1 /* 2 * linux/fs/ext4/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@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/fs.h> 17 #include <linux/jbd2.h> 18 #include <linux/stat.h> 19 #include <linux/string.h> 20 #include <linux/quotaops.h> 21 #include <linux/buffer_head.h> 22 #include <linux/random.h> 23 #include <linux/bitops.h> 24 #include <linux/blkdev.h> 25 #include <asm/byteorder.h> 26 27 #include "ext4.h" 28 #include "ext4_jbd2.h" 29 #include "xattr.h" 30 #include "acl.h" 31 32 #include <trace/events/ext4.h> 33 34 /* 35 * ialloc.c contains the inodes allocation and deallocation routines 36 */ 37 38 /* 39 * The free inodes are managed by bitmaps. A file system contains several 40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 41 * block for inodes, N blocks for the inode table and data blocks. 42 * 43 * The file system contains group descriptors which are located after the 44 * super block. Each descriptor contains the number of the bitmap block and 45 * the free blocks count in the block. 46 */ 47 48 /* 49 * To avoid calling the atomic setbit hundreds or thousands of times, we only 50 * need to use it within a single byte (to ensure we get endianness right). 51 * We can use memset for the rest of the bitmap as there are no other users. 52 */ 53 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap) 54 { 55 int i; 56 57 if (start_bit >= end_bit) 58 return; 59 60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); 61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) 62 ext4_set_bit(i, bitmap); 63 if (i < end_bit) 64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); 65 } 66 67 /* Initializes an uninitialized inode bitmap */ 68 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh, 69 ext4_group_t block_group, 70 struct ext4_group_desc *gdp) 71 { 72 struct ext4_sb_info *sbi = EXT4_SB(sb); 73 74 J_ASSERT_BH(bh, buffer_locked(bh)); 75 76 /* If checksum is bad mark all blocks and inodes use to prevent 77 * allocation, essentially implementing a per-group read-only flag. */ 78 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { 79 ext4_error(sb, "Checksum bad for group %u", block_group); 80 ext4_free_blks_set(sb, gdp, 0); 81 ext4_free_inodes_set(sb, gdp, 0); 82 ext4_itable_unused_set(sb, gdp, 0); 83 memset(bh->b_data, 0xff, sb->s_blocksize); 84 return 0; 85 } 86 87 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); 88 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, 89 bh->b_data); 90 91 return EXT4_INODES_PER_GROUP(sb); 92 } 93 94 /* 95 * Read the inode allocation bitmap for a given block_group, reading 96 * into the specified slot in the superblock's bitmap cache. 97 * 98 * Return buffer_head of bitmap on success or NULL. 99 */ 100 static struct buffer_head * 101 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) 102 { 103 struct ext4_group_desc *desc; 104 struct buffer_head *bh = NULL; 105 ext4_fsblk_t bitmap_blk; 106 107 desc = ext4_get_group_desc(sb, block_group, NULL); 108 if (!desc) 109 return NULL; 110 bitmap_blk = ext4_inode_bitmap(sb, desc); 111 bh = sb_getblk(sb, bitmap_blk); 112 if (unlikely(!bh)) { 113 ext4_error(sb, "Cannot read inode bitmap - " 114 "block_group = %u, inode_bitmap = %llu", 115 block_group, bitmap_blk); 116 return NULL; 117 } 118 if (bitmap_uptodate(bh)) 119 return bh; 120 121 lock_buffer(bh); 122 if (bitmap_uptodate(bh)) { 123 unlock_buffer(bh); 124 return bh; 125 } 126 ext4_lock_group(sb, block_group); 127 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { 128 ext4_init_inode_bitmap(sb, bh, block_group, desc); 129 set_bitmap_uptodate(bh); 130 set_buffer_uptodate(bh); 131 ext4_unlock_group(sb, block_group); 132 unlock_buffer(bh); 133 return bh; 134 } 135 ext4_unlock_group(sb, block_group); 136 if (buffer_uptodate(bh)) { 137 /* 138 * if not uninit if bh is uptodate, 139 * bitmap is also uptodate 140 */ 141 set_bitmap_uptodate(bh); 142 unlock_buffer(bh); 143 return bh; 144 } 145 /* 146 * submit the buffer_head for read. We can 147 * safely mark the bitmap as uptodate now. 148 * We do it here so the bitmap uptodate bit 149 * get set with buffer lock held. 150 */ 151 set_bitmap_uptodate(bh); 152 if (bh_submit_read(bh) < 0) { 153 put_bh(bh); 154 ext4_error(sb, "Cannot read inode bitmap - " 155 "block_group = %u, inode_bitmap = %llu", 156 block_group, bitmap_blk); 157 return NULL; 158 } 159 return bh; 160 } 161 162 /* 163 * NOTE! When we get the inode, we're the only people 164 * that have access to it, and as such there are no 165 * race conditions we have to worry about. The inode 166 * is not on the hash-lists, and it cannot be reached 167 * through the filesystem because the directory entry 168 * has been deleted earlier. 169 * 170 * HOWEVER: we must make sure that we get no aliases, 171 * which means that we have to call "clear_inode()" 172 * _before_ we mark the inode not in use in the inode 173 * bitmaps. Otherwise a newly created file might use 174 * the same inode number (not actually the same pointer 175 * though), and then we'd have two inodes sharing the 176 * same inode number and space on the harddisk. 177 */ 178 void ext4_free_inode(handle_t *handle, struct inode *inode) 179 { 180 struct super_block *sb = inode->i_sb; 181 int is_directory; 182 unsigned long ino; 183 struct buffer_head *bitmap_bh = NULL; 184 struct buffer_head *bh2; 185 ext4_group_t block_group; 186 unsigned long bit; 187 struct ext4_group_desc *gdp; 188 struct ext4_super_block *es; 189 struct ext4_sb_info *sbi; 190 int fatal = 0, err, count, cleared; 191 192 if (atomic_read(&inode->i_count) > 1) { 193 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n", 194 atomic_read(&inode->i_count)); 195 return; 196 } 197 if (inode->i_nlink) { 198 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n", 199 inode->i_nlink); 200 return; 201 } 202 if (!sb) { 203 printk(KERN_ERR "ext4_free_inode: inode on " 204 "nonexistent device\n"); 205 return; 206 } 207 sbi = EXT4_SB(sb); 208 209 ino = inode->i_ino; 210 ext4_debug("freeing inode %lu\n", ino); 211 trace_ext4_free_inode(inode); 212 213 /* 214 * Note: we must free any quota before locking the superblock, 215 * as writing the quota to disk may need the lock as well. 216 */ 217 dquot_initialize(inode); 218 ext4_xattr_delete_inode(handle, inode); 219 dquot_free_inode(inode); 220 dquot_drop(inode); 221 222 is_directory = S_ISDIR(inode->i_mode); 223 224 /* Do this BEFORE marking the inode not in use or returning an error */ 225 ext4_clear_inode(inode); 226 227 es = EXT4_SB(sb)->s_es; 228 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { 229 ext4_error(sb, "reserved or nonexistent inode %lu", ino); 230 goto error_return; 231 } 232 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 233 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 234 bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 235 if (!bitmap_bh) 236 goto error_return; 237 238 BUFFER_TRACE(bitmap_bh, "get_write_access"); 239 fatal = ext4_journal_get_write_access(handle, bitmap_bh); 240 if (fatal) 241 goto error_return; 242 243 fatal = -ESRCH; 244 gdp = ext4_get_group_desc(sb, block_group, &bh2); 245 if (gdp) { 246 BUFFER_TRACE(bh2, "get_write_access"); 247 fatal = ext4_journal_get_write_access(handle, bh2); 248 } 249 ext4_lock_group(sb, block_group); 250 cleared = ext4_clear_bit(bit, bitmap_bh->b_data); 251 if (fatal || !cleared) { 252 ext4_unlock_group(sb, block_group); 253 goto out; 254 } 255 256 count = ext4_free_inodes_count(sb, gdp) + 1; 257 ext4_free_inodes_set(sb, gdp, count); 258 if (is_directory) { 259 count = ext4_used_dirs_count(sb, gdp) - 1; 260 ext4_used_dirs_set(sb, gdp, count); 261 percpu_counter_dec(&sbi->s_dirs_counter); 262 } 263 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp); 264 ext4_unlock_group(sb, block_group); 265 266 percpu_counter_inc(&sbi->s_freeinodes_counter); 267 if (sbi->s_log_groups_per_flex) { 268 ext4_group_t f = ext4_flex_group(sbi, block_group); 269 270 atomic_inc(&sbi->s_flex_groups[f].free_inodes); 271 if (is_directory) 272 atomic_dec(&sbi->s_flex_groups[f].used_dirs); 273 } 274 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); 275 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); 276 out: 277 if (cleared) { 278 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); 279 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); 280 if (!fatal) 281 fatal = err; 282 ext4_mark_super_dirty(sb); 283 } else 284 ext4_error(sb, "bit already cleared for inode %lu", ino); 285 286 error_return: 287 brelse(bitmap_bh); 288 ext4_std_error(sb, fatal); 289 } 290 291 /* 292 * There are two policies for allocating an inode. If the new inode is 293 * a directory, then a forward search is made for a block group with both 294 * free space and a low directory-to-inode ratio; if that fails, then of 295 * the groups with above-average free space, that group with the fewest 296 * directories already is chosen. 297 * 298 * For other inodes, search forward from the parent directory\'s block 299 * group to find a free inode. 300 */ 301 static int find_group_dir(struct super_block *sb, struct inode *parent, 302 ext4_group_t *best_group) 303 { 304 ext4_group_t ngroups = ext4_get_groups_count(sb); 305 unsigned int freei, avefreei; 306 struct ext4_group_desc *desc, *best_desc = NULL; 307 ext4_group_t group; 308 int ret = -1; 309 310 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); 311 avefreei = freei / ngroups; 312 313 for (group = 0; group < ngroups; group++) { 314 desc = ext4_get_group_desc(sb, group, NULL); 315 if (!desc || !ext4_free_inodes_count(sb, desc)) 316 continue; 317 if (ext4_free_inodes_count(sb, desc) < avefreei) 318 continue; 319 if (!best_desc || 320 (ext4_free_blks_count(sb, desc) > 321 ext4_free_blks_count(sb, best_desc))) { 322 *best_group = group; 323 best_desc = desc; 324 ret = 0; 325 } 326 } 327 return ret; 328 } 329 330 #define free_block_ratio 10 331 332 static int find_group_flex(struct super_block *sb, struct inode *parent, 333 ext4_group_t *best_group) 334 { 335 struct ext4_sb_info *sbi = EXT4_SB(sb); 336 struct ext4_group_desc *desc; 337 struct flex_groups *flex_group = sbi->s_flex_groups; 338 ext4_group_t parent_group = EXT4_I(parent)->i_block_group; 339 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); 340 ext4_group_t ngroups = ext4_get_groups_count(sb); 341 int flex_size = ext4_flex_bg_size(sbi); 342 ext4_group_t best_flex = parent_fbg_group; 343 int blocks_per_flex = sbi->s_blocks_per_group * flex_size; 344 int flexbg_free_blocks; 345 int flex_freeb_ratio; 346 ext4_group_t n_fbg_groups; 347 ext4_group_t i; 348 349 n_fbg_groups = (ngroups + flex_size - 1) >> 350 sbi->s_log_groups_per_flex; 351 352 find_close_to_parent: 353 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks); 354 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; 355 if (atomic_read(&flex_group[best_flex].free_inodes) && 356 flex_freeb_ratio > free_block_ratio) 357 goto found_flexbg; 358 359 if (best_flex && best_flex == parent_fbg_group) { 360 best_flex--; 361 goto find_close_to_parent; 362 } 363 364 for (i = 0; i < n_fbg_groups; i++) { 365 if (i == parent_fbg_group || i == parent_fbg_group - 1) 366 continue; 367 368 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks); 369 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; 370 371 if (flex_freeb_ratio > free_block_ratio && 372 (atomic_read(&flex_group[i].free_inodes))) { 373 best_flex = i; 374 goto found_flexbg; 375 } 376 377 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) || 378 ((atomic_read(&flex_group[i].free_blocks) > 379 atomic_read(&flex_group[best_flex].free_blocks)) && 380 atomic_read(&flex_group[i].free_inodes))) 381 best_flex = i; 382 } 383 384 if (!atomic_read(&flex_group[best_flex].free_inodes) || 385 !atomic_read(&flex_group[best_flex].free_blocks)) 386 return -1; 387 388 found_flexbg: 389 for (i = best_flex * flex_size; i < ngroups && 390 i < (best_flex + 1) * flex_size; i++) { 391 desc = ext4_get_group_desc(sb, i, NULL); 392 if (ext4_free_inodes_count(sb, desc)) { 393 *best_group = i; 394 goto out; 395 } 396 } 397 398 return -1; 399 out: 400 return 0; 401 } 402 403 struct orlov_stats { 404 __u32 free_inodes; 405 __u32 free_blocks; 406 __u32 used_dirs; 407 }; 408 409 /* 410 * Helper function for Orlov's allocator; returns critical information 411 * for a particular block group or flex_bg. If flex_size is 1, then g 412 * is a block group number; otherwise it is flex_bg number. 413 */ 414 void get_orlov_stats(struct super_block *sb, ext4_group_t g, 415 int flex_size, struct orlov_stats *stats) 416 { 417 struct ext4_group_desc *desc; 418 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; 419 420 if (flex_size > 1) { 421 stats->free_inodes = atomic_read(&flex_group[g].free_inodes); 422 stats->free_blocks = atomic_read(&flex_group[g].free_blocks); 423 stats->used_dirs = atomic_read(&flex_group[g].used_dirs); 424 return; 425 } 426 427 desc = ext4_get_group_desc(sb, g, NULL); 428 if (desc) { 429 stats->free_inodes = ext4_free_inodes_count(sb, desc); 430 stats->free_blocks = ext4_free_blks_count(sb, desc); 431 stats->used_dirs = ext4_used_dirs_count(sb, desc); 432 } else { 433 stats->free_inodes = 0; 434 stats->free_blocks = 0; 435 stats->used_dirs = 0; 436 } 437 } 438 439 /* 440 * Orlov's allocator for directories. 441 * 442 * We always try to spread first-level directories. 443 * 444 * If there are blockgroups with both free inodes and free blocks counts 445 * not worse than average we return one with smallest directory count. 446 * Otherwise we simply return a random group. 447 * 448 * For the rest rules look so: 449 * 450 * It's OK to put directory into a group unless 451 * it has too many directories already (max_dirs) or 452 * it has too few free inodes left (min_inodes) or 453 * it has too few free blocks left (min_blocks) or 454 * Parent's group is preferred, if it doesn't satisfy these 455 * conditions we search cyclically through the rest. If none 456 * of the groups look good we just look for a group with more 457 * free inodes than average (starting at parent's group). 458 */ 459 460 static int find_group_orlov(struct super_block *sb, struct inode *parent, 461 ext4_group_t *group, int mode, 462 const struct qstr *qstr) 463 { 464 ext4_group_t parent_group = EXT4_I(parent)->i_block_group; 465 struct ext4_sb_info *sbi = EXT4_SB(sb); 466 ext4_group_t real_ngroups = ext4_get_groups_count(sb); 467 int inodes_per_group = EXT4_INODES_PER_GROUP(sb); 468 unsigned int freei, avefreei; 469 ext4_fsblk_t freeb, avefreeb; 470 unsigned int ndirs; 471 int max_dirs, min_inodes; 472 ext4_grpblk_t min_blocks; 473 ext4_group_t i, grp, g, ngroups; 474 struct ext4_group_desc *desc; 475 struct orlov_stats stats; 476 int flex_size = ext4_flex_bg_size(sbi); 477 struct dx_hash_info hinfo; 478 479 ngroups = real_ngroups; 480 if (flex_size > 1) { 481 ngroups = (real_ngroups + flex_size - 1) >> 482 sbi->s_log_groups_per_flex; 483 parent_group >>= sbi->s_log_groups_per_flex; 484 } 485 486 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); 487 avefreei = freei / ngroups; 488 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); 489 avefreeb = freeb; 490 do_div(avefreeb, ngroups); 491 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); 492 493 if (S_ISDIR(mode) && 494 ((parent == sb->s_root->d_inode) || 495 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { 496 int best_ndir = inodes_per_group; 497 int ret = -1; 498 499 if (qstr) { 500 hinfo.hash_version = DX_HASH_HALF_MD4; 501 hinfo.seed = sbi->s_hash_seed; 502 ext4fs_dirhash(qstr->name, qstr->len, &hinfo); 503 grp = hinfo.hash; 504 } else 505 get_random_bytes(&grp, sizeof(grp)); 506 parent_group = (unsigned)grp % ngroups; 507 for (i = 0; i < ngroups; i++) { 508 g = (parent_group + i) % ngroups; 509 get_orlov_stats(sb, g, flex_size, &stats); 510 if (!stats.free_inodes) 511 continue; 512 if (stats.used_dirs >= best_ndir) 513 continue; 514 if (stats.free_inodes < avefreei) 515 continue; 516 if (stats.free_blocks < avefreeb) 517 continue; 518 grp = g; 519 ret = 0; 520 best_ndir = stats.used_dirs; 521 } 522 if (ret) 523 goto fallback; 524 found_flex_bg: 525 if (flex_size == 1) { 526 *group = grp; 527 return 0; 528 } 529 530 /* 531 * We pack inodes at the beginning of the flexgroup's 532 * inode tables. Block allocation decisions will do 533 * something similar, although regular files will 534 * start at 2nd block group of the flexgroup. See 535 * ext4_ext_find_goal() and ext4_find_near(). 536 */ 537 grp *= flex_size; 538 for (i = 0; i < flex_size; i++) { 539 if (grp+i >= real_ngroups) 540 break; 541 desc = ext4_get_group_desc(sb, grp+i, NULL); 542 if (desc && ext4_free_inodes_count(sb, desc)) { 543 *group = grp+i; 544 return 0; 545 } 546 } 547 goto fallback; 548 } 549 550 max_dirs = ndirs / ngroups + inodes_per_group / 16; 551 min_inodes = avefreei - inodes_per_group*flex_size / 4; 552 if (min_inodes < 1) 553 min_inodes = 1; 554 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4; 555 556 /* 557 * Start looking in the flex group where we last allocated an 558 * inode for this parent directory 559 */ 560 if (EXT4_I(parent)->i_last_alloc_group != ~0) { 561 parent_group = EXT4_I(parent)->i_last_alloc_group; 562 if (flex_size > 1) 563 parent_group >>= sbi->s_log_groups_per_flex; 564 } 565 566 for (i = 0; i < ngroups; i++) { 567 grp = (parent_group + i) % ngroups; 568 get_orlov_stats(sb, grp, flex_size, &stats); 569 if (stats.used_dirs >= max_dirs) 570 continue; 571 if (stats.free_inodes < min_inodes) 572 continue; 573 if (stats.free_blocks < min_blocks) 574 continue; 575 goto found_flex_bg; 576 } 577 578 fallback: 579 ngroups = real_ngroups; 580 avefreei = freei / ngroups; 581 fallback_retry: 582 parent_group = EXT4_I(parent)->i_block_group; 583 for (i = 0; i < ngroups; i++) { 584 grp = (parent_group + i) % ngroups; 585 desc = ext4_get_group_desc(sb, grp, NULL); 586 if (desc && ext4_free_inodes_count(sb, desc) && 587 ext4_free_inodes_count(sb, desc) >= avefreei) { 588 *group = grp; 589 return 0; 590 } 591 } 592 593 if (avefreei) { 594 /* 595 * The free-inodes counter is approximate, and for really small 596 * filesystems the above test can fail to find any blockgroups 597 */ 598 avefreei = 0; 599 goto fallback_retry; 600 } 601 602 return -1; 603 } 604 605 static int find_group_other(struct super_block *sb, struct inode *parent, 606 ext4_group_t *group, int mode) 607 { 608 ext4_group_t parent_group = EXT4_I(parent)->i_block_group; 609 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); 610 struct ext4_group_desc *desc; 611 int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); 612 613 /* 614 * Try to place the inode is the same flex group as its 615 * parent. If we can't find space, use the Orlov algorithm to 616 * find another flex group, and store that information in the 617 * parent directory's inode information so that use that flex 618 * group for future allocations. 619 */ 620 if (flex_size > 1) { 621 int retry = 0; 622 623 try_again: 624 parent_group &= ~(flex_size-1); 625 last = parent_group + flex_size; 626 if (last > ngroups) 627 last = ngroups; 628 for (i = parent_group; i < last; i++) { 629 desc = ext4_get_group_desc(sb, i, NULL); 630 if (desc && ext4_free_inodes_count(sb, desc)) { 631 *group = i; 632 return 0; 633 } 634 } 635 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { 636 retry = 1; 637 parent_group = EXT4_I(parent)->i_last_alloc_group; 638 goto try_again; 639 } 640 /* 641 * If this didn't work, use the Orlov search algorithm 642 * to find a new flex group; we pass in the mode to 643 * avoid the topdir algorithms. 644 */ 645 *group = parent_group + flex_size; 646 if (*group > ngroups) 647 *group = 0; 648 return find_group_orlov(sb, parent, group, mode, 0); 649 } 650 651 /* 652 * Try to place the inode in its parent directory 653 */ 654 *group = parent_group; 655 desc = ext4_get_group_desc(sb, *group, NULL); 656 if (desc && ext4_free_inodes_count(sb, desc) && 657 ext4_free_blks_count(sb, desc)) 658 return 0; 659 660 /* 661 * We're going to place this inode in a different blockgroup from its 662 * parent. We want to cause files in a common directory to all land in 663 * the same blockgroup. But we want files which are in a different 664 * directory which shares a blockgroup with our parent to land in a 665 * different blockgroup. 666 * 667 * So add our directory's i_ino into the starting point for the hash. 668 */ 669 *group = (*group + parent->i_ino) % ngroups; 670 671 /* 672 * Use a quadratic hash to find a group with a free inode and some free 673 * blocks. 674 */ 675 for (i = 1; i < ngroups; i <<= 1) { 676 *group += i; 677 if (*group >= ngroups) 678 *group -= ngroups; 679 desc = ext4_get_group_desc(sb, *group, NULL); 680 if (desc && ext4_free_inodes_count(sb, desc) && 681 ext4_free_blks_count(sb, desc)) 682 return 0; 683 } 684 685 /* 686 * That failed: try linear search for a free inode, even if that group 687 * has no free blocks. 688 */ 689 *group = parent_group; 690 for (i = 0; i < ngroups; i++) { 691 if (++*group >= ngroups) 692 *group = 0; 693 desc = ext4_get_group_desc(sb, *group, NULL); 694 if (desc && ext4_free_inodes_count(sb, desc)) 695 return 0; 696 } 697 698 return -1; 699 } 700 701 /* 702 * claim the inode from the inode bitmap. If the group 703 * is uninit we need to take the groups's ext4_group_lock 704 * and clear the uninit flag. The inode bitmap update 705 * and group desc uninit flag clear should be done 706 * after holding ext4_group_lock so that ext4_read_inode_bitmap 707 * doesn't race with the ext4_claim_inode 708 */ 709 static int ext4_claim_inode(struct super_block *sb, 710 struct buffer_head *inode_bitmap_bh, 711 unsigned long ino, ext4_group_t group, int mode) 712 { 713 int free = 0, retval = 0, count; 714 struct ext4_sb_info *sbi = EXT4_SB(sb); 715 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); 716 717 ext4_lock_group(sb, group); 718 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { 719 /* not a free inode */ 720 retval = 1; 721 goto err_ret; 722 } 723 ino++; 724 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || 725 ino > EXT4_INODES_PER_GROUP(sb)) { 726 ext4_unlock_group(sb, group); 727 ext4_error(sb, "reserved inode or inode > inodes count - " 728 "block_group = %u, inode=%lu", group, 729 ino + group * EXT4_INODES_PER_GROUP(sb)); 730 return 1; 731 } 732 /* If we didn't allocate from within the initialized part of the inode 733 * table then we need to initialize up to this inode. */ 734 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { 735 736 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { 737 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); 738 /* When marking the block group with 739 * ~EXT4_BG_INODE_UNINIT we don't want to depend 740 * on the value of bg_itable_unused even though 741 * mke2fs could have initialized the same for us. 742 * Instead we calculated the value below 743 */ 744 745 free = 0; 746 } else { 747 free = EXT4_INODES_PER_GROUP(sb) - 748 ext4_itable_unused_count(sb, gdp); 749 } 750 751 /* 752 * Check the relative inode number against the last used 753 * relative inode number in this group. if it is greater 754 * we need to update the bg_itable_unused count 755 * 756 */ 757 if (ino > free) 758 ext4_itable_unused_set(sb, gdp, 759 (EXT4_INODES_PER_GROUP(sb) - ino)); 760 } 761 count = ext4_free_inodes_count(sb, gdp) - 1; 762 ext4_free_inodes_set(sb, gdp, count); 763 if (S_ISDIR(mode)) { 764 count = ext4_used_dirs_count(sb, gdp) + 1; 765 ext4_used_dirs_set(sb, gdp, count); 766 if (sbi->s_log_groups_per_flex) { 767 ext4_group_t f = ext4_flex_group(sbi, group); 768 769 atomic_inc(&sbi->s_flex_groups[f].used_dirs); 770 } 771 } 772 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); 773 err_ret: 774 ext4_unlock_group(sb, group); 775 return retval; 776 } 777 778 /* 779 * There are two policies for allocating an inode. If the new inode is 780 * a directory, then a forward search is made for a block group with both 781 * free space and a low directory-to-inode ratio; if that fails, then of 782 * the groups with above-average free space, that group with the fewest 783 * directories already is chosen. 784 * 785 * For other inodes, search forward from the parent directory's block 786 * group to find a free inode. 787 */ 788 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode, 789 const struct qstr *qstr, __u32 goal) 790 { 791 struct super_block *sb; 792 struct buffer_head *inode_bitmap_bh = NULL; 793 struct buffer_head *group_desc_bh; 794 ext4_group_t ngroups, group = 0; 795 unsigned long ino = 0; 796 struct inode *inode; 797 struct ext4_group_desc *gdp = NULL; 798 struct ext4_inode_info *ei; 799 struct ext4_sb_info *sbi; 800 int ret2, err = 0; 801 struct inode *ret; 802 ext4_group_t i; 803 int free = 0; 804 static int once = 1; 805 ext4_group_t flex_group; 806 807 /* Cannot create files in a deleted directory */ 808 if (!dir || !dir->i_nlink) 809 return ERR_PTR(-EPERM); 810 811 sb = dir->i_sb; 812 ngroups = ext4_get_groups_count(sb); 813 trace_ext4_request_inode(dir, mode); 814 inode = new_inode(sb); 815 if (!inode) 816 return ERR_PTR(-ENOMEM); 817 ei = EXT4_I(inode); 818 sbi = EXT4_SB(sb); 819 820 if (!goal) 821 goal = sbi->s_inode_goal; 822 823 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { 824 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); 825 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); 826 ret2 = 0; 827 goto got_group; 828 } 829 830 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) { 831 ret2 = find_group_flex(sb, dir, &group); 832 if (ret2 == -1) { 833 ret2 = find_group_other(sb, dir, &group, mode); 834 if (ret2 == 0 && once) { 835 once = 0; 836 printk(KERN_NOTICE "ext4: find_group_flex " 837 "failed, fallback succeeded dir %lu\n", 838 dir->i_ino); 839 } 840 } 841 goto got_group; 842 } 843 844 if (S_ISDIR(mode)) { 845 if (test_opt(sb, OLDALLOC)) 846 ret2 = find_group_dir(sb, dir, &group); 847 else 848 ret2 = find_group_orlov(sb, dir, &group, mode, qstr); 849 } else 850 ret2 = find_group_other(sb, dir, &group, mode); 851 852 got_group: 853 EXT4_I(dir)->i_last_alloc_group = group; 854 err = -ENOSPC; 855 if (ret2 == -1) 856 goto out; 857 858 for (i = 0; i < ngroups; i++, ino = 0) { 859 err = -EIO; 860 861 gdp = ext4_get_group_desc(sb, group, &group_desc_bh); 862 if (!gdp) 863 goto fail; 864 865 brelse(inode_bitmap_bh); 866 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); 867 if (!inode_bitmap_bh) 868 goto fail; 869 870 repeat_in_this_group: 871 ino = ext4_find_next_zero_bit((unsigned long *) 872 inode_bitmap_bh->b_data, 873 EXT4_INODES_PER_GROUP(sb), ino); 874 875 if (ino < EXT4_INODES_PER_GROUP(sb)) { 876 877 BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); 878 err = ext4_journal_get_write_access(handle, 879 inode_bitmap_bh); 880 if (err) 881 goto fail; 882 883 BUFFER_TRACE(group_desc_bh, "get_write_access"); 884 err = ext4_journal_get_write_access(handle, 885 group_desc_bh); 886 if (err) 887 goto fail; 888 if (!ext4_claim_inode(sb, inode_bitmap_bh, 889 ino, group, mode)) { 890 /* we won it */ 891 BUFFER_TRACE(inode_bitmap_bh, 892 "call ext4_handle_dirty_metadata"); 893 err = ext4_handle_dirty_metadata(handle, 894 NULL, 895 inode_bitmap_bh); 896 if (err) 897 goto fail; 898 /* zero bit is inode number 1*/ 899 ino++; 900 goto got; 901 } 902 /* we lost it */ 903 ext4_handle_release_buffer(handle, inode_bitmap_bh); 904 ext4_handle_release_buffer(handle, group_desc_bh); 905 906 if (++ino < EXT4_INODES_PER_GROUP(sb)) 907 goto repeat_in_this_group; 908 } 909 910 /* 911 * This case is possible in concurrent environment. It is very 912 * rare. We cannot repeat the find_group_xxx() call because 913 * that will simply return the same blockgroup, because the 914 * group descriptor metadata has not yet been updated. 915 * So we just go onto the next blockgroup. 916 */ 917 if (++group == ngroups) 918 group = 0; 919 } 920 err = -ENOSPC; 921 goto out; 922 923 got: 924 /* We may have to initialize the block bitmap if it isn't already */ 925 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && 926 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 927 struct buffer_head *block_bitmap_bh; 928 929 block_bitmap_bh = ext4_read_block_bitmap(sb, group); 930 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); 931 err = ext4_journal_get_write_access(handle, block_bitmap_bh); 932 if (err) { 933 brelse(block_bitmap_bh); 934 goto fail; 935 } 936 937 free = 0; 938 ext4_lock_group(sb, group); 939 /* recheck and clear flag under lock if we still need to */ 940 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 941 free = ext4_free_blocks_after_init(sb, group, gdp); 942 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); 943 ext4_free_blks_set(sb, gdp, free); 944 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, 945 gdp); 946 } 947 ext4_unlock_group(sb, group); 948 949 /* Don't need to dirty bitmap block if we didn't change it */ 950 if (free) { 951 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); 952 err = ext4_handle_dirty_metadata(handle, 953 NULL, block_bitmap_bh); 954 } 955 956 brelse(block_bitmap_bh); 957 if (err) 958 goto fail; 959 } 960 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); 961 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); 962 if (err) 963 goto fail; 964 965 percpu_counter_dec(&sbi->s_freeinodes_counter); 966 if (S_ISDIR(mode)) 967 percpu_counter_inc(&sbi->s_dirs_counter); 968 ext4_mark_super_dirty(sb); 969 970 if (sbi->s_log_groups_per_flex) { 971 flex_group = ext4_flex_group(sbi, group); 972 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); 973 } 974 975 if (test_opt(sb, GRPID)) { 976 inode->i_mode = mode; 977 inode->i_uid = current_fsuid(); 978 inode->i_gid = dir->i_gid; 979 } else 980 inode_init_owner(inode, dir, mode); 981 982 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); 983 /* This is the optimal IO size (for stat), not the fs block size */ 984 inode->i_blocks = 0; 985 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = 986 ext4_current_time(inode); 987 988 memset(ei->i_data, 0, sizeof(ei->i_data)); 989 ei->i_dir_start_lookup = 0; 990 ei->i_disksize = 0; 991 992 /* 993 * Don't inherit extent flag from directory, amongst others. We set 994 * extent flag on newly created directory and file only if -o extent 995 * mount option is specified 996 */ 997 ei->i_flags = 998 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); 999 ei->i_file_acl = 0; 1000 ei->i_dtime = 0; 1001 ei->i_block_group = group; 1002 ei->i_last_alloc_group = ~0; 1003 1004 ext4_set_inode_flags(inode); 1005 if (IS_DIRSYNC(inode)) 1006 ext4_handle_sync(handle); 1007 if (insert_inode_locked(inode) < 0) { 1008 err = -EINVAL; 1009 goto fail_drop; 1010 } 1011 spin_lock(&sbi->s_next_gen_lock); 1012 inode->i_generation = sbi->s_next_generation++; 1013 spin_unlock(&sbi->s_next_gen_lock); 1014 1015 ei->i_state_flags = 0; 1016 ext4_set_inode_state(inode, EXT4_STATE_NEW); 1017 1018 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; 1019 1020 ret = inode; 1021 dquot_initialize(inode); 1022 err = dquot_alloc_inode(inode); 1023 if (err) 1024 goto fail_drop; 1025 1026 err = ext4_init_acl(handle, inode, dir); 1027 if (err) 1028 goto fail_free_drop; 1029 1030 err = ext4_init_security(handle, inode, dir); 1031 if (err) 1032 goto fail_free_drop; 1033 1034 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { 1035 /* set extent flag only for directory, file and normal symlink*/ 1036 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { 1037 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); 1038 ext4_ext_tree_init(handle, inode); 1039 } 1040 } 1041 1042 err = ext4_mark_inode_dirty(handle, inode); 1043 if (err) { 1044 ext4_std_error(sb, err); 1045 goto fail_free_drop; 1046 } 1047 1048 ext4_debug("allocating inode %lu\n", inode->i_ino); 1049 trace_ext4_allocate_inode(inode, dir, mode); 1050 goto really_out; 1051 fail: 1052 ext4_std_error(sb, err); 1053 out: 1054 iput(inode); 1055 ret = ERR_PTR(err); 1056 really_out: 1057 brelse(inode_bitmap_bh); 1058 return ret; 1059 1060 fail_free_drop: 1061 dquot_free_inode(inode); 1062 1063 fail_drop: 1064 dquot_drop(inode); 1065 inode->i_flags |= S_NOQUOTA; 1066 inode->i_nlink = 0; 1067 unlock_new_inode(inode); 1068 iput(inode); 1069 brelse(inode_bitmap_bh); 1070 return ERR_PTR(err); 1071 } 1072 1073 /* Verify that we are loading a valid orphan from disk */ 1074 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) 1075 { 1076 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); 1077 ext4_group_t block_group; 1078 int bit; 1079 struct buffer_head *bitmap_bh; 1080 struct inode *inode = NULL; 1081 long err = -EIO; 1082 1083 /* Error cases - e2fsck has already cleaned up for us */ 1084 if (ino > max_ino) { 1085 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); 1086 goto error; 1087 } 1088 1089 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 1090 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 1091 bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 1092 if (!bitmap_bh) { 1093 ext4_warning(sb, "inode bitmap error for orphan %lu", ino); 1094 goto error; 1095 } 1096 1097 /* Having the inode bit set should be a 100% indicator that this 1098 * is a valid orphan (no e2fsck run on fs). Orphans also include 1099 * inodes that were being truncated, so we can't check i_nlink==0. 1100 */ 1101 if (!ext4_test_bit(bit, bitmap_bh->b_data)) 1102 goto bad_orphan; 1103 1104 inode = ext4_iget(sb, ino); 1105 if (IS_ERR(inode)) 1106 goto iget_failed; 1107 1108 /* 1109 * If the orphans has i_nlinks > 0 then it should be able to be 1110 * truncated, otherwise it won't be removed from the orphan list 1111 * during processing and an infinite loop will result. 1112 */ 1113 if (inode->i_nlink && !ext4_can_truncate(inode)) 1114 goto bad_orphan; 1115 1116 if (NEXT_ORPHAN(inode) > max_ino) 1117 goto bad_orphan; 1118 brelse(bitmap_bh); 1119 return inode; 1120 1121 iget_failed: 1122 err = PTR_ERR(inode); 1123 inode = NULL; 1124 bad_orphan: 1125 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); 1126 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", 1127 bit, (unsigned long long)bitmap_bh->b_blocknr, 1128 ext4_test_bit(bit, bitmap_bh->b_data)); 1129 printk(KERN_NOTICE "inode=%p\n", inode); 1130 if (inode) { 1131 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", 1132 is_bad_inode(inode)); 1133 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", 1134 NEXT_ORPHAN(inode)); 1135 printk(KERN_NOTICE "max_ino=%lu\n", max_ino); 1136 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); 1137 /* Avoid freeing blocks if we got a bad deleted inode */ 1138 if (inode->i_nlink == 0) 1139 inode->i_blocks = 0; 1140 iput(inode); 1141 } 1142 brelse(bitmap_bh); 1143 error: 1144 return ERR_PTR(err); 1145 } 1146 1147 unsigned long ext4_count_free_inodes(struct super_block *sb) 1148 { 1149 unsigned long desc_count; 1150 struct ext4_group_desc *gdp; 1151 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1152 #ifdef EXT4FS_DEBUG 1153 struct ext4_super_block *es; 1154 unsigned long bitmap_count, x; 1155 struct buffer_head *bitmap_bh = NULL; 1156 1157 es = EXT4_SB(sb)->s_es; 1158 desc_count = 0; 1159 bitmap_count = 0; 1160 gdp = NULL; 1161 for (i = 0; i < ngroups; i++) { 1162 gdp = ext4_get_group_desc(sb, i, NULL); 1163 if (!gdp) 1164 continue; 1165 desc_count += ext4_free_inodes_count(sb, gdp); 1166 brelse(bitmap_bh); 1167 bitmap_bh = ext4_read_inode_bitmap(sb, i); 1168 if (!bitmap_bh) 1169 continue; 1170 1171 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); 1172 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", 1173 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); 1174 bitmap_count += x; 1175 } 1176 brelse(bitmap_bh); 1177 printk(KERN_DEBUG "ext4_count_free_inodes: " 1178 "stored = %u, computed = %lu, %lu\n", 1179 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); 1180 return desc_count; 1181 #else 1182 desc_count = 0; 1183 for (i = 0; i < ngroups; i++) { 1184 gdp = ext4_get_group_desc(sb, i, NULL); 1185 if (!gdp) 1186 continue; 1187 desc_count += ext4_free_inodes_count(sb, gdp); 1188 cond_resched(); 1189 } 1190 return desc_count; 1191 #endif 1192 } 1193 1194 /* Called at mount-time, super-block is locked */ 1195 unsigned long ext4_count_dirs(struct super_block * sb) 1196 { 1197 unsigned long count = 0; 1198 ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1199 1200 for (i = 0; i < ngroups; i++) { 1201 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 1202 if (!gdp) 1203 continue; 1204 count += ext4_used_dirs_count(sb, gdp); 1205 } 1206 return count; 1207 } 1208