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