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 /*
 307  * Return the block number which was discovered to be invalid, or 0 if
 308  * the block bitmap is valid.
 309  */
 310 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 311                                             struct ext4_group_desc *desc,
 312                                             ext4_group_t block_group,
 313                                             struct buffer_head *bh)
 314 {
 315         struct ext4_sb_info *sbi = EXT4_SB(sb);
 316         ext4_grpblk_t offset;
 317         ext4_grpblk_t next_zero_bit;
 318         ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
 319         ext4_fsblk_t blk;
 320         ext4_fsblk_t group_first_block;
 321
 322         if (ext4_has_feature_flex_bg(sb)) {
 323                 /* with FLEX_BG, the inode/block bitmaps and itable
 324                  * blocks may not be in the group at all
 325                  * so the bitmap validation will be skipped for those groups
 326                  * or it has to also read the block group where the bitmaps
 327                  * are located to verify they are set.
 328                  */
 329                 return 0;
 330         }
 331         group_first_block = ext4_group_first_block_no(sb, block_group);
 332
 333         /* check whether block bitmap block number is set */
 334         blk = ext4_block_bitmap(sb, desc);
 335         offset = blk - group_first_block;
 336         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 337             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 338                 /* bad block bitmap */
 339                 return blk;
 340
 341         /* check whether the inode bitmap block number is set */
 342         blk = ext4_inode_bitmap(sb, desc);
 343         offset = blk - group_first_block;
 344         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 345             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 346                 /* bad block bitmap */
 347                 return blk;
 348
 349         /* check whether the inode table block number is set */
 350         blk = ext4_inode_table(sb, desc);
 351         offset = blk - group_first_block;
 352         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 353             EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
 354                 return blk;
 355         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 356                         EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
 357                         EXT4_B2C(sbi, offset));
 358         if (next_zero_bit <
 359             EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
 360                 /* bad bitmap for inode tables */
 361                 return blk;
 362         return 0;
 363 }
 364
 365 static int ext4_validate_block_bitmap(struct super_block *sb,
 366                                       struct ext4_group_desc *desc,
 367                                       ext4_group_t block_group,
 368                                       struct buffer_head *bh)
 369 {
 370         ext4_fsblk_t    blk;
 371         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
 372
 373         if (buffer_verified(bh))
 374                 return 0;
 375         if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 376                 return -EFSCORRUPTED;
 377
 378         ext4_lock_group(sb, block_group);
 379         if (buffer_verified(bh))
 380                 goto verified;
 381         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 382                                                     desc, bh) ||
 383                      ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
 384                 ext4_unlock_group(sb, block_group);
 385                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 386                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 387                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 388                 return -EFSBADCRC;
 389         }
 390         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
 391         if (unlikely(blk != 0)) {
 392                 ext4_unlock_group(sb, block_group);
 393                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
 394                            block_group, blk);
 395                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 396                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 397                 return -EFSCORRUPTED;
 398         }
 399         set_buffer_verified(bh);
 400 verified:
 401         ext4_unlock_group(sb, block_group);
 402         return 0;
 403 }
 404
 405 /**
 406  * ext4_read_block_bitmap_nowait()
 407  * @sb:                 super block
 408  * @block_group:        given block group
 409  *
 410  * Read the bitmap for a given block_group,and validate the
 411  * bits for block/inode/inode tables are set in the bitmaps
 412  *
 413  * Return buffer_head on success or NULL in case of failure.
 414  */
 415 struct buffer_head *
 416 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 417 {
 418         struct ext4_group_desc *desc;
 419         struct ext4_sb_info *sbi = EXT4_SB(sb);
 420         struct buffer_head *bh;
 421         ext4_fsblk_t bitmap_blk;
 422         int err;
 423
 424         desc = ext4_get_group_desc(sb, block_group, NULL);
 425         if (!desc)
 426                 return ERR_PTR(-EFSCORRUPTED);
 427         bitmap_blk = ext4_block_bitmap(sb, desc);
 428         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 429             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 430                 ext4_error(sb, "Invalid block bitmap block %llu in "
 431                            "block_group %u", bitmap_blk, block_group);
 432                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 433                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 434                 return ERR_PTR(-EFSCORRUPTED);
 435         }
 436         bh = sb_getblk(sb, bitmap_blk);
 437         if (unlikely(!bh)) {
 438                 ext4_warning(sb, "Cannot get buffer for block bitmap - "
 439                              "block_group = %u, block_bitmap = %llu",
 440                              block_group, bitmap_blk);
 441                 return ERR_PTR(-ENOMEM);
 442         }
 443
 444         if (bitmap_uptodate(bh))
 445                 goto verify;
 446
 447         lock_buffer(bh);
 448         if (bitmap_uptodate(bh)) {
 449                 unlock_buffer(bh);
 450                 goto verify;
 451         }
 452         ext4_lock_group(sb, block_group);
 453         if (ext4_has_group_desc_csum(sb) &&
 454             (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 455                 if (block_group == 0) {
 456                         ext4_unlock_group(sb, block_group);
 457                         unlock_buffer(bh);
 458                         ext4_error(sb, "Block bitmap for bg 0 marked "
 459                                    "uninitialized");
 460                         err = -EFSCORRUPTED;
 461                         goto out;
 462                 }
 463                 err = ext4_init_block_bitmap(sb, bh, block_group, desc);
 464                 set_bitmap_uptodate(bh);
 465                 set_buffer_uptodate(bh);
 466                 set_buffer_verified(bh);
 467                 ext4_unlock_group(sb, block_group);
 468                 unlock_buffer(bh);
 469                 if (err) {
 470                         ext4_error(sb, "Failed to init block bitmap for group "
 471                                    "%u: %d", block_group, err);
 472                         goto out;
 473                 }
 474                 goto verify;
 475         }
 476         ext4_unlock_group(sb, block_group);
 477         if (buffer_uptodate(bh)) {
 478                 /*
 479                  * if not uninit if bh is uptodate,
 480                  * bitmap is also uptodate
 481                  */
 482                 set_bitmap_uptodate(bh);
 483                 unlock_buffer(bh);
 484                 goto verify;
 485         }
 486         /*
 487          * submit the buffer_head for reading
 488          */
 489         set_buffer_new(bh);
 490         trace_ext4_read_block_bitmap_load(sb, block_group);
 491         bh->b_end_io = ext4_end_bitmap_read;
 492         get_bh(bh);
 493         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 494         return bh;
 495 verify:
 496         err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
 497         if (err)
 498                 goto out;
 499         return bh;
 500 out:
 501         put_bh(bh);
 502         return ERR_PTR(err);
 503 }
 504
 505 /* Returns 0 on success, 1 on error */
 506 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
 507                            struct buffer_head *bh)
 508 {
 509         struct ext4_group_desc *desc;
 510
 511         if (!buffer_new(bh))
 512                 return 0;
 513         desc = ext4_get_group_desc(sb, block_group, NULL);
 514         if (!desc)
 515                 return -EFSCORRUPTED;
 516         wait_on_buffer(bh);
 517         ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO);
 518         if (!buffer_uptodate(bh)) {
 519                 ext4_error_err(sb, EIO, "Cannot read block bitmap - "
 520                                "block_group = %u, block_bitmap = %llu",
 521                                block_group, (unsigned long long) bh->b_blocknr);
 522                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 523                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 524                 return -EIO;
 525         }
 526         clear_buffer_new(bh);
 527         /* Panic or remount fs read-only if block bitmap is invalid */
 528         return ext4_validate_block_bitmap(sb, desc, block_group, bh);
 529 }
 530
 531 struct buffer_head *
 532 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 533 {
 534         struct buffer_head *bh;
 535         int err;
 536
 537         bh = ext4_read_block_bitmap_nowait(sb, block_group);
 538         if (IS_ERR(bh))
 539                 return bh;
 540         err = ext4_wait_block_bitmap(sb, block_group, bh);
 541         if (err) {
 542                 put_bh(bh);
 543                 return ERR_PTR(err);
 544         }
 545         return bh;
 546 }
 547
 548 /**
 549  * ext4_has_free_clusters()
 550  * @sbi:        in-core super block structure.
 551  * @nclusters:  number of needed blocks
 552  * @flags:      flags from ext4_mb_new_blocks()
 553  *
 554  * Check if filesystem has nclusters free & available for allocation.
 555  * On success return 1, return 0 on failure.
 556  */
 557 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 558                                   s64 nclusters, unsigned int flags)
 559 {
 560         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 561         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 562         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 563
 564         free_clusters  = percpu_counter_read_positive(fcc);
 565         dirty_clusters = percpu_counter_read_positive(dcc);
 566         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 567
 568         /*
 569          * r_blocks_count should always be multiple of the cluster ratio so
 570          * we are safe to do a plane bit shift only.
 571          */
 572         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 573               resv_clusters;
 574
 575         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 576                                         EXT4_FREECLUSTERS_WATERMARK) {
 577                 free_clusters  = percpu_counter_sum_positive(fcc);
 578                 dirty_clusters = percpu_counter_sum_positive(dcc);
 579         }
 580         /* Check whether we have space after accounting for current
 581          * dirty clusters & root reserved clusters.
 582          */
 583         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 584                 return 1;
 585
 586         /* Hm, nope.  Are (enough) root reserved clusters available? */
 587         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 588             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 589             capable(CAP_SYS_RESOURCE) ||
 590             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 591
 592                 if (free_clusters >= (nclusters + dirty_clusters +
 593                                       resv_clusters))
 594                         return 1;
 595         }
 596         /* No free blocks. Let's see if we can dip into reserved pool */
 597         if (flags & EXT4_MB_USE_RESERVED) {
 598                 if (free_clusters >= (nclusters + dirty_clusters))
 599                         return 1;
 600         }
 601
 602         return 0;
 603 }
 604
 605 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 606                              s64 nclusters, unsigned int flags)
 607 {
 608         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 609                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 610                 return 0;
 611         } else
 612                 return -ENOSPC;
 613 }
 614
 615 /**
 616  * ext4_should_retry_alloc() - check if a block allocation should be retried
 617  * @sb:                 super block
 618  * @retries:            number of attemps has been made
 619  *
 620  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 621  * it is profitable to retry the operation, this function will wait
 622  * for the current or committing transaction to complete, and then
 623  * return TRUE.  We will only retry once.
 624  */
 625 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 626 {
 627         if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
 628             (*retries)++ > 1 ||
 629             !EXT4_SB(sb)->s_journal)
 630                 return 0;
 631
 632         smp_mb();
 633         if (EXT4_SB(sb)->s_mb_free_pending == 0)
 634                 return 0;
 635
 636         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 637         jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 638         return 1;
 639 }
 640
 641 /*
 642  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 643  *
 644  * @handle:             handle to this transaction
 645  * @inode:              file inode
 646  * @goal:               given target block(filesystem wide)
 647  * @count:              pointer to total number of clusters needed
 648  * @errp:               error code
 649  *
 650  * Return 1st allocated block number on success, *count stores total account
 651  * error stores in errp pointer
 652  */
 653 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 654                                   ext4_fsblk_t goal, unsigned int flags,
 655                                   unsigned long *count, int *errp)
 656 {
 657         struct ext4_allocation_request ar;
 658         ext4_fsblk_t ret;
 659
 660         memset(&ar, 0, sizeof(ar));
 661         /* Fill with neighbour allocated blocks */
 662         ar.inode = inode;
 663         ar.goal = goal;
 664         ar.len = count ? *count : 1;
 665         ar.flags = flags;
 666
 667         ret = ext4_mb_new_blocks(handle, &ar, errp);
 668         if (count)
 669                 *count = ar.len;
 670         /*
 671          * Account for the allocated meta blocks.  We will never
 672          * fail EDQUOT for metdata, but we do account for it.
 673          */
 674         if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
 675                 dquot_alloc_block_nofail(inode,
 676                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 677         }
 678         return ret;
 679 }
 680
 681 /**
 682  * ext4_count_free_clusters() -- count filesystem free clusters
 683  * @sb:         superblock
 684  *
 685  * Adds up the number of free clusters from each block group.
 686  */
 687 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 688 {
 689         ext4_fsblk_t desc_count;
 690         struct ext4_group_desc *gdp;
 691         ext4_group_t i;
 692         ext4_group_t ngroups = ext4_get_groups_count(sb);
 693         struct ext4_group_info *grp;
 694 #ifdef EXT4FS_DEBUG
 695         struct ext4_super_block *es;
 696         ext4_fsblk_t bitmap_count;
 697         unsigned int x;
 698         struct buffer_head *bitmap_bh = NULL;
 699
 700         es = EXT4_SB(sb)->s_es;
 701         desc_count = 0;
 702         bitmap_count = 0;
 703         gdp = NULL;
 704
 705         for (i = 0; i < ngroups; i++) {
 706                 gdp = ext4_get_group_desc(sb, i, NULL);
 707                 if (!gdp)
 708                         continue;
 709                 grp = NULL;
 710                 if (EXT4_SB(sb)->s_group_info)
 711                         grp = ext4_get_group_info(sb, i);
 712                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 713                         desc_count += ext4_free_group_clusters(sb, gdp);
 714                 brelse(bitmap_bh);
 715                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 716                 if (IS_ERR(bitmap_bh)) {
 717                         bitmap_bh = NULL;
 718                         continue;
 719                 }
 720
 721                 x = ext4_count_free(bitmap_bh->b_data,
 722                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
 723                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 724                         i, ext4_free_group_clusters(sb, gdp), x);
 725                 bitmap_count += x;
 726         }
 727         brelse(bitmap_bh);
 728         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 729                ", computed = %llu, %llu\n",
 730                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 731                desc_count, bitmap_count);
 732         return bitmap_count;
 733 #else
 734         desc_count = 0;
 735         for (i = 0; i < ngroups; i++) {
 736                 gdp = ext4_get_group_desc(sb, i, NULL);
 737                 if (!gdp)
 738                         continue;
 739                 grp = NULL;
 740                 if (EXT4_SB(sb)->s_group_info)
 741                         grp = ext4_get_group_info(sb, i);
 742                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 743                         desc_count += ext4_free_group_clusters(sb, gdp);
 744         }
 745
 746         return desc_count;
 747 #endif
 748 }
 749
 750 static inline int test_root(ext4_group_t a, int b)
 751 {
 752         while (1) {
 753                 if (a < b)
 754                         return 0;
 755                 if (a == b)
 756                         return 1;
 757                 if ((a % b) != 0)
 758                         return 0;
 759                 a = a / b;
 760         }
 761 }
 762
 763 /**
 764  *      ext4_bg_has_super - number of blocks used by the superblock in group
 765  *      @sb: superblock for filesystem
 766  *      @group: group number to check
 767  *
 768  *      Return the number of blocks used by the superblock (primary or backup)
 769  *      in this group.  Currently this will be only 0 or 1.
 770  */
 771 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 772 {
 773         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 774
 775         if (group == 0)
 776                 return 1;
 777         if (ext4_has_feature_sparse_super2(sb)) {
 778                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
 779                     group == le32_to_cpu(es->s_backup_bgs[1]))
 780                         return 1;
 781                 return 0;
 782         }
 783         if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
 784                 return 1;
 785         if (!(group & 1))
 786                 return 0;
 787         if (test_root(group, 3) || (test_root(group, 5)) ||
 788             test_root(group, 7))
 789                 return 1;
 790
 791         return 0;
 792 }
 793
 794 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 795                                         ext4_group_t group)
 796 {
 797         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 798         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 799         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 800
 801         if (group == first || group == first + 1 || group == last)
 802                 return 1;
 803         return 0;
 804 }
 805
 806 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 807                                         ext4_group_t group)
 808 {
 809         if (!ext4_bg_has_super(sb, group))
 810                 return 0;
 811
 812         if (ext4_has_feature_meta_bg(sb))
 813                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 814         else
 815                 return EXT4_SB(sb)->s_gdb_count;
 816 }
 817
 818 /**
 819  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 820  *      @sb: superblock for filesystem
 821  *      @group: group number to check
 822  *
 823  *      Return the number of blocks used by the group descriptor table
 824  *      (primary or backup) in this group.  In the future there may be a
 825  *      different number of descriptor blocks in each group.
 826  */
 827 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 828 {
 829         unsigned long first_meta_bg =
 830                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 831         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 832
 833         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
 834                 return ext4_bg_num_gdb_nometa(sb, group);
 835
 836         return ext4_bg_num_gdb_meta(sb,group);
 837
 838 }
 839
 840 /*
 841  * This function returns the number of file system metadata clusters at
 842  * the beginning of a block group, including the reserved gdt blocks.
 843  */
 844 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
 845                                      ext4_group_t block_group)
 846 {
 847         struct ext4_sb_info *sbi = EXT4_SB(sb);
 848         unsigned num;
 849
 850         /* Check for superblock and gdt backups in this group */
 851         num = ext4_bg_has_super(sb, block_group);
 852
 853         if (!ext4_has_feature_meta_bg(sb) ||
 854             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 855                           sbi->s_desc_per_block) {
 856                 if (num) {
 857                         num += ext4_bg_num_gdb(sb, block_group);
 858                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 859                 }
 860         } else { /* For META_BG_BLOCK_GROUPS */
 861                 num += ext4_bg_num_gdb(sb, block_group);
 862         }
 863         return EXT4_NUM_B2C(sbi, num);
 864 }
 865 /**
 866  *      ext4_inode_to_goal_block - return a hint for block allocation
 867  *      @inode: inode for block allocation
 868  *
 869  *      Return the ideal location to start allocating blocks for a
 870  *      newly created inode.
 871  */
 872 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 873 {
 874         struct ext4_inode_info *ei = EXT4_I(inode);
 875         ext4_group_t block_group;
 876         ext4_grpblk_t colour;
 877         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 878         ext4_fsblk_t bg_start;
 879         ext4_fsblk_t last_block;
 880
 881         block_group = ei->i_block_group;
 882         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 883                 /*
 884                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 885                  * block groups per flexgroup, reserve the first block
 886                  * group for directories and special files.  Regular
 887                  * files will start at the second block group.  This
 888                  * tends to speed up directory access and improves
 889                  * fsck times.
 890                  */
 891                 block_group &= ~(flex_size-1);
 892                 if (S_ISREG(inode->i_mode))
 893                         block_group++;
 894         }
 895         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 896         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 897
 898         /*
 899          * If we are doing delayed allocation, we don't need take
 900          * colour into account.
 901          */
 902         if (test_opt(inode->i_sb, DELALLOC))
 903                 return bg_start;
 904
 905         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 906                 colour = (current->pid % 16) *
 907                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 908         else
 909                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 910         return bg_start + colour;
 911 }
 912