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>
25 #include <linux/bitops.h>
27 #include <asm/byteorder.h>
34 * ialloc.c contains the inodes allocation and deallocation routines
38 * The free inodes are managed by bitmaps. A file system contains several
39 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
40 * block for inodes, N blocks for the inode table and data blocks.
42 * The file system contains group descriptors which are located after the
43 * super block. Each descriptor contains the number of the bitmap block and
44 * the free blocks count in the block.
49 * Read the inode allocation bitmap for a given block_group, reading
50 * into the specified slot in the superblock's bitmap cache.
52 * Return buffer_head of bitmap on success or NULL.
54 static struct buffer_head *
55 read_inode_bitmap(struct super_block * sb, unsigned long block_group)
57 struct ext3_group_desc *desc;
58 struct buffer_head *bh = NULL;
60 desc = ext3_get_group_desc(sb, block_group, NULL);
64 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
66 ext3_error(sb, "read_inode_bitmap",
67 "Cannot read inode bitmap - "
68 "block_group = %lu, inode_bitmap = %u",
69 block_group, le32_to_cpu(desc->bg_inode_bitmap));
75 * NOTE! When we get the inode, we're the only people
76 * that have access to it, and as such there are no
77 * race conditions we have to worry about. The inode
78 * is not on the hash-lists, and it cannot be reached
79 * through the filesystem because the directory entry
80 * has been deleted earlier.
82 * HOWEVER: we must make sure that we get no aliases,
83 * which means that we have to call "clear_inode()"
84 * _before_ we mark the inode not in use in the inode
85 * bitmaps. Otherwise a newly created file might use
86 * the same inode number (not actually the same pointer
87 * though), and then we'd have two inodes sharing the
88 * same inode number and space on the harddisk.
90 void ext3_free_inode (handle_t *handle, struct inode * inode)
92 struct super_block * sb = inode->i_sb;
95 struct buffer_head *bitmap_bh = NULL;
96 struct buffer_head *bh2;
97 unsigned long block_group;
99 struct ext3_group_desc * gdp;
100 struct ext3_super_block * es;
101 struct ext3_sb_info *sbi;
104 if (atomic_read(&inode->i_count) > 1) {
105 printk ("ext3_free_inode: inode has count=%d\n",
106 atomic_read(&inode->i_count));
109 if (inode->i_nlink) {
110 printk ("ext3_free_inode: inode has nlink=%d\n",
115 printk("ext3_free_inode: inode on nonexistent device\n");
121 ext3_debug ("freeing inode %lu\n", ino);
124 * Note: we must free any quota before locking the superblock,
125 * as writing the quota to disk may need the lock as well.
128 ext3_xattr_delete_inode(handle, inode);
129 DQUOT_FREE_INODE(inode);
132 is_directory = S_ISDIR(inode->i_mode);
134 /* Do this BEFORE marking the inode not in use or returning an error */
137 es = EXT3_SB(sb)->s_es;
138 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
139 ext3_error (sb, "ext3_free_inode",
140 "reserved or nonexistent inode %lu", ino);
143 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
144 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
145 bitmap_bh = read_inode_bitmap(sb, block_group);
149 BUFFER_TRACE(bitmap_bh, "get_write_access");
150 fatal = ext3_journal_get_write_access(handle, bitmap_bh);
154 /* Ok, now we can actually update the inode bitmaps.. */
155 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
156 bit, bitmap_bh->b_data))
157 ext3_error (sb, "ext3_free_inode",
158 "bit already cleared for inode %lu", ino);
160 gdp = ext3_get_group_desc (sb, block_group, &bh2);
162 BUFFER_TRACE(bh2, "get_write_access");
163 fatal = ext3_journal_get_write_access(handle, bh2);
164 if (fatal) goto error_return;
167 spin_lock(sb_bgl_lock(sbi, block_group));
168 gdp->bg_free_inodes_count = cpu_to_le16(
169 le16_to_cpu(gdp->bg_free_inodes_count) + 1);
171 gdp->bg_used_dirs_count = cpu_to_le16(
172 le16_to_cpu(gdp->bg_used_dirs_count) - 1);
173 spin_unlock(sb_bgl_lock(sbi, block_group));
174 percpu_counter_inc(&sbi->s_freeinodes_counter);
176 percpu_counter_dec(&sbi->s_dirs_counter);
179 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
180 err = ext3_journal_dirty_metadata(handle, bh2);
181 if (!fatal) fatal = err;
183 BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata");
184 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
190 ext3_std_error(sb, fatal);
194 * There are two policies for allocating an inode. If the new inode is
195 * a directory, then a forward search is made for a block group with both
196 * free space and a low directory-to-inode ratio; if that fails, then of
197 * the groups with above-average free space, that group with the fewest
198 * directories already is chosen.
200 * For other inodes, search forward from the parent directory\'s block
201 * group to find a free inode.
203 static int find_group_dir(struct super_block *sb, struct inode *parent)
205 int ngroups = EXT3_SB(sb)->s_groups_count;
207 struct ext3_group_desc *desc, *best_desc = NULL;
208 struct buffer_head *bh;
209 int group, best_group = -1;
211 freei = percpu_counter_read_positive(&EXT3_SB(sb)->s_freeinodes_counter);
212 avefreei = freei / ngroups;
214 for (group = 0; group < ngroups; group++) {
215 desc = ext3_get_group_desc (sb, group, &bh);
216 if (!desc || !desc->bg_free_inodes_count)
218 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
221 (le16_to_cpu(desc->bg_free_blocks_count) >
222 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 prefered, 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 = EXT3_I(parent)->i_block_group;
261 struct ext3_sb_info *sbi = EXT3_SB(sb);
262 struct ext3_super_block *es = sbi->s_es;
263 int ngroups = sbi->s_groups_count;
264 int inodes_per_group = EXT3_INODES_PER_GROUP(sb);
267 int blocks_per_dir, ndirs;
268 int max_debt, max_dirs, min_blocks, min_inodes;
270 struct ext3_group_desc *desc;
271 struct buffer_head *bh;
273 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
274 avefreei = freei / ngroups;
275 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
276 avefreeb = freeb / ngroups;
277 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
279 if ((parent == sb->s_root->d_inode) ||
280 (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) {
281 int best_ndir = inodes_per_group;
284 get_random_bytes(&group, sizeof(group));
285 parent_group = (unsigned)group % ngroups;
286 for (i = 0; i < ngroups; i++) {
287 group = (parent_group + i) % ngroups;
288 desc = ext3_get_group_desc (sb, group, &bh);
289 if (!desc || !desc->bg_free_inodes_count)
291 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
293 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
295 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
298 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
305 blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs;
307 max_dirs = ndirs / ngroups + inodes_per_group / 16;
308 min_inodes = avefreei - inodes_per_group / 4;
309 min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4;
311 max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
312 if (max_debt * INODE_COST > inodes_per_group)
313 max_debt = inodes_per_group / INODE_COST;
319 for (i = 0; i < ngroups; i++) {
320 group = (parent_group + i) % ngroups;
321 desc = ext3_get_group_desc (sb, group, &bh);
322 if (!desc || !desc->bg_free_inodes_count)
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++) {
466 gdp = ext3_get_group_desc(sb, group, &bh2);
471 bitmap_bh = read_inode_bitmap(sb, group);
477 repeat_in_this_group:
478 ino = ext3_find_next_zero_bit((unsigned long *)
479 bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino);
480 if (ino < EXT3_INODES_PER_GROUP(sb)) {
482 BUFFER_TRACE(bitmap_bh, "get_write_access");
483 err = ext3_journal_get_write_access(handle, bitmap_bh);
487 if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group),
488 ino, bitmap_bh->b_data)) {
490 BUFFER_TRACE(bitmap_bh,
491 "call ext3_journal_dirty_metadata");
492 err = ext3_journal_dirty_metadata(handle,
499 journal_release_buffer(handle, bitmap_bh);
501 if (++ino < EXT3_INODES_PER_GROUP(sb))
502 goto repeat_in_this_group;
506 * This case is possible in concurrent environment. It is very
507 * rare. We cannot repeat the find_group_xxx() call because
508 * that will simply return the same blockgroup, because the
509 * group descriptor metadata has not yet been updated.
510 * So we just go onto the next blockgroup.
512 if (++group == sbi->s_groups_count)
519 ino += group * EXT3_INODES_PER_GROUP(sb) + 1;
520 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
521 ext3_error (sb, "ext3_new_inode",
522 "reserved inode or inode > inodes count - "
523 "block_group = %d, inode=%lu", group, ino);
528 BUFFER_TRACE(bh2, "get_write_access");
529 err = ext3_journal_get_write_access(handle, bh2);
531 spin_lock(sb_bgl_lock(sbi, group));
532 gdp->bg_free_inodes_count =
533 cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
535 gdp->bg_used_dirs_count =
536 cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
538 spin_unlock(sb_bgl_lock(sbi, group));
539 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
540 err = ext3_journal_dirty_metadata(handle, bh2);
543 percpu_counter_dec(&sbi->s_freeinodes_counter);
545 percpu_counter_inc(&sbi->s_dirs_counter);
548 inode->i_uid = current->fsuid;
549 if (test_opt (sb, GRPID))
550 inode->i_gid = dir->i_gid;
551 else if (dir->i_mode & S_ISGID) {
552 inode->i_gid = dir->i_gid;
556 inode->i_gid = current->fsgid;
557 inode->i_mode = mode;
560 /* This is the optimal IO size (for stat), not the fs block size */
561 inode->i_blksize = PAGE_SIZE;
563 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
565 memset(ei->i_data, 0, sizeof(ei->i_data));
566 ei->i_dir_start_lookup = 0;
569 ei->i_flags = EXT3_I(dir)->i_flags & ~EXT3_INDEX_FL;
571 ei->i_flags &= ~(EXT3_IMMUTABLE_FL|EXT3_APPEND_FL);
572 /* dirsync only applies to directories */
574 ei->i_flags &= ~EXT3_DIRSYNC_FL;
575 #ifdef EXT3_FRAGMENTS
583 ei->i_block_alloc_info = NULL;
584 ei->i_block_group = group;
586 ext3_set_inode_flags(inode);
587 if (IS_DIRSYNC(inode))
589 insert_inode_hash(inode);
590 spin_lock(&sbi->s_next_gen_lock);
591 inode->i_generation = sbi->s_next_generation++;
592 spin_unlock(&sbi->s_next_gen_lock);
594 ei->i_state = EXT3_STATE_NEW;
596 (EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) ?
597 sizeof(struct ext3_inode) - EXT3_GOOD_OLD_INODE_SIZE : 0;
600 if(DQUOT_ALLOC_INODE(inode)) {
605 err = ext3_init_acl(handle, inode, dir);
609 err = ext3_init_security(handle,inode, dir);
613 err = ext3_mark_inode_dirty(handle, inode);
615 ext3_std_error(sb, err);
619 ext3_debug("allocating inode %lu\n", inode->i_ino);
622 ext3_std_error(sb, err);
631 DQUOT_FREE_INODE(inode);
635 inode->i_flags |= S_NOQUOTA;
642 /* Verify that we are loading a valid orphan from disk */
643 struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino)
645 unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count);
646 unsigned long block_group;
648 struct buffer_head *bitmap_bh = NULL;
649 struct inode *inode = NULL;
651 /* Error cases - e2fsck has already cleaned up for us */
653 ext3_warning(sb, __FUNCTION__,
654 "bad orphan ino %lu! e2fsck was run?", ino);
658 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
659 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
660 bitmap_bh = read_inode_bitmap(sb, block_group);
662 ext3_warning(sb, __FUNCTION__,
663 "inode bitmap error for orphan %lu", ino);
667 /* Having the inode bit set should be a 100% indicator that this
668 * is a valid orphan (no e2fsck run on fs). Orphans also include
669 * inodes that were being truncated, so we can't check i_nlink==0.
671 if (!ext3_test_bit(bit, bitmap_bh->b_data) ||
672 !(inode = iget(sb, ino)) || is_bad_inode(inode) ||
673 NEXT_ORPHAN(inode) > max_ino) {
674 ext3_warning(sb, __FUNCTION__,
675 "bad orphan inode %lu! e2fsck was run?", ino);
676 printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n",
677 bit, (unsigned long long)bitmap_bh->b_blocknr,
678 ext3_test_bit(bit, bitmap_bh->b_data));
679 printk(KERN_NOTICE "inode=%p\n", inode);
681 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
682 is_bad_inode(inode));
683 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
685 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
687 /* Avoid freeing blocks if we got a bad deleted inode */
688 if (inode && inode->i_nlink == 0)
698 unsigned long ext3_count_free_inodes (struct super_block * sb)
700 unsigned long desc_count;
701 struct ext3_group_desc *gdp;
704 struct ext3_super_block *es;
705 unsigned long bitmap_count, x;
706 struct buffer_head *bitmap_bh = NULL;
708 es = EXT3_SB(sb)->s_es;
712 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
713 gdp = ext3_get_group_desc (sb, i, NULL);
716 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
718 bitmap_bh = read_inode_bitmap(sb, i);
722 x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8);
723 printk("group %d: stored = %d, counted = %lu\n",
724 i, le16_to_cpu(gdp->bg_free_inodes_count), x);
728 printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n",
729 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
733 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
734 gdp = ext3_get_group_desc (sb, i, NULL);
737 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
744 /* Called at mount-time, super-block is locked */
745 unsigned long ext3_count_dirs (struct super_block * sb)
747 unsigned long count = 0;
750 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
751 struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL);
754 count += le16_to_cpu(gdp->bg_used_dirs_count);