1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext2/ialloc.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@dcs.ed.ac.uk), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
16 #include <linux/quotaops.h>
17 #include <linux/sched.h>
18 #include <linux/backing-dev.h>
19 #include <linux/buffer_head.h>
20 #include <linux/random.h>
26 * ialloc.c contains the inodes allocation and deallocation routines
30 * The free inodes are managed by bitmaps. A file system contains several
31 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
32 * block for inodes, N blocks for the inode table and data blocks.
34 * The file system contains group descriptors which are located after the
35 * super block. Each descriptor contains the number of the bitmap block and
36 * the free blocks count in the block.
41 * Read the inode allocation bitmap for a given block_group, reading
42 * into the specified slot in the superblock's bitmap cache.
44 * Return buffer_head of bitmap on success or NULL.
46 static struct buffer_head
*
47 read_inode_bitmap(struct super_block
* sb
, unsigned long block_group
)
49 struct ext2_group_desc
*desc
;
50 struct buffer_head
*bh
= NULL
;
52 desc
= ext2_get_group_desc(sb
, block_group
, NULL
);
56 bh
= sb_bread(sb
, le32_to_cpu(desc
->bg_inode_bitmap
));
58 ext2_error(sb
, "read_inode_bitmap",
59 "Cannot read inode bitmap - "
60 "block_group = %lu, inode_bitmap = %u",
61 block_group
, le32_to_cpu(desc
->bg_inode_bitmap
));
66 static void ext2_release_inode(struct super_block
*sb
, int group
, int dir
)
68 struct ext2_group_desc
* desc
;
69 struct buffer_head
*bh
;
71 desc
= ext2_get_group_desc(sb
, group
, &bh
);
73 ext2_error(sb
, "ext2_release_inode",
74 "can't get descriptor for group %d", group
);
78 spin_lock(sb_bgl_lock(EXT2_SB(sb
), group
));
79 le16_add_cpu(&desc
->bg_free_inodes_count
, 1);
81 le16_add_cpu(&desc
->bg_used_dirs_count
, -1);
82 spin_unlock(sb_bgl_lock(EXT2_SB(sb
), group
));
83 percpu_counter_inc(&EXT2_SB(sb
)->s_freeinodes_counter
);
85 percpu_counter_dec(&EXT2_SB(sb
)->s_dirs_counter
);
86 mark_buffer_dirty(bh
);
90 * NOTE! When we get the inode, we're the only people
91 * that have access to it, and as such there are no
92 * race conditions we have to worry about. The inode
93 * is not on the hash-lists, and it cannot be reached
94 * through the filesystem because the directory entry
95 * has been deleted earlier.
97 * HOWEVER: we must make sure that we get no aliases,
98 * which means that we have to call "clear_inode()"
99 * _before_ we mark the inode not in use in the inode
100 * bitmaps. Otherwise a newly created file might use
101 * the same inode number (not actually the same pointer
102 * though), and then we'd have two inodes sharing the
103 * same inode number and space on the harddisk.
105 void ext2_free_inode (struct inode
* inode
)
107 struct super_block
* sb
= inode
->i_sb
;
110 struct buffer_head
*bitmap_bh
;
111 unsigned long block_group
;
113 struct ext2_super_block
* es
;
116 ext2_debug ("freeing inode %lu\n", ino
);
119 * Note: we must free any quota before locking the superblock,
120 * as writing the quota to disk may need the lock as well.
122 /* Quota is already initialized in iput() */
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
& SB_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_to_bdi(inode
);
176 if (bdi_rw_congested(bdi
))
179 block_group
= (inode
->i_ino
- 1) / EXT2_INODES_PER_GROUP(inode
->i_sb
);
180 gdp
= ext2_get_group_desc(inode
->i_sb
, block_group
, NULL
);
185 * Figure out the offset within the block group inode table
187 offset
= ((inode
->i_ino
- 1) % EXT2_INODES_PER_GROUP(inode
->i_sb
)) *
188 EXT2_INODE_SIZE(inode
->i_sb
);
189 block
= le32_to_cpu(gdp
->bg_inode_table
) +
190 (offset
>> EXT2_BLOCK_SIZE_BITS(inode
->i_sb
));
191 sb_breadahead(inode
->i_sb
, block
);
195 * There are two policies for allocating an inode. If the new inode is
196 * a directory, then a forward search is made for a block group with both
197 * free space and a low directory-to-inode ratio; if that fails, then of
198 * the groups with above-average free space, that group with the fewest
199 * directories already is chosen.
201 * For other inodes, search forward from the parent directory\'s block
202 * group to find a free inode.
204 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
)
206 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
207 int avefreei
= ext2_count_free_inodes(sb
) / ngroups
;
208 struct ext2_group_desc
*desc
, *best_desc
= NULL
;
209 int group
, best_group
= -1;
211 for (group
= 0; group
< ngroups
; group
++) {
212 desc
= ext2_get_group_desc (sb
, group
, NULL
);
213 if (!desc
|| !desc
->bg_free_inodes_count
)
215 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
218 (le16_to_cpu(desc
->bg_free_blocks_count
) >
219 le16_to_cpu(best_desc
->bg_free_blocks_count
))) {
229 * Orlov's allocator for directories.
231 * We always try to spread first-level directories.
233 * If there are blockgroups with both free inodes and free blocks counts
234 * not worse than average we return one with smallest directory count.
235 * Otherwise we simply return a random group.
237 * For the rest rules look so:
239 * It's OK to put directory into a group unless
240 * it has too many directories already (max_dirs) or
241 * it has too few free inodes left (min_inodes) or
242 * it has too few free blocks left (min_blocks) or
243 * it's already running too large debt (max_debt).
244 * Parent's group is preferred, if it doesn't satisfy these
245 * conditions we search cyclically through the rest. If none
246 * of the groups look good we just look for a group with more
247 * free inodes than average (starting at parent's group).
249 * Debt is incremented each time we allocate a directory and decremented
250 * when we allocate an inode, within 0--255.
253 #define INODE_COST 64
254 #define BLOCK_COST 256
256 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
258 int parent_group
= EXT2_I(parent
)->i_block_group
;
259 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
260 struct ext2_super_block
*es
= sbi
->s_es
;
261 int ngroups
= sbi
->s_groups_count
;
262 int inodes_per_group
= EXT2_INODES_PER_GROUP(sb
);
269 int max_debt
, max_dirs
, min_blocks
, min_inodes
;
271 struct ext2_group_desc
*desc
;
273 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
274 avefreei
= freei
/ ngroups
;
275 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
276 avefreeb
= free_blocks
/ ngroups
;
277 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
279 if ((parent
== d_inode(sb
->s_root
)) ||
280 (EXT2_I(parent
)->i_flags
& EXT2_TOPDIR_FL
)) {
281 struct ext2_group_desc
*best_desc
= NULL
;
282 int best_ndir
= inodes_per_group
;
285 group
= prandom_u32();
286 parent_group
= (unsigned)group
% ngroups
;
287 for (i
= 0; i
< ngroups
; i
++) {
288 group
= (parent_group
+ i
) % ngroups
;
289 desc
= ext2_get_group_desc (sb
, group
, NULL
);
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
);
302 if (best_group
>= 0) {
311 ndirs
= 1; /* percpu_counters are approximate... */
313 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
)-free_blocks
) / ndirs
;
315 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
316 min_inodes
= avefreei
- inodes_per_group
/ 4;
317 min_blocks
= avefreeb
- EXT2_BLOCKS_PER_GROUP(sb
) / 4;
319 max_debt
= EXT2_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, BLOCK_COST
);
320 if (max_debt
* INODE_COST
> inodes_per_group
)
321 max_debt
= inodes_per_group
/ INODE_COST
;
327 for (i
= 0; i
< ngroups
; i
++) {
328 group
= (parent_group
+ i
) % ngroups
;
329 desc
= ext2_get_group_desc (sb
, group
, NULL
);
330 if (!desc
|| !desc
->bg_free_inodes_count
)
332 if (sbi
->s_debts
[group
] >= max_debt
)
334 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
336 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
338 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
344 for (i
= 0; i
< ngroups
; i
++) {
345 group
= (parent_group
+ i
) % ngroups
;
346 desc
= ext2_get_group_desc (sb
, group
, NULL
);
347 if (!desc
|| !desc
->bg_free_inodes_count
)
349 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
355 * The free-inodes counter is approximate, and for really small
356 * filesystems the above test can fail to find any blockgroups
368 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
370 int parent_group
= EXT2_I(parent
)->i_block_group
;
371 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
372 struct ext2_group_desc
*desc
;
376 * Try to place the inode in its parent directory
378 group
= parent_group
;
379 desc
= ext2_get_group_desc (sb
, group
, NULL
);
380 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
381 le16_to_cpu(desc
->bg_free_blocks_count
))
385 * We're going to place this inode in a different blockgroup from its
386 * parent. We want to cause files in a common directory to all land in
387 * the same blockgroup. But we want files which are in a different
388 * directory which shares a blockgroup with our parent to land in a
389 * different blockgroup.
391 * So add our directory's i_ino into the starting point for the hash.
393 group
= (group
+ parent
->i_ino
) % ngroups
;
396 * Use a quadratic hash to find a group with a free inode and some
399 for (i
= 1; i
< ngroups
; i
<<= 1) {
401 if (group
>= ngroups
)
403 desc
= ext2_get_group_desc (sb
, group
, NULL
);
404 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
405 le16_to_cpu(desc
->bg_free_blocks_count
))
410 * That failed: try linear search for a free inode, even if that group
411 * has no free blocks.
413 group
= parent_group
;
414 for (i
= 0; i
< ngroups
; i
++) {
415 if (++group
>= ngroups
)
417 desc
= ext2_get_group_desc (sb
, group
, NULL
);
418 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
428 struct inode
*ext2_new_inode(struct inode
*dir
, umode_t mode
,
429 const struct qstr
*qstr
)
431 struct super_block
*sb
;
432 struct buffer_head
*bitmap_bh
= NULL
;
433 struct buffer_head
*bh2
;
436 struct inode
* inode
;
437 struct ext2_group_desc
*gdp
;
438 struct ext2_super_block
*es
;
439 struct ext2_inode_info
*ei
;
440 struct ext2_sb_info
*sbi
;
444 inode
= new_inode(sb
);
446 return ERR_PTR(-ENOMEM
);
452 if (test_opt(sb
, OLDALLOC
))
453 group
= find_group_dir(sb
, dir
);
455 group
= find_group_orlov(sb
, dir
);
457 group
= find_group_other(sb
, dir
);
464 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
465 gdp
= ext2_get_group_desc(sb
, group
, &bh2
);
467 if (++group
== sbi
->s_groups_count
)
472 bitmap_bh
= read_inode_bitmap(sb
, group
);
479 repeat_in_this_group
:
480 ino
= ext2_find_next_zero_bit((unsigned long *)bitmap_bh
->b_data
,
481 EXT2_INODES_PER_GROUP(sb
), ino
);
482 if (ino
>= EXT2_INODES_PER_GROUP(sb
)) {
484 * Rare race: find_group_xx() decided that there were
485 * free inodes in this group, but by the time we tried
486 * to allocate one, they're all gone. This can also
487 * occur because the counters which find_group_orlov()
488 * uses are approximate. So just go and search the
491 if (++group
== sbi
->s_groups_count
)
495 if (ext2_set_bit_atomic(sb_bgl_lock(sbi
, group
),
496 ino
, bitmap_bh
->b_data
)) {
497 /* we lost this inode */
498 if (++ino
>= EXT2_INODES_PER_GROUP(sb
)) {
499 /* this group is exhausted, try next group */
500 if (++group
== sbi
->s_groups_count
)
504 /* try to find free inode in the same group */
505 goto repeat_in_this_group
;
511 * Scanned all blockgroups.
517 mark_buffer_dirty(bitmap_bh
);
518 if (sb
->s_flags
& SB_SYNCHRONOUS
)
519 sync_dirty_buffer(bitmap_bh
);
522 ino
+= group
* EXT2_INODES_PER_GROUP(sb
) + 1;
523 if (ino
< EXT2_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
524 ext2_error (sb
, "ext2_new_inode",
525 "reserved inode or inode > inodes count - "
526 "block_group = %d,inode=%lu", group
,
527 (unsigned long) ino
);
532 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
534 percpu_counter_inc(&sbi
->s_dirs_counter
);
536 spin_lock(sb_bgl_lock(sbi
, group
));
537 le16_add_cpu(&gdp
->bg_free_inodes_count
, -1);
539 if (sbi
->s_debts
[group
] < 255)
540 sbi
->s_debts
[group
]++;
541 le16_add_cpu(&gdp
->bg_used_dirs_count
, 1);
543 if (sbi
->s_debts
[group
])
544 sbi
->s_debts
[group
]--;
546 spin_unlock(sb_bgl_lock(sbi
, group
));
548 mark_buffer_dirty(bh2
);
549 if (test_opt(sb
, GRPID
)) {
550 inode
->i_mode
= mode
;
551 inode
->i_uid
= current_fsuid();
552 inode
->i_gid
= dir
->i_gid
;
554 inode_init_owner(inode
, dir
, mode
);
558 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= current_time(inode
);
559 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
561 ext2_mask_flags(mode
, EXT2_I(dir
)->i_flags
& EXT2_FL_INHERITED
);
568 ei
->i_block_alloc_info
= NULL
;
569 ei
->i_block_group
= group
;
570 ei
->i_dir_start_lookup
= 0;
571 ei
->i_state
= EXT2_STATE_NEW
;
572 ext2_set_inode_flags(inode
);
573 spin_lock(&sbi
->s_next_gen_lock
);
574 inode
->i_generation
= sbi
->s_next_generation
++;
575 spin_unlock(&sbi
->s_next_gen_lock
);
576 if (insert_inode_locked(inode
) < 0) {
577 ext2_error(sb
, "ext2_new_inode",
578 "inode number already in use - inode=%lu",
579 (unsigned long) ino
);
584 err
= dquot_initialize(inode
);
588 err
= dquot_alloc_inode(inode
);
592 err
= ext2_init_acl(inode
, dir
);
596 err
= ext2_init_security(inode
, dir
, qstr
);
600 mark_inode_dirty(inode
);
601 ext2_debug("allocating inode %lu\n", inode
->i_ino
);
602 ext2_preread_inode(inode
);
606 dquot_free_inode(inode
);
610 inode
->i_flags
|= S_NOQUOTA
;
612 discard_new_inode(inode
);
616 make_bad_inode(inode
);
621 unsigned long ext2_count_free_inodes (struct super_block
* sb
)
623 struct ext2_group_desc
*desc
;
624 unsigned long desc_count
= 0;
628 struct ext2_super_block
*es
;
629 unsigned long bitmap_count
= 0;
630 struct buffer_head
*bitmap_bh
= NULL
;
632 es
= EXT2_SB(sb
)->s_es
;
633 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
636 desc
= ext2_get_group_desc (sb
, i
, NULL
);
639 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
641 bitmap_bh
= read_inode_bitmap(sb
, i
);
645 x
= ext2_count_free(bitmap_bh
, EXT2_INODES_PER_GROUP(sb
) / 8);
646 printk("group %d: stored = %d, counted = %u\n",
647 i
, le16_to_cpu(desc
->bg_free_inodes_count
), x
);
651 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
653 percpu_counter_read(&EXT2_SB(sb
)->s_freeinodes_counter
),
654 desc_count
, bitmap_count
);
657 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
658 desc
= ext2_get_group_desc (sb
, i
, NULL
);
661 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
667 /* Called at mount-time, super-block is locked */
668 unsigned long ext2_count_dirs (struct super_block
* sb
)
670 unsigned long count
= 0;
673 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
674 struct ext2_group_desc
*gdp
= ext2_get_group_desc (sb
, i
, NULL
);
677 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);