1 /* 2 * linux/fs/ext4/balloc.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993 10 * Big-endian to little-endian byte-swapping/bitmaps by 11 * David S. Miller (davem@caip.rutgers.edu), 1995 12 */ 13 14 #include <linux/time.h> 15 #include <linux/capability.h> 16 #include <linux/fs.h> 17 #include <linux/jbd2.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 the block group number and offset into the block/cluster 34 * allocation bitmap, given a block number 35 */ 36 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr, 37 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp) 38 { 39 struct ext4_super_block *es = EXT4_SB(sb)->s_es; 40 ext4_grpblk_t offset; 41 42 blocknr = blocknr - le32_to_cpu(es->s_first_data_block); 43 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >> 44 EXT4_SB(sb)->s_cluster_bits; 45 if (offsetp) 46 *offsetp = offset; 47 if (blockgrpp) 48 *blockgrpp = blocknr; 49 50 } 51 52 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block, 53 ext4_group_t block_group) 54 { 55 ext4_group_t actual_group; 56 ext4_get_group_no_and_offset(sb, block, &actual_group, NULL); 57 if (actual_group == block_group) 58 return 1; 59 return 0; 60 } 61 62 /* Return the number of clusters used for file system metadata; this 63 * represents the overhead needed by the file system. 64 */ 65 unsigned ext4_num_overhead_clusters(struct super_block *sb, 66 ext4_group_t block_group, 67 struct ext4_group_desc *gdp) 68 { 69 unsigned num_clusters; 70 int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c; 71 ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group); 72 ext4_fsblk_t itbl_blk; 73 struct ext4_sb_info *sbi = EXT4_SB(sb); 74 75 /* This is the number of clusters used by the superblock, 76 * block group descriptors, and reserved block group 77 * descriptor blocks */ 78 num_clusters = ext4_num_base_meta_clusters(sb, block_group); 79 80 /* 81 * For the allocation bitmaps and inode table, we first need 82 * to check to see if the block is in the block group. If it 83 * is, then check to see if the cluster is already accounted 84 * for in the clusters used for the base metadata cluster, or 85 * if we can increment the base metadata cluster to include 86 * that block. Otherwise, we will have to track the cluster 87 * used for the allocation bitmap or inode table explicitly. 88 * Normally all of these blocks are contiguous, so the special 89 * case handling shouldn't be necessary except for *very* 90 * unusual file system layouts. 91 */ 92 if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) { 93 block_cluster = EXT4_B2C(sbi, 94 ext4_block_bitmap(sb, gdp) - start); 95 if (block_cluster < num_clusters) 96 block_cluster = -1; 97 else if (block_cluster == num_clusters) { 98 num_clusters++; 99 block_cluster = -1; 100 } 101 } 102 103 if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) { 104 inode_cluster = EXT4_B2C(sbi, 105 ext4_inode_bitmap(sb, gdp) - start); 106 if (inode_cluster < num_clusters) 107 inode_cluster = -1; 108 else if (inode_cluster == num_clusters) { 109 num_clusters++; 110 inode_cluster = -1; 111 } 112 } 113 114 itbl_blk = ext4_inode_table(sb, gdp); 115 for (i = 0; i < sbi->s_itb_per_group; i++) { 116 if (ext4_block_in_group(sb, itbl_blk + i, block_group)) { 117 c = EXT4_B2C(sbi, itbl_blk + i - start); 118 if ((c < num_clusters) || (c == inode_cluster) || 119 (c == block_cluster) || (c == itbl_cluster)) 120 continue; 121 if (c == num_clusters) { 122 num_clusters++; 123 continue; 124 } 125 num_clusters++; 126 itbl_cluster = c; 127 } 128 } 129 130 if (block_cluster != -1) 131 num_clusters++; 132 if (inode_cluster != -1) 133 num_clusters++; 134 135 return num_clusters; 136 } 137 138 static unsigned int num_clusters_in_group(struct super_block *sb, 139 ext4_group_t block_group) 140 { 141 unsigned int blocks; 142 143 if (block_group == ext4_get_groups_count(sb) - 1) { 144 /* 145 * Even though mke2fs always initializes the first and 146 * last group, just in case some other tool was used, 147 * we need to make sure we calculate the right free 148 * blocks. 149 */ 150 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) - 151 ext4_group_first_block_no(sb, block_group); 152 } else 153 blocks = EXT4_BLOCKS_PER_GROUP(sb); 154 return EXT4_NUM_B2C(EXT4_SB(sb), blocks); 155 } 156 157 /* Initializes an uninitialized block bitmap */ 158 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh, 159 ext4_group_t block_group, 160 struct ext4_group_desc *gdp) 161 { 162 unsigned int bit, bit_max; 163 struct ext4_sb_info *sbi = EXT4_SB(sb); 164 ext4_fsblk_t start, tmp; 165 int flex_bg = 0; 166 167 J_ASSERT_BH(bh, buffer_locked(bh)); 168 169 /* If checksum is bad mark all blocks used to prevent allocation 170 * essentially implementing a per-group read-only flag. */ 171 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { 172 ext4_error(sb, "Checksum bad for group %u", block_group); 173 ext4_free_group_clusters_set(sb, gdp, 0); 174 ext4_free_inodes_set(sb, gdp, 0); 175 ext4_itable_unused_set(sb, gdp, 0); 176 memset(bh->b_data, 0xff, sb->s_blocksize); 177 ext4_block_bitmap_csum_set(sb, block_group, gdp, bh, 178 EXT4_BLOCKS_PER_GROUP(sb) / 8); 179 return; 180 } 181 memset(bh->b_data, 0, sb->s_blocksize); 182 183 bit_max = ext4_num_base_meta_clusters(sb, block_group); 184 for (bit = 0; bit < bit_max; bit++) 185 ext4_set_bit(bit, bh->b_data); 186 187 start = ext4_group_first_block_no(sb, block_group); 188 189 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) 190 flex_bg = 1; 191 192 /* Set bits for block and inode bitmaps, and inode table */ 193 tmp = ext4_block_bitmap(sb, gdp); 194 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) 195 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 196 197 tmp = ext4_inode_bitmap(sb, gdp); 198 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) 199 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 200 201 tmp = ext4_inode_table(sb, gdp); 202 for (; tmp < ext4_inode_table(sb, gdp) + 203 sbi->s_itb_per_group; tmp++) { 204 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) 205 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); 206 } 207 208 /* 209 * Also if the number of blocks within the group is less than 210 * the blocksize * 8 ( which is the size of bitmap ), set rest 211 * of the block bitmap to 1 212 */ 213 ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group), 214 sb->s_blocksize * 8, bh->b_data); 215 ext4_block_bitmap_csum_set(sb, block_group, gdp, bh, 216 EXT4_BLOCKS_PER_GROUP(sb) / 8); 217 ext4_group_desc_csum_set(sb, block_group, gdp); 218 } 219 220 /* Return the number of free blocks in a block group. It is used when 221 * the block bitmap is uninitialized, so we can't just count the bits 222 * in the bitmap. */ 223 unsigned ext4_free_clusters_after_init(struct super_block *sb, 224 ext4_group_t block_group, 225 struct ext4_group_desc *gdp) 226 { 227 return num_clusters_in_group(sb, block_group) - 228 ext4_num_overhead_clusters(sb, block_group, gdp); 229 } 230 231 /* 232 * The free blocks are managed by bitmaps. A file system contains several 233 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 234 * block for inodes, N blocks for the inode table and data blocks. 235 * 236 * The file system contains group descriptors which are located after the 237 * super block. Each descriptor contains the number of the bitmap block and 238 * the free blocks count in the block. The descriptors are loaded in memory 239 * when a file system is mounted (see ext4_fill_super). 240 */ 241 242 /** 243 * ext4_get_group_desc() -- load group descriptor from disk 244 * @sb: super block 245 * @block_group: given block group 246 * @bh: pointer to the buffer head to store the block 247 * group descriptor 248 */ 249 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb, 250 ext4_group_t block_group, 251 struct buffer_head **bh) 252 { 253 unsigned int group_desc; 254 unsigned int offset; 255 ext4_group_t ngroups = ext4_get_groups_count(sb); 256 struct ext4_group_desc *desc; 257 struct ext4_sb_info *sbi = EXT4_SB(sb); 258 259 if (block_group >= ngroups) { 260 ext4_error(sb, "block_group >= groups_count - block_group = %u," 261 " groups_count = %u", block_group, ngroups); 262 263 return NULL; 264 } 265 266 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); 267 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); 268 if (!sbi->s_group_desc[group_desc]) { 269 ext4_error(sb, "Group descriptor not loaded - " 270 "block_group = %u, group_desc = %u, desc = %u", 271 block_group, group_desc, offset); 272 return NULL; 273 } 274 275 desc = (struct ext4_group_desc *)( 276 (__u8 *)sbi->s_group_desc[group_desc]->b_data + 277 offset * EXT4_DESC_SIZE(sb)); 278 if (bh) 279 *bh = sbi->s_group_desc[group_desc]; 280 return desc; 281 } 282 283 static int ext4_valid_block_bitmap(struct super_block *sb, 284 struct ext4_group_desc *desc, 285 unsigned int block_group, 286 struct buffer_head *bh) 287 { 288 ext4_grpblk_t offset; 289 ext4_grpblk_t next_zero_bit; 290 ext4_fsblk_t bitmap_blk; 291 ext4_fsblk_t group_first_block; 292 293 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { 294 /* with FLEX_BG, the inode/block bitmaps and itable 295 * blocks may not be in the group at all 296 * so the bitmap validation will be skipped for those groups 297 * or it has to also read the block group where the bitmaps 298 * are located to verify they are set. 299 */ 300 return 1; 301 } 302 group_first_block = ext4_group_first_block_no(sb, block_group); 303 304 /* check whether block bitmap block number is set */ 305 bitmap_blk = ext4_block_bitmap(sb, desc); 306 offset = bitmap_blk - group_first_block; 307 if (!ext4_test_bit(offset, bh->b_data)) 308 /* bad block bitmap */ 309 goto err_out; 310 311 /* check whether the inode bitmap block number is set */ 312 bitmap_blk = ext4_inode_bitmap(sb, desc); 313 offset = bitmap_blk - group_first_block; 314 if (!ext4_test_bit(offset, bh->b_data)) 315 /* bad block bitmap */ 316 goto err_out; 317 318 /* check whether the inode table block number is set */ 319 bitmap_blk = ext4_inode_table(sb, desc); 320 offset = bitmap_blk - group_first_block; 321 next_zero_bit = ext4_find_next_zero_bit(bh->b_data, 322 offset + EXT4_SB(sb)->s_itb_per_group, 323 offset); 324 if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group) 325 /* good bitmap for inode tables */ 326 return 1; 327 328 err_out: 329 ext4_error(sb, "Invalid block bitmap - block_group = %d, block = %llu", 330 block_group, bitmap_blk); 331 return 0; 332 } 333 334 void ext4_validate_block_bitmap(struct super_block *sb, 335 struct ext4_group_desc *desc, 336 unsigned int block_group, 337 struct buffer_head *bh) 338 { 339 if (buffer_verified(bh)) 340 return; 341 342 ext4_lock_group(sb, block_group); 343 if (ext4_valid_block_bitmap(sb, desc, block_group, bh) && 344 ext4_block_bitmap_csum_verify(sb, block_group, desc, bh, 345 EXT4_BLOCKS_PER_GROUP(sb) / 8)) 346 set_buffer_verified(bh); 347 ext4_unlock_group(sb, block_group); 348 } 349 350 /** 351 * ext4_read_block_bitmap() 352 * @sb: super block 353 * @block_group: given block group 354 * 355 * Read the bitmap for a given block_group,and validate the 356 * bits for block/inode/inode tables are set in the bitmaps 357 * 358 * Return buffer_head on success or NULL in case of failure. 359 */ 360 struct buffer_head * 361 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group) 362 { 363 struct ext4_group_desc *desc; 364 struct buffer_head *bh; 365 ext4_fsblk_t bitmap_blk; 366 367 desc = ext4_get_group_desc(sb, block_group, NULL); 368 if (!desc) 369 return NULL; 370 bitmap_blk = ext4_block_bitmap(sb, desc); 371 bh = sb_getblk(sb, bitmap_blk); 372 if (unlikely(!bh)) { 373 ext4_error(sb, "Cannot get buffer for block bitmap - " 374 "block_group = %u, block_bitmap = %llu", 375 block_group, bitmap_blk); 376 return NULL; 377 } 378 379 if (bitmap_uptodate(bh)) 380 goto verify; 381 382 lock_buffer(bh); 383 if (bitmap_uptodate(bh)) { 384 unlock_buffer(bh); 385 goto verify; 386 } 387 ext4_lock_group(sb, block_group); 388 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 389 ext4_init_block_bitmap(sb, bh, block_group, desc); 390 set_bitmap_uptodate(bh); 391 set_buffer_uptodate(bh); 392 ext4_unlock_group(sb, block_group); 393 unlock_buffer(bh); 394 return bh; 395 } 396 ext4_unlock_group(sb, block_group); 397 if (buffer_uptodate(bh)) { 398 /* 399 * if not uninit if bh is uptodate, 400 * bitmap is also uptodate 401 */ 402 set_bitmap_uptodate(bh); 403 unlock_buffer(bh); 404 goto verify; 405 } 406 /* 407 * submit the buffer_head for reading 408 */ 409 set_buffer_new(bh); 410 trace_ext4_read_block_bitmap_load(sb, block_group); 411 bh->b_end_io = ext4_end_bitmap_read; 412 get_bh(bh); 413 submit_bh(READ, bh); 414 return bh; 415 verify: 416 ext4_validate_block_bitmap(sb, desc, block_group, bh); 417 return bh; 418 } 419 420 /* Returns 0 on success, 1 on error */ 421 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group, 422 struct buffer_head *bh) 423 { 424 struct ext4_group_desc *desc; 425 426 if (!buffer_new(bh)) 427 return 0; 428 desc = ext4_get_group_desc(sb, block_group, NULL); 429 if (!desc) 430 return 1; 431 wait_on_buffer(bh); 432 if (!buffer_uptodate(bh)) { 433 ext4_error(sb, "Cannot read block bitmap - " 434 "block_group = %u, block_bitmap = %llu", 435 block_group, (unsigned long long) bh->b_blocknr); 436 return 1; 437 } 438 clear_buffer_new(bh); 439 /* Panic or remount fs read-only if block bitmap is invalid */ 440 ext4_validate_block_bitmap(sb, desc, block_group, bh); 441 return 0; 442 } 443 444 struct buffer_head * 445 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group) 446 { 447 struct buffer_head *bh; 448 449 bh = ext4_read_block_bitmap_nowait(sb, block_group); 450 if (ext4_wait_block_bitmap(sb, block_group, bh)) { 451 put_bh(bh); 452 return NULL; 453 } 454 return bh; 455 } 456 457 /** 458 * ext4_has_free_clusters() 459 * @sbi: in-core super block structure. 460 * @nclusters: number of needed blocks 461 * @flags: flags from ext4_mb_new_blocks() 462 * 463 * Check if filesystem has nclusters free & available for allocation. 464 * On success return 1, return 0 on failure. 465 */ 466 static int ext4_has_free_clusters(struct ext4_sb_info *sbi, 467 s64 nclusters, unsigned int flags) 468 { 469 s64 free_clusters, dirty_clusters, root_clusters; 470 struct percpu_counter *fcc = &sbi->s_freeclusters_counter; 471 struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter; 472 473 free_clusters = percpu_counter_read_positive(fcc); 474 dirty_clusters = percpu_counter_read_positive(dcc); 475 root_clusters = EXT4_B2C(sbi, ext4_r_blocks_count(sbi->s_es)); 476 477 if (free_clusters - (nclusters + root_clusters + dirty_clusters) < 478 EXT4_FREECLUSTERS_WATERMARK) { 479 free_clusters = EXT4_C2B(sbi, percpu_counter_sum_positive(fcc)); 480 dirty_clusters = percpu_counter_sum_positive(dcc); 481 } 482 /* Check whether we have space after accounting for current 483 * dirty clusters & root reserved clusters. 484 */ 485 if (free_clusters >= ((root_clusters + nclusters) + dirty_clusters)) 486 return 1; 487 488 /* Hm, nope. Are (enough) root reserved clusters available? */ 489 if (uid_eq(sbi->s_resuid, current_fsuid()) || 490 (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) || 491 capable(CAP_SYS_RESOURCE) || 492 (flags & EXT4_MB_USE_ROOT_BLOCKS)) { 493 494 if (free_clusters >= (nclusters + dirty_clusters)) 495 return 1; 496 } 497 498 return 0; 499 } 500 501 int ext4_claim_free_clusters(struct ext4_sb_info *sbi, 502 s64 nclusters, unsigned int flags) 503 { 504 if (ext4_has_free_clusters(sbi, nclusters, flags)) { 505 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters); 506 return 0; 507 } else 508 return -ENOSPC; 509 } 510 511 /** 512 * ext4_should_retry_alloc() 513 * @sb: super block 514 * @retries number of attemps has been made 515 * 516 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if 517 * it is profitable to retry the operation, this function will wait 518 * for the current or committing transaction to complete, and then 519 * return TRUE. 520 * 521 * if the total number of retries exceed three times, return FALSE. 522 */ 523 int ext4_should_retry_alloc(struct super_block *sb, int *retries) 524 { 525 if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) || 526 (*retries)++ > 3 || 527 !EXT4_SB(sb)->s_journal) 528 return 0; 529 530 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); 531 532 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal); 533 } 534 535 /* 536 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks 537 * 538 * @handle: handle to this transaction 539 * @inode: file inode 540 * @goal: given target block(filesystem wide) 541 * @count: pointer to total number of clusters needed 542 * @errp: error code 543 * 544 * Return 1st allocated block number on success, *count stores total account 545 * error stores in errp pointer 546 */ 547 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode, 548 ext4_fsblk_t goal, unsigned int flags, 549 unsigned long *count, int *errp) 550 { 551 struct ext4_allocation_request ar; 552 ext4_fsblk_t ret; 553 554 memset(&ar, 0, sizeof(ar)); 555 /* Fill with neighbour allocated blocks */ 556 ar.inode = inode; 557 ar.goal = goal; 558 ar.len = count ? *count : 1; 559 ar.flags = flags; 560 561 ret = ext4_mb_new_blocks(handle, &ar, errp); 562 if (count) 563 *count = ar.len; 564 /* 565 * Account for the allocated meta blocks. We will never 566 * fail EDQUOT for metdata, but we do account for it. 567 */ 568 if (!(*errp) && 569 ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) { 570 spin_lock(&EXT4_I(inode)->i_block_reservation_lock); 571 EXT4_I(inode)->i_allocated_meta_blocks += ar.len; 572 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 573 dquot_alloc_block_nofail(inode, 574 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len)); 575 } 576 return ret; 577 } 578 579 /** 580 * ext4_count_free_clusters() -- count filesystem free clusters 581 * @sb: superblock 582 * 583 * Adds up the number of free clusters from each block group. 584 */ 585 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb) 586 { 587 ext4_fsblk_t desc_count; 588 struct ext4_group_desc *gdp; 589 ext4_group_t i; 590 ext4_group_t ngroups = ext4_get_groups_count(sb); 591 #ifdef EXT4FS_DEBUG 592 struct ext4_super_block *es; 593 ext4_fsblk_t bitmap_count; 594 unsigned int x; 595 struct buffer_head *bitmap_bh = NULL; 596 597 es = EXT4_SB(sb)->s_es; 598 desc_count = 0; 599 bitmap_count = 0; 600 gdp = NULL; 601 602 for (i = 0; i < ngroups; i++) { 603 gdp = ext4_get_group_desc(sb, i, NULL); 604 if (!gdp) 605 continue; 606 desc_count += ext4_free_group_clusters(sb, gdp); 607 brelse(bitmap_bh); 608 bitmap_bh = ext4_read_block_bitmap(sb, i); 609 if (bitmap_bh == NULL) 610 continue; 611 612 x = ext4_count_free(bitmap_bh, sb->s_blocksize); 613 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n", 614 i, ext4_free_group_clusters(sb, gdp), x); 615 bitmap_count += x; 616 } 617 brelse(bitmap_bh); 618 printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu" 619 ", computed = %llu, %llu\n", 620 EXT4_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)), 621 desc_count, bitmap_count); 622 return bitmap_count; 623 #else 624 desc_count = 0; 625 for (i = 0; i < ngroups; i++) { 626 gdp = ext4_get_group_desc(sb, i, NULL); 627 if (!gdp) 628 continue; 629 desc_count += ext4_free_group_clusters(sb, gdp); 630 } 631 632 return desc_count; 633 #endif 634 } 635 636 static inline int test_root(ext4_group_t a, int b) 637 { 638 int num = b; 639 640 while (a > num) 641 num *= b; 642 return num == a; 643 } 644 645 static int ext4_group_sparse(ext4_group_t group) 646 { 647 if (group <= 1) 648 return 1; 649 if (!(group & 1)) 650 return 0; 651 return (test_root(group, 7) || test_root(group, 5) || 652 test_root(group, 3)); 653 } 654 655 /** 656 * ext4_bg_has_super - number of blocks used by the superblock in group 657 * @sb: superblock for filesystem 658 * @group: group number to check 659 * 660 * Return the number of blocks used by the superblock (primary or backup) 661 * in this group. Currently this will be only 0 or 1. 662 */ 663 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) 664 { 665 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, 666 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && 667 !ext4_group_sparse(group)) 668 return 0; 669 return 1; 670 } 671 672 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, 673 ext4_group_t group) 674 { 675 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); 676 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); 677 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; 678 679 if (group == first || group == first + 1 || group == last) 680 return 1; 681 return 0; 682 } 683 684 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, 685 ext4_group_t group) 686 { 687 if (!ext4_bg_has_super(sb, group)) 688 return 0; 689 690 if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG)) 691 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); 692 else 693 return EXT4_SB(sb)->s_gdb_count; 694 } 695 696 /** 697 * ext4_bg_num_gdb - number of blocks used by the group table in group 698 * @sb: superblock for filesystem 699 * @group: group number to check 700 * 701 * Return the number of blocks used by the group descriptor table 702 * (primary or backup) in this group. In the future there may be a 703 * different number of descriptor blocks in each group. 704 */ 705 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) 706 { 707 unsigned long first_meta_bg = 708 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); 709 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); 710 711 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || 712 metagroup < first_meta_bg) 713 return ext4_bg_num_gdb_nometa(sb, group); 714 715 return ext4_bg_num_gdb_meta(sb,group); 716 717 } 718 719 /* 720 * This function returns the number of file system metadata clusters at 721 * the beginning of a block group, including the reserved gdt blocks. 722 */ 723 static unsigned ext4_num_base_meta_clusters(struct super_block *sb, 724 ext4_group_t block_group) 725 { 726 struct ext4_sb_info *sbi = EXT4_SB(sb); 727 unsigned num; 728 729 /* Check for superblock and gdt backups in this group */ 730 num = ext4_bg_has_super(sb, block_group); 731 732 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) || 733 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * 734 sbi->s_desc_per_block) { 735 if (num) { 736 num += ext4_bg_num_gdb(sb, block_group); 737 num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); 738 } 739 } else { /* For META_BG_BLOCK_GROUPS */ 740 num += ext4_bg_num_gdb(sb, block_group); 741 } 742 return EXT4_NUM_B2C(sbi, num); 743 } 744 /** 745 * ext4_inode_to_goal_block - return a hint for block allocation 746 * @inode: inode for block allocation 747 * 748 * Return the ideal location to start allocating blocks for a 749 * newly created inode. 750 */ 751 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode) 752 { 753 struct ext4_inode_info *ei = EXT4_I(inode); 754 ext4_group_t block_group; 755 ext4_grpblk_t colour; 756 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); 757 ext4_fsblk_t bg_start; 758 ext4_fsblk_t last_block; 759 760 block_group = ei->i_block_group; 761 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { 762 /* 763 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME 764 * block groups per flexgroup, reserve the first block 765 * group for directories and special files. Regular 766 * files will start at the second block group. This 767 * tends to speed up directory access and improves 768 * fsck times. 769 */ 770 block_group &= ~(flex_size-1); 771 if (S_ISREG(inode->i_mode)) 772 block_group++; 773 } 774 bg_start = ext4_group_first_block_no(inode->i_sb, block_group); 775 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; 776 777 /* 778 * If we are doing delayed allocation, we don't need take 779 * colour into account. 780 */ 781 if (test_opt(inode->i_sb, DELALLOC)) 782 return bg_start; 783 784 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) 785 colour = (current->pid % 16) * 786 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); 787 else 788 colour = (current->pid % 16) * ((last_block - bg_start) / 16); 789 return bg_start + colour; 790 } 791 792