1/* 2 * linux/fs/ext2/ialloc.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 * BSD ufs-inspired inode and directory allocation by 10 * Stephen Tweedie (sct@dcs.ed.ac.uk), 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/quotaops.h> 16#include <linux/sched.h> 17#include <linux/backing-dev.h> 18#include <linux/buffer_head.h> 19#include <linux/random.h> 20#include "ext2.h" 21#include "xattr.h" 22#include "acl.h" 23 24/* 25 * ialloc.c contains the inodes allocation and deallocation routines 26 */ 27 28/* 29 * The free inodes are managed by bitmaps. A file system contains several 30 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 31 * block for inodes, N blocks for the inode table and data blocks. 32 * 33 * The file system contains group descriptors which are located after the 34 * super block. Each descriptor contains the number of the bitmap block and 35 * the free blocks count in the block. 36 */ 37 38 39/* 40 * Read the inode allocation bitmap for a given block_group, reading 41 * into the specified slot in the superblock's bitmap cache. 42 * 43 * Return buffer_head of bitmap on success or NULL. 44 */ 45static struct buffer_head * 46read_inode_bitmap(struct super_block * sb, unsigned long block_group) 47{ 48 struct ext2_group_desc *desc; 49 struct buffer_head *bh = NULL; 50 51 desc = ext2_get_group_desc(sb, block_group, NULL); 52 if (!desc) 53 goto error_out; 54 55 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap)); 56 if (!bh) 57 ext2_error(sb, "read_inode_bitmap", 58 "Cannot read inode bitmap - " 59 "block_group = %lu, inode_bitmap = %u", 60 block_group, le32_to_cpu(desc->bg_inode_bitmap)); 61error_out: 62 return bh; 63} 64 65static void ext2_release_inode(struct super_block *sb, int group, int dir) 66{ 67 struct ext2_group_desc * desc; 68 struct buffer_head *bh; 69 70 desc = ext2_get_group_desc(sb, group, &bh); 71 if (!desc) { 72 ext2_error(sb, "ext2_release_inode", 73 "can't get descriptor for group %d", group); 74 return; 75 } 76 77 spin_lock(sb_bgl_lock(EXT2_SB(sb), group)); 78 le16_add_cpu(&desc->bg_free_inodes_count, 1); 79 if (dir) 80 le16_add_cpu(&desc->bg_used_dirs_count, -1); 81 spin_unlock(sb_bgl_lock(EXT2_SB(sb), group)); 82 if (dir) 83 percpu_counter_dec(&EXT2_SB(sb)->s_dirs_counter); 84 sb->s_dirt = 1; 85 mark_buffer_dirty(bh); 86} 87 88/* 89 * NOTE! When we get the inode, we're the only people 90 * that have access to it, and as such there are no 91 * race conditions we have to worry about. The inode 92 * is not on the hash-lists, and it cannot be reached 93 * through the filesystem because the directory entry 94 * has been deleted earlier. 95 * 96 * HOWEVER: we must make sure that we get no aliases, 97 * which means that we have to call "clear_inode()" 98 * _before_ we mark the inode not in use in the inode 99 * bitmaps. Otherwise a newly created file might use 100 * the same inode number (not actually the same pointer 101 * though), and then we'd have two inodes sharing the 102 * same inode number and space on the harddisk. 103 */ 104void ext2_free_inode (struct inode * inode) 105{ 106 struct super_block * sb = inode->i_sb; 107 int is_directory; 108 unsigned long ino; 109 struct buffer_head *bitmap_bh; 110 unsigned long block_group; 111 unsigned long bit; 112 struct ext2_super_block * es; 113 114 ino = inode->i_ino; 115 ext2_debug ("freeing inode %lu\n", ino); 116 117 /* 118 * Note: we must free any quota before locking the superblock, 119 * as writing the quota to disk may need the lock as well. 120 */ 121 /* Quota is already initialized in iput() */ 122 ext2_xattr_delete_inode(inode); 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 & MS_SYNCHRONOUS) 150 sync_dirty_buffer(bitmap_bh); 151 152 brelse(bitmap_bh); 153} 154 155static void ext2_preread_inode(struct inode *inode) 156{ 157 unsigned long block_group; 158 unsigned long offset; 159 unsigned long block; 160 struct ext2_group_desc * gdp; 161 struct backing_dev_info *bdi; 162 163 bdi = inode->i_mapping->backing_dev_info; 164 if (bdi_read_congested(bdi)) 165 return; 166 if (bdi_write_congested(bdi)) 167 return; 168 169 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 170 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL); 171 if (gdp == NULL) 172 return; 173 174 /* 175 * Figure out the offset within the block group inode table 176 */ 177 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) * 178 EXT2_INODE_SIZE(inode->i_sb); 179 block = le32_to_cpu(gdp->bg_inode_table) + 180 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb)); 181 sb_breadahead(inode->i_sb, block); 182} 183 184/* 185 * There are two policies for allocating an inode. If the new inode is 186 * a directory, then a forward search is made for a block group with both 187 * free space and a low directory-to-inode ratio; if that fails, then of 188 * the groups with above-average free space, that group with the fewest 189 * directories already is chosen. 190 * 191 * For other inodes, search forward from the parent directory\'s block 192 * group to find a free inode. 193 */ 194static int find_group_dir(struct super_block *sb, struct inode *parent) 195{ 196 int ngroups = EXT2_SB(sb)->s_groups_count; 197 int avefreei = ext2_count_free_inodes(sb) / ngroups; 198 struct ext2_group_desc *desc, *best_desc = NULL; 199 int group, best_group = -1; 200 201 for (group = 0; group < ngroups; group++) { 202 desc = ext2_get_group_desc (sb, group, NULL); 203 if (!desc || !desc->bg_free_inodes_count) 204 continue; 205 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) 206 continue; 207 if (!best_desc || 208 (le16_to_cpu(desc->bg_free_blocks_count) > 209 le16_to_cpu(best_desc->bg_free_blocks_count))) { 210 best_group = group; 211 best_desc = desc; 212 } 213 } 214 if (!best_desc) 215 return -1; 216 217 return best_group; 218} 219 220/* 221 * Orlov's allocator for directories. 222 * 223 * We always try to spread first-level directories. 224 * 225 * If there are blockgroups with both free inodes and free blocks counts 226 * not worse than average we return one with smallest directory count. 227 * Otherwise we simply return a random group. 228 * 229 * For the rest rules look so: 230 * 231 * It's OK to put directory into a group unless 232 * it has too many directories already (max_dirs) or 233 * it has too few free inodes left (min_inodes) or 234 * it has too few free blocks left (min_blocks) or 235 * it's already running too large debt (max_debt). 236 * Parent's group is preferred, if it doesn't satisfy these 237 * conditions we search cyclically through the rest. If none 238 * of the groups look good we just look for a group with more 239 * free inodes than average (starting at parent's group). 240 * 241 * Debt is incremented each time we allocate a directory and decremented 242 * when we allocate an inode, within 0--255. 243 */ 244 245#define INODE_COST 64 246#define BLOCK_COST 256 247 248static int find_group_orlov(struct super_block *sb, struct inode *parent) 249{ 250 int parent_group = EXT2_I(parent)->i_block_group; 251 struct ext2_sb_info *sbi = EXT2_SB(sb); 252 struct ext2_super_block *es = sbi->s_es; 253 int ngroups = sbi->s_groups_count; 254 int inodes_per_group = EXT2_INODES_PER_GROUP(sb); 255 int freei; 256 int avefreei; 257 int free_blocks; 258 int avefreeb; 259 int blocks_per_dir; 260 int ndirs; 261 int max_debt, max_dirs, min_blocks, min_inodes; 262 int group = -1, i; 263 struct ext2_group_desc *desc; 264 265 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); 266 avefreei = freei / ngroups; 267 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); 268 avefreeb = free_blocks / ngroups; 269 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); 270 271 if ((parent == sb->s_root->d_inode) || 272 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) { 273 struct ext2_group_desc *best_desc = NULL; 274 int best_ndir = inodes_per_group; 275 int best_group = -1; 276 277 get_random_bytes(&group, sizeof(group)); 278 parent_group = (unsigned)group % ngroups; 279 for (i = 0; i < ngroups; i++) { 280 group = (parent_group + i) % ngroups; 281 desc = ext2_get_group_desc (sb, group, NULL); 282 if (!desc || !desc->bg_free_inodes_count) 283 continue; 284 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) 285 continue; 286 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) 287 continue; 288 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) 289 continue; 290 best_group = group; 291 best_ndir = le16_to_cpu(desc->bg_used_dirs_count); 292 best_desc = desc; 293 } 294 if (best_group >= 0) { 295 desc = best_desc; 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 335fallback: 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 356found: 357 return group; 358} 359 360static 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 + 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 416found: 417 return group; 418} 419 420struct inode *ext2_new_inode(struct inode *dir, int mode) 421{ 422 struct super_block *sb; 423 struct buffer_head *bitmap_bh = NULL; 424 struct buffer_head *bh2; 425 int group, i; 426 ino_t ino = 0; 427 struct inode * inode; 428 struct ext2_group_desc *gdp; 429 struct ext2_super_block *es; 430 struct ext2_inode_info *ei; 431 struct ext2_sb_info *sbi; 432 int err; 433 434 sb = dir->i_sb; 435 inode = new_inode(sb); 436 if (!inode) 437 return ERR_PTR(-ENOMEM); 438 439 ei = EXT2_I(inode); 440 sbi = EXT2_SB(sb); 441 es = sbi->s_es; 442 if (S_ISDIR(mode)) { 443 if (test_opt(sb, OLDALLOC)) 444 group = find_group_dir(sb, dir); 445 else 446 group = find_group_orlov(sb, dir); 447 } else 448 group = find_group_other(sb, dir); 449 450 if (group == -1) { 451 err = -ENOSPC; 452 goto fail; 453 } 454 455 for (i = 0; i < sbi->s_groups_count; i++) { 456 gdp = ext2_get_group_desc(sb, group, &bh2); 457 brelse(bitmap_bh); 458 bitmap_bh = read_inode_bitmap(sb, group); 459 if (!bitmap_bh) { 460 err = -EIO; 461 goto fail; 462 } 463 ino = 0; 464 465repeat_in_this_group: 466 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data, 467 EXT2_INODES_PER_GROUP(sb), ino); 468 if (ino >= EXT2_INODES_PER_GROUP(sb)) { 469 /* 470 * Rare race: find_group_xx() decided that there were 471 * free inodes in this group, but by the time we tried 472 * to allocate one, they're all gone. This can also 473 * occur because the counters which find_group_orlov() 474 * uses are approximate. So just go and search the 475 * next block group. 476 */ 477 if (++group == sbi->s_groups_count) 478 group = 0; 479 continue; 480 } 481 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group), 482 ino, bitmap_bh->b_data)) { 483 /* we lost this inode */ 484 if (++ino >= EXT2_INODES_PER_GROUP(sb)) { 485 /* this group is exhausted, try next group */ 486 if (++group == sbi->s_groups_count) 487 group = 0; 488 continue; 489 } 490 /* try to find free inode in the same group */ 491 goto repeat_in_this_group; 492 } 493 goto got; 494 } 495 496 /* 497 * Scanned all blockgroups. 498 */ 499 err = -ENOSPC; 500 goto fail; 501got: 502 mark_buffer_dirty(bitmap_bh); 503 if (sb->s_flags & MS_SYNCHRONOUS) 504 sync_dirty_buffer(bitmap_bh); 505 brelse(bitmap_bh); 506 507 ino += group * EXT2_INODES_PER_GROUP(sb) + 1; 508 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { 509 ext2_error (sb, "ext2_new_inode", 510 "reserved inode or inode > inodes count - " 511 "block_group = %d,inode=%lu", group, 512 (unsigned long) ino); 513 err = -EIO; 514 goto fail; 515 } 516 517 percpu_counter_add(&sbi->s_freeinodes_counter, -1); 518 if (S_ISDIR(mode)) 519 percpu_counter_inc(&sbi->s_dirs_counter); 520 521 spin_lock(sb_bgl_lock(sbi, group)); 522 le16_add_cpu(&gdp->bg_free_inodes_count, -1); 523 if (S_ISDIR(mode)) { 524 if (sbi->s_debts[group] < 255) 525 sbi->s_debts[group]++; 526 le16_add_cpu(&gdp->bg_used_dirs_count, 1); 527 } else { 528 if (sbi->s_debts[group]) 529 sbi->s_debts[group]--; 530 } 531 spin_unlock(sb_bgl_lock(sbi, group)); 532 533 sb->s_dirt = 1; 534 mark_buffer_dirty(bh2); 535 if (test_opt(sb, GRPID)) { 536 inode->i_mode = mode; 537 inode->i_uid = current_fsuid(); 538 inode->i_gid = dir->i_gid; 539 } else 540 inode_init_owner(inode, dir, mode); 541 542 inode->i_ino = ino; 543 inode->i_blocks = 0; 544 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC; 545 memset(ei->i_data, 0, sizeof(ei->i_data)); 546 ei->i_flags = 547 ext2_mask_flags(mode, EXT2_I(dir)->i_flags & EXT2_FL_INHERITED); 548 ei->i_faddr = 0; 549 ei->i_frag_no = 0; 550 ei->i_frag_size = 0; 551 ei->i_file_acl = 0; 552 ei->i_dir_acl = 0; 553 ei->i_dtime = 0; 554 ei->i_block_alloc_info = NULL; 555 ei->i_block_group = group; 556 ei->i_dir_start_lookup = 0; 557 ei->i_state = EXT2_STATE_NEW; 558 ext2_set_inode_flags(inode); 559 spin_lock(&sbi->s_next_gen_lock); 560 inode->i_generation = sbi->s_next_generation++; 561 spin_unlock(&sbi->s_next_gen_lock); 562 if (insert_inode_locked(inode) < 0) { 563 err = -EINVAL; 564 goto fail_drop; 565 } 566 567 dquot_initialize(inode); 568 err = dquot_alloc_inode(inode); 569 if (err) 570 goto fail_drop; 571 572 err = ext2_init_acl(inode, dir); 573 if (err) 574 goto fail_free_drop; 575 576 err = ext2_init_security(inode,dir); 577 if (err) 578 goto fail_free_drop; 579 580 mark_inode_dirty(inode); 581 ext2_debug("allocating inode %lu\n", inode->i_ino); 582 ext2_preread_inode(inode); 583 return inode; 584 585fail_free_drop: 586 dquot_free_inode(inode); 587 588fail_drop: 589 dquot_drop(inode); 590 inode->i_flags |= S_NOQUOTA; 591 inode->i_nlink = 0; 592 unlock_new_inode(inode); 593 iput(inode); 594 return ERR_PTR(err); 595 596fail: 597 make_bad_inode(inode); 598 iput(inode); 599 return ERR_PTR(err); 600} 601 602unsigned long ext2_count_free_inodes (struct super_block * sb) 603{ 604 struct ext2_group_desc *desc; 605 unsigned long desc_count = 0; 606 int i; 607 608#ifdef EXT2FS_DEBUG 609 struct ext2_super_block *es; 610 unsigned long bitmap_count = 0; 611 struct buffer_head *bitmap_bh = NULL; 612 613 es = EXT2_SB(sb)->s_es; 614 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 615 unsigned x; 616 617 desc = ext2_get_group_desc (sb, i, NULL); 618 if (!desc) 619 continue; 620 desc_count += le16_to_cpu(desc->bg_free_inodes_count); 621 brelse(bitmap_bh); 622 bitmap_bh = read_inode_bitmap(sb, i); 623 if (!bitmap_bh) 624 continue; 625 626 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8); 627 printk("group %d: stored = %d, counted = %u\n", 628 i, le16_to_cpu(desc->bg_free_inodes_count), x); 629 bitmap_count += x; 630 } 631 brelse(bitmap_bh); 632 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n", 633 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter), 634 desc_count, bitmap_count); 635 return desc_count; 636#else 637 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 638 desc = ext2_get_group_desc (sb, i, NULL); 639 if (!desc) 640 continue; 641 desc_count += le16_to_cpu(desc->bg_free_inodes_count); 642 } 643 return desc_count; 644#endif 645} 646 647/* Called at mount-time, super-block is locked */ 648unsigned long ext2_count_dirs (struct super_block * sb) 649{ 650 unsigned long count = 0; 651 int i; 652 653 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) { 654 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL); 655 if (!gdp) 656 continue; 657 count += le16_to_cpu(gdp->bg_used_dirs_count); 658 } 659 return count; 660} 661