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 #include <kunit/static_stub.h> 26 27 static unsigned ext4_num_base_meta_clusters(struct super_block *sb, 28 ext4_group_t block_group); 29 /* 30 * balloc.c contains the blocks allocation and deallocation routines 31 */ 32 33 /* 34 * Calculate block group number for a given block number 35 */ 36 ext4_group_t ext4_get_group_number(struct super_block *sb, 37 ext4_fsblk_t block) 38 { 39 ext4_group_t group; 40 41 if (test_opt2(sb, STD_GROUP_SIZE)) 42 group = (block - 43 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >> 44 (EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3); 45 else 46 ext4_get_group_no_and_offset(sb, block, &group, NULL); 47 return group; 48 } 49 50 /* 51 * Calculate the block group number and offset into the block/cluster 52 * allocation bitmap, given a block number 53 */ 54 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr, 55 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp) 56 { 57 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 58 ext4_grpblk_t offset; 59 60 blocknr = blocknr - le32_to_cpu(es->s_first_data_block); 61 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >> 62 EXT4_SB(sb)->s_cluster_bits; 63 if (offsetp) 64 *offsetp = offset; 65 if (blockgrpp) 66 *blockgrpp = blocknr; 67 68 } 69 70 /* 71 * Check whether the 'block' lives within the 'block_group'. Returns 1 if so 72 * and 0 otherwise. 73 */ 74 static inline int ext4_block_in_group(struct super_block *sb, 75 ext4_fsblk_t block, 76 ext4_group_t block_group) 77 { 78 ext4_group_t actual_group; 79 80 actual_group = ext4_get_group_number(sb, block); 81 return (actual_group == block_group) ? 1 : 0; 82 } 83 84 /* 85 * Return the number of clusters used for file system metadata; this 86 * represents the overhead needed by the file system. 87 */ 88 static unsigned ext4_num_overhead_clusters(struct super_block *sb, 89 ext4_group_t block_group, 90 struct ext4_group_desc *gdp) 91 { 92 unsigned base_clusters, num_clusters; 93 int block_cluster = -1, inode_cluster; 94 int itbl_cluster_start = -1, itbl_cluster_end = -1; 95 ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group); 96 ext4_fsblk_t end = start + EXT4_BLOCKS_PER_GROUP(sb) - 1; 97 ext4_fsblk_t itbl_blk_start, itbl_blk_end; 98 struct ext4_sb_info *sbi = EXT4_SB(sb); 99 100 /* This is the number of clusters used by the superblock, 101 * block group descriptors, and reserved block group 102 * descriptor blocks */ 103 base_clusters = ext4_num_base_meta_clusters(sb, block_group); 104 num_clusters = base_clusters; 105 106 /* 107 * Account and record inode table clusters if any cluster 108 * is in the block group, or inode table cluster range is 109 * [-1, -1] and won't overlap with block/inode bitmap cluster 110 * accounted below. 111 */ 112 itbl_blk_start = ext4_inode_table(sb, gdp); 113 itbl_blk_end = itbl_blk_start + sbi->s_itb_per_group - 1; 114 if (itbl_blk_start <= end && itbl_blk_end >= start) { 115 itbl_blk_start = max(itbl_blk_start, start); 116 itbl_blk_end = min(itbl_blk_end, end); 117 118 itbl_cluster_start = EXT4_B2C(sbi, itbl_blk_start - start); 119 itbl_cluster_end = EXT4_B2C(sbi, itbl_blk_end - start); 120 121 num_clusters += itbl_cluster_end - itbl_cluster_start + 1; 122 /* check if border cluster is overlapped */ 123 if (itbl_cluster_start == base_clusters - 1) 124 num_clusters--; 125 } 126 127 /* 128 * For the allocation bitmaps, we first need to check to see 129 * if the block is in the block group. If it is, then check 130 * to see if the cluster is already accounted for in the clusters 131 * used for the base metadata cluster and inode tables cluster. 132 * Normally all of these blocks are contiguous, so the special 133 * case handling shouldn't be necessary except for *very* 134 * unusual file system layouts. 135 */ 136 if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) { 137 block_cluster = EXT4_B2C(sbi, 138 ext4_block_bitmap(sb, gdp) - start); 139 if (block_cluster >= base_clusters && 140 (block_cluster < itbl_cluster_start || 141 block_cluster > itbl_cluster_end)) 142 num_clusters++; 143 } 144 145 if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) { 146 inode_cluster = EXT4_B2C(sbi, 147 ext4_inode_bitmap(sb, gdp) - start); 148 /* 149 * Additional check if inode bitmap is in just accounted 150 * block_cluster 151 */ 152 if (inode_cluster != block_cluster && 153 inode_cluster >= base_clusters && 154 (inode_cluster < itbl_cluster_start || 155 inode_cluster > itbl_cluster_end)) 156 num_clusters++; 157 } 158 159 return num_clusters; 160 } 161 162 static unsigned int num_clusters_in_group(struct super_block *sb, 163 ext4_group_t block_group) 164 { 165 unsigned int blocks; 166 167 if (block_group == ext4_get_groups_count(sb) - 1) { 168 /* 169 * Even though mke2fs always initializes the first and 170 * last group, just in case some other tool was used, 171 * we need to make sure we calculate the right free 172 * blocks. 173 */ 174 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) - 175 ext4_group_first_block_no(sb, block_group); 176 } else 177 blocks = EXT4_BLOCKS_PER_GROUP(sb); 178 return EXT4_NUM_B2C(EXT4_SB(sb), blocks); 179 } 180 181 /* Initializes an uninitialized block bitmap */ 182 static int ext4_init_block_bitmap(struct super_block *sb, 183 struct buffer_head *bh, 184 ext4_group_t block_group, 185 struct ext4_group_desc *gdp) 186 { 187 unsigned int bit, bit_max; 188 struct ext4_sb_info *sbi = EXT4_SB(sb); 189 ext4_fsblk_t start, tmp; 190 191 ASSERT(buffer_locked(bh)); 192 193 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { 194 ext4_mark_group_bitmap_corrupted(sb, block_group, 195 EXT4_GROUP_INFO_BBITMAP_CORRUPT | 196 EXT4_GROUP_INFO_IBITMAP_CORRUPT); 197 return -EFSBADCRC; 198 } 199 memset(bh->b_data, 0, sb->s_blocksize); 200 201 bit_max = ext4_num_base_meta_clusters(sb, block_group); 202 if ((bit_max >> 3) >= bh->b_size) 203 return -EFSCORRUPTED; 204 205 for (bit = 0; bit < bit_max; bit++) 206 ext4_set_bit(bit, bh->b_data); 207 208 start = ext4_group_first_block_no(sb, block_group); 209 210 /* Set bits for block and inode bitmaps, and inode table */ 211 tmp = ext4_block_bitmap(sb, gdp); 212 if (ext4_block_in_group(sb, tmp, block_group)) 213 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 214 215 tmp = ext4_inode_bitmap(sb, gdp); 216 if (ext4_block_in_group(sb, tmp, block_group)) 217 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 218 219 tmp = ext4_inode_table(sb, gdp); 220 for (; tmp < ext4_inode_table(sb, gdp) + 221 sbi->s_itb_per_group; tmp++) { 222 if (ext4_block_in_group(sb, tmp, block_group)) 223 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 224 } 225 226 /* 227 * Also if the number of blocks within the group is less than 228 * the blocksize * 8 ( which is the size of bitmap ), set rest 229 * of the block bitmap to 1 230 */ 231 ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group), 232 sb->s_blocksize * 8, bh->b_data); 233 return 0; 234 } 235 236 /* Return the number of free blocks in a block group. It is used when 237 * the block bitmap is uninitialized, so we can't just count the bits 238 * in the bitmap. */ 239 unsigned ext4_free_clusters_after_init(struct super_block *sb, 240 ext4_group_t block_group, 241 struct ext4_group_desc *gdp) 242 { 243 return num_clusters_in_group(sb, block_group) - 244 ext4_num_overhead_clusters(sb, block_group, gdp); 245 } 246 247 /* 248 * The free blocks are managed by bitmaps. A file system contains several 249 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 250 * block for inodes, N blocks for the inode table and data blocks. 251 * 252 * The file system contains group descriptors which are located after the 253 * super block. Each descriptor contains the number of the bitmap block and 254 * the free blocks count in the block. The descriptors are loaded in memory 255 * when a file system is mounted (see ext4_fill_super). 256 */ 257 258 /** 259 * ext4_get_group_desc() -- load group descriptor from disk 260 * @sb: super block 261 * @block_group: given block group 262 * @bh: pointer to the buffer head to store the block 263 * group descriptor 264 */ 265 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb, 266 ext4_group_t block_group, 267 struct buffer_head **bh) 268 { 269 unsigned int group_desc; 270 unsigned int offset; 271 ext4_group_t ngroups = ext4_get_groups_count(sb); 272 struct ext4_group_desc *desc; 273 struct ext4_sb_info *sbi = EXT4_SB(sb); 274 struct buffer_head *bh_p; 275 276 KUNIT_STATIC_STUB_REDIRECT(ext4_get_group_desc, 277 sb, block_group, bh); 278 279 if (block_group >= ngroups) { 280 ext4_error(sb, "block_group >= groups_count - block_group = %u," 281 " groups_count = %u", block_group, ngroups); 282 283 return NULL; 284 } 285 286 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); 287 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); 288 bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc); 289 /* 290 * sbi_array_rcu_deref returns with rcu unlocked, this is ok since 291 * the pointer being dereferenced won't be dereferenced again. By 292 * looking at the usage in add_new_gdb() the value isn't modified, 293 * just the pointer, and so it remains valid. 294 */ 295 if (!bh_p) { 296 ext4_error(sb, "Group descriptor not loaded - " 297 "block_group = %u, group_desc = %u, desc = %u", 298 block_group, group_desc, offset); 299 return NULL; 300 } 301 302 desc = (struct ext4_group_desc *)( 303 (__u8 *)bh_p->b_data + 304 offset * EXT4_DESC_SIZE(sb)); 305 if (bh) 306 *bh = bh_p; 307 return desc; 308 } 309 310 static ext4_fsblk_t ext4_valid_block_bitmap_padding(struct super_block *sb, 311 ext4_group_t block_group, 312 struct buffer_head *bh) 313 { 314 ext4_grpblk_t next_zero_bit; 315 unsigned long bitmap_size = sb->s_blocksize * 8; 316 unsigned int offset = num_clusters_in_group(sb, block_group); 317 318 if (bitmap_size <= offset) 319 return 0; 320 321 next_zero_bit = ext4_find_next_zero_bit(bh->b_data, bitmap_size, offset); 322 323 return (next_zero_bit < bitmap_size ? next_zero_bit : 0); 324 } 325 326 struct ext4_group_info *ext4_get_group_info(struct super_block *sb, 327 ext4_group_t group) 328 { 329 struct ext4_group_info **grp_info; 330 long indexv, indexh; 331 332 if (unlikely(group >= EXT4_SB(sb)->s_groups_count)) 333 return NULL; 334 if (unlikely(!EXT4_SB(sb)->s_group_info)) 335 return NULL; 336 indexv = group >> (EXT4_DESC_PER_BLOCK_BITS(sb)); 337 indexh = group & ((EXT4_DESC_PER_BLOCK(sb)) - 1); 338 grp_info = sbi_array_rcu_deref(EXT4_SB(sb), s_group_info, indexv); 339 if (unlikely(!grp_info)) 340 return NULL; 341 return grp_info[indexh]; 342 } 343 344 /* 345 * Return the block number which was discovered to be invalid, or 0 if 346 * the block bitmap is valid. 347 */ 348 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb, 349 struct ext4_group_desc *desc, 350 ext4_group_t block_group, 351 struct buffer_head *bh) 352 { 353 struct ext4_sb_info *sbi = EXT4_SB(sb); 354 ext4_grpblk_t offset; 355 ext4_grpblk_t next_zero_bit; 356 ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb); 357 ext4_fsblk_t blk; 358 ext4_fsblk_t group_first_block; 359 360 if (ext4_has_feature_flex_bg(sb)) { 361 /* with FLEX_BG, the inode/block bitmaps and itable 362 * blocks may not be in the group at all 363 * so the bitmap validation will be skipped for those groups 364 * or it has to also read the block group where the bitmaps 365 * are located to verify they are set. 366 */ 367 return 0; 368 } 369 group_first_block = ext4_group_first_block_no(sb, block_group); 370 371 /* check whether block bitmap block number is set */ 372 blk = ext4_block_bitmap(sb, desc); 373 offset = blk - group_first_block; 374 if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit || 375 !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data)) 376 /* bad block bitmap */ 377 return blk; 378 379 /* check whether the inode bitmap block number is set */ 380 blk = ext4_inode_bitmap(sb, desc); 381 offset = blk - group_first_block; 382 if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit || 383 !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data)) 384 /* bad block bitmap */ 385 return blk; 386 387 /* check whether the inode table block number is set */ 388 blk = ext4_inode_table(sb, desc); 389 offset = blk - group_first_block; 390 if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit || 391 EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) >= max_bit) 392 return blk; 393 next_zero_bit = ext4_find_next_zero_bit(bh->b_data, 394 EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1, 395 EXT4_B2C(sbi, offset)); 396 if (next_zero_bit < 397 EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1) 398 /* bad bitmap for inode tables */ 399 return blk; 400 return 0; 401 } 402 403 static int ext4_validate_block_bitmap(struct super_block *sb, 404 struct ext4_group_desc *desc, 405 ext4_group_t block_group, 406 struct buffer_head *bh) 407 { 408 ext4_fsblk_t blk; 409 struct ext4_group_info *grp; 410 411 if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) 412 return 0; 413 414 grp = ext4_get_group_info(sb, block_group); 415 416 if (buffer_verified(bh)) 417 return 0; 418 if (!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) 419 return -EFSCORRUPTED; 420 421 ext4_lock_group(sb, block_group); 422 if (buffer_verified(bh)) 423 goto verified; 424 if (unlikely(!ext4_block_bitmap_csum_verify(sb, desc, bh) || 425 ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) { 426 ext4_unlock_group(sb, block_group); 427 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group); 428 ext4_mark_group_bitmap_corrupted(sb, block_group, 429 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 430 return -EFSBADCRC; 431 } 432 blk = ext4_valid_block_bitmap(sb, desc, block_group, bh); 433 if (unlikely(blk != 0)) { 434 ext4_unlock_group(sb, block_group); 435 ext4_error(sb, "bg %u: block %llu: invalid block bitmap", 436 block_group, blk); 437 ext4_mark_group_bitmap_corrupted(sb, block_group, 438 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 439 return -EFSCORRUPTED; 440 } 441 blk = ext4_valid_block_bitmap_padding(sb, block_group, bh); 442 if (unlikely(blk != 0)) { 443 ext4_unlock_group(sb, block_group); 444 ext4_error(sb, "bg %u: block %llu: padding at end of block bitmap is not set", 445 block_group, blk); 446 ext4_mark_group_bitmap_corrupted(sb, block_group, 447 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 448 return -EFSCORRUPTED; 449 } 450 set_buffer_verified(bh); 451 verified: 452 ext4_unlock_group(sb, block_group); 453 return 0; 454 } 455 456 /** 457 * ext4_read_block_bitmap_nowait() 458 * @sb: super block 459 * @block_group: given block group 460 * @ignore_locked: ignore locked buffers 461 * 462 * Read the bitmap for a given block_group,and validate the 463 * bits for block/inode/inode tables are set in the bitmaps 464 * 465 * Return buffer_head on success or an ERR_PTR in case of failure. 466 */ 467 struct buffer_head * 468 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group, 469 bool ignore_locked) 470 { 471 struct ext4_group_desc *desc; 472 struct ext4_sb_info *sbi = EXT4_SB(sb); 473 struct buffer_head *bh; 474 ext4_fsblk_t bitmap_blk; 475 int err; 476 477 KUNIT_STATIC_STUB_REDIRECT(ext4_read_block_bitmap_nowait, 478 sb, block_group, ignore_locked); 479 480 desc = ext4_get_group_desc(sb, block_group, NULL); 481 if (!desc) 482 return ERR_PTR(-EFSCORRUPTED); 483 bitmap_blk = ext4_block_bitmap(sb, desc); 484 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || 485 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { 486 ext4_error(sb, "Invalid block bitmap block %llu in " 487 "block_group %u", bitmap_blk, block_group); 488 ext4_mark_group_bitmap_corrupted(sb, block_group, 489 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 490 return ERR_PTR(-EFSCORRUPTED); 491 } 492 bh = sb_getblk(sb, bitmap_blk); 493 if (unlikely(!bh)) { 494 ext4_warning(sb, "Cannot get buffer for block bitmap - " 495 "block_group = %u, block_bitmap = %llu", 496 block_group, bitmap_blk); 497 return ERR_PTR(-ENOMEM); 498 } 499 500 if (ignore_locked && buffer_locked(bh)) { 501 /* buffer under IO already, return if called for prefetching */ 502 put_bh(bh); 503 return NULL; 504 } 505 506 if (bitmap_uptodate(bh)) 507 goto verify; 508 509 lock_buffer(bh); 510 if (bitmap_uptodate(bh)) { 511 unlock_buffer(bh); 512 goto verify; 513 } 514 ext4_lock_group(sb, block_group); 515 if (ext4_has_group_desc_csum(sb) && 516 (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { 517 if (block_group == 0) { 518 ext4_unlock_group(sb, block_group); 519 unlock_buffer(bh); 520 ext4_error(sb, "Block bitmap for bg 0 marked " 521 "uninitialized"); 522 err = -EFSCORRUPTED; 523 goto out; 524 } 525 err = ext4_init_block_bitmap(sb, bh, block_group, desc); 526 if (err) { 527 ext4_unlock_group(sb, block_group); 528 unlock_buffer(bh); 529 ext4_error(sb, "Failed to init block bitmap for group " 530 "%u: %d", block_group, err); 531 goto out; 532 } 533 set_bitmap_uptodate(bh); 534 set_buffer_uptodate(bh); 535 set_buffer_verified(bh); 536 ext4_unlock_group(sb, block_group); 537 unlock_buffer(bh); 538 return bh; 539 } 540 ext4_unlock_group(sb, block_group); 541 if (buffer_uptodate(bh)) { 542 /* 543 * if not uninit if bh is uptodate, 544 * bitmap is also uptodate 545 */ 546 set_bitmap_uptodate(bh); 547 unlock_buffer(bh); 548 goto verify; 549 } 550 /* 551 * submit the buffer_head for reading 552 */ 553 set_buffer_new(bh); 554 trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked); 555 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO | 556 (ignore_locked ? REQ_RAHEAD : 0), 557 ext4_end_bitmap_read, 558 ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_EIO)); 559 return bh; 560 verify: 561 err = ext4_validate_block_bitmap(sb, desc, block_group, bh); 562 if (err) 563 goto out; 564 return bh; 565 out: 566 put_bh(bh); 567 return ERR_PTR(err); 568 } 569 570 /* Returns 0 on success, -errno on error */ 571 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group, 572 struct buffer_head *bh) 573 { 574 struct ext4_group_desc *desc; 575 576 KUNIT_STATIC_STUB_REDIRECT(ext4_wait_block_bitmap, 577 sb, block_group, bh); 578 579 if (!buffer_new(bh)) 580 return 0; 581 desc = ext4_get_group_desc(sb, block_group, NULL); 582 if (!desc) 583 return -EFSCORRUPTED; 584 wait_on_buffer(bh); 585 if (!buffer_uptodate(bh)) { 586 ext4_error_err(sb, EIO, "Cannot read block bitmap - " 587 "block_group = %u, block_bitmap = %llu", 588 block_group, (unsigned long long) bh->b_blocknr); 589 ext4_mark_group_bitmap_corrupted(sb, block_group, 590 EXT4_GROUP_INFO_BBITMAP_CORRUPT); 591 return -EIO; 592 } 593 clear_buffer_new(bh); 594 /* Panic or remount fs read-only if block bitmap is invalid */ 595 return ext4_validate_block_bitmap(sb, desc, block_group, bh); 596 } 597 598 struct buffer_head * 599 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group) 600 { 601 struct buffer_head *bh; 602 int err; 603 604 bh = ext4_read_block_bitmap_nowait(sb, block_group, false); 605 if (IS_ERR(bh)) 606 return bh; 607 err = ext4_wait_block_bitmap(sb, block_group, bh); 608 if (err) { 609 put_bh(bh); 610 return ERR_PTR(err); 611 } 612 return bh; 613 } 614 615 /** 616 * ext4_has_free_clusters() 617 * @sbi: in-core super block structure. 618 * @nclusters: number of needed blocks 619 * @flags: flags from ext4_mb_new_blocks() 620 * 621 * Check if filesystem has nclusters free & available for allocation. 622 * On success return 1, return 0 on failure. 623 */ 624 static int ext4_has_free_clusters(struct ext4_sb_info *sbi, 625 s64 nclusters, unsigned int flags) 626 { 627 s64 free_clusters, dirty_clusters, rsv, resv_clusters; 628 struct percpu_counter *fcc = &sbi->s_freeclusters_counter; 629 struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter; 630 631 free_clusters = percpu_counter_read_positive(fcc); 632 dirty_clusters = percpu_counter_read_positive(dcc); 633 resv_clusters = atomic64_read(&sbi->s_resv_clusters); 634 635 /* 636 * r_blocks_count should always be multiple of the cluster ratio so 637 * we are safe to do a plane bit shift only. 638 */ 639 rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) + 640 resv_clusters; 641 642 if (free_clusters - (nclusters + rsv + dirty_clusters) < 643 EXT4_FREECLUSTERS_WATERMARK) { 644 free_clusters = percpu_counter_sum_positive(fcc); 645 dirty_clusters = percpu_counter_sum_positive(dcc); 646 } 647 /* Check whether we have space after accounting for current 648 * dirty clusters & root reserved clusters. 649 */ 650 if (free_clusters >= (rsv + nclusters + dirty_clusters)) 651 return 1; 652 653 /* Hm, nope. Are (enough) root reserved clusters available? */ 654 if (uid_eq(sbi->s_resuid, current_fsuid()) || 655 (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) || 656 (flags & EXT4_MB_USE_ROOT_BLOCKS) || 657 capable(CAP_SYS_RESOURCE)) { 658 659 if (free_clusters >= (nclusters + dirty_clusters + 660 resv_clusters)) 661 return 1; 662 } 663 /* No free blocks. Let's see if we can dip into reserved pool */ 664 if (flags & EXT4_MB_USE_RESERVED) { 665 if (free_clusters >= (nclusters + dirty_clusters)) 666 return 1; 667 } 668 669 return 0; 670 } 671 672 int ext4_claim_free_clusters(struct ext4_sb_info *sbi, 673 s64 nclusters, unsigned int flags) 674 { 675 if (ext4_has_free_clusters(sbi, nclusters, flags)) { 676 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters); 677 return 0; 678 } else 679 return -ENOSPC; 680 } 681 682 /** 683 * ext4_should_retry_alloc() - check if a block allocation should be retried 684 * @sb: superblock 685 * @retries: number of retry attempts made so far 686 * 687 * ext4_should_retry_alloc() is called when ENOSPC is returned while 688 * attempting to allocate blocks. If there's an indication that a pending 689 * journal transaction might free some space and allow another attempt to 690 * succeed, this function will wait for the current or committing transaction 691 * to complete and then return TRUE. 692 */ 693 int ext4_should_retry_alloc(struct super_block *sb, int *retries) 694 { 695 struct ext4_sb_info *sbi = EXT4_SB(sb); 696 697 if (!sbi->s_journal) 698 return 0; 699 700 if (++(*retries) > 3) { 701 percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit); 702 return 0; 703 } 704 705 /* 706 * if there's no indication that blocks are about to be freed it's 707 * possible we just missed a transaction commit that did so 708 */ 709 smp_mb(); 710 if (atomic_read(&sbi->s_mb_free_pending) == 0) { 711 if (test_opt(sb, DISCARD)) { 712 atomic_inc(&sbi->s_retry_alloc_pending); 713 flush_work(&sbi->s_discard_work); 714 atomic_dec(&sbi->s_retry_alloc_pending); 715 } 716 return ext4_has_free_clusters(sbi, 1, 0); 717 } 718 719 /* 720 * it's possible we've just missed a transaction commit here, 721 * so ignore the returned status 722 */ 723 ext4_debug("%s: retrying operation after ENOSPC\n", sb->s_id); 724 (void) jbd2_journal_force_commit_nested(sbi->s_journal); 725 return 1; 726 } 727 728 /* 729 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks 730 * 731 * @handle: handle to this transaction 732 * @inode: file inode 733 * @goal: given target block(filesystem wide) 734 * @count: pointer to total number of clusters needed 735 * @errp: error code 736 * 737 * Return 1st allocated block number on success, *count stores total account 738 * error stores in errp pointer 739 */ 740 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode, 741 ext4_fsblk_t goal, unsigned int flags, 742 unsigned long *count, int *errp) 743 { 744 struct ext4_allocation_request ar; 745 ext4_fsblk_t ret; 746 747 memset(&ar, 0, sizeof(ar)); 748 /* Fill with neighbour allocated blocks */ 749 ar.inode = inode; 750 ar.goal = goal; 751 ar.len = count ? *count : 1; 752 ar.flags = flags; 753 754 ret = ext4_mb_new_blocks(handle, &ar, errp); 755 if (count) 756 *count = ar.len; 757 /* 758 * Account for the allocated meta blocks. We will never 759 * fail EDQUOT for metadata, but we do account for it. 760 */ 761 if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) { 762 dquot_alloc_block_nofail(inode, 763 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len)); 764 } 765 return ret; 766 } 767 768 /** 769 * ext4_count_free_clusters() -- count filesystem free clusters 770 * @sb: superblock 771 * 772 * Adds up the number of free clusters from each block group. 773 */ 774 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb) 775 { 776 ext4_fsblk_t desc_count; 777 struct ext4_group_desc *gdp; 778 ext4_group_t i; 779 ext4_group_t ngroups = ext4_get_groups_count(sb); 780 struct ext4_group_info *grp; 781 #ifdef EXT4FS_DEBUG 782 struct ext4_super_block *es; 783 ext4_fsblk_t bitmap_count; 784 unsigned int x; 785 struct buffer_head *bitmap_bh = NULL; 786 787 es = EXT4_SB(sb)->s_es; 788 desc_count = 0; 789 bitmap_count = 0; 790 gdp = NULL; 791 792 for (i = 0; i < ngroups; i++) { 793 gdp = ext4_get_group_desc(sb, i, NULL); 794 if (!gdp) 795 continue; 796 grp = NULL; 797 if (EXT4_SB(sb)->s_group_info) 798 grp = ext4_get_group_info(sb, i); 799 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) 800 desc_count += ext4_free_group_clusters(sb, gdp); 801 brelse(bitmap_bh); 802 bitmap_bh = ext4_read_block_bitmap(sb, i); 803 if (IS_ERR(bitmap_bh)) { 804 bitmap_bh = NULL; 805 continue; 806 } 807 808 x = ext4_count_free(bitmap_bh->b_data, 809 EXT4_CLUSTERS_PER_GROUP(sb) / 8); 810 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n", 811 i, ext4_free_group_clusters(sb, gdp), x); 812 bitmap_count += x; 813 } 814 brelse(bitmap_bh); 815 printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu" 816 ", computed = %llu, %llu\n", 817 EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)), 818 desc_count, bitmap_count); 819 return bitmap_count; 820 #else 821 desc_count = 0; 822 for (i = 0; i < ngroups; i++) { 823 gdp = ext4_get_group_desc(sb, i, NULL); 824 if (!gdp) 825 continue; 826 grp = NULL; 827 if (EXT4_SB(sb)->s_group_info) 828 grp = ext4_get_group_info(sb, i); 829 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) 830 desc_count += ext4_free_group_clusters(sb, gdp); 831 } 832 833 return desc_count; 834 #endif 835 } 836 837 static inline int test_root(ext4_group_t a, int b) 838 { 839 while (1) { 840 if (a < b) 841 return 0; 842 if (a == b) 843 return 1; 844 if ((a % b) != 0) 845 return 0; 846 a = a / b; 847 } 848 } 849 850 /** 851 * ext4_bg_has_super - number of blocks used by the superblock in group 852 * @sb: superblock for filesystem 853 * @group: group number to check 854 * 855 * Return the number of blocks used by the superblock (primary or backup) 856 * in this group. Currently this will be only 0 or 1. 857 */ 858 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) 859 { 860 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 861 862 if (group == 0) 863 return 1; 864 if (ext4_has_feature_sparse_super2(sb)) { 865 if (group == le32_to_cpu(es->s_backup_bgs[0]) || 866 group == le32_to_cpu(es->s_backup_bgs[1])) 867 return 1; 868 return 0; 869 } 870 if ((group <= 1) || !ext4_has_feature_sparse_super(sb)) 871 return 1; 872 if (!(group & 1)) 873 return 0; 874 if (test_root(group, 3) || (test_root(group, 5)) || 875 test_root(group, 7)) 876 return 1; 877 878 return 0; 879 } 880 881 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, 882 ext4_group_t group) 883 { 884 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); 885 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); 886 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; 887 888 if (group == first || group == first + 1 || group == last) 889 return 1; 890 return 0; 891 } 892 893 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, 894 ext4_group_t group) 895 { 896 if (!ext4_bg_has_super(sb, group)) 897 return 0; 898 899 if (ext4_has_feature_meta_bg(sb)) 900 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); 901 else 902 return EXT4_SB(sb)->s_gdb_count; 903 } 904 905 /** 906 * ext4_bg_num_gdb - number of blocks used by the group table in group 907 * @sb: superblock for filesystem 908 * @group: group number to check 909 * 910 * Return the number of blocks used by the group descriptor table 911 * (primary or backup) in this group. In the future there may be a 912 * different number of descriptor blocks in each group. 913 */ 914 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) 915 { 916 unsigned long first_meta_bg = 917 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); 918 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); 919 920 if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) 921 return ext4_bg_num_gdb_nometa(sb, group); 922 923 return ext4_bg_num_gdb_meta(sb,group); 924 925 } 926 927 /* 928 * This function returns the number of file system metadata blocks at 929 * the beginning of a block group, including the reserved gdt blocks. 930 */ 931 unsigned int ext4_num_base_meta_blocks(struct super_block *sb, 932 ext4_group_t block_group) 933 { 934 struct ext4_sb_info *sbi = EXT4_SB(sb); 935 unsigned num; 936 937 /* Check for superblock and gdt backups in this group */ 938 num = ext4_bg_has_super(sb, block_group); 939 940 if (!ext4_has_feature_meta_bg(sb) || 941 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * 942 sbi->s_desc_per_block) { 943 if (num) { 944 num += ext4_bg_num_gdb_nometa(sb, block_group); 945 num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); 946 } 947 } else { /* For META_BG_BLOCK_GROUPS */ 948 num += ext4_bg_num_gdb_meta(sb, block_group); 949 } 950 return num; 951 } 952 953 static unsigned int ext4_num_base_meta_clusters(struct super_block *sb, 954 ext4_group_t block_group) 955 { 956 return EXT4_NUM_B2C(EXT4_SB(sb), ext4_num_base_meta_blocks(sb, block_group)); 957 } 958 959 /** 960 * ext4_inode_to_goal_block - return a hint for block allocation 961 * @inode: inode for block allocation 962 * 963 * Return the ideal location to start allocating blocks for a 964 * newly created inode. 965 */ 966 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode) 967 { 968 struct ext4_inode_info *ei = EXT4_I(inode); 969 ext4_group_t block_group; 970 ext4_grpblk_t colour; 971 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); 972 ext4_fsblk_t bg_start; 973 ext4_fsblk_t last_block; 974 975 block_group = ei->i_block_group; 976 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { 977 /* 978 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME 979 * block groups per flexgroup, reserve the first block 980 * group for directories and special files. Regular 981 * files will start at the second block group. This 982 * tends to speed up directory access and improves 983 * fsck times. 984 */ 985 block_group &= ~(flex_size-1); 986 if (S_ISREG(inode->i_mode)) 987 block_group++; 988 } 989 bg_start = ext4_group_first_block_no(inode->i_sb, block_group); 990 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; 991 992 /* 993 * If we are doing delayed allocation, we don't need take 994 * colour into account. 995 */ 996 if (test_opt(inode->i_sb, DELALLOC)) 997 return bg_start; 998 999 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) 1000 colour = (task_pid_nr(current) % 16) * 1001 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); 1002 else 1003 colour = (task_pid_nr(current) % 16) * 1004 ((last_block - bg_start) / 16); 1005 return bg_start + colour; 1006 } 1007 1008