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 /*
 307  * Return the block number which was discovered to be invalid, or 0 if
 308  * the block bitmap is valid.
 309  */
 310 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
 311                                             struct ext4_group_desc *desc,
 312                                             ext4_group_t block_group,
 313                                             struct buffer_head *bh)
 314 {
 315         struct ext4_sb_info *sbi = EXT4_SB(sb);
 316         ext4_grpblk_t offset;
 317         ext4_grpblk_t next_zero_bit;
 318         ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
 319         ext4_fsblk_t blk;
 320         ext4_fsblk_t group_first_block;
 321
 322         if (ext4_has_feature_flex_bg(sb)) {
 323                 /* with FLEX_BG, the inode/block bitmaps and itable
 324                  * blocks may not be in the group at all
 325                  * so the bitmap validation will be skipped for those groups
 326                  * or it has to also read the block group where the bitmaps
 327                  * are located to verify they are set.
 328                  */
 329                 return 0;
 330         }
 331         group_first_block = ext4_group_first_block_no(sb, block_group);
 332
 333         /* check whether block bitmap block number is set */
 334         blk = ext4_block_bitmap(sb, desc);
 335         offset = blk - group_first_block;
 336         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 337             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 338                 /* bad block bitmap */
 339                 return blk;
 340
 341         /* check whether the inode bitmap block number is set */
 342         blk = ext4_inode_bitmap(sb, desc);
 343         offset = blk - group_first_block;
 344         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 345             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
 346                 /* bad block bitmap */
 347                 return blk;
 348
 349         /* check whether the inode table block number is set */
 350         blk = ext4_inode_table(sb, desc);
 351         offset = blk - group_first_block;
 352         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
 353             EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
 354                 return blk;
 355         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
 356                         EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
 357                         EXT4_B2C(sbi, offset));
 358         if (next_zero_bit <
 359             EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
 360                 /* bad bitmap for inode tables */
 361                 return blk;
 362         return 0;
 363 }
 364
 365 static int ext4_validate_block_bitmap(struct super_block *sb,
 366                                       struct ext4_group_desc *desc,
 367                                       ext4_group_t block_group,
 368                                       struct buffer_head *bh)
 369 {
 370         ext4_fsblk_t    blk;
 371         struct ext4_group_info *grp;
 372
 373         if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
 374                 return 0;
 375
 376         grp = ext4_get_group_info(sb, block_group);
 377
 378         if (buffer_verified(bh))
 379                 return 0;
 380         if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 381                 return -EFSCORRUPTED;
 382
 383         ext4_lock_group(sb, block_group);
 384         if (buffer_verified(bh))
 385                 goto verified;
 386         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
 387                                                     desc, bh) ||
 388                      ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
 389                 ext4_unlock_group(sb, block_group);
 390                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
 391                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 392                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 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                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 401                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 402                 return -EFSCORRUPTED;
 403         }
 404         set_buffer_verified(bh);
 405 verified:
 406         ext4_unlock_group(sb, block_group);
 407         return 0;
 408 }
 409
 410 /**
 411  * ext4_read_block_bitmap_nowait()
 412  * @sb:                 super block
 413  * @block_group:        given block group
 414  *
 415  * Read the bitmap for a given block_group,and validate the
 416  * bits for block/inode/inode tables are set in the bitmaps
 417  *
 418  * Return buffer_head on success or an ERR_PTR in case of failure.
 419  */
 420 struct buffer_head *
 421 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group,
 422                               bool ignore_locked)
 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                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 439                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 440                 return ERR_PTR(-EFSCORRUPTED);
 441         }
 442         bh = sb_getblk(sb, bitmap_blk);
 443         if (unlikely(!bh)) {
 444                 ext4_warning(sb, "Cannot get buffer for block bitmap - "
 445                              "block_group = %u, block_bitmap = %llu",
 446                              block_group, bitmap_blk);
 447                 return ERR_PTR(-ENOMEM);
 448         }
 449
 450         if (ignore_locked && buffer_locked(bh)) {
 451                 /* buffer under IO already, return if called for prefetching */
 452                 put_bh(bh);
 453                 return NULL;
 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, ignore_locked);
 503         ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO |
 504                             (ignore_locked ? REQ_RAHEAD : 0),
 505                             ext4_end_bitmap_read);
 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, -errno 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         ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO);
 530         if (!buffer_uptodate(bh)) {
 531                 ext4_error_err(sb, EIO, "Cannot read block bitmap - "
 532                                "block_group = %u, block_bitmap = %llu",
 533                                block_group, (unsigned long long) bh->b_blocknr);
 534                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 535                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
 536                 return -EIO;
 537         }
 538         clear_buffer_new(bh);
 539         /* Panic or remount fs read-only if block bitmap is invalid */
 540         return ext4_validate_block_bitmap(sb, desc, block_group, bh);
 541 }
 542
 543 struct buffer_head *
 544 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 545 {
 546         struct buffer_head *bh;
 547         int err;
 548
 549         bh = ext4_read_block_bitmap_nowait(sb, block_group, false);
 550         if (IS_ERR(bh))
 551                 return bh;
 552         err = ext4_wait_block_bitmap(sb, block_group, bh);
 553         if (err) {
 554                 put_bh(bh);
 555                 return ERR_PTR(err);
 556         }
 557         return bh;
 558 }
 559
 560 /**
 561  * ext4_has_free_clusters()
 562  * @sbi:        in-core super block structure.
 563  * @nclusters:  number of needed blocks
 564  * @flags:      flags from ext4_mb_new_blocks()
 565  *
 566  * Check if filesystem has nclusters free & available for allocation.
 567  * On success return 1, return 0 on failure.
 568  */
 569 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
 570                                   s64 nclusters, unsigned int flags)
 571 {
 572         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
 573         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
 574         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
 575
 576         free_clusters  = percpu_counter_read_positive(fcc);
 577         dirty_clusters = percpu_counter_read_positive(dcc);
 578         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
 579
 580         /*
 581          * r_blocks_count should always be multiple of the cluster ratio so
 582          * we are safe to do a plane bit shift only.
 583          */
 584         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
 585               resv_clusters;
 586
 587         if (free_clusters - (nclusters + rsv + dirty_clusters) <
 588                                         EXT4_FREECLUSTERS_WATERMARK) {
 589                 free_clusters  = percpu_counter_sum_positive(fcc);
 590                 dirty_clusters = percpu_counter_sum_positive(dcc);
 591         }
 592         /* Check whether we have space after accounting for current
 593          * dirty clusters & root reserved clusters.
 594          */
 595         if (free_clusters >= (rsv + nclusters + dirty_clusters))
 596                 return 1;
 597
 598         /* Hm, nope.  Are (enough) root reserved clusters available? */
 599         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
 600             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
 601             capable(CAP_SYS_RESOURCE) ||
 602             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
 603
 604                 if (free_clusters >= (nclusters + dirty_clusters +
 605                                       resv_clusters))
 606                         return 1;
 607         }
 608         /* No free blocks. Let's see if we can dip into reserved pool */
 609         if (flags & EXT4_MB_USE_RESERVED) {
 610                 if (free_clusters >= (nclusters + dirty_clusters))
 611                         return 1;
 612         }
 613
 614         return 0;
 615 }
 616
 617 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
 618                              s64 nclusters, unsigned int flags)
 619 {
 620         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
 621                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
 622                 return 0;
 623         } else
 624                 return -ENOSPC;
 625 }
 626
 627 /**
 628  * ext4_should_retry_alloc() - check if a block allocation should be retried
 629  * @sb:                 superblock
 630  * @retries:            number of retry attempts made so far
 631  *
 632  * ext4_should_retry_alloc() is called when ENOSPC is returned while
 633  * attempting to allocate blocks.  If there's an indication that a pending
 634  * journal transaction might free some space and allow another attempt to
 635  * succeed, this function will wait for the current or committing transaction
 636  * to complete and then return TRUE.
 637  */
 638 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
 639 {
 640         struct ext4_sb_info *sbi = EXT4_SB(sb);
 641
 642         if (!sbi->s_journal)
 643                 return 0;
 644
 645         if (++(*retries) > 3) {
 646                 percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit);
 647                 return 0;
 648         }
 649
 650         /*
 651          * if there's no indication that blocks are about to be freed it's
 652          * possible we just missed a transaction commit that did so
 653          */
 654         smp_mb();
 655         if (sbi->s_mb_free_pending == 0) {
 656                 if (test_opt(sb, DISCARD)) {
 657                         atomic_inc(&sbi->s_retry_alloc_pending);
 658                         flush_work(&sbi->s_discard_work);
 659                         atomic_dec(&sbi->s_retry_alloc_pending);
 660                 }
 661                 return ext4_has_free_clusters(sbi, 1, 0);
 662         }
 663
 664         /*
 665          * it's possible we've just missed a transaction commit here,
 666          * so ignore the returned status
 667          */
 668         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
 669         (void) jbd2_journal_force_commit_nested(sbi->s_journal);
 670         return 1;
 671 }
 672
 673 /*
 674  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
 675  *
 676  * @handle:             handle to this transaction
 677  * @inode:              file inode
 678  * @goal:               given target block(filesystem wide)
 679  * @count:              pointer to total number of clusters needed
 680  * @errp:               error code
 681  *
 682  * Return 1st allocated block number on success, *count stores total account
 683  * error stores in errp pointer
 684  */
 685 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
 686                                   ext4_fsblk_t goal, unsigned int flags,
 687                                   unsigned long *count, int *errp)
 688 {
 689         struct ext4_allocation_request ar;
 690         ext4_fsblk_t ret;
 691
 692         memset(&ar, 0, sizeof(ar));
 693         /* Fill with neighbour allocated blocks */
 694         ar.inode = inode;
 695         ar.goal = goal;
 696         ar.len = count ? *count : 1;
 697         ar.flags = flags;
 698
 699         ret = ext4_mb_new_blocks(handle, &ar, errp);
 700         if (count)
 701                 *count = ar.len;
 702         /*
 703          * Account for the allocated meta blocks.  We will never
 704          * fail EDQUOT for metdata, but we do account for it.
 705          */
 706         if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
 707                 dquot_alloc_block_nofail(inode,
 708                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
 709         }
 710         return ret;
 711 }
 712
 713 /**
 714  * ext4_count_free_clusters() -- count filesystem free clusters
 715  * @sb:         superblock
 716  *
 717  * Adds up the number of free clusters from each block group.
 718  */
 719 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
 720 {
 721         ext4_fsblk_t desc_count;
 722         struct ext4_group_desc *gdp;
 723         ext4_group_t i;
 724         ext4_group_t ngroups = ext4_get_groups_count(sb);
 725         struct ext4_group_info *grp;
 726 #ifdef EXT4FS_DEBUG
 727         struct ext4_super_block *es;
 728         ext4_fsblk_t bitmap_count;
 729         unsigned int x;
 730         struct buffer_head *bitmap_bh = NULL;
 731
 732         es = EXT4_SB(sb)->s_es;
 733         desc_count = 0;
 734         bitmap_count = 0;
 735         gdp = NULL;
 736
 737         for (i = 0; i < ngroups; i++) {
 738                 gdp = ext4_get_group_desc(sb, i, NULL);
 739                 if (!gdp)
 740                         continue;
 741                 grp = NULL;
 742                 if (EXT4_SB(sb)->s_group_info)
 743                         grp = ext4_get_group_info(sb, i);
 744                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 745                         desc_count += ext4_free_group_clusters(sb, gdp);
 746                 brelse(bitmap_bh);
 747                 bitmap_bh = ext4_read_block_bitmap(sb, i);
 748                 if (IS_ERR(bitmap_bh)) {
 749                         bitmap_bh = NULL;
 750                         continue;
 751                 }
 752
 753                 x = ext4_count_free(bitmap_bh->b_data,
 754                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
 755                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
 756                         i, ext4_free_group_clusters(sb, gdp), x);
 757                 bitmap_count += x;
 758         }
 759         brelse(bitmap_bh);
 760         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
 761                ", computed = %llu, %llu\n",
 762                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
 763                desc_count, bitmap_count);
 764         return bitmap_count;
 765 #else
 766         desc_count = 0;
 767         for (i = 0; i < ngroups; i++) {
 768                 gdp = ext4_get_group_desc(sb, i, NULL);
 769                 if (!gdp)
 770                         continue;
 771                 grp = NULL;
 772                 if (EXT4_SB(sb)->s_group_info)
 773                         grp = ext4_get_group_info(sb, i);
 774                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
 775                         desc_count += ext4_free_group_clusters(sb, gdp);
 776         }
 777
 778         return desc_count;
 779 #endif
 780 }
 781
 782 static inline int test_root(ext4_group_t a, int b)
 783 {
 784         while (1) {
 785                 if (a < b)
 786                         return 0;
 787                 if (a == b)
 788                         return 1;
 789                 if ((a % b) != 0)
 790                         return 0;
 791                 a = a / b;
 792         }
 793 }
 794
 795 /**
 796  *      ext4_bg_has_super - number of blocks used by the superblock in group
 797  *      @sb: superblock for filesystem
 798  *      @group: group number to check
 799  *
 800  *      Return the number of blocks used by the superblock (primary or backup)
 801  *      in this group.  Currently this will be only 0 or 1.
 802  */
 803 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
 804 {
 805         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 806
 807         if (group == 0)
 808                 return 1;
 809         if (ext4_has_feature_sparse_super2(sb)) {
 810                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
 811                     group == le32_to_cpu(es->s_backup_bgs[1]))
 812                         return 1;
 813                 return 0;
 814         }
 815         if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
 816                 return 1;
 817         if (!(group & 1))
 818                 return 0;
 819         if (test_root(group, 3) || (test_root(group, 5)) ||
 820             test_root(group, 7))
 821                 return 1;
 822
 823         return 0;
 824 }
 825
 826 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
 827                                         ext4_group_t group)
 828 {
 829         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 830         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
 831         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
 832
 833         if (group == first || group == first + 1 || group == last)
 834                 return 1;
 835         return 0;
 836 }
 837
 838 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
 839                                         ext4_group_t group)
 840 {
 841         if (!ext4_bg_has_super(sb, group))
 842                 return 0;
 843
 844         if (ext4_has_feature_meta_bg(sb))
 845                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 846         else
 847                 return EXT4_SB(sb)->s_gdb_count;
 848 }
 849
 850 /**
 851  *      ext4_bg_num_gdb - number of blocks used by the group table in group
 852  *      @sb: superblock for filesystem
 853  *      @group: group number to check
 854  *
 855  *      Return the number of blocks used by the group descriptor table
 856  *      (primary or backup) in this group.  In the future there may be a
 857  *      different number of descriptor blocks in each group.
 858  */
 859 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
 860 {
 861         unsigned long first_meta_bg =
 862                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
 863         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
 864
 865         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
 866                 return ext4_bg_num_gdb_nometa(sb, group);
 867
 868         return ext4_bg_num_gdb_meta(sb,group);
 869
 870 }
 871
 872 /*
 873  * This function returns the number of file system metadata clusters at
 874  * the beginning of a block group, including the reserved gdt blocks.
 875  */
 876 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
 877                                      ext4_group_t block_group)
 878 {
 879         struct ext4_sb_info *sbi = EXT4_SB(sb);
 880         unsigned num;
 881
 882         /* Check for superblock and gdt backups in this group */
 883         num = ext4_bg_has_super(sb, block_group);
 884
 885         if (!ext4_has_feature_meta_bg(sb) ||
 886             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
 887                           sbi->s_desc_per_block) {
 888                 if (num) {
 889                         num += ext4_bg_num_gdb(sb, block_group);
 890                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
 891                 }
 892         } else { /* For META_BG_BLOCK_GROUPS */
 893                 num += ext4_bg_num_gdb(sb, block_group);
 894         }
 895         return EXT4_NUM_B2C(sbi, num);
 896 }
 897 /**
 898  *      ext4_inode_to_goal_block - return a hint for block allocation
 899  *      @inode: inode for block allocation
 900  *
 901  *      Return the ideal location to start allocating blocks for a
 902  *      newly created inode.
 903  */
 904 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
 905 {
 906         struct ext4_inode_info *ei = EXT4_I(inode);
 907         ext4_group_t block_group;
 908         ext4_grpblk_t colour;
 909         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 910         ext4_fsblk_t bg_start;
 911         ext4_fsblk_t last_block;
 912
 913         block_group = ei->i_block_group;
 914         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 915                 /*
 916                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 917                  * block groups per flexgroup, reserve the first block
 918                  * group for directories and special files.  Regular
 919                  * files will start at the second block group.  This
 920                  * tends to speed up directory access and improves
 921                  * fsck times.
 922                  */
 923                 block_group &= ~(flex_size-1);
 924                 if (S_ISREG(inode->i_mode))
 925                         block_group++;
 926         }
 927         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 928         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 929
 930         /*
 931          * If we are doing delayed allocation, we don't need take
 932          * colour into account.
 933          */
 934         if (test_opt(inode->i_sb, DELALLOC))
 935                 return bg_start;
 936
 937         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 938                 colour = (task_pid_nr(current) % 16) *
 939                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 940         else
 941                 colour = (task_pid_nr(current) % 16) *
 942                         ((last_block - bg_start) / 16);
 943         return bg_start + colour;
 944 }
 945