2 * linux/fs/ext2/balloc.c
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)
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
14 #include <linux/config.h>
16 #include <linux/quotaops.h>
17 #include <linux/sched.h>
18 #include <linux/buffer_head.h>
19 #include <linux/vs_dlimit.h>
22 * balloc.c contains the blocks allocation and deallocation routines
26 * The free blocks are managed by bitmaps. A file system contains several
27 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
28 * block for inodes, N blocks for the inode table and data blocks.
30 * The file system contains group descriptors which are located after the
31 * super block. Each descriptor contains the number of the bitmap block and
32 * the free blocks count in the block. The descriptors are loaded in memory
33 * when a file system is mounted (see ext2_read_super).
37 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
39 struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
40 unsigned int block_group,
41 struct buffer_head ** bh)
43 unsigned long group_desc;
45 struct ext2_group_desc * desc;
46 struct ext2_sb_info *sbi = EXT2_SB(sb);
48 if (block_group >= sbi->s_groups_count) {
49 ext2_error (sb, "ext2_get_group_desc",
50 "block_group >= groups_count - "
51 "block_group = %d, groups_count = %lu",
52 block_group, sbi->s_groups_count);
57 group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
58 offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
59 if (!sbi->s_group_desc[group_desc]) {
60 ext2_error (sb, "ext2_get_group_desc",
61 "Group descriptor not loaded - "
62 "block_group = %d, group_desc = %lu, desc = %lu",
63 block_group, group_desc, offset);
67 desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
69 *bh = sbi->s_group_desc[group_desc];
74 * Read the bitmap for a given block_group, reading into the specified
75 * slot in the superblock's bitmap cache.
77 * Return buffer_head on success or NULL in case of failure.
79 static struct buffer_head *
80 read_block_bitmap(struct super_block *sb, unsigned int block_group)
82 struct ext2_group_desc * desc;
83 struct buffer_head * bh = NULL;
85 desc = ext2_get_group_desc (sb, block_group, NULL);
88 bh = sb_bread(sb, le32_to_cpu(desc->bg_block_bitmap));
90 ext2_error (sb, "read_block_bitmap",
91 "Cannot read block bitmap - "
92 "block_group = %d, block_bitmap = %u",
93 block_group, le32_to_cpu(desc->bg_block_bitmap));
99 * Set sb->s_dirt here because the superblock was "logically" altered. We
100 * need to recalculate its free blocks count and flush it out.
102 static int reserve_blocks(struct super_block *sb, int count)
104 struct ext2_sb_info *sbi = EXT2_SB(sb);
105 struct ext2_super_block *es = sbi->s_es;
106 unsigned free_blocks;
107 unsigned root_blocks;
109 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
110 root_blocks = le32_to_cpu(es->s_r_blocks_count);
112 DLIMIT_ADJUST_BLOCK(sb, vx_current_xid(), &free_blocks, &root_blocks);
114 if (free_blocks < count)
117 if (free_blocks < root_blocks + count && !capable(CAP_SYS_RESOURCE) &&
118 sbi->s_resuid != current->fsuid &&
119 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
121 * We are too close to reserve and we are not privileged.
122 * Can we allocate anything at all?
124 if (free_blocks > root_blocks)
125 count = free_blocks - root_blocks;
130 percpu_counter_mod(&sbi->s_freeblocks_counter, -count);
135 static void release_blocks(struct super_block *sb, int count)
138 struct ext2_sb_info *sbi = EXT2_SB(sb);
140 percpu_counter_mod(&sbi->s_freeblocks_counter, count);
145 static int group_reserve_blocks(struct ext2_sb_info *sbi, int group_no,
146 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
148 unsigned free_blocks;
150 if (!desc->bg_free_blocks_count)
153 spin_lock(sb_bgl_lock(sbi, group_no));
154 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
155 if (free_blocks < count)
157 desc->bg_free_blocks_count = cpu_to_le16(free_blocks - count);
158 spin_unlock(sb_bgl_lock(sbi, group_no));
159 mark_buffer_dirty(bh);
163 static void group_release_blocks(struct super_block *sb, int group_no,
164 struct ext2_group_desc *desc, struct buffer_head *bh, int count)
167 struct ext2_sb_info *sbi = EXT2_SB(sb);
168 unsigned free_blocks;
170 spin_lock(sb_bgl_lock(sbi, group_no));
171 free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
172 desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
173 spin_unlock(sb_bgl_lock(sbi, group_no));
175 mark_buffer_dirty(bh);
179 /* Free given blocks, update quota and i_blocks field */
180 void ext2_free_blocks (struct inode * inode, unsigned long block,
183 struct buffer_head *bitmap_bh = NULL;
184 struct buffer_head * bh2;
185 unsigned long block_group;
188 unsigned long overflow;
189 struct super_block * sb = inode->i_sb;
190 struct ext2_sb_info * sbi = EXT2_SB(sb);
191 struct ext2_group_desc * desc;
192 struct ext2_super_block * es = sbi->s_es;
193 unsigned freed = 0, group_freed;
195 if (block < le32_to_cpu(es->s_first_data_block) ||
196 block + count < block ||
197 block + count > le32_to_cpu(es->s_blocks_count)) {
198 ext2_error (sb, "ext2_free_blocks",
199 "Freeing blocks not in datazone - "
200 "block = %lu, count = %lu", block, count);
204 ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
208 block_group = (block - le32_to_cpu(es->s_first_data_block)) /
209 EXT2_BLOCKS_PER_GROUP(sb);
210 bit = (block - le32_to_cpu(es->s_first_data_block)) %
211 EXT2_BLOCKS_PER_GROUP(sb);
213 * Check to see if we are freeing blocks across a group
216 if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
217 overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
221 bitmap_bh = read_block_bitmap(sb, block_group);
225 desc = ext2_get_group_desc (sb, block_group, &bh2);
229 if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
230 in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
231 in_range (block, le32_to_cpu(desc->bg_inode_table),
232 sbi->s_itb_per_group) ||
233 in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
234 sbi->s_itb_per_group))
235 ext2_error (sb, "ext2_free_blocks",
236 "Freeing blocks in system zones - "
237 "Block = %lu, count = %lu",
240 for (i = 0, group_freed = 0; i < count; i++) {
241 if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
242 bit + i, bitmap_bh->b_data)) {
243 ext2_error(sb, __FUNCTION__,
244 "bit already cleared for block %lu", block + i);
250 mark_buffer_dirty(bitmap_bh);
251 if (sb->s_flags & MS_SYNCHRONOUS)
252 sync_dirty_buffer(bitmap_bh);
254 group_release_blocks(sb, block_group, desc, bh2, group_freed);
255 freed += group_freed;
264 DLIMIT_FREE_BLOCK(sb, inode->i_xid, freed);
265 release_blocks(sb, freed);
266 DQUOT_FREE_BLOCK(inode, freed);
269 static int grab_block(spinlock_t *lock, char *map, unsigned size, int goal)
274 if (!ext2_test_bit(goal, map))
280 * The goal was occupied; search forward for a free
281 * block within the next XX blocks.
283 * end_goal is more or less random, but it has to be
284 * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
285 * next 64-bit boundary is simple..
287 k = (goal + 63) & ~63;
288 goal = ext2_find_next_zero_bit(map, k, goal);
292 * Search in the remainder of the current group.
296 p = map + (goal >> 3);
297 r = memscan(p, 0, (size - goal + 7) >> 3);
301 * We have succeeded in finding a free byte in the block
302 * bitmap. Now search backwards to find the start of this
303 * group of free blocks - won't take more than 7 iterations.
305 for (goal = k; goal && !ext2_test_bit (goal - 1, map); goal--)
310 k = ext2_find_next_zero_bit ((u32 *)map, size, goal);
317 if (ext2_set_bit_atomic(lock, goal, (void *) map))
323 * ext2_new_block uses a goal block to assist allocation. If the goal is
324 * free, or there is a free block within 32 blocks of the goal, that block
325 * is allocated. Otherwise a forward search is made for a free block; within
326 * each block group the search first looks for an entire free byte in the block
327 * bitmap, and then for any free bit if that fails.
328 * This function also updates quota and i_blocks field.
330 int ext2_new_block(struct inode *inode, unsigned long goal,
331 u32 *prealloc_count, u32 *prealloc_block, int *err)
333 struct buffer_head *bitmap_bh = NULL;
334 struct buffer_head *gdp_bh; /* bh2 */
335 struct ext2_group_desc *desc;
336 int group_no; /* i */
337 int ret_block; /* j */
338 int group_idx; /* k */
339 int target_block; /* tmp */
341 struct super_block *sb = inode->i_sb;
342 struct ext2_sb_info *sbi = EXT2_SB(sb);
343 struct ext2_super_block *es = sbi->s_es;
344 unsigned group_size = EXT2_BLOCKS_PER_GROUP(sb);
345 unsigned prealloc_goal = es->s_prealloc_blocks;
346 unsigned group_alloc = 0, es_alloc, dq_alloc;
347 int nr_scanned_groups;
349 if (!prealloc_goal--)
350 prealloc_goal = EXT2_DEFAULT_PREALLOC_BLOCKS - 1;
351 if (!prealloc_count || *prealloc_count)
354 if (DQUOT_ALLOC_BLOCK(inode, 1)) {
359 while (prealloc_goal && DQUOT_PREALLOC_BLOCK(inode, prealloc_goal))
362 dq_alloc = prealloc_goal + 1;
363 es_alloc = reserve_blocks(sb, dq_alloc);
368 if (DLIMIT_ALLOC_BLOCK(sb, inode->i_xid, es_alloc)) {
373 ext2_debug ("goal=%lu.\n", goal);
375 if (goal < le32_to_cpu(es->s_first_data_block) ||
376 goal >= le32_to_cpu(es->s_blocks_count))
377 goal = le32_to_cpu(es->s_first_data_block);
378 group_no = (goal - le32_to_cpu(es->s_first_data_block)) / group_size;
379 desc = ext2_get_group_desc (sb, group_no, &gdp_bh);
382 * gdp_bh may still be uninitialised. But group_release_blocks
383 * will not touch it because group_alloc is zero.
388 group_alloc = group_reserve_blocks(sbi, group_no, desc,
391 ret_block = ((goal - le32_to_cpu(es->s_first_data_block)) %
394 bitmap_bh = read_block_bitmap(sb, group_no);
398 ext2_debug("goal is at %d:%d.\n", group_no, ret_block);
400 ret_block = grab_block(sb_bgl_lock(sbi, group_no),
401 bitmap_bh->b_data, group_size, ret_block);
404 group_release_blocks(sb, group_no, desc, gdp_bh, group_alloc);
408 ext2_debug ("Bit not found in block group %d.\n", group_no);
411 * Now search the rest of the groups. We assume that
412 * i and desc correctly point to the last group visited.
414 nr_scanned_groups = 0;
416 for (group_idx = 0; !group_alloc &&
417 group_idx < sbi->s_groups_count; group_idx++) {
419 if (group_no >= sbi->s_groups_count)
421 desc = ext2_get_group_desc(sb, group_no, &gdp_bh);
424 group_alloc = group_reserve_blocks(sbi, group_no, desc,
432 bitmap_bh = read_block_bitmap(sb, group_no);
436 ret_block = grab_block(sb_bgl_lock(sbi, group_no), bitmap_bh->b_data,
440 * If a free block counter is corrupted we can loop inifintely.
444 if (nr_scanned_groups > 2 * sbi->s_groups_count) {
445 ext2_error(sb, "ext2_new_block",
446 "corrupted free blocks counters");
450 * Someone else grabbed the last free block in this blockgroup
451 * before us. Retry the scan.
453 group_release_blocks(sb, group_no, desc, gdp_bh, group_alloc);
459 ext2_debug("using block group %d(%d)\n",
460 group_no, desc->bg_free_blocks_count);
462 target_block = ret_block + group_no * group_size +
463 le32_to_cpu(es->s_first_data_block);
465 if (target_block == le32_to_cpu(desc->bg_block_bitmap) ||
466 target_block == le32_to_cpu(desc->bg_inode_bitmap) ||
467 in_range(target_block, le32_to_cpu(desc->bg_inode_table),
468 sbi->s_itb_per_group))
469 ext2_error (sb, "ext2_new_block",
470 "Allocating block in system zone - "
471 "block = %u", target_block);
473 if (target_block >= le32_to_cpu(es->s_blocks_count)) {
474 ext2_error (sb, "ext2_new_block",
475 "block(%d) >= blocks count(%d) - "
476 "block_group = %d, es == %p ", ret_block,
477 le32_to_cpu(es->s_blocks_count), group_no, es);
480 block = target_block;
482 /* OK, we _had_ allocated something */
483 ext2_debug("found bit %d\n", ret_block);
490 * Do block preallocation now if required.
492 write_lock(&EXT2_I(inode)->i_meta_lock);
493 if (group_alloc && !*prealloc_count) {
496 for (n = 0; n < group_alloc && ++ret_block < group_size; n++) {
497 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group_no),
499 (void*) bitmap_bh->b_data))
502 *prealloc_block = block + 1;
508 write_unlock(&EXT2_I(inode)->i_meta_lock);
510 mark_buffer_dirty(bitmap_bh);
511 if (sb->s_flags & MS_SYNCHRONOUS)
512 sync_dirty_buffer(bitmap_bh);
514 ext2_debug ("allocating block %d. ", block);
518 group_release_blocks(sb, group_no, desc, gdp_bh, group_alloc);
519 DLIMIT_FREE_BLOCK(sb, inode->i_xid, es_alloc);
521 release_blocks(sb, es_alloc);
523 DQUOT_FREE_BLOCK(inode, dq_alloc);
533 unsigned long ext2_count_free_blocks (struct super_block * sb)
535 struct ext2_group_desc * desc;
536 unsigned long desc_count = 0;
539 unsigned long bitmap_count, x;
540 struct ext2_super_block *es;
543 es = EXT2_SB(sb)->s_es;
547 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
548 struct buffer_head *bitmap_bh;
549 desc = ext2_get_group_desc (sb, i, NULL);
552 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
553 bitmap_bh = read_block_bitmap(sb, i);
557 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
558 printk ("group %d: stored = %d, counted = %lu\n",
559 i, le16_to_cpu(desc->bg_free_blocks_count), x);
563 printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
564 (long)le32_to_cpu(es->s_free_blocks_count),
565 desc_count, bitmap_count);
569 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
570 desc = ext2_get_group_desc (sb, i, NULL);
573 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
580 block_in_use(unsigned long block, struct super_block *sb, unsigned char *map)
582 return ext2_test_bit ((block -
583 le32_to_cpu(EXT2_SB(sb)->s_es->s_first_data_block)) %
584 EXT2_BLOCKS_PER_GROUP(sb), map);
587 static inline int test_root(int a, int b)
596 static int ext2_group_sparse(int group)
600 return (test_root(group, 3) || test_root(group, 5) ||
601 test_root(group, 7));
605 * ext2_bg_has_super - number of blocks used by the superblock in group
606 * @sb: superblock for filesystem
607 * @group: group number to check
609 * Return the number of blocks used by the superblock (primary or backup)
610 * in this group. Currently this will be only 0 or 1.
612 int ext2_bg_has_super(struct super_block *sb, int group)
614 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
615 !ext2_group_sparse(group))
621 * ext2_bg_num_gdb - number of blocks used by the group table in group
622 * @sb: superblock for filesystem
623 * @group: group number to check
625 * Return the number of blocks used by the group descriptor table
626 * (primary or backup) in this group. In the future there may be a
627 * different number of descriptor blocks in each group.
629 unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
631 if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
632 !ext2_group_sparse(group))
634 return EXT2_SB(sb)->s_gdb_count;
637 #ifdef CONFIG_EXT2_CHECK
638 /* Called at mount-time, super-block is locked */
639 void ext2_check_blocks_bitmap (struct super_block * sb)
641 struct buffer_head *bitmap_bh = NULL;
642 struct ext2_super_block * es;
643 unsigned long desc_count, bitmap_count, x, j;
644 unsigned long desc_blocks;
645 struct ext2_group_desc * desc;
648 es = EXT2_SB(sb)->s_es;
652 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
653 desc = ext2_get_group_desc (sb, i, NULL);
656 desc_count += le16_to_cpu(desc->bg_free_blocks_count);
658 bitmap_bh = read_block_bitmap(sb, i);
662 if (ext2_bg_has_super(sb, i) &&
663 !ext2_test_bit(0, bitmap_bh->b_data))
664 ext2_error(sb, __FUNCTION__,
665 "Superblock in group %d is marked free", i);
667 desc_blocks = ext2_bg_num_gdb(sb, i);
668 for (j = 0; j < desc_blocks; j++)
669 if (!ext2_test_bit(j + 1, bitmap_bh->b_data))
670 ext2_error(sb, __FUNCTION__,
671 "Descriptor block #%ld in group "
672 "%d is marked free", j, i);
674 if (!block_in_use(le32_to_cpu(desc->bg_block_bitmap),
675 sb, bitmap_bh->b_data))
676 ext2_error(sb, "ext2_check_blocks_bitmap",
677 "Block bitmap for group %d is marked free",
680 if (!block_in_use(le32_to_cpu(desc->bg_inode_bitmap),
681 sb, bitmap_bh->b_data))
682 ext2_error(sb, "ext2_check_blocks_bitmap",
683 "Inode bitmap for group %d is marked free",
686 for (j = 0; j < EXT2_SB(sb)->s_itb_per_group; j++)
687 if (!block_in_use(le32_to_cpu(desc->bg_inode_table) + j,
688 sb, bitmap_bh->b_data))
689 ext2_error (sb, "ext2_check_blocks_bitmap",
690 "Block #%ld of the inode table in "
691 "group %d is marked free", j, i);
693 x = ext2_count_free(bitmap_bh, sb->s_blocksize);
694 if (le16_to_cpu(desc->bg_free_blocks_count) != x)
695 ext2_error (sb, "ext2_check_blocks_bitmap",
696 "Wrong free blocks count for group %d, "
697 "stored = %d, counted = %lu", i,
698 le16_to_cpu(desc->bg_free_blocks_count), x);
701 if (le32_to_cpu(es->s_free_blocks_count) != bitmap_count)
702 ext2_error (sb, "ext2_check_blocks_bitmap",
703 "Wrong free blocks count in super block, "
704 "stored = %lu, counted = %lu",
705 (unsigned long)le32_to_cpu(es->s_free_blocks_count),