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         struct ext4_group_info *grp;
 188
 189         J_ASSERT_BH(bh, buffer_locked(bh));
 190
 191         /* If checksum is bad mark all blocks used to prevent allocation
 192          * essentially implementing a per-group read-only flag. */
 193         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
 194                 grp = ext4_get_group_info(sb, block_group);
 195                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 196                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 197                                            grp->bb_free);
 198                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
 199                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 200                         int count;
 201                         count = ext4_free_inodes_count(sb, gdp);
 202                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 203                                            count);
 204                 }
 205                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 206                 return -EFSBADCRC;
 207         }
 208         memset(bh->b_data, 0, sb->s_blocksize);
 209
 210         bit_max = ext4_num_base_meta_clusters(sb, block_group);
 211         if ((bit_max >> 3) >= bh->b_size)
 212                 return -EFSCORRUPTED;
 213
 214         for (bit = 0; bit < bit_max; bit++)
 215                 ext4_set_bit(bit, bh->b_data);
 216
 217         start = ext4_group_first_block_no(sb, block_group);
 218
 219         /* Set bits for block and inode bitmaps, and inode table */
 220         tmp = ext4_block_bitmap(sb, gdp);
 221         if (ext4_block_in_group(sb, tmp, block_group))
 222                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 223
 224         tmp = ext4_inode_bitmap(sb, gdp);
 225         if (ext4_block_in_group(sb, tmp, block_group))
 226                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 227
 228         tmp = ext4_inode_table(sb, gdp);
 229         for (; tmp < ext4_inode_table(sb, gdp) +
 230                      sbi->s_itb_per_group; tmp++) {
 231                 if (ext4_block_in_group(sb, tmp, block_group))
 232                         ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 233         }
 234
 235         /*
 236          * Also if the number of blocks within the group is less than
 237          * the blocksize * 8 ( which is the size of bitmap ), set rest
 238          * of the block bitmap to 1
 239          */
 240         ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
 241                              sb->s_blocksize * 8, bh->b_data);
 242         return 0;
 243 }
 244
 245 /* Return the number of free blocks in a block group.  It is used when
 246  * the block bitmap is uninitialized, so we can't just count the bits
 247  * in the bitmap. */
 248 unsigned ext4_free_clusters_after_init(struct super_block *sb,
 249                                        ext4_group_t block_group,
 250                                        struct ext4_group_desc *gdp)
 251 {
 252         return num_clusters_in_group(sb, block_group) -
 253                 ext4_num_overhead_clusters(sb, block_group, gdp);
 254 }
 255
 256 /*
 257  * The free blocks are managed by bitmaps.  A file system contains several
 258  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 259  * block for inodes, N blocks for the inode table and data blocks.
 260  *
 261  * The file system contains group descriptors which are located after the
 262  * super block.  Each descriptor contains the number of the bitmap block and
 263  * the free blocks count in the block.  The descriptors are loaded in memory
 264  * when a file system is mounted (see ext4_fill_super).
 265  */
 266
 267 /**
 268  * ext4_get_group_desc() -- load group descriptor from disk
 269  * @sb:                 super block
 270  * @block_group:        given block group
 271  * @bh:                 pointer to the buffer head to store the block
 272  *                      group descriptor
 273  */
 274 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 275                                              ext4_group_t block_group,
 276                                              struct buffer_head **bh)
 277 {
 278         unsigned int group_desc;
 279         unsigned int offset;
 280         ext4_group_t ngroups = ext4_get_groups_count(sb);
 281         struct ext4_group_desc *desc;
 282         struct ext4_sb_info *sbi = EXT4_SB(sb);
 283         struct buffer_head *bh_p;
 284
 285         if (block_group >= ngroups) {
 286                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
 287                            " groups_count = %u", block_group, ngroups);
 288
 289                 return NULL;
 290         }
 291
 292         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 293         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 294         bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
 295         /*
 296          * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
 297          * the pointer being dereferenced won't be dereferenced again. By
 298          * looking at the usage in add_new_gdb() the value isn't modified,
 299          * just the pointer, and so it remains valid.
 300          */
 301         if (!bh_p) {
 302                 ext4_error(sb, "Group descriptor not loaded - "
 303                            "block_group = %u, group_desc = %u, desc = %u",
 304                            block_group, group_desc, offset);
 305                 return NULL;
 306         }
 307
 308         desc = (struct ext4_group_desc *)(
 309                 (__u8 *)bh_p->b_data +
 310                 offset * EXT4_DESC_SIZE(sb));
 311         if (bh)
 312                 *bh = bh_p;
 313         return desc;
 314 }
 315
 316 /*
 317  * Return the block number which was discovered to be invalid, or 0 if
 318  * the block bitmap is valid.
 319  */
 320 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 321                                             struct ext4_group_desc *desc,
 322                                             ext4_group_t block_group,
 323                                             struct buffer_head *bh)
 324 {
 325         struct ext4_sb_info *sbi = EXT4_SB(sb);
 326         ext4_grpblk_t offset;
 327         ext4_grpblk_t next_zero_bit;
 328         ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
 329         ext4_fsblk_t blk;
 330         ext4_fsblk_t group_first_block;
 331
 332         if (ext4_has_feature_flex_bg(sb)) {
 333                 /* with FLEX_BG, the inode/block bitmaps and itable
 334                  * blocks may not be in the group at all
 335                  * so the bitmap validation will be skipped for those groups
 336                  * or it has to also read the block group where the bitmaps
 337                  * are located to verify they are set.
 338                  */
 339                 return 0;
 340         }
 341         group_first_block = ext4_group_first_block_no(sb, block_group);
 342
 343         /* check whether block bitmap block number is set */
 344         blk = ext4_block_bitmap(sb, desc);
 345         offset = blk - group_first_block;
 346         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 347             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 348                 /* bad block bitmap */
 349                 return blk;
 350
 351         /* check whether the inode bitmap block number is set */
 352         blk = ext4_inode_bitmap(sb, desc);
 353         offset = blk - group_first_block;
 354         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 355             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 356                 /* bad block bitmap */
 357                 return blk;
 358
 359         /* check whether the inode table block number is set */
 360         blk = ext4_inode_table(sb, desc);
 361         offset = blk - group_first_block;
 362         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 363             EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
 364                 return blk;
 365         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 366                         EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
 367                         EXT4_B2C(sbi, offset));
 368         if (next_zero_bit <
 369             EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
 370                 /* bad bitmap for inode tables */
 371                 return blk;
 372         return 0;
 373 }
 374
 375 static int ext4_validate_block_bitmap(struct super_block *sb,
 376                                       struct ext4_group_desc *desc,
 377                                       ext4_group_t block_group,
 378                                       struct buffer_head *bh)
 379 {
 380         ext4_fsblk_t    blk;
 381         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
 382         struct ext4_sb_info *sbi = EXT4_SB(sb);
 383
 384         if (buffer_verified(bh))
 385                 return 0;
 386         if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 387                 return -EFSCORRUPTED;
 388
 389         ext4_lock_group(sb, block_group);
 390         if (buffer_verified(bh))
 391                 goto verified;
 392         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 393                         desc, bh))) {
 394                 ext4_unlock_group(sb, block_group);
 395                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 396                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 397                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 398                                            grp->bb_free);
 399                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
 400                 return -EFSBADCRC;
 401         }
 402         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
 403         if (unlikely(blk != 0)) {
 404                 ext4_unlock_group(sb, block_group);
 405                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
 406                            block_group, blk);
 407                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 408                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 409                                            grp->bb_free);
 410                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
 411                 return -EFSCORRUPTED;
 412         }
 413         set_buffer_verified(bh);
 414 verified:
 415         ext4_unlock_group(sb, block_group);
 416         return 0;
 417 }
 418
 419 /**
 420  * ext4_read_block_bitmap_nowait()
 421  * @sb:                 super block
 422  * @block_group:        given block group
 423  *
 424  * Read the bitmap for a given block_group,and validate the
 425  * bits for block/inode/inode tables are set in the bitmaps
 426  *
 427  * Return buffer_head on success or NULL in case of failure.
 428  */
 429 struct buffer_head *
 430 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 431 {
 432         struct ext4_group_desc *desc;
 433         struct ext4_sb_info *sbi = EXT4_SB(sb);
 434         struct buffer_head *bh;
 435         ext4_fsblk_t bitmap_blk;
 436         int err;
 437
 438         desc = ext4_get_group_desc(sb, block_group, NULL);
 439         if (!desc)
 440                 return ERR_PTR(-EFSCORRUPTED);
 441         bitmap_blk = ext4_block_bitmap(sb, desc);
 442         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 443             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 444                 ext4_error(sb, "Invalid block bitmap block %llu in "
 445                            "block_group %u", bitmap_blk, block_group);
 446                 return ERR_PTR(-EFSCORRUPTED);
 447         }
 448         bh = sb_getblk(sb, bitmap_blk);
 449         if (unlikely(!bh)) {
 450                 ext4_error(sb, "Cannot get buffer for block bitmap - "
 451                            "block_group = %u, block_bitmap = %llu",
 452                            block_group, bitmap_blk);
 453                 return ERR_PTR(-ENOMEM);
 454         }
 455
 456         if (bitmap_uptodate(bh))
 457                 goto verify;
 458
 459         lock_buffer(bh);
 460         if (bitmap_uptodate(bh)) {
 461                 unlock_buffer(bh);
 462                 goto verify;
 463         }
 464         ext4_lock_group(sb, block_group);
 465         if (ext4_has_group_desc_csum(sb) &&
 466             (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 467                 if (block_group == 0) {
 468                         ext4_unlock_group(sb, block_group);
 469                         unlock_buffer(bh);
 470                         ext4_error(sb, "Block bitmap for bg 0 marked "
 471                                    "uninitialized");
 472                         err = -EFSCORRUPTED;
 473                         goto out;
 474                 }
 475                 err = ext4_init_block_bitmap(sb, bh, block_group, desc);
 476                 set_bitmap_uptodate(bh);
 477                 set_buffer_uptodate(bh);
 478                 set_buffer_verified(bh);
 479                 ext4_unlock_group(sb, block_group);
 480                 unlock_buffer(bh);
 481                 if (err) {
 482                         ext4_error(sb, "Failed to init block bitmap for group "
 483                                    "%u: %d", block_group, err);
 484                         goto out;
 485                 }
 486                 goto verify;
 487         }
 488         ext4_unlock_group(sb, block_group);
 489         if (buffer_uptodate(bh)) {
 490                 /*
 491                  * if not uninit if bh is uptodate,
 492                  * bitmap is also uptodate
 493                  */
 494                 set_bitmap_uptodate(bh);
 495                 unlock_buffer(bh);
 496                 goto verify;
 497         }
 498         /*
 499          * submit the buffer_head for reading
 500          */
 501         set_buffer_new(bh);
 502         trace_ext4_read_block_bitmap_load(sb, block_group);
 503         bh->b_end_io = ext4_end_bitmap_read;
 504         get_bh(bh);
 505         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 506         return bh;
 507 verify:
 508         err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
 509         if (err)
 510                 goto out;
 511         return bh;
 512 out:
 513         put_bh(bh);
 514         return ERR_PTR(err);
 515 }
 516
 517 /* Returns 0 on success, 1 on error */
 518 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
 519                            struct buffer_head *bh)
 520 {
 521         struct ext4_group_desc *desc;
 522
 523         if (!buffer_new(bh))
 524                 return 0;
 525         desc = ext4_get_group_desc(sb, block_group, NULL);
 526         if (!desc)
 527                 return -EFSCORRUPTED;
 528         wait_on_buffer(bh);
 529         if (!buffer_uptodate(bh)) {
 530                 ext4_error(sb, "Cannot read block bitmap - "
 531                            "block_group = %u, block_bitmap = %llu",
 532                            block_group, (unsigned long long) bh->b_blocknr);
 533                 return -EIO;
 534         }
 535         clear_buffer_new(bh);
 536         /* Panic or remount fs read-only if block bitmap is invalid */
 537         return ext4_validate_block_bitmap(sb, desc, block_group, bh);
 538 }
 539
 540 struct buffer_head *
 541 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 542 {
 543         struct buffer_head *bh;
 544         int err;
 545
 546         bh = ext4_read_block_bitmap_nowait(sb, block_group);
 547         if (IS_ERR(bh))
 548                 return bh;
 549         err = ext4_wait_block_bitmap(sb, block_group, bh);
 550         if (err) {
 551                 put_bh(bh);
 552                 return ERR_PTR(err);
 553         }
 554         return bh;
 555 }
 556
 557 /**
 558  * ext4_has_free_clusters()
 559  * @sbi:        in-core super block structure.
 560  * @nclusters:  number of needed blocks
 561  * @flags:      flags from ext4_mb_new_blocks()
 562  *
 563  * Check if filesystem has nclusters free & available for allocation.
 564  * On success return 1, return 0 on failure.
 565  */
 566 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 567                                   s64 nclusters, unsigned int flags)
 568 {
 569         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 570         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 571         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 572
 573         free_clusters  = percpu_counter_read_positive(fcc);
 574         dirty_clusters = percpu_counter_read_positive(dcc);
 575         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 576
 577         /*
 578          * r_blocks_count should always be multiple of the cluster ratio so
 579          * we are safe to do a plane bit shift only.
 580          */
 581         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 582               resv_clusters;
 583
 584         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 585                                         EXT4_FREECLUSTERS_WATERMARK) {
 586                 free_clusters  = percpu_counter_sum_positive(fcc);
 587                 dirty_clusters = percpu_counter_sum_positive(dcc);
 588         }
 589         /* Check whether we have space after accounting for current
 590          * dirty clusters & root reserved clusters.
 591          */
 592         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 593                 return 1;
 594
 595         /* Hm, nope.  Are (enough) root reserved clusters available? */
 596         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 597             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 598             capable(CAP_SYS_RESOURCE) ||
 599             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 600
 601                 if (free_clusters >= (nclusters + dirty_clusters +
 602                                       resv_clusters))
 603                         return 1;
 604         }
 605         /* No free blocks. Let's see if we can dip into reserved pool */
 606         if (flags & EXT4_MB_USE_RESERVED) {
 607                 if (free_clusters >= (nclusters + dirty_clusters))
 608                         return 1;
 609         }
 610
 611         return 0;
 612 }
 613
 614 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 615                              s64 nclusters, unsigned int flags)
 616 {
 617         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 618                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 619                 return 0;
 620         } else
 621                 return -ENOSPC;
 622 }
 623
 624 /**
 625  * ext4_should_retry_alloc()
 626  * @sb:                 super block
 627  * @retries             number of attemps has been made
 628  *
 629  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
 630  * it is profitable to retry the operation, this function will wait
 631  * for the current or committing transaction to complete, and then
 632  * return TRUE.
 633  *
 634  * if the total number of retries exceed three times, return FALSE.
 635  */
 636 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 637 {
 638         if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
 639             (*retries)++ > 3 ||
 640             !EXT4_SB(sb)->s_journal)
 641                 return 0;
 642
 643         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 644
 645         smp_mb();
 646         if (EXT4_SB(sb)->s_mb_free_pending)
 647                 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
 648         return 1;
 649 }
 650
 651 /*
 652  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 653  *
 654  * @handle:             handle to this transaction
 655  * @inode:              file inode
 656  * @goal:               given target block(filesystem wide)
 657  * @count:              pointer to total number of clusters needed
 658  * @errp:               error code
 659  *
 660  * Return 1st allocated block number on success, *count stores total account
 661  * error stores in errp pointer
 662  */
 663 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 664                                   ext4_fsblk_t goal, unsigned int flags,
 665                                   unsigned long *count, int *errp)
 666 {
 667         struct ext4_allocation_request ar;
 668         ext4_fsblk_t ret;
 669
 670         memset(&ar, 0, sizeof(ar));
 671         /* Fill with neighbour allocated blocks */
 672         ar.inode = inode;
 673         ar.goal = goal;
 674         ar.len = count ? *count : 1;
 675         ar.flags = flags;
 676
 677         ret = ext4_mb_new_blocks(handle, &ar, errp);
 678         if (count)
 679                 *count = ar.len;
 680         /*
 681          * Account for the allocated meta blocks.  We will never
 682          * fail EDQUOT for metdata, but we do account for it.
 683          */
 684         if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
 685                 dquot_alloc_block_nofail(inode,
 686                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 687         }
 688         return ret;
 689 }
 690
 691 /**
 692  * ext4_count_free_clusters() -- count filesystem free clusters
 693  * @sb:         superblock
 694  *
 695  * Adds up the number of free clusters from each block group.
 696  */
 697 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 698 {
 699         ext4_fsblk_t desc_count;
 700         struct ext4_group_desc *gdp;
 701         ext4_group_t i;
 702         ext4_group_t ngroups = ext4_get_groups_count(sb);
 703         struct ext4_group_info *grp;
 704 #ifdef EXT4FS_DEBUG
 705         struct ext4_super_block *es;
 706         ext4_fsblk_t bitmap_count;
 707         unsigned int x;
 708         struct buffer_head *bitmap_bh = NULL;
 709
 710         es = EXT4_SB(sb)->s_es;
 711         desc_count = 0;
 712         bitmap_count = 0;
 713         gdp = NULL;
 714
 715         for (i = 0; i < ngroups; i++) {
 716                 gdp = ext4_get_group_desc(sb, i, NULL);
 717                 if (!gdp)
 718                         continue;
 719                 grp = NULL;
 720                 if (EXT4_SB(sb)->s_group_info)
 721                         grp = ext4_get_group_info(sb, i);
 722                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 723                         desc_count += ext4_free_group_clusters(sb, gdp);
 724                 brelse(bitmap_bh);
 725                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 726                 if (IS_ERR(bitmap_bh)) {
 727                         bitmap_bh = NULL;
 728                         continue;
 729                 }
 730
 731                 x = ext4_count_free(bitmap_bh->b_data,
 732                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
 733                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 734                         i, ext4_free_group_clusters(sb, gdp), x);
 735                 bitmap_count += x;
 736         }
 737         brelse(bitmap_bh);
 738         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 739                ", computed = %llu, %llu\n",
 740                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 741                desc_count, bitmap_count);
 742         return bitmap_count;
 743 #else
 744         desc_count = 0;
 745         for (i = 0; i < ngroups; i++) {
 746                 gdp = ext4_get_group_desc(sb, i, NULL);
 747                 if (!gdp)
 748                         continue;
 749                 grp = NULL;
 750                 if (EXT4_SB(sb)->s_group_info)
 751                         grp = ext4_get_group_info(sb, i);
 752                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 753                         desc_count += ext4_free_group_clusters(sb, gdp);
 754         }
 755
 756         return desc_count;
 757 #endif
 758 }
 759
 760 static inline int test_root(ext4_group_t a, int b)
 761 {
 762         while (1) {
 763                 if (a < b)
 764                         return 0;
 765                 if (a == b)
 766                         return 1;
 767                 if ((a % b) != 0)
 768                         return 0;
 769                 a = a / b;
 770         }
 771 }
 772
 773 /**
 774  *      ext4_bg_has_super - number of blocks used by the superblock in group
 775  *      @sb: superblock for filesystem
 776  *      @group: group number to check
 777  *
 778  *      Return the number of blocks used by the superblock (primary or backup)
 779  *      in this group.  Currently this will be only 0 or 1.
 780  */
 781 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 782 {
 783         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 784
 785         if (group == 0)
 786                 return 1;
 787         if (ext4_has_feature_sparse_super2(sb)) {
 788                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
 789                     group == le32_to_cpu(es->s_backup_bgs[1]))
 790                         return 1;
 791                 return 0;
 792         }
 793         if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
 794                 return 1;
 795         if (!(group & 1))
 796                 return 0;
 797         if (test_root(group, 3) || (test_root(group, 5)) ||
 798             test_root(group, 7))
 799                 return 1;
 800
 801         return 0;
 802 }
 803
 804 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 805                                         ext4_group_t group)
 806 {
 807         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 808         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 809         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 810
 811         if (group == first || group == first + 1 || group == last)
 812                 return 1;
 813         return 0;
 814 }
 815
 816 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 817                                         ext4_group_t group)
 818 {
 819         if (!ext4_bg_has_super(sb, group))
 820                 return 0;
 821
 822         if (ext4_has_feature_meta_bg(sb))
 823                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 824         else
 825                 return EXT4_SB(sb)->s_gdb_count;
 826 }
 827
 828 /**
 829  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 830  *      @sb: superblock for filesystem
 831  *      @group: group number to check
 832  *
 833  *      Return the number of blocks used by the group descriptor table
 834  *      (primary or backup) in this group.  In the future there may be a
 835  *      different number of descriptor blocks in each group.
 836  */
 837 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 838 {
 839         unsigned long first_meta_bg =
 840                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 841         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 842
 843         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
 844                 return ext4_bg_num_gdb_nometa(sb, group);
 845
 846         return ext4_bg_num_gdb_meta(sb,group);
 847
 848 }
 849
 850 /*
 851  * This function returns the number of file system metadata clusters at
 852  * the beginning of a block group, including the reserved gdt blocks.
 853  */
 854 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
 855                                      ext4_group_t block_group)
 856 {
 857         struct ext4_sb_info *sbi = EXT4_SB(sb);
 858         unsigned num;
 859
 860         /* Check for superblock and gdt backups in this group */
 861         num = ext4_bg_has_super(sb, block_group);
 862
 863         if (!ext4_has_feature_meta_bg(sb) ||
 864             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 865                           sbi->s_desc_per_block) {
 866                 if (num) {
 867                         num += ext4_bg_num_gdb(sb, block_group);
 868                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 869                 }
 870         } else { /* For META_BG_BLOCK_GROUPS */
 871                 num += ext4_bg_num_gdb(sb, block_group);
 872         }
 873         return EXT4_NUM_B2C(sbi, num);
 874 }
 875 /**
 876  *      ext4_inode_to_goal_block - return a hint for block allocation
 877  *      @inode: inode for block allocation
 878  *
 879  *      Return the ideal location to start allocating blocks for a
 880  *      newly created inode.
 881  */
 882 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 883 {
 884         struct ext4_inode_info *ei = EXT4_I(inode);
 885         ext4_group_t block_group;
 886         ext4_grpblk_t colour;
 887         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 888         ext4_fsblk_t bg_start;
 889         ext4_fsblk_t last_block;
 890
 891         block_group = ei->i_block_group;
 892         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 893                 /*
 894                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 895                  * block groups per flexgroup, reserve the first block
 896                  * group for directories and special files.  Regular
 897                  * files will start at the second block group.  This
 898                  * tends to speed up directory access and improves
 899                  * fsck times.
 900                  */
 901                 block_group &= ~(flex_size-1);
 902                 if (S_ISREG(inode->i_mode))
 903                         block_group++;
 904         }
 905         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 906         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 907
 908         /*
 909          * If we are doing delayed allocation, we don't need take
 910          * colour into account.
 911          */
 912         if (test_opt(inode->i_sb, DELALLOC))
 913                 return bg_start;
 914
 915         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 916                 colour = (current->pid % 16) *
 917                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 918         else
 919                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 920         return bg_start + colour;
 921 }
 922