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         ASSERT(buffer_locked(bh));
 189
 190         /* If checksum is bad mark all blocks used to prevent allocation
 191          * essentially implementing a per-group read-only flag. */
 192         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
 193                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 194                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT |
 195                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 196                 return -EFSBADCRC;
 197         }
 198         memset(bh->b_data, 0, sb->s_blocksize);
 199
 200         bit_max = ext4_num_base_meta_clusters(sb, block_group);
 201         if ((bit_max >> 3) >= bh->b_size)
 202                 return -EFSCORRUPTED;
 203
 204         for (bit = 0; bit < bit_max; bit++)
 205                 ext4_set_bit(bit, bh->b_data);
 206
 207         start = ext4_group_first_block_no(sb, block_group);
 208
 209         /* Set bits for block and inode bitmaps, and inode table */
 210         tmp = ext4_block_bitmap(sb, gdp);
 211         if (ext4_block_in_group(sb, tmp, block_group))
 212                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 213
 214         tmp = ext4_inode_bitmap(sb, gdp);
 215         if (ext4_block_in_group(sb, tmp, block_group))
 216                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 217
 218         tmp = ext4_inode_table(sb, gdp);
 219         for (; tmp < ext4_inode_table(sb, gdp) +
 220                      sbi->s_itb_per_group; tmp++) {
 221                 if (ext4_block_in_group(sb, tmp, block_group))
 222                         ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
 223         }
 224
 225         /*
 226          * Also if the number of blocks within the group is less than
 227          * the blocksize * 8 ( which is the size of bitmap ), set rest
 228          * of the block bitmap to 1
 229          */
 230         ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
 231                              sb->s_blocksize * 8, bh->b_data);
 232         return 0;
 233 }
 234
 235 /* Return the number of free blocks in a block group.  It is used when
 236  * the block bitmap is uninitialized, so we can't just count the bits
 237  * in the bitmap. */
 238 unsigned ext4_free_clusters_after_init(struct super_block *sb,
 239                                        ext4_group_t block_group,
 240                                        struct ext4_group_desc *gdp)
 241 {
 242         return num_clusters_in_group(sb, block_group) -
 243                 ext4_num_overhead_clusters(sb, block_group, gdp);
 244 }
 245
 246 /*
 247  * The free blocks are managed by bitmaps.  A file system contains several
 248  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
 249  * block for inodes, N blocks for the inode table and data blocks.
 250  *
 251  * The file system contains group descriptors which are located after the
 252  * super block.  Each descriptor contains the number of the bitmap block and
 253  * the free blocks count in the block.  The descriptors are loaded in memory
 254  * when a file system is mounted (see ext4_fill_super).
 255  */
 256
 257 /**
 258  * ext4_get_group_desc() -- load group descriptor from disk
 259  * @sb:                 super block
 260  * @block_group:        given block group
 261  * @bh:                 pointer to the buffer head to store the block
 262  *                      group descriptor
 263  */
 264 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
 265                                              ext4_group_t block_group,
 266                                              struct buffer_head **bh)
 267 {
 268         unsigned int group_desc;
 269         unsigned int offset;
 270         ext4_group_t ngroups = ext4_get_groups_count(sb);
 271         struct ext4_group_desc *desc;
 272         struct ext4_sb_info *sbi = EXT4_SB(sb);
 273         struct buffer_head *bh_p;
 274
 275         if (block_group >= ngroups) {
 276                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
 277                            " groups_count = %u", block_group, ngroups);
 278
 279                 return NULL;
 280         }
 281
 282         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
 283         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 284         bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
 285         /*
 286          * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
 287          * the pointer being dereferenced won't be dereferenced again. By
 288          * looking at the usage in add_new_gdb() the value isn't modified,
 289          * just the pointer, and so it remains valid.
 290          */
 291         if (!bh_p) {
 292                 ext4_error(sb, "Group descriptor not loaded - "
 293                            "block_group = %u, group_desc = %u, desc = %u",
 294                            block_group, group_desc, offset);
 295                 return NULL;
 296         }
 297
 298         desc = (struct ext4_group_desc *)(
 299                 (__u8 *)bh_p->b_data +
 300                 offset * EXT4_DESC_SIZE(sb));
 301         if (bh)
 302                 *bh = bh_p;
 303         return desc;
 304 }
 305
 306 static ext4_fsblk_t ext4_valid_block_bitmap_padding(struct super_block *sb,
 307                                                     ext4_group_t block_group,
 308                                                     struct buffer_head *bh)
 309 {
 310         ext4_grpblk_t next_zero_bit;
 311         unsigned long bitmap_size = sb->s_blocksize * 8;
 312         unsigned int offset = num_clusters_in_group(sb, block_group);
 313
 314         if (bitmap_size <= offset)
 315                 return 0;
 316
 317         next_zero_bit = ext4_find_next_zero_bit(bh->b_data, bitmap_size, offset);
 318
 319         return (next_zero_bit < bitmap_size ? next_zero_bit : 0);
 320 }
 321
 322 struct ext4_group_info *ext4_get_group_info(struct super_block *sb,
 323                                             ext4_group_t group)
 324 {
 325         struct ext4_group_info **grp_info;
 326         long indexv, indexh;
 327
 328         if (unlikely(group >= EXT4_SB(sb)->s_groups_count))
 329                 return NULL;
 330         indexv = group >> (EXT4_DESC_PER_BLOCK_BITS(sb));
 331         indexh = group & ((EXT4_DESC_PER_BLOCK(sb)) - 1);
 332         grp_info = sbi_array_rcu_deref(EXT4_SB(sb), s_group_info, indexv);
 333         return grp_info[indexh];
 334 }
 335
 336 /*
 337  * Return the block number which was discovered to be invalid, or 0 if
 338  * the block bitmap is valid.
 339  */
 340 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 341                                             struct ext4_group_desc *desc,
 342                                             ext4_group_t block_group,
 343                                             struct buffer_head *bh)
 344 {
 345         struct ext4_sb_info *sbi = EXT4_SB(sb);
 346         ext4_grpblk_t offset;
 347         ext4_grpblk_t next_zero_bit;
 348         ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
 349         ext4_fsblk_t blk;
 350         ext4_fsblk_t group_first_block;
 351
 352         if (ext4_has_feature_flex_bg(sb)) {
 353                 /* with FLEX_BG, the inode/block bitmaps and itable
 354                  * blocks may not be in the group at all
 355                  * so the bitmap validation will be skipped for those groups
 356                  * or it has to also read the block group where the bitmaps
 357                  * are located to verify they are set.
 358                  */
 359                 return 0;
 360         }
 361         group_first_block = ext4_group_first_block_no(sb, block_group);
 362
 363         /* check whether block bitmap block number is set */
 364         blk = ext4_block_bitmap(sb, desc);
 365         offset = blk - group_first_block;
 366         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 367             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 368                 /* bad block bitmap */
 369                 return blk;
 370
 371         /* check whether the inode bitmap block number is set */
 372         blk = ext4_inode_bitmap(sb, desc);
 373         offset = blk - group_first_block;
 374         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 375             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 376                 /* bad block bitmap */
 377                 return blk;
 378
 379         /* check whether the inode table block number is set */
 380         blk = ext4_inode_table(sb, desc);
 381         offset = blk - group_first_block;
 382         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 383             EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
 384                 return blk;
 385         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 386                         EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
 387                         EXT4_B2C(sbi, offset));
 388         if (next_zero_bit <
 389             EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
 390                 /* bad bitmap for inode tables */
 391                 return blk;
 392         return 0;
 393 }
 394
 395 static int ext4_validate_block_bitmap(struct super_block *sb,
 396                                       struct ext4_group_desc *desc,
 397                                       ext4_group_t block_group,
 398                                       struct buffer_head *bh)
 399 {
 400         ext4_fsblk_t    blk;
 401         struct ext4_group_info *grp;
 402
 403         if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
 404                 return 0;
 405
 406         grp = ext4_get_group_info(sb, block_group);
 407
 408         if (buffer_verified(bh))
 409                 return 0;
 410         if (!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 411                 return -EFSCORRUPTED;
 412
 413         ext4_lock_group(sb, block_group);
 414         if (buffer_verified(bh))
 415                 goto verified;
 416         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 417                                                     desc, bh) ||
 418                      ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
 419                 ext4_unlock_group(sb, block_group);
 420                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 421                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 422                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 423                 return -EFSBADCRC;
 424         }
 425         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
 426         if (unlikely(blk != 0)) {
 427                 ext4_unlock_group(sb, block_group);
 428                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
 429                            block_group, blk);
 430                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 431                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 432                 return -EFSCORRUPTED;
 433         }
 434         blk = ext4_valid_block_bitmap_padding(sb, block_group, bh);
 435         if (unlikely(blk != 0)) {
 436                 ext4_unlock_group(sb, block_group);
 437                 ext4_error(sb, "bg %u: block %llu: padding at end of block bitmap is not set",
 438                            block_group, blk);
 439                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 440                                                  EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 441                 return -EFSCORRUPTED;
 442         }
 443         set_buffer_verified(bh);
 444 verified:
 445         ext4_unlock_group(sb, block_group);
 446         return 0;
 447 }
 448
 449 /**
 450  * ext4_read_block_bitmap_nowait()
 451  * @sb:                 super block
 452  * @block_group:        given block group
 453  *
 454  * Read the bitmap for a given block_group,and validate the
 455  * bits for block/inode/inode tables are set in the bitmaps
 456  *
 457  * Return buffer_head on success or an ERR_PTR in case of failure.
 458  */
 459 struct buffer_head *
 460 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group,
 461                               bool ignore_locked)
 462 {
 463         struct ext4_group_desc *desc;
 464         struct ext4_sb_info *sbi = EXT4_SB(sb);
 465         struct buffer_head *bh;
 466         ext4_fsblk_t bitmap_blk;
 467         int err;
 468
 469         desc = ext4_get_group_desc(sb, block_group, NULL);
 470         if (!desc)
 471                 return ERR_PTR(-EFSCORRUPTED);
 472         bitmap_blk = ext4_block_bitmap(sb, desc);
 473         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 474             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 475                 ext4_error(sb, "Invalid block bitmap block %llu in "
 476                            "block_group %u", bitmap_blk, block_group);
 477                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 478                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 479                 return ERR_PTR(-EFSCORRUPTED);
 480         }
 481         bh = sb_getblk(sb, bitmap_blk);
 482         if (unlikely(!bh)) {
 483                 ext4_warning(sb, "Cannot get buffer for block bitmap - "
 484                              "block_group = %u, block_bitmap = %llu",
 485                              block_group, bitmap_blk);
 486                 return ERR_PTR(-ENOMEM);
 487         }
 488
 489         if (ignore_locked && buffer_locked(bh)) {
 490                 /* buffer under IO already, return if called for prefetching */
 491                 put_bh(bh);
 492                 return NULL;
 493         }
 494
 495         if (bitmap_uptodate(bh))
 496                 goto verify;
 497
 498         lock_buffer(bh);
 499         if (bitmap_uptodate(bh)) {
 500                 unlock_buffer(bh);
 501                 goto verify;
 502         }
 503         ext4_lock_group(sb, block_group);
 504         if (ext4_has_group_desc_csum(sb) &&
 505             (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 506                 if (block_group == 0) {
 507                         ext4_unlock_group(sb, block_group);
 508                         unlock_buffer(bh);
 509                         ext4_error(sb, "Block bitmap for bg 0 marked "
 510                                    "uninitialized");
 511                         err = -EFSCORRUPTED;
 512                         goto out;
 513                 }
 514                 err = ext4_init_block_bitmap(sb, bh, block_group, desc);
 515                 set_bitmap_uptodate(bh);
 516                 set_buffer_uptodate(bh);
 517                 set_buffer_verified(bh);
 518                 ext4_unlock_group(sb, block_group);
 519                 unlock_buffer(bh);
 520                 if (err) {
 521                         ext4_error(sb, "Failed to init block bitmap for group "
 522                                    "%u: %d", block_group, err);
 523                         goto out;
 524                 }
 525                 goto verify;
 526         }
 527         ext4_unlock_group(sb, block_group);
 528         if (buffer_uptodate(bh)) {
 529                 /*
 530                  * if not uninit if bh is uptodate,
 531                  * bitmap is also uptodate
 532                  */
 533                 set_bitmap_uptodate(bh);
 534                 unlock_buffer(bh);
 535                 goto verify;
 536         }
 537         /*
 538          * submit the buffer_head for reading
 539          */
 540         set_buffer_new(bh);
 541         trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked);
 542         ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO |
 543                             (ignore_locked ? REQ_RAHEAD : 0),
 544                             ext4_end_bitmap_read);
 545         return bh;
 546 verify:
 547         err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
 548         if (err)
 549                 goto out;
 550         return bh;
 551 out:
 552         put_bh(bh);
 553         return ERR_PTR(err);
 554 }
 555
 556 /* Returns 0 on success, -errno on error */
 557 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
 558                            struct buffer_head *bh)
 559 {
 560         struct ext4_group_desc *desc;
 561
 562         if (!buffer_new(bh))
 563                 return 0;
 564         desc = ext4_get_group_desc(sb, block_group, NULL);
 565         if (!desc)
 566                 return -EFSCORRUPTED;
 567         wait_on_buffer(bh);
 568         ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO);
 569         if (!buffer_uptodate(bh)) {
 570                 ext4_error_err(sb, EIO, "Cannot read block bitmap - "
 571                                "block_group = %u, block_bitmap = %llu",
 572                                block_group, (unsigned long long) bh->b_blocknr);
 573                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 574                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 575                 return -EIO;
 576         }
 577         clear_buffer_new(bh);
 578         /* Panic or remount fs read-only if block bitmap is invalid */
 579         return ext4_validate_block_bitmap(sb, desc, block_group, bh);
 580 }
 581
 582 struct buffer_head *
 583 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 584 {
 585         struct buffer_head *bh;
 586         int err;
 587
 588         bh = ext4_read_block_bitmap_nowait(sb, block_group, false);
 589         if (IS_ERR(bh))
 590                 return bh;
 591         err = ext4_wait_block_bitmap(sb, block_group, bh);
 592         if (err) {
 593                 put_bh(bh);
 594                 return ERR_PTR(err);
 595         }
 596         return bh;
 597 }
 598
 599 /**
 600  * ext4_has_free_clusters()
 601  * @sbi:        in-core super block structure.
 602  * @nclusters:  number of needed blocks
 603  * @flags:      flags from ext4_mb_new_blocks()
 604  *
 605  * Check if filesystem has nclusters free & available for allocation.
 606  * On success return 1, return 0 on failure.
 607  */
 608 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 609                                   s64 nclusters, unsigned int flags)
 610 {
 611         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 612         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 613         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 614
 615         free_clusters  = percpu_counter_read_positive(fcc);
 616         dirty_clusters = percpu_counter_read_positive(dcc);
 617         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 618
 619         /*
 620          * r_blocks_count should always be multiple of the cluster ratio so
 621          * we are safe to do a plane bit shift only.
 622          */
 623         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 624               resv_clusters;
 625
 626         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 627                                         EXT4_FREECLUSTERS_WATERMARK) {
 628                 free_clusters  = percpu_counter_sum_positive(fcc);
 629                 dirty_clusters = percpu_counter_sum_positive(dcc);
 630         }
 631         /* Check whether we have space after accounting for current
 632          * dirty clusters & root reserved clusters.
 633          */
 634         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 635                 return 1;
 636
 637         /* Hm, nope.  Are (enough) root reserved clusters available? */
 638         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 639             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 640             capable(CAP_SYS_RESOURCE) ||
 641             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 642
 643                 if (free_clusters >= (nclusters + dirty_clusters +
 644                                       resv_clusters))
 645                         return 1;
 646         }
 647         /* No free blocks. Let's see if we can dip into reserved pool */
 648         if (flags & EXT4_MB_USE_RESERVED) {
 649                 if (free_clusters >= (nclusters + dirty_clusters))
 650                         return 1;
 651         }
 652
 653         return 0;
 654 }
 655
 656 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 657                              s64 nclusters, unsigned int flags)
 658 {
 659         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 660                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 661                 return 0;
 662         } else
 663                 return -ENOSPC;
 664 }
 665
 666 /**
 667  * ext4_should_retry_alloc() - check if a block allocation should be retried
 668  * @sb:                 superblock
 669  * @retries:            number of retry attempts made so far
 670  *
 671  * ext4_should_retry_alloc() is called when ENOSPC is returned while
 672  * attempting to allocate blocks.  If there's an indication that a pending
 673  * journal transaction might free some space and allow another attempt to
 674  * succeed, this function will wait for the current or committing transaction
 675  * to complete and then return TRUE.
 676  */
 677 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 678 {
 679         struct ext4_sb_info *sbi = EXT4_SB(sb);
 680
 681         if (!sbi->s_journal)
 682                 return 0;
 683
 684         if (++(*retries) > 3) {
 685                 percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit);
 686                 return 0;
 687         }
 688
 689         /*
 690          * if there's no indication that blocks are about to be freed it's
 691          * possible we just missed a transaction commit that did so
 692          */
 693         smp_mb();
 694         if (sbi->s_mb_free_pending == 0) {
 695                 if (test_opt(sb, DISCARD)) {
 696                         atomic_inc(&sbi->s_retry_alloc_pending);
 697                         flush_work(&sbi->s_discard_work);
 698                         atomic_dec(&sbi->s_retry_alloc_pending);
 699                 }
 700                 return ext4_has_free_clusters(sbi, 1, 0);
 701         }
 702
 703         /*
 704          * it's possible we've just missed a transaction commit here,
 705          * so ignore the returned status
 706          */
 707         ext4_debug("%s: retrying operation after ENOSPC\n", sb->s_id);
 708         (void) jbd2_journal_force_commit_nested(sbi->s_journal);
 709         return 1;
 710 }
 711
 712 /*
 713  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 714  *
 715  * @handle:             handle to this transaction
 716  * @inode:              file inode
 717  * @goal:               given target block(filesystem wide)
 718  * @count:              pointer to total number of clusters needed
 719  * @errp:               error code
 720  *
 721  * Return 1st allocated block number on success, *count stores total account
 722  * error stores in errp pointer
 723  */
 724 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 725                                   ext4_fsblk_t goal, unsigned int flags,
 726                                   unsigned long *count, int *errp)
 727 {
 728         struct ext4_allocation_request ar;
 729         ext4_fsblk_t ret;
 730
 731         memset(&ar, 0, sizeof(ar));
 732         /* Fill with neighbour allocated blocks */
 733         ar.inode = inode;
 734         ar.goal = goal;
 735         ar.len = count ? *count : 1;
 736         ar.flags = flags;
 737
 738         ret = ext4_mb_new_blocks(handle, &ar, errp);
 739         if (count)
 740                 *count = ar.len;
 741         /*
 742          * Account for the allocated meta blocks.  We will never
 743          * fail EDQUOT for metdata, but we do account for it.
 744          */
 745         if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
 746                 dquot_alloc_block_nofail(inode,
 747                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 748         }
 749         return ret;
 750 }
 751
 752 /**
 753  * ext4_count_free_clusters() -- count filesystem free clusters
 754  * @sb:         superblock
 755  *
 756  * Adds up the number of free clusters from each block group.
 757  */
 758 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 759 {
 760         ext4_fsblk_t desc_count;
 761         struct ext4_group_desc *gdp;
 762         ext4_group_t i;
 763         ext4_group_t ngroups = ext4_get_groups_count(sb);
 764         struct ext4_group_info *grp;
 765 #ifdef EXT4FS_DEBUG
 766         struct ext4_super_block *es;
 767         ext4_fsblk_t bitmap_count;
 768         unsigned int x;
 769         struct buffer_head *bitmap_bh = NULL;
 770
 771         es = EXT4_SB(sb)->s_es;
 772         desc_count = 0;
 773         bitmap_count = 0;
 774         gdp = NULL;
 775
 776         for (i = 0; i < ngroups; i++) {
 777                 gdp = ext4_get_group_desc(sb, i, NULL);
 778                 if (!gdp)
 779                         continue;
 780                 grp = NULL;
 781                 if (EXT4_SB(sb)->s_group_info)
 782                         grp = ext4_get_group_info(sb, i);
 783                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 784                         desc_count += ext4_free_group_clusters(sb, gdp);
 785                 brelse(bitmap_bh);
 786                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 787                 if (IS_ERR(bitmap_bh)) {
 788                         bitmap_bh = NULL;
 789                         continue;
 790                 }
 791
 792                 x = ext4_count_free(bitmap_bh->b_data,
 793                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
 794                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 795                         i, ext4_free_group_clusters(sb, gdp), x);
 796                 bitmap_count += x;
 797         }
 798         brelse(bitmap_bh);
 799         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 800                ", computed = %llu, %llu\n",
 801                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 802                desc_count, bitmap_count);
 803         return bitmap_count;
 804 #else
 805         desc_count = 0;
 806         for (i = 0; i < ngroups; i++) {
 807                 gdp = ext4_get_group_desc(sb, i, NULL);
 808                 if (!gdp)
 809                         continue;
 810                 grp = NULL;
 811                 if (EXT4_SB(sb)->s_group_info)
 812                         grp = ext4_get_group_info(sb, i);
 813                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 814                         desc_count += ext4_free_group_clusters(sb, gdp);
 815         }
 816
 817         return desc_count;
 818 #endif
 819 }
 820
 821 static inline int test_root(ext4_group_t a, int b)
 822 {
 823         while (1) {
 824                 if (a < b)
 825                         return 0;
 826                 if (a == b)
 827                         return 1;
 828                 if ((a % b) != 0)
 829                         return 0;
 830                 a = a / b;
 831         }
 832 }
 833
 834 /**
 835  *      ext4_bg_has_super - number of blocks used by the superblock in group
 836  *      @sb: superblock for filesystem
 837  *      @group: group number to check
 838  *
 839  *      Return the number of blocks used by the superblock (primary or backup)
 840  *      in this group.  Currently this will be only 0 or 1.
 841  */
 842 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 843 {
 844         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 845
 846         if (group == 0)
 847                 return 1;
 848         if (ext4_has_feature_sparse_super2(sb)) {
 849                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
 850                     group == le32_to_cpu(es->s_backup_bgs[1]))
 851                         return 1;
 852                 return 0;
 853         }
 854         if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
 855                 return 1;
 856         if (!(group & 1))
 857                 return 0;
 858         if (test_root(group, 3) || (test_root(group, 5)) ||
 859             test_root(group, 7))
 860                 return 1;
 861
 862         return 0;
 863 }
 864
 865 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 866                                         ext4_group_t group)
 867 {
 868         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 869         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 870         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 871
 872         if (group == first || group == first + 1 || group == last)
 873                 return 1;
 874         return 0;
 875 }
 876
 877 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 878                                         ext4_group_t group)
 879 {
 880         if (!ext4_bg_has_super(sb, group))
 881                 return 0;
 882
 883         if (ext4_has_feature_meta_bg(sb))
 884                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 885         else
 886                 return EXT4_SB(sb)->s_gdb_count;
 887 }
 888
 889 /**
 890  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 891  *      @sb: superblock for filesystem
 892  *      @group: group number to check
 893  *
 894  *      Return the number of blocks used by the group descriptor table
 895  *      (primary or backup) in this group.  In the future there may be a
 896  *      different number of descriptor blocks in each group.
 897  */
 898 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 899 {
 900         unsigned long first_meta_bg =
 901                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 902         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 903
 904         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
 905                 return ext4_bg_num_gdb_nometa(sb, group);
 906
 907         return ext4_bg_num_gdb_meta(sb,group);
 908
 909 }
 910
 911 /*
 912  * This function returns the number of file system metadata blocks at
 913  * the beginning of a block group, including the reserved gdt blocks.
 914  */
 915 unsigned int ext4_num_base_meta_blocks(struct super_block *sb,
 916                                        ext4_group_t block_group)
 917 {
 918         struct ext4_sb_info *sbi = EXT4_SB(sb);
 919         unsigned num;
 920
 921         /* Check for superblock and gdt backups in this group */
 922         num = ext4_bg_has_super(sb, block_group);
 923
 924         if (!ext4_has_feature_meta_bg(sb) ||
 925             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 926                           sbi->s_desc_per_block) {
 927                 if (num) {
 928                         num += ext4_bg_num_gdb(sb, block_group);
 929                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 930                 }
 931         } else { /* For META_BG_BLOCK_GROUPS */
 932                 num += ext4_bg_num_gdb(sb, block_group);
 933         }
 934         return num;
 935 }
 936
 937 static unsigned int ext4_num_base_meta_clusters(struct super_block *sb,
 938                                                 ext4_group_t block_group)
 939 {
 940         return EXT4_NUM_B2C(EXT4_SB(sb), ext4_num_base_meta_blocks(sb, block_group));
 941 }
 942
 943 /**
 944  *      ext4_inode_to_goal_block - return a hint for block allocation
 945  *      @inode: inode for block allocation
 946  *
 947  *      Return the ideal location to start allocating blocks for a
 948  *      newly created inode.
 949  */
 950 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 951 {
 952         struct ext4_inode_info *ei = EXT4_I(inode);
 953         ext4_group_t block_group;
 954         ext4_grpblk_t colour;
 955         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 956         ext4_fsblk_t bg_start;
 957         ext4_fsblk_t last_block;
 958
 959         block_group = ei->i_block_group;
 960         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 961                 /*
 962                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 963                  * block groups per flexgroup, reserve the first block
 964                  * group for directories and special files.  Regular
 965                  * files will start at the second block group.  This
 966                  * tends to speed up directory access and improves
 967                  * fsck times.
 968                  */
 969                 block_group &= ~(flex_size-1);
 970                 if (S_ISREG(inode->i_mode))
 971                         block_group++;
 972         }
 973         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 974         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 975
 976         /*
 977          * If we are doing delayed allocation, we don't need take
 978          * colour into account.
 979          */
 980         if (test_opt(inode->i_sb, DELALLOC))
 981                 return bg_start;
 982
 983         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 984                 colour = (task_pid_nr(current) % 16) *
 985                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 986         else
 987                 colour = (task_pid_nr(current) % 16) *
 988                         ((last_block - bg_start) / 16);
 989         return bg_start + colour;
 990 }
 991