1/* 2 * linux/fs/ext3/balloc.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993 10 * Big-endian to little-endian byte-swapping/bitmaps by 11 * David S. Miller (davem@caip.rutgers.edu), 1995 12 */ 13 14#include <linux/config.h> 15#include <linux/sched.h> 16#include <linux/fs.h> 17#include <linux/jbd.h> 18#include <linux/ext3_fs.h> 19#include <linux/ext3_jbd.h> 20#include <linux/locks.h> 21#include <linux/quotaops.h> 22 23/* 24 * balloc.c contains the blocks allocation and deallocation routines 25 */ 26 27/* 28 * The free blocks are managed by bitmaps. A file system contains several 29 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 30 * block for inodes, N blocks for the inode table and data blocks. 31 * 32 * The file system contains group descriptors which are located after the 33 * super block. Each descriptor contains the number of the bitmap block and 34 * the free blocks count in the block. The descriptors are loaded in memory 35 * when a file system is mounted (see ext3_read_super). 36 */ 37 38 39#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) 40 41struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb, 42 unsigned int block_group, 43 struct buffer_head ** bh) 44{ 45 unsigned long group_desc; 46 unsigned long desc; 47 struct ext3_group_desc * gdp; 48 49 if (block_group >= sb->u.ext3_sb.s_groups_count) { 50 ext3_error (sb, "ext3_get_group_desc", 51 "block_group >= groups_count - " 52 "block_group = %d, groups_count = %lu", 53 block_group, sb->u.ext3_sb.s_groups_count); 54 55 return NULL; 56 } 57 58 group_desc = block_group / EXT3_DESC_PER_BLOCK(sb); 59 desc = block_group % EXT3_DESC_PER_BLOCK(sb); 60 if (!sb->u.ext3_sb.s_group_desc[group_desc]) { 61 ext3_error (sb, "ext3_get_group_desc", 62 "Group descriptor not loaded - " 63 "block_group = %d, group_desc = %lu, desc = %lu", 64 block_group, group_desc, desc); 65 return NULL; 66 } 67 68 gdp = (struct ext3_group_desc *) 69 sb->u.ext3_sb.s_group_desc[group_desc]->b_data; 70 if (bh) 71 *bh = sb->u.ext3_sb.s_group_desc[group_desc]; 72 return gdp + desc; 73} 74 75/* 76 * Read the bitmap for a given block_group, reading into the specified 77 * slot in the superblock's bitmap cache. 78 * 79 * Return >=0 on success or a -ve error code. 80 */ 81 82static int read_block_bitmap (struct super_block * sb, 83 unsigned int block_group, 84 unsigned long bitmap_nr) 85{ 86 struct ext3_group_desc * gdp; 87 struct buffer_head * bh = NULL; 88 int retval = -EIO; 89 90 gdp = ext3_get_group_desc (sb, block_group, NULL); 91 if (!gdp) 92 goto error_out; 93 retval = 0; 94 bh = sb_bread(sb, le32_to_cpu(gdp->bg_block_bitmap)); 95 if (!bh) { 96 ext3_error (sb, "read_block_bitmap", 97 "Cannot read block bitmap - " 98 "block_group = %d, block_bitmap = %lu", 99 block_group, (unsigned long) gdp->bg_block_bitmap); 100 retval = -EIO; 101 } 102 /* 103 * On IO error, just leave a zero in the superblock's block pointer for 104 * this group. The IO will be retried next time. 105 */ 106error_out: 107 sb->u.ext3_sb.s_block_bitmap_number[bitmap_nr] = block_group; 108 sb->u.ext3_sb.s_block_bitmap[bitmap_nr] = bh; 109 return retval; 110} 111 112/* 113 * load_block_bitmap loads the block bitmap for a blocks group 114 * 115 * It maintains a cache for the last bitmaps loaded. This cache is managed 116 * with a LRU algorithm. 117 * 118 * Notes: 119 * 1/ There is one cache per mounted file system. 120 * 2/ If the file system contains less than EXT3_MAX_GROUP_LOADED groups, 121 * this function reads the bitmap without maintaining a LRU cache. 122 * 123 * Return the slot used to store the bitmap, or a -ve error code. 124 */ 125static int __load_block_bitmap (struct super_block * sb, 126 unsigned int block_group) 127{ 128 int i, j, retval = 0; 129 unsigned long block_bitmap_number; 130 struct buffer_head * block_bitmap; 131 132 if (block_group >= sb->u.ext3_sb.s_groups_count) 133 ext3_panic (sb, "load_block_bitmap", 134 "block_group >= groups_count - " 135 "block_group = %d, groups_count = %lu", 136 block_group, sb->u.ext3_sb.s_groups_count); 137 138 if (sb->u.ext3_sb.s_groups_count <= EXT3_MAX_GROUP_LOADED) { 139 if (sb->u.ext3_sb.s_block_bitmap[block_group]) { 140 if (sb->u.ext3_sb.s_block_bitmap_number[block_group] == 141 block_group) 142 return block_group; 143 ext3_error (sb, "__load_block_bitmap", 144 "block_group != block_bitmap_number"); 145 } 146 retval = read_block_bitmap (sb, block_group, block_group); 147 if (retval < 0) 148 return retval; 149 return block_group; 150 } 151 152 for (i = 0; i < sb->u.ext3_sb.s_loaded_block_bitmaps && 153 sb->u.ext3_sb.s_block_bitmap_number[i] != block_group; i++) 154 ; 155 if (i < sb->u.ext3_sb.s_loaded_block_bitmaps && 156 sb->u.ext3_sb.s_block_bitmap_number[i] == block_group) { 157 block_bitmap_number = sb->u.ext3_sb.s_block_bitmap_number[i]; 158 block_bitmap = sb->u.ext3_sb.s_block_bitmap[i]; 159 for (j = i; j > 0; j--) { 160 sb->u.ext3_sb.s_block_bitmap_number[j] = 161 sb->u.ext3_sb.s_block_bitmap_number[j - 1]; 162 sb->u.ext3_sb.s_block_bitmap[j] = 163 sb->u.ext3_sb.s_block_bitmap[j - 1]; 164 } 165 sb->u.ext3_sb.s_block_bitmap_number[0] = block_bitmap_number; 166 sb->u.ext3_sb.s_block_bitmap[0] = block_bitmap; 167 168 /* 169 * There's still one special case here --- if block_bitmap == 0 170 * then our last attempt to read the bitmap failed and we have 171 * just ended up caching that failure. Try again to read it. 172 */ 173 if (!block_bitmap) 174 retval = read_block_bitmap (sb, block_group, 0); 175 } else { 176 if (sb->u.ext3_sb.s_loaded_block_bitmaps<EXT3_MAX_GROUP_LOADED) 177 sb->u.ext3_sb.s_loaded_block_bitmaps++; 178 else 179 brelse (sb->u.ext3_sb.s_block_bitmap 180 [EXT3_MAX_GROUP_LOADED - 1]); 181 for (j = sb->u.ext3_sb.s_loaded_block_bitmaps - 1; 182 j > 0; j--) { 183 sb->u.ext3_sb.s_block_bitmap_number[j] = 184 sb->u.ext3_sb.s_block_bitmap_number[j - 1]; 185 sb->u.ext3_sb.s_block_bitmap[j] = 186 sb->u.ext3_sb.s_block_bitmap[j - 1]; 187 } 188 retval = read_block_bitmap (sb, block_group, 0); 189 } 190 return retval; 191} 192 193/* 194 * Load the block bitmap for a given block group. First of all do a couple 195 * of fast lookups for common cases and then pass the request onto the guts 196 * of the bitmap loader. 197 * 198 * Return the slot number of the group in the superblock bitmap cache's on 199 * success, or a -ve error code. 200 * 201 * There is still one inconsistency here --- if the number of groups in this 202 * filesystems is <= EXT3_MAX_GROUP_LOADED, then we have no way of 203 * differentiating between a group for which we have never performed a bitmap 204 * IO request, and a group for which the last bitmap read request failed. 205 */ 206static inline int load_block_bitmap (struct super_block * sb, 207 unsigned int block_group) 208{ 209 int slot; 210 211 /* 212 * Do the lookup for the slot. First of all, check if we're asking 213 * for the same slot as last time, and did we succeed that last time? 214 */ 215 if (sb->u.ext3_sb.s_loaded_block_bitmaps > 0 && 216 sb->u.ext3_sb.s_block_bitmap_number[0] == block_group && 217 sb->u.ext3_sb.s_block_bitmap[0]) { 218 return 0; 219 } 220 /* 221 * Or can we do a fast lookup based on a loaded group on a filesystem 222 * small enough to be mapped directly into the superblock? 223 */ 224 else if (sb->u.ext3_sb.s_groups_count <= EXT3_MAX_GROUP_LOADED && 225 sb->u.ext3_sb.s_block_bitmap_number[block_group]==block_group 226 && sb->u.ext3_sb.s_block_bitmap[block_group]) { 227 slot = block_group; 228 } 229 /* 230 * If not, then do a full lookup for this block group. 231 */ 232 else { 233 slot = __load_block_bitmap (sb, block_group); 234 } 235 236 /* 237 * <0 means we just got an error 238 */ 239 if (slot < 0) 240 return slot; 241 242 /* 243 * If it's a valid slot, we may still have cached a previous IO error, 244 * in which case the bh in the superblock cache will be zero. 245 */ 246 if (!sb->u.ext3_sb.s_block_bitmap[slot]) 247 return -EIO; 248 249 /* 250 * Must have been read in OK to get this far. 251 */ 252 return slot; 253} 254 255/* Free given blocks, update quota and i_blocks field */ 256void ext3_free_blocks (handle_t *handle, struct inode * inode, 257 unsigned long block, unsigned long count) 258{ 259 struct buffer_head *bitmap_bh; 260 struct buffer_head *gd_bh; 261 unsigned long block_group; 262 unsigned long bit; 263 unsigned long i; 264 int bitmap_nr; 265 unsigned long overflow; 266 struct super_block * sb; 267 struct ext3_group_desc * gdp; 268 struct ext3_super_block * es; 269 int err = 0, ret; 270 int dquot_freed_blocks = 0; 271 272 sb = inode->i_sb; 273 if (!sb) { 274 printk ("ext3_free_blocks: nonexistent device"); 275 return; 276 } 277 lock_super (sb); 278 es = sb->u.ext3_sb.s_es; 279 if (block < le32_to_cpu(es->s_first_data_block) || 280 (block + count) > le32_to_cpu(es->s_blocks_count)) { 281 ext3_error (sb, "ext3_free_blocks", 282 "Freeing blocks not in datazone - " 283 "block = %lu, count = %lu", block, count); 284 goto error_return; 285 } 286 287 ext3_debug ("freeing block %lu\n", block); 288 289do_more: 290 overflow = 0; 291 block_group = (block - le32_to_cpu(es->s_first_data_block)) / 292 EXT3_BLOCKS_PER_GROUP(sb); 293 bit = (block - le32_to_cpu(es->s_first_data_block)) % 294 EXT3_BLOCKS_PER_GROUP(sb); 295 /* 296 * Check to see if we are freeing blocks across a group 297 * boundary. 298 */ 299 if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) { 300 overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb); 301 count -= overflow; 302 } 303 bitmap_nr = load_block_bitmap (sb, block_group); 304 if (bitmap_nr < 0) 305 goto error_return; 306 307 bitmap_bh = sb->u.ext3_sb.s_block_bitmap[bitmap_nr]; 308 gdp = ext3_get_group_desc (sb, block_group, &gd_bh); 309 if (!gdp) 310 goto error_return; 311 312 if (in_range (le32_to_cpu(gdp->bg_block_bitmap), block, count) || 313 in_range (le32_to_cpu(gdp->bg_inode_bitmap), block, count) || 314 in_range (block, le32_to_cpu(gdp->bg_inode_table), 315 sb->u.ext3_sb.s_itb_per_group) || 316 in_range (block + count - 1, le32_to_cpu(gdp->bg_inode_table), 317 sb->u.ext3_sb.s_itb_per_group)) 318 ext3_error (sb, "ext3_free_blocks", 319 "Freeing blocks in system zones - " 320 "Block = %lu, count = %lu", 321 block, count); 322 323 /* 324 * We are about to start releasing blocks in the bitmap, 325 * so we need undo access. 326 */ 327 /* @@@ check errors */ 328 BUFFER_TRACE(bitmap_bh, "getting undo access"); 329 err = ext3_journal_get_undo_access(handle, bitmap_bh); 330 if (err) 331 goto error_return; 332 333 /* 334 * We are about to modify some metadata. Call the journal APIs 335 * to unshare ->b_data if a currently-committing transaction is 336 * using it 337 */ 338 BUFFER_TRACE(gd_bh, "get_write_access"); 339 err = ext3_journal_get_write_access(handle, gd_bh); 340 if (err) 341 goto error_return; 342 343 BUFFER_TRACE(sb->u.ext3_sb.s_sbh, "get_write_access"); 344 err = ext3_journal_get_write_access(handle, sb->u.ext3_sb.s_sbh); 345 if (err) 346 goto error_return; 347 348 for (i = 0; i < count; i++) { 349 /* 350 * An HJ special. This is expensive... 351 */ 352#ifdef CONFIG_JBD_DEBUG 353 { 354 struct buffer_head *debug_bh; 355 debug_bh = sb_get_hash_table(sb, block + i); 356 if (debug_bh) { 357 BUFFER_TRACE(debug_bh, "Deleted!"); 358 if (!bh2jh(bitmap_bh)->b_committed_data) 359 BUFFER_TRACE(debug_bh, 360 "No commited data in bitmap"); 361 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap"); 362 __brelse(debug_bh); 363 } 364 } 365#endif 366 BUFFER_TRACE(bitmap_bh, "clear bit"); 367 if (!ext3_clear_bit (bit + i, bitmap_bh->b_data)) { 368 ext3_error (sb, __FUNCTION__, 369 "bit already cleared for block %lu", 370 block + i); 371 BUFFER_TRACE(bitmap_bh, "bit already cleared"); 372 } else { 373 dquot_freed_blocks++; 374 gdp->bg_free_blocks_count = 375 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count)+1); 376 es->s_free_blocks_count = 377 cpu_to_le32(le32_to_cpu(es->s_free_blocks_count)+1); 378 } 379 /* @@@ This prevents newly-allocated data from being 380 * freed and then reallocated within the same 381 * transaction. 382 * 383 * Ideally we would want to allow that to happen, but to 384 * do so requires making journal_forget() capable of 385 * revoking the queued write of a data block, which 386 * implies blocking on the journal lock. *forget() 387 * cannot block due to truncate races. 388 * 389 * Eventually we can fix this by making journal_forget() 390 * return a status indicating whether or not it was able 391 * to revoke the buffer. On successful revoke, it is 392 * safe not to set the allocation bit in the committed 393 * bitmap, because we know that there is no outstanding 394 * activity on the buffer any more and so it is safe to 395 * reallocate it. 396 */ 397 BUFFER_TRACE(bitmap_bh, "clear in b_committed_data"); 398 J_ASSERT_BH(bitmap_bh, 399 bh2jh(bitmap_bh)->b_committed_data != NULL); 400 ext3_set_bit(bit + i, bh2jh(bitmap_bh)->b_committed_data); 401 } 402 403 /* We dirtied the bitmap block */ 404 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); 405 err = ext3_journal_dirty_metadata(handle, bitmap_bh); 406 407 /* And the group descriptor block */ 408 BUFFER_TRACE(gd_bh, "dirtied group descriptor block"); 409 ret = ext3_journal_dirty_metadata(handle, gd_bh); 410 if (!err) err = ret; 411 412 /* And the superblock */ 413 BUFFER_TRACE(sb->u.ext3_sb.s_sbh, "dirtied superblock"); 414 ret = ext3_journal_dirty_metadata(handle, sb->u.ext3_sb.s_sbh); 415 if (!err) err = ret; 416 417 if (overflow && !err) { 418 block += count; 419 count = overflow; 420 goto do_more; 421 } 422 sb->s_dirt = 1; 423error_return: 424 ext3_std_error(sb, err); 425 unlock_super(sb); 426 if (dquot_freed_blocks) 427 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks); 428 return; 429} 430 431/* For ext3 allocations, we must not reuse any blocks which are 432 * allocated in the bitmap buffer's "last committed data" copy. This 433 * prevents deletes from freeing up the page for reuse until we have 434 * committed the delete transaction. 435 * 436 * If we didn't do this, then deleting something and reallocating it as 437 * data would allow the old block to be overwritten before the 438 * transaction committed (because we force data to disk before commit). 439 * This would lead to corruption if we crashed between overwriting the 440 * data and committing the delete. 441 * 442 * @@@ We may want to make this allocation behaviour conditional on 443 * data-writes at some point, and disable it for metadata allocations or 444 * sync-data inodes. 445 */ 446static int ext3_test_allocatable(int nr, struct buffer_head *bh) 447{ 448 if (ext3_test_bit(nr, bh->b_data)) 449 return 0; 450 if (!buffer_jbd(bh) || !bh2jh(bh)->b_committed_data) 451 return 1; 452 return !ext3_test_bit(nr, bh2jh(bh)->b_committed_data); 453} 454 455/* 456 * Find an allocatable block in a bitmap. We honour both the bitmap and 457 * its last-committed copy (if that exists), and perform the "most 458 * appropriate allocation" algorithm of looking for a free block near 459 * the initial goal; then for a free byte somewhere in the bitmap; then 460 * for any free bit in the bitmap. 461 */ 462static int find_next_usable_block(int start, 463 struct buffer_head *bh, int maxblocks) 464{ 465 int here, next; 466 char *p, *r; 467 468 if (start > 0) { 469 /* 470 * The goal was occupied; search forward for a free 471 * block within the next XX blocks. 472 * 473 * end_goal is more or less random, but it has to be 474 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the 475 * next 64-bit boundary is simple.. 476 */ 477 int end_goal = (start + 63) & ~63; 478 here = ext3_find_next_zero_bit(bh->b_data, end_goal, start); 479 if (here < end_goal && ext3_test_allocatable(here, bh)) 480 return here; 481 482 ext3_debug ("Bit not found near goal\n"); 483 484 } 485 486 here = start; 487 if (here < 0) 488 here = 0; 489 490 /* 491 * There has been no free block found in the near vicinity of 492 * the goal: do a search forward through the block groups, 493 * searching in each group first for an entire free byte in the 494 * bitmap and then for any free bit. 495 * 496 * Search first in the remainder of the current group 497 */ 498 p = ((char *) bh->b_data) + (here >> 3); 499 r = memscan(p, 0, (maxblocks - here + 7) >> 3); 500 next = (r - ((char *) bh->b_data)) << 3; 501 502 if (next < maxblocks && ext3_test_allocatable(next, bh)) 503 return next; 504 505 /* The bitmap search --- search forward alternately 506 * through the actual bitmap and the last-committed copy 507 * until we find a bit free in both. */ 508 509 while (here < maxblocks) { 510 next = ext3_find_next_zero_bit ((unsigned long *) bh->b_data, 511 maxblocks, here); 512 if (next >= maxblocks) 513 return -1; 514 if (ext3_test_allocatable(next, bh)) 515 return next; 516 517 J_ASSERT_BH(bh, bh2jh(bh)->b_committed_data); 518 here = ext3_find_next_zero_bit 519 ((unsigned long *) bh2jh(bh)->b_committed_data, 520 maxblocks, next); 521 } 522 return -1; 523} 524 525/* 526 * ext3_new_block uses a goal block to assist allocation. If the goal is 527 * free, or there is a free block within 32 blocks of the goal, that block 528 * is allocated. Otherwise a forward search is made for a free block; within 529 * each block group the search first looks for an entire free byte in the block 530 * bitmap, and then for any free bit if that fails. 531 * This function also updates quota and i_blocks field. 532 */ 533int ext3_new_block (handle_t *handle, struct inode * inode, 534 unsigned long goal, u32 * prealloc_count, 535 u32 * prealloc_block, int * errp) 536{ 537 struct buffer_head * bh, *bhtmp; 538 struct buffer_head * bh2; 539 int i, j, k, tmp, alloctmp; 540 int bitmap_nr; 541 int fatal = 0, err; 542 int performed_allocation = 0; 543 struct super_block * sb; 544 struct ext3_group_desc * gdp; 545 struct ext3_super_block * es; 546#ifdef EXT3FS_DEBUG 547 static int goal_hits = 0, goal_attempts = 0; 548#endif 549 *errp = -ENOSPC; 550 sb = inode->i_sb; 551 if (!sb) { 552 printk ("ext3_new_block: nonexistent device"); 553 return 0; 554 } 555 556 /* 557 * Check quota for allocation of this block. 558 */ 559 if (DQUOT_ALLOC_BLOCK(inode, 1)) { 560 *errp = -EDQUOT; 561 return 0; 562 } 563 564 lock_super (sb); 565 es = sb->u.ext3_sb.s_es; 566 if (le32_to_cpu(es->s_free_blocks_count) <= 567 le32_to_cpu(es->s_r_blocks_count) && 568 ((sb->u.ext3_sb.s_resuid != current->fsuid) && 569 (sb->u.ext3_sb.s_resgid == 0 || 570 !in_group_p (sb->u.ext3_sb.s_resgid)) && 571 !capable(CAP_SYS_RESOURCE))) 572 goto out; 573 574 ext3_debug ("goal=%lu.\n", goal); 575 576 /* 577 * First, test whether the goal block is free. 578 */ 579 if (goal < le32_to_cpu(es->s_first_data_block) || 580 goal >= le32_to_cpu(es->s_blocks_count)) 581 goal = le32_to_cpu(es->s_first_data_block); 582 i = (goal - le32_to_cpu(es->s_first_data_block)) / 583 EXT3_BLOCKS_PER_GROUP(sb); 584 gdp = ext3_get_group_desc (sb, i, &bh2); 585 if (!gdp) 586 goto io_error; 587 588 if (le16_to_cpu(gdp->bg_free_blocks_count) > 0) { 589 j = ((goal - le32_to_cpu(es->s_first_data_block)) % 590 EXT3_BLOCKS_PER_GROUP(sb)); 591#ifdef EXT3FS_DEBUG 592 if (j) 593 goal_attempts++; 594#endif 595 bitmap_nr = load_block_bitmap (sb, i); 596 if (bitmap_nr < 0) 597 goto io_error; 598 599 bh = sb->u.ext3_sb.s_block_bitmap[bitmap_nr]; 600 601 ext3_debug ("goal is at %d:%d.\n", i, j); 602 603 if (ext3_test_allocatable(j, bh)) { 604#ifdef EXT3FS_DEBUG 605 goal_hits++; 606 ext3_debug ("goal bit allocated.\n"); 607#endif 608 goto got_block; 609 } 610 611 j = find_next_usable_block(j, bh, EXT3_BLOCKS_PER_GROUP(sb)); 612 if (j >= 0) 613 goto search_back; 614 } 615 616 ext3_debug ("Bit not found in block group %d.\n", i); 617 618 /* 619 * Now search the rest of the groups. We assume that 620 * i and gdp correctly point to the last group visited. 621 */ 622 for (k = 0; k < sb->u.ext3_sb.s_groups_count; k++) { 623 i++; 624 if (i >= sb->u.ext3_sb.s_groups_count) 625 i = 0; 626 gdp = ext3_get_group_desc (sb, i, &bh2); 627 if (!gdp) { 628 *errp = -EIO; 629 goto out; 630 } 631 if (le16_to_cpu(gdp->bg_free_blocks_count) > 0) { 632 bitmap_nr = load_block_bitmap (sb, i); 633 if (bitmap_nr < 0) 634 goto io_error; 635 636 bh = sb->u.ext3_sb.s_block_bitmap[bitmap_nr]; 637 j = find_next_usable_block(-1, bh, 638 EXT3_BLOCKS_PER_GROUP(sb)); 639 if (j >= 0) 640 goto search_back; 641 } 642 } 643 644 /* No space left on the device */ 645 goto out; 646 647search_back: 648 /* 649 * We have succeeded in finding a free byte in the block 650 * bitmap. Now search backwards up to 7 bits to find the 651 * start of this group of free blocks. 652 */ 653 for ( k = 0; 654 k < 7 && j > 0 && ext3_test_allocatable(j - 1, bh); 655 k++, j--) 656 ; 657 658got_block: 659 660 ext3_debug ("using block group %d(%d)\n", i, gdp->bg_free_blocks_count); 661 662 /* Make sure we use undo access for the bitmap, because it is 663 critical that we do the frozen_data COW on bitmap buffers in 664 all cases even if the buffer is in BJ_Forget state in the 665 committing transaction. */ 666 BUFFER_TRACE(bh, "get undo access for marking new block"); 667 fatal = ext3_journal_get_undo_access(handle, bh); 668 if (fatal) goto out; 669 670 BUFFER_TRACE(bh2, "get_write_access"); 671 fatal = ext3_journal_get_write_access(handle, bh2); 672 if (fatal) goto out; 673 674 BUFFER_TRACE(sb->u.ext3_sb.s_sbh, "get_write_access"); 675 fatal = ext3_journal_get_write_access(handle, sb->u.ext3_sb.s_sbh); 676 if (fatal) goto out; 677 678 tmp = j + i * EXT3_BLOCKS_PER_GROUP(sb) 679 + le32_to_cpu(es->s_first_data_block); 680 681 if (tmp == le32_to_cpu(gdp->bg_block_bitmap) || 682 tmp == le32_to_cpu(gdp->bg_inode_bitmap) || 683 in_range (tmp, le32_to_cpu(gdp->bg_inode_table), 684 sb->u.ext3_sb.s_itb_per_group)) 685 ext3_error (sb, "ext3_new_block", 686 "Allocating block in system zone - " 687 "block = %u", tmp); 688 689 /* The superblock lock should guard against anybody else beating 690 * us to this point! */ 691 J_ASSERT_BH(bh, !ext3_test_bit(j, bh->b_data)); 692 BUFFER_TRACE(bh, "setting bitmap bit"); 693 ext3_set_bit(j, bh->b_data); 694 performed_allocation = 1; 695 696#ifdef CONFIG_JBD_DEBUG 697 { 698 struct buffer_head *debug_bh; 699 700 /* Record bitmap buffer state in the newly allocated block */ 701 debug_bh = sb_get_hash_table(sb, tmp); 702 if (debug_bh) { 703 BUFFER_TRACE(debug_bh, "state when allocated"); 704 BUFFER_TRACE2(debug_bh, bh, "bitmap state"); 705 brelse(debug_bh); 706 } 707 } 708#endif 709 if (buffer_jbd(bh) && bh2jh(bh)->b_committed_data) 710 J_ASSERT_BH(bh, !ext3_test_bit(j, bh2jh(bh)->b_committed_data)); 711 bhtmp = bh; 712 alloctmp = j; 713 714 ext3_debug ("found bit %d\n", j); 715 716 /* 717 * Do block preallocation now if required. 718 */ 719#ifdef EXT3_PREALLOCATE 720 /* 721 * akpm: this is not enabled for ext3. Need to use 722 * ext3_test_allocatable() 723 */ 724 /* Writer: ->i_prealloc* */ 725 if (prealloc_count && !*prealloc_count) { 726 int prealloc_goal; 727 unsigned long next_block = tmp + 1; 728 729 prealloc_goal = es->s_prealloc_blocks ? 730 es->s_prealloc_blocks : EXT3_DEFAULT_PREALLOC_BLOCKS; 731 732 *prealloc_block = next_block; 733 /* Writer: end */ 734 for (k = 1; 735 k < prealloc_goal && (j + k) < EXT3_BLOCKS_PER_GROUP(sb); 736 k++, next_block++) { 737 if (DQUOT_PREALLOC_BLOCK(inode, 1)) 738 break; 739 /* Writer: ->i_prealloc* */ 740 if (*prealloc_block + *prealloc_count != next_block || 741 ext3_set_bit (j + k, bh->b_data)) { 742 /* Writer: end */ 743 DQUOT_FREE_BLOCK(inode, 1); 744 break; 745 } 746 (*prealloc_count)++; 747 /* Writer: end */ 748 } 749 /* 750 * As soon as we go for per-group spinlocks we'll need these 751 * done inside the loop above. 752 */ 753 gdp->bg_free_blocks_count = 754 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) - 755 (k - 1)); 756 es->s_free_blocks_count = 757 cpu_to_le32(le32_to_cpu(es->s_free_blocks_count) - 758 (k - 1)); 759 ext3_debug ("Preallocated a further %lu bits.\n", 760 (k - 1)); 761 } 762#endif 763 764 j = tmp; 765 766 BUFFER_TRACE(bh, "journal_dirty_metadata for bitmap block"); 767 err = ext3_journal_dirty_metadata(handle, bh); 768 if (!fatal) fatal = err; 769 770 if (j >= le32_to_cpu(es->s_blocks_count)) { 771 ext3_error (sb, "ext3_new_block", 772 "block(%d) >= blocks count(%d) - " 773 "block_group = %d, es == %p ",j, 774 le32_to_cpu(es->s_blocks_count), i, es); 775 goto out; 776 } 777 778 /* 779 * It is up to the caller to add the new buffer to a journal 780 * list of some description. We don't know in advance whether 781 * the caller wants to use it as metadata or data. 782 */ 783 784 ext3_debug ("allocating block %d. " 785 "Goal hits %d of %d.\n", j, goal_hits, goal_attempts); 786 787 gdp->bg_free_blocks_count = 788 cpu_to_le16(le16_to_cpu(gdp->bg_free_blocks_count) - 1); 789 es->s_free_blocks_count = 790 cpu_to_le32(le32_to_cpu(es->s_free_blocks_count) - 1); 791 792 BUFFER_TRACE(bh2, "journal_dirty_metadata for group descriptor"); 793 err = ext3_journal_dirty_metadata(handle, bh2); 794 if (!fatal) fatal = err; 795 796 BUFFER_TRACE(bh, "journal_dirty_metadata for superblock"); 797 err = ext3_journal_dirty_metadata(handle, sb->u.ext3_sb.s_sbh); 798 if (!fatal) fatal = err; 799 800 sb->s_dirt = 1; 801 if (fatal) 802 goto out; 803 804 unlock_super (sb); 805 *errp = 0; 806 return j; 807 808io_error: 809 *errp = -EIO; 810out: 811 if (fatal) { 812 *errp = fatal; 813 ext3_std_error(sb, fatal); 814 } 815 unlock_super (sb); 816 /* 817 * Undo the block allocation 818 */ 819 if (!performed_allocation) 820 DQUOT_FREE_BLOCK(inode, 1); 821 return 0; 822 823} 824 825unsigned long ext3_count_free_blocks (struct super_block * sb) 826{ 827#ifdef EXT3FS_DEBUG 828 struct ext3_super_block * es; 829 unsigned long desc_count, bitmap_count, x; 830 int bitmap_nr; 831 struct ext3_group_desc * gdp; 832 int i; 833 834 lock_super (sb); 835 es = sb->u.ext3_sb.s_es; 836 desc_count = 0; 837 bitmap_count = 0; 838 gdp = NULL; 839 for (i = 0; i < sb->u.ext3_sb.s_groups_count; i++) { 840 gdp = ext3_get_group_desc (sb, i, NULL); 841 if (!gdp) 842 continue; 843 desc_count += le16_to_cpu(gdp->bg_free_blocks_count); 844 bitmap_nr = load_block_bitmap (sb, i); 845 if (bitmap_nr < 0) 846 continue; 847 848 x = ext3_count_free (sb->u.ext3_sb.s_block_bitmap[bitmap_nr], 849 sb->s_blocksize); 850 printk ("group %d: stored = %d, counted = %lu\n", 851 i, le16_to_cpu(gdp->bg_free_blocks_count), x); 852 bitmap_count += x; 853 } 854 printk("ext3_count_free_blocks: stored = %lu, computed = %lu, %lu\n", 855 le32_to_cpu(es->s_free_blocks_count), desc_count, bitmap_count); 856 unlock_super (sb); 857 return bitmap_count; 858#else 859 return le32_to_cpu(sb->u.ext3_sb.s_es->s_free_blocks_count); 860#endif 861} 862 863static inline int block_in_use (unsigned long block, 864 struct super_block * sb, 865 unsigned char * map) 866{ 867 return ext3_test_bit ((block - 868 le32_to_cpu(sb->u.ext3_sb.s_es->s_first_data_block)) % 869 EXT3_BLOCKS_PER_GROUP(sb), map); 870} 871 872static inline int test_root(int a, int b) 873{ 874 if (a == 0) 875 return 1; 876 while (1) { 877 if (a == 1) 878 return 1; 879 if (a % b) 880 return 0; 881 a = a / b; 882 } 883} 884 885int ext3_group_sparse(int group) 886{ 887 return (test_root(group, 3) || test_root(group, 5) || 888 test_root(group, 7)); 889} 890 891/** 892 * ext3_bg_has_super - number of blocks used by the superblock in group 893 * @sb: superblock for filesystem 894 * @group: group number to check 895 * 896 * Return the number of blocks used by the superblock (primary or backup) 897 * in this group. Currently this will be only 0 or 1. 898 */ 899int ext3_bg_has_super(struct super_block *sb, int group) 900{ 901 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)&& 902 !ext3_group_sparse(group)) 903 return 0; 904 return 1; 905} 906 907/** 908 * ext3_bg_num_gdb - number of blocks used by the group table in group 909 * @sb: superblock for filesystem 910 * @group: group number to check 911 * 912 * Return the number of blocks used by the group descriptor table 913 * (primary or backup) in this group. In the future there may be a 914 * different number of descriptor blocks in each group. 915 */ 916unsigned long ext3_bg_num_gdb(struct super_block *sb, int group) 917{ 918 if (EXT3_HAS_RO_COMPAT_FEATURE(sb,EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)&& 919 !ext3_group_sparse(group)) 920 return 0; 921 return EXT3_SB(sb)->s_gdb_count; 922} 923 924#ifdef CONFIG_EXT3_CHECK 925/* Called at mount-time, super-block is locked */ 926void ext3_check_blocks_bitmap (struct super_block * sb) 927{ 928 struct buffer_head * bh; 929 struct ext3_super_block * es; 930 unsigned long desc_count, bitmap_count, x, j; 931 unsigned long desc_blocks; 932 int bitmap_nr; 933 struct ext3_group_desc * gdp; 934 int i; 935 936 es = sb->u.ext3_sb.s_es; 937 desc_count = 0; 938 bitmap_count = 0; 939 gdp = NULL; 940 for (i = 0; i < sb->u.ext3_sb.s_groups_count; i++) { 941 gdp = ext3_get_group_desc (sb, i, NULL); 942 if (!gdp) 943 continue; 944 desc_count += le16_to_cpu(gdp->bg_free_blocks_count); 945 bitmap_nr = load_block_bitmap (sb, i); 946 if (bitmap_nr < 0) 947 continue; 948 949 bh = EXT3_SB(sb)->s_block_bitmap[bitmap_nr]; 950 951 if (ext3_bg_has_super(sb, i) && !ext3_test_bit(0, bh->b_data)) 952 ext3_error(sb, __FUNCTION__, 953 "Superblock in group %d is marked free", i); 954 955 desc_blocks = ext3_bg_num_gdb(sb, i); 956 for (j = 0; j < desc_blocks; j++) 957 if (!ext3_test_bit(j + 1, bh->b_data)) 958 ext3_error(sb, __FUNCTION__, 959 "Descriptor block #%ld in group " 960 "%d is marked free", j, i); 961 962 if (!block_in_use (le32_to_cpu(gdp->bg_block_bitmap), 963 sb, bh->b_data)) 964 ext3_error (sb, "ext3_check_blocks_bitmap", 965 "Block bitmap for group %d is marked free", 966 i); 967 968 if (!block_in_use (le32_to_cpu(gdp->bg_inode_bitmap), 969 sb, bh->b_data)) 970 ext3_error (sb, "ext3_check_blocks_bitmap", 971 "Inode bitmap for group %d is marked free", 972 i); 973 974 for (j = 0; j < sb->u.ext3_sb.s_itb_per_group; j++) 975 if (!block_in_use (le32_to_cpu(gdp->bg_inode_table) + j, 976 sb, bh->b_data)) 977 ext3_error (sb, "ext3_check_blocks_bitmap", 978 "Block #%d of the inode table in " 979 "group %d is marked free", j, i); 980 981 x = ext3_count_free (bh, sb->s_blocksize); 982 if (le16_to_cpu(gdp->bg_free_blocks_count) != x) 983 ext3_error (sb, "ext3_check_blocks_bitmap", 984 "Wrong free blocks count for group %d, " 985 "stored = %d, counted = %lu", i, 986 le16_to_cpu(gdp->bg_free_blocks_count), x); 987 bitmap_count += x; 988 } 989 if (le32_to_cpu(es->s_free_blocks_count) != bitmap_count) 990 ext3_error (sb, "ext3_check_blocks_bitmap", 991 "Wrong free blocks count in super block, " 992 "stored = %lu, counted = %lu", 993 (unsigned long)le32_to_cpu(es->s_free_blocks_count), 994 bitmap_count); 995} 996#endif 997