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));
83 percpu_counter_dec(&EXT2_SB(sb)->s_dirs_counter);
84 mark_buffer_dirty(bh);
88 * NOTE! When we get the inode, we're the only people
89 * that have access to it, and as such there are no
90 * race conditions we have to worry about. The inode
91 * is not on the hash-lists, and it cannot be reached
92 * through the filesystem because the directory entry
93 * has been deleted earlier.
95 * HOWEVER: we must make sure that we get no aliases,
96 * which means that we have to call "clear_inode()"
97 * _before_ we mark the inode not in use in the inode
98 * bitmaps. Otherwise a newly created file might use
99 * the same inode number (not actually the same pointer
100 * though), and then we'd have two inodes sharing the
101 * same inode number and space on the harddisk.
103 void ext2_free_inode (struct inode * inode)
105 struct super_block * sb = inode->i_sb;
108 struct buffer_head *bitmap_bh;
109 unsigned long block_group;
111 struct ext2_super_block * es;
114 ext2_debug ("freeing inode %lu\n", ino);
117 * Note: we must free any quota before locking the superblock,
118 * as writing the quota to disk may need the lock as well.
120 /* Quota is already initialized in iput() */
121 dquot_free_inode(inode);
124 es = EXT2_SB(sb)->s_es;
125 is_directory = S_ISDIR(inode->i_mode);
127 if (ino < EXT2_FIRST_INO(sb) ||
128 ino > le32_to_cpu(es->s_inodes_count)) {
129 ext2_error (sb, "ext2_free_inode",
130 "reserved or nonexistent inode %lu", ino);
133 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
134 bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
135 bitmap_bh = read_inode_bitmap(sb, block_group);
139 /* Ok, now we can actually update the inode bitmaps.. */
140 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb), block_group),
141 bit, (void *) bitmap_bh->b_data))
142 ext2_error (sb, "ext2_free_inode",
143 "bit already cleared for inode %lu", ino);
145 ext2_release_inode(sb, block_group, is_directory);
146 mark_buffer_dirty(bitmap_bh);
147 if (sb->s_flags & MS_SYNCHRONOUS)
148 sync_dirty_buffer(bitmap_bh);
154 * We perform asynchronous prereading of the new inode's inode block when
155 * we create the inode, in the expectation that the inode will be written
156 * back soon. There are two reasons:
158 * - When creating a large number of files, the async prereads will be
159 * nicely merged into large reads
160 * - When writing out a large number of inodes, we don't need to keep on
161 * stalling the writes while we read the inode block.
163 * FIXME: ext2_get_group_desc() needs to be simplified.
165 static void ext2_preread_inode(struct inode *inode)
167 unsigned long block_group;
168 unsigned long offset;
170 struct ext2_group_desc * gdp;
171 struct backing_dev_info *bdi;
173 bdi = inode->i_mapping->backing_dev_info;
174 if (bdi_read_congested(bdi))
176 if (bdi_write_congested(bdi))
179 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
180 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL);
185 * Figure out the offset within the block group inode table
187 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) *
188 EXT2_INODE_SIZE(inode->i_sb);
189 block = le32_to_cpu(gdp->bg_inode_table) +
190 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb));
191 sb_breadahead(inode->i_sb, block);
195 * There are two policies for allocating an inode. If the new inode is
196 * a directory, then a forward search is made for a block group with both
197 * free space and a low directory-to-inode ratio; if that fails, then of
198 * the groups with above-average free space, that group with the fewest
199 * directories already is chosen.
201 * For other inodes, search forward from the parent directory\'s block
202 * group to find a free inode.
204 static int find_group_dir(struct super_block *sb, struct inode *parent)
206 int ngroups = EXT2_SB(sb)->s_groups_count;
207 int avefreei = ext2_count_free_inodes(sb) / ngroups;
208 struct ext2_group_desc *desc, *best_desc = NULL;
209 int group, best_group = -1;
211 for (group = 0; group < ngroups; group++) {
212 desc = ext2_get_group_desc (sb, group, NULL);
213 if (!desc || !desc->bg_free_inodes_count)
215 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
218 (le16_to_cpu(desc->bg_free_blocks_count) >
219 le16_to_cpu(best_desc->bg_free_blocks_count))) {
231 * Orlov's allocator for directories.
233 * We always try to spread first-level directories.
235 * If there are blockgroups with both free inodes and free blocks counts
236 * not worse than average we return one with smallest directory count.
237 * Otherwise we simply return a random group.
239 * For the rest rules look so:
241 * It's OK to put directory into a group unless
242 * it has too many directories already (max_dirs) or
243 * it has too few free inodes left (min_inodes) or
244 * it has too few free blocks left (min_blocks) or
245 * it's already running too large debt (max_debt).
246 * Parent's group is preferred, if it doesn't satisfy these
247 * conditions we search cyclically through the rest. If none
248 * of the groups look good we just look for a group with more
249 * free inodes than average (starting at parent's group).
251 * Debt is incremented each time we allocate a directory and decremented
252 * when we allocate an inode, within 0--255.
255 #define INODE_COST 64
256 #define BLOCK_COST 256
258 static int find_group_orlov(struct super_block *sb, struct inode *parent)
260 int parent_group = EXT2_I(parent)->i_block_group;
261 struct ext2_sb_info *sbi = EXT2_SB(sb);
262 struct ext2_super_block *es = sbi->s_es;
263 int ngroups = sbi->s_groups_count;
264 int inodes_per_group = EXT2_INODES_PER_GROUP(sb);
271 int max_debt, max_dirs, min_blocks, min_inodes;
273 struct ext2_group_desc *desc;
275 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
276 avefreei = freei / ngroups;
277 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
278 avefreeb = free_blocks / ngroups;
279 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
281 if ((parent == sb->s_root->d_inode) ||
282 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) {
283 struct ext2_group_desc *best_desc = NULL;
284 int best_ndir = inodes_per_group;
287 get_random_bytes(&group, sizeof(group));
288 parent_group = (unsigned)group % ngroups;
289 for (i = 0; i < ngroups; i++) {
290 group = (parent_group + i) % ngroups;
291 desc = ext2_get_group_desc (sb, group, NULL);
292 if (!desc || !desc->bg_free_inodes_count)
294 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
296 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
298 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
301 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
304 if (best_group >= 0) {
313 ndirs = 1; /* percpu_counters are approximate... */
315 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs;
317 max_dirs = ndirs / ngroups + inodes_per_group / 16;
318 min_inodes = avefreei - inodes_per_group / 4;
319 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4;
321 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
322 if (max_debt * INODE_COST > inodes_per_group)
323 max_debt = inodes_per_group / INODE_COST;
329 for (i = 0; i < ngroups; i++) {
330 group = (parent_group + i) % ngroups;
331 desc = ext2_get_group_desc (sb, group, NULL);
332 if (!desc || !desc->bg_free_inodes_count)
334 if (sbi->s_debts[group] >= max_debt)
336 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
338 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
340 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
346 for (i = 0; i < ngroups; i++) {
347 group = (parent_group + i) % ngroups;
348 desc = ext2_get_group_desc (sb, group, NULL);
349 if (!desc || !desc->bg_free_inodes_count)
351 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
357 * The free-inodes counter is approximate, and for really small
358 * filesystems the above test can fail to find any blockgroups
370 static int find_group_other(struct super_block *sb, struct inode *parent)
372 int parent_group = EXT2_I(parent)->i_block_group;
373 int ngroups = EXT2_SB(sb)->s_groups_count;
374 struct ext2_group_desc *desc;
378 * Try to place the inode in its parent directory
380 group = parent_group;
381 desc = ext2_get_group_desc (sb, group, NULL);
382 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
383 le16_to_cpu(desc->bg_free_blocks_count))
387 * We're going to place this inode in a different blockgroup from its
388 * parent. We want to cause files in a common directory to all land in
389 * the same blockgroup. But we want files which are in a different
390 * directory which shares a blockgroup with our parent to land in a
391 * different blockgroup.
393 * So add our directory's i_ino into the starting point for the hash.
395 group = (group + parent->i_ino) % ngroups;
398 * Use a quadratic hash to find a group with a free inode and some
401 for (i = 1; i < ngroups; i <<= 1) {
403 if (group >= ngroups)
405 desc = ext2_get_group_desc (sb, group, NULL);
406 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
407 le16_to_cpu(desc->bg_free_blocks_count))
412 * That failed: try linear search for a free inode, even if that group
413 * has no free blocks.
415 group = parent_group;
416 for (i = 0; i < ngroups; i++) {
417 if (++group >= ngroups)
419 desc = ext2_get_group_desc (sb, group, NULL);
420 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
430 struct inode *ext2_new_inode(struct inode *dir, umode_t mode,
431 const struct qstr *qstr)
433 struct super_block *sb;
434 struct buffer_head *bitmap_bh = NULL;
435 struct buffer_head *bh2;
438 struct inode * inode;
439 struct ext2_group_desc *gdp;
440 struct ext2_super_block *es;
441 struct ext2_inode_info *ei;
442 struct ext2_sb_info *sbi;
446 inode = new_inode(sb);
448 return ERR_PTR(-ENOMEM);
454 if (test_opt(sb, OLDALLOC))
455 group = find_group_dir(sb, dir);
457 group = find_group_orlov(sb, dir);
459 group = find_group_other(sb, dir);
466 for (i = 0; i < sbi->s_groups_count; i++) {
467 gdp = ext2_get_group_desc(sb, group, &bh2);
469 bitmap_bh = read_inode_bitmap(sb, group);
476 repeat_in_this_group:
477 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data,
478 EXT2_INODES_PER_GROUP(sb), ino);
479 if (ino >= EXT2_INODES_PER_GROUP(sb)) {
481 * Rare race: find_group_xx() decided that there were
482 * free inodes in this group, but by the time we tried
483 * to allocate one, they're all gone. This can also
484 * occur because the counters which find_group_orlov()
485 * uses are approximate. So just go and search the
488 if (++group == sbi->s_groups_count)
492 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group),
493 ino, bitmap_bh->b_data)) {
494 /* we lost this inode */
495 if (++ino >= EXT2_INODES_PER_GROUP(sb)) {
496 /* this group is exhausted, try next group */
497 if (++group == sbi->s_groups_count)
501 /* try to find free inode in the same group */
502 goto repeat_in_this_group;
508 * Scanned all blockgroups.
513 mark_buffer_dirty(bitmap_bh);
514 if (sb->s_flags & MS_SYNCHRONOUS)
515 sync_dirty_buffer(bitmap_bh);
518 ino += group * EXT2_INODES_PER_GROUP(sb) + 1;
519 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
520 ext2_error (sb, "ext2_new_inode",
521 "reserved inode or inode > inodes count - "
522 "block_group = %d,inode=%lu", group,
523 (unsigned long) ino);
528 percpu_counter_add(&sbi->s_freeinodes_counter, -1);
530 percpu_counter_inc(&sbi->s_dirs_counter);
532 spin_lock(sb_bgl_lock(sbi, group));
533 le16_add_cpu(&gdp->bg_free_inodes_count, -1);
535 if (sbi->s_debts[group] < 255)
536 sbi->s_debts[group]++;
537 le16_add_cpu(&gdp->bg_used_dirs_count, 1);
539 if (sbi->s_debts[group])
540 sbi->s_debts[group]--;
542 spin_unlock(sb_bgl_lock(sbi, group));
544 mark_buffer_dirty(bh2);
545 if (test_opt(sb, GRPID)) {
546 inode->i_mode = mode;
547 inode->i_uid = current_fsuid();
548 inode->i_gid = dir->i_gid;
550 inode_init_owner(inode, dir, mode);
554 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
555 memset(ei->i_data, 0, sizeof(ei->i_data));
557 ext2_mask_flags(mode, EXT2_I(dir)->i_flags & EXT2_FL_INHERITED);
564 ei->i_block_alloc_info = NULL;
565 ei->i_block_group = group;
566 ei->i_dir_start_lookup = 0;
567 ei->i_state = EXT2_STATE_NEW;
568 ext2_set_inode_flags(inode);
569 spin_lock(&sbi->s_next_gen_lock);
570 inode->i_generation = sbi->s_next_generation++;
571 spin_unlock(&sbi->s_next_gen_lock);
572 if (insert_inode_locked(inode) < 0) {
573 ext2_error(sb, "ext2_new_inode",
574 "inode number already in use - inode=%lu",
575 (unsigned long) ino);
580 dquot_initialize(inode);
581 err = dquot_alloc_inode(inode);
585 err = ext2_init_acl(inode, dir);
589 err = ext2_init_security(inode, dir, qstr);
593 mark_inode_dirty(inode);
594 ext2_debug("allocating inode %lu\n", inode->i_ino);
595 ext2_preread_inode(inode);
599 dquot_free_inode(inode);
603 inode->i_flags |= S_NOQUOTA;
605 unlock_new_inode(inode);
610 make_bad_inode(inode);
615 unsigned long ext2_count_free_inodes (struct super_block * sb)
617 struct ext2_group_desc *desc;
618 unsigned long desc_count = 0;
622 struct ext2_super_block *es;
623 unsigned long bitmap_count = 0;
624 struct buffer_head *bitmap_bh = NULL;
626 es = EXT2_SB(sb)->s_es;
627 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
630 desc = ext2_get_group_desc (sb, i, NULL);
633 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
635 bitmap_bh = read_inode_bitmap(sb, i);
639 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8);
640 printk("group %d: stored = %d, counted = %u\n",
641 i, le16_to_cpu(desc->bg_free_inodes_count), x);
645 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
647 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter),
648 desc_count, bitmap_count);
651 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
652 desc = ext2_get_group_desc (sb, i, NULL);
655 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
661 /* Called at mount-time, super-block is locked */
662 unsigned long ext2_count_dirs (struct super_block * sb)
664 unsigned long count = 0;
667 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
668 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL);
671 count += le16_to_cpu(gdp->bg_used_dirs_count);