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
));
84 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 dquot_free_inode(inode
);
125 es
= EXT2_SB(sb
)->s_es
;
126 is_directory
= S_ISDIR(inode
->i_mode
);
128 if (ino
< EXT2_FIRST_INO(sb
) ||
129 ino
> le32_to_cpu(es
->s_inodes_count
)) {
130 ext2_error (sb
, "ext2_free_inode",
131 "reserved or nonexistent inode %lu", ino
);
134 block_group
= (ino
- 1) / EXT2_INODES_PER_GROUP(sb
);
135 bit
= (ino
- 1) % EXT2_INODES_PER_GROUP(sb
);
136 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
140 /* Ok, now we can actually update the inode bitmaps.. */
141 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb
), block_group
),
142 bit
, (void *) bitmap_bh
->b_data
))
143 ext2_error (sb
, "ext2_free_inode",
144 "bit already cleared for inode %lu", ino
);
146 ext2_release_inode(sb
, block_group
, is_directory
);
147 mark_buffer_dirty(bitmap_bh
);
148 if (sb
->s_flags
& SB_SYNCHRONOUS
)
149 sync_dirty_buffer(bitmap_bh
);
155 * We perform asynchronous prereading of the new inode's inode block when
156 * we create the inode, in the expectation that the inode will be written
157 * back soon. There are two reasons:
159 * - When creating a large number of files, the async prereads will be
160 * nicely merged into large reads
161 * - When writing out a large number of inodes, we don't need to keep on
162 * stalling the writes while we read the inode block.
164 * FIXME: ext2_get_group_desc() needs to be simplified.
166 static void ext2_preread_inode(struct inode
*inode
)
168 unsigned long block_group
;
169 unsigned long offset
;
171 struct ext2_group_desc
* gdp
;
172 struct backing_dev_info
*bdi
;
174 bdi
= inode_to_bdi(inode
);
175 if (bdi_rw_congested(bdi
))
178 block_group
= (inode
->i_ino
- 1) / EXT2_INODES_PER_GROUP(inode
->i_sb
);
179 gdp
= ext2_get_group_desc(inode
->i_sb
, block_group
, NULL
);
184 * Figure out the offset within the block group inode table
186 offset
= ((inode
->i_ino
- 1) % EXT2_INODES_PER_GROUP(inode
->i_sb
)) *
187 EXT2_INODE_SIZE(inode
->i_sb
);
188 block
= le32_to_cpu(gdp
->bg_inode_table
) +
189 (offset
>> EXT2_BLOCK_SIZE_BITS(inode
->i_sb
));
190 sb_breadahead(inode
->i_sb
, block
);
194 * There are two policies for allocating an inode. If the new inode is
195 * a directory, then a forward search is made for a block group with both
196 * free space and a low directory-to-inode ratio; if that fails, then of
197 * the groups with above-average free space, that group with the fewest
198 * directories already is chosen.
200 * For other inodes, search forward from the parent directory\'s block
201 * group to find a free inode.
203 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
)
205 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
206 int avefreei
= ext2_count_free_inodes(sb
) / ngroups
;
207 struct ext2_group_desc
*desc
, *best_desc
= NULL
;
208 int group
, best_group
= -1;
210 for (group
= 0; group
< ngroups
; group
++) {
211 desc
= ext2_get_group_desc (sb
, group
, NULL
);
212 if (!desc
|| !desc
->bg_free_inodes_count
)
214 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
217 (le16_to_cpu(desc
->bg_free_blocks_count
) >
218 le16_to_cpu(best_desc
->bg_free_blocks_count
))) {
228 * Orlov's allocator for directories.
230 * We always try to spread first-level directories.
232 * If there are blockgroups with both free inodes and free blocks counts
233 * not worse than average we return one with smallest directory count.
234 * Otherwise we simply return a random group.
236 * For the rest rules look so:
238 * It's OK to put directory into a group unless
239 * it has too many directories already (max_dirs) or
240 * it has too few free inodes left (min_inodes) or
241 * it has too few free blocks left (min_blocks) or
242 * it's already running too large debt (max_debt).
243 * Parent's group is preferred, if it doesn't satisfy these
244 * conditions we search cyclically through the rest. If none
245 * of the groups look good we just look for a group with more
246 * free inodes than average (starting at parent's group).
248 * Debt is incremented each time we allocate a directory and decremented
249 * when we allocate an inode, within 0--255.
252 #define INODE_COST 64
253 #define BLOCK_COST 256
255 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
257 int parent_group
= EXT2_I(parent
)->i_block_group
;
258 struct ext2_sb_info
*sbi
= EXT2_SB(sb
);
259 struct ext2_super_block
*es
= sbi
->s_es
;
260 int ngroups
= sbi
->s_groups_count
;
261 int inodes_per_group
= EXT2_INODES_PER_GROUP(sb
);
268 int max_debt
, max_dirs
, min_blocks
, min_inodes
;
270 struct ext2_group_desc
*desc
;
272 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
273 avefreei
= freei
/ ngroups
;
274 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
275 avefreeb
= free_blocks
/ ngroups
;
276 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
278 if ((parent
== d_inode(sb
->s_root
)) ||
279 (EXT2_I(parent
)->i_flags
& EXT2_TOPDIR_FL
)) {
280 struct ext2_group_desc
*best_desc
= NULL
;
281 int best_ndir
= inodes_per_group
;
284 group
= prandom_u32();
285 parent_group
= (unsigned)group
% ngroups
;
286 for (i
= 0; i
< ngroups
; i
++) {
287 group
= (parent_group
+ i
) % ngroups
;
288 desc
= ext2_get_group_desc (sb
, group
, NULL
);
289 if (!desc
|| !desc
->bg_free_inodes_count
)
291 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
293 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
295 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
298 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
301 if (best_group
>= 0) {
310 ndirs
= 1; /* percpu_counters are approximate... */
312 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
)-free_blocks
) / ndirs
;
314 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
315 min_inodes
= avefreei
- inodes_per_group
/ 4;
316 min_blocks
= avefreeb
- EXT2_BLOCKS_PER_GROUP(sb
) / 4;
318 max_debt
= EXT2_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, BLOCK_COST
);
319 if (max_debt
* INODE_COST
> inodes_per_group
)
320 max_debt
= inodes_per_group
/ INODE_COST
;
326 for (i
= 0; i
< ngroups
; i
++) {
327 group
= (parent_group
+ i
) % ngroups
;
328 desc
= ext2_get_group_desc (sb
, group
, NULL
);
329 if (!desc
|| !desc
->bg_free_inodes_count
)
331 if (sbi
->s_debts
[group
] >= max_debt
)
333 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
335 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
337 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
343 for (i
= 0; i
< ngroups
; i
++) {
344 group
= (parent_group
+ i
) % ngroups
;
345 desc
= ext2_get_group_desc (sb
, group
, NULL
);
346 if (!desc
|| !desc
->bg_free_inodes_count
)
348 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
354 * The free-inodes counter is approximate, and for really small
355 * filesystems the above test can fail to find any blockgroups
367 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
369 int parent_group
= EXT2_I(parent
)->i_block_group
;
370 int ngroups
= EXT2_SB(sb
)->s_groups_count
;
371 struct ext2_group_desc
*desc
;
375 * Try to place the inode in its parent directory
377 group
= parent_group
;
378 desc
= ext2_get_group_desc (sb
, group
, NULL
);
379 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
380 le16_to_cpu(desc
->bg_free_blocks_count
))
384 * We're going to place this inode in a different blockgroup from its
385 * parent. We want to cause files in a common directory to all land in
386 * the same blockgroup. But we want files which are in a different
387 * directory which shares a blockgroup with our parent to land in a
388 * different blockgroup.
390 * So add our directory's i_ino into the starting point for the hash.
392 group
= (group
+ parent
->i_ino
) % ngroups
;
395 * Use a quadratic hash to find a group with a free inode and some
398 for (i
= 1; i
< ngroups
; i
<<= 1) {
400 if (group
>= ngroups
)
402 desc
= ext2_get_group_desc (sb
, group
, NULL
);
403 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
404 le16_to_cpu(desc
->bg_free_blocks_count
))
409 * That failed: try linear search for a free inode, even if that group
410 * has no free blocks.
412 group
= parent_group
;
413 for (i
= 0; i
< ngroups
; i
++) {
414 if (++group
>= ngroups
)
416 desc
= ext2_get_group_desc (sb
, group
, NULL
);
417 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
427 struct inode
*ext2_new_inode(struct inode
*dir
, umode_t mode
,
428 const struct qstr
*qstr
)
430 struct super_block
*sb
;
431 struct buffer_head
*bitmap_bh
= NULL
;
432 struct buffer_head
*bh2
;
435 struct inode
* inode
;
436 struct ext2_group_desc
*gdp
;
437 struct ext2_super_block
*es
;
438 struct ext2_inode_info
*ei
;
439 struct ext2_sb_info
*sbi
;
443 inode
= new_inode(sb
);
445 return ERR_PTR(-ENOMEM
);
451 if (test_opt(sb
, OLDALLOC
))
452 group
= find_group_dir(sb
, dir
);
454 group
= find_group_orlov(sb
, dir
);
456 group
= find_group_other(sb
, dir
);
463 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
464 gdp
= ext2_get_group_desc(sb
, group
, &bh2
);
466 if (++group
== sbi
->s_groups_count
)
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.
516 mark_buffer_dirty(bitmap_bh
);
517 if (sb
->s_flags
& SB_SYNCHRONOUS
)
518 sync_dirty_buffer(bitmap_bh
);
521 ino
+= group
* EXT2_INODES_PER_GROUP(sb
) + 1;
522 if (ino
< EXT2_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
523 ext2_error (sb
, "ext2_new_inode",
524 "reserved inode or inode > inodes count - "
525 "block_group = %d,inode=%lu", group
,
526 (unsigned long) ino
);
531 percpu_counter_add(&sbi
->s_freeinodes_counter
, -1);
533 percpu_counter_inc(&sbi
->s_dirs_counter
);
535 spin_lock(sb_bgl_lock(sbi
, group
));
536 le16_add_cpu(&gdp
->bg_free_inodes_count
, -1);
538 if (sbi
->s_debts
[group
] < 255)
539 sbi
->s_debts
[group
]++;
540 le16_add_cpu(&gdp
->bg_used_dirs_count
, 1);
542 if (sbi
->s_debts
[group
])
543 sbi
->s_debts
[group
]--;
545 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(inode
);
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 err
= dquot_initialize(inode
);
587 err
= dquot_alloc_inode(inode
);
591 err
= ext2_init_acl(inode
, dir
);
595 err
= ext2_init_security(inode
, dir
, qstr
);
599 mark_inode_dirty(inode
);
600 ext2_debug("allocating inode %lu\n", inode
->i_ino
);
601 ext2_preread_inode(inode
);
605 dquot_free_inode(inode
);
609 inode
->i_flags
|= S_NOQUOTA
;
611 discard_new_inode(inode
);
615 make_bad_inode(inode
);
620 unsigned long ext2_count_free_inodes (struct super_block
* sb
)
622 struct ext2_group_desc
*desc
;
623 unsigned long desc_count
= 0;
627 struct ext2_super_block
*es
;
628 unsigned long bitmap_count
= 0;
629 struct buffer_head
*bitmap_bh
= NULL
;
631 es
= EXT2_SB(sb
)->s_es
;
632 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
635 desc
= ext2_get_group_desc (sb
, i
, NULL
);
638 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
640 bitmap_bh
= read_inode_bitmap(sb
, i
);
644 x
= ext2_count_free(bitmap_bh
, EXT2_INODES_PER_GROUP(sb
) / 8);
645 printk("group %d: stored = %d, counted = %u\n",
646 i
, le16_to_cpu(desc
->bg_free_inodes_count
), x
);
650 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
652 percpu_counter_read(&EXT2_SB(sb
)->s_freeinodes_counter
),
653 desc_count
, bitmap_count
);
656 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
657 desc
= ext2_get_group_desc (sb
, i
, NULL
);
660 desc_count
+= le16_to_cpu(desc
->bg_free_inodes_count
);
666 /* Called at mount-time, super-block is locked */
667 unsigned long ext2_count_dirs (struct super_block
* sb
)
669 unsigned long count
= 0;
672 for (i
= 0; i
< EXT2_SB(sb
)->s_groups_count
; i
++) {
673 struct ext2_group_desc
*gdp
= ext2_get_group_desc (sb
, i
, NULL
);
676 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);