2 * linux/fs/ext3/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@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
15 #include <linux/time.h>
17 #include <linux/jbd.h>
18 #include <linux/ext3_fs.h>
19 #include <linux/ext3_jbd.h>
20 #include <linux/stat.h>
21 #include <linux/string.h>
22 #include <linux/quotaops.h>
23 #include <linux/buffer_head.h>
24 #include <linux/random.h>
26 #include <asm/bitops.h>
27 #include <asm/byteorder.h>
33 * ialloc.c contains the inodes allocation and deallocation routines
37 * The free inodes are managed by bitmaps. A file system contains several
38 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
39 * block for inodes, N blocks for the inode table and data blocks.
41 * The file system contains group descriptors which are located after the
42 * super block. Each descriptor contains the number of the bitmap block and
43 * the free blocks count in the block.
48 * Read the inode allocation bitmap for a given block_group, reading
49 * into the specified slot in the superblock's bitmap cache.
51 * Return buffer_head of bitmap on success or NULL.
53 static struct buffer_head *
54 read_inode_bitmap(struct super_block * sb, unsigned long block_group)
56 struct ext3_group_desc *desc;
57 struct buffer_head *bh = NULL;
59 desc = ext3_get_group_desc(sb, block_group, NULL);
63 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
65 ext3_error(sb, "read_inode_bitmap",
66 "Cannot read inode bitmap - "
67 "block_group = %lu, inode_bitmap = %lu",
68 block_group, (unsigned long) desc->bg_inode_bitmap);
74 * NOTE! When we get the inode, we're the only people
75 * that have access to it, and as such there are no
76 * race conditions we have to worry about. The inode
77 * is not on the hash-lists, and it cannot be reached
78 * through the filesystem because the directory entry
79 * has been deleted earlier.
81 * HOWEVER: we must make sure that we get no aliases,
82 * which means that we have to call "clear_inode()"
83 * _before_ we mark the inode not in use in the inode
84 * bitmaps. Otherwise a newly created file might use
85 * the same inode number (not actually the same pointer
86 * though), and then we'd have two inodes sharing the
87 * same inode number and space on the harddisk.
89 void ext3_free_inode (handle_t *handle, struct inode * inode)
91 struct super_block * sb = inode->i_sb;
94 struct buffer_head *bitmap_bh = NULL;
95 struct buffer_head *bh2;
96 unsigned long block_group;
98 struct ext3_group_desc * gdp;
99 struct ext3_super_block * es;
100 struct ext3_sb_info *sbi = EXT3_SB(sb);
103 if (atomic_read(&inode->i_count) > 1) {
104 printk ("ext3_free_inode: inode has count=%d\n",
105 atomic_read(&inode->i_count));
108 if (inode->i_nlink) {
109 printk ("ext3_free_inode: inode has nlink=%d\n",
114 printk("ext3_free_inode: inode on nonexistent device\n");
119 ext3_debug ("freeing inode %lu\n", ino);
122 * Note: we must free any quota before locking the superblock,
123 * as writing the quota to disk may need the lock as well.
126 ext3_xattr_delete_inode(handle, inode);
127 DQUOT_FREE_INODE(inode);
130 is_directory = S_ISDIR(inode->i_mode);
132 /* Do this BEFORE marking the inode not in use or returning an error */
135 es = EXT3_SB(sb)->s_es;
136 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
137 ext3_error (sb, "ext3_free_inode",
138 "reserved or nonexistent inode %lu", ino);
141 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
142 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
143 bitmap_bh = read_inode_bitmap(sb, block_group);
147 BUFFER_TRACE(bitmap_bh, "get_write_access");
148 fatal = ext3_journal_get_write_access(handle, bitmap_bh);
152 /* Ok, now we can actually update the inode bitmaps.. */
153 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
154 bit, bitmap_bh->b_data))
155 ext3_error (sb, "ext3_free_inode",
156 "bit already cleared for inode %lu", ino);
158 gdp = ext3_get_group_desc (sb, block_group, &bh2);
160 BUFFER_TRACE(bh2, "get_write_access");
161 fatal = ext3_journal_get_write_access(handle, bh2);
162 if (fatal) goto error_return;
165 spin_lock(sb_bgl_lock(sbi, block_group));
166 gdp->bg_free_inodes_count = cpu_to_le16(
167 le16_to_cpu(gdp->bg_free_inodes_count) + 1);
169 gdp->bg_used_dirs_count = cpu_to_le16(
170 le16_to_cpu(gdp->bg_used_dirs_count) - 1);
171 spin_unlock(sb_bgl_lock(sbi, block_group));
172 percpu_counter_inc(&sbi->s_freeinodes_counter);
174 percpu_counter_dec(&sbi->s_dirs_counter);
177 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
178 err = ext3_journal_dirty_metadata(handle, bh2);
179 if (!fatal) fatal = err;
181 BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata");
182 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
188 ext3_std_error(sb, fatal);
192 * There are two policies for allocating an inode. If the new inode is
193 * a directory, then a forward search is made for a block group with both
194 * free space and a low directory-to-inode ratio; if that fails, then of
195 * the groups with above-average free space, that group with the fewest
196 * directories already is chosen.
198 * For other inodes, search forward from the parent directory\'s block
199 * group to find a free inode.
201 static int find_group_dir(struct super_block *sb, struct inode *parent)
203 int ngroups = EXT3_SB(sb)->s_groups_count;
205 struct ext3_group_desc *desc, *best_desc = NULL;
206 struct buffer_head *bh;
207 int group, best_group = -1;
209 freei = percpu_counter_read_positive(&EXT3_SB(sb)->s_freeinodes_counter);
210 avefreei = freei / ngroups;
212 for (group = 0; group < ngroups; group++) {
213 desc = ext3_get_group_desc (sb, group, &bh);
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))) {
229 * Orlov's allocator for directories.
231 * We always try to spread first-level directories.
233 * If there are blockgroups with both free inodes and free blocks counts
234 * not worse than average we return one with smallest directory count.
235 * Otherwise we simply return a random group.
237 * For the rest rules look so:
239 * It's OK to put directory into a group unless
240 * it has too many directories already (max_dirs) or
241 * it has too few free inodes left (min_inodes) or
242 * it has too few free blocks left (min_blocks) or
243 * it's already running too large debt (max_debt).
244 * Parent's group is prefered, if it doesn't satisfy these
245 * conditions we search cyclically through the rest. If none
246 * of the groups look good we just look for a group with more
247 * free inodes than average (starting at parent's group).
249 * Debt is incremented each time we allocate a directory and decremented
250 * when we allocate an inode, within 0--255.
253 #define INODE_COST 64
254 #define BLOCK_COST 256
256 static int find_group_orlov(struct super_block *sb, struct inode *parent)
258 int parent_group = EXT3_I(parent)->i_block_group;
259 struct ext3_sb_info *sbi = EXT3_SB(sb);
260 struct ext3_super_block *es = sbi->s_es;
261 int ngroups = sbi->s_groups_count;
262 int inodes_per_group = EXT3_INODES_PER_GROUP(sb);
265 int blocks_per_dir, ndirs;
266 int max_debt, max_dirs, min_blocks, min_inodes;
268 struct ext3_group_desc *desc;
269 struct buffer_head *bh;
271 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
272 avefreei = freei / ngroups;
273 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
274 avefreeb = freeb / ngroups;
275 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
277 if ((parent == sb->s_root->d_inode) ||
278 (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) {
279 int best_ndir = inodes_per_group;
282 get_random_bytes(&group, sizeof(group));
283 parent_group = (unsigned)group % ngroups;
284 for (i = 0; i < ngroups; i++) {
285 group = (parent_group + i) % ngroups;
286 desc = ext3_get_group_desc (sb, group, &bh);
287 if (!desc || !desc->bg_free_inodes_count)
289 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
291 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
293 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
296 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
303 blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs;
305 max_dirs = ndirs / ngroups + inodes_per_group / 16;
306 min_inodes = avefreei - inodes_per_group / 4;
307 min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4;
309 max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
310 if (max_debt * INODE_COST > inodes_per_group)
311 max_debt = inodes_per_group / INODE_COST;
317 for (i = 0; i < ngroups; i++) {
318 group = (parent_group + i) % ngroups;
319 desc = ext3_get_group_desc (sb, group, &bh);
320 if (!desc || !desc->bg_free_inodes_count)
322 if (sbi->s_debts[group] >= max_debt)
324 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
326 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
328 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
334 for (i = 0; i < ngroups; i++) {
335 group = (parent_group + i) % ngroups;
336 desc = ext3_get_group_desc (sb, group, &bh);
337 if (!desc || !desc->bg_free_inodes_count)
339 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
345 * The free-inodes counter is approximate, and for really small
346 * filesystems the above test can fail to find any blockgroups
355 static int find_group_other(struct super_block *sb, struct inode *parent)
357 int parent_group = EXT3_I(parent)->i_block_group;
358 int ngroups = EXT3_SB(sb)->s_groups_count;
359 struct ext3_group_desc *desc;
360 struct buffer_head *bh;
364 * Try to place the inode in its parent directory
366 group = parent_group;
367 desc = ext3_get_group_desc (sb, group, &bh);
368 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
369 le16_to_cpu(desc->bg_free_blocks_count))
373 * We're going to place this inode in a different blockgroup from its
374 * parent. We want to cause files in a common directory to all land in
375 * the same blockgroup. But we want files which are in a different
376 * directory which shares a blockgroup with our parent to land in a
377 * different blockgroup.
379 * So add our directory's i_ino into the starting point for the hash.
381 group = (group + parent->i_ino) % ngroups;
384 * Use a quadratic hash to find a group with a free inode and some free
387 for (i = 1; i < ngroups; i <<= 1) {
389 if (group >= ngroups)
391 desc = ext3_get_group_desc (sb, group, &bh);
392 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
393 le16_to_cpu(desc->bg_free_blocks_count))
398 * That failed: try linear search for a free inode, even if that group
399 * has no free blocks.
401 group = parent_group;
402 for (i = 0; i < ngroups; i++) {
403 if (++group >= ngroups)
405 desc = ext3_get_group_desc (sb, group, &bh);
406 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
414 * There are two policies for allocating an inode. If the new inode is
415 * a directory, then a forward search is made for a block group with both
416 * free space and a low directory-to-inode ratio; if that fails, then of
417 * the groups with above-average free space, that group with the fewest
418 * directories already is chosen.
420 * For other inodes, search forward from the parent directory's block
421 * group to find a free inode.
423 struct inode *ext3_new_inode(handle_t *handle, struct inode * dir, int mode)
425 struct super_block *sb;
426 struct buffer_head *bitmap_bh = NULL;
427 struct buffer_head *bh2;
429 unsigned long ino = 0;
430 struct inode * inode;
431 struct ext3_group_desc * gdp = NULL;
432 struct ext3_super_block * es;
433 struct ext3_inode_info *ei;
434 struct ext3_sb_info *sbi;
439 /* Cannot create files in a deleted directory */
440 if (!dir || !dir->i_nlink)
441 return ERR_PTR(-EPERM);
444 inode = new_inode(sb);
446 return ERR_PTR(-ENOMEM);
452 if (test_opt (sb, OLDALLOC))
453 group = find_group_dir(sb, dir);
455 group = find_group_orlov(sb, dir);
457 group = find_group_other(sb, dir);
463 for (i = 0; i < sbi->s_groups_count; i++) {
464 gdp = ext3_get_group_desc(sb, group, &bh2);
468 bitmap_bh = read_inode_bitmap(sb, group);
474 repeat_in_this_group:
475 ino = ext3_find_next_zero_bit((unsigned long *)
476 bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino);
477 if (ino < EXT3_INODES_PER_GROUP(sb)) {
480 BUFFER_TRACE(bitmap_bh, "get_write_access");
481 err = ext3_journal_get_write_access_credits(handle,
482 bitmap_bh, &credits);
486 if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group),
487 ino, bitmap_bh->b_data)) {
489 BUFFER_TRACE(bitmap_bh,
490 "call ext3_journal_dirty_metadata");
491 err = ext3_journal_dirty_metadata(handle,
498 journal_release_buffer(handle, bitmap_bh, credits);
500 if (++ino < EXT3_INODES_PER_GROUP(sb))
501 goto repeat_in_this_group;
505 * This case is possible in concurrent environment. It is very
506 * rare. We cannot repeat the find_group_xxx() call because
507 * that will simply return the same blockgroup, because the
508 * group descriptor metadata has not yet been updated.
509 * So we just go onto the next blockgroup.
511 if (++group == sbi->s_groups_count)
518 ino += group * EXT3_INODES_PER_GROUP(sb) + 1;
519 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
520 ext3_error (sb, "ext3_new_inode",
521 "reserved inode or inode > inodes count - "
522 "block_group = %d, inode=%lu", group, ino);
527 BUFFER_TRACE(bh2, "get_write_access");
528 err = ext3_journal_get_write_access(handle, bh2);
530 spin_lock(sb_bgl_lock(sbi, group));
531 gdp->bg_free_inodes_count =
532 cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
534 gdp->bg_used_dirs_count =
535 cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
537 spin_unlock(sb_bgl_lock(sbi, group));
538 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
539 err = ext3_journal_dirty_metadata(handle, bh2);
542 percpu_counter_dec(&sbi->s_freeinodes_counter);
544 percpu_counter_inc(&sbi->s_dirs_counter);
547 inode->i_uid = current->fsuid;
548 if (test_opt (sb, GRPID))
549 inode->i_gid = dir->i_gid;
550 else if (dir->i_mode & S_ISGID) {
551 inode->i_gid = dir->i_gid;
555 inode->i_gid = current->fsgid;
556 inode->i_mode = mode;
559 /* This is the optimal IO size (for stat), not the fs block size */
560 inode->i_blksize = PAGE_SIZE;
562 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
564 memset(ei->i_data, 0, sizeof(ei->i_data));
565 ei->i_next_alloc_block = 0;
566 ei->i_next_alloc_goal = 0;
567 ei->i_dir_start_lookup = 0;
570 ei->i_flags = EXT3_I(dir)->i_flags &
571 ~(EXT3_INDEX_FL|EXT3_IUNLINK_FL|EXT3_BARRIER_FL);
573 ei->i_flags &= ~(EXT3_IMMUTABLE_FL|EXT3_APPEND_FL);
574 /* dirsync only applies to directories */
576 ei->i_flags &= ~EXT3_DIRSYNC_FL;
577 #ifdef EXT3_FRAGMENTS
585 #ifdef EXT3_PREALLOCATE
586 ei->i_prealloc_block = 0;
587 ei->i_prealloc_count = 0;
589 ei->i_block_group = group;
591 ext3_set_inode_flags(inode);
592 if (IS_DIRSYNC(inode))
594 insert_inode_hash(inode);
595 spin_lock(&sbi->s_next_gen_lock);
596 inode->i_generation = sbi->s_next_generation++;
597 spin_unlock(&sbi->s_next_gen_lock);
599 ei->i_state = EXT3_STATE_NEW;
602 if(DQUOT_ALLOC_INODE(inode)) {
607 err = ext3_init_acl(handle, inode, dir);
609 DQUOT_FREE_INODE(inode);
612 err = ext3_mark_inode_dirty(handle, inode);
614 ext3_std_error(sb, err);
615 DQUOT_FREE_INODE(inode);
619 ext3_debug("allocating inode %lu\n", inode->i_ino);
622 ext3_std_error(sb, err);
631 inode->i_flags |= S_NOQUOTA;
638 /* Verify that we are loading a valid orphan from disk */
639 struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino)
641 unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count);
642 unsigned long block_group;
644 struct buffer_head *bitmap_bh = NULL;
645 struct inode *inode = NULL;
647 /* Error cases - e2fsck has already cleaned up for us */
649 ext3_warning(sb, __FUNCTION__,
650 "bad orphan ino %lu! e2fsck was run?\n", ino);
654 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
655 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
656 bitmap_bh = read_inode_bitmap(sb, block_group);
658 ext3_warning(sb, __FUNCTION__,
659 "inode bitmap error for orphan %lu\n", ino);
663 /* Having the inode bit set should be a 100% indicator that this
664 * is a valid orphan (no e2fsck run on fs). Orphans also include
665 * inodes that were being truncated, so we can't check i_nlink==0.
667 if (!ext3_test_bit(bit, bitmap_bh->b_data) ||
668 !(inode = iget(sb, ino)) || is_bad_inode(inode) ||
669 NEXT_ORPHAN(inode) > max_ino) {
670 ext3_warning(sb, __FUNCTION__,
671 "bad orphan inode %lu! e2fsck was run?\n", ino);
672 printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n",
673 bit, (unsigned long long)bitmap_bh->b_blocknr,
674 ext3_test_bit(bit, bitmap_bh->b_data));
675 printk(KERN_NOTICE "inode=%p\n", inode);
677 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
678 is_bad_inode(inode));
679 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
681 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
683 /* Avoid freeing blocks if we got a bad deleted inode */
684 if (inode && inode->i_nlink == 0)
694 unsigned long ext3_count_free_inodes (struct super_block * sb)
696 unsigned long desc_count;
697 struct ext3_group_desc *gdp;
700 struct ext3_super_block *es;
701 unsigned long bitmap_count, x;
702 struct buffer_head *bitmap_bh = NULL;
705 es = EXT3_SB(sb)->s_es;
709 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
710 gdp = ext3_get_group_desc (sb, i, NULL);
713 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
715 bitmap_bh = read_inode_bitmap(sb, i);
719 x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8);
720 printk("group %d: stored = %d, counted = %lu\n",
721 i, le16_to_cpu(gdp->bg_free_inodes_count), x);
725 printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n",
726 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
731 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
732 gdp = ext3_get_group_desc (sb, i, NULL);
735 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
741 /* Called at mount-time, super-block is locked */
742 unsigned long ext3_count_dirs (struct super_block * sb)
744 unsigned long count = 0;
747 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
748 struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL);
751 count += le16_to_cpu(gdp->bg_used_dirs_count);
756 #ifdef CONFIG_EXT3_CHECK
757 /* Called at mount-time, super-block is locked */
758 void ext3_check_inodes_bitmap (struct super_block * sb)
760 struct ext3_super_block * es;
761 unsigned long desc_count, bitmap_count, x;
762 struct buffer_head *bitmap_bh = NULL;
763 struct ext3_group_desc * gdp;
766 es = EXT3_SB(sb)->s_es;
770 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
771 gdp = ext3_get_group_desc (sb, i, NULL);
774 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
776 bitmap_bh = read_inode_bitmap(sb, i);
780 x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8);
781 if (le16_to_cpu(gdp->bg_free_inodes_count) != x)
782 ext3_error (sb, "ext3_check_inodes_bitmap",
783 "Wrong free inodes count in group %d, "
784 "stored = %d, counted = %lu", i,
785 le16_to_cpu(gdp->bg_free_inodes_count), x);
789 if (le32_to_cpu(es->s_free_inodes_count) != bitmap_count)
790 ext3_error (sb, "ext3_check_inodes_bitmap",
791 "Wrong free inodes count in super block, "
792 "stored = %lu, counted = %lu",
793 (unsigned long)le32_to_cpu(es->s_free_inodes_count),