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>
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 = %u",
68 block_group, le32_to_cpu(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;
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");
120 ext3_debug ("freeing inode %lu\n", ino);
123 * Note: we must free any quota before locking the superblock,
124 * as writing the quota to disk may need the lock as well.
127 ext3_xattr_delete_inode(handle, inode);
128 DQUOT_FREE_INODE(inode);
131 is_directory = S_ISDIR(inode->i_mode);
133 /* Do this BEFORE marking the inode not in use or returning an error */
136 es = EXT3_SB(sb)->s_es;
137 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
138 ext3_error (sb, "ext3_free_inode",
139 "reserved or nonexistent inode %lu", ino);
142 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
143 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
144 bitmap_bh = read_inode_bitmap(sb, block_group);
148 BUFFER_TRACE(bitmap_bh, "get_write_access");
149 fatal = ext3_journal_get_write_access(handle, bitmap_bh);
153 /* Ok, now we can actually update the inode bitmaps.. */
154 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
155 bit, bitmap_bh->b_data))
156 ext3_error (sb, "ext3_free_inode",
157 "bit already cleared for inode %lu", ino);
159 gdp = ext3_get_group_desc (sb, block_group, &bh2);
161 BUFFER_TRACE(bh2, "get_write_access");
162 fatal = ext3_journal_get_write_access(handle, bh2);
163 if (fatal) goto error_return;
166 spin_lock(sb_bgl_lock(sbi, block_group));
167 le16_add_cpu(&gdp->bg_free_inodes_count, 1);
169 le16_add_cpu(&gdp->bg_used_dirs_count, -1);
170 spin_unlock(sb_bgl_lock(sbi, block_group));
171 percpu_counter_inc(&sbi->s_freeinodes_counter);
173 percpu_counter_dec(&sbi->s_dirs_counter);
176 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
177 err = ext3_journal_dirty_metadata(handle, bh2);
178 if (!fatal) fatal = err;
180 BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata");
181 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
187 ext3_std_error(sb, fatal);
191 * There are two policies for allocating an inode. If the new inode is
192 * a directory, then a forward search is made for a block group with both
193 * free space and a low directory-to-inode ratio; if that fails, then of
194 * the groups with above-average free space, that group with the fewest
195 * directories already is chosen.
197 * For other inodes, search forward from the parent directory\'s block
198 * group to find a free inode.
200 static int find_group_dir(struct super_block *sb, struct inode *parent)
202 int ngroups = EXT3_SB(sb)->s_groups_count;
203 unsigned int freei, avefreei;
204 struct ext3_group_desc *desc, *best_desc = NULL;
205 int group, best_group = -1;
207 freei = percpu_counter_read_positive(&EXT3_SB(sb)->s_freeinodes_counter);
208 avefreei = freei / ngroups;
210 for (group = 0; group < ngroups; group++) {
211 desc = ext3_get_group_desc (sb, group, NULL);
212 if (!desc || !desc->bg_free_inodes_count)
214 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
217 (le16_to_cpu(desc->bg_free_blocks_count) >
218 le16_to_cpu(best_desc->bg_free_blocks_count))) {
227 * Orlov's allocator for directories.
229 * We always try to spread first-level directories.
231 * If there are blockgroups with both free inodes and free blocks counts
232 * not worse than average we return one with smallest directory count.
233 * Otherwise we simply return a random group.
235 * For the rest rules look so:
237 * It's OK to put directory into a group unless
238 * it has too many directories already (max_dirs) or
239 * it has too few free inodes left (min_inodes) or
240 * it has too few free blocks left (min_blocks) or
241 * it's already running too large debt (max_debt).
242 * Parent's group is preferred, if it doesn't satisfy these
243 * conditions we search cyclically through the rest. If none
244 * of the groups look good we just look for a group with more
245 * free inodes than average (starting at parent's group).
247 * Debt is incremented each time we allocate a directory and decremented
248 * when we allocate an inode, within 0--255.
251 #define INODE_COST 64
252 #define BLOCK_COST 256
254 static int find_group_orlov(struct super_block *sb, struct inode *parent)
256 int parent_group = EXT3_I(parent)->i_block_group;
257 struct ext3_sb_info *sbi = EXT3_SB(sb);
258 struct ext3_super_block *es = sbi->s_es;
259 int ngroups = sbi->s_groups_count;
260 int inodes_per_group = EXT3_INODES_PER_GROUP(sb);
261 unsigned int freei, avefreei;
262 ext3_fsblk_t freeb, avefreeb;
263 ext3_fsblk_t blocks_per_dir;
265 int max_debt, max_dirs, min_inodes;
266 ext3_grpblk_t min_blocks;
268 struct ext3_group_desc *desc;
270 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
271 avefreei = freei / ngroups;
272 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
273 avefreeb = freeb / ngroups;
274 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
276 if ((parent == sb->s_root->d_inode) ||
277 (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) {
278 int best_ndir = inodes_per_group;
281 get_random_bytes(&group, sizeof(group));
282 parent_group = (unsigned)group % ngroups;
283 for (i = 0; i < ngroups; i++) {
284 group = (parent_group + i) % ngroups;
285 desc = ext3_get_group_desc (sb, group, NULL);
286 if (!desc || !desc->bg_free_inodes_count)
288 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
290 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
292 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
295 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
302 blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs;
304 max_dirs = ndirs / ngroups + inodes_per_group / 16;
305 min_inodes = avefreei - inodes_per_group / 4;
306 min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4;
308 max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, (ext3_fsblk_t)BLOCK_COST);
309 if (max_debt * INODE_COST > inodes_per_group)
310 max_debt = inodes_per_group / INODE_COST;
316 for (i = 0; i < ngroups; i++) {
317 group = (parent_group + i) % ngroups;
318 desc = ext3_get_group_desc (sb, group, NULL);
319 if (!desc || !desc->bg_free_inodes_count)
321 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
323 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
325 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
331 for (i = 0; i < ngroups; i++) {
332 group = (parent_group + i) % ngroups;
333 desc = ext3_get_group_desc (sb, group, NULL);
334 if (!desc || !desc->bg_free_inodes_count)
336 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
342 * The free-inodes counter is approximate, and for really small
343 * filesystems the above test can fail to find any blockgroups
352 static int find_group_other(struct super_block *sb, struct inode *parent)
354 int parent_group = EXT3_I(parent)->i_block_group;
355 int ngroups = EXT3_SB(sb)->s_groups_count;
356 struct ext3_group_desc *desc;
360 * Try to place the inode in its parent directory
362 group = parent_group;
363 desc = ext3_get_group_desc (sb, group, NULL);
364 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
365 le16_to_cpu(desc->bg_free_blocks_count))
369 * We're going to place this inode in a different blockgroup from its
370 * parent. We want to cause files in a common directory to all land in
371 * the same blockgroup. But we want files which are in a different
372 * directory which shares a blockgroup with our parent to land in a
373 * different blockgroup.
375 * So add our directory's i_ino into the starting point for the hash.
377 group = (group + parent->i_ino) % ngroups;
380 * Use a quadratic hash to find a group with a free inode and some free
383 for (i = 1; i < ngroups; i <<= 1) {
385 if (group >= ngroups)
387 desc = ext3_get_group_desc (sb, group, NULL);
388 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
389 le16_to_cpu(desc->bg_free_blocks_count))
394 * That failed: try linear search for a free inode, even if that group
395 * has no free blocks.
397 group = parent_group;
398 for (i = 0; i < ngroups; i++) {
399 if (++group >= ngroups)
401 desc = ext3_get_group_desc (sb, group, NULL);
402 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
410 * There are two policies for allocating an inode. If the new inode is
411 * a directory, then a forward search is made for a block group with both
412 * free space and a low directory-to-inode ratio; if that fails, then of
413 * the groups with above-average free space, that group with the fewest
414 * directories already is chosen.
416 * For other inodes, search forward from the parent directory's block
417 * group to find a free inode.
419 struct inode *ext3_new_inode(handle_t *handle, struct inode * dir, int mode)
421 struct super_block *sb;
422 struct buffer_head *bitmap_bh = NULL;
423 struct buffer_head *bh2;
425 unsigned long ino = 0;
426 struct inode * inode;
427 struct ext3_group_desc * gdp = NULL;
428 struct ext3_super_block * es;
429 struct ext3_inode_info *ei;
430 struct ext3_sb_info *sbi;
435 /* Cannot create files in a deleted directory */
436 if (!dir || !dir->i_nlink)
437 return ERR_PTR(-EPERM);
440 inode = new_inode(sb);
442 return ERR_PTR(-ENOMEM);
448 if (test_opt (sb, OLDALLOC))
449 group = find_group_dir(sb, dir);
451 group = find_group_orlov(sb, dir);
453 group = find_group_other(sb, dir);
459 for (i = 0; i < sbi->s_groups_count; i++) {
462 gdp = ext3_get_group_desc(sb, group, &bh2);
467 bitmap_bh = read_inode_bitmap(sb, group);
473 repeat_in_this_group:
474 ino = ext3_find_next_zero_bit((unsigned long *)
475 bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino);
476 if (ino < EXT3_INODES_PER_GROUP(sb)) {
478 BUFFER_TRACE(bitmap_bh, "get_write_access");
479 err = ext3_journal_get_write_access(handle, bitmap_bh);
483 if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group),
484 ino, bitmap_bh->b_data)) {
486 BUFFER_TRACE(bitmap_bh,
487 "call ext3_journal_dirty_metadata");
488 err = ext3_journal_dirty_metadata(handle,
495 journal_release_buffer(handle, bitmap_bh);
497 if (++ino < EXT3_INODES_PER_GROUP(sb))
498 goto repeat_in_this_group;
502 * This case is possible in concurrent environment. It is very
503 * rare. We cannot repeat the find_group_xxx() call because
504 * that will simply return the same blockgroup, because the
505 * group descriptor metadata has not yet been updated.
506 * So we just go onto the next blockgroup.
508 if (++group == sbi->s_groups_count)
515 ino += group * EXT3_INODES_PER_GROUP(sb) + 1;
516 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
517 ext3_error (sb, "ext3_new_inode",
518 "reserved inode or inode > inodes count - "
519 "block_group = %d, inode=%lu", group, ino);
524 BUFFER_TRACE(bh2, "get_write_access");
525 err = ext3_journal_get_write_access(handle, bh2);
527 spin_lock(sb_bgl_lock(sbi, group));
528 le16_add_cpu(&gdp->bg_free_inodes_count, -1);
530 le16_add_cpu(&gdp->bg_used_dirs_count, 1);
532 spin_unlock(sb_bgl_lock(sbi, group));
533 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
534 err = ext3_journal_dirty_metadata(handle, bh2);
537 percpu_counter_dec(&sbi->s_freeinodes_counter);
539 percpu_counter_inc(&sbi->s_dirs_counter);
542 inode->i_uid = current->fsuid;
543 if (test_opt (sb, GRPID))
544 inode->i_gid = dir->i_gid;
545 else if (dir->i_mode & S_ISGID) {
546 inode->i_gid = dir->i_gid;
550 inode->i_gid = current->fsgid;
551 inode->i_mode = mode;
554 /* This is the optimal IO size (for stat), not the fs block size */
556 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
558 memset(ei->i_data, 0, sizeof(ei->i_data));
559 ei->i_dir_start_lookup = 0;
562 ei->i_flags = EXT3_I(dir)->i_flags & ~EXT3_INDEX_FL;
564 ei->i_flags &= ~(EXT3_IMMUTABLE_FL|EXT3_APPEND_FL);
565 /* dirsync only applies to directories */
567 ei->i_flags &= ~EXT3_DIRSYNC_FL;
568 #ifdef EXT3_FRAGMENTS
576 ei->i_block_alloc_info = NULL;
577 ei->i_block_group = group;
579 ext3_set_inode_flags(inode);
580 if (IS_DIRSYNC(inode))
582 insert_inode_hash(inode);
583 spin_lock(&sbi->s_next_gen_lock);
584 inode->i_generation = sbi->s_next_generation++;
585 spin_unlock(&sbi->s_next_gen_lock);
587 ei->i_state = EXT3_STATE_NEW;
589 (EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) ?
590 sizeof(struct ext3_inode) - EXT3_GOOD_OLD_INODE_SIZE : 0;
593 if(DQUOT_ALLOC_INODE(inode)) {
598 err = ext3_init_acl(handle, inode, dir);
602 err = ext3_init_security(handle,inode, dir);
606 err = ext3_mark_inode_dirty(handle, inode);
608 ext3_std_error(sb, err);
612 ext3_debug("allocating inode %lu\n", inode->i_ino);
615 ext3_std_error(sb, err);
624 DQUOT_FREE_INODE(inode);
628 inode->i_flags |= S_NOQUOTA;
635 /* Verify that we are loading a valid orphan from disk */
636 struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino)
638 unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count);
639 unsigned long block_group;
641 struct buffer_head *bitmap_bh;
642 struct inode *inode = NULL;
645 /* Error cases - e2fsck has already cleaned up for us */
647 ext3_warning(sb, __func__,
648 "bad orphan ino %lu! e2fsck was run?", ino);
652 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
653 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
654 bitmap_bh = read_inode_bitmap(sb, block_group);
656 ext3_warning(sb, __func__,
657 "inode bitmap error for orphan %lu", ino);
661 /* Having the inode bit set should be a 100% indicator that this
662 * is a valid orphan (no e2fsck run on fs). Orphans also include
663 * inodes that were being truncated, so we can't check i_nlink==0.
665 if (!ext3_test_bit(bit, bitmap_bh->b_data))
668 inode = ext3_iget(sb, ino);
672 if (NEXT_ORPHAN(inode) > max_ino)
678 err = PTR_ERR(inode);
681 ext3_warning(sb, __func__,
682 "bad orphan inode %lu! e2fsck was run?", ino);
683 printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n",
684 bit, (unsigned long long)bitmap_bh->b_blocknr,
685 ext3_test_bit(bit, bitmap_bh->b_data));
686 printk(KERN_NOTICE "inode=%p\n", inode);
688 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
689 is_bad_inode(inode));
690 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
692 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
693 /* Avoid freeing blocks if we got a bad deleted inode */
694 if (inode->i_nlink == 0)
703 unsigned long ext3_count_free_inodes (struct super_block * sb)
705 unsigned long desc_count;
706 struct ext3_group_desc *gdp;
709 struct ext3_super_block *es;
710 unsigned long bitmap_count, x;
711 struct buffer_head *bitmap_bh = NULL;
713 es = EXT3_SB(sb)->s_es;
717 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
718 gdp = ext3_get_group_desc (sb, i, NULL);
721 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
723 bitmap_bh = read_inode_bitmap(sb, i);
727 x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8);
728 printk("group %d: stored = %d, counted = %lu\n",
729 i, le16_to_cpu(gdp->bg_free_inodes_count), x);
733 printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n",
734 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
738 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
739 gdp = ext3_get_group_desc (sb, i, NULL);
742 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
749 /* Called at mount-time, super-block is locked */
750 unsigned long ext3_count_dirs (struct super_block * sb)
752 unsigned long count = 0;
755 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
756 struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL);
759 count += le16_to_cpu(gdp->bg_used_dirs_count);