2 * linux/fs/ext3/ialloc.c
4 * Copyright (C) 1992, 1993, 1994, 1995
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
9 * BSD ufs-inspired inode and directory allocation by
11 * Big-endian to little-endian byte-swapping/bitmaps by
15 #include <linux/quotaops.h>
16 #include <linux/random.h>
23 * ialloc.c contains the inodes allocation and deallocation routines
27 * The free inodes are managed by bitmaps. A file system contains several
28 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
29 * block for inodes, N blocks for the inode table and data blocks.
31 * The file system contains group descriptors which are located after the
32 * super block. Each descriptor contains the number of the bitmap block and
33 * the free blocks count in the block.
38 * Read the inode allocation bitmap for a given block_group, reading
39 * into the specified slot in the superblock's bitmap cache.
41 * Return buffer_head of bitmap on success or NULL.
43 static struct buffer_head *
44 read_inode_bitmap(struct super_block * sb, unsigned long block_group)
46 struct ext3_group_desc *desc;
47 struct buffer_head *bh = NULL;
49 desc = ext3_get_group_desc(sb, block_group, NULL);
53 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
55 ext3_error(sb, "read_inode_bitmap",
56 "Cannot read inode bitmap - "
57 "block_group = %lu, inode_bitmap = %u",
58 block_group, le32_to_cpu(desc->bg_inode_bitmap));
64 * NOTE! When we get the inode, we're the only people
65 * that have access to it, and as such there are no
66 * race conditions we have to worry about. The inode
67 * is not on the hash-lists, and it cannot be reached
68 * through the filesystem because the directory entry
69 * has been deleted earlier.
71 * HOWEVER: we must make sure that we get no aliases,
72 * which means that we have to call "clear_inode()"
73 * _before_ we mark the inode not in use in the inode
74 * bitmaps. Otherwise a newly created file might use
75 * the same inode number (not actually the same pointer
76 * though), and then we'd have two inodes sharing the
77 * same inode number and space on the harddisk.
79 void ext3_free_inode (handle_t *handle, struct inode * inode)
81 struct super_block * sb = inode->i_sb;
84 struct buffer_head *bitmap_bh = NULL;
85 struct buffer_head *bh2;
86 unsigned long block_group;
88 struct ext3_group_desc * gdp;
89 struct ext3_super_block * es;
90 struct ext3_sb_info *sbi;
93 if (atomic_read(&inode->i_count) > 1) {
94 printk ("ext3_free_inode: inode has count=%d\n",
95 atomic_read(&inode->i_count));
99 printk ("ext3_free_inode: inode has nlink=%d\n",
104 printk("ext3_free_inode: inode on nonexistent device\n");
110 ext3_debug ("freeing inode %lu\n", ino);
111 trace_ext3_free_inode(inode);
113 is_directory = S_ISDIR(inode->i_mode);
115 es = EXT3_SB(sb)->s_es;
116 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
117 ext3_error (sb, "ext3_free_inode",
118 "reserved or nonexistent inode %lu", ino);
121 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
122 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
123 bitmap_bh = read_inode_bitmap(sb, block_group);
127 BUFFER_TRACE(bitmap_bh, "get_write_access");
128 fatal = ext3_journal_get_write_access(handle, bitmap_bh);
132 /* Ok, now we can actually update the inode bitmaps.. */
133 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
134 bit, bitmap_bh->b_data))
135 ext3_error (sb, "ext3_free_inode",
136 "bit already cleared for inode %lu", ino);
138 gdp = ext3_get_group_desc (sb, block_group, &bh2);
140 BUFFER_TRACE(bh2, "get_write_access");
141 fatal = ext3_journal_get_write_access(handle, bh2);
142 if (fatal) goto error_return;
145 spin_lock(sb_bgl_lock(sbi, block_group));
146 le16_add_cpu(&gdp->bg_free_inodes_count, 1);
148 le16_add_cpu(&gdp->bg_used_dirs_count, -1);
149 spin_unlock(sb_bgl_lock(sbi, block_group));
150 percpu_counter_inc(&sbi->s_freeinodes_counter);
152 percpu_counter_dec(&sbi->s_dirs_counter);
155 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
156 err = ext3_journal_dirty_metadata(handle, bh2);
157 if (!fatal) fatal = err;
159 BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata");
160 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
166 ext3_std_error(sb, fatal);
170 * Orlov's allocator for directories.
172 * We always try to spread first-level directories.
174 * If there are blockgroups with both free inodes and free blocks counts
175 * not worse than average we return one with smallest directory count.
176 * Otherwise we simply return a random group.
178 * For the rest rules look so:
180 * It's OK to put directory into a group unless
181 * it has too many directories already (max_dirs) or
182 * it has too few free inodes left (min_inodes) or
183 * it has too few free blocks left (min_blocks) or
184 * it's already running too large debt (max_debt).
185 * Parent's group is preferred, if it doesn't satisfy these
186 * conditions we search cyclically through the rest. If none
187 * of the groups look good we just look for a group with more
188 * free inodes than average (starting at parent's group).
190 * Debt is incremented each time we allocate a directory and decremented
191 * when we allocate an inode, within 0--255.
194 #define INODE_COST 64
195 #define BLOCK_COST 256
197 static int find_group_orlov(struct super_block *sb, struct inode *parent)
199 int parent_group = EXT3_I(parent)->i_block_group;
200 struct ext3_sb_info *sbi = EXT3_SB(sb);
201 struct ext3_super_block *es = sbi->s_es;
202 int ngroups = sbi->s_groups_count;
203 int inodes_per_group = EXT3_INODES_PER_GROUP(sb);
204 unsigned int freei, avefreei;
205 ext3_fsblk_t freeb, avefreeb;
206 ext3_fsblk_t blocks_per_dir;
208 int max_debt, max_dirs, min_inodes;
209 ext3_grpblk_t min_blocks;
211 struct ext3_group_desc *desc;
213 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
214 avefreei = freei / ngroups;
215 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
216 avefreeb = freeb / ngroups;
217 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
219 if ((parent == sb->s_root->d_inode) ||
220 (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) {
221 int best_ndir = inodes_per_group;
224 get_random_bytes(&group, sizeof(group));
225 parent_group = (unsigned)group % ngroups;
226 for (i = 0; i < ngroups; i++) {
227 group = (parent_group + i) % ngroups;
228 desc = ext3_get_group_desc (sb, group, NULL);
229 if (!desc || !desc->bg_free_inodes_count)
231 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
233 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
235 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
238 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
245 blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs;
247 max_dirs = ndirs / ngroups + inodes_per_group / 16;
248 min_inodes = avefreei - inodes_per_group / 4;
249 min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4;
251 max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, (ext3_fsblk_t)BLOCK_COST);
252 if (max_debt * INODE_COST > inodes_per_group)
253 max_debt = inodes_per_group / INODE_COST;
259 for (i = 0; i < ngroups; i++) {
260 group = (parent_group + i) % ngroups;
261 desc = ext3_get_group_desc (sb, group, NULL);
262 if (!desc || !desc->bg_free_inodes_count)
264 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
266 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
268 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
274 for (i = 0; i < ngroups; i++) {
275 group = (parent_group + i) % ngroups;
276 desc = ext3_get_group_desc (sb, group, NULL);
277 if (!desc || !desc->bg_free_inodes_count)
279 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
285 * The free-inodes counter is approximate, and for really small
286 * filesystems the above test can fail to find any blockgroups
295 static int find_group_other(struct super_block *sb, struct inode *parent)
297 int parent_group = EXT3_I(parent)->i_block_group;
298 int ngroups = EXT3_SB(sb)->s_groups_count;
299 struct ext3_group_desc *desc;
303 * Try to place the inode in its parent directory
305 group = parent_group;
306 desc = ext3_get_group_desc (sb, group, NULL);
307 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
308 le16_to_cpu(desc->bg_free_blocks_count))
312 * We're going to place this inode in a different blockgroup from its
313 * parent. We want to cause files in a common directory to all land in
314 * the same blockgroup. But we want files which are in a different
315 * directory which shares a blockgroup with our parent to land in a
316 * different blockgroup.
318 * So add our directory's i_ino into the starting point for the hash.
320 group = (group + parent->i_ino) % ngroups;
323 * Use a quadratic hash to find a group with a free inode and some free
326 for (i = 1; i < ngroups; i <<= 1) {
328 if (group >= ngroups)
330 desc = ext3_get_group_desc (sb, group, NULL);
331 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
332 le16_to_cpu(desc->bg_free_blocks_count))
337 * That failed: try linear search for a free inode, even if that group
338 * has no free blocks.
340 group = parent_group;
341 for (i = 0; i < ngroups; i++) {
342 if (++group >= ngroups)
344 desc = ext3_get_group_desc (sb, group, NULL);
345 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
353 * There are two policies for allocating an inode. If the new inode is
354 * a directory, then a forward search is made for a block group with both
355 * free space and a low directory-to-inode ratio; if that fails, then of
356 * the groups with above-average free space, that group with the fewest
357 * directories already is chosen.
359 * For other inodes, search forward from the parent directory's block
360 * group to find a free inode.
362 struct inode *ext3_new_inode(handle_t *handle, struct inode * dir,
363 const struct qstr *qstr, umode_t mode)
365 struct super_block *sb;
366 struct buffer_head *bitmap_bh = NULL;
367 struct buffer_head *bh2;
369 unsigned long ino = 0;
370 struct inode * inode;
371 struct ext3_group_desc * gdp = NULL;
372 struct ext3_super_block * es;
373 struct ext3_inode_info *ei;
374 struct ext3_sb_info *sbi;
379 /* Cannot create files in a deleted directory */
380 if (!dir || !dir->i_nlink)
381 return ERR_PTR(-EPERM);
384 trace_ext3_request_inode(dir, mode);
385 inode = new_inode(sb);
387 return ERR_PTR(-ENOMEM);
393 group = find_group_orlov(sb, dir);
395 group = find_group_other(sb, dir);
401 for (i = 0; i < sbi->s_groups_count; i++) {
404 gdp = ext3_get_group_desc(sb, group, &bh2);
409 bitmap_bh = read_inode_bitmap(sb, group);
415 repeat_in_this_group:
416 ino = ext3_find_next_zero_bit((unsigned long *)
417 bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino);
418 if (ino < EXT3_INODES_PER_GROUP(sb)) {
420 BUFFER_TRACE(bitmap_bh, "get_write_access");
421 err = ext3_journal_get_write_access(handle, bitmap_bh);
425 if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group),
426 ino, bitmap_bh->b_data)) {
428 BUFFER_TRACE(bitmap_bh,
429 "call ext3_journal_dirty_metadata");
430 err = ext3_journal_dirty_metadata(handle,
437 journal_release_buffer(handle, bitmap_bh);
439 if (++ino < EXT3_INODES_PER_GROUP(sb))
440 goto repeat_in_this_group;
444 * This case is possible in concurrent environment. It is very
445 * rare. We cannot repeat the find_group_xxx() call because
446 * that will simply return the same blockgroup, because the
447 * group descriptor metadata has not yet been updated.
448 * So we just go onto the next blockgroup.
450 if (++group == sbi->s_groups_count)
457 ino += group * EXT3_INODES_PER_GROUP(sb) + 1;
458 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
459 ext3_error (sb, "ext3_new_inode",
460 "reserved inode or inode > inodes count - "
461 "block_group = %d, inode=%lu", group, ino);
466 BUFFER_TRACE(bh2, "get_write_access");
467 err = ext3_journal_get_write_access(handle, bh2);
469 spin_lock(sb_bgl_lock(sbi, group));
470 le16_add_cpu(&gdp->bg_free_inodes_count, -1);
472 le16_add_cpu(&gdp->bg_used_dirs_count, 1);
474 spin_unlock(sb_bgl_lock(sbi, group));
475 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
476 err = ext3_journal_dirty_metadata(handle, bh2);
479 percpu_counter_dec(&sbi->s_freeinodes_counter);
481 percpu_counter_inc(&sbi->s_dirs_counter);
484 if (test_opt(sb, GRPID)) {
485 inode->i_mode = mode;
486 inode->i_uid = current_fsuid();
487 inode->i_gid = dir->i_gid;
489 inode_init_owner(inode, dir, mode);
492 /* This is the optimal IO size (for stat), not the fs block size */
494 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
496 memset(ei->i_data, 0, sizeof(ei->i_data));
497 ei->i_dir_start_lookup = 0;
501 ext3_mask_flags(mode, EXT3_I(dir)->i_flags & EXT3_FL_INHERITED);
502 #ifdef EXT3_FRAGMENTS
510 ei->i_block_alloc_info = NULL;
511 ei->i_block_group = group;
513 ext3_set_inode_flags(inode);
514 if (IS_DIRSYNC(inode))
516 if (insert_inode_locked(inode) < 0) {
518 * Likely a bitmap corruption causing inode to be allocated
524 spin_lock(&sbi->s_next_gen_lock);
525 inode->i_generation = sbi->s_next_generation++;
526 spin_unlock(&sbi->s_next_gen_lock);
528 ei->i_state_flags = 0;
529 ext3_set_inode_state(inode, EXT3_STATE_NEW);
531 /* See comment in ext3_iget for explanation */
532 if (ino >= EXT3_FIRST_INO(sb) + 1 &&
533 EXT3_INODE_SIZE(sb) > EXT3_GOOD_OLD_INODE_SIZE) {
535 sizeof(struct ext3_inode) - EXT3_GOOD_OLD_INODE_SIZE;
537 ei->i_extra_isize = 0;
541 dquot_initialize(inode);
542 err = dquot_alloc_inode(inode);
546 err = ext3_init_acl(handle, inode, dir);
550 err = ext3_init_security(handle, inode, dir, qstr);
554 err = ext3_mark_inode_dirty(handle, inode);
556 ext3_std_error(sb, err);
560 ext3_debug("allocating inode %lu\n", inode->i_ino);
561 trace_ext3_allocate_inode(inode, dir, mode);
564 ext3_std_error(sb, err);
573 dquot_free_inode(inode);
577 inode->i_flags |= S_NOQUOTA;
579 unlock_new_inode(inode);
585 /* Verify that we are loading a valid orphan from disk */
586 struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino)
588 unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count);
589 unsigned long block_group;
591 struct buffer_head *bitmap_bh;
592 struct inode *inode = NULL;
595 /* Error cases - e2fsck has already cleaned up for us */
597 ext3_warning(sb, __func__,
598 "bad orphan ino %lu! e2fsck was run?", ino);
602 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
603 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
604 bitmap_bh = read_inode_bitmap(sb, block_group);
606 ext3_warning(sb, __func__,
607 "inode bitmap error for orphan %lu", ino);
611 /* Having the inode bit set should be a 100% indicator that this
612 * is a valid orphan (no e2fsck run on fs). Orphans also include
613 * inodes that were being truncated, so we can't check i_nlink==0.
615 if (!ext3_test_bit(bit, bitmap_bh->b_data))
618 inode = ext3_iget(sb, ino);
623 * If the orphans has i_nlinks > 0 then it should be able to be
624 * truncated, otherwise it won't be removed from the orphan list
625 * during processing and an infinite loop will result.
627 if (inode->i_nlink && !ext3_can_truncate(inode))
630 if (NEXT_ORPHAN(inode) > max_ino)
636 err = PTR_ERR(inode);
639 ext3_warning(sb, __func__,
640 "bad orphan inode %lu! e2fsck was run?", ino);
641 printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n",
642 bit, (unsigned long long)bitmap_bh->b_blocknr,
643 ext3_test_bit(bit, bitmap_bh->b_data));
644 printk(KERN_NOTICE "inode=%p\n", inode);
646 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
647 is_bad_inode(inode));
648 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
650 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
651 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
652 /* Avoid freeing blocks if we got a bad deleted inode */
653 if (inode->i_nlink == 0)
662 unsigned long ext3_count_free_inodes (struct super_block * sb)
664 unsigned long desc_count;
665 struct ext3_group_desc *gdp;
668 struct ext3_super_block *es;
669 unsigned long bitmap_count, x;
670 struct buffer_head *bitmap_bh = NULL;
672 es = EXT3_SB(sb)->s_es;
676 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
677 gdp = ext3_get_group_desc (sb, i, NULL);
680 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
682 bitmap_bh = read_inode_bitmap(sb, i);
686 x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8);
687 printk("group %d: stored = %d, counted = %lu\n",
688 i, le16_to_cpu(gdp->bg_free_inodes_count), x);
692 printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n",
693 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
697 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
698 gdp = ext3_get_group_desc (sb, i, NULL);
701 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
708 /* Called at mount-time, super-block is locked */
709 unsigned long ext3_count_dirs (struct super_block * sb)
711 unsigned long count = 0;
714 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
715 struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL);
718 count += le16_to_cpu(gdp->bg_used_dirs_count);