fs/ext4/balloc.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/balloc.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
  11  *  Big-endian to little-endian byte-swapping/bitmaps by
  12  *        David S. Miller (davem@caip.rutgers.edu), 1995
  13  */
  14
  15 #include <linux/time.h>
  16 #include <linux/capability.h>
  17 #include <linux/fs.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 int 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
 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_mark_group_bitmap_corrupted(sb, block_group,
 194                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT |
 195                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 196                 return -EFSBADCRC;
 197         }
 198         memset(bh->b_data, 0, sb->s_blocksize);
 199
 200         bit_max = ext4_num_base_meta_clusters(sb, block_group);
 201         if ((bit_max >> 3) >= bh->b_size)
 202                 return -EFSCORRUPTED;
 203
 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         /* Set bits for block and inode bitmaps, and inode table */
 210         tmp = ext4_block_bitmap(sb, gdp);
 211         if (ext4_block_in_group(sb, tmp, block_group))
 212                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 213
 214         tmp = ext4_inode_bitmap(sb, gdp);
 215         if (ext4_block_in_group(sb, tmp, block_group))
 216                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 217
 218         tmp = ext4_inode_table(sb, gdp);
 219         for (; tmp < ext4_inode_table(sb, gdp) +
 220                      sbi->s_itb_per_group; tmp++) {
 221                 if (ext4_block_in_group(sb, tmp, block_group))
 222                         ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 223         }
 224
 225         /*
 226          * Also if the number of blocks within the group is less than
 227          * the blocksize * 8 ( which is the size of bitmap ), set rest
 228          * of the block bitmap to 1
 229          */
 230         ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
 231                              sb->s_blocksize * 8, bh->b_data);
 232         return 0;
 233 }
 234
 235 /* Return the number of free blocks in a block group.  It is used when
 236  * the block bitmap is uninitialized, so we can't just count the bits
 237  * in the bitmap. */
 238 unsigned ext4_free_clusters_after_init(struct super_block *sb,
 239                                        ext4_group_t block_group,
 240                                        struct ext4_group_desc *gdp)
 241 {
 242         return num_clusters_in_group(sb, block_group) -
 243                 ext4_num_overhead_clusters(sb, block_group, gdp);
 244 }
 245
 246 /*
 247  * The free blocks are managed by bitmaps.  A file system contains several
 248  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 249  * block for inodes, N blocks for the inode table and data blocks.
 250  *
 251  * The file system contains group descriptors which are located after the
 252  * super block.  Each descriptor contains the number of the bitmap block and
 253  * the free blocks count in the block.  The descriptors are loaded in memory
 254  * when a file system is mounted (see ext4_fill_super).
 255  */
 256
 257 /**
 258  * ext4_get_group_desc() -- load group descriptor from disk
 259  * @sb:                 super block
 260  * @block_group:        given block group
 261  * @bh:                 pointer to the buffer head to store the block
 262  *                      group descriptor
 263  */
 264 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 265                                              ext4_group_t block_group,
 266                                              struct buffer_head **bh)
 267 {
 268         unsigned int group_desc;
 269         unsigned int offset;
 270         ext4_group_t ngroups = ext4_get_groups_count(sb);
 271         struct ext4_group_desc *desc;
 272         struct ext4_sb_info *sbi = EXT4_SB(sb);
 273         struct buffer_head *bh_p;
 274
 275         if (block_group >= ngroups) {
 276                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
 277                            " groups_count = %u", block_group, ngroups);
 278
 279                 return NULL;
 280         }
 281
 282         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 283         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 284         bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
 285         /*
 286          * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
 287          * the pointer being dereferenced won't be dereferenced again. By
 288          * looking at the usage in add_new_gdb() the value isn't modified,
 289          * just the pointer, and so it remains valid.
 290          */
 291         if (!bh_p) {
 292                 ext4_error(sb, "Group descriptor not loaded - "
 293                            "block_group = %u, group_desc = %u, desc = %u",
 294                            block_group, group_desc, offset);
 295                 return NULL;
 296         }
 297
 298         desc = (struct ext4_group_desc *)(
 299                 (__u8 *)bh_p->b_data +
 300                 offset * EXT4_DESC_SIZE(sb));
 301         if (bh)
 302                 *bh = bh_p;
 303         return desc;
 304 }
 305
 306 static ext4_fsblk_t ext4_valid_block_bitmap_padding(struct super_block *sb,
 307                                                     ext4_group_t block_group,
 308                                                     struct buffer_head *bh)
 309 {
 310         ext4_grpblk_t next_zero_bit;
 311         unsigned long bitmap_size = sb->s_blocksize * 8;
 312         unsigned int offset = num_clusters_in_group(sb, block_group);
 313
 314         if (bitmap_size <= offset)
 315                 return 0;
 316
 317         next_zero_bit = ext4_find_next_zero_bit(bh->b_data, bitmap_size, offset);
 318
 319         return (next_zero_bit < bitmap_size ? next_zero_bit : 0);
 320 }
 321
 322 /*
 323  * Return the block number which was discovered to be invalid, or 0 if
 324  * the block bitmap is valid.
 325  */
 326 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 327                                             struct ext4_group_desc *desc,
 328                                             ext4_group_t block_group,
 329                                             struct buffer_head *bh)
 330 {
 331         struct ext4_sb_info *sbi = EXT4_SB(sb);
 332         ext4_grpblk_t offset;
 333         ext4_grpblk_t next_zero_bit;
 334         ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
 335         ext4_fsblk_t blk;
 336         ext4_fsblk_t group_first_block;
 337
 338         if (ext4_has_feature_flex_bg(sb)) {
 339                 /* with FLEX_BG, the inode/block bitmaps and itable
 340                  * blocks may not be in the group at all
 341                  * so the bitmap validation will be skipped for those groups
 342                  * or it has to also read the block group where the bitmaps
 343                  * are located to verify they are set.
 344                  */
 345                 return 0;
 346         }
 347         group_first_block = ext4_group_first_block_no(sb, block_group);
 348
 349         /* check whether block bitmap block number is set */
 350         blk = ext4_block_bitmap(sb, desc);
 351         offset = blk - group_first_block;
 352         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 353             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 354                 /* bad block bitmap */
 355                 return blk;
 356
 357         /* check whether the inode bitmap block number is set */
 358         blk = ext4_inode_bitmap(sb, desc);
 359         offset = blk - group_first_block;
 360         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 361             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 362                 /* bad block bitmap */
 363                 return blk;
 364
 365         /* check whether the inode table block number is set */
 366         blk = ext4_inode_table(sb, desc);
 367         offset = blk - group_first_block;
 368         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 369             EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
 370                 return blk;
 371         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 372                         EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
 373                         EXT4_B2C(sbi, offset));
 374         if (next_zero_bit <
 375             EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
 376                 /* bad bitmap for inode tables */
 377                 return blk;
 378         return 0;
 379 }
 380
 381 static int ext4_validate_block_bitmap(struct super_block *sb,
 382                                       struct ext4_group_desc *desc,
 383                                       ext4_group_t block_group,
 384                                       struct buffer_head *bh)
 385 {
 386         ext4_fsblk_t    blk;
 387         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
 388
 389         if (buffer_verified(bh))
 390                 return 0;
 391         if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 392                 return -EFSCORRUPTED;
 393
 394         ext4_lock_group(sb, block_group);
 395         if (buffer_verified(bh))
 396                 goto verified;
 397         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 398                         desc, bh))) {
 399                 ext4_unlock_group(sb, block_group);
 400                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 401                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 402                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 403                 return -EFSBADCRC;
 404         }
 405         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
 406         if (unlikely(blk != 0)) {
 407                 ext4_unlock_group(sb, block_group);
 408                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
 409                            block_group, blk);
 410                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 411                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 412                 return -EFSCORRUPTED;
 413         }
 414         blk = ext4_valid_block_bitmap_padding(sb, block_group, bh);
 415         if (unlikely(blk != 0)) {
 416                 ext4_unlock_group(sb, block_group);
 417                 ext4_error(sb, "bg %u: block %llu: padding at end of block bitmap is not set",
 418                            block_group, blk);
 419                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 420                                                  EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 421                 return -EFSCORRUPTED;
 422         }
 423         set_buffer_verified(bh);
 424 verified:
 425         ext4_unlock_group(sb, block_group);
 426         return 0;
 427 }
 428
 429 /**
 430  * ext4_read_block_bitmap_nowait()
 431  * @sb:                 super block
 432  * @block_group:        given block group
 433  *
 434  * Read the bitmap for a given block_group,and validate the
 435  * bits for block/inode/inode tables are set in the bitmaps
 436  *
 437  * Return buffer_head on success or NULL in case of failure.
 438  */
 439 struct buffer_head *
 440 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 441 {
 442         struct ext4_group_desc *desc;
 443         struct ext4_sb_info *sbi = EXT4_SB(sb);
 444         struct buffer_head *bh;
 445         ext4_fsblk_t bitmap_blk;
 446         int err;
 447
 448         desc = ext4_get_group_desc(sb, block_group, NULL);
 449         if (!desc)
 450                 return ERR_PTR(-EFSCORRUPTED);
 451         bitmap_blk = ext4_block_bitmap(sb, desc);
 452         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 453             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 454                 ext4_error(sb, "Invalid block bitmap block %llu in "
 455                            "block_group %u", bitmap_blk, block_group);
 456                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 457                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 458                 return ERR_PTR(-EFSCORRUPTED);
 459         }
 460         bh = sb_getblk(sb, bitmap_blk);
 461         if (unlikely(!bh)) {
 462                 ext4_warning(sb, "Cannot get buffer for block bitmap - "
 463                              "block_group = %u, block_bitmap = %llu",
 464                              block_group, bitmap_blk);
 465                 return ERR_PTR(-ENOMEM);
 466         }
 467
 468         if (bitmap_uptodate(bh))
 469                 goto verify;
 470
 471         lock_buffer(bh);
 472         if (bitmap_uptodate(bh)) {
 473                 unlock_buffer(bh);
 474                 goto verify;
 475         }
 476         ext4_lock_group(sb, block_group);
 477         if (ext4_has_group_desc_csum(sb) &&
 478             (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 479                 if (block_group == 0) {
 480                         ext4_unlock_group(sb, block_group);
 481                         unlock_buffer(bh);
 482                         ext4_error(sb, "Block bitmap for bg 0 marked "
 483                                    "uninitialized");
 484                         err = -EFSCORRUPTED;
 485                         goto out;
 486                 }
 487                 err = ext4_init_block_bitmap(sb, bh, block_group, desc);
 488                 set_bitmap_uptodate(bh);
 489                 set_buffer_uptodate(bh);
 490                 set_buffer_verified(bh);
 491                 ext4_unlock_group(sb, block_group);
 492                 unlock_buffer(bh);
 493                 if (err) {
 494                         ext4_error(sb, "Failed to init block bitmap for group "
 495                                    "%u: %d", block_group, err);
 496                         goto out;
 497                 }
 498                 goto verify;
 499         }
 500         ext4_unlock_group(sb, block_group);
 501         if (buffer_uptodate(bh)) {
 502                 /*
 503                  * if not uninit if bh is uptodate,
 504                  * bitmap is also uptodate
 505                  */
 506                 set_bitmap_uptodate(bh);
 507                 unlock_buffer(bh);
 508                 goto verify;
 509         }
 510         /*
 511          * submit the buffer_head for reading
 512          */
 513         set_buffer_new(bh);
 514         trace_ext4_read_block_bitmap_load(sb, block_group);
 515         bh->b_end_io = ext4_end_bitmap_read;
 516         get_bh(bh);
 517         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 518         return bh;
 519 verify:
 520         err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
 521         if (err)
 522                 goto out;
 523         return bh;
 524 out:
 525         put_bh(bh);
 526         return ERR_PTR(err);
 527 }
 528
 529 /* Returns 0 on success, 1 on error */
 530 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
 531                            struct buffer_head *bh)
 532 {
 533         struct ext4_group_desc *desc;
 534
 535         if (!buffer_new(bh))
 536                 return 0;
 537         desc = ext4_get_group_desc(sb, block_group, NULL);
 538         if (!desc)
 539                 return -EFSCORRUPTED;
 540         wait_on_buffer(bh);
 541         if (!buffer_uptodate(bh)) {
 542                 ext4_error(sb, "Cannot read block bitmap - "
 543                            "block_group = %u, block_bitmap = %llu",
 544                            block_group, (unsigned long long) bh->b_blocknr);
 545                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 546                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 547                 return -EIO;
 548         }
 549         clear_buffer_new(bh);
 550         /* Panic or remount fs read-only if block bitmap is invalid */
 551         return ext4_validate_block_bitmap(sb, desc, block_group, bh);
 552 }
 553
 554 struct buffer_head *
 555 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 556 {
 557         struct buffer_head *bh;
 558         int err;
 559
 560         bh = ext4_read_block_bitmap_nowait(sb, block_group);
 561         if (IS_ERR(bh))
 562                 return bh;
 563         err = ext4_wait_block_bitmap(sb, block_group, bh);
 564         if (err) {
 565                 put_bh(bh);
 566                 return ERR_PTR(err);
 567         }
 568         return bh;
 569 }
 570
 571 /**
 572  * ext4_has_free_clusters()
 573  * @sbi:        in-core super block structure.
 574  * @nclusters:  number of needed blocks
 575  * @flags:      flags from ext4_mb_new_blocks()
 576  *
 577  * Check if filesystem has nclusters free & available for allocation.
 578  * On success return 1, return 0 on failure.
 579  */
 580 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 581                                   s64 nclusters, unsigned int flags)
 582 {
 583         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 584         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 585         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 586
 587         free_clusters  = percpu_counter_read_positive(fcc);
 588         dirty_clusters = percpu_counter_read_positive(dcc);
 589         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 590
 591         /*
 592          * r_blocks_count should always be multiple of the cluster ratio so
 593          * we are safe to do a plane bit shift only.
 594          */
 595         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 596               resv_clusters;
 597
 598         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 599                                         EXT4_FREECLUSTERS_WATERMARK) {
 600                 free_clusters  = percpu_counter_sum_positive(fcc);
 601                 dirty_clusters = percpu_counter_sum_positive(dcc);
 602         }
 603         /* Check whether we have space after accounting for current
 604          * dirty clusters & root reserved clusters.
 605          */
 606         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 607                 return 1;
 608
 609         /* Hm, nope.  Are (enough) root reserved clusters available? */
 610         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 611             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 612             capable(CAP_SYS_RESOURCE) ||
 613             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 614
 615                 if (free_clusters >= (nclusters + dirty_clusters +
 616                                       resv_clusters))
 617                         return 1;
 618         }
 619         /* No free blocks. Let's see if we can dip into reserved pool */
 620         if (flags & EXT4_MB_USE_RESERVED) {
 621                 if (free_clusters >= (nclusters + dirty_clusters))
 622                         return 1;
 623         }
 624
 625         return 0;
 626 }
 627
 628 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 629                              s64 nclusters, unsigned int flags)
 630 {
 631         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 632                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 633                 return 0;
 634         } else
 635                 return -ENOSPC;
 636 }
 637
 638 /**
 639  * ext4_should_retry_alloc() - check if a block allocation should be retried
 640  * @sb:                 superblock
 641  * @retries:            number of retry attempts made so far
 642  *
 643  * ext4_should_retry_alloc() is called when ENOSPC is returned while
 644  * attempting to allocate blocks.  If there's an indication that a pending
 645  * journal transaction might free some space and allow another attempt to
 646  * succeed, this function will wait for the current or committing transaction
 647  * to complete and then return TRUE.
 648  */
 649 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 650 {
 651         struct ext4_sb_info *sbi = EXT4_SB(sb);
 652
 653         if (!sbi->s_journal)
 654                 return 0;
 655
 656         if (++(*retries) > 3) {
 657                 percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit);
 658                 return 0;
 659         }
 660
 661         /*
 662          * if there's no indication that blocks are about to be freed it's
 663          * possible we just missed a transaction commit that did so
 664          */
 665         smp_mb();
 666         if (sbi->s_mb_free_pending == 0)
 667                 return ext4_has_free_clusters(sbi, 1, 0);
 668
 669         /*
 670          * it's possible we've just missed a transaction commit here,
 671          * so ignore the returned status
 672          */
 673         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 674         (void) jbd2_journal_force_commit_nested(sbi->s_journal);
 675         return 1;
 676 }
 677
 678 /*
 679  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 680  *
 681  * @handle:             handle to this transaction
 682  * @inode:              file inode
 683  * @goal:               given target block(filesystem wide)
 684  * @count:              pointer to total number of clusters needed
 685  * @errp:               error code
 686  *
 687  * Return 1st allocated block number on success, *count stores total account
 688  * error stores in errp pointer
 689  */
 690 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 691                                   ext4_fsblk_t goal, unsigned int flags,
 692                                   unsigned long *count, int *errp)
 693 {
 694         struct ext4_allocation_request ar;
 695         ext4_fsblk_t ret;
 696
 697         memset(&ar, 0, sizeof(ar));
 698         /* Fill with neighbour allocated blocks */
 699         ar.inode = inode;
 700         ar.goal = goal;
 701         ar.len = count ? *count : 1;
 702         ar.flags = flags;
 703
 704         ret = ext4_mb_new_blocks(handle, &ar, errp);
 705         if (count)
 706                 *count = ar.len;
 707         /*
 708          * Account for the allocated meta blocks.  We will never
 709          * fail EDQUOT for metdata, but we do account for it.
 710          */
 711         if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
 712                 dquot_alloc_block_nofail(inode,
 713                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 714         }
 715         return ret;
 716 }
 717
 718 /**
 719  * ext4_count_free_clusters() -- count filesystem free clusters
 720  * @sb:         superblock
 721  *
 722  * Adds up the number of free clusters from each block group.
 723  */
 724 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 725 {
 726         ext4_fsblk_t desc_count;
 727         struct ext4_group_desc *gdp;
 728         ext4_group_t i;
 729         ext4_group_t ngroups = ext4_get_groups_count(sb);
 730         struct ext4_group_info *grp;
 731 #ifdef EXT4FS_DEBUG
 732         struct ext4_super_block *es;
 733         ext4_fsblk_t bitmap_count;
 734         unsigned int x;
 735         struct buffer_head *bitmap_bh = NULL;
 736
 737         es = EXT4_SB(sb)->s_es;
 738         desc_count = 0;
 739         bitmap_count = 0;
 740         gdp = NULL;
 741
 742         for (i = 0; i < ngroups; i++) {
 743                 gdp = ext4_get_group_desc(sb, i, NULL);
 744                 if (!gdp)
 745                         continue;
 746                 grp = NULL;
 747                 if (EXT4_SB(sb)->s_group_info)
 748                         grp = ext4_get_group_info(sb, i);
 749                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 750                         desc_count += ext4_free_group_clusters(sb, gdp);
 751                 brelse(bitmap_bh);
 752                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 753                 if (IS_ERR(bitmap_bh)) {
 754                         bitmap_bh = NULL;
 755                         continue;
 756                 }
 757
 758                 x = ext4_count_free(bitmap_bh->b_data,
 759                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
 760                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 761                         i, ext4_free_group_clusters(sb, gdp), x);
 762                 bitmap_count += x;
 763         }
 764         brelse(bitmap_bh);
 765         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 766                ", computed = %llu, %llu\n",
 767                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 768                desc_count, bitmap_count);
 769         return bitmap_count;
 770 #else
 771         desc_count = 0;
 772         for (i = 0; i < ngroups; i++) {
 773                 gdp = ext4_get_group_desc(sb, i, NULL);
 774                 if (!gdp)
 775                         continue;
 776                 grp = NULL;
 777                 if (EXT4_SB(sb)->s_group_info)
 778                         grp = ext4_get_group_info(sb, i);
 779                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 780                         desc_count += ext4_free_group_clusters(sb, gdp);
 781         }
 782
 783         return desc_count;
 784 #endif
 785 }
 786
 787 static inline int test_root(ext4_group_t a, int b)
 788 {
 789         while (1) {
 790                 if (a < b)
 791                         return 0;
 792                 if (a == b)
 793                         return 1;
 794                 if ((a % b) != 0)
 795                         return 0;
 796                 a = a / b;
 797         }
 798 }
 799
 800 /**
 801  *      ext4_bg_has_super - number of blocks used by the superblock in group
 802  *      @sb: superblock for filesystem
 803  *      @group: group number to check
 804  *
 805  *      Return the number of blocks used by the superblock (primary or backup)
 806  *      in this group.  Currently this will be only 0 or 1.
 807  */
 808 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 809 {
 810         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 811
 812         if (group == 0)
 813                 return 1;
 814         if (ext4_has_feature_sparse_super2(sb)) {
 815                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
 816                     group == le32_to_cpu(es->s_backup_bgs[1]))
 817                         return 1;
 818                 return 0;
 819         }
 820         if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
 821                 return 1;
 822         if (!(group & 1))
 823                 return 0;
 824         if (test_root(group, 3) || (test_root(group, 5)) ||
 825             test_root(group, 7))
 826                 return 1;
 827
 828         return 0;
 829 }
 830
 831 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 832                                         ext4_group_t group)
 833 {
 834         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 835         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 836         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 837
 838         if (group == first || group == first + 1 || group == last)
 839                 return 1;
 840         return 0;
 841 }
 842
 843 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 844                                         ext4_group_t group)
 845 {
 846         if (!ext4_bg_has_super(sb, group))
 847                 return 0;
 848
 849         if (ext4_has_feature_meta_bg(sb))
 850                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 851         else
 852                 return EXT4_SB(sb)->s_gdb_count;
 853 }
 854
 855 /**
 856  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 857  *      @sb: superblock for filesystem
 858  *      @group: group number to check
 859  *
 860  *      Return the number of blocks used by the group descriptor table
 861  *      (primary or backup) in this group.  In the future there may be a
 862  *      different number of descriptor blocks in each group.
 863  */
 864 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 865 {
 866         unsigned long first_meta_bg =
 867                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 868         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 869
 870         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
 871                 return ext4_bg_num_gdb_nometa(sb, group);
 872
 873         return ext4_bg_num_gdb_meta(sb,group);
 874
 875 }
 876
 877 /*
 878  * This function returns the number of file system metadata clusters at
 879  * the beginning of a block group, including the reserved gdt blocks.
 880  */
 881 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
 882                                      ext4_group_t block_group)
 883 {
 884         struct ext4_sb_info *sbi = EXT4_SB(sb);
 885         unsigned num;
 886
 887         /* Check for superblock and gdt backups in this group */
 888         num = ext4_bg_has_super(sb, block_group);
 889
 890         if (!ext4_has_feature_meta_bg(sb) ||
 891             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 892                           sbi->s_desc_per_block) {
 893                 if (num) {
 894                         num += ext4_bg_num_gdb(sb, block_group);
 895                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 896                 }
 897         } else { /* For META_BG_BLOCK_GROUPS */
 898                 num += ext4_bg_num_gdb(sb, block_group);
 899         }
 900         return EXT4_NUM_B2C(sbi, num);
 901 }
 902 /**
 903  *      ext4_inode_to_goal_block - return a hint for block allocation
 904  *      @inode: inode for block allocation
 905  *
 906  *      Return the ideal location to start allocating blocks for a
 907  *      newly created inode.
 908  */
 909 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 910 {
 911         struct ext4_inode_info *ei = EXT4_I(inode);
 912         ext4_group_t block_group;
 913         ext4_grpblk_t colour;
 914         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 915         ext4_fsblk_t bg_start;
 916         ext4_fsblk_t last_block;
 917
 918         block_group = ei->i_block_group;
 919         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 920                 /*
 921                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 922                  * block groups per flexgroup, reserve the first block
 923                  * group for directories and special files.  Regular
 924                  * files will start at the second block group.  This
 925                  * tends to speed up directory access and improves
 926                  * fsck times.
 927                  */
 928                 block_group &= ~(flex_size-1);
 929                 if (S_ISREG(inode->i_mode))
 930                         block_group++;
 931         }
 932         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 933         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 934
 935         /*
 936          * If we are doing delayed allocation, we don't need take
 937          * colour into account.
 938          */
 939         if (test_opt(inode->i_sb, DELALLOC))
 940                 return bg_start;
 941
 942         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 943                 colour = (current->pid % 16) *
 944                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 945         else
 946                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 947         return bg_start + colour;
 948 }
 949