2 * linux/fs/ext4/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@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
15 #include <linux/time.h>
17 #include <linux/jbd2.h>
18 #include <linux/stat.h>
19 #include <linux/string.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
22 #include <linux/random.h>
23 #include <linux/bitops.h>
24 #include <linux/blkdev.h>
25 #include <asm/byteorder.h>
28 #include "ext4_jbd2.h"
32 #include <trace/events/ext4.h>
35 * ialloc.c contains the inodes allocation and deallocation routines
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
53 void mark_bitmap_end(int start_bit
, int end_bit
, char *bitmap
)
57 if (start_bit
>= end_bit
)
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit
, end_bit
);
61 for (i
= start_bit
; i
< ((start_bit
+ 7) & ~7UL); i
++)
62 ext4_set_bit(i
, bitmap
);
64 memset(bitmap
+ (i
>> 3), 0xff, (end_bit
- i
) >> 3);
67 /* Initializes an uninitialized inode bitmap */
68 unsigned ext4_init_inode_bitmap(struct super_block
*sb
, struct buffer_head
*bh
,
69 ext4_group_t block_group
,
70 struct ext4_group_desc
*gdp
)
72 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
74 J_ASSERT_BH(bh
, buffer_locked(bh
));
76 /* If checksum is bad mark all blocks and inodes use to prevent
77 * allocation, essentially implementing a per-group read-only flag. */
78 if (!ext4_group_desc_csum_verify(sbi
, block_group
, gdp
)) {
79 ext4_error(sb
, "Checksum bad for group %u", block_group
);
80 ext4_free_blks_set(sb
, gdp
, 0);
81 ext4_free_inodes_set(sb
, gdp
, 0);
82 ext4_itable_unused_set(sb
, gdp
, 0);
83 memset(bh
->b_data
, 0xff, sb
->s_blocksize
);
87 memset(bh
->b_data
, 0, (EXT4_INODES_PER_GROUP(sb
) + 7) / 8);
88 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
91 return EXT4_INODES_PER_GROUP(sb
);
95 * Read the inode allocation bitmap for a given block_group, reading
96 * into the specified slot in the superblock's bitmap cache.
98 * Return buffer_head of bitmap on success or NULL.
100 static struct buffer_head
*
101 ext4_read_inode_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
103 struct ext4_group_desc
*desc
;
104 struct buffer_head
*bh
= NULL
;
105 ext4_fsblk_t bitmap_blk
;
107 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
110 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
111 bh
= sb_getblk(sb
, bitmap_blk
);
113 ext4_error(sb
, "Cannot read inode bitmap - "
114 "block_group = %u, inode_bitmap = %llu",
115 block_group
, bitmap_blk
);
118 if (bitmap_uptodate(bh
))
122 if (bitmap_uptodate(bh
)) {
126 ext4_lock_group(sb
, block_group
);
127 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
128 ext4_init_inode_bitmap(sb
, bh
, block_group
, desc
);
129 set_bitmap_uptodate(bh
);
130 set_buffer_uptodate(bh
);
131 ext4_unlock_group(sb
, block_group
);
135 ext4_unlock_group(sb
, block_group
);
136 if (buffer_uptodate(bh
)) {
138 * if not uninit if bh is uptodate,
139 * bitmap is also uptodate
141 set_bitmap_uptodate(bh
);
146 * submit the buffer_head for read. We can
147 * safely mark the bitmap as uptodate now.
148 * We do it here so the bitmap uptodate bit
149 * get set with buffer lock held.
151 set_bitmap_uptodate(bh
);
152 if (bh_submit_read(bh
) < 0) {
154 ext4_error(sb
, "Cannot read inode bitmap - "
155 "block_group = %u, inode_bitmap = %llu",
156 block_group
, bitmap_blk
);
163 * NOTE! When we get the inode, we're the only people
164 * that have access to it, and as such there are no
165 * race conditions we have to worry about. The inode
166 * is not on the hash-lists, and it cannot be reached
167 * through the filesystem because the directory entry
168 * has been deleted earlier.
170 * HOWEVER: we must make sure that we get no aliases,
171 * which means that we have to call "clear_inode()"
172 * _before_ we mark the inode not in use in the inode
173 * bitmaps. Otherwise a newly created file might use
174 * the same inode number (not actually the same pointer
175 * though), and then we'd have two inodes sharing the
176 * same inode number and space on the harddisk.
178 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
180 struct super_block
*sb
= inode
->i_sb
;
183 struct buffer_head
*bitmap_bh
= NULL
;
184 struct buffer_head
*bh2
;
185 ext4_group_t block_group
;
187 struct ext4_group_desc
*gdp
;
188 struct ext4_super_block
*es
;
189 struct ext4_sb_info
*sbi
;
190 int fatal
= 0, err
, count
, cleared
;
192 if (atomic_read(&inode
->i_count
) > 1) {
193 printk(KERN_ERR
"ext4_free_inode: inode has count=%d\n",
194 atomic_read(&inode
->i_count
));
197 if (inode
->i_nlink
) {
198 printk(KERN_ERR
"ext4_free_inode: inode has nlink=%d\n",
203 printk(KERN_ERR
"ext4_free_inode: inode on "
204 "nonexistent device\n");
210 ext4_debug("freeing inode %lu\n", ino
);
211 trace_ext4_free_inode(inode
);
214 * Note: we must free any quota before locking the superblock,
215 * as writing the quota to disk may need the lock as well.
217 dquot_initialize(inode
);
218 ext4_xattr_delete_inode(handle
, inode
);
219 dquot_free_inode(inode
);
222 is_directory
= S_ISDIR(inode
->i_mode
);
224 /* Do this BEFORE marking the inode not in use or returning an error */
227 es
= EXT4_SB(sb
)->s_es
;
228 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
229 ext4_error(sb
, "reserved or nonexistent inode %lu", ino
);
232 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
233 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
234 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
238 BUFFER_TRACE(bitmap_bh
, "get_write_access");
239 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
244 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
246 BUFFER_TRACE(bh2
, "get_write_access");
247 fatal
= ext4_journal_get_write_access(handle
, bh2
);
249 ext4_lock_group(sb
, block_group
);
250 cleared
= ext4_clear_bit(bit
, bitmap_bh
->b_data
);
251 if (fatal
|| !cleared
) {
252 ext4_unlock_group(sb
, block_group
);
256 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
257 ext4_free_inodes_set(sb
, gdp
, count
);
259 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
260 ext4_used_dirs_set(sb
, gdp
, count
);
261 percpu_counter_dec(&sbi
->s_dirs_counter
);
263 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, block_group
, gdp
);
264 ext4_unlock_group(sb
, block_group
);
266 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
267 if (sbi
->s_log_groups_per_flex
) {
268 ext4_group_t f
= ext4_flex_group(sbi
, block_group
);
270 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
272 atomic_dec(&sbi
->s_flex_groups
[f
].used_dirs
);
274 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
275 fatal
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
278 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
279 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
284 ext4_error(sb
, "bit already cleared for inode %lu", ino
);
288 ext4_std_error(sb
, fatal
);
292 * There are two policies for allocating an inode. If the new inode is
293 * a directory, then a forward search is made for a block group with both
294 * free space and a low directory-to-inode ratio; if that fails, then of
295 * the groups with above-average free space, that group with the fewest
296 * directories already is chosen.
298 * For other inodes, search forward from the parent directory\'s block
299 * group to find a free inode.
301 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
,
302 ext4_group_t
*best_group
)
304 ext4_group_t ngroups
= ext4_get_groups_count(sb
);
305 unsigned int freei
, avefreei
;
306 struct ext4_group_desc
*desc
, *best_desc
= NULL
;
310 freei
= percpu_counter_read_positive(&EXT4_SB(sb
)->s_freeinodes_counter
);
311 avefreei
= freei
/ ngroups
;
313 for (group
= 0; group
< ngroups
; group
++) {
314 desc
= ext4_get_group_desc(sb
, group
, NULL
);
315 if (!desc
|| !ext4_free_inodes_count(sb
, desc
))
317 if (ext4_free_inodes_count(sb
, desc
) < avefreei
)
320 (ext4_free_blks_count(sb
, desc
) >
321 ext4_free_blks_count(sb
, best_desc
))) {
330 #define free_block_ratio 10
332 static int find_group_flex(struct super_block
*sb
, struct inode
*parent
,
333 ext4_group_t
*best_group
)
335 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
336 struct ext4_group_desc
*desc
;
337 struct flex_groups
*flex_group
= sbi
->s_flex_groups
;
338 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
339 ext4_group_t parent_fbg_group
= ext4_flex_group(sbi
, parent_group
);
340 ext4_group_t ngroups
= ext4_get_groups_count(sb
);
341 int flex_size
= ext4_flex_bg_size(sbi
);
342 ext4_group_t best_flex
= parent_fbg_group
;
343 int blocks_per_flex
= sbi
->s_blocks_per_group
* flex_size
;
344 int flexbg_free_blocks
;
345 int flex_freeb_ratio
;
346 ext4_group_t n_fbg_groups
;
349 n_fbg_groups
= (ngroups
+ flex_size
- 1) >>
350 sbi
->s_log_groups_per_flex
;
352 find_close_to_parent
:
353 flexbg_free_blocks
= atomic_read(&flex_group
[best_flex
].free_blocks
);
354 flex_freeb_ratio
= flexbg_free_blocks
* 100 / blocks_per_flex
;
355 if (atomic_read(&flex_group
[best_flex
].free_inodes
) &&
356 flex_freeb_ratio
> free_block_ratio
)
359 if (best_flex
&& best_flex
== parent_fbg_group
) {
361 goto find_close_to_parent
;
364 for (i
= 0; i
< n_fbg_groups
; i
++) {
365 if (i
== parent_fbg_group
|| i
== parent_fbg_group
- 1)
368 flexbg_free_blocks
= atomic_read(&flex_group
[i
].free_blocks
);
369 flex_freeb_ratio
= flexbg_free_blocks
* 100 / blocks_per_flex
;
371 if (flex_freeb_ratio
> free_block_ratio
&&
372 (atomic_read(&flex_group
[i
].free_inodes
))) {
377 if ((atomic_read(&flex_group
[best_flex
].free_inodes
) == 0) ||
378 ((atomic_read(&flex_group
[i
].free_blocks
) >
379 atomic_read(&flex_group
[best_flex
].free_blocks
)) &&
380 atomic_read(&flex_group
[i
].free_inodes
)))
384 if (!atomic_read(&flex_group
[best_flex
].free_inodes
) ||
385 !atomic_read(&flex_group
[best_flex
].free_blocks
))
389 for (i
= best_flex
* flex_size
; i
< ngroups
&&
390 i
< (best_flex
+ 1) * flex_size
; i
++) {
391 desc
= ext4_get_group_desc(sb
, i
, NULL
);
392 if (ext4_free_inodes_count(sb
, desc
)) {
410 * Helper function for Orlov's allocator; returns critical information
411 * for a particular block group or flex_bg. If flex_size is 1, then g
412 * is a block group number; otherwise it is flex_bg number.
414 void get_orlov_stats(struct super_block
*sb
, ext4_group_t g
,
415 int flex_size
, struct orlov_stats
*stats
)
417 struct ext4_group_desc
*desc
;
418 struct flex_groups
*flex_group
= EXT4_SB(sb
)->s_flex_groups
;
421 stats
->free_inodes
= atomic_read(&flex_group
[g
].free_inodes
);
422 stats
->free_blocks
= atomic_read(&flex_group
[g
].free_blocks
);
423 stats
->used_dirs
= atomic_read(&flex_group
[g
].used_dirs
);
427 desc
= ext4_get_group_desc(sb
, g
, NULL
);
429 stats
->free_inodes
= ext4_free_inodes_count(sb
, desc
);
430 stats
->free_blocks
= ext4_free_blks_count(sb
, desc
);
431 stats
->used_dirs
= ext4_used_dirs_count(sb
, desc
);
433 stats
->free_inodes
= 0;
434 stats
->free_blocks
= 0;
435 stats
->used_dirs
= 0;
440 * Orlov's allocator for directories.
442 * We always try to spread first-level directories.
444 * If there are blockgroups with both free inodes and free blocks counts
445 * not worse than average we return one with smallest directory count.
446 * Otherwise we simply return a random group.
448 * For the rest rules look so:
450 * It's OK to put directory into a group unless
451 * it has too many directories already (max_dirs) or
452 * it has too few free inodes left (min_inodes) or
453 * it has too few free blocks left (min_blocks) or
454 * Parent's group is preferred, if it doesn't satisfy these
455 * conditions we search cyclically through the rest. If none
456 * of the groups look good we just look for a group with more
457 * free inodes than average (starting at parent's group).
460 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
461 ext4_group_t
*group
, int mode
,
462 const struct qstr
*qstr
)
464 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
465 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
466 ext4_group_t real_ngroups
= ext4_get_groups_count(sb
);
467 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
468 unsigned int freei
, avefreei
;
469 ext4_fsblk_t freeb
, avefreeb
;
471 int max_dirs
, min_inodes
;
472 ext4_grpblk_t min_blocks
;
473 ext4_group_t i
, grp
, g
, ngroups
;
474 struct ext4_group_desc
*desc
;
475 struct orlov_stats stats
;
476 int flex_size
= ext4_flex_bg_size(sbi
);
477 struct dx_hash_info hinfo
;
479 ngroups
= real_ngroups
;
481 ngroups
= (real_ngroups
+ flex_size
- 1) >>
482 sbi
->s_log_groups_per_flex
;
483 parent_group
>>= sbi
->s_log_groups_per_flex
;
486 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
487 avefreei
= freei
/ ngroups
;
488 freeb
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
490 do_div(avefreeb
, ngroups
);
491 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
494 ((parent
== sb
->s_root
->d_inode
) ||
495 (EXT4_I(parent
)->i_flags
& EXT4_TOPDIR_FL
))) {
496 int best_ndir
= inodes_per_group
;
500 hinfo
.hash_version
= DX_HASH_HALF_MD4
;
501 hinfo
.seed
= sbi
->s_hash_seed
;
502 ext4fs_dirhash(qstr
->name
, qstr
->len
, &hinfo
);
505 get_random_bytes(&grp
, sizeof(grp
));
506 parent_group
= (unsigned)grp
% ngroups
;
507 for (i
= 0; i
< ngroups
; i
++) {
508 g
= (parent_group
+ i
) % ngroups
;
509 get_orlov_stats(sb
, g
, flex_size
, &stats
);
510 if (!stats
.free_inodes
)
512 if (stats
.used_dirs
>= best_ndir
)
514 if (stats
.free_inodes
< avefreei
)
516 if (stats
.free_blocks
< avefreeb
)
520 best_ndir
= stats
.used_dirs
;
525 if (flex_size
== 1) {
531 * We pack inodes at the beginning of the flexgroup's
532 * inode tables. Block allocation decisions will do
533 * something similar, although regular files will
534 * start at 2nd block group of the flexgroup. See
535 * ext4_ext_find_goal() and ext4_find_near().
538 for (i
= 0; i
< flex_size
; i
++) {
539 if (grp
+i
>= real_ngroups
)
541 desc
= ext4_get_group_desc(sb
, grp
+i
, NULL
);
542 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
550 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
551 min_inodes
= avefreei
- inodes_per_group
*flex_size
/ 4;
554 min_blocks
= avefreeb
- EXT4_BLOCKS_PER_GROUP(sb
)*flex_size
/ 4;
557 * Start looking in the flex group where we last allocated an
558 * inode for this parent directory
560 if (EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
561 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
563 parent_group
>>= sbi
->s_log_groups_per_flex
;
566 for (i
= 0; i
< ngroups
; i
++) {
567 grp
= (parent_group
+ i
) % ngroups
;
568 get_orlov_stats(sb
, grp
, flex_size
, &stats
);
569 if (stats
.used_dirs
>= max_dirs
)
571 if (stats
.free_inodes
< min_inodes
)
573 if (stats
.free_blocks
< min_blocks
)
579 ngroups
= real_ngroups
;
580 avefreei
= freei
/ ngroups
;
582 parent_group
= EXT4_I(parent
)->i_block_group
;
583 for (i
= 0; i
< ngroups
; i
++) {
584 grp
= (parent_group
+ i
) % ngroups
;
585 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
586 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
587 ext4_free_inodes_count(sb
, desc
) >= avefreei
) {
595 * The free-inodes counter is approximate, and for really small
596 * filesystems the above test can fail to find any blockgroups
605 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
606 ext4_group_t
*group
, int mode
)
608 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
609 ext4_group_t i
, last
, ngroups
= ext4_get_groups_count(sb
);
610 struct ext4_group_desc
*desc
;
611 int flex_size
= ext4_flex_bg_size(EXT4_SB(sb
));
614 * Try to place the inode is the same flex group as its
615 * parent. If we can't find space, use the Orlov algorithm to
616 * find another flex group, and store that information in the
617 * parent directory's inode information so that use that flex
618 * group for future allocations.
624 parent_group
&= ~(flex_size
-1);
625 last
= parent_group
+ flex_size
;
628 for (i
= parent_group
; i
< last
; i
++) {
629 desc
= ext4_get_group_desc(sb
, i
, NULL
);
630 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
635 if (!retry
&& EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
637 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
641 * If this didn't work, use the Orlov search algorithm
642 * to find a new flex group; we pass in the mode to
643 * avoid the topdir algorithms.
645 *group
= parent_group
+ flex_size
;
646 if (*group
> ngroups
)
648 return find_group_orlov(sb
, parent
, group
, mode
, 0);
652 * Try to place the inode in its parent directory
654 *group
= parent_group
;
655 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
656 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
657 ext4_free_blks_count(sb
, desc
))
661 * We're going to place this inode in a different blockgroup from its
662 * parent. We want to cause files in a common directory to all land in
663 * the same blockgroup. But we want files which are in a different
664 * directory which shares a blockgroup with our parent to land in a
665 * different blockgroup.
667 * So add our directory's i_ino into the starting point for the hash.
669 *group
= (*group
+ parent
->i_ino
) % ngroups
;
672 * Use a quadratic hash to find a group with a free inode and some free
675 for (i
= 1; i
< ngroups
; i
<<= 1) {
677 if (*group
>= ngroups
)
679 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
680 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
681 ext4_free_blks_count(sb
, desc
))
686 * That failed: try linear search for a free inode, even if that group
687 * has no free blocks.
689 *group
= parent_group
;
690 for (i
= 0; i
< ngroups
; i
++) {
691 if (++*group
>= ngroups
)
693 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
694 if (desc
&& ext4_free_inodes_count(sb
, desc
))
702 * claim the inode from the inode bitmap. If the group
703 * is uninit we need to take the groups's ext4_group_lock
704 * and clear the uninit flag. The inode bitmap update
705 * and group desc uninit flag clear should be done
706 * after holding ext4_group_lock so that ext4_read_inode_bitmap
707 * doesn't race with the ext4_claim_inode
709 static int ext4_claim_inode(struct super_block
*sb
,
710 struct buffer_head
*inode_bitmap_bh
,
711 unsigned long ino
, ext4_group_t group
, int mode
)
713 int free
= 0, retval
= 0, count
;
714 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
715 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, group
, NULL
);
717 ext4_lock_group(sb
, group
);
718 if (ext4_set_bit(ino
, inode_bitmap_bh
->b_data
)) {
719 /* not a free inode */
724 if ((group
== 0 && ino
< EXT4_FIRST_INO(sb
)) ||
725 ino
> EXT4_INODES_PER_GROUP(sb
)) {
726 ext4_unlock_group(sb
, group
);
727 ext4_error(sb
, "reserved inode or inode > inodes count - "
728 "block_group = %u, inode=%lu", group
,
729 ino
+ group
* EXT4_INODES_PER_GROUP(sb
));
732 /* If we didn't allocate from within the initialized part of the inode
733 * table then we need to initialize up to this inode. */
734 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
)) {
736 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
737 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
738 /* When marking the block group with
739 * ~EXT4_BG_INODE_UNINIT we don't want to depend
740 * on the value of bg_itable_unused even though
741 * mke2fs could have initialized the same for us.
742 * Instead we calculated the value below
747 free
= EXT4_INODES_PER_GROUP(sb
) -
748 ext4_itable_unused_count(sb
, gdp
);
752 * Check the relative inode number against the last used
753 * relative inode number in this group. if it is greater
754 * we need to update the bg_itable_unused count
758 ext4_itable_unused_set(sb
, gdp
,
759 (EXT4_INODES_PER_GROUP(sb
) - ino
));
761 count
= ext4_free_inodes_count(sb
, gdp
) - 1;
762 ext4_free_inodes_set(sb
, gdp
, count
);
764 count
= ext4_used_dirs_count(sb
, gdp
) + 1;
765 ext4_used_dirs_set(sb
, gdp
, count
);
766 if (sbi
->s_log_groups_per_flex
) {
767 ext4_group_t f
= ext4_flex_group(sbi
, group
);
769 atomic_inc(&sbi
->s_flex_groups
[f
].used_dirs
);
772 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
, gdp
);
774 ext4_unlock_group(sb
, group
);
779 * There are two policies for allocating an inode. If the new inode is
780 * a directory, then a forward search is made for a block group with both
781 * free space and a low directory-to-inode ratio; if that fails, then of
782 * the groups with above-average free space, that group with the fewest
783 * directories already is chosen.
785 * For other inodes, search forward from the parent directory's block
786 * group to find a free inode.
788 struct inode
*ext4_new_inode(handle_t
*handle
, struct inode
*dir
, int mode
,
789 const struct qstr
*qstr
, __u32 goal
)
791 struct super_block
*sb
;
792 struct buffer_head
*inode_bitmap_bh
= NULL
;
793 struct buffer_head
*group_desc_bh
;
794 ext4_group_t ngroups
, group
= 0;
795 unsigned long ino
= 0;
797 struct ext4_group_desc
*gdp
= NULL
;
798 struct ext4_inode_info
*ei
;
799 struct ext4_sb_info
*sbi
;
805 ext4_group_t flex_group
;
807 /* Cannot create files in a deleted directory */
808 if (!dir
|| !dir
->i_nlink
)
809 return ERR_PTR(-EPERM
);
812 ngroups
= ext4_get_groups_count(sb
);
813 trace_ext4_request_inode(dir
, mode
);
814 inode
= new_inode(sb
);
816 return ERR_PTR(-ENOMEM
);
821 goal
= sbi
->s_inode_goal
;
823 if (goal
&& goal
<= le32_to_cpu(sbi
->s_es
->s_inodes_count
)) {
824 group
= (goal
- 1) / EXT4_INODES_PER_GROUP(sb
);
825 ino
= (goal
- 1) % EXT4_INODES_PER_GROUP(sb
);
830 if (sbi
->s_log_groups_per_flex
&& test_opt(sb
, OLDALLOC
)) {
831 ret2
= find_group_flex(sb
, dir
, &group
);
833 ret2
= find_group_other(sb
, dir
, &group
, mode
);
834 if (ret2
== 0 && once
) {
836 printk(KERN_NOTICE
"ext4: find_group_flex "
837 "failed, fallback succeeded dir %lu\n",
845 if (test_opt(sb
, OLDALLOC
))
846 ret2
= find_group_dir(sb
, dir
, &group
);
848 ret2
= find_group_orlov(sb
, dir
, &group
, mode
, qstr
);
850 ret2
= find_group_other(sb
, dir
, &group
, mode
);
853 EXT4_I(dir
)->i_last_alloc_group
= group
;
858 for (i
= 0; i
< ngroups
; i
++, ino
= 0) {
861 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
865 brelse(inode_bitmap_bh
);
866 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
867 if (!inode_bitmap_bh
)
870 repeat_in_this_group
:
871 ino
= ext4_find_next_zero_bit((unsigned long *)
872 inode_bitmap_bh
->b_data
,
873 EXT4_INODES_PER_GROUP(sb
), ino
);
875 if (ino
< EXT4_INODES_PER_GROUP(sb
)) {
877 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
878 err
= ext4_journal_get_write_access(handle
,
883 BUFFER_TRACE(group_desc_bh
, "get_write_access");
884 err
= ext4_journal_get_write_access(handle
,
888 if (!ext4_claim_inode(sb
, inode_bitmap_bh
,
891 BUFFER_TRACE(inode_bitmap_bh
,
892 "call ext4_handle_dirty_metadata");
893 err
= ext4_handle_dirty_metadata(handle
,
898 /* zero bit is inode number 1*/
903 ext4_handle_release_buffer(handle
, inode_bitmap_bh
);
904 ext4_handle_release_buffer(handle
, group_desc_bh
);
906 if (++ino
< EXT4_INODES_PER_GROUP(sb
))
907 goto repeat_in_this_group
;
911 * This case is possible in concurrent environment. It is very
912 * rare. We cannot repeat the find_group_xxx() call because
913 * that will simply return the same blockgroup, because the
914 * group descriptor metadata has not yet been updated.
915 * So we just go onto the next blockgroup.
917 if (++group
== ngroups
)
924 /* We may have to initialize the block bitmap if it isn't already */
925 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
) &&
926 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
927 struct buffer_head
*block_bitmap_bh
;
929 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
930 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
931 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
933 brelse(block_bitmap_bh
);
938 ext4_lock_group(sb
, group
);
939 /* recheck and clear flag under lock if we still need to */
940 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
941 free
= ext4_free_blocks_after_init(sb
, group
, gdp
);
942 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
943 ext4_free_blks_set(sb
, gdp
, free
);
944 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
,
947 ext4_unlock_group(sb
, group
);
949 /* Don't need to dirty bitmap block if we didn't change it */
951 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
952 err
= ext4_handle_dirty_metadata(handle
,
953 NULL
, block_bitmap_bh
);
956 brelse(block_bitmap_bh
);
960 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
961 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
965 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
967 percpu_counter_inc(&sbi
->s_dirs_counter
);
970 if (sbi
->s_log_groups_per_flex
) {
971 flex_group
= ext4_flex_group(sbi
, group
);
972 atomic_dec(&sbi
->s_flex_groups
[flex_group
].free_inodes
);
975 inode
->i_uid
= current_fsuid();
976 if (test_opt(sb
, GRPID
))
977 inode
->i_gid
= dir
->i_gid
;
978 else if (dir
->i_mode
& S_ISGID
) {
979 inode
->i_gid
= dir
->i_gid
;
983 inode
->i_gid
= current_fsgid();
984 inode
->i_mode
= mode
;
986 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
987 /* This is the optimal IO size (for stat), not the fs block size */
989 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
990 ext4_current_time(inode
);
992 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
993 ei
->i_dir_start_lookup
= 0;
997 * Don't inherit extent flag from directory, amongst others. We set
998 * extent flag on newly created directory and file only if -o extent
999 * mount option is specified
1002 ext4_mask_flags(mode
, EXT4_I(dir
)->i_flags
& EXT4_FL_INHERITED
);
1005 ei
->i_block_group
= group
;
1006 ei
->i_last_alloc_group
= ~0;
1008 ext4_set_inode_flags(inode
);
1009 if (IS_DIRSYNC(inode
))
1010 ext4_handle_sync(handle
);
1011 if (insert_inode_locked(inode
) < 0) {
1015 spin_lock(&sbi
->s_next_gen_lock
);
1016 inode
->i_generation
= sbi
->s_next_generation
++;
1017 spin_unlock(&sbi
->s_next_gen_lock
);
1019 ei
->i_state_flags
= 0;
1020 ext4_set_inode_state(inode
, EXT4_STATE_NEW
);
1022 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
1025 dquot_initialize(inode
);
1026 err
= dquot_alloc_inode(inode
);
1030 err
= ext4_init_acl(handle
, inode
, dir
);
1032 goto fail_free_drop
;
1034 err
= ext4_init_security(handle
, inode
, dir
);
1036 goto fail_free_drop
;
1038 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
1039 /* set extent flag only for directory, file and normal symlink*/
1040 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
1041 EXT4_I(inode
)->i_flags
|= EXT4_EXTENTS_FL
;
1042 ext4_ext_tree_init(handle
, inode
);
1046 err
= ext4_mark_inode_dirty(handle
, inode
);
1048 ext4_std_error(sb
, err
);
1049 goto fail_free_drop
;
1052 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
1053 trace_ext4_allocate_inode(inode
, dir
, mode
);
1056 ext4_std_error(sb
, err
);
1061 brelse(inode_bitmap_bh
);
1065 dquot_free_inode(inode
);
1069 inode
->i_flags
|= S_NOQUOTA
;
1071 unlock_new_inode(inode
);
1073 brelse(inode_bitmap_bh
);
1074 return ERR_PTR(err
);
1077 /* Verify that we are loading a valid orphan from disk */
1078 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
1080 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
1081 ext4_group_t block_group
;
1083 struct buffer_head
*bitmap_bh
;
1084 struct inode
*inode
= NULL
;
1087 /* Error cases - e2fsck has already cleaned up for us */
1088 if (ino
> max_ino
) {
1089 ext4_warning(sb
, "bad orphan ino %lu! e2fsck was run?", ino
);
1093 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
1094 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
1095 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
1097 ext4_warning(sb
, "inode bitmap error for orphan %lu", ino
);
1101 /* Having the inode bit set should be a 100% indicator that this
1102 * is a valid orphan (no e2fsck run on fs). Orphans also include
1103 * inodes that were being truncated, so we can't check i_nlink==0.
1105 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
1108 inode
= ext4_iget(sb
, ino
);
1113 * If the orphans has i_nlinks > 0 then it should be able to be
1114 * truncated, otherwise it won't be removed from the orphan list
1115 * during processing and an infinite loop will result.
1117 if (inode
->i_nlink
&& !ext4_can_truncate(inode
))
1120 if (NEXT_ORPHAN(inode
) > max_ino
)
1126 err
= PTR_ERR(inode
);
1129 ext4_warning(sb
, "bad orphan inode %lu! e2fsck was run?", ino
);
1130 printk(KERN_NOTICE
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
1131 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
1132 ext4_test_bit(bit
, bitmap_bh
->b_data
));
1133 printk(KERN_NOTICE
"inode=%p\n", inode
);
1135 printk(KERN_NOTICE
"is_bad_inode(inode)=%d\n",
1136 is_bad_inode(inode
));
1137 printk(KERN_NOTICE
"NEXT_ORPHAN(inode)=%u\n",
1138 NEXT_ORPHAN(inode
));
1139 printk(KERN_NOTICE
"max_ino=%lu\n", max_ino
);
1140 printk(KERN_NOTICE
"i_nlink=%u\n", inode
->i_nlink
);
1141 /* Avoid freeing blocks if we got a bad deleted inode */
1142 if (inode
->i_nlink
== 0)
1143 inode
->i_blocks
= 0;
1148 return ERR_PTR(err
);
1151 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
1153 unsigned long desc_count
;
1154 struct ext4_group_desc
*gdp
;
1155 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1157 struct ext4_super_block
*es
;
1158 unsigned long bitmap_count
, x
;
1159 struct buffer_head
*bitmap_bh
= NULL
;
1161 es
= EXT4_SB(sb
)->s_es
;
1165 for (i
= 0; i
< ngroups
; i
++) {
1166 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1169 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1171 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1175 x
= ext4_count_free(bitmap_bh
, EXT4_INODES_PER_GROUP(sb
) / 8);
1176 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1177 (unsigned long) i
, ext4_free_inodes_count(sb
, gdp
), x
);
1181 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1182 "stored = %u, computed = %lu, %lu\n",
1183 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1187 for (i
= 0; i
< ngroups
; i
++) {
1188 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1191 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1198 /* Called at mount-time, super-block is locked */
1199 unsigned long ext4_count_dirs(struct super_block
* sb
)
1201 unsigned long count
= 0;
1202 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1204 for (i
= 0; i
< ngroups
; i
++) {
1205 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1208 count
+= ext4_used_dirs_count(sb
, gdp
);