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 desc
->bg_free_inodes_count
=
79 cpu_to_le16(le16_to_cpu(desc
->bg_free_inodes_count
) + 1);
81 desc
->bg_used_dirs_count
=
82 cpu_to_le16(le16_to_cpu(desc
->bg_used_dirs_count
) - 1);
83 spin_unlock(sb_bgl_lock(EXT2_SB(sb
), group
));
85 percpu_counter_dec(&EXT2_SB(sb
)->s_dirs_counter
);
87 mark_buffer_dirty(bh
);
91 * NOTE! When we get the inode, we're the only people
92 * that have access to it, and as such there are no
93 * race conditions we have to worry about. The inode
94 * is not on the hash-lists, and it cannot be reached
95 * through the filesystem because the directory entry
96 * has been deleted earlier.
98 * HOWEVER: we must make sure that we get no aliases,
99 * which means that we have to call "clear_inode()"
100 * _before_ we mark the inode not in use in the inode
101 * bitmaps. Otherwise a newly created file might use
102 * the same inode number (not actually the same pointer
103 * though), and then we'd have two inodes sharing the
104 * same inode number and space on the harddisk.
106 void ext2_free_inode (struct inode
* inode
)
108 struct super_block
* sb
= inode
->i_sb
;
111 struct buffer_head
*bitmap_bh
= NULL
;
112 unsigned long block_group
;
114 struct ext2_super_block
* es
;
117 ext2_debug ("freeing inode %lu\n", ino
);
120 * Note: we must free any quota before locking the superblock,
121 * as writing the quota to disk may need the lock as well.
123 if (!is_bad_inode(inode
)) {
124 /* Quota is already initialized in iput() */
125 ext2_xattr_delete_inode(inode
);
126 DQUOT_FREE_INODE(inode
);
130 es
= EXT2_SB(sb
)->s_es
;
131 is_directory
= S_ISDIR(inode
->i_mode
);
133 /* Do this BEFORE marking the inode not in use or returning an error */
136 if (ino
< EXT2_FIRST_INO(sb
) ||
137 ino
> le32_to_cpu(es
->s_inodes_count
)) {
138 ext2_error (sb
, "ext2_free_inode",
139 "reserved or nonexistent inode %lu", ino
);
142 block_group
= (ino
- 1) / EXT2_INODES_PER_GROUP(sb
);
143 bit
= (ino
- 1) % EXT2_INODES_PER_GROUP(sb
);
145 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
149 /* Ok, now we can actually update the inode bitmaps.. */
150 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), block_group
),
151 bit
, (void *) bitmap_bh
->b_data
))
152 ext2_error (sb
, "ext2_free_inode",
153 "bit already cleared for inode %lu", ino
);
155 ext2_release_inode(sb
, block_group
, is_directory
);
156 mark_buffer_dirty(bitmap_bh
);
157 if (sb
->s_flags
& MS_SYNCHRONOUS
)
158 sync_dirty_buffer(bitmap_bh
);
164 * We perform asynchronous prereading of the new inode's inode block when
165 * we create the inode, in the expectation that the inode will be written
166 * back soon. There are two reasons:
168 * - When creating a large number of files, the async prereads will be
169 * nicely merged into large reads
170 * - When writing out a large number of inodes, we don't need to keep on
171 * stalling the writes while we read the inode block.
173 * FIXME: ext2_get_group_desc() needs to be simplified.
175 static void ext2_preread_inode(struct inode
*inode
)
177 unsigned long block_group
;
178 unsigned long offset
;
180 struct buffer_head
*bh
;
181 struct ext2_group_desc
* gdp
;
182 struct backing_dev_info
*bdi
;
184 bdi
= inode
->i_mapping
->backing_dev_info
;
185 if (bdi_read_congested(bdi
))
187 if (bdi_write_congested(bdi
))
190 block_group
= (inode
->i_ino
- 1) / EXT2_INODES_PER_GROUP(inode
->i_sb
);
191 gdp
= ext2_get_group_desc(inode
->i_sb
, block_group
, &bh
);
196 * Figure out the offset within the block group inode table
198 offset
= ((inode
->i_ino
- 1) % EXT2_INODES_PER_GROUP(inode
->i_sb
)) *
199 EXT2_INODE_SIZE(inode
->i_sb
);
200 block
= le32_to_cpu(gdp
->bg_inode_table
) +
201 (offset
>> EXT2_BLOCK_SIZE_BITS(inode
->i_sb
));
202 sb_breadahead(inode
->i_sb
, block
);
206 * There are two policies for allocating an inode. If the new inode is
207 * a directory, then a forward search is made for a block group with both
208 * free space and a low directory-to-inode ratio; if that fails, then of
209 * the groups with above-average free space, that group with the fewest
210 * directories already is chosen.
212 * For other inodes, search forward from the parent directory\'s block
213 * group to find a free inode.
215 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
)
217 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
218 int avefreei
= ext2_count_free_inodes(sb
) / ngroups
;
219 struct ext2_group_desc
*desc
, *best_desc
= NULL
;
220 struct buffer_head
*bh
, *best_bh
= NULL
;
221 int group
, best_group
= -1;
223 for (group
= 0; group
< ngroups
; group
++) {
224 desc
= ext2_get_group_desc (sb
, group
, &bh
);
225 if (!desc
|| !desc
->bg_free_inodes_count
)
227 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
230 (le16_to_cpu(desc
->bg_free_blocks_count
) >
231 le16_to_cpu(best_desc
->bg_free_blocks_count
))) {
244 * Orlov's allocator for directories.
246 * We always try to spread first-level directories.
248 * If there are blockgroups with both free inodes and free blocks counts
249 * not worse than average we return one with smallest directory count.
250 * Otherwise we simply return a random group.
252 * For the rest rules look so:
254 * It's OK to put directory into a group unless
255 * it has too many directories already (max_dirs) or
256 * it has too few free inodes left (min_inodes) or
257 * it has too few free blocks left (min_blocks) or
258 * it's already running too large debt (max_debt).
259 * Parent's group is prefered, if it doesn't satisfy these
260 * conditions we search cyclically through the rest. If none
261 * of the groups look good we just look for a group with more
262 * free inodes than average (starting at parent's group).
264 * Debt is incremented each time we allocate a directory and decremented
265 * when we allocate an inode, within 0--255.
268 #define INODE_COST 64
269 #define BLOCK_COST 256
271 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
273 int parent_group
= EXT2_I(parent
)->i_block_group
;
274 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
275 struct ext2_super_block
*es
= sbi
->s_es
;
276 int ngroups
= sbi
->s_groups_count
;
277 int inodes_per_group
= EXT2_INODES_PER_GROUP(sb
);
284 int max_debt
, max_dirs
, min_blocks
, min_inodes
;
286 struct ext2_group_desc
*desc
;
287 struct buffer_head
*bh
;
289 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
290 avefreei
= freei
/ ngroups
;
291 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
292 avefreeb
= free_blocks
/ ngroups
;
293 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
295 if ((parent
== sb
->s_root
->d_inode
) ||
296 (EXT2_I(parent
)->i_flags
& EXT2_TOPDIR_FL
)) {
297 struct ext2_group_desc
*best_desc
= NULL
;
298 struct buffer_head
*best_bh
= NULL
;
299 int best_ndir
= inodes_per_group
;
302 get_random_bytes(&group
, sizeof(group
));
303 parent_group
= (unsigned)group
% ngroups
;
304 for (i
= 0; i
< ngroups
; i
++) {
305 group
= (parent_group
+ i
) % ngroups
;
306 desc
= ext2_get_group_desc (sb
, group
, &bh
);
307 if (!desc
|| !desc
->bg_free_inodes_count
)
309 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
311 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
313 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
316 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
320 if (best_group
>= 0) {
330 ndirs
= 1; /* percpu_counters are approximate... */
332 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
)-free_blocks
) / ndirs
;
334 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
335 min_inodes
= avefreei
- inodes_per_group
/ 4;
336 min_blocks
= avefreeb
- EXT2_BLOCKS_PER_GROUP(sb
) / 4;
338 max_debt
= EXT2_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, BLOCK_COST
);
339 if (max_debt
* INODE_COST
> inodes_per_group
)
340 max_debt
= inodes_per_group
/ INODE_COST
;
346 for (i
= 0; i
< ngroups
; i
++) {
347 group
= (parent_group
+ i
) % ngroups
;
348 desc
= ext2_get_group_desc (sb
, group
, &bh
);
349 if (!desc
|| !desc
->bg_free_inodes_count
)
351 if (sbi
->s_debts
[group
] >= max_debt
)
353 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
355 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
357 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
363 for (i
= 0; i
< ngroups
; i
++) {
364 group
= (parent_group
+ i
) % ngroups
;
365 desc
= ext2_get_group_desc (sb
, group
, &bh
);
366 if (!desc
|| !desc
->bg_free_inodes_count
)
368 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
374 * The free-inodes counter is approximate, and for really small
375 * filesystems the above test can fail to find any blockgroups
387 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
389 int parent_group
= EXT2_I(parent
)->i_block_group
;
390 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
391 struct ext2_group_desc
*desc
;
392 struct buffer_head
*bh
;
396 * Try to place the inode in its parent directory
398 group
= parent_group
;
399 desc
= ext2_get_group_desc (sb
, group
, &bh
);
400 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
401 le16_to_cpu(desc
->bg_free_blocks_count
))
405 * We're going to place this inode in a different blockgroup from its
406 * parent. We want to cause files in a common directory to all land in
407 * the same blockgroup. But we want files which are in a different
408 * directory which shares a blockgroup with our parent to land in a
409 * different blockgroup.
411 * So add our directory's i_ino into the starting point for the hash.
413 group
= (group
+ parent
->i_ino
) % ngroups
;
416 * Use a quadratic hash to find a group with a free inode and some
419 for (i
= 1; i
< ngroups
; i
<<= 1) {
421 if (group
>= ngroups
)
423 desc
= ext2_get_group_desc (sb
, group
, &bh
);
424 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
425 le16_to_cpu(desc
->bg_free_blocks_count
))
430 * That failed: try linear search for a free inode, even if that group
431 * has no free blocks.
433 group
= parent_group
;
434 for (i
= 0; i
< ngroups
; i
++) {
435 if (++group
>= ngroups
)
437 desc
= ext2_get_group_desc (sb
, group
, &bh
);
438 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
448 struct inode
*ext2_new_inode(struct inode
*dir
, int mode
)
450 struct super_block
*sb
;
451 struct buffer_head
*bitmap_bh
= NULL
;
452 struct buffer_head
*bh2
;
455 struct inode
* inode
;
456 struct ext2_group_desc
*gdp
;
457 struct ext2_super_block
*es
;
458 struct ext2_inode_info
*ei
;
459 struct ext2_sb_info
*sbi
;
463 inode
= new_inode(sb
);
465 return ERR_PTR(-ENOMEM
);
471 if (test_opt(sb
, OLDALLOC
))
472 group
= find_group_dir(sb
, dir
);
474 group
= find_group_orlov(sb
, dir
);
476 group
= find_group_other(sb
, dir
);
483 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
484 gdp
= ext2_get_group_desc(sb
, group
, &bh2
);
486 bitmap_bh
= read_inode_bitmap(sb
, group
);
493 repeat_in_this_group
:
494 ino
= ext2_find_next_zero_bit((unsigned long *)bitmap_bh
->b_data
,
495 EXT2_INODES_PER_GROUP(sb
), ino
);
496 if (ino
>= EXT2_INODES_PER_GROUP(sb
)) {
498 * Rare race: find_group_xx() decided that there were
499 * free inodes in this group, but by the time we tried
500 * to allocate one, they're all gone. This can also
501 * occur because the counters which find_group_orlov()
502 * uses are approximate. So just go and search the
505 if (++group
== sbi
->s_groups_count
)
509 if (ext2_set_bit_atomic(sb_bgl_lock(sbi
, group
),
510 ino
, bitmap_bh
->b_data
)) {
511 /* we lost this inode */
512 if (++ino
>= EXT2_INODES_PER_GROUP(sb
)) {
513 /* this group is exhausted, try next group */
514 if (++group
== sbi
->s_groups_count
)
518 /* try to find free inode in the same group */
519 goto repeat_in_this_group
;
525 * Scanned all blockgroups.
530 mark_buffer_dirty(bitmap_bh
);
531 if (sb
->s_flags
& MS_SYNCHRONOUS
)
532 sync_dirty_buffer(bitmap_bh
);
535 ino
+= group
* EXT2_INODES_PER_GROUP(sb
) + 1;
536 if (ino
< EXT2_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
537 ext2_error (sb
, "ext2_new_inode",
538 "reserved inode or inode > inodes count - "
539 "block_group = %d,inode=%lu", group
,
540 (unsigned long) ino
);
545 percpu_counter_mod(&sbi
->s_freeinodes_counter
, -1);
547 percpu_counter_inc(&sbi
->s_dirs_counter
);
549 spin_lock(sb_bgl_lock(sbi
, group
));
550 gdp
->bg_free_inodes_count
=
551 cpu_to_le16(le16_to_cpu(gdp
->bg_free_inodes_count
) - 1);
553 if (sbi
->s_debts
[group
] < 255)
554 sbi
->s_debts
[group
]++;
555 gdp
->bg_used_dirs_count
=
556 cpu_to_le16(le16_to_cpu(gdp
->bg_used_dirs_count
) + 1);
558 if (sbi
->s_debts
[group
])
559 sbi
->s_debts
[group
]--;
561 spin_unlock(sb_bgl_lock(sbi
, group
));
564 mark_buffer_dirty(bh2
);
565 inode
->i_uid
= current
->fsuid
;
566 if (test_opt (sb
, GRPID
))
567 inode
->i_gid
= dir
->i_gid
;
568 else if (dir
->i_mode
& S_ISGID
) {
569 inode
->i_gid
= dir
->i_gid
;
573 inode
->i_gid
= current
->fsgid
;
574 inode
->i_mode
= mode
;
577 inode
->i_blksize
= PAGE_SIZE
; /* This is the optimal IO size (for stat), not the fs block size */
579 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
580 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
581 ei
->i_flags
= EXT2_I(dir
)->i_flags
& ~EXT2_BTREE_FL
;
583 ei
->i_flags
&= ~(EXT2_IMMUTABLE_FL
|EXT2_APPEND_FL
);
584 /* dirsync is only applied to directories */
586 ei
->i_flags
&= ~EXT2_DIRSYNC_FL
;
593 ei
->i_block_group
= group
;
594 ei
->i_next_alloc_block
= 0;
595 ei
->i_next_alloc_goal
= 0;
596 ei
->i_prealloc_block
= 0;
597 ei
->i_prealloc_count
= 0;
598 ei
->i_dir_start_lookup
= 0;
599 ei
->i_state
= EXT2_STATE_NEW
;
600 ext2_set_inode_flags(inode
);
601 spin_lock(&sbi
->s_next_gen_lock
);
602 inode
->i_generation
= sbi
->s_next_generation
++;
603 spin_unlock(&sbi
->s_next_gen_lock
);
604 insert_inode_hash(inode
);
606 if (DQUOT_ALLOC_INODE(inode
)) {
611 err
= ext2_init_acl(inode
, dir
);
615 err
= ext2_init_security(inode
,dir
);
619 mark_inode_dirty(inode
);
620 ext2_debug("allocating inode %lu\n", inode
->i_ino
);
621 ext2_preread_inode(inode
);
625 DQUOT_FREE_INODE(inode
);
629 inode
->i_flags
|= S_NOQUOTA
;
635 make_bad_inode(inode
);
640 unsigned long ext2_count_free_inodes (struct super_block
* sb
)
642 struct ext2_group_desc
*desc
;
643 unsigned long desc_count
= 0;
647 struct ext2_super_block
*es
;
648 unsigned long bitmap_count
= 0;
649 struct buffer_head
*bitmap_bh
= NULL
;
651 es
= EXT2_SB(sb
)->s_es
;
652 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
655 desc
= ext2_get_group_desc (sb
, i
, NULL
);
658 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
660 bitmap_bh
= read_inode_bitmap(sb
, i
);
664 x
= ext2_count_free(bitmap_bh
, EXT2_INODES_PER_GROUP(sb
) / 8);
665 printk("group %d: stored = %d, counted = %u\n",
666 i
, le16_to_cpu(desc
->bg_free_inodes_count
), x
);
670 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
671 percpu_counter_read(&EXT2_SB(sb
)->s_freeinodes_counter
),
672 desc_count
, bitmap_count
);
675 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
676 desc
= ext2_get_group_desc (sb
, i
, NULL
);
679 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
685 /* Called at mount-time, super-block is locked */
686 unsigned long ext2_count_dirs (struct super_block
* sb
)
688 unsigned long count
= 0;
691 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
692 struct ext2_group_desc
*gdp
= ext2_get_group_desc (sb
, i
, NULL
);
695 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);