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 /* 84 * Return the number of clusters used for file system metadata; this 85 * represents the overhead needed by the file system. 86 */ 87 static unsigned ext4_num_overhead_clusters(struct super_block *sb, 88 ext4_group_t block_group, 89 struct ext4_group_desc *gdp) 90 { 91 unsigned base_clusters, num_clusters; 92 int block_cluster = -1, inode_cluster; 93 int itbl_cluster_start = -1, itbl_cluster_end = -1; 94 ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group); 95 ext4_fsblk_t end = start + EXT4_BLOCKS_PER_GROUP(sb) - 1; 96 ext4_fsblk_t itbl_blk_start, itbl_blk_end; 97 struct ext4_sb_info *sbi = EXT4_SB(sb); 98 99 /* This is the number of clusters used by the superblock, 100 * block group descriptors, and reserved block group 101 * descriptor blocks */ 102 base_clusters = ext4_num_base_meta_clusters(sb, block_group); 103 num_clusters = base_clusters; 104 105 /* 106 * Account and record inode table clusters if any cluster 107 * is in the block group, or inode table cluster range is 108 * [-1, -1] and won't overlap with block/inode bitmap cluster 109 * accounted below. 110 */ 111 itbl_blk_start = ext4_inode_table(sb, gdp); 112 itbl_blk_end = itbl_blk_start + sbi->s_itb_per_group - 1; 113 if (itbl_blk_start <= end && itbl_blk_end >= start) { 114 itbl_blk_start = itbl_blk_start >= start ? 115 itbl_blk_start : start; 116 itbl_blk_end = itbl_blk_end <= end ? 117 itbl_blk_end : end; 118 119 itbl_cluster_start = EXT4_B2C(sbi, itbl_blk_start - start); 120 itbl_cluster_end = EXT4_B2C(sbi, itbl_blk_end - start); 121 122 num_clusters += itbl_cluster_end - itbl_cluster_start + 1; 123 /* check if border cluster is overlapped */ 124 if (itbl_cluster_start == base_clusters - 1) 125 num_clusters--; 126 } 127 128 /* 129 * For the allocation bitmaps, we first need to check to see 130 * if the block is in the block group. If it is, then check 131 * to see if the cluster is already accounted for in the clusters 132 * used for the base metadata cluster and inode tables cluster. 133 * Normally all of these blocks are contiguous, so the special 134 * case handling shouldn't be necessary except for *very* 135 * unusual file system layouts. 136 */ 137 if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) { 138 block_cluster = EXT4_B2C(sbi, 139 ext4_block_bitmap(sb, gdp) - start); 140 if (block_cluster >= base_clusters && 141 (block_cluster < itbl_cluster_start || 142 block_cluster > itbl_cluster_end)) 143 num_clusters++; 144 } 145 146 if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) { 147 inode_cluster = EXT4_B2C(sbi, 148 ext4_inode_bitmap(sb, gdp) - start); 149 /* 150 * Additional check if inode bitmap is in just accounted 151 * block_cluster 152 */ 153 if (inode_cluster != block_cluster && 154 inode_cluster >= base_clusters && 155 (inode_cluster < itbl_cluster_start || 156 inode_cluster > itbl_cluster_end)) 157 num_clusters++; 158 } 159 160 return num_clusters; 161 } 162 163 static unsigned int num_clusters_in_group(struct super_block *sb, 164 ext4_group_t block_group) 165 { 166 unsigned int blocks; 167 168 if (block_group == ext4_get_groups_count(sb) - 1) { 169 /* 170 * Even though mke2fs always initializes the first and 171 * last group, just in case some other tool was used, 172 * we need to make sure we calculate the right free 173 * blocks. 174 */ 175 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) - 176 ext4_group_first_block_no(sb, block_group); 177 } else 178 blocks = EXT4_BLOCKS_PER_GROUP(sb); 179 return EXT4_NUM_B2C(EXT4_SB(sb), blocks); 180 } 181 182 /* Initializes an uninitialized block bitmap */ 183 static int ext4_init_block_bitmap(struct super_block *sb, 184 struct buffer_head *bh, 185 ext4_group_t block_group, 186 struct ext4_group_desc *gdp) 187 { 188 unsigned int bit, bit_max; 189 struct ext4_sb_info *sbi = EXT4_SB(sb); 190 ext4_fsblk_t start, tmp; 191 192 ASSERT(buffer_locked(bh)); 193 194 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { 195 ext4_mark_group_bitmap_corrupted(sb, block_group, 196 EXT4_GROUP_INFO_BBITMAP_CORRUPT | 197 EXT4_GROUP_INFO_IBITMAP_CORRUPT); 198 return -EFSBADCRC; 199 } 200 memset(bh->b_data, 0, sb->s_blocksize); 201 202 bit_max = ext4_num_base_meta_clusters(sb, block_group); 203 if ((bit_max >> 3) >= bh->b_size) 204 return -EFSCORRUPTED; 205 206 for (bit = 0; bit < bit_max; bit++) 207 ext4_set_bit(bit, bh->b_data); 208 209 start = ext4_group_first_block_no(sb, block_group); 210 211 /* Set bits for block and inode bitmaps, and inode table */ 212 tmp = ext4_block_bitmap(sb, gdp); 213 if (ext4_block_in_group(sb, tmp, block_group)) 214 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 215 216 tmp = ext4_inode_bitmap(sb, gdp); 217 if (ext4_block_in_group(sb, tmp, block_group)) 218 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 219 220 tmp = ext4_inode_table(sb, gdp); 221 for (; tmp < ext4_inode_table(sb, gdp) + 222 sbi->s_itb_per_group; tmp++) { 223 if (ext4_block_in_group(sb, tmp, block_group)) 224 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 225 } 226 227 /* 228 * Also if the number of blocks within the group is less than 229 * the blocksize * 8 ( which is the size of bitmap ), set rest 230 * of the block bitmap to 1 231 */ 232 ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group), 233 sb->s_blocksize * 8, bh->b_data); 234 return 0; 235 } 236 237 /* Return the number of free blocks in a block group. It is used when 238 * the block bitmap is uninitialized, so we can't just count the bits 239 * in the bitmap. */ 240 unsigned ext4_free_clusters_after_init(struct super_block *sb, 241 ext4_group_t block_group, 242 struct ext4_group_desc *gdp) 243 { 244 return num_clusters_in_group(sb, block_group) - 245 ext4_num_overhead_clusters(sb, block_group, gdp); 246 } 247 248 /* 249 * The free blocks are managed by bitmaps. A file system contains several 250 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 251 * block for inodes, N blocks for the inode table and data blocks. 252 * 253 * The file system contains group descriptors which are located after the 254 * super block. Each descriptor contains the number of the bitmap block and 255 * the free blocks count in the block. The descriptors are loaded in memory 256 * when a file system is mounted (see ext4_fill_super). 257 */ 258 259 /** 260 * ext4_get_group_desc() -- load group descriptor from disk 261 * @sb: super block 262 * @block_group: given block group 263 * @bh: pointer to the buffer head to store the block 264 * group descriptor 265 */ 266 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb, 267 ext4_group_t block_group, 268 struct buffer_head **bh) 269 { 270 unsigned int group_desc; 271 unsigned int offset; 272 ext4_group_t ngroups = ext4_get_groups_count(sb); 273 struct ext4_group_desc *desc; 274 struct ext4_sb_info *sbi = EXT4_SB(sb); 275 struct buffer_head *bh_p; 276 277 if (block_group >= ngroups) { 278 ext4_error(sb, "block_group >= groups_count - block_group = %u," 279 " groups_count = %u", block_group, ngroups); 280 281 return NULL; 282 } 283 284 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); 285 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); 286 bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc); 287 /* 288 * sbi_array_rcu_deref returns with rcu unlocked, this is ok since 289 * the pointer being dereferenced won't be dereferenced again. By 290 * looking at the usage in add_new_gdb() the value isn't modified, 291 * just the pointer, and so it remains valid. 292 */ 293 if (!bh_p) { 294 ext4_error(sb, "Group descriptor not loaded - " 295 "block_group = %u, group_desc = %u, desc = %u", 296 block_group, group_desc, offset); 297 return NULL; 298 } 299 300 desc = (struct ext4_group_desc *)( 301 (__u8 *)bh_p->b_data + 302 offset * EXT4_DESC_SIZE(sb)); 303 if (bh) 304 *bh = bh_p; 305 return desc; 306 } 307 308 static ext4_fsblk_t ext4_valid_block_bitmap_padding(struct super_block *sb, 309 ext4_group_t block_group, 310 struct buffer_head *bh) 311 { 312 ext4_grpblk_t next_zero_bit; 313 unsigned long bitmap_size = sb->s_blocksize * 8; 314 unsigned int offset = num_clusters_in_group(sb, block_group); 315 316 if (bitmap_size <= offset) 317 return 0; 318 319 next_zero_bit = ext4_find_next_zero_bit(bh->b_data, bitmap_size, offset); 320 321 return (next_zero_bit < bitmap_size ? next_zero_bit : 0); 322 } 323 324 struct ext4_group_info *ext4_get_group_info(struct super_block *sb, 325 ext4_group_t group) 326 { 327 struct ext4_group_info **grp_info; 328 long indexv, indexh; 329 330 if (unlikely(group >= EXT4_SB(sb)->s_groups_count)) { 331 ext4_error(sb, "invalid group %u", group); 332 return NULL; 333 } 334 indexv = group >> (EXT4_DESC_PER_BLOCK_BITS(sb)); 335 indexh = group & ((EXT4_DESC_PER_BLOCK(sb)) - 1); 336 grp_info = sbi_array_rcu_deref(EXT4_SB(sb), s_group_info, indexv); 337 return grp_info[indexh]; 338 } 339 340 /* 341 * Return the block number which was discovered to be invalid, or 0 if 342 * the block bitmap is valid. 343 */ 344 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb, 345 struct ext4_group_desc *desc, 346 ext4_group_t block_group, 347 struct buffer_head *bh) 348 { 349 struct ext4_sb_info *sbi = EXT4_SB(sb); 350 ext4_grpblk_t offset; 351 ext4_grpblk_t next_zero_bit; 352 ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb); 353 ext4_fsblk_t blk; 354 ext4_fsblk_t group_first_block; 355 356 if (ext4_has_feature_flex_bg(sb)) { 357 /* with FLEX_BG, the inode/block bitmaps and itable 358 * blocks may not be in the group at all 359 * so the bitmap validation will be skipped for those groups 360 * or it has to also read the block group where the bitmaps 361 * are located to verify they are set. 362 */ 363 return 0; 364 } 365 group_first_block = ext4_group_first_block_no(sb, block_group); 366 367 /* check whether block bitmap block number is set */ 368 blk = ext4_block_bitmap(sb, desc); 369 offset = blk - group_first_block; 370 if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit || 371 !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data)) 372 /* bad block bitmap */ 373 return blk; 374 375 /* check whether the inode bitmap block number is set */ 376 blk = ext4_inode_bitmap(sb, desc); 377 offset = blk - group_first_block; 378 if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit || 379 !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data)) 380 /* bad block bitmap */ 381 return blk; 382 383 /* check whether the inode table block number is set */ 384 blk = ext4_inode_table(sb, desc); 385 offset = blk - group_first_block; 386 if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit || 387 EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) >= max_bit) 388 return blk; 389 next_zero_bit = ext4_find_next_zero_bit(bh->b_data, 390 EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1, 391 EXT4_B2C(sbi, offset)); 392 if (next_zero_bit < 393 EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1) 394 /* bad bitmap for inode tables */ 395 return blk; 396 return 0; 397 } 398 399 static int ext4_validate_block_bitmap(struct super_block *sb, 400 struct ext4_group_desc *desc, 401 ext4_group_t block_group, 402 struct buffer_head *bh) 403 { 404 ext4_fsblk_t blk; 405 struct ext4_group_info *grp; 406 407 if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) 408 return 0; 409 410 grp = ext4_get_group_info(sb, block_group); 411 412 if (buffer_verified(bh)) 413 return 0; 414 if (!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) 415 return -EFSCORRUPTED; 416 417 ext4_lock_group(sb, block_group); 418 if (buffer_verified(bh)) 419 goto verified; 420 if (unlikely(!ext4_block_bitmap_csum_verify(sb, desc, bh) || 421 ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) { 422 ext4_unlock_group(sb, block_group); 423 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group); 424 ext4_mark_group_bitmap_corrupted(sb, block_group, 425 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 426 return -EFSBADCRC; 427 } 428 blk = ext4_valid_block_bitmap(sb, desc, block_group, bh); 429 if (unlikely(blk != 0)) { 430 ext4_unlock_group(sb, block_group); 431 ext4_error(sb, "bg %u: block %llu: invalid block bitmap", 432 block_group, blk); 433 ext4_mark_group_bitmap_corrupted(sb, block_group, 434 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 435 return -EFSCORRUPTED; 436 } 437 blk = ext4_valid_block_bitmap_padding(sb, block_group, bh); 438 if (unlikely(blk != 0)) { 439 ext4_unlock_group(sb, block_group); 440 ext4_error(sb, "bg %u: block %llu: padding at end of block bitmap is not set", 441 block_group, blk); 442 ext4_mark_group_bitmap_corrupted(sb, block_group, 443 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 444 return -EFSCORRUPTED; 445 } 446 set_buffer_verified(bh); 447 verified: 448 ext4_unlock_group(sb, block_group); 449 return 0; 450 } 451 452 /** 453 * ext4_read_block_bitmap_nowait() 454 * @sb: super block 455 * @block_group: given block group 456 * @ignore_locked: ignore locked buffers 457 * 458 * Read the bitmap for a given block_group,and validate the 459 * bits for block/inode/inode tables are set in the bitmaps 460 * 461 * Return buffer_head on success or an ERR_PTR in case of failure. 462 */ 463 struct buffer_head * 464 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group, 465 bool ignore_locked) 466 { 467 struct ext4_group_desc *desc; 468 struct ext4_sb_info *sbi = EXT4_SB(sb); 469 struct buffer_head *bh; 470 ext4_fsblk_t bitmap_blk; 471 int err; 472 473 desc = ext4_get_group_desc(sb, block_group, NULL); 474 if (!desc) 475 return ERR_PTR(-EFSCORRUPTED); 476 bitmap_blk = ext4_block_bitmap(sb, desc); 477 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || 478 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { 479 ext4_error(sb, "Invalid block bitmap block %llu in " 480 "block_group %u", bitmap_blk, block_group); 481 ext4_mark_group_bitmap_corrupted(sb, block_group, 482 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 483 return ERR_PTR(-EFSCORRUPTED); 484 } 485 bh = sb_getblk(sb, bitmap_blk); 486 if (unlikely(!bh)) { 487 ext4_warning(sb, "Cannot get buffer for block bitmap - " 488 "block_group = %u, block_bitmap = %llu", 489 block_group, bitmap_blk); 490 return ERR_PTR(-ENOMEM); 491 } 492 493 if (ignore_locked && buffer_locked(bh)) { 494 /* buffer under IO already, return if called for prefetching */ 495 put_bh(bh); 496 return NULL; 497 } 498 499 if (bitmap_uptodate(bh)) 500 goto verify; 501 502 lock_buffer(bh); 503 if (bitmap_uptodate(bh)) { 504 unlock_buffer(bh); 505 goto verify; 506 } 507 ext4_lock_group(sb, block_group); 508 if (ext4_has_group_desc_csum(sb) && 509 (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { 510 if (block_group == 0) { 511 ext4_unlock_group(sb, block_group); 512 unlock_buffer(bh); 513 ext4_error(sb, "Block bitmap for bg 0 marked " 514 "uninitialized"); 515 err = -EFSCORRUPTED; 516 goto out; 517 } 518 err = ext4_init_block_bitmap(sb, bh, block_group, desc); 519 if (err) { 520 ext4_unlock_group(sb, block_group); 521 unlock_buffer(bh); 522 ext4_error(sb, "Failed to init block bitmap for group " 523 "%u: %d", block_group, err); 524 goto out; 525 } 526 set_bitmap_uptodate(bh); 527 set_buffer_uptodate(bh); 528 set_buffer_verified(bh); 529 ext4_unlock_group(sb, block_group); 530 unlock_buffer(bh); 531 return bh; 532 } 533 ext4_unlock_group(sb, block_group); 534 if (buffer_uptodate(bh)) { 535 /* 536 * if not uninit if bh is uptodate, 537 * bitmap is also uptodate 538 */ 539 set_bitmap_uptodate(bh); 540 unlock_buffer(bh); 541 goto verify; 542 } 543 /* 544 * submit the buffer_head for reading 545 */ 546 set_buffer_new(bh); 547 trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked); 548 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO | 549 (ignore_locked ? REQ_RAHEAD : 0), 550 ext4_end_bitmap_read); 551 return bh; 552 verify: 553 err = ext4_validate_block_bitmap(sb, desc, block_group, bh); 554 if (err) 555 goto out; 556 return bh; 557 out: 558 put_bh(bh); 559 return ERR_PTR(err); 560 } 561 562 /* Returns 0 on success, -errno on error */ 563 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group, 564 struct buffer_head *bh) 565 { 566 struct ext4_group_desc *desc; 567 568 if (!buffer_new(bh)) 569 return 0; 570 desc = ext4_get_group_desc(sb, block_group, NULL); 571 if (!desc) 572 return -EFSCORRUPTED; 573 wait_on_buffer(bh); 574 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO); 575 if (!buffer_uptodate(bh)) { 576 ext4_error_err(sb, EIO, "Cannot read block bitmap - " 577 "block_group = %u, block_bitmap = %llu", 578 block_group, (unsigned long long) bh->b_blocknr); 579 ext4_mark_group_bitmap_corrupted(sb, block_group, 580 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 581 return -EIO; 582 } 583 clear_buffer_new(bh); 584 /* Panic or remount fs read-only if block bitmap is invalid */ 585 return ext4_validate_block_bitmap(sb, desc, block_group, bh); 586 } 587 588 struct buffer_head * 589 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group) 590 { 591 struct buffer_head *bh; 592 int err; 593 594 bh = ext4_read_block_bitmap_nowait(sb, block_group, false); 595 if (IS_ERR(bh)) 596 return bh; 597 err = ext4_wait_block_bitmap(sb, block_group, bh); 598 if (err) { 599 put_bh(bh); 600 return ERR_PTR(err); 601 } 602 return bh; 603 } 604 605 /** 606 * ext4_has_free_clusters() 607 * @sbi: in-core super block structure. 608 * @nclusters: number of needed blocks 609 * @flags: flags from ext4_mb_new_blocks() 610 * 611 * Check if filesystem has nclusters free & available for allocation. 612 * On success return 1, return 0 on failure. 613 */ 614 static int ext4_has_free_clusters(struct ext4_sb_info *sbi, 615 s64 nclusters, unsigned int flags) 616 { 617 s64 free_clusters, dirty_clusters, rsv, resv_clusters; 618 struct percpu_counter *fcc = &sbi->s_freeclusters_counter; 619 struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter; 620 621 free_clusters = percpu_counter_read_positive(fcc); 622 dirty_clusters = percpu_counter_read_positive(dcc); 623 resv_clusters = atomic64_read(&sbi->s_resv_clusters); 624 625 /* 626 * r_blocks_count should always be multiple of the cluster ratio so 627 * we are safe to do a plane bit shift only. 628 */ 629 rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) + 630 resv_clusters; 631 632 if (free_clusters - (nclusters + rsv + dirty_clusters) < 633 EXT4_FREECLUSTERS_WATERMARK) { 634 free_clusters = percpu_counter_sum_positive(fcc); 635 dirty_clusters = percpu_counter_sum_positive(dcc); 636 } 637 /* Check whether we have space after accounting for current 638 * dirty clusters & root reserved clusters. 639 */ 640 if (free_clusters >= (rsv + nclusters + dirty_clusters)) 641 return 1; 642 643 /* Hm, nope. Are (enough) root reserved clusters available? */ 644 if (uid_eq(sbi->s_resuid, current_fsuid()) || 645 (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) || 646 capable(CAP_SYS_RESOURCE) || 647 (flags & EXT4_MB_USE_ROOT_BLOCKS)) { 648 649 if (free_clusters >= (nclusters + dirty_clusters + 650 resv_clusters)) 651 return 1; 652 } 653 /* No free blocks. Let's see if we can dip into reserved pool */ 654 if (flags & EXT4_MB_USE_RESERVED) { 655 if (free_clusters >= (nclusters + dirty_clusters)) 656 return 1; 657 } 658 659 return 0; 660 } 661 662 int ext4_claim_free_clusters(struct ext4_sb_info *sbi, 663 s64 nclusters, unsigned int flags) 664 { 665 if (ext4_has_free_clusters(sbi, nclusters, flags)) { 666 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters); 667 return 0; 668 } else 669 return -ENOSPC; 670 } 671 672 /** 673 * ext4_should_retry_alloc() - check if a block allocation should be retried 674 * @sb: superblock 675 * @retries: number of retry attempts made so far 676 * 677 * ext4_should_retry_alloc() is called when ENOSPC is returned while 678 * attempting to allocate blocks. If there's an indication that a pending 679 * journal transaction might free some space and allow another attempt to 680 * succeed, this function will wait for the current or committing transaction 681 * to complete and then return TRUE. 682 */ 683 int ext4_should_retry_alloc(struct super_block *sb, int *retries) 684 { 685 struct ext4_sb_info *sbi = EXT4_SB(sb); 686 687 if (!sbi->s_journal) 688 return 0; 689 690 if (++(*retries) > 3) { 691 percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit); 692 return 0; 693 } 694 695 /* 696 * if there's no indication that blocks are about to be freed it's 697 * possible we just missed a transaction commit that did so 698 */ 699 smp_mb(); 700 if (sbi->s_mb_free_pending == 0) { 701 if (test_opt(sb, DISCARD)) { 702 atomic_inc(&sbi->s_retry_alloc_pending); 703 flush_work(&sbi->s_discard_work); 704 atomic_dec(&sbi->s_retry_alloc_pending); 705 } 706 return ext4_has_free_clusters(sbi, 1, 0); 707 } 708 709 /* 710 * it's possible we've just missed a transaction commit here, 711 * so ignore the returned status 712 */ 713 ext4_debug("%s: retrying operation after ENOSPC\n", sb->s_id); 714 (void) jbd2_journal_force_commit_nested(sbi->s_journal); 715 return 1; 716 } 717 718 /* 719 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks 720 * 721 * @handle: handle to this transaction 722 * @inode: file inode 723 * @goal: given target block(filesystem wide) 724 * @count: pointer to total number of clusters needed 725 * @errp: error code 726 * 727 * Return 1st allocated block number on success, *count stores total account 728 * error stores in errp pointer 729 */ 730 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode, 731 ext4_fsblk_t goal, unsigned int flags, 732 unsigned long *count, int *errp) 733 { 734 struct ext4_allocation_request ar; 735 ext4_fsblk_t ret; 736 737 memset(&ar, 0, sizeof(ar)); 738 /* Fill with neighbour allocated blocks */ 739 ar.inode = inode; 740 ar.goal = goal; 741 ar.len = count ? *count : 1; 742 ar.flags = flags; 743 744 ret = ext4_mb_new_blocks(handle, &ar, errp); 745 if (count) 746 *count = ar.len; 747 /* 748 * Account for the allocated meta blocks. We will never 749 * fail EDQUOT for metdata, but we do account for it. 750 */ 751 if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) { 752 dquot_alloc_block_nofail(inode, 753 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len)); 754 } 755 return ret; 756 } 757 758 /** 759 * ext4_count_free_clusters() -- count filesystem free clusters 760 * @sb: superblock 761 * 762 * Adds up the number of free clusters from each block group. 763 */ 764 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb) 765 { 766 ext4_fsblk_t desc_count; 767 struct ext4_group_desc *gdp; 768 ext4_group_t i; 769 ext4_group_t ngroups = ext4_get_groups_count(sb); 770 struct ext4_group_info *grp; 771 #ifdef EXT4FS_DEBUG 772 struct ext4_super_block *es; 773 ext4_fsblk_t bitmap_count; 774 unsigned int x; 775 struct buffer_head *bitmap_bh = NULL; 776 777 es = EXT4_SB(sb)->s_es; 778 desc_count = 0; 779 bitmap_count = 0; 780 gdp = NULL; 781 782 for (i = 0; i < ngroups; i++) { 783 gdp = ext4_get_group_desc(sb, i, NULL); 784 if (!gdp) 785 continue; 786 grp = NULL; 787 if (EXT4_SB(sb)->s_group_info) 788 grp = ext4_get_group_info(sb, i); 789 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) 790 desc_count += ext4_free_group_clusters(sb, gdp); 791 brelse(bitmap_bh); 792 bitmap_bh = ext4_read_block_bitmap(sb, i); 793 if (IS_ERR(bitmap_bh)) { 794 bitmap_bh = NULL; 795 continue; 796 } 797 798 x = ext4_count_free(bitmap_bh->b_data, 799 EXT4_CLUSTERS_PER_GROUP(sb) / 8); 800 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n", 801 i, ext4_free_group_clusters(sb, gdp), x); 802 bitmap_count += x; 803 } 804 brelse(bitmap_bh); 805 printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu" 806 ", computed = %llu, %llu\n", 807 EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)), 808 desc_count, bitmap_count); 809 return bitmap_count; 810 #else 811 desc_count = 0; 812 for (i = 0; i < ngroups; i++) { 813 gdp = ext4_get_group_desc(sb, i, NULL); 814 if (!gdp) 815 continue; 816 grp = NULL; 817 if (EXT4_SB(sb)->s_group_info) 818 grp = ext4_get_group_info(sb, i); 819 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) 820 desc_count += ext4_free_group_clusters(sb, gdp); 821 } 822 823 return desc_count; 824 #endif 825 } 826 827 static inline int test_root(ext4_group_t a, int b) 828 { 829 while (1) { 830 if (a < b) 831 return 0; 832 if (a == b) 833 return 1; 834 if ((a % b) != 0) 835 return 0; 836 a = a / b; 837 } 838 } 839 840 /** 841 * ext4_bg_has_super - number of blocks used by the superblock in group 842 * @sb: superblock for filesystem 843 * @group: group number to check 844 * 845 * Return the number of blocks used by the superblock (primary or backup) 846 * in this group. Currently this will be only 0 or 1. 847 */ 848 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) 849 { 850 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 851 852 if (group == 0) 853 return 1; 854 if (ext4_has_feature_sparse_super2(sb)) { 855 if (group == le32_to_cpu(es->s_backup_bgs[0]) || 856 group == le32_to_cpu(es->s_backup_bgs[1])) 857 return 1; 858 return 0; 859 } 860 if ((group <= 1) || !ext4_has_feature_sparse_super(sb)) 861 return 1; 862 if (!(group & 1)) 863 return 0; 864 if (test_root(group, 3) || (test_root(group, 5)) || 865 test_root(group, 7)) 866 return 1; 867 868 return 0; 869 } 870 871 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, 872 ext4_group_t group) 873 { 874 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); 875 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); 876 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; 877 878 if (group == first || group == first + 1 || group == last) 879 return 1; 880 return 0; 881 } 882 883 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, 884 ext4_group_t group) 885 { 886 if (!ext4_bg_has_super(sb, group)) 887 return 0; 888 889 return EXT4_SB(sb)->s_gdb_count; 890 } 891 892 /** 893 * ext4_bg_num_gdb - number of blocks used by the group table in group 894 * @sb: superblock for filesystem 895 * @group: group number to check 896 * 897 * Return the number of blocks used by the group descriptor table 898 * (primary or backup) in this group. In the future there may be a 899 * different number of descriptor blocks in each group. 900 */ 901 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) 902 { 903 unsigned long first_meta_bg = 904 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); 905 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); 906 907 if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) 908 return ext4_bg_num_gdb_nometa(sb, group); 909 910 return ext4_bg_num_gdb_meta(sb,group); 911 912 } 913 914 /* 915 * This function returns the number of file system metadata clusters at 916 * the beginning of a block group, including the reserved gdt blocks. 917 */ 918 static unsigned ext4_num_base_meta_clusters(struct super_block *sb, 919 ext4_group_t block_group) 920 { 921 struct ext4_sb_info *sbi = EXT4_SB(sb); 922 unsigned num; 923 924 /* Check for superblock and gdt backups in this group */ 925 num = ext4_bg_has_super(sb, block_group); 926 927 if (!ext4_has_feature_meta_bg(sb) || 928 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * 929 sbi->s_desc_per_block) { 930 if (num) { 931 num += ext4_bg_num_gdb_nometa(sb, block_group); 932 num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); 933 } 934 } else { /* For META_BG_BLOCK_GROUPS */ 935 num += ext4_bg_num_gdb_meta(sb, block_group); 936 } 937 return EXT4_NUM_B2C(sbi, num); 938 } 939 /** 940 * ext4_inode_to_goal_block - return a hint for block allocation 941 * @inode: inode for block allocation 942 * 943 * Return the ideal location to start allocating blocks for a 944 * newly created inode. 945 */ 946 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode) 947 { 948 struct ext4_inode_info *ei = EXT4_I(inode); 949 ext4_group_t block_group; 950 ext4_grpblk_t colour; 951 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); 952 ext4_fsblk_t bg_start; 953 ext4_fsblk_t last_block; 954 955 block_group = ei->i_block_group; 956 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { 957 /* 958 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME 959 * block groups per flexgroup, reserve the first block 960 * group for directories and special files. Regular 961 * files will start at the second block group. This 962 * tends to speed up directory access and improves 963 * fsck times. 964 */ 965 block_group &= ~(flex_size-1); 966 if (S_ISREG(inode->i_mode)) 967 block_group++; 968 } 969 bg_start = ext4_group_first_block_no(inode->i_sb, block_group); 970 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; 971 972 /* 973 * If we are doing delayed allocation, we don't need take 974 * colour into account. 975 */ 976 if (test_opt(inode->i_sb, DELALLOC)) 977 return bg_start; 978 979 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) 980 colour = (task_pid_nr(current) % 16) * 981 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); 982 else 983 colour = (task_pid_nr(current) % 16) * 984 ((last_block - bg_start) / 16); 985 return bg_start + colour; 986 } 987 988