2 * linux/fs/ext2/ialloc.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 * 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
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>
25 * ialloc.c contains the inodes allocation and deallocation routines
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.
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.
40 * Read the inode allocation bitmap for a given block_group, reading
41 * into the specified slot in the superblock's bitmap cache.
43 * Return buffer_head of bitmap on success or NULL.
45 static struct buffer_head *
46 read_inode_bitmap(struct super_block * sb, unsigned long block_group)
48 struct ext2_group_desc *desc;
49 struct buffer_head *bh = NULL;
51 desc = ext2_get_group_desc(sb, block_group, NULL);
55 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
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));
65 static void ext2_release_inode(struct super_block *sb, int group, int dir)
67 struct ext2_group_desc * desc;
68 struct buffer_head *bh;
70 desc = ext2_get_group_desc(sb, group, &bh);
72 ext2_error(sb, "ext2_release_inode",
73 "can't get descriptor for group %d", group);
77 spin_lock(sb_bgl_lock(EXT2_SB(sb), group));
78 le16_add_cpu(&desc->bg_free_inodes_count, 1);
80 le16_add_cpu(&desc->bg_used_dirs_count, -1);
81 spin_unlock(sb_bgl_lock(EXT2_SB(sb), group));
82 percpu_counter_inc(&EXT2_SB(sb)->s_freeinodes_counter);
84 percpu_counter_dec(&EXT2_SB(sb)->s_dirs_counter);
85 mark_buffer_dirty(bh);
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.
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.
104 void ext2_free_inode (struct inode * inode)
106 struct super_block * sb = inode->i_sb;
109 struct buffer_head *bitmap_bh;
110 unsigned long block_group;
112 struct ext2_super_block * es;
115 ext2_debug ("freeing inode %lu\n", ino);
118 * Note: we must free any quota before locking the superblock,
119 * as writing the quota to disk may need the lock as well.
121 /* Quota is already initialized in iput() */
122 dquot_free_inode(inode);
125 es = EXT2_SB(sb)->s_es;
126 is_directory = S_ISDIR(inode->i_mode);
128 if (ino < EXT2_FIRST_INO(sb) ||
129 ino > le32_to_cpu(es->s_inodes_count)) {
130 ext2_error (sb, "ext2_free_inode",
131 "reserved or nonexistent inode %lu", ino);
134 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
135 bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
136 bitmap_bh = read_inode_bitmap(sb, block_group);
140 /* Ok, now we can actually update the inode bitmaps.. */
141 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb), block_group),
142 bit, (void *) bitmap_bh->b_data))
143 ext2_error (sb, "ext2_free_inode",
144 "bit already cleared for inode %lu", ino);
146 ext2_release_inode(sb, block_group, is_directory);
147 mark_buffer_dirty(bitmap_bh);
148 if (sb->s_flags & MS_SYNCHRONOUS)
149 sync_dirty_buffer(bitmap_bh);
155 * We perform asynchronous prereading of the new inode's inode block when
156 * we create the inode, in the expectation that the inode will be written
157 * back soon. There are two reasons:
159 * - When creating a large number of files, the async prereads will be
160 * nicely merged into large reads
161 * - When writing out a large number of inodes, we don't need to keep on
162 * stalling the writes while we read the inode block.
164 * FIXME: ext2_get_group_desc() needs to be simplified.
166 static void ext2_preread_inode(struct inode *inode)
168 unsigned long block_group;
169 unsigned long offset;
171 struct ext2_group_desc * gdp;
172 struct backing_dev_info *bdi;
174 bdi = inode_to_bdi(inode);
175 if (bdi_read_congested(bdi))
177 if (bdi_write_congested(bdi))
180 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
181 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL);
186 * Figure out the offset within the block group inode table
188 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) *
189 EXT2_INODE_SIZE(inode->i_sb);
190 block = le32_to_cpu(gdp->bg_inode_table) +
191 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb));
192 sb_breadahead(inode->i_sb, block);
196 * There are two policies for allocating an inode. If the new inode is
197 * a directory, then a forward search is made for a block group with both
198 * free space and a low directory-to-inode ratio; if that fails, then of
199 * the groups with above-average free space, that group with the fewest
200 * directories already is chosen.
202 * For other inodes, search forward from the parent directory\'s block
203 * group to find a free inode.
205 static int find_group_dir(struct super_block *sb, struct inode *parent)
207 int ngroups = EXT2_SB(sb)->s_groups_count;
208 int avefreei = ext2_count_free_inodes(sb) / ngroups;
209 struct ext2_group_desc *desc, *best_desc = NULL;
210 int group, best_group = -1;
212 for (group = 0; group < ngroups; group++) {
213 desc = ext2_get_group_desc (sb, group, NULL);
214 if (!desc || !desc->bg_free_inodes_count)
216 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
219 (le16_to_cpu(desc->bg_free_blocks_count) >
220 le16_to_cpu(best_desc->bg_free_blocks_count))) {
232 * Orlov's allocator for directories.
234 * We always try to spread first-level directories.
236 * If there are blockgroups with both free inodes and free blocks counts
237 * not worse than average we return one with smallest directory count.
238 * Otherwise we simply return a random group.
240 * For the rest rules look so:
242 * It's OK to put directory into a group unless
243 * it has too many directories already (max_dirs) or
244 * it has too few free inodes left (min_inodes) or
245 * it has too few free blocks left (min_blocks) or
246 * it's already running too large debt (max_debt).
247 * Parent's group is preferred, if it doesn't satisfy these
248 * conditions we search cyclically through the rest. If none
249 * of the groups look good we just look for a group with more
250 * free inodes than average (starting at parent's group).
252 * Debt is incremented each time we allocate a directory and decremented
253 * when we allocate an inode, within 0--255.
256 #define INODE_COST 64
257 #define BLOCK_COST 256
259 static int find_group_orlov(struct super_block *sb, struct inode *parent)
261 int parent_group = EXT2_I(parent)->i_block_group;
262 struct ext2_sb_info *sbi = EXT2_SB(sb);
263 struct ext2_super_block *es = sbi->s_es;
264 int ngroups = sbi->s_groups_count;
265 int inodes_per_group = EXT2_INODES_PER_GROUP(sb);
272 int max_debt, max_dirs, min_blocks, min_inodes;
274 struct ext2_group_desc *desc;
276 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
277 avefreei = freei / ngroups;
278 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
279 avefreeb = free_blocks / ngroups;
280 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
282 if ((parent == d_inode(sb->s_root)) ||
283 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) {
284 struct ext2_group_desc *best_desc = NULL;
285 int best_ndir = inodes_per_group;
288 group = prandom_u32();
289 parent_group = (unsigned)group % ngroups;
290 for (i = 0; i < ngroups; i++) {
291 group = (parent_group + i) % ngroups;
292 desc = ext2_get_group_desc (sb, group, NULL);
293 if (!desc || !desc->bg_free_inodes_count)
295 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
297 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
299 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
302 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
305 if (best_group >= 0) {
314 ndirs = 1; /* percpu_counters are approximate... */
316 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs;
318 max_dirs = ndirs / ngroups + inodes_per_group / 16;
319 min_inodes = avefreei - inodes_per_group / 4;
320 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4;
322 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
323 if (max_debt * INODE_COST > inodes_per_group)
324 max_debt = inodes_per_group / INODE_COST;
330 for (i = 0; i < ngroups; i++) {
331 group = (parent_group + i) % ngroups;
332 desc = ext2_get_group_desc (sb, group, NULL);
333 if (!desc || !desc->bg_free_inodes_count)
335 if (sbi->s_debts[group] >= max_debt)
337 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
339 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
341 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
347 for (i = 0; i < ngroups; i++) {
348 group = (parent_group + i) % ngroups;
349 desc = ext2_get_group_desc (sb, group, NULL);
350 if (!desc || !desc->bg_free_inodes_count)
352 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
358 * The free-inodes counter is approximate, and for really small
359 * filesystems the above test can fail to find any blockgroups
371 static int find_group_other(struct super_block *sb, struct inode *parent)
373 int parent_group = EXT2_I(parent)->i_block_group;
374 int ngroups = EXT2_SB(sb)->s_groups_count;
375 struct ext2_group_desc *desc;
379 * Try to place the inode in its parent directory
381 group = parent_group;
382 desc = ext2_get_group_desc (sb, group, NULL);
383 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
384 le16_to_cpu(desc->bg_free_blocks_count))
388 * We're going to place this inode in a different blockgroup from its
389 * parent. We want to cause files in a common directory to all land in
390 * the same blockgroup. But we want files which are in a different
391 * directory which shares a blockgroup with our parent to land in a
392 * different blockgroup.
394 * So add our directory's i_ino into the starting point for the hash.
396 group = (group + parent->i_ino) % ngroups;
399 * Use a quadratic hash to find a group with a free inode and some
402 for (i = 1; i < ngroups; i <<= 1) {
404 if (group >= ngroups)
406 desc = ext2_get_group_desc (sb, group, NULL);
407 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
408 le16_to_cpu(desc->bg_free_blocks_count))
413 * That failed: try linear search for a free inode, even if that group
414 * has no free blocks.
416 group = parent_group;
417 for (i = 0; i < ngroups; i++) {
418 if (++group >= ngroups)
420 desc = ext2_get_group_desc (sb, group, NULL);
421 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
431 struct inode *ext2_new_inode(struct inode *dir, umode_t mode,
432 const struct qstr *qstr)
434 struct super_block *sb;
435 struct buffer_head *bitmap_bh = NULL;
436 struct buffer_head *bh2;
439 struct inode * inode;
440 struct ext2_group_desc *gdp;
441 struct ext2_super_block *es;
442 struct ext2_inode_info *ei;
443 struct ext2_sb_info *sbi;
447 inode = new_inode(sb);
449 return ERR_PTR(-ENOMEM);
455 if (test_opt(sb, OLDALLOC))
456 group = find_group_dir(sb, dir);
458 group = find_group_orlov(sb, dir);
460 group = find_group_other(sb, dir);
467 for (i = 0; i < sbi->s_groups_count; i++) {
468 gdp = ext2_get_group_desc(sb, group, &bh2);
470 if (++group == sbi->s_groups_count)
475 bitmap_bh = read_inode_bitmap(sb, group);
482 repeat_in_this_group:
483 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data,
484 EXT2_INODES_PER_GROUP(sb), ino);
485 if (ino >= EXT2_INODES_PER_GROUP(sb)) {
487 * Rare race: find_group_xx() decided that there were
488 * free inodes in this group, but by the time we tried
489 * to allocate one, they're all gone. This can also
490 * occur because the counters which find_group_orlov()
491 * uses are approximate. So just go and search the
494 if (++group == sbi->s_groups_count)
498 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group),
499 ino, bitmap_bh->b_data)) {
500 /* we lost this inode */
501 if (++ino >= EXT2_INODES_PER_GROUP(sb)) {
502 /* this group is exhausted, try next group */
503 if (++group == sbi->s_groups_count)
507 /* try to find free inode in the same group */
508 goto repeat_in_this_group;
514 * Scanned all blockgroups.
519 mark_buffer_dirty(bitmap_bh);
520 if (sb->s_flags & MS_SYNCHRONOUS)
521 sync_dirty_buffer(bitmap_bh);
524 ino += group * EXT2_INODES_PER_GROUP(sb) + 1;
525 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
526 ext2_error (sb, "ext2_new_inode",
527 "reserved inode or inode > inodes count - "
528 "block_group = %d,inode=%lu", group,
529 (unsigned long) ino);
534 percpu_counter_dec(&sbi->s_freeinodes_counter);
536 percpu_counter_inc(&sbi->s_dirs_counter);
538 spin_lock(sb_bgl_lock(sbi, group));
539 le16_add_cpu(&gdp->bg_free_inodes_count, -1);
541 if (sbi->s_debts[group] < 255)
542 sbi->s_debts[group]++;
543 le16_add_cpu(&gdp->bg_used_dirs_count, 1);
545 if (sbi->s_debts[group])
546 sbi->s_debts[group]--;
548 spin_unlock(sb_bgl_lock(sbi, group));
550 mark_buffer_dirty(bh2);
551 if (test_opt(sb, GRPID)) {
552 inode->i_mode = mode;
553 inode->i_uid = current_fsuid();
554 inode->i_gid = dir->i_gid;
556 inode_init_owner(inode, dir, mode);
560 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
561 memset(ei->i_data, 0, sizeof(ei->i_data));
563 ext2_mask_flags(mode, EXT2_I(dir)->i_flags & EXT2_FL_INHERITED);
570 ei->i_block_alloc_info = NULL;
571 ei->i_block_group = group;
572 ei->i_dir_start_lookup = 0;
573 ei->i_state = EXT2_STATE_NEW;
574 ext2_set_inode_flags(inode);
575 spin_lock(&sbi->s_next_gen_lock);
576 inode->i_generation = sbi->s_next_generation++;
577 spin_unlock(&sbi->s_next_gen_lock);
578 if (insert_inode_locked(inode) < 0) {
579 ext2_error(sb, "ext2_new_inode",
580 "inode number already in use - inode=%lu",
581 (unsigned long) ino);
586 err = dquot_initialize(inode);
590 err = dquot_alloc_inode(inode);
594 err = ext2_init_acl(inode, dir);
598 err = ext2_init_security(inode, dir, qstr);
602 mark_inode_dirty(inode);
603 ext2_debug("allocating inode %lu\n", inode->i_ino);
604 ext2_preread_inode(inode);
608 dquot_free_inode(inode);
612 inode->i_flags |= S_NOQUOTA;
614 unlock_new_inode(inode);
619 make_bad_inode(inode);
624 unsigned long ext2_count_free_inodes (struct super_block * sb)
626 struct ext2_group_desc *desc;
627 unsigned long desc_count = 0;
631 struct ext2_super_block *es;
632 unsigned long bitmap_count = 0;
633 struct buffer_head *bitmap_bh = NULL;
635 es = EXT2_SB(sb)->s_es;
636 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
639 desc = ext2_get_group_desc (sb, i, NULL);
642 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
644 bitmap_bh = read_inode_bitmap(sb, i);
648 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8);
649 printk("group %d: stored = %d, counted = %u\n",
650 i, le16_to_cpu(desc->bg_free_inodes_count), x);
654 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
656 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter),
657 desc_count, bitmap_count);
660 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
661 desc = ext2_get_group_desc (sb, i, NULL);
664 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
670 /* Called at mount-time, super-block is locked */
671 unsigned long ext2_count_dirs (struct super_block * sb)
673 unsigned long count = 0;
676 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
677 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL);
680 count += le16_to_cpu(gdp->bg_used_dirs_count);