| 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 | * @ignore_locked: ignore locked buffers |
| 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 an ERR_PTR 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 | bool ignore_locked) |
| 424 | { |
| 425 | struct ext4_group_desc *desc; |
| 426 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 427 | struct buffer_head *bh; |
| 428 | ext4_fsblk_t bitmap_blk; |
| 429 | int err; |
| 430 | |
| 431 | desc = ext4_get_group_desc(sb, block_group, NULL); |
| 432 | if (!desc) |
| 433 | return ERR_PTR(-EFSCORRUPTED); |
| 434 | bitmap_blk = ext4_block_bitmap(sb, desc); |
| 435 | if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || |
| 436 | (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { |
| 437 | ext4_error(sb, "Invalid block bitmap block %llu in " |
| 438 | "block_group %u", bitmap_blk, block_group); |
| 439 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
| 440 | EXT4_GROUP_INFO_BBITMAP_CORRUPT); |
| 441 | return ERR_PTR(-EFSCORRUPTED); |
| 442 | } |
| 443 | bh = sb_getblk(sb, bitmap_blk); |
| 444 | if (unlikely(!bh)) { |
| 445 | ext4_warning(sb, "Cannot get buffer for block bitmap - " |
| 446 | "block_group = %u, block_bitmap = %llu", |
| 447 | block_group, bitmap_blk); |
| 448 | return ERR_PTR(-ENOMEM); |
| 449 | } |
| 450 | |
| 451 | if (ignore_locked && buffer_locked(bh)) { |
| 452 | /* buffer under IO already, return if called for prefetching */ |
| 453 | put_bh(bh); |
| 454 | return NULL; |
| 455 | } |
| 456 | |
| 457 | if (bitmap_uptodate(bh)) |
| 458 | goto verify; |
| 459 | |
| 460 | lock_buffer(bh); |
| 461 | if (bitmap_uptodate(bh)) { |
| 462 | unlock_buffer(bh); |
| 463 | goto verify; |
| 464 | } |
| 465 | ext4_lock_group(sb, block_group); |
| 466 | if (ext4_has_group_desc_csum(sb) && |
| 467 | (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { |
| 468 | if (block_group == 0) { |
| 469 | ext4_unlock_group(sb, block_group); |
| 470 | unlock_buffer(bh); |
| 471 | ext4_error(sb, "Block bitmap for bg 0 marked " |
| 472 | "uninitialized"); |
| 473 | err = -EFSCORRUPTED; |
| 474 | goto out; |
| 475 | } |
| 476 | err = ext4_init_block_bitmap(sb, bh, block_group, desc); |
| 477 | set_bitmap_uptodate(bh); |
| 478 | set_buffer_uptodate(bh); |
| 479 | set_buffer_verified(bh); |
| 480 | ext4_unlock_group(sb, block_group); |
| 481 | unlock_buffer(bh); |
| 482 | if (err) { |
| 483 | ext4_error(sb, "Failed to init block bitmap for group " |
| 484 | "%u: %d", block_group, err); |
| 485 | goto out; |
| 486 | } |
| 487 | goto verify; |
| 488 | } |
| 489 | ext4_unlock_group(sb, block_group); |
| 490 | if (buffer_uptodate(bh)) { |
| 491 | /* |
| 492 | * if not uninit if bh is uptodate, |
| 493 | * bitmap is also uptodate |
| 494 | */ |
| 495 | set_bitmap_uptodate(bh); |
| 496 | unlock_buffer(bh); |
| 497 | goto verify; |
| 498 | } |
| 499 | /* |
| 500 | * submit the buffer_head for reading |
| 501 | */ |
| 502 | set_buffer_new(bh); |
| 503 | trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked); |
| 504 | ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO | |
| 505 | (ignore_locked ? REQ_RAHEAD : 0), |
| 506 | ext4_end_bitmap_read); |
| 507 | return bh; |
| 508 | verify: |
| 509 | err = ext4_validate_block_bitmap(sb, desc, block_group, bh); |
| 510 | if (err) |
| 511 | goto out; |
| 512 | return bh; |
| 513 | out: |
| 514 | put_bh(bh); |
| 515 | return ERR_PTR(err); |
| 516 | } |
| 517 | |
| 518 | /* Returns 0 on success, -errno on error */ |
| 519 | int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group, |
| 520 | struct buffer_head *bh) |
| 521 | { |
| 522 | struct ext4_group_desc *desc; |
| 523 | |
| 524 | if (!buffer_new(bh)) |
| 525 | return 0; |
| 526 | desc = ext4_get_group_desc(sb, block_group, NULL); |
| 527 | if (!desc) |
| 528 | return -EFSCORRUPTED; |
| 529 | wait_on_buffer(bh); |
| 530 | ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO); |
| 531 | if (!buffer_uptodate(bh)) { |
| 532 | ext4_error_err(sb, EIO, "Cannot read block bitmap - " |
| 533 | "block_group = %u, block_bitmap = %llu", |
| 534 | block_group, (unsigned long long) bh->b_blocknr); |
| 535 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
| 536 | EXT4_GROUP_INFO_BBITMAP_CORRUPT); |
| 537 | return -EIO; |
| 538 | } |
| 539 | clear_buffer_new(bh); |
| 540 | /* Panic or remount fs read-only if block bitmap is invalid */ |
| 541 | return ext4_validate_block_bitmap(sb, desc, block_group, bh); |
| 542 | } |
| 543 | |
| 544 | struct buffer_head * |
| 545 | ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group) |
| 546 | { |
| 547 | struct buffer_head *bh; |
| 548 | int err; |
| 549 | |
| 550 | bh = ext4_read_block_bitmap_nowait(sb, block_group, false); |
| 551 | if (IS_ERR(bh)) |
| 552 | return bh; |
| 553 | err = ext4_wait_block_bitmap(sb, block_group, bh); |
| 554 | if (err) { |
| 555 | put_bh(bh); |
| 556 | return ERR_PTR(err); |
| 557 | } |
| 558 | return bh; |
| 559 | } |
| 560 | |
| 561 | /** |
| 562 | * ext4_has_free_clusters() |
| 563 | * @sbi: in-core super block structure. |
| 564 | * @nclusters: number of needed blocks |
| 565 | * @flags: flags from ext4_mb_new_blocks() |
| 566 | * |
| 567 | * Check if filesystem has nclusters free & available for allocation. |
| 568 | * On success return 1, return 0 on failure. |
| 569 | */ |
| 570 | static int ext4_has_free_clusters(struct ext4_sb_info *sbi, |
| 571 | s64 nclusters, unsigned int flags) |
| 572 | { |
| 573 | s64 free_clusters, dirty_clusters, rsv, resv_clusters; |
| 574 | struct percpu_counter *fcc = &sbi->s_freeclusters_counter; |
| 575 | struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter; |
| 576 | |
| 577 | free_clusters = percpu_counter_read_positive(fcc); |
| 578 | dirty_clusters = percpu_counter_read_positive(dcc); |
| 579 | resv_clusters = atomic64_read(&sbi->s_resv_clusters); |
| 580 | |
| 581 | /* |
| 582 | * r_blocks_count should always be multiple of the cluster ratio so |
| 583 | * we are safe to do a plane bit shift only. |
| 584 | */ |
| 585 | rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) + |
| 586 | resv_clusters; |
| 587 | |
| 588 | if (free_clusters - (nclusters + rsv + dirty_clusters) < |
| 589 | EXT4_FREECLUSTERS_WATERMARK) { |
| 590 | free_clusters = percpu_counter_sum_positive(fcc); |
| 591 | dirty_clusters = percpu_counter_sum_positive(dcc); |
| 592 | } |
| 593 | /* Check whether we have space after accounting for current |
| 594 | * dirty clusters & root reserved clusters. |
| 595 | */ |
| 596 | if (free_clusters >= (rsv + nclusters + dirty_clusters)) |
| 597 | return 1; |
| 598 | |
| 599 | /* Hm, nope. Are (enough) root reserved clusters available? */ |
| 600 | if (uid_eq(sbi->s_resuid, current_fsuid()) || |
| 601 | (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) || |
| 602 | capable(CAP_SYS_RESOURCE) || |
| 603 | (flags & EXT4_MB_USE_ROOT_BLOCKS)) { |
| 604 | |
| 605 | if (free_clusters >= (nclusters + dirty_clusters + |
| 606 | resv_clusters)) |
| 607 | return 1; |
| 608 | } |
| 609 | /* No free blocks. Let's see if we can dip into reserved pool */ |
| 610 | if (flags & EXT4_MB_USE_RESERVED) { |
| 611 | if (free_clusters >= (nclusters + dirty_clusters)) |
| 612 | return 1; |
| 613 | } |
| 614 | |
| 615 | return 0; |
| 616 | } |
| 617 | |
| 618 | int ext4_claim_free_clusters(struct ext4_sb_info *sbi, |
| 619 | s64 nclusters, unsigned int flags) |
| 620 | { |
| 621 | if (ext4_has_free_clusters(sbi, nclusters, flags)) { |
| 622 | percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters); |
| 623 | return 0; |
| 624 | } else |
| 625 | return -ENOSPC; |
| 626 | } |
| 627 | |
| 628 | /** |
| 629 | * ext4_should_retry_alloc() - check if a block allocation should be retried |
| 630 | * @sb: superblock |
| 631 | * @retries: number of retry attempts made so far |
| 632 | * |
| 633 | * ext4_should_retry_alloc() is called when ENOSPC is returned while |
| 634 | * attempting to allocate blocks. If there's an indication that a pending |
| 635 | * journal transaction might free some space and allow another attempt to |
| 636 | * succeed, this function will wait for the current or committing transaction |
| 637 | * to complete and then return TRUE. |
| 638 | */ |
| 639 | int ext4_should_retry_alloc(struct super_block *sb, int *retries) |
| 640 | { |
| 641 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 642 | |
| 643 | if (!sbi->s_journal) |
| 644 | return 0; |
| 645 | |
| 646 | if (++(*retries) > 3) { |
| 647 | percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit); |
| 648 | return 0; |
| 649 | } |
| 650 | |
| 651 | /* |
| 652 | * if there's no indication that blocks are about to be freed it's |
| 653 | * possible we just missed a transaction commit that did so |
| 654 | */ |
| 655 | smp_mb(); |
| 656 | if (sbi->s_mb_free_pending == 0) { |
| 657 | if (test_opt(sb, DISCARD)) { |
| 658 | atomic_inc(&sbi->s_retry_alloc_pending); |
| 659 | flush_work(&sbi->s_discard_work); |
| 660 | atomic_dec(&sbi->s_retry_alloc_pending); |
| 661 | } |
| 662 | return ext4_has_free_clusters(sbi, 1, 0); |
| 663 | } |
| 664 | |
| 665 | /* |
| 666 | * it's possible we've just missed a transaction commit here, |
| 667 | * so ignore the returned status |
| 668 | */ |
| 669 | ext4_debug("%s: retrying operation after ENOSPC\n", sb->s_id); |
| 670 | (void) jbd2_journal_force_commit_nested(sbi->s_journal); |
| 671 | return 1; |
| 672 | } |
| 673 | |
| 674 | /* |
| 675 | * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks |
| 676 | * |
| 677 | * @handle: handle to this transaction |
| 678 | * @inode: file inode |
| 679 | * @goal: given target block(filesystem wide) |
| 680 | * @count: pointer to total number of clusters needed |
| 681 | * @errp: error code |
| 682 | * |
| 683 | * Return 1st allocated block number on success, *count stores total account |
| 684 | * error stores in errp pointer |
| 685 | */ |
| 686 | ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode, |
| 687 | ext4_fsblk_t goal, unsigned int flags, |
| 688 | unsigned long *count, int *errp) |
| 689 | { |
| 690 | struct ext4_allocation_request ar; |
| 691 | ext4_fsblk_t ret; |
| 692 | |
| 693 | memset(&ar, 0, sizeof(ar)); |
| 694 | /* Fill with neighbour allocated blocks */ |
| 695 | ar.inode = inode; |
| 696 | ar.goal = goal; |
| 697 | ar.len = count ? *count : 1; |
| 698 | ar.flags = flags; |
| 699 | |
| 700 | ret = ext4_mb_new_blocks(handle, &ar, errp); |
| 701 | if (count) |
| 702 | *count = ar.len; |
| 703 | /* |
| 704 | * Account for the allocated meta blocks. We will never |
| 705 | * fail EDQUOT for metdata, but we do account for it. |
| 706 | */ |
| 707 | if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) { |
| 708 | dquot_alloc_block_nofail(inode, |
| 709 | EXT4_C2B(EXT4_SB(inode->i_sb), ar.len)); |
| 710 | } |
| 711 | return ret; |
| 712 | } |
| 713 | |
| 714 | /** |
| 715 | * ext4_count_free_clusters() -- count filesystem free clusters |
| 716 | * @sb: superblock |
| 717 | * |
| 718 | * Adds up the number of free clusters from each block group. |
| 719 | */ |
| 720 | ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb) |
| 721 | { |
| 722 | ext4_fsblk_t desc_count; |
| 723 | struct ext4_group_desc *gdp; |
| 724 | ext4_group_t i; |
| 725 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
| 726 | struct ext4_group_info *grp; |
| 727 | #ifdef EXT4FS_DEBUG |
| 728 | struct ext4_super_block *es; |
| 729 | ext4_fsblk_t bitmap_count; |
| 730 | unsigned int x; |
| 731 | struct buffer_head *bitmap_bh = NULL; |
| 732 | |
| 733 | es = EXT4_SB(sb)->s_es; |
| 734 | desc_count = 0; |
| 735 | bitmap_count = 0; |
| 736 | gdp = NULL; |
| 737 | |
| 738 | for (i = 0; i < ngroups; i++) { |
| 739 | gdp = ext4_get_group_desc(sb, i, NULL); |
| 740 | if (!gdp) |
| 741 | continue; |
| 742 | grp = NULL; |
| 743 | if (EXT4_SB(sb)->s_group_info) |
| 744 | grp = ext4_get_group_info(sb, i); |
| 745 | if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
| 746 | desc_count += ext4_free_group_clusters(sb, gdp); |
| 747 | brelse(bitmap_bh); |
| 748 | bitmap_bh = ext4_read_block_bitmap(sb, i); |
| 749 | if (IS_ERR(bitmap_bh)) { |
| 750 | bitmap_bh = NULL; |
| 751 | continue; |
| 752 | } |
| 753 | |
| 754 | x = ext4_count_free(bitmap_bh->b_data, |
| 755 | EXT4_CLUSTERS_PER_GROUP(sb) / 8); |
| 756 | printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n", |
| 757 | i, ext4_free_group_clusters(sb, gdp), x); |
| 758 | bitmap_count += x; |
| 759 | } |
| 760 | brelse(bitmap_bh); |
| 761 | printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu" |
| 762 | ", computed = %llu, %llu\n", |
| 763 | EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)), |
| 764 | desc_count, bitmap_count); |
| 765 | return bitmap_count; |
| 766 | #else |
| 767 | desc_count = 0; |
| 768 | for (i = 0; i < ngroups; i++) { |
| 769 | gdp = ext4_get_group_desc(sb, i, NULL); |
| 770 | if (!gdp) |
| 771 | continue; |
| 772 | grp = NULL; |
| 773 | if (EXT4_SB(sb)->s_group_info) |
| 774 | grp = ext4_get_group_info(sb, i); |
| 775 | if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
| 776 | desc_count += ext4_free_group_clusters(sb, gdp); |
| 777 | } |
| 778 | |
| 779 | return desc_count; |
| 780 | #endif |
| 781 | } |
| 782 | |
| 783 | static inline int test_root(ext4_group_t a, int b) |
| 784 | { |
| 785 | while (1) { |
| 786 | if (a < b) |
| 787 | return 0; |
| 788 | if (a == b) |
| 789 | return 1; |
| 790 | if ((a % b) != 0) |
| 791 | return 0; |
| 792 | a = a / b; |
| 793 | } |
| 794 | } |
| 795 | |
| 796 | /** |
| 797 | * ext4_bg_has_super - number of blocks used by the superblock in group |
| 798 | * @sb: superblock for filesystem |
| 799 | * @group: group number to check |
| 800 | * |
| 801 | * Return the number of blocks used by the superblock (primary or backup) |
| 802 | * in this group. Currently this will be only 0 or 1. |
| 803 | */ |
| 804 | int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) |
| 805 | { |
| 806 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
| 807 | |
| 808 | if (group == 0) |
| 809 | return 1; |
| 810 | if (ext4_has_feature_sparse_super2(sb)) { |
| 811 | if (group == le32_to_cpu(es->s_backup_bgs[0]) || |
| 812 | group == le32_to_cpu(es->s_backup_bgs[1])) |
| 813 | return 1; |
| 814 | return 0; |
| 815 | } |
| 816 | if ((group <= 1) || !ext4_has_feature_sparse_super(sb)) |
| 817 | return 1; |
| 818 | if (!(group & 1)) |
| 819 | return 0; |
| 820 | if (test_root(group, 3) || (test_root(group, 5)) || |
| 821 | test_root(group, 7)) |
| 822 | return 1; |
| 823 | |
| 824 | return 0; |
| 825 | } |
| 826 | |
| 827 | static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, |
| 828 | ext4_group_t group) |
| 829 | { |
| 830 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
| 831 | ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); |
| 832 | ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; |
| 833 | |
| 834 | if (group == first || group == first + 1 || group == last) |
| 835 | return 1; |
| 836 | return 0; |
| 837 | } |
| 838 | |
| 839 | static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, |
| 840 | ext4_group_t group) |
| 841 | { |
| 842 | if (!ext4_bg_has_super(sb, group)) |
| 843 | return 0; |
| 844 | |
| 845 | if (ext4_has_feature_meta_bg(sb)) |
| 846 | return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
| 847 | else |
| 848 | return EXT4_SB(sb)->s_gdb_count; |
| 849 | } |
| 850 | |
| 851 | /** |
| 852 | * ext4_bg_num_gdb - number of blocks used by the group table in group |
| 853 | * @sb: superblock for filesystem |
| 854 | * @group: group number to check |
| 855 | * |
| 856 | * Return the number of blocks used by the group descriptor table |
| 857 | * (primary or backup) in this group. In the future there may be a |
| 858 | * different number of descriptor blocks in each group. |
| 859 | */ |
| 860 | unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) |
| 861 | { |
| 862 | unsigned long first_meta_bg = |
| 863 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
| 864 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
| 865 | |
| 866 | if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) |
| 867 | return ext4_bg_num_gdb_nometa(sb, group); |
| 868 | |
| 869 | return ext4_bg_num_gdb_meta(sb,group); |
| 870 | |
| 871 | } |
| 872 | |
| 873 | /* |
| 874 | * This function returns the number of file system metadata clusters at |
| 875 | * the beginning of a block group, including the reserved gdt blocks. |
| 876 | */ |
| 877 | static unsigned ext4_num_base_meta_clusters(struct super_block *sb, |
| 878 | ext4_group_t block_group) |
| 879 | { |
| 880 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 881 | unsigned num; |
| 882 | |
| 883 | /* Check for superblock and gdt backups in this group */ |
| 884 | num = ext4_bg_has_super(sb, block_group); |
| 885 | |
| 886 | if (!ext4_has_feature_meta_bg(sb) || |
| 887 | block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * |
| 888 | sbi->s_desc_per_block) { |
| 889 | if (num) { |
| 890 | num += ext4_bg_num_gdb(sb, block_group); |
| 891 | num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); |
| 892 | } |
| 893 | } else { /* For META_BG_BLOCK_GROUPS */ |
| 894 | num += ext4_bg_num_gdb(sb, block_group); |
| 895 | } |
| 896 | return EXT4_NUM_B2C(sbi, num); |
| 897 | } |
| 898 | /** |
| 899 | * ext4_inode_to_goal_block - return a hint for block allocation |
| 900 | * @inode: inode for block allocation |
| 901 | * |
| 902 | * Return the ideal location to start allocating blocks for a |
| 903 | * newly created inode. |
| 904 | */ |
| 905 | ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode) |
| 906 | { |
| 907 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 908 | ext4_group_t block_group; |
| 909 | ext4_grpblk_t colour; |
| 910 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); |
| 911 | ext4_fsblk_t bg_start; |
| 912 | ext4_fsblk_t last_block; |
| 913 | |
| 914 | block_group = ei->i_block_group; |
| 915 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { |
| 916 | /* |
| 917 | * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME |
| 918 | * block groups per flexgroup, reserve the first block |
| 919 | * group for directories and special files. Regular |
| 920 | * files will start at the second block group. This |
| 921 | * tends to speed up directory access and improves |
| 922 | * fsck times. |
| 923 | */ |
| 924 | block_group &= ~(flex_size-1); |
| 925 | if (S_ISREG(inode->i_mode)) |
| 926 | block_group++; |
| 927 | } |
| 928 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); |
| 929 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
| 930 | |
| 931 | /* |
| 932 | * If we are doing delayed allocation, we don't need take |
| 933 | * colour into account. |
| 934 | */ |
| 935 | if (test_opt(inode->i_sb, DELALLOC)) |
| 936 | return bg_start; |
| 937 | |
| 938 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
| 939 | colour = (task_pid_nr(current) % 16) * |
| 940 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
| 941 | else |
| 942 | colour = (task_pid_nr(current) % 16) * |
| 943 | ((last_block - bg_start) / 16); |
| 944 | return bg_start + colour; |
| 945 | } |
| 946 | |