1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext2/ialloc.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 * BSD ufs-inspired inode and directory allocation by 11 * Stephen Tweedie (sct@dcs.ed.ac.uk), 1993 12 * Big-endian to little-endian byte-swapping/bitmaps by 13 * David S. Miller (davem@caip.rutgers.edu), 1995 14 */ 15 16 #include <linux/quotaops.h> 17 #include <linux/sched.h> 18 #include <linux/backing-dev.h> 19 #include <linux/buffer_head.h> 20 #include <linux/random.h> 21 #include "ext2.h" 22 #include "xattr.h" 23 #include "acl.h" 24 25 /* 26 * ialloc.c contains the inodes allocation and deallocation routines 27 */ 28 29 /* 30 * The free inodes are managed by bitmaps. A file system contains several 31 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 32 * block for inodes, N blocks for the inode table and data blocks. 33 * 34 * The file system contains group descriptors which are located after the 35 * super block. Each descriptor contains the number of the bitmap block and 36 * the free blocks count in the block. 37 */ 38 39 40 /* 41 * Read the inode allocation bitmap for a given block_group, reading 42 * into the specified slot in the superblock's bitmap cache. 43 * 44 * Return buffer_head of bitmap on success or NULL. 45 */ 46 static struct buffer_head * 47 read_inode_bitmap(struct super_block * sb, unsigned long block_group) 48 { 49 struct ext2_group_desc *desc; 50 struct buffer_head *bh = NULL; 51 52 desc = ext2_get_group_desc(sb, block_group, NULL); 53 if (!desc) 54 goto error_out; 55 56 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap)); 57 if (!bh) 58 ext2_error(sb, "read_inode_bitmap", 59 "Cannot read inode bitmap - " 60 "block_group = %lu, inode_bitmap = %u", 61 block_group, le32_to_cpu(desc->bg_inode_bitmap)); 62 error_out: 63 return bh; 64 } 65 66 static void ext2_release_inode(struct super_block *sb, int group, int dir) 67 { 68 struct ext2_group_desc * desc; 69 struct buffer_head *bh; 70 71 desc = ext2_get_group_desc(sb, group, &bh); 72 if (!desc) { 73 ext2_error(sb, "ext2_release_inode", 74 "can't get descriptor for group %d", group); 75 return; 76 } 77 78 spin_lock(sb_bgl_lock(EXT2_SB(sb), group)); 79 le16_add_cpu(&desc->bg_free_inodes_count, 1); 80 if (dir) 81 le16_add_cpu(&desc->bg_used_dirs_count, -1); 82 spin_unlock(sb_bgl_lock(EXT2_SB(sb), group)); 83 percpu_counter_inc(&EXT2_SB(sb)->s_freeinodes_counter); 84 if (dir) 85 percpu_counter_dec(&EXT2_SB(sb)->s_dirs_counter); 86 mark_buffer_dirty(bh); 87 } 88 89 /* 90 * NOTE! When we get the inode, we're the only people 91 * that have access to it, and as such there are no 92 * race conditions we have to worry about. The inode 93 * is not on the hash-lists, and it cannot be reached 94 * through the filesystem because the directory entry 95 * has been deleted earlier. 96 * 97 * HOWEVER: we must make sure that we get no aliases, 98 * which means that we have to call "clear_inode()" 99 * _before_ we mark the inode not in use in the inode 100 * bitmaps. Otherwise a newly created file might use 101 * the same inode number (not actually the same pointer 102 * though), and then we'd have two inodes sharing the 103 * same inode number and space on the harddisk. 104 */ 105 void ext2_free_inode (struct inode * inode) 106 { 107 struct super_block * sb = inode->i_sb; 108 int is_directory; 109 unsigned long ino; 110 struct buffer_head *bitmap_bh; 111 unsigned long block_group; 112 unsigned long bit; 113 struct ext2_super_block * es; 114 115 ino = inode->i_ino; 116 ext2_debug ("freeing inode %lu\n", ino); 117 118 /* 119 * Note: we must free any quota before locking the superblock, 120 * as writing the quota to disk may need the lock as well. 121 */ 122 /* Quota is already initialized in iput() */ 123 dquot_free_inode(inode); 124 dquot_drop(inode); 125 126 es = EXT2_SB(sb)->s_es; 127 is_directory = S_ISDIR(inode->i_mode); 128 129 if (ino < EXT2_FIRST_INO(sb) || 130 ino > le32_to_cpu(es->s_inodes_count)) { 131 ext2_error (sb, "ext2_free_inode", 132 "reserved or nonexistent inode %lu", ino); 133 return; 134 } 135 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb); 136 bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb); 137 bitmap_bh = read_inode_bitmap(sb, block_group); 138 if (!bitmap_bh) 139 return; 140 141 /* Ok, now we can actually update the inode bitmaps.. */ 142 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb), block_group), 143 bit, (void *) bitmap_bh->b_data)) 144 ext2_error (sb, "ext2_free_inode", 145 "bit already cleared for inode %lu", ino); 146 else 147 ext2_release_inode(sb, block_group, is_directory); 148 mark_buffer_dirty(bitmap_bh); 149 if (sb->s_flags & SB_SYNCHRONOUS) 150 sync_dirty_buffer(bitmap_bh); 151 152 brelse(bitmap_bh); 153 } 154 155 /* 156 * We perform asynchronous prereading of the new inode's inode block when 157 * we create the inode, in the expectation that the inode will be written 158 * back soon. There are two reasons: 159 * 160 * - When creating a large number of files, the async prereads will be 161 * nicely merged into large reads 162 * - When writing out a large number of inodes, we don't need to keep on 163 * stalling the writes while we read the inode block. 164 * 165 * FIXME: ext2_get_group_desc() needs to be simplified. 166 */ 167 static void ext2_preread_inode(struct inode *inode) 168 { 169 unsigned long block_group; 170 unsigned long offset; 171 unsigned long block; 172 unsigned int ino = inode->i_ino; 173 struct ext2_group_desc * gdp; 174 175 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 176 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL); 177 if (gdp == NULL) 178 return; 179 180 /* 181 * Figure out the offset within the block group inode table 182 */ 183 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) * 184 EXT2_INODE_SIZE(inode->i_sb); 185 block = le32_to_cpu(gdp->bg_inode_table) + 186 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb)); 187 sb_breadahead(inode->i_sb, block); 188 } 189 190 /* 191 * There are two policies for allocating an inode. If the new inode is 192 * a directory, then a forward search is made for a block group with both 193 * free space and a low directory-to-inode ratio; if that fails, then of 194 * the groups with above-average free space, that group with the fewest 195 * directories already is chosen. 196 * 197 * For other inodes, search forward from the parent directory\'s block 198 * group to find a free inode. 199 */ 200 static int find_group_dir(struct super_block *sb, struct inode *parent) 201 { 202 int ngroups = EXT2_SB(sb)->s_groups_count; 203 int avefreei = ext2_count_free_inodes(sb) / ngroups; 204 struct ext2_group_desc *desc, *best_desc = NULL; 205 int group, best_group = -1; 206 207 for (group = 0; group < ngroups; group++) { 208 desc = ext2_get_group_desc (sb, group, NULL); 209 if (!desc || !desc->bg_free_inodes_count) 210 continue; 211 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) 212 continue; 213 if (!best_desc || 214 (le16_to_cpu(desc->bg_free_blocks_count) > 215 le16_to_cpu(best_desc->bg_free_blocks_count))) { 216 best_group = group; 217 best_desc = desc; 218 } 219 } 220 221 return best_group; 222 } 223 224 /* 225 * Orlov's allocator for directories. 226 * 227 * We always try to spread first-level directories. 228 * 229 * If there are blockgroups with both free inodes and free blocks counts 230 * not worse than average we return one with smallest directory count. 231 * Otherwise we simply return a random group. 232 * 233 * For the rest rules look so: 234 * 235 * It's OK to put directory into a group unless 236 * it has too many directories already (max_dirs) or 237 * it has too few free inodes left (min_inodes) or 238 * it has too few free blocks left (min_blocks) or 239 * it's already running too large debt (max_debt). 240 * Parent's group is preferred, if it doesn't satisfy these 241 * conditions we search cyclically through the rest. If none 242 * of the groups look good we just look for a group with more 243 * free inodes than average (starting at parent's group). 244 * 245 * Debt is incremented each time we allocate a directory and decremented 246 * when we allocate an inode, within 0--255. 247 */ 248 249 #define INODE_COST 64 250 #define BLOCK_COST 256 251 252 static int find_group_orlov(struct super_block *sb, struct inode *parent) 253 { 254 int parent_group = EXT2_I(parent)->i_block_group; 255 struct ext2_sb_info *sbi = EXT2_SB(sb); 256 struct ext2_super_block *es = sbi->s_es; 257 int ngroups = sbi->s_groups_count; 258 int inodes_per_group = EXT2_INODES_PER_GROUP(sb); 259 int freei; 260 int avefreei; 261 int free_blocks; 262 int avefreeb; 263 int blocks_per_dir; 264 int ndirs; 265 int max_debt, max_dirs, min_blocks, min_inodes; 266 int group = -1, i; 267 struct ext2_group_desc *desc; 268 269 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); 270 avefreei = freei / ngroups; 271 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); 272 avefreeb = free_blocks / ngroups; 273 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); 274 275 if ((parent == d_inode(sb->s_root)) || 276 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) { 277 int best_ndir = inodes_per_group; 278 int best_group = -1; 279 280 parent_group = get_random_u32_below(ngroups); 281 for (i = 0; i < ngroups; i++) { 282 group = (parent_group + i) % ngroups; 283 desc = ext2_get_group_desc (sb, group, NULL); 284 if (!desc || !desc->bg_free_inodes_count) 285 continue; 286 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) 287 continue; 288 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) 289 continue; 290 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) 291 continue; 292 best_group = group; 293 best_ndir = le16_to_cpu(desc->bg_used_dirs_count); 294 } 295 if (best_group >= 0) { 296 group = best_group; 297 goto found; 298 } 299 goto fallback; 300 } 301 302 if (ndirs == 0) 303 ndirs = 1; /* percpu_counters are approximate... */ 304 305 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs; 306 307 max_dirs = ndirs / ngroups + inodes_per_group / 16; 308 min_inodes = avefreei - inodes_per_group / 4; 309 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4; 310 311 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST); 312 if (max_debt * INODE_COST > inodes_per_group) 313 max_debt = inodes_per_group / INODE_COST; 314 if (max_debt > 255) 315 max_debt = 255; 316 if (max_debt == 0) 317 max_debt = 1; 318 319 for (i = 0; i < ngroups; i++) { 320 group = (parent_group + i) % ngroups; 321 desc = ext2_get_group_desc (sb, group, NULL); 322 if (!desc || !desc->bg_free_inodes_count) 323 continue; 324 if (sbi->s_debts[group] >= max_debt) 325 continue; 326 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) 327 continue; 328 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) 329 continue; 330 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) 331 continue; 332 goto found; 333 } 334 335 fallback: 336 for (i = 0; i < ngroups; i++) { 337 group = (parent_group + i) % ngroups; 338 desc = ext2_get_group_desc (sb, group, NULL); 339 if (!desc || !desc->bg_free_inodes_count) 340 continue; 341 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) 342 goto found; 343 } 344 345 if (avefreei) { 346 /* 347 * The free-inodes counter is approximate, and for really small 348 * filesystems the above test can fail to find any blockgroups 349 */ 350 avefreei = 0; 351 goto fallback; 352 } 353 354 return -1; 355 356 found: 357 return group; 358 } 359 360 static int find_group_other(struct super_block *sb, struct inode *parent) 361 { 362 int parent_group = EXT2_I(parent)->i_block_group; 363 int ngroups = EXT2_SB(sb)->s_groups_count; 364 struct ext2_group_desc *desc; 365 int group, i; 366 367 /* 368 * Try to place the inode in its parent directory 369 */ 370 group = parent_group; 371 desc = ext2_get_group_desc (sb, group, NULL); 372 if (desc && le16_to_cpu(desc->bg_free_inodes_count) && 373 le16_to_cpu(desc->bg_free_blocks_count)) 374 goto found; 375 376 /* 377 * We're going to place this inode in a different blockgroup from its 378 * parent. We want to cause files in a common directory to all land in 379 * the same blockgroup. But we want files which are in a different 380 * directory which shares a blockgroup with our parent to land in a 381 * different blockgroup. 382 * 383 * So add our directory's i_ino into the starting point for the hash. 384 */ 385 group = (group + (unsigned int)parent->i_ino) % ngroups; 386 387 /* 388 * Use a quadratic hash to find a group with a free inode and some 389 * free blocks. 390 */ 391 for (i = 1; i < ngroups; i <<= 1) { 392 group += i; 393 if (group >= ngroups) 394 group -= ngroups; 395 desc = ext2_get_group_desc (sb, group, NULL); 396 if (desc && le16_to_cpu(desc->bg_free_inodes_count) && 397 le16_to_cpu(desc->bg_free_blocks_count)) 398 goto found; 399 } 400 401 /* 402 * That failed: try linear search for a free inode, even if that group 403 * has no free blocks. 404 */ 405 group = parent_group; 406 for (i = 0; i < ngroups; i++) { 407 if (++group >= ngroups) 408 group = 0; 409 desc = ext2_get_group_desc (sb, group, NULL); 410 if (desc && le16_to_cpu(desc->bg_free_inodes_count)) 411 goto found; 412 } 413 414 return -1; 415 416 found: 417 return group; 418 } 419 420 struct inode *ext2_new_inode(struct inode *dir, umode_t mode, 421 const struct qstr *qstr) 422 { 423 struct super_block *sb; 424 struct buffer_head *bitmap_bh = NULL; 425 struct buffer_head *bh2; 426 int group, i; 427 ino_t ino = 0; 428 struct inode * inode; 429 struct ext2_group_desc *gdp; 430 struct ext2_super_block *es; 431 struct ext2_inode_info *ei; 432 struct ext2_sb_info *sbi; 433 int err; 434 435 sb = dir->i_sb; 436 inode = new_inode(sb); 437 if (!inode) 438 return ERR_PTR(-ENOMEM); 439 440 ei = EXT2_I(inode); 441 sbi = EXT2_SB(sb); 442 es = sbi->s_es; 443 if (S_ISDIR(mode)) { 444 if (test_opt(sb, OLDALLOC)) 445 group = find_group_dir(sb, dir); 446 else 447 group = find_group_orlov(sb, dir); 448 } else 449 group = find_group_other(sb, dir); 450 451 if (group == -1) { 452 err = -ENOSPC; 453 goto fail; 454 } 455 456 for (i = 0; i < sbi->s_groups_count; i++) { 457 gdp = ext2_get_group_desc(sb, group, &bh2); 458 if (!gdp) { 459 if (++group == sbi->s_groups_count) 460 group = 0; 461 continue; 462 } 463 brelse(bitmap_bh); 464 bitmap_bh = read_inode_bitmap(sb, group); 465 if (!bitmap_bh) { 466 err = -EIO; 467 goto fail; 468 } 469 ino = 0; 470 471 repeat_in_this_group: 472 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data, 473 EXT2_INODES_PER_GROUP(sb), ino); 474 if (ino >= EXT2_INODES_PER_GROUP(sb)) { 475 /* 476 * Rare race: find_group_xx() decided that there were 477 * free inodes in this group, but by the time we tried 478 * to allocate one, they're all gone. This can also 479 * occur because the counters which find_group_orlov() 480 * uses are approximate. So just go and search the 481 * next block group. 482 */ 483 if (++group == sbi->s_groups_count) 484 group = 0; 485 continue; 486 } 487 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group), 488 ino, bitmap_bh->b_data)) { 489 /* we lost this inode */ 490 if (++ino >= EXT2_INODES_PER_GROUP(sb)) { 491 /* this group is exhausted, try next group */ 492 if (++group == sbi->s_groups_count) 493 group = 0; 494 continue; 495 } 496 /* try to find free inode in the same group */ 497 goto repeat_in_this_group; 498 } 499 goto got; 500 } 501 502 /* 503 * Scanned all blockgroups. 504 */ 505 brelse(bitmap_bh); 506 err = -ENOSPC; 507 goto fail; 508 got: 509 mark_buffer_dirty(bitmap_bh); 510 if (sb->s_flags & SB_SYNCHRONOUS) 511 sync_dirty_buffer(bitmap_bh); 512 brelse(bitmap_bh); 513 514 ino += group * EXT2_INODES_PER_GROUP(sb) + 1; 515 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { 516 ext2_error (sb, "ext2_new_inode", 517 "reserved inode or inode > inodes count - " 518 "block_group = %d,inode=%lu", group, 519 (unsigned long) ino); 520 err = -EIO; 521 goto fail; 522 } 523 524 percpu_counter_dec(&sbi->s_freeinodes_counter); 525 if (S_ISDIR(mode)) 526 percpu_counter_inc(&sbi->s_dirs_counter); 527 528 spin_lock(sb_bgl_lock(sbi, group)); 529 le16_add_cpu(&gdp->bg_free_inodes_count, -1); 530 if (S_ISDIR(mode)) { 531 if (sbi->s_debts[group] < 255) 532 sbi->s_debts[group]++; 533 le16_add_cpu(&gdp->bg_used_dirs_count, 1); 534 } else { 535 if (sbi->s_debts[group]) 536 sbi->s_debts[group]--; 537 } 538 spin_unlock(sb_bgl_lock(sbi, group)); 539 540 mark_buffer_dirty(bh2); 541 if (test_opt(sb, GRPID)) { 542 inode->i_mode = mode; 543 inode->i_uid = current_fsuid(); 544 inode->i_gid = dir->i_gid; 545 } else 546 inode_init_owner(&nop_mnt_idmap, inode, dir, mode); 547 548 inode->i_ino = ino; 549 inode->i_blocks = 0; 550 simple_inode_init_ts(inode); 551 memset(ei->i_data, 0, sizeof(ei->i_data)); 552 ei->i_flags = 553 ext2_mask_flags(mode, EXT2_I(dir)->i_flags & EXT2_FL_INHERITED); 554 ei->i_faddr = 0; 555 ei->i_frag_no = 0; 556 ei->i_frag_size = 0; 557 ei->i_file_acl = 0; 558 ei->i_dir_acl = 0; 559 ei->i_dtime = 0; 560 ei->i_block_alloc_info = NULL; 561 ei->i_block_group = group; 562 ei->i_dir_start_lookup = 0; 563 ei->i_state = EXT2_STATE_NEW; 564 ext2_set_inode_flags(inode); 565 spin_lock(&sbi->s_next_gen_lock); 566 inode->i_generation = sbi->s_next_generation++; 567 spin_unlock(&sbi->s_next_gen_lock); 568 if (insert_inode_locked(inode) < 0) { 569 ext2_error(sb, "ext2_new_inode", 570 "inode number already in use - inode=%lu", 571 (unsigned long) ino); 572 err = -EIO; 573 goto fail; 574 } 575 576 err = dquot_initialize(inode); 577 if (err) 578 goto fail_drop; 579 580 err = dquot_alloc_inode(inode); 581 if (err) 582 goto fail_drop; 583 584 err = ext2_init_acl(inode, dir); 585 if (err) 586 goto fail_free_drop; 587 588 err = ext2_init_security(inode, dir, qstr); 589 if (err) 590 goto fail_free_drop; 591 592 mark_inode_dirty(inode); 593 ext2_debug("allocating inode %llu\n", inode->i_ino); 594 ext2_preread_inode(inode); 595 return inode; 596 597 fail_free_drop: 598 dquot_free_inode(inode); 599 600 fail_drop: 601 dquot_drop(inode); 602 inode->i_flags |= S_NOQUOTA; 603 clear_nlink(inode); 604 discard_new_inode(inode); 605 return ERR_PTR(err); 606 607 fail: 608 make_bad_inode(inode); 609 iput(inode); 610 return ERR_PTR(err); 611 } 612 613 unsigned long ext2_count_free_inodes (struct super_block * sb) 614 { 615 struct ext2_group_desc *desc; 616 unsigned long desc_count = 0; 617 int i; 618 619 #ifdef EXT2FS_DEBUG 620 struct ext2_super_block *es; 621 unsigned long bitmap_count = 0; 622 struct buffer_head *bitmap_bh = NULL; 623 624 es = EXT2_SB(sb)->s_es; 625 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 626 unsigned x; 627 628 desc = ext2_get_group_desc (sb, i, NULL); 629 if (!desc) 630 continue; 631 desc_count += le16_to_cpu(desc->bg_free_inodes_count); 632 brelse(bitmap_bh); 633 bitmap_bh = read_inode_bitmap(sb, i); 634 if (!bitmap_bh) 635 continue; 636 637 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8); 638 printk("group %d: stored = %d, counted = %u\n", 639 i, le16_to_cpu(desc->bg_free_inodes_count), x); 640 bitmap_count += x; 641 } 642 brelse(bitmap_bh); 643 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n", 644 (unsigned long) 645 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter), 646 desc_count, bitmap_count); 647 return desc_count; 648 #else 649 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 650 desc = ext2_get_group_desc (sb, i, NULL); 651 if (!desc) 652 continue; 653 desc_count += le16_to_cpu(desc->bg_free_inodes_count); 654 } 655 return desc_count; 656 #endif 657 } 658 659 /* Called at mount-time, super-block is locked */ 660 unsigned long ext2_count_dirs (struct super_block * sb) 661 { 662 unsigned long count = 0; 663 int i; 664 665 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 666 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL); 667 if (!gdp) 668 continue; 669 count += le16_to_cpu(gdp->bg_used_dirs_count); 670 } 671 return count; 672 } 673 674