2 * linux/fs/ext2/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@dcs.ed.ac.uk), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
15 #include <linux/quotaops.h>
16 #include <linux/sched.h>
17 #include <linux/backing-dev.h>
18 #include <linux/buffer_head.h>
19 #include <linux/random.h>
25 * ialloc.c contains the inodes allocation and deallocation routines
29 * The free inodes are managed by bitmaps. A file system contains several
30 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
31 * block for inodes, N blocks for the inode table and data blocks.
33 * The file system contains group descriptors which are located after the
34 * super block. Each descriptor contains the number of the bitmap block and
35 * the free blocks count in the block.
40 * Read the inode allocation bitmap for a given block_group, reading
41 * into the specified slot in the superblock's bitmap cache.
43 * Return buffer_head of bitmap on success or NULL.
45 static struct buffer_head
*
46 read_inode_bitmap(struct super_block
* sb
, unsigned long block_group
)
48 struct ext2_group_desc
*desc
;
49 struct buffer_head
*bh
= NULL
;
51 desc
= ext2_get_group_desc(sb
, block_group
, NULL
);
55 bh
= sb_bread(sb
, le32_to_cpu(desc
->bg_inode_bitmap
));
57 ext2_error(sb
, "read_inode_bitmap",
58 "Cannot read inode bitmap - "
59 "block_group = %lu, inode_bitmap = %u",
60 block_group
, le32_to_cpu(desc
->bg_inode_bitmap
));
65 static void ext2_release_inode(struct super_block
*sb
, int group
, int dir
)
67 struct ext2_group_desc
* desc
;
68 struct buffer_head
*bh
;
70 desc
= ext2_get_group_desc(sb
, group
, &bh
);
72 ext2_error(sb
, "ext2_release_inode",
73 "can't get descriptor for group %d", group
);
77 spin_lock(sb_bgl_lock(EXT2_SB(sb
), group
));
78 le16_add_cpu(&desc
->bg_free_inodes_count
, 1);
80 le16_add_cpu(&desc
->bg_used_dirs_count
, -1);
81 spin_unlock(sb_bgl_lock(EXT2_SB(sb
), group
));
83 percpu_counter_dec(&EXT2_SB(sb
)->s_dirs_counter
);
85 mark_buffer_dirty(bh
);
89 * NOTE! When we get the inode, we're the only people
90 * that have access to it, and as such there are no
91 * race conditions we have to worry about. The inode
92 * is not on the hash-lists, and it cannot be reached
93 * through the filesystem because the directory entry
94 * has been deleted earlier.
96 * HOWEVER: we must make sure that we get no aliases,
97 * which means that we have to call "clear_inode()"
98 * _before_ we mark the inode not in use in the inode
99 * bitmaps. Otherwise a newly created file might use
100 * the same inode number (not actually the same pointer
101 * though), and then we'd have two inodes sharing the
102 * same inode number and space on the harddisk.
104 void ext2_free_inode (struct inode
* inode
)
106 struct super_block
* sb
= inode
->i_sb
;
109 struct buffer_head
*bitmap_bh
= NULL
;
110 unsigned long block_group
;
112 struct ext2_super_block
* es
;
115 ext2_debug ("freeing inode %lu\n", ino
);
118 * Note: we must free any quota before locking the superblock,
119 * as writing the quota to disk may need the lock as well.
121 if (!is_bad_inode(inode
)) {
122 /* Quota is already initialized in iput() */
123 ext2_xattr_delete_inode(inode
);
124 DQUOT_FREE_INODE(inode
);
128 es
= EXT2_SB(sb
)->s_es
;
129 is_directory
= S_ISDIR(inode
->i_mode
);
131 /* Do this BEFORE marking the inode not in use or returning an error */
134 if (ino
< EXT2_FIRST_INO(sb
) ||
135 ino
> le32_to_cpu(es
->s_inodes_count
)) {
136 ext2_error (sb
, "ext2_free_inode",
137 "reserved or nonexistent inode %lu", ino
);
140 block_group
= (ino
- 1) / EXT2_INODES_PER_GROUP(sb
);
141 bit
= (ino
- 1) % EXT2_INODES_PER_GROUP(sb
);
143 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
147 /* Ok, now we can actually update the inode bitmaps.. */
148 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), block_group
),
149 bit
, (void *) bitmap_bh
->b_data
))
150 ext2_error (sb
, "ext2_free_inode",
151 "bit already cleared for inode %lu", ino
);
153 ext2_release_inode(sb
, block_group
, is_directory
);
154 mark_buffer_dirty(bitmap_bh
);
155 if (sb
->s_flags
& MS_SYNCHRONOUS
)
156 sync_dirty_buffer(bitmap_bh
);
162 * We perform asynchronous prereading of the new inode's inode block when
163 * we create the inode, in the expectation that the inode will be written
164 * back soon. There are two reasons:
166 * - When creating a large number of files, the async prereads will be
167 * nicely merged into large reads
168 * - When writing out a large number of inodes, we don't need to keep on
169 * stalling the writes while we read the inode block.
171 * FIXME: ext2_get_group_desc() needs to be simplified.
173 static void ext2_preread_inode(struct inode
*inode
)
175 unsigned long block_group
;
176 unsigned long offset
;
178 struct ext2_group_desc
* gdp
;
179 struct backing_dev_info
*bdi
;
181 bdi
= inode
->i_mapping
->backing_dev_info
;
182 if (bdi_read_congested(bdi
))
184 if (bdi_write_congested(bdi
))
187 block_group
= (inode
->i_ino
- 1) / EXT2_INODES_PER_GROUP(inode
->i_sb
);
188 gdp
= ext2_get_group_desc(inode
->i_sb
, block_group
, NULL
);
193 * Figure out the offset within the block group inode table
195 offset
= ((inode
->i_ino
- 1) % EXT2_INODES_PER_GROUP(inode
->i_sb
)) *
196 EXT2_INODE_SIZE(inode
->i_sb
);
197 block
= le32_to_cpu(gdp
->bg_inode_table
) +
198 (offset
>> EXT2_BLOCK_SIZE_BITS(inode
->i_sb
));
199 sb_breadahead(inode
->i_sb
, block
);
203 * There are two policies for allocating an inode. If the new inode is
204 * a directory, then a forward search is made for a block group with both
205 * free space and a low directory-to-inode ratio; if that fails, then of
206 * the groups with above-average free space, that group with the fewest
207 * directories already is chosen.
209 * For other inodes, search forward from the parent directory\'s block
210 * group to find a free inode.
212 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
)
214 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
215 int avefreei
= ext2_count_free_inodes(sb
) / ngroups
;
216 struct ext2_group_desc
*desc
, *best_desc
= NULL
;
217 int group
, best_group
= -1;
219 for (group
= 0; group
< ngroups
; group
++) {
220 desc
= ext2_get_group_desc (sb
, group
, NULL
);
221 if (!desc
|| !desc
->bg_free_inodes_count
)
223 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
226 (le16_to_cpu(desc
->bg_free_blocks_count
) >
227 le16_to_cpu(best_desc
->bg_free_blocks_count
))) {
239 * Orlov's allocator for directories.
241 * We always try to spread first-level directories.
243 * If there are blockgroups with both free inodes and free blocks counts
244 * not worse than average we return one with smallest directory count.
245 * Otherwise we simply return a random group.
247 * For the rest rules look so:
249 * It's OK to put directory into a group unless
250 * it has too many directories already (max_dirs) or
251 * it has too few free inodes left (min_inodes) or
252 * it has too few free blocks left (min_blocks) or
253 * it's already running too large debt (max_debt).
254 * Parent's group is preferred, if it doesn't satisfy these
255 * conditions we search cyclically through the rest. If none
256 * of the groups look good we just look for a group with more
257 * free inodes than average (starting at parent's group).
259 * Debt is incremented each time we allocate a directory and decremented
260 * when we allocate an inode, within 0--255.
263 #define INODE_COST 64
264 #define BLOCK_COST 256
266 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
268 int parent_group
= EXT2_I(parent
)->i_block_group
;
269 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
270 struct ext2_super_block
*es
= sbi
->s_es
;
271 int ngroups
= sbi
->s_groups_count
;
272 int inodes_per_group
= EXT2_INODES_PER_GROUP(sb
);
279 int max_debt
, max_dirs
, min_blocks
, min_inodes
;
281 struct ext2_group_desc
*desc
;
283 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
284 avefreei
= freei
/ ngroups
;
285 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
286 avefreeb
= free_blocks
/ ngroups
;
287 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
289 if ((parent
== sb
->s_root
->d_inode
) ||
290 (EXT2_I(parent
)->i_flags
& EXT2_TOPDIR_FL
)) {
291 struct ext2_group_desc
*best_desc
= NULL
;
292 int best_ndir
= inodes_per_group
;
295 get_random_bytes(&group
, sizeof(group
));
296 parent_group
= (unsigned)group
% ngroups
;
297 for (i
= 0; i
< ngroups
; i
++) {
298 group
= (parent_group
+ i
) % ngroups
;
299 desc
= ext2_get_group_desc (sb
, group
, NULL
);
300 if (!desc
|| !desc
->bg_free_inodes_count
)
302 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
304 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
306 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
309 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
312 if (best_group
>= 0) {
321 ndirs
= 1; /* percpu_counters are approximate... */
323 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
)-free_blocks
) / ndirs
;
325 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
326 min_inodes
= avefreei
- inodes_per_group
/ 4;
327 min_blocks
= avefreeb
- EXT2_BLOCKS_PER_GROUP(sb
) / 4;
329 max_debt
= EXT2_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, BLOCK_COST
);
330 if (max_debt
* INODE_COST
> inodes_per_group
)
331 max_debt
= inodes_per_group
/ INODE_COST
;
337 for (i
= 0; i
< ngroups
; i
++) {
338 group
= (parent_group
+ i
) % ngroups
;
339 desc
= ext2_get_group_desc (sb
, group
, NULL
);
340 if (!desc
|| !desc
->bg_free_inodes_count
)
342 if (sbi
->s_debts
[group
] >= max_debt
)
344 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
346 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
348 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
354 for (i
= 0; i
< ngroups
; i
++) {
355 group
= (parent_group
+ i
) % ngroups
;
356 desc
= ext2_get_group_desc (sb
, group
, NULL
);
357 if (!desc
|| !desc
->bg_free_inodes_count
)
359 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
365 * The free-inodes counter is approximate, and for really small
366 * filesystems the above test can fail to find any blockgroups
378 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
380 int parent_group
= EXT2_I(parent
)->i_block_group
;
381 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
382 struct ext2_group_desc
*desc
;
386 * Try to place the inode in its parent directory
388 group
= parent_group
;
389 desc
= ext2_get_group_desc (sb
, group
, NULL
);
390 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
391 le16_to_cpu(desc
->bg_free_blocks_count
))
395 * We're going to place this inode in a different blockgroup from its
396 * parent. We want to cause files in a common directory to all land in
397 * the same blockgroup. But we want files which are in a different
398 * directory which shares a blockgroup with our parent to land in a
399 * different blockgroup.
401 * So add our directory's i_ino into the starting point for the hash.
403 group
= (group
+ parent
->i_ino
) % ngroups
;
406 * Use a quadratic hash to find a group with a free inode and some
409 for (i
= 1; i
< ngroups
; i
<<= 1) {
411 if (group
>= ngroups
)
413 desc
= ext2_get_group_desc (sb
, group
, NULL
);
414 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
415 le16_to_cpu(desc
->bg_free_blocks_count
))
420 * That failed: try linear search for a free inode, even if that group
421 * has no free blocks.
423 group
= parent_group
;
424 for (i
= 0; i
< ngroups
; i
++) {
425 if (++group
>= ngroups
)
427 desc
= ext2_get_group_desc (sb
, group
, NULL
);
428 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
438 struct inode
*ext2_new_inode(struct inode
*dir
, int mode
)
440 struct super_block
*sb
;
441 struct buffer_head
*bitmap_bh
= NULL
;
442 struct buffer_head
*bh2
;
445 struct inode
* inode
;
446 struct ext2_group_desc
*gdp
;
447 struct ext2_super_block
*es
;
448 struct ext2_inode_info
*ei
;
449 struct ext2_sb_info
*sbi
;
453 inode
= new_inode(sb
);
455 return ERR_PTR(-ENOMEM
);
461 if (test_opt(sb
, OLDALLOC
))
462 group
= find_group_dir(sb
, dir
);
464 group
= find_group_orlov(sb
, dir
);
466 group
= find_group_other(sb
, dir
);
473 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
474 gdp
= ext2_get_group_desc(sb
, group
, &bh2
);
476 bitmap_bh
= read_inode_bitmap(sb
, group
);
483 repeat_in_this_group
:
484 ino
= ext2_find_next_zero_bit((unsigned long *)bitmap_bh
->b_data
,
485 EXT2_INODES_PER_GROUP(sb
), ino
);
486 if (ino
>= EXT2_INODES_PER_GROUP(sb
)) {
488 * Rare race: find_group_xx() decided that there were
489 * free inodes in this group, but by the time we tried
490 * to allocate one, they're all gone. This can also
491 * occur because the counters which find_group_orlov()
492 * uses are approximate. So just go and search the
495 if (++group
== sbi
->s_groups_count
)
499 if (ext2_set_bit_atomic(sb_bgl_lock(sbi
, group
),
500 ino
, bitmap_bh
->b_data
)) {
501 /* we lost this inode */
502 if (++ino
>= EXT2_INODES_PER_GROUP(sb
)) {
503 /* this group is exhausted, try next group */
504 if (++group
== sbi
->s_groups_count
)
508 /* try to find free inode in the same group */
509 goto repeat_in_this_group
;
515 * Scanned all blockgroups.
520 mark_buffer_dirty(bitmap_bh
);
521 if (sb
->s_flags
& MS_SYNCHRONOUS
)
522 sync_dirty_buffer(bitmap_bh
);
525 ino
+= group
* EXT2_INODES_PER_GROUP(sb
) + 1;
526 if (ino
< EXT2_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
527 ext2_error (sb
, "ext2_new_inode",
528 "reserved inode or inode > inodes count - "
529 "block_group = %d,inode=%lu", group
,
530 (unsigned long) ino
);
535 percpu_counter_add(&sbi
->s_freeinodes_counter
, -1);
537 percpu_counter_inc(&sbi
->s_dirs_counter
);
539 spin_lock(sb_bgl_lock(sbi
, group
));
540 le16_add_cpu(&gdp
->bg_free_inodes_count
, -1);
542 if (sbi
->s_debts
[group
] < 255)
543 sbi
->s_debts
[group
]++;
544 le16_add_cpu(&gdp
->bg_used_dirs_count
, 1);
546 if (sbi
->s_debts
[group
])
547 sbi
->s_debts
[group
]--;
549 spin_unlock(sb_bgl_lock(sbi
, group
));
552 mark_buffer_dirty(bh2
);
553 inode
->i_uid
= current_fsuid();
554 if (test_opt (sb
, GRPID
))
555 inode
->i_gid
= dir
->i_gid
;
556 else if (dir
->i_mode
& S_ISGID
) {
557 inode
->i_gid
= dir
->i_gid
;
561 inode
->i_gid
= current_fsgid();
562 inode
->i_mode
= mode
;
566 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
567 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
568 ei
->i_flags
= EXT2_I(dir
)->i_flags
& ~EXT2_BTREE_FL
;
570 ei
->i_flags
&= ~(EXT2_IMMUTABLE_FL
|EXT2_APPEND_FL
);
571 /* dirsync is only applied to directories */
573 ei
->i_flags
&= ~EXT2_DIRSYNC_FL
;
580 ei
->i_block_alloc_info
= NULL
;
581 ei
->i_block_group
= group
;
582 ei
->i_dir_start_lookup
= 0;
583 ei
->i_state
= EXT2_STATE_NEW
;
584 ext2_set_inode_flags(inode
);
585 spin_lock(&sbi
->s_next_gen_lock
);
586 inode
->i_generation
= sbi
->s_next_generation
++;
587 spin_unlock(&sbi
->s_next_gen_lock
);
588 if (insert_inode_locked(inode
) < 0) {
593 if (DQUOT_ALLOC_INODE(inode
)) {
598 err
= ext2_init_acl(inode
, dir
);
602 err
= ext2_init_security(inode
,dir
);
606 mark_inode_dirty(inode
);
607 ext2_debug("allocating inode %lu\n", inode
->i_ino
);
608 ext2_preread_inode(inode
);
612 DQUOT_FREE_INODE(inode
);
616 inode
->i_flags
|= S_NOQUOTA
;
618 unlock_new_inode(inode
);
623 make_bad_inode(inode
);
628 unsigned long ext2_count_free_inodes (struct super_block
* sb
)
630 struct ext2_group_desc
*desc
;
631 unsigned long desc_count
= 0;
635 struct ext2_super_block
*es
;
636 unsigned long bitmap_count
= 0;
637 struct buffer_head
*bitmap_bh
= NULL
;
639 es
= EXT2_SB(sb
)->s_es
;
640 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
643 desc
= ext2_get_group_desc (sb
, i
, NULL
);
646 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
648 bitmap_bh
= read_inode_bitmap(sb
, i
);
652 x
= ext2_count_free(bitmap_bh
, EXT2_INODES_PER_GROUP(sb
) / 8);
653 printk("group %d: stored = %d, counted = %u\n",
654 i
, le16_to_cpu(desc
->bg_free_inodes_count
), x
);
658 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
659 percpu_counter_read(&EXT2_SB(sb
)->s_freeinodes_counter
),
660 desc_count
, bitmap_count
);
663 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
664 desc
= ext2_get_group_desc (sb
, i
, NULL
);
667 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
673 /* Called at mount-time, super-block is locked */
674 unsigned long ext2_count_dirs (struct super_block
* sb
)
676 unsigned long count
= 0;
679 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
680 struct ext2_group_desc
*gdp
= ext2_get_group_desc (sb
, i
, NULL
);
683 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);