ext4_bg_num_gdb_nometa(sb, group);
 803
 804         return ext4_bg_num_gdb_meta(sb,group);
 805
 806 }
 807
 808 /*
 809  * This function returns the number of file system metadata clusters at
 810  * the beginning of a block group, including the reserved gdt blocks.
 811  */
 812 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
 813                                      ext4_group_t block_group)
 814 {
 815         struct ext4_sb_info *sbi = EXT4_SB(sb);
 816         unsigned num;
 817
 818         /* Check for superblock and gdt backups in this group */
 819         num = ext4_bg_has_super(sb, block_group);
 820
 821         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
 822             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 823                           sbi->s_desc_per_block) {
 824                 if (num) {
 825                         num += ext4_bg_num_gdb(sb, block_group);
 826                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 827                 }
 828         } else { /* For META_BG_BLOCK_GROUPS */
 829                 num += ext4_bg_num_gdb(sb, block_group);
 830         }
 831         return EXT4_NUM_B2C(sbi, num);
 832 }
 833 /**
 834  *      ext4_inode_to_goal_block - return a hint for block allocation
 835  *      @inode: inode for block allocation
 836  *
 837  *      Return the ideal location to start allocating blocks for a
 838  *      newly created inode.
 839  */
 840 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 841 {
 842         struct ext4_inode_info *ei = EXT4_I(inode);
 843         ext4_group_t block_group;
 844         ext4_grpblk_t colour;
 845         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 846         ext4_fsblk_t bg_start;
 847         ext4_fsblk_t last_block;
 848
 849         block_group = ei->i_block_group;
 850         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 851                 /*
 852                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 853                  * block groups per flexgroup, reserve the first block
 854                  * group for directories and special files.  Regular
 855                  * files will start at the second block group.  This
 856                  * tends to speed up directory access and improves
 857                  * fsck times.
 858                  */
 859                 block_group &= ~(flex_size-1);
 860                 if (S_ISREG(inode->i_mode))
 861                         block_group++;
 862         }
 863         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 864         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 865
 866         /*
 867          * If we are doing delayed allocation, we don't need take
 868          * colour into account.
 869          */
 870         if (test_opt(inode->i_sb, DELALLOC))
 871                 return bg_start;
 872
 873         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 874                 colour = (current->pid % 16) *
 875                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 876         else
 877                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 878         return bg_start + colour;
 879 }
 880