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
;
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 /* Quota is already initialized in iput() */
122 ext2_xattr_delete_inode(inode
);
123 dquot_free_inode(inode
);
126 es
= EXT2_SB(sb
)->s_es
;
127 is_directory
= S_ISDIR(inode
->i_mode
);
129 if (ino
< EXT2_FIRST_INO(sb
) ||
130 ino
> le32_to_cpu(es
->s_inodes_count
)) {
131 ext2_error (sb
, "ext2_free_inode",
132 "reserved or nonexistent inode %lu", ino
);
135 block_group
= (ino
- 1) / EXT2_INODES_PER_GROUP(sb
);
136 bit
= (ino
- 1) % EXT2_INODES_PER_GROUP(sb
);
137 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
141 /* Ok, now we can actually update the inode bitmaps.. */
142 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), block_group
),
143 bit
, (void *) bitmap_bh
->b_data
))
144 ext2_error (sb
, "ext2_free_inode",
145 "bit already cleared for inode %lu", ino
);
147 ext2_release_inode(sb
, block_group
, is_directory
);
148 mark_buffer_dirty(bitmap_bh
);
149 if (sb
->s_flags
& MS_SYNCHRONOUS
)
150 sync_dirty_buffer(bitmap_bh
);
156 * We perform asynchronous prereading of the new inode's inode block when
157 * we create the inode, in the expectation that the inode will be written
158 * back soon. There are two reasons:
160 * - When creating a large number of files, the async prereads will be
161 * nicely merged into large reads
162 * - When writing out a large number of inodes, we don't need to keep on
163 * stalling the writes while we read the inode block.
165 * FIXME: ext2_get_group_desc() needs to be simplified.
167 static void ext2_preread_inode(struct inode
*inode
)
169 unsigned long block_group
;
170 unsigned long offset
;
172 struct ext2_group_desc
* gdp
;
173 struct backing_dev_info
*bdi
;
175 bdi
= inode
->i_mapping
->backing_dev_info
;
176 if (bdi_read_congested(bdi
))
178 if (bdi_write_congested(bdi
))
181 block_group
= (inode
->i_ino
- 1) / EXT2_INODES_PER_GROUP(inode
->i_sb
);
182 gdp
= ext2_get_group_desc(inode
->i_sb
, block_group
, NULL
);
187 * Figure out the offset within the block group inode table
189 offset
= ((inode
->i_ino
- 1) % EXT2_INODES_PER_GROUP(inode
->i_sb
)) *
190 EXT2_INODE_SIZE(inode
->i_sb
);
191 block
= le32_to_cpu(gdp
->bg_inode_table
) +
192 (offset
>> EXT2_BLOCK_SIZE_BITS(inode
->i_sb
));
193 sb_breadahead(inode
->i_sb
, block
);
197 * There are two policies for allocating an inode. If the new inode is
198 * a directory, then a forward search is made for a block group with both
199 * free space and a low directory-to-inode ratio; if that fails, then of
200 * the groups with above-average free space, that group with the fewest
201 * directories already is chosen.
203 * For other inodes, search forward from the parent directory\'s block
204 * group to find a free inode.
206 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
)
208 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
209 int avefreei
= ext2_count_free_inodes(sb
) / ngroups
;
210 struct ext2_group_desc
*desc
, *best_desc
= NULL
;
211 int group
, best_group
= -1;
213 for (group
= 0; group
< ngroups
; group
++) {
214 desc
= ext2_get_group_desc (sb
, group
, NULL
);
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
))) {
233 * Orlov's allocator for directories.
235 * We always try to spread first-level directories.
237 * If there are blockgroups with both free inodes and free blocks counts
238 * not worse than average we return one with smallest directory count.
239 * Otherwise we simply return a random group.
241 * For the rest rules look so:
243 * It's OK to put directory into a group unless
244 * it has too many directories already (max_dirs) or
245 * it has too few free inodes left (min_inodes) or
246 * it has too few free blocks left (min_blocks) or
247 * it's already running too large debt (max_debt).
248 * Parent's group is preferred, if it doesn't satisfy these
249 * conditions we search cyclically through the rest. If none
250 * of the groups look good we just look for a group with more
251 * free inodes than average (starting at parent's group).
253 * Debt is incremented each time we allocate a directory and decremented
254 * when we allocate an inode, within 0--255.
257 #define INODE_COST 64
258 #define BLOCK_COST 256
260 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
262 int parent_group
= EXT2_I(parent
)->i_block_group
;
263 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
264 struct ext2_super_block
*es
= sbi
->s_es
;
265 int ngroups
= sbi
->s_groups_count
;
266 int inodes_per_group
= EXT2_INODES_PER_GROUP(sb
);
273 int max_debt
, max_dirs
, min_blocks
, min_inodes
;
275 struct ext2_group_desc
*desc
;
277 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
278 avefreei
= freei
/ ngroups
;
279 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
280 avefreeb
= free_blocks
/ ngroups
;
281 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
283 if ((parent
== sb
->s_root
->d_inode
) ||
284 (EXT2_I(parent
)->i_flags
& EXT2_TOPDIR_FL
)) {
285 struct ext2_group_desc
*best_desc
= NULL
;
286 int best_ndir
= inodes_per_group
;
289 get_random_bytes(&group
, sizeof(group
));
290 parent_group
= (unsigned)group
% ngroups
;
291 for (i
= 0; i
< ngroups
; i
++) {
292 group
= (parent_group
+ i
) % ngroups
;
293 desc
= ext2_get_group_desc (sb
, group
, NULL
);
294 if (!desc
|| !desc
->bg_free_inodes_count
)
296 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
298 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
300 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
303 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
306 if (best_group
>= 0) {
315 ndirs
= 1; /* percpu_counters are approximate... */
317 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
)-free_blocks
) / ndirs
;
319 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
320 min_inodes
= avefreei
- inodes_per_group
/ 4;
321 min_blocks
= avefreeb
- EXT2_BLOCKS_PER_GROUP(sb
) / 4;
323 max_debt
= EXT2_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, BLOCK_COST
);
324 if (max_debt
* INODE_COST
> inodes_per_group
)
325 max_debt
= inodes_per_group
/ INODE_COST
;
331 for (i
= 0; i
< ngroups
; i
++) {
332 group
= (parent_group
+ i
) % ngroups
;
333 desc
= ext2_get_group_desc (sb
, group
, NULL
);
334 if (!desc
|| !desc
->bg_free_inodes_count
)
336 if (sbi
->s_debts
[group
] >= max_debt
)
338 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
340 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
342 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
348 for (i
= 0; i
< ngroups
; i
++) {
349 group
= (parent_group
+ i
) % ngroups
;
350 desc
= ext2_get_group_desc (sb
, group
, NULL
);
351 if (!desc
|| !desc
->bg_free_inodes_count
)
353 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
359 * The free-inodes counter is approximate, and for really small
360 * filesystems the above test can fail to find any blockgroups
372 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
374 int parent_group
= EXT2_I(parent
)->i_block_group
;
375 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
376 struct ext2_group_desc
*desc
;
380 * Try to place the inode in its parent directory
382 group
= parent_group
;
383 desc
= ext2_get_group_desc (sb
, group
, NULL
);
384 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
385 le16_to_cpu(desc
->bg_free_blocks_count
))
389 * We're going to place this inode in a different blockgroup from its
390 * parent. We want to cause files in a common directory to all land in
391 * the same blockgroup. But we want files which are in a different
392 * directory which shares a blockgroup with our parent to land in a
393 * different blockgroup.
395 * So add our directory's i_ino into the starting point for the hash.
397 group
= (group
+ parent
->i_ino
) % ngroups
;
400 * Use a quadratic hash to find a group with a free inode and some
403 for (i
= 1; i
< ngroups
; i
<<= 1) {
405 if (group
>= ngroups
)
407 desc
= ext2_get_group_desc (sb
, group
, NULL
);
408 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
409 le16_to_cpu(desc
->bg_free_blocks_count
))
414 * That failed: try linear search for a free inode, even if that group
415 * has no free blocks.
417 group
= parent_group
;
418 for (i
= 0; i
< ngroups
; i
++) {
419 if (++group
>= ngroups
)
421 desc
= ext2_get_group_desc (sb
, group
, NULL
);
422 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
432 struct inode
*ext2_new_inode(struct inode
*dir
, umode_t mode
,
433 const struct qstr
*qstr
)
435 struct super_block
*sb
;
436 struct buffer_head
*bitmap_bh
= NULL
;
437 struct buffer_head
*bh2
;
440 struct inode
* inode
;
441 struct ext2_group_desc
*gdp
;
442 struct ext2_super_block
*es
;
443 struct ext2_inode_info
*ei
;
444 struct ext2_sb_info
*sbi
;
448 inode
= new_inode(sb
);
450 return ERR_PTR(-ENOMEM
);
456 if (test_opt(sb
, OLDALLOC
))
457 group
= find_group_dir(sb
, dir
);
459 group
= find_group_orlov(sb
, dir
);
461 group
= find_group_other(sb
, dir
);
468 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
469 gdp
= ext2_get_group_desc(sb
, group
, &bh2
);
471 bitmap_bh
= read_inode_bitmap(sb
, group
);
478 repeat_in_this_group
:
479 ino
= ext2_find_next_zero_bit((unsigned long *)bitmap_bh
->b_data
,
480 EXT2_INODES_PER_GROUP(sb
), ino
);
481 if (ino
>= EXT2_INODES_PER_GROUP(sb
)) {
483 * Rare race: find_group_xx() decided that there were
484 * free inodes in this group, but by the time we tried
485 * to allocate one, they're all gone. This can also
486 * occur because the counters which find_group_orlov()
487 * uses are approximate. So just go and search the
490 if (++group
== sbi
->s_groups_count
)
494 if (ext2_set_bit_atomic(sb_bgl_lock(sbi
, group
),
495 ino
, bitmap_bh
->b_data
)) {
496 /* we lost this inode */
497 if (++ino
>= EXT2_INODES_PER_GROUP(sb
)) {
498 /* this group is exhausted, try next group */
499 if (++group
== sbi
->s_groups_count
)
503 /* try to find free inode in the same group */
504 goto repeat_in_this_group
;
510 * Scanned all blockgroups.
515 mark_buffer_dirty(bitmap_bh
);
516 if (sb
->s_flags
& MS_SYNCHRONOUS
)
517 sync_dirty_buffer(bitmap_bh
);
520 ino
+= group
* EXT2_INODES_PER_GROUP(sb
) + 1;
521 if (ino
< EXT2_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
522 ext2_error (sb
, "ext2_new_inode",
523 "reserved inode or inode > inodes count - "
524 "block_group = %d,inode=%lu", group
,
525 (unsigned long) ino
);
530 percpu_counter_add(&sbi
->s_freeinodes_counter
, -1);
532 percpu_counter_inc(&sbi
->s_dirs_counter
);
534 spin_lock(sb_bgl_lock(sbi
, group
));
535 le16_add_cpu(&gdp
->bg_free_inodes_count
, -1);
537 if (sbi
->s_debts
[group
] < 255)
538 sbi
->s_debts
[group
]++;
539 le16_add_cpu(&gdp
->bg_used_dirs_count
, 1);
541 if (sbi
->s_debts
[group
])
542 sbi
->s_debts
[group
]--;
544 spin_unlock(sb_bgl_lock(sbi
, group
));
547 mark_buffer_dirty(bh2
);
548 if (test_opt(sb
, GRPID
)) {
549 inode
->i_mode
= mode
;
550 inode
->i_uid
= current_fsuid();
551 inode
->i_gid
= dir
->i_gid
;
553 inode_init_owner(inode
, dir
, mode
);
557 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
558 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
560 ext2_mask_flags(mode
, EXT2_I(dir
)->i_flags
& EXT2_FL_INHERITED
);
567 ei
->i_block_alloc_info
= NULL
;
568 ei
->i_block_group
= group
;
569 ei
->i_dir_start_lookup
= 0;
570 ei
->i_state
= EXT2_STATE_NEW
;
571 ext2_set_inode_flags(inode
);
572 spin_lock(&sbi
->s_next_gen_lock
);
573 inode
->i_generation
= sbi
->s_next_generation
++;
574 spin_unlock(&sbi
->s_next_gen_lock
);
575 if (insert_inode_locked(inode
) < 0) {
576 ext2_error(sb
, "ext2_new_inode",
577 "inode number already in use - inode=%lu",
578 (unsigned long) ino
);
583 dquot_initialize(inode
);
584 err
= dquot_alloc_inode(inode
);
588 err
= ext2_init_acl(inode
, dir
);
592 err
= ext2_init_security(inode
, dir
, qstr
);
596 mark_inode_dirty(inode
);
597 ext2_debug("allocating inode %lu\n", inode
->i_ino
);
598 ext2_preread_inode(inode
);
602 dquot_free_inode(inode
);
606 inode
->i_flags
|= S_NOQUOTA
;
608 unlock_new_inode(inode
);
613 make_bad_inode(inode
);
618 unsigned long ext2_count_free_inodes (struct super_block
* sb
)
620 struct ext2_group_desc
*desc
;
621 unsigned long desc_count
= 0;
625 struct ext2_super_block
*es
;
626 unsigned long bitmap_count
= 0;
627 struct buffer_head
*bitmap_bh
= NULL
;
629 es
= EXT2_SB(sb
)->s_es
;
630 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
633 desc
= ext2_get_group_desc (sb
, i
, NULL
);
636 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
638 bitmap_bh
= read_inode_bitmap(sb
, i
);
642 x
= ext2_count_free(bitmap_bh
, EXT2_INODES_PER_GROUP(sb
) / 8);
643 printk("group %d: stored = %d, counted = %u\n",
644 i
, le16_to_cpu(desc
->bg_free_inodes_count
), x
);
648 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
649 percpu_counter_read(&EXT2_SB(sb
)->s_freeinodes_counter
),
650 desc_count
, bitmap_count
);
653 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
654 desc
= ext2_get_group_desc (sb
, i
, NULL
);
657 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
663 /* Called at mount-time, super-block is locked */
664 unsigned long ext2_count_dirs (struct super_block
* sb
)
666 unsigned long count
= 0;
669 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
670 struct ext2_group_desc
*gdp
= ext2_get_group_desc (sb
, i
, NULL
);
673 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);