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 struct ext2_group_desc * gdp; 173 struct backing_dev_info *bdi; 174 175 bdi = inode_to_bdi(inode); 176 if (bdi_rw_congested(bdi)) 177 return; 178 179 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 180 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL); 181 if (gdp == NULL) 182 return; 183 184 /* 185 * Figure out the offset within the block group inode table 186 */ 187 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) * 188 EXT2_INODE_SIZE(inode->i_sb); 189 block = le32_to_cpu(gdp->bg_inode_table) + 190 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb)); 191 sb_breadahead(inode->i_sb, block); 192 } 193 194 /* 195 * There are two policies for allocating an inode. If the new inode is 196 * a directory, then a forward search is made for a block group with both 197 * free space and a low directory-to-inode ratio; if that fails, then of 198 * the groups with above-average free space, that group with the fewest 199 * directories already is chosen. 200 * 201 * For other inodes, search forward from the parent directory\'s block 202 * group to find a free inode. 203 */ 204 static int find_group_dir(struct super_block *sb, struct inode *parent) 205 { 206 int ngroups = EXT2_SB(sb)->s_groups_count; 207 int avefreei = ext2_count_free_inodes(sb) / ngroups; 208 struct ext2_group_desc *desc, *best_desc = NULL; 209 int group, best_group = -1; 210 211 for (group = 0; group < ngroups; group++) { 212 desc = ext2_get_group_desc (sb, group, NULL); 213 if (!desc || !desc->bg_free_inodes_count) 214 continue; 215 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) 216 continue; 217 if (!best_desc || 218 (le16_to_cpu(desc->bg_free_blocks_count) > 219 le16_to_cpu(best_desc->bg_free_blocks_count))) { 220 best_group = group; 221 best_desc = desc; 222 } 223 } 224 225 return best_group; 226 } 227 228 /* 229 * Orlov's allocator for directories. 230 * 231 * We always try to spread first-level directories. 232 * 233 * If there are blockgroups with both free inodes and free blocks counts 234 * not worse than average we return one with smallest directory count. 235 * Otherwise we simply return a random group. 236 * 237 * For the rest rules look so: 238 * 239 * It's OK to put directory into a group unless 240 * it has too many directories already (max_dirs) or 241 * it has too few free inodes left (min_inodes) or 242 * it has too few free blocks left (min_blocks) or 243 * it's already running too large debt (max_debt). 244 * Parent's group is preferred, if it doesn't satisfy these 245 * conditions we search cyclically through the rest. If none 246 * of the groups look good we just look for a group with more 247 * free inodes than average (starting at parent's group). 248 * 249 * Debt is incremented each time we allocate a directory and decremented 250 * when we allocate an inode, within 0--255. 251 */ 252 253 #define INODE_COST 64 254 #define BLOCK_COST 256 255 256 static int find_group_orlov(struct super_block *sb, struct inode *parent) 257 { 258 int parent_group = EXT2_I(parent)->i_block_group; 259 struct ext2_sb_info *sbi = EXT2_SB(sb); 260 struct ext2_super_block *es = sbi->s_es; 261 int ngroups = sbi->s_groups_count; 262 int inodes_per_group = EXT2_INODES_PER_GROUP(sb); 263 int freei; 264 int avefreei; 265 int free_blocks; 266 int avefreeb; 267 int blocks_per_dir; 268 int ndirs; 269 int max_debt, max_dirs, min_blocks, min_inodes; 270 int group = -1, i; 271 struct ext2_group_desc *desc; 272 273 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); 274 avefreei = freei / ngroups; 275 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); 276 avefreeb = free_blocks / ngroups; 277 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); 278 279 if ((parent == d_inode(sb->s_root)) || 280 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) { 281 struct ext2_group_desc *best_desc = NULL; 282 int best_ndir = inodes_per_group; 283 int best_group = -1; 284 285 group = prandom_u32(); 286 parent_group = (unsigned)group % ngroups; 287 for (i = 0; i < ngroups; i++) { 288 group = (parent_group + i) % ngroups; 289 desc = ext2_get_group_desc (sb, group, NULL); 290 if (!desc || !desc->bg_free_inodes_count) 291 continue; 292 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) 293 continue; 294 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) 295 continue; 296 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) 297 continue; 298 best_group = group; 299 best_ndir = le16_to_cpu(desc->bg_used_dirs_count); 300 best_desc = desc; 301 } 302 if (best_group >= 0) { 303 desc = best_desc; 304 group = best_group; 305 goto found; 306 } 307 goto fallback; 308 } 309 310 if (ndirs == 0) 311 ndirs = 1; /* percpu_counters are approximate... */ 312 313 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs; 314 315 max_dirs = ndirs / ngroups + inodes_per_group / 16; 316 min_inodes = avefreei - inodes_per_group / 4; 317 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4; 318 319 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST); 320 if (max_debt * INODE_COST > inodes_per_group) 321 max_debt = inodes_per_group / INODE_COST; 322 if (max_debt > 255) 323 max_debt = 255; 324 if (max_debt == 0) 325 max_debt = 1; 326 327 for (i = 0; i < ngroups; i++) { 328 group = (parent_group + i) % ngroups; 329 desc = ext2_get_group_desc (sb, group, NULL); 330 if (!desc || !desc->bg_free_inodes_count) 331 continue; 332 if (sbi->s_debts[group] >= max_debt) 333 continue; 334 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) 335 continue; 336 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) 337 continue; 338 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) 339 continue; 340 goto found; 341 } 342 343 fallback: 344 for (i = 0; i < ngroups; i++) { 345 group = (parent_group + i) % ngroups; 346 desc = ext2_get_group_desc (sb, group, NULL); 347 if (!desc || !desc->bg_free_inodes_count) 348 continue; 349 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) 350 goto found; 351 } 352 353 if (avefreei) { 354 /* 355 * The free-inodes counter is approximate, and for really small 356 * filesystems the above test can fail to find any blockgroups 357 */ 358 avefreei = 0; 359 goto fallback; 360 } 361 362 return -1; 363 364 found: 365 return group; 366 } 367 368 static int find_group_other(struct super_block *sb, struct inode *parent) 369 { 370 int parent_group = EXT2_I(parent)->i_block_group; 371 int ngroups = EXT2_SB(sb)->s_groups_count; 372 struct ext2_group_desc *desc; 373 int group, i; 374 375 /* 376 * Try to place the inode in its parent directory 377 */ 378 group = parent_group; 379 desc = ext2_get_group_desc (sb, group, NULL); 380 if (desc && le16_to_cpu(desc->bg_free_inodes_count) && 381 le16_to_cpu(desc->bg_free_blocks_count)) 382 goto found; 383 384 /* 385 * We're going to place this inode in a different blockgroup from its 386 * parent. We want to cause files in a common directory to all land in 387 * the same blockgroup. But we want files which are in a different 388 * directory which shares a blockgroup with our parent to land in a 389 * different blockgroup. 390 * 391 * So add our directory's i_ino into the starting point for the hash. 392 */ 393 group = (group + parent->i_ino) % ngroups; 394 395 /* 396 * Use a quadratic hash to find a group with a free inode and some 397 * free blocks. 398 */ 399 for (i = 1; i < ngroups; i <<= 1) { 400 group += i; 401 if (group >= ngroups) 402 group -= ngroups; 403 desc = ext2_get_group_desc (sb, group, NULL); 404 if (desc && le16_to_cpu(desc->bg_free_inodes_count) && 405 le16_to_cpu(desc->bg_free_blocks_count)) 406 goto found; 407 } 408 409 /* 410 * That failed: try linear search for a free inode, even if that group 411 * has no free blocks. 412 */ 413 group = parent_group; 414 for (i = 0; i < ngroups; i++) { 415 if (++group >= ngroups) 416 group = 0; 417 desc = ext2_get_group_desc (sb, group, NULL); 418 if (desc && le16_to_cpu(desc->bg_free_inodes_count)) 419 goto found; 420 } 421 422 return -1; 423 424 found: 425 return group; 426 } 427 428 struct inode *ext2_new_inode(struct inode *dir, umode_t mode, 429 const struct qstr *qstr) 430 { 431 struct super_block *sb; 432 struct buffer_head *bitmap_bh = NULL; 433 struct buffer_head *bh2; 434 int group, i; 435 ino_t ino = 0; 436 struct inode * inode; 437 struct ext2_group_desc *gdp; 438 struct ext2_super_block *es; 439 struct ext2_inode_info *ei; 440 struct ext2_sb_info *sbi; 441 int err; 442 443 sb = dir->i_sb; 444 inode = new_inode(sb); 445 if (!inode) 446 return ERR_PTR(-ENOMEM); 447 448 ei = EXT2_I(inode); 449 sbi = EXT2_SB(sb); 450 es = sbi->s_es; 451 if (S_ISDIR(mode)) { 452 if (test_opt(sb, OLDALLOC)) 453 group = find_group_dir(sb, dir); 454 else 455 group = find_group_orlov(sb, dir); 456 } else 457 group = find_group_other(sb, dir); 458 459 if (group == -1) { 460 err = -ENOSPC; 461 goto fail; 462 } 463 464 for (i = 0; i < sbi->s_groups_count; i++) { 465 gdp = ext2_get_group_desc(sb, group, &bh2); 466 if (!gdp) { 467 if (++group == sbi->s_groups_count) 468 group = 0; 469 continue; 470 } 471 brelse(bitmap_bh); 472 bitmap_bh = read_inode_bitmap(sb, group); 473 if (!bitmap_bh) { 474 err = -EIO; 475 goto fail; 476 } 477 ino = 0; 478 479 repeat_in_this_group: 480 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data, 481 EXT2_INODES_PER_GROUP(sb), ino); 482 if (ino >= EXT2_INODES_PER_GROUP(sb)) { 483 /* 484 * Rare race: find_group_xx() decided that there were 485 * free inodes in this group, but by the time we tried 486 * to allocate one, they're all gone. This can also 487 * occur because the counters which find_group_orlov() 488 * uses are approximate. So just go and search the 489 * next block group. 490 */ 491 if (++group == sbi->s_groups_count) 492 group = 0; 493 continue; 494 } 495 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group), 496 ino, bitmap_bh->b_data)) { 497 /* we lost this inode */ 498 if (++ino >= EXT2_INODES_PER_GROUP(sb)) { 499 /* this group is exhausted, try next group */ 500 if (++group == sbi->s_groups_count) 501 group = 0; 502 continue; 503 } 504 /* try to find free inode in the same group */ 505 goto repeat_in_this_group; 506 } 507 goto got; 508 } 509 510 /* 511 * Scanned all blockgroups. 512 */ 513 brelse(bitmap_bh); 514 err = -ENOSPC; 515 goto fail; 516 got: 517 mark_buffer_dirty(bitmap_bh); 518 if (sb->s_flags & SB_SYNCHRONOUS) 519 sync_dirty_buffer(bitmap_bh); 520 brelse(bitmap_bh); 521 522 ino += group * EXT2_INODES_PER_GROUP(sb) + 1; 523 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { 524 ext2_error (sb, "ext2_new_inode", 525 "reserved inode or inode > inodes count - " 526 "block_group = %d,inode=%lu", group, 527 (unsigned long) ino); 528 err = -EIO; 529 goto fail; 530 } 531 532 percpu_counter_dec(&sbi->s_freeinodes_counter); 533 if (S_ISDIR(mode)) 534 percpu_counter_inc(&sbi->s_dirs_counter); 535 536 spin_lock(sb_bgl_lock(sbi, group)); 537 le16_add_cpu(&gdp->bg_free_inodes_count, -1); 538 if (S_ISDIR(mode)) { 539 if (sbi->s_debts[group] < 255) 540 sbi->s_debts[group]++; 541 le16_add_cpu(&gdp->bg_used_dirs_count, 1); 542 } else { 543 if (sbi->s_debts[group]) 544 sbi->s_debts[group]--; 545 } 546 spin_unlock(sb_bgl_lock(sbi, group)); 547 548 mark_buffer_dirty(bh2); 549 if (test_opt(sb, GRPID)) { 550 inode->i_mode = mode; 551 inode->i_uid = current_fsuid(); 552 inode->i_gid = dir->i_gid; 553 } else 554 inode_init_owner(inode, dir, mode); 555 556 inode->i_ino = ino; 557 inode->i_blocks = 0; 558 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); 559 memset(ei->i_data, 0, sizeof(ei->i_data)); 560 ei->i_flags = 561 ext2_mask_flags(mode, EXT2_I(dir)->i_flags & EXT2_FL_INHERITED); 562 ei->i_faddr = 0; 563 ei->i_frag_no = 0; 564 ei->i_frag_size = 0; 565 ei->i_file_acl = 0; 566 ei->i_dir_acl = 0; 567 ei->i_dtime = 0; 568 ei->i_block_alloc_info = NULL; 569 ei->i_block_group = group; 570 ei->i_dir_start_lookup = 0; 571 ei->i_state = EXT2_STATE_NEW; 572 ext2_set_inode_flags(inode); 573 spin_lock(&sbi->s_next_gen_lock); 574 inode->i_generation = sbi->s_next_generation++; 575 spin_unlock(&sbi->s_next_gen_lock); 576 if (insert_inode_locked(inode) < 0) { 577 ext2_error(sb, "ext2_new_inode", 578 "inode number already in use - inode=%lu", 579 (unsigned long) ino); 580 err = -EIO; 581 goto fail; 582 } 583 584 err = dquot_initialize(inode); 585 if (err) 586 goto fail_drop; 587 588 err = dquot_alloc_inode(inode); 589 if (err) 590 goto fail_drop; 591 592 err = ext2_init_acl(inode, dir); 593 if (err) 594 goto fail_free_drop; 595 596 err = ext2_init_security(inode, dir, qstr); 597 if (err) 598 goto fail_free_drop; 599 600 mark_inode_dirty(inode); 601 ext2_debug("allocating inode %lu\n", inode->i_ino); 602 ext2_preread_inode(inode); 603 return inode; 604 605 fail_free_drop: 606 dquot_free_inode(inode); 607 608 fail_drop: 609 dquot_drop(inode); 610 inode->i_flags |= S_NOQUOTA; 611 clear_nlink(inode); 612 discard_new_inode(inode); 613 return ERR_PTR(err); 614 615 fail: 616 make_bad_inode(inode); 617 iput(inode); 618 return ERR_PTR(err); 619 } 620 621 unsigned long ext2_count_free_inodes (struct super_block * sb) 622 { 623 struct ext2_group_desc *desc; 624 unsigned long desc_count = 0; 625 int i; 626 627 #ifdef EXT2FS_DEBUG 628 struct ext2_super_block *es; 629 unsigned long bitmap_count = 0; 630 struct buffer_head *bitmap_bh = NULL; 631 632 es = EXT2_SB(sb)->s_es; 633 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 634 unsigned x; 635 636 desc = ext2_get_group_desc (sb, i, NULL); 637 if (!desc) 638 continue; 639 desc_count += le16_to_cpu(desc->bg_free_inodes_count); 640 brelse(bitmap_bh); 641 bitmap_bh = read_inode_bitmap(sb, i); 642 if (!bitmap_bh) 643 continue; 644 645 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8); 646 printk("group %d: stored = %d, counted = %u\n", 647 i, le16_to_cpu(desc->bg_free_inodes_count), x); 648 bitmap_count += x; 649 } 650 brelse(bitmap_bh); 651 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n", 652 (unsigned long) 653 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter), 654 desc_count, bitmap_count); 655 return desc_count; 656 #else 657 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 658 desc = ext2_get_group_desc (sb, i, NULL); 659 if (!desc) 660 continue; 661 desc_count += le16_to_cpu(desc->bg_free_inodes_count); 662 } 663 return desc_count; 664 #endif 665 } 666 667 /* Called at mount-time, super-block is locked */ 668 unsigned long ext2_count_dirs (struct super_block * sb) 669 { 670 unsigned long count = 0; 671 int i; 672 673 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 674 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL); 675 if (!gdp) 676 continue; 677 count += le16_to_cpu(gdp->bg_used_dirs_count); 678 } 679 return count; 680 } 681 682