2 * linux/fs/ext4/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/jbd2.h>
18 #include <linux/ext4_fs.h>
19 #include <linux/ext4_jbd2.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 ext4_group_desc
*desc
;
57 struct buffer_head
*bh
= NULL
;
59 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
63 bh
= sb_bread(sb
, le32_to_cpu(desc
->bg_inode_bitmap
));
65 ext4_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 ext4_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 ext4_group_desc
* gdp
;
99 struct ext4_super_block
* es
;
100 struct ext4_sb_info
*sbi
;
103 if (atomic_read(&inode
->i_count
) > 1) {
104 printk ("ext4_free_inode: inode has count=%d\n",
105 atomic_read(&inode
->i_count
));
108 if (inode
->i_nlink
) {
109 printk ("ext4_free_inode: inode has nlink=%d\n",
114 printk("ext4_free_inode: inode on nonexistent device\n");
120 ext4_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 ext4_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
= EXT4_SB(sb
)->s_es
;
137 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
138 ext4_error (sb
, "ext4_free_inode",
139 "reserved or nonexistent inode %lu", ino
);
142 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
143 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
144 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
148 BUFFER_TRACE(bitmap_bh
, "get_write_access");
149 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
153 /* Ok, now we can actually update the inode bitmaps.. */
154 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
155 bit
, bitmap_bh
->b_data
))
156 ext4_error (sb
, "ext4_free_inode",
157 "bit already cleared for inode %lu", ino
);
159 gdp
= ext4_get_group_desc (sb
, block_group
, &bh2
);
161 BUFFER_TRACE(bh2
, "get_write_access");
162 fatal
= ext4_journal_get_write_access(handle
, bh2
);
163 if (fatal
) goto error_return
;
166 spin_lock(sb_bgl_lock(sbi
, block_group
));
167 gdp
->bg_free_inodes_count
= cpu_to_le16(
168 le16_to_cpu(gdp
->bg_free_inodes_count
) + 1);
170 gdp
->bg_used_dirs_count
= cpu_to_le16(
171 le16_to_cpu(gdp
->bg_used_dirs_count
) - 1);
172 spin_unlock(sb_bgl_lock(sbi
, block_group
));
173 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
175 percpu_counter_dec(&sbi
->s_dirs_counter
);
178 BUFFER_TRACE(bh2
, "call ext4_journal_dirty_metadata");
179 err
= ext4_journal_dirty_metadata(handle
, bh2
);
180 if (!fatal
) fatal
= err
;
182 BUFFER_TRACE(bitmap_bh
, "call ext4_journal_dirty_metadata");
183 err
= ext4_journal_dirty_metadata(handle
, bitmap_bh
);
189 ext4_std_error(sb
, fatal
);
193 * There are two policies for allocating an inode. If the new inode is
194 * a directory, then a forward search is made for a block group with both
195 * free space and a low directory-to-inode ratio; if that fails, then of
196 * the groups with above-average free space, that group with the fewest
197 * directories already is chosen.
199 * For other inodes, search forward from the parent directory\'s block
200 * group to find a free inode.
202 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
)
204 int ngroups
= EXT4_SB(sb
)->s_groups_count
;
205 unsigned int freei
, avefreei
;
206 struct ext4_group_desc
*desc
, *best_desc
= NULL
;
207 struct buffer_head
*bh
;
208 int group
, best_group
= -1;
210 freei
= percpu_counter_read_positive(&EXT4_SB(sb
)->s_freeinodes_counter
);
211 avefreei
= freei
/ ngroups
;
213 for (group
= 0; group
< ngroups
; group
++) {
214 desc
= ext4_get_group_desc (sb
, group
, &bh
);
215 if (!desc
|| !desc
->bg_free_inodes_count
)
217 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
220 (le16_to_cpu(desc
->bg_free_blocks_count
) >
221 le16_to_cpu(best_desc
->bg_free_blocks_count
))) {
230 * Orlov's allocator for directories.
232 * We always try to spread first-level directories.
234 * If there are blockgroups with both free inodes and free blocks counts
235 * not worse than average we return one with smallest directory count.
236 * Otherwise we simply return a random group.
238 * For the rest rules look so:
240 * It's OK to put directory into a group unless
241 * it has too many directories already (max_dirs) or
242 * it has too few free inodes left (min_inodes) or
243 * it has too few free blocks left (min_blocks) or
244 * it's already running too large debt (max_debt).
245 * Parent's group is prefered, if it doesn't satisfy these
246 * conditions we search cyclically through the rest. If none
247 * of the groups look good we just look for a group with more
248 * free inodes than average (starting at parent's group).
250 * Debt is incremented each time we allocate a directory and decremented
251 * when we allocate an inode, within 0--255.
254 #define INODE_COST 64
255 #define BLOCK_COST 256
257 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
259 int parent_group
= EXT4_I(parent
)->i_block_group
;
260 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
261 struct ext4_super_block
*es
= sbi
->s_es
;
262 int ngroups
= sbi
->s_groups_count
;
263 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
264 unsigned int freei
, avefreei
;
265 ext4_fsblk_t freeb
, avefreeb
;
266 ext4_fsblk_t blocks_per_dir
;
268 int max_debt
, max_dirs
, min_inodes
;
269 ext4_grpblk_t min_blocks
;
271 struct ext4_group_desc
*desc
;
272 struct buffer_head
*bh
;
274 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
275 avefreei
= freei
/ ngroups
;
276 freeb
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
277 avefreeb
= freeb
/ ngroups
;
278 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
280 if ((parent
== sb
->s_root
->d_inode
) ||
281 (EXT4_I(parent
)->i_flags
& EXT4_TOPDIR_FL
)) {
282 int best_ndir
= inodes_per_group
;
285 get_random_bytes(&group
, sizeof(group
));
286 parent_group
= (unsigned)group
% ngroups
;
287 for (i
= 0; i
< ngroups
; i
++) {
288 group
= (parent_group
+ i
) % ngroups
;
289 desc
= ext4_get_group_desc (sb
, group
, &bh
);
290 if (!desc
|| !desc
->bg_free_inodes_count
)
292 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
294 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
296 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
299 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
306 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
) - freeb
) / ndirs
;
308 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
309 min_inodes
= avefreei
- inodes_per_group
/ 4;
310 min_blocks
= avefreeb
- EXT4_BLOCKS_PER_GROUP(sb
) / 4;
312 max_debt
= EXT4_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, (ext4_fsblk_t
)BLOCK_COST
);
313 if (max_debt
* INODE_COST
> inodes_per_group
)
314 max_debt
= inodes_per_group
/ INODE_COST
;
320 for (i
= 0; i
< ngroups
; i
++) {
321 group
= (parent_group
+ i
) % ngroups
;
322 desc
= ext4_get_group_desc (sb
, group
, &bh
);
323 if (!desc
|| !desc
->bg_free_inodes_count
)
325 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
327 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
329 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
335 for (i
= 0; i
< ngroups
; i
++) {
336 group
= (parent_group
+ i
) % ngroups
;
337 desc
= ext4_get_group_desc (sb
, group
, &bh
);
338 if (!desc
|| !desc
->bg_free_inodes_count
)
340 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
346 * The free-inodes counter is approximate, and for really small
347 * filesystems the above test can fail to find any blockgroups
356 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
358 int parent_group
= EXT4_I(parent
)->i_block_group
;
359 int ngroups
= EXT4_SB(sb
)->s_groups_count
;
360 struct ext4_group_desc
*desc
;
361 struct buffer_head
*bh
;
365 * Try to place the inode in its parent directory
367 group
= parent_group
;
368 desc
= ext4_get_group_desc (sb
, group
, &bh
);
369 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
370 le16_to_cpu(desc
->bg_free_blocks_count
))
374 * We're going to place this inode in a different blockgroup from its
375 * parent. We want to cause files in a common directory to all land in
376 * the same blockgroup. But we want files which are in a different
377 * directory which shares a blockgroup with our parent to land in a
378 * different blockgroup.
380 * So add our directory's i_ino into the starting point for the hash.
382 group
= (group
+ parent
->i_ino
) % ngroups
;
385 * Use a quadratic hash to find a group with a free inode and some free
388 for (i
= 1; i
< ngroups
; i
<<= 1) {
390 if (group
>= ngroups
)
392 desc
= ext4_get_group_desc (sb
, group
, &bh
);
393 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
394 le16_to_cpu(desc
->bg_free_blocks_count
))
399 * That failed: try linear search for a free inode, even if that group
400 * has no free blocks.
402 group
= parent_group
;
403 for (i
= 0; i
< ngroups
; i
++) {
404 if (++group
>= ngroups
)
406 desc
= ext4_get_group_desc (sb
, group
, &bh
);
407 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
415 * There are two policies for allocating an inode. If the new inode is
416 * a directory, then a forward search is made for a block group with both
417 * free space and a low directory-to-inode ratio; if that fails, then of
418 * the groups with above-average free space, that group with the fewest
419 * directories already is chosen.
421 * For other inodes, search forward from the parent directory's block
422 * group to find a free inode.
424 struct inode
*ext4_new_inode(handle_t
*handle
, struct inode
* dir
, int mode
)
426 struct super_block
*sb
;
427 struct buffer_head
*bitmap_bh
= NULL
;
428 struct buffer_head
*bh2
;
430 unsigned long ino
= 0;
431 struct inode
* inode
;
432 struct ext4_group_desc
* gdp
= NULL
;
433 struct ext4_super_block
* es
;
434 struct ext4_inode_info
*ei
;
435 struct ext4_sb_info
*sbi
;
440 /* Cannot create files in a deleted directory */
441 if (!dir
|| !dir
->i_nlink
)
442 return ERR_PTR(-EPERM
);
445 inode
= new_inode(sb
);
447 return ERR_PTR(-ENOMEM
);
453 if (test_opt (sb
, OLDALLOC
))
454 group
= find_group_dir(sb
, dir
);
456 group
= find_group_orlov(sb
, dir
);
458 group
= find_group_other(sb
, dir
);
464 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
467 gdp
= ext4_get_group_desc(sb
, group
, &bh2
);
472 bitmap_bh
= read_inode_bitmap(sb
, group
);
478 repeat_in_this_group
:
479 ino
= ext4_find_next_zero_bit((unsigned long *)
480 bitmap_bh
->b_data
, EXT4_INODES_PER_GROUP(sb
), ino
);
481 if (ino
< EXT4_INODES_PER_GROUP(sb
)) {
483 BUFFER_TRACE(bitmap_bh
, "get_write_access");
484 err
= ext4_journal_get_write_access(handle
, bitmap_bh
);
488 if (!ext4_set_bit_atomic(sb_bgl_lock(sbi
, group
),
489 ino
, bitmap_bh
->b_data
)) {
491 BUFFER_TRACE(bitmap_bh
,
492 "call ext4_journal_dirty_metadata");
493 err
= ext4_journal_dirty_metadata(handle
,
500 jbd2_journal_release_buffer(handle
, bitmap_bh
);
502 if (++ino
< EXT4_INODES_PER_GROUP(sb
))
503 goto repeat_in_this_group
;
507 * This case is possible in concurrent environment. It is very
508 * rare. We cannot repeat the find_group_xxx() call because
509 * that will simply return the same blockgroup, because the
510 * group descriptor metadata has not yet been updated.
511 * So we just go onto the next blockgroup.
513 if (++group
== sbi
->s_groups_count
)
520 ino
+= group
* EXT4_INODES_PER_GROUP(sb
) + 1;
521 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
522 ext4_error (sb
, "ext4_new_inode",
523 "reserved inode or inode > inodes count - "
524 "block_group = %d, inode=%lu", group
, ino
);
529 BUFFER_TRACE(bh2
, "get_write_access");
530 err
= ext4_journal_get_write_access(handle
, bh2
);
532 spin_lock(sb_bgl_lock(sbi
, group
));
533 gdp
->bg_free_inodes_count
=
534 cpu_to_le16(le16_to_cpu(gdp
->bg_free_inodes_count
) - 1);
536 gdp
->bg_used_dirs_count
=
537 cpu_to_le16(le16_to_cpu(gdp
->bg_used_dirs_count
) + 1);
539 spin_unlock(sb_bgl_lock(sbi
, group
));
540 BUFFER_TRACE(bh2
, "call ext4_journal_dirty_metadata");
541 err
= ext4_journal_dirty_metadata(handle
, bh2
);
544 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
546 percpu_counter_inc(&sbi
->s_dirs_counter
);
549 inode
->i_uid
= current
->fsuid
;
550 if (test_opt (sb
, GRPID
))
551 inode
->i_gid
= dir
->i_gid
;
552 else if (dir
->i_mode
& S_ISGID
) {
553 inode
->i_gid
= dir
->i_gid
;
557 inode
->i_gid
= current
->fsgid
;
558 inode
->i_mode
= mode
;
561 /* This is the optimal IO size (for stat), not the fs block 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
= EXT4_I(dir
)->i_flags
& ~EXT4_INDEX_FL
;
571 ei
->i_flags
&= ~(EXT4_IMMUTABLE_FL
|EXT4_APPEND_FL
);
572 /* dirsync only applies to directories */
574 ei
->i_flags
&= ~EXT4_DIRSYNC_FL
;
575 #ifdef EXT4_FRAGMENTS
583 ei
->i_block_alloc_info
= NULL
;
584 ei
->i_block_group
= group
;
586 ext4_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
= EXT4_STATE_NEW
;
596 (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) ?
597 sizeof(struct ext4_inode
) - EXT4_GOOD_OLD_INODE_SIZE
: 0;
600 if(DQUOT_ALLOC_INODE(inode
)) {
605 err
= ext4_init_acl(handle
, inode
, dir
);
609 err
= ext4_init_security(handle
,inode
, dir
);
613 err
= ext4_mark_inode_dirty(handle
, inode
);
615 ext4_std_error(sb
, err
);
619 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
622 ext4_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
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
645 unsigned long max_ino
= le32_to_cpu(EXT4_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 ext4_warning(sb
, __FUNCTION__
,
654 "bad orphan ino %lu! e2fsck was run?", ino
);
658 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
659 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
660 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
662 ext4_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 (!ext4_test_bit(bit
, bitmap_bh
->b_data
) ||
672 !(inode
= iget(sb
, ino
)) || is_bad_inode(inode
) ||
673 NEXT_ORPHAN(inode
) > max_ino
) {
674 ext4_warning(sb
, __FUNCTION__
,
675 "bad orphan inode %lu! e2fsck was run?", ino
);
676 printk(KERN_NOTICE
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
677 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
678 ext4_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 ext4_count_free_inodes (struct super_block
* sb
)
700 unsigned long desc_count
;
701 struct ext4_group_desc
*gdp
;
704 struct ext4_super_block
*es
;
705 unsigned long bitmap_count
, x
;
706 struct buffer_head
*bitmap_bh
= NULL
;
708 es
= EXT4_SB(sb
)->s_es
;
712 for (i
= 0; i
< EXT4_SB(sb
)->s_groups_count
; i
++) {
713 gdp
= ext4_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
= ext4_count_free(bitmap_bh
, EXT4_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("ext4_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
< EXT4_SB(sb
)->s_groups_count
; i
++) {
734 gdp
= ext4_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 ext4_count_dirs (struct super_block
* sb
)
747 unsigned long count
= 0;
750 for (i
= 0; i
< EXT4_SB(sb
)->s_groups_count
; i
++) {
751 struct ext4_group_desc
*gdp
= ext4_get_group_desc (sb
, i
, NULL
);
754 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);