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
 274         if (block_group >= ngroups) {
 275                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
 276                            " groups_count = %u", block_group, ngroups);
 277
 278                 return NULL;
 279         }
 280
 281         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 282         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 283         if (!sbi->s_group_desc[group_desc]) {
 284                 ext4_error(sb, "Group descriptor not loaded - "
 285                            "block_group = %u, group_desc = %u, desc = %u",
 286                            block_group, group_desc, offset);
 287                 return NULL;
 288         }
 289
 290         desc = (struct ext4_group_desc *)(
 291                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
 292                 offset * EXT4_DESC_SIZE(sb));
 293         if (bh)
 294                 *bh = sbi->s_group_desc[group_desc];
 295         return desc;
 296 }
 297
 298 /*
 299  * Return the block number which was discovered to be invalid, or 0 if
 300  * the block bitmap is valid.
 301  */
 302 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 303                                             struct ext4_group_desc *desc,
 304                                             ext4_group_t block_group,
 305                                             struct buffer_head *bh)
 306 {
 307         struct ext4_sb_info *sbi = EXT4_SB(sb);
 308         ext4_grpblk_t offset;
 309         ext4_grpblk_t next_zero_bit;
 310         ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
 311         ext4_fsblk_t blk;
 312         ext4_fsblk_t group_first_block;
 313
 314         if (ext4_has_feature_flex_bg(sb)) {
 315                 /* with FLEX_BG, the inode/block bitmaps and itable
 316                  * blocks may not be in the group at all
 317                  * so the bitmap validation will be skipped for those groups
 318                  * or it has to also read the block group where the bitmaps
 319                  * are located to verify they are set.
 320                  */
 321                 return 0;
 322         }
 323         group_first_block = ext4_group_first_block_no(sb, block_group);
 324
 325         /* check whether block bitmap block number is set */
 326         blk = ext4_block_bitmap(sb, desc);
 327         offset = blk - group_first_block;
 328         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 329             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 330                 /* bad block bitmap */
 331                 return blk;
 332
 333         /* check whether the inode bitmap block number is set */
 334         blk = ext4_inode_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 table block number is set */
 342         blk = ext4_inode_table(sb, desc);
 343         offset = blk - group_first_block;
 344         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 345             EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
 346                 return blk;
 347         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 348                         EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
 349                         EXT4_B2C(sbi, offset));
 350         if (next_zero_bit <
 351             EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
 352                 /* bad bitmap for inode tables */
 353                 return blk;
 354         return 0;
 355 }
 356
 357 static int ext4_validate_block_bitmap(struct super_block *sb,
 358                                       struct ext4_group_desc *desc,
 359                                       ext4_group_t block_group,
 360                                       struct buffer_head *bh)
 361 {
 362         ext4_fsblk_t    blk;
 363         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
 364
 365         if (buffer_verified(bh))
 366                 return 0;
 367         if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 368                 return -EFSCORRUPTED;
 369
 370         ext4_lock_group(sb, block_group);
 371         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 372                         desc, bh))) {
 373                 ext4_unlock_group(sb, block_group);
 374                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 375                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 376                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 377                 return -EFSBADCRC;
 378         }
 379         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
 380         if (unlikely(blk != 0)) {
 381                 ext4_unlock_group(sb, block_group);
 382                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
 383                            block_group, blk);
 384                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 385                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 386                 return -EFSCORRUPTED;
 387         }
 388         set_buffer_verified(bh);
 389         ext4_unlock_group(sb, block_group);
 390         return 0;
 391 }
 392
 393 /**
 394  * ext4_read_block_bitmap_nowait()
 395  * @sb:                 super block
 396  * @block_group:        given block group
 397  *
 398  * Read the bitmap for a given block_group,and validate the
 399  * bits for block/inode/inode tables are set in the bitmaps
 400  *
 401  * Return buffer_head on success or NULL in case of failure.
 402  */
 403 struct buffer_head *
 404 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 405 {
 406         struct ext4_group_desc *desc;
 407         struct ext4_sb_info *sbi = EXT4_SB(sb);
 408         struct buffer_head *bh;
 409         ext4_fsblk_t bitmap_blk;
 410         int err;
 411
 412         desc = ext4_get_group_desc(sb, block_group, NULL);
 413         if (!desc)
 414                 return ERR_PTR(-EFSCORRUPTED);
 415         bitmap_blk = ext4_block_bitmap(sb, desc);
 416         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 417             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 418                 ext4_error(sb, "Invalid block bitmap block %llu in "
 419                            "block_group %u", bitmap_blk, block_group);
 420                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 421                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 422                 return ERR_PTR(-EFSCORRUPTED);
 423         }
 424         bh = sb_getblk(sb, bitmap_blk);
 425         if (unlikely(!bh)) {
 426                 ext4_error(sb, "Cannot get buffer for block bitmap - "
 427                            "block_group = %u, block_bitmap = %llu",
 428                            block_group, bitmap_blk);
 429                 return ERR_PTR(-ENOMEM);
 430         }
 431
 432         if (bitmap_uptodate(bh))
 433                 goto verify;
 434
 435         lock_buffer(bh);
 436         if (bitmap_uptodate(bh)) {
 437                 unlock_buffer(bh);
 438                 goto verify;
 439         }
 440         ext4_lock_group(sb, block_group);
 441         if (ext4_has_group_desc_csum(sb) &&
 442             (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 443                 if (block_group == 0) {
 444                         ext4_unlock_group(sb, block_group);
 445                         unlock_buffer(bh);
 446                         ext4_error(sb, "Block bitmap for bg 0 marked "
 447                                    "uninitialized");
 448                         err = -EFSCORRUPTED;
 449                         goto out;
 450                 }
 451                 err = ext4_init_block_bitmap(sb, bh, block_group, desc);
 452                 set_bitmap_uptodate(bh);
 453                 set_buffer_uptodate(bh);
 454                 set_buffer_verified(bh);
 455                 ext4_unlock_group(sb, block_group);
 456                 unlock_buffer(bh);
 457                 if (err) {
 458                         ext4_error(sb, "Failed to init block bitmap for group "
 459                                    "%u: %d", block_group, err);
 460                         goto out;
 461                 }
 462                 goto verify;
 463         }
 464         ext4_unlock_group(sb, block_group);
 465         if (buffer_uptodate(bh)) {
 466                 /*
 467                  * if not uninit if bh is uptodate,
 468                  * bitmap is also uptodate
 469                  */
 470                 set_bitmap_uptodate(bh);
 471                 unlock_buffer(bh);
 472                 goto verify;
 473         }
 474         /*
 475          * submit the buffer_head for reading
 476          */
 477         set_buffer_new(bh);
 478         trace_ext4_read_block_bitmap_load(sb, block_group);
 479         bh->b_end_io = ext4_end_bitmap_read;
 480         get_bh(bh);
 481         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 482         return bh;
 483 verify:
 484         err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
 485         if (err)
 486                 goto out;
 487         return bh;
 488 out:
 489         put_bh(bh);
 490         return ERR_PTR(err);
 491 }
 492
 493 /* Returns 0 on success, 1 on error */
 494 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
 495                            struct buffer_head *bh)
 496 {
 497         struct ext4_group_desc *desc;
 498
 499         if (!buffer_new(bh))
 500                 return 0;
 501         desc = ext4_get_group_desc(sb, block_group, NULL);
 502         if (!desc)
 503                 return -EFSCORRUPTED;
 504         wait_on_buffer(bh);
 505         if (!buffer_uptodate(bh)) {
 506                 ext4_error(sb, "Cannot read block bitmap - "
 507                            "block_group = %u, block_bitmap = %llu",
 508                            block_group, (unsigned long long) bh->b_blocknr);
 509                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 510                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 511                 return -EIO;
 512         }
 513         clear_buffer_new(bh);
 514         /* Panic or remount fs read-only if block bitmap is invalid */
 515         return ext4_validate_block_bitmap(sb, desc, block_group, bh);
 516 }
 517
 518 struct buffer_head *
 519 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 520 {
 521         struct buffer_head *bh;
 522         int err;
 523
 524         bh = ext4_read_block_bitmap_nowait(sb, block_group);
 525         if (IS_ERR(bh))
 526                 return bh;
 527         err = ext4_wait_block_bitmap(sb, block_group, bh);
 528         if (err) {
 529                 put_bh(bh);
 530                 return ERR_PTR(err);
 531         }
 532         return bh;
 533 }
 534
 535 /**
 536  * ext4_has_free_clusters()
 537  * @sbi:        in-core super block structure.
 538  * @nclusters:  number of needed blocks
 539  * @flags:      flags from ext4_mb_new_blocks()
 540  *
 541  * Check if filesystem has nclusters free & available for allocation.
 542  * On success return 1, return 0 on failure.
 543  */
 544 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 545                                   s64 nclusters, unsigned int flags)
 546 {
 547         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 548         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 549         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 550
 551         free_clusters  = percpu_counter_read_positive(fcc);
 552         dirty_clusters = percpu_counter_read_positive(dcc);
 553         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 554
 555         /*
 556          * r_blocks_count should always be multiple of the cluster ratio so
 557          * we are safe to do a plane bit shift only.
 558          */
 559         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 560               resv_clusters;
 561
 562         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 563                                         EXT4_FREECLUSTERS_WATERMARK) {
 564                 free_clusters  = percpu_counter_sum_positive(fcc);
 565                 dirty_clusters = percpu_counter_sum_positive(dcc);
 566         }
 567         /* Check whether we have space after accounting for current
 568          * dirty clusters & root reserved clusters.
 569          */
 570         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 571                 return 1;
 572
 573         /* Hm, nope.  Are (enough) root reserved clusters available? */
 574         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 575             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 576             capable(CAP_SYS_RESOURCE) ||
 577             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 578
 579                 if (free_clusters >= (nclusters + dirty_clusters +
 580                                       resv_clusters))
 581                         return 1;
 582         }
 583         /* No free blocks. Let's see if we can dip into reserved pool */
 584         if (flags & EXT4_MB_USE_RESERVED) {
 585                 if (free_clusters >= (nclusters + dirty_clusters))
 586                         return 1;
 587         }
 588
 589         return 0;
 590 }
 591
 592 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 593                              s64 nclusters, unsigned int flags)
 594 {
 595         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 596                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 597                 return 0;
 598         } else
 599                 return -ENOSPC;
 600 }
 601
 602 /**
 603  * ext4_should_retry_alloc()
 604  * @sb:                 super block
 605  * @retries             number of attemps has been made
 606  *
 607  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 608  * it is profitable to retry the operation, this function will wait
 609  * for the current or committing transaction to complete, and then
 610  * return TRUE.  We will only retry once.
 611  */
 612 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 613 {
 614         if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
 615             (*retries)++ > 1 ||
 616             !EXT4_SB(sb)->s_journal)
 617                 return 0;
 618
 619         smp_mb();
 620         if (EXT4_SB(sb)->s_mb_free_pending == 0)
 621                 return 0;
 622
 623         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 624         jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 625         return 1;
 626 }
 627
 628 /*
 629  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 630  *
 631  * @handle:             handle to this transaction
 632  * @inode:              file inode
 633  * @goal:               given target block(filesystem wide)
 634  * @count:              pointer to total number of clusters needed
 635  * @errp:               error code
 636  *
 637  * Return 1st allocated block number on success, *count stores total account
 638  * error stores in errp pointer
 639  */
 640 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 641                                   ext4_fsblk_t goal, unsigned int flags,
 642                                   unsigned long *count, int *errp)
 643 {
 644         struct ext4_allocation_request ar;
 645         ext4_fsblk_t ret;
 646
 647         memset(&ar, 0, sizeof(ar));
 648         /* Fill with neighbour allocated blocks */
 649         ar.inode = inode;
 650         ar.goal = goal;
 651         ar.len = count ? *count : 1;
 652         ar.flags = flags;
 653
 654         ret = ext4_mb_new_blocks(handle, &ar, errp);
 655         if (count)
 656                 *count = ar.len;
 657         /*
 658          * Account for the allocated meta blocks.  We will never
 659          * fail EDQUOT for metdata, but we do account for it.
 660          */
 661         if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
 662                 dquot_alloc_block_nofail(inode,
 663                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 664         }
 665         return ret;
 666 }
 667
 668 /**
 669  * ext4_count_free_clusters() -- count filesystem free clusters
 670  * @sb:         superblock
 671  *
 672  * Adds up the number of free clusters from each block group.
 673  */
 674 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 675 {
 676         ext4_fsblk_t desc_count;
 677         struct ext4_group_desc *gdp;
 678         ext4_group_t i;
 679         ext4_group_t ngroups = ext4_get_groups_count(sb);
 680         struct ext4_group_info *grp;
 681 #ifdef EXT4FS_DEBUG
 682         struct ext4_super_block *es;
 683         ext4_fsblk_t bitmap_count;
 684         unsigned int x;
 685         struct buffer_head *bitmap_bh = NULL;
 686
 687         es = EXT4_SB(sb)->s_es;
 688         desc_count = 0;
 689         bitmap_count = 0;
 690         gdp = NULL;
 691
 692         for (i = 0; i < ngroups; i++) {
 693                 gdp = ext4_get_group_desc(sb, i, NULL);
 694                 if (!gdp)
 695                         continue;
 696                 grp = NULL;
 697                 if (EXT4_SB(sb)->s_group_info)
 698                         grp = ext4_get_group_info(sb, i);
 699                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 700                         desc_count += ext4_free_group_clusters(sb, gdp);
 701                 brelse(bitmap_bh);
 702                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 703                 if (IS_ERR(bitmap_bh)) {
 704                         bitmap_bh = NULL;
 705                         continue;
 706                 }
 707
 708                 x = ext4_count_free(bitmap_bh->b_data,
 709                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
 710                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 711                         i, ext4_free_group_clusters(sb, gdp), x);
 712                 bitmap_count += x;
 713         }
 714         brelse(bitmap_bh);
 715         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 716                ", computed = %llu, %llu\n",
 717                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 718                desc_count, bitmap_count);
 719         return bitmap_count;
 720 #else
 721         desc_count = 0;
 722         for (i = 0; i < ngroups; i++) {
 723                 gdp = ext4_get_group_desc(sb, i, NULL);
 724                 if (!gdp)
 725                         continue;
 726                 grp = NULL;
 727                 if (EXT4_SB(sb)->s_group_info)
 728                         grp = ext4_get_group_info(sb, i);
 729                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 730                         desc_count += ext4_free_group_clusters(sb, gdp);
 731         }
 732
 733         return desc_count;
 734 #endif
 735 }
 736
 737 static inline int test_root(ext4_group_t a, int b)
 738 {
 739         while (1) {
 740                 if (a < b)
 741                         return 0;
 742                 if (a == b)
 743                         return 1;
 744                 if ((a % b) != 0)
 745                         return 0;
 746                 a = a / b;
 747         }
 748 }
 749
 750 /**
 751  *      ext4_bg_has_super - number of blocks used by the superblock in group
 752  *      @sb: superblock for filesystem
 753  *      @group: group number to check
 754  *
 755  *      Return the number of blocks used by the superblock (primary or backup)
 756  *      in this group.  Currently this will be only 0 or 1.
 757  */
 758 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 759 {
 760         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 761
 762         if (group == 0)
 763                 return 1;
 764         if (ext4_has_feature_sparse_super2(sb)) {
 765                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
 766                     group == le32_to_cpu(es->s_backup_bgs[1]))
 767                         return 1;
 768                 return 0;
 769         }
 770         if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
 771                 return 1;
 772         if (!(group & 1))
 773                 return 0;
 774         if (test_root(group, 3) || (test_root(group, 5)) ||
 775             test_root(group, 7))
 776                 return 1;
 777
 778         return 0;
 779 }
 780
 781 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 782                                         ext4_group_t group)
 783 {
 784         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 785         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 786         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 787
 788         if (group == first || group == first + 1 || group == last)
 789                 return 1;
 790         return 0;
 791 }
 792
 793 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 794                                         ext4_group_t group)
 795 {
 796         if (!ext4_bg_has_super(sb, group))
 797                 return 0;
 798
 799         if (ext4_has_feature_meta_bg(sb))
 800                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 801         else
 802                 return EXT4_SB(sb)->s_gdb_count;
 803 }
 804
 805 /**
 806  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 807  *      @sb: superblock for filesystem
 808  *      @group: group number to check
 809  *
 810  *      Return the number of blocks used by the group descriptor table
 811  *      (primary or backup) in this group.  In the future there may be a
 812  *      different number of descriptor blocks in each group.
 813  */
 814 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 815 {
 816         unsigned long first_meta_bg =
 817                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 818         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 819
 820         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
 821                 return ext4_bg_num_gdb_nometa(sb, group);
 822
 823         return ext4_bg_num_gdb_meta(sb,group);
 824
 825 }
 826
 827 /*
 828  * This function returns the number of file system metadata clusters at
 829  * the beginning of a block group, including the reserved gdt blocks.
 830  */
 831 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
 832                                      ext4_group_t block_group)
 833 {
 834         struct ext4_sb_info *sbi = EXT4_SB(sb);
 835         unsigned num;
 836
 837         /* Check for superblock and gdt backups in this group */
 838         num = ext4_bg_has_super(sb, block_group);
 839
 840         if (!ext4_has_feature_meta_bg(sb) ||
 841             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 842                           sbi->s_desc_per_block) {
 843                 if (num) {
 844                         num += ext4_bg_num_gdb(sb, block_group);
 845                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 846                 }
 847         } else { /* For META_BG_BLOCK_GROUPS */
 848                 num += ext4_bg_num_gdb(sb, block_group);
 849         }
 850         return EXT4_NUM_B2C(sbi, num);
 851 }
 852 /**
 853  *      ext4_inode_to_goal_block - return a hint for block allocation
 854  *      @inode: inode for block allocation
 855  *
 856  *      Return the ideal location to start allocating blocks for a
 857  *      newly created inode.
 858  */
 859 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 860 {
 861         struct ext4_inode_info *ei = EXT4_I(inode);
 862         ext4_group_t block_group;
 863         ext4_grpblk_t colour;
 864         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 865         ext4_fsblk_t bg_start;
 866         ext4_fsblk_t last_block;
 867
 868         block_group = ei->i_block_group;
 869         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 870                 /*
 871                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 872                  * block groups per flexgroup, reserve the first block
 873                  * group for directories and special files.  Regular
 874                  * files will start at the second block group.  This
 875                  * tends to speed up directory access and improves
 876                  * fsck times.
 877                  */
 878                 block_group &= ~(flex_size-1);
 879                 if (S_ISREG(inode->i_mode))
 880                         block_group++;
 881         }
 882         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 883         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 884
 885         /*
 886          * If we are doing delayed allocation, we don't need take
 887          * colour into account.
 888          */
 889         if (test_opt(inode->i_sb, DELALLOC))
 890                 return bg_start;
 891
 892         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 893                 colour = (current->pid % 16) *
 894                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 895         else
 896                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 897         return bg_start + colour;
 898 }
 899