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/stat.h>
18 #include <linux/string.h>
19 #include <linux/quotaops.h>
20 #include <linux/buffer_head.h>
21 #include <linux/random.h>
22 #include <linux/bitops.h>
23 #include <linux/blkdev.h>
24 #include <asm/byteorder.h>
27 #include "ext4_jbd2.h"
31 #include <trace/events/ext4.h>
34 * ialloc.c contains the inodes allocation and deallocation routines
38 * The free inodes are managed by bitmaps. A file system contains several
39 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
40 * block for inodes, N blocks for the inode table and data blocks.
42 * The file system contains group descriptors which are located after the
43 * super block. Each descriptor contains the number of the bitmap block and
44 * the free blocks count in the block.
48 * To avoid calling the atomic setbit hundreds or thousands of times, we only
49 * need to use it within a single byte (to ensure we get endianness right).
50 * We can use memset for the rest of the bitmap as there are no other users.
52 void ext4_mark_bitmap_end(int start_bit
, int end_bit
, char *bitmap
)
56 if (start_bit
>= end_bit
)
59 ext4_debug("mark end bits +%d through +%d used\n", start_bit
, end_bit
);
60 for (i
= start_bit
; i
< ((start_bit
+ 7) & ~7UL); i
++)
61 ext4_set_bit(i
, bitmap
);
63 memset(bitmap
+ (i
>> 3), 0xff, (end_bit
- i
) >> 3);
66 /* Initializes an uninitialized inode bitmap */
67 static int ext4_init_inode_bitmap(struct super_block
*sb
,
68 struct buffer_head
*bh
,
69 ext4_group_t block_group
,
70 struct ext4_group_desc
*gdp
)
72 struct ext4_group_info
*grp
;
73 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(sb
, block_group
, gdp
)) {
79 ext4_error(sb
, "Checksum bad for group %u", block_group
);
80 grp
= ext4_get_group_info(sb
, block_group
);
81 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp
))
82 percpu_counter_sub(&sbi
->s_freeclusters_counter
,
84 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
85 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
87 count
= ext4_free_inodes_count(sb
, gdp
);
88 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
91 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
95 memset(bh
->b_data
, 0, (EXT4_INODES_PER_GROUP(sb
) + 7) / 8);
96 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
98 ext4_inode_bitmap_csum_set(sb
, block_group
, gdp
, bh
,
99 EXT4_INODES_PER_GROUP(sb
) / 8);
100 ext4_group_desc_csum_set(sb
, block_group
, gdp
);
105 void ext4_end_bitmap_read(struct buffer_head
*bh
, int uptodate
)
108 set_buffer_uptodate(bh
);
109 set_bitmap_uptodate(bh
);
115 static int ext4_validate_inode_bitmap(struct super_block
*sb
,
116 struct ext4_group_desc
*desc
,
117 ext4_group_t block_group
,
118 struct buffer_head
*bh
)
121 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, block_group
);
122 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
124 if (buffer_verified(bh
))
126 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))
127 return -EFSCORRUPTED
;
129 ext4_lock_group(sb
, block_group
);
130 blk
= ext4_inode_bitmap(sb
, desc
);
131 if (!ext4_inode_bitmap_csum_verify(sb
, block_group
, desc
, bh
,
132 EXT4_INODES_PER_GROUP(sb
) / 8)) {
133 ext4_unlock_group(sb
, block_group
);
134 ext4_error(sb
, "Corrupt inode bitmap - block_group = %u, "
135 "inode_bitmap = %llu", block_group
, blk
);
136 grp
= ext4_get_group_info(sb
, block_group
);
137 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
139 count
= ext4_free_inodes_count(sb
, desc
);
140 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
143 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
146 set_buffer_verified(bh
);
147 ext4_unlock_group(sb
, block_group
);
152 * Read the inode allocation bitmap for a given block_group, reading
153 * into the specified slot in the superblock's bitmap cache.
155 * Return buffer_head of bitmap on success or NULL.
157 static struct buffer_head
*
158 ext4_read_inode_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
160 struct ext4_group_desc
*desc
;
161 struct buffer_head
*bh
= NULL
;
162 ext4_fsblk_t bitmap_blk
;
165 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
167 return ERR_PTR(-EFSCORRUPTED
);
169 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
170 bh
= sb_getblk(sb
, bitmap_blk
);
172 ext4_error(sb
, "Cannot read inode bitmap - "
173 "block_group = %u, inode_bitmap = %llu",
174 block_group
, bitmap_blk
);
175 return ERR_PTR(-EIO
);
177 if (bitmap_uptodate(bh
))
181 if (bitmap_uptodate(bh
)) {
186 ext4_lock_group(sb
, block_group
);
187 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
188 err
= ext4_init_inode_bitmap(sb
, bh
, block_group
, desc
);
189 set_bitmap_uptodate(bh
);
190 set_buffer_uptodate(bh
);
191 set_buffer_verified(bh
);
192 ext4_unlock_group(sb
, block_group
);
198 ext4_unlock_group(sb
, block_group
);
200 if (buffer_uptodate(bh
)) {
202 * if not uninit if bh is uptodate,
203 * bitmap is also uptodate
205 set_bitmap_uptodate(bh
);
210 * submit the buffer_head for reading
212 trace_ext4_load_inode_bitmap(sb
, block_group
);
213 bh
->b_end_io
= ext4_end_bitmap_read
;
215 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
217 if (!buffer_uptodate(bh
)) {
219 ext4_error(sb
, "Cannot read inode bitmap - "
220 "block_group = %u, inode_bitmap = %llu",
221 block_group
, bitmap_blk
);
222 return ERR_PTR(-EIO
);
226 err
= ext4_validate_inode_bitmap(sb
, desc
, block_group
, bh
);
236 * NOTE! When we get the inode, we're the only people
237 * that have access to it, and as such there are no
238 * race conditions we have to worry about. The inode
239 * is not on the hash-lists, and it cannot be reached
240 * through the filesystem because the directory entry
241 * has been deleted earlier.
243 * HOWEVER: we must make sure that we get no aliases,
244 * which means that we have to call "clear_inode()"
245 * _before_ we mark the inode not in use in the inode
246 * bitmaps. Otherwise a newly created file might use
247 * the same inode number (not actually the same pointer
248 * though), and then we'd have two inodes sharing the
249 * same inode number and space on the harddisk.
251 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
253 struct super_block
*sb
= inode
->i_sb
;
256 struct buffer_head
*bitmap_bh
= NULL
;
257 struct buffer_head
*bh2
;
258 ext4_group_t block_group
;
260 struct ext4_group_desc
*gdp
;
261 struct ext4_super_block
*es
;
262 struct ext4_sb_info
*sbi
;
263 int fatal
= 0, err
, count
, cleared
;
264 struct ext4_group_info
*grp
;
267 printk(KERN_ERR
"EXT4-fs: %s:%d: inode on "
268 "nonexistent device\n", __func__
, __LINE__
);
271 if (atomic_read(&inode
->i_count
) > 1) {
272 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: count=%d",
273 __func__
, __LINE__
, inode
->i_ino
,
274 atomic_read(&inode
->i_count
));
277 if (inode
->i_nlink
) {
278 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: nlink=%d\n",
279 __func__
, __LINE__
, inode
->i_ino
, inode
->i_nlink
);
285 ext4_debug("freeing inode %lu\n", ino
);
286 trace_ext4_free_inode(inode
);
289 * Note: we must free any quota before locking the superblock,
290 * as writing the quota to disk may need the lock as well.
292 dquot_initialize(inode
);
293 ext4_xattr_delete_inode(handle
, inode
);
294 dquot_free_inode(inode
);
297 is_directory
= S_ISDIR(inode
->i_mode
);
299 /* Do this BEFORE marking the inode not in use or returning an error */
300 ext4_clear_inode(inode
);
302 es
= EXT4_SB(sb
)->s_es
;
303 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
304 ext4_error(sb
, "reserved or nonexistent inode %lu", ino
);
307 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
308 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
309 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
310 /* Don't bother if the inode bitmap is corrupt. */
311 grp
= ext4_get_group_info(sb
, block_group
);
312 if (IS_ERR(bitmap_bh
)) {
313 fatal
= PTR_ERR(bitmap_bh
);
317 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))) {
318 fatal
= -EFSCORRUPTED
;
322 BUFFER_TRACE(bitmap_bh
, "get_write_access");
323 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
328 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
330 BUFFER_TRACE(bh2
, "get_write_access");
331 fatal
= ext4_journal_get_write_access(handle
, bh2
);
333 ext4_lock_group(sb
, block_group
);
334 cleared
= ext4_test_and_clear_bit(bit
, bitmap_bh
->b_data
);
335 if (fatal
|| !cleared
) {
336 ext4_unlock_group(sb
, block_group
);
340 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
341 ext4_free_inodes_set(sb
, gdp
, count
);
343 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
344 ext4_used_dirs_set(sb
, gdp
, count
);
345 percpu_counter_dec(&sbi
->s_dirs_counter
);
347 ext4_inode_bitmap_csum_set(sb
, block_group
, gdp
, bitmap_bh
,
348 EXT4_INODES_PER_GROUP(sb
) / 8);
349 ext4_group_desc_csum_set(sb
, block_group
, gdp
);
350 ext4_unlock_group(sb
, block_group
);
352 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
353 if (sbi
->s_log_groups_per_flex
) {
354 ext4_group_t f
= ext4_flex_group(sbi
, block_group
);
356 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
358 atomic_dec(&sbi
->s_flex_groups
[f
].used_dirs
);
360 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
361 fatal
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
364 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
365 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
369 ext4_error(sb
, "bit already cleared for inode %lu", ino
);
370 if (gdp
&& !EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
372 count
= ext4_free_inodes_count(sb
, gdp
);
373 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
376 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
381 ext4_std_error(sb
, fatal
);
391 * Helper function for Orlov's allocator; returns critical information
392 * for a particular block group or flex_bg. If flex_size is 1, then g
393 * is a block group number; otherwise it is flex_bg number.
395 static void get_orlov_stats(struct super_block
*sb
, ext4_group_t g
,
396 int flex_size
, struct orlov_stats
*stats
)
398 struct ext4_group_desc
*desc
;
399 struct flex_groups
*flex_group
= EXT4_SB(sb
)->s_flex_groups
;
402 stats
->free_inodes
= atomic_read(&flex_group
[g
].free_inodes
);
403 stats
->free_clusters
= atomic64_read(&flex_group
[g
].free_clusters
);
404 stats
->used_dirs
= atomic_read(&flex_group
[g
].used_dirs
);
408 desc
= ext4_get_group_desc(sb
, g
, NULL
);
410 stats
->free_inodes
= ext4_free_inodes_count(sb
, desc
);
411 stats
->free_clusters
= ext4_free_group_clusters(sb
, desc
);
412 stats
->used_dirs
= ext4_used_dirs_count(sb
, desc
);
414 stats
->free_inodes
= 0;
415 stats
->free_clusters
= 0;
416 stats
->used_dirs
= 0;
421 * Orlov's allocator for directories.
423 * We always try to spread first-level directories.
425 * If there are blockgroups with both free inodes and free blocks counts
426 * not worse than average we return one with smallest directory count.
427 * Otherwise we simply return a random group.
429 * For the rest rules look so:
431 * It's OK to put directory into a group unless
432 * it has too many directories already (max_dirs) or
433 * it has too few free inodes left (min_inodes) or
434 * it has too few free blocks left (min_blocks) or
435 * Parent's group is preferred, if it doesn't satisfy these
436 * conditions we search cyclically through the rest. If none
437 * of the groups look good we just look for a group with more
438 * free inodes than average (starting at parent's group).
441 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
442 ext4_group_t
*group
, umode_t mode
,
443 const struct qstr
*qstr
)
445 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
446 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
447 ext4_group_t real_ngroups
= ext4_get_groups_count(sb
);
448 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
449 unsigned int freei
, avefreei
, grp_free
;
450 ext4_fsblk_t freeb
, avefreec
;
452 int max_dirs
, min_inodes
;
453 ext4_grpblk_t min_clusters
;
454 ext4_group_t i
, grp
, g
, ngroups
;
455 struct ext4_group_desc
*desc
;
456 struct orlov_stats stats
;
457 int flex_size
= ext4_flex_bg_size(sbi
);
458 struct dx_hash_info hinfo
;
460 ngroups
= real_ngroups
;
462 ngroups
= (real_ngroups
+ flex_size
- 1) >>
463 sbi
->s_log_groups_per_flex
;
464 parent_group
>>= sbi
->s_log_groups_per_flex
;
467 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
468 avefreei
= freei
/ ngroups
;
469 freeb
= EXT4_C2B(sbi
,
470 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
472 do_div(avefreec
, ngroups
);
473 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
476 ((parent
== d_inode(sb
->s_root
)) ||
477 (ext4_test_inode_flag(parent
, EXT4_INODE_TOPDIR
)))) {
478 int best_ndir
= inodes_per_group
;
482 hinfo
.hash_version
= DX_HASH_HALF_MD4
;
483 hinfo
.seed
= sbi
->s_hash_seed
;
484 ext4fs_dirhash(qstr
->name
, qstr
->len
, &hinfo
);
488 parent_group
= (unsigned)grp
% ngroups
;
489 for (i
= 0; i
< ngroups
; i
++) {
490 g
= (parent_group
+ i
) % ngroups
;
491 get_orlov_stats(sb
, g
, flex_size
, &stats
);
492 if (!stats
.free_inodes
)
494 if (stats
.used_dirs
>= best_ndir
)
496 if (stats
.free_inodes
< avefreei
)
498 if (stats
.free_clusters
< avefreec
)
502 best_ndir
= stats
.used_dirs
;
507 if (flex_size
== 1) {
513 * We pack inodes at the beginning of the flexgroup's
514 * inode tables. Block allocation decisions will do
515 * something similar, although regular files will
516 * start at 2nd block group of the flexgroup. See
517 * ext4_ext_find_goal() and ext4_find_near().
520 for (i
= 0; i
< flex_size
; i
++) {
521 if (grp
+i
>= real_ngroups
)
523 desc
= ext4_get_group_desc(sb
, grp
+i
, NULL
);
524 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
532 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
533 min_inodes
= avefreei
- inodes_per_group
*flex_size
/ 4;
536 min_clusters
= avefreec
- EXT4_CLUSTERS_PER_GROUP(sb
)*flex_size
/ 4;
539 * Start looking in the flex group where we last allocated an
540 * inode for this parent directory
542 if (EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
543 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
545 parent_group
>>= sbi
->s_log_groups_per_flex
;
548 for (i
= 0; i
< ngroups
; i
++) {
549 grp
= (parent_group
+ i
) % ngroups
;
550 get_orlov_stats(sb
, grp
, flex_size
, &stats
);
551 if (stats
.used_dirs
>= max_dirs
)
553 if (stats
.free_inodes
< min_inodes
)
555 if (stats
.free_clusters
< min_clusters
)
561 ngroups
= real_ngroups
;
562 avefreei
= freei
/ ngroups
;
564 parent_group
= EXT4_I(parent
)->i_block_group
;
565 for (i
= 0; i
< ngroups
; i
++) {
566 grp
= (parent_group
+ i
) % ngroups
;
567 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
569 grp_free
= ext4_free_inodes_count(sb
, desc
);
570 if (grp_free
&& grp_free
>= avefreei
) {
579 * The free-inodes counter is approximate, and for really small
580 * filesystems the above test can fail to find any blockgroups
589 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
590 ext4_group_t
*group
, umode_t mode
)
592 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
593 ext4_group_t i
, last
, ngroups
= ext4_get_groups_count(sb
);
594 struct ext4_group_desc
*desc
;
595 int flex_size
= ext4_flex_bg_size(EXT4_SB(sb
));
598 * Try to place the inode is the same flex group as its
599 * parent. If we can't find space, use the Orlov algorithm to
600 * find another flex group, and store that information in the
601 * parent directory's inode information so that use that flex
602 * group for future allocations.
608 parent_group
&= ~(flex_size
-1);
609 last
= parent_group
+ flex_size
;
612 for (i
= parent_group
; i
< last
; i
++) {
613 desc
= ext4_get_group_desc(sb
, i
, NULL
);
614 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
619 if (!retry
&& EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
621 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
625 * If this didn't work, use the Orlov search algorithm
626 * to find a new flex group; we pass in the mode to
627 * avoid the topdir algorithms.
629 *group
= parent_group
+ flex_size
;
630 if (*group
> ngroups
)
632 return find_group_orlov(sb
, parent
, group
, mode
, NULL
);
636 * Try to place the inode in its parent directory
638 *group
= parent_group
;
639 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
640 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
641 ext4_free_group_clusters(sb
, desc
))
645 * We're going to place this inode in a different blockgroup from its
646 * parent. We want to cause files in a common directory to all land in
647 * the same blockgroup. But we want files which are in a different
648 * directory which shares a blockgroup with our parent to land in a
649 * different blockgroup.
651 * So add our directory's i_ino into the starting point for the hash.
653 *group
= (*group
+ parent
->i_ino
) % ngroups
;
656 * Use a quadratic hash to find a group with a free inode and some free
659 for (i
= 1; i
< ngroups
; i
<<= 1) {
661 if (*group
>= ngroups
)
663 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
664 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
665 ext4_free_group_clusters(sb
, desc
))
670 * That failed: try linear search for a free inode, even if that group
671 * has no free blocks.
673 *group
= parent_group
;
674 for (i
= 0; i
< ngroups
; i
++) {
675 if (++*group
>= ngroups
)
677 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
678 if (desc
&& ext4_free_inodes_count(sb
, desc
))
686 * In no journal mode, if an inode has recently been deleted, we want
687 * to avoid reusing it until we're reasonably sure the inode table
688 * block has been written back to disk. (Yes, these values are
689 * somewhat arbitrary...)
691 #define RECENTCY_MIN 5
692 #define RECENTCY_DIRTY 30
694 static int recently_deleted(struct super_block
*sb
, ext4_group_t group
, int ino
)
696 struct ext4_group_desc
*gdp
;
697 struct ext4_inode
*raw_inode
;
698 struct buffer_head
*bh
;
699 unsigned long dtime
, now
;
700 int inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
701 int offset
, ret
= 0, recentcy
= RECENTCY_MIN
;
703 gdp
= ext4_get_group_desc(sb
, group
, NULL
);
707 bh
= sb_getblk(sb
, ext4_inode_table(sb
, gdp
) +
708 (ino
/ inodes_per_block
));
709 if (unlikely(!bh
) || !buffer_uptodate(bh
))
711 * If the block is not in the buffer cache, then it
712 * must have been written out.
716 offset
= (ino
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
717 raw_inode
= (struct ext4_inode
*) (bh
->b_data
+ offset
);
718 dtime
= le32_to_cpu(raw_inode
->i_dtime
);
720 if (buffer_dirty(bh
))
721 recentcy
+= RECENTCY_DIRTY
;
723 if (dtime
&& (dtime
< now
) && (now
< dtime
+ recentcy
))
731 * There are two policies for allocating an inode. If the new inode is
732 * a directory, then a forward search is made for a block group with both
733 * free space and a low directory-to-inode ratio; if that fails, then of
734 * the groups with above-average free space, that group with the fewest
735 * directories already is chosen.
737 * For other inodes, search forward from the parent directory's block
738 * group to find a free inode.
740 struct inode
*__ext4_new_inode(handle_t
*handle
, struct inode
*dir
,
741 umode_t mode
, const struct qstr
*qstr
,
742 __u32 goal
, uid_t
*owner
, int handle_type
,
743 unsigned int line_no
, int nblocks
)
745 struct super_block
*sb
;
746 struct buffer_head
*inode_bitmap_bh
= NULL
;
747 struct buffer_head
*group_desc_bh
;
748 ext4_group_t ngroups
, group
= 0;
749 unsigned long ino
= 0;
751 struct ext4_group_desc
*gdp
= NULL
;
752 struct ext4_inode_info
*ei
;
753 struct ext4_sb_info
*sbi
;
757 ext4_group_t flex_group
;
758 struct ext4_group_info
*grp
;
761 /* Cannot create files in a deleted directory */
762 if (!dir
|| !dir
->i_nlink
)
763 return ERR_PTR(-EPERM
);
765 if ((ext4_encrypted_inode(dir
) ||
766 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir
->i_sb
))) &&
767 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
768 err
= ext4_get_encryption_info(dir
);
771 if (ext4_encryption_info(dir
) == NULL
)
772 return ERR_PTR(-EPERM
);
774 nblocks
+= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
);
779 ngroups
= ext4_get_groups_count(sb
);
780 trace_ext4_request_inode(dir
, mode
);
781 inode
= new_inode(sb
);
783 return ERR_PTR(-ENOMEM
);
788 * Initalize owners and quota early so that we don't have to account
789 * for quota initialization worst case in standard inode creating
793 inode
->i_mode
= mode
;
794 i_uid_write(inode
, owner
[0]);
795 i_gid_write(inode
, owner
[1]);
796 } else if (test_opt(sb
, GRPID
)) {
797 inode
->i_mode
= mode
;
798 inode
->i_uid
= current_fsuid();
799 inode
->i_gid
= dir
->i_gid
;
801 inode_init_owner(inode
, dir
, mode
);
802 err
= dquot_initialize(inode
);
807 goal
= sbi
->s_inode_goal
;
809 if (goal
&& goal
<= le32_to_cpu(sbi
->s_es
->s_inodes_count
)) {
810 group
= (goal
- 1) / EXT4_INODES_PER_GROUP(sb
);
811 ino
= (goal
- 1) % EXT4_INODES_PER_GROUP(sb
);
817 ret2
= find_group_orlov(sb
, dir
, &group
, mode
, qstr
);
819 ret2
= find_group_other(sb
, dir
, &group
, mode
);
822 EXT4_I(dir
)->i_last_alloc_group
= group
;
828 * Normally we will only go through one pass of this loop,
829 * unless we get unlucky and it turns out the group we selected
830 * had its last inode grabbed by someone else.
832 for (i
= 0; i
< ngroups
; i
++, ino
= 0) {
835 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
840 * Check free inodes count before loading bitmap.
842 if (ext4_free_inodes_count(sb
, gdp
) == 0) {
843 if (++group
== ngroups
)
848 grp
= ext4_get_group_info(sb
, group
);
849 /* Skip groups with already-known suspicious inode tables */
850 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
851 if (++group
== ngroups
)
856 brelse(inode_bitmap_bh
);
857 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
858 /* Skip groups with suspicious inode tables */
859 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
) ||
860 IS_ERR(inode_bitmap_bh
)) {
861 inode_bitmap_bh
= NULL
;
862 if (++group
== ngroups
)
867 repeat_in_this_group
:
868 ino
= ext4_find_next_zero_bit((unsigned long *)
869 inode_bitmap_bh
->b_data
,
870 EXT4_INODES_PER_GROUP(sb
), ino
);
871 if (ino
>= EXT4_INODES_PER_GROUP(sb
))
873 if (group
== 0 && (ino
+1) < EXT4_FIRST_INO(sb
)) {
874 ext4_error(sb
, "reserved inode found cleared - "
875 "inode=%lu", ino
+ 1);
878 if ((EXT4_SB(sb
)->s_journal
== NULL
) &&
879 recently_deleted(sb
, group
, ino
)) {
884 BUG_ON(nblocks
<= 0);
885 handle
= __ext4_journal_start_sb(dir
->i_sb
, line_no
,
886 handle_type
, nblocks
,
888 if (IS_ERR(handle
)) {
889 err
= PTR_ERR(handle
);
890 ext4_std_error(sb
, err
);
894 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
895 err
= ext4_journal_get_write_access(handle
, inode_bitmap_bh
);
897 ext4_std_error(sb
, err
);
900 ext4_lock_group(sb
, group
);
901 ret2
= ext4_test_and_set_bit(ino
, inode_bitmap_bh
->b_data
);
902 ext4_unlock_group(sb
, group
);
903 ino
++; /* the inode bitmap is zero-based */
905 goto got
; /* we grabbed the inode! */
907 if (ino
< EXT4_INODES_PER_GROUP(sb
))
908 goto repeat_in_this_group
;
910 if (++group
== ngroups
)
917 BUFFER_TRACE(inode_bitmap_bh
, "call ext4_handle_dirty_metadata");
918 err
= ext4_handle_dirty_metadata(handle
, NULL
, inode_bitmap_bh
);
920 ext4_std_error(sb
, err
);
924 BUFFER_TRACE(group_desc_bh
, "get_write_access");
925 err
= ext4_journal_get_write_access(handle
, group_desc_bh
);
927 ext4_std_error(sb
, err
);
931 /* We may have to initialize the block bitmap if it isn't already */
932 if (ext4_has_group_desc_csum(sb
) &&
933 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
934 struct buffer_head
*block_bitmap_bh
;
936 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
937 if (IS_ERR(block_bitmap_bh
)) {
938 err
= PTR_ERR(block_bitmap_bh
);
941 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
942 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
944 brelse(block_bitmap_bh
);
945 ext4_std_error(sb
, err
);
949 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
950 err
= ext4_handle_dirty_metadata(handle
, NULL
, block_bitmap_bh
);
952 /* recheck and clear flag under lock if we still need to */
953 ext4_lock_group(sb
, group
);
954 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
955 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
956 ext4_free_group_clusters_set(sb
, gdp
,
957 ext4_free_clusters_after_init(sb
, group
, gdp
));
958 ext4_block_bitmap_csum_set(sb
, group
, gdp
,
960 ext4_group_desc_csum_set(sb
, group
, gdp
);
962 ext4_unlock_group(sb
, group
);
963 brelse(block_bitmap_bh
);
966 ext4_std_error(sb
, err
);
971 /* Update the relevant bg descriptor fields */
972 if (ext4_has_group_desc_csum(sb
)) {
974 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
976 down_read(&grp
->alloc_sem
); /* protect vs itable lazyinit */
977 ext4_lock_group(sb
, group
); /* while we modify the bg desc */
978 free
= EXT4_INODES_PER_GROUP(sb
) -
979 ext4_itable_unused_count(sb
, gdp
);
980 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
981 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
985 * Check the relative inode number against the last used
986 * relative inode number in this group. if it is greater
987 * we need to update the bg_itable_unused count
990 ext4_itable_unused_set(sb
, gdp
,
991 (EXT4_INODES_PER_GROUP(sb
) - ino
));
992 up_read(&grp
->alloc_sem
);
994 ext4_lock_group(sb
, group
);
997 ext4_free_inodes_set(sb
, gdp
, ext4_free_inodes_count(sb
, gdp
) - 1);
999 ext4_used_dirs_set(sb
, gdp
, ext4_used_dirs_count(sb
, gdp
) + 1);
1000 if (sbi
->s_log_groups_per_flex
) {
1001 ext4_group_t f
= ext4_flex_group(sbi
, group
);
1003 atomic_inc(&sbi
->s_flex_groups
[f
].used_dirs
);
1006 if (ext4_has_group_desc_csum(sb
)) {
1007 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, inode_bitmap_bh
,
1008 EXT4_INODES_PER_GROUP(sb
) / 8);
1009 ext4_group_desc_csum_set(sb
, group
, gdp
);
1011 ext4_unlock_group(sb
, group
);
1013 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
1014 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
1016 ext4_std_error(sb
, err
);
1020 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
1022 percpu_counter_inc(&sbi
->s_dirs_counter
);
1024 if (sbi
->s_log_groups_per_flex
) {
1025 flex_group
= ext4_flex_group(sbi
, group
);
1026 atomic_dec(&sbi
->s_flex_groups
[flex_group
].free_inodes
);
1029 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
1030 /* This is the optimal IO size (for stat), not the fs block size */
1031 inode
->i_blocks
= 0;
1032 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
1033 ext4_current_time(inode
);
1035 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
1036 ei
->i_dir_start_lookup
= 0;
1039 /* Don't inherit extent flag from directory, amongst others. */
1041 ext4_mask_flags(mode
, EXT4_I(dir
)->i_flags
& EXT4_FL_INHERITED
);
1044 ei
->i_block_group
= group
;
1045 ei
->i_last_alloc_group
= ~0;
1047 ext4_set_inode_flags(inode
);
1048 if (IS_DIRSYNC(inode
))
1049 ext4_handle_sync(handle
);
1050 if (insert_inode_locked(inode
) < 0) {
1052 * Likely a bitmap corruption causing inode to be allocated
1056 ext4_error(sb
, "failed to insert inode %lu: doubly allocated?",
1060 spin_lock(&sbi
->s_next_gen_lock
);
1061 inode
->i_generation
= sbi
->s_next_generation
++;
1062 spin_unlock(&sbi
->s_next_gen_lock
);
1064 /* Precompute checksum seed for inode metadata */
1065 if (ext4_has_metadata_csum(sb
)) {
1067 __le32 inum
= cpu_to_le32(inode
->i_ino
);
1068 __le32 gen
= cpu_to_le32(inode
->i_generation
);
1069 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
1071 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
1075 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
1076 ext4_set_inode_state(inode
, EXT4_STATE_NEW
);
1078 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
1079 ei
->i_inline_off
= 0;
1080 if (ext4_has_feature_inline_data(sb
))
1081 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1083 err
= dquot_alloc_inode(inode
);
1087 err
= ext4_init_acl(handle
, inode
, dir
);
1089 goto fail_free_drop
;
1091 err
= ext4_init_security(handle
, inode
, dir
, qstr
);
1093 goto fail_free_drop
;
1095 if (ext4_has_feature_extents(sb
)) {
1096 /* set extent flag only for directory, file and normal symlink*/
1097 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
1098 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
1099 ext4_ext_tree_init(handle
, inode
);
1103 if (ext4_handle_valid(handle
)) {
1104 ei
->i_sync_tid
= handle
->h_transaction
->t_tid
;
1105 ei
->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1109 err
= ext4_inherit_context(dir
, inode
);
1111 goto fail_free_drop
;
1114 err
= ext4_mark_inode_dirty(handle
, inode
);
1116 ext4_std_error(sb
, err
);
1117 goto fail_free_drop
;
1120 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
1121 trace_ext4_allocate_inode(inode
, dir
, mode
);
1122 brelse(inode_bitmap_bh
);
1126 dquot_free_inode(inode
);
1129 unlock_new_inode(inode
);
1132 inode
->i_flags
|= S_NOQUOTA
;
1134 brelse(inode_bitmap_bh
);
1135 return ERR_PTR(err
);
1138 /* Verify that we are loading a valid orphan from disk */
1139 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
1141 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
1142 ext4_group_t block_group
;
1144 struct buffer_head
*bitmap_bh
;
1145 struct inode
*inode
= NULL
;
1148 /* Error cases - e2fsck has already cleaned up for us */
1149 if (ino
> max_ino
) {
1150 ext4_warning(sb
, "bad orphan ino %lu! e2fsck was run?", ino
);
1151 err
= -EFSCORRUPTED
;
1155 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
1156 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
1157 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
1158 if (IS_ERR(bitmap_bh
)) {
1159 err
= PTR_ERR(bitmap_bh
);
1160 ext4_warning(sb
, "inode bitmap error %ld for orphan %lu",
1165 /* Having the inode bit set should be a 100% indicator that this
1166 * is a valid orphan (no e2fsck run on fs). Orphans also include
1167 * inodes that were being truncated, so we can't check i_nlink==0.
1169 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
1172 inode
= ext4_iget(sb
, ino
);
1177 * If the orphans has i_nlinks > 0 then it should be able to be
1178 * truncated, otherwise it won't be removed from the orphan list
1179 * during processing and an infinite loop will result.
1181 if (inode
->i_nlink
&& !ext4_can_truncate(inode
))
1184 if (NEXT_ORPHAN(inode
) > max_ino
)
1190 err
= PTR_ERR(inode
);
1193 ext4_warning(sb
, "bad orphan inode %lu! e2fsck was run?", ino
);
1194 printk(KERN_WARNING
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
1195 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
1196 ext4_test_bit(bit
, bitmap_bh
->b_data
));
1197 printk(KERN_WARNING
"inode=%p\n", inode
);
1199 printk(KERN_WARNING
"is_bad_inode(inode)=%d\n",
1200 is_bad_inode(inode
));
1201 printk(KERN_WARNING
"NEXT_ORPHAN(inode)=%u\n",
1202 NEXT_ORPHAN(inode
));
1203 printk(KERN_WARNING
"max_ino=%lu\n", max_ino
);
1204 printk(KERN_WARNING
"i_nlink=%u\n", inode
->i_nlink
);
1205 /* Avoid freeing blocks if we got a bad deleted inode */
1206 if (inode
->i_nlink
== 0)
1207 inode
->i_blocks
= 0;
1212 return ERR_PTR(err
);
1215 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
1217 unsigned long desc_count
;
1218 struct ext4_group_desc
*gdp
;
1219 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1221 struct ext4_super_block
*es
;
1222 unsigned long bitmap_count
, x
;
1223 struct buffer_head
*bitmap_bh
= NULL
;
1225 es
= EXT4_SB(sb
)->s_es
;
1229 for (i
= 0; i
< ngroups
; i
++) {
1230 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1233 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1235 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1236 if (IS_ERR(bitmap_bh
)) {
1241 x
= ext4_count_free(bitmap_bh
->b_data
,
1242 EXT4_INODES_PER_GROUP(sb
) / 8);
1243 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1244 (unsigned long) i
, ext4_free_inodes_count(sb
, gdp
), x
);
1248 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1249 "stored = %u, computed = %lu, %lu\n",
1250 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1254 for (i
= 0; i
< ngroups
; i
++) {
1255 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1258 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1265 /* Called at mount-time, super-block is locked */
1266 unsigned long ext4_count_dirs(struct super_block
* sb
)
1268 unsigned long count
= 0;
1269 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1271 for (i
= 0; i
< ngroups
; i
++) {
1272 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1275 count
+= ext4_used_dirs_count(sb
, gdp
);
1281 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1282 * inode table. Must be called without any spinlock held. The only place
1283 * where it is called from on active part of filesystem is ext4lazyinit
1284 * thread, so we do not need any special locks, however we have to prevent
1285 * inode allocation from the current group, so we take alloc_sem lock, to
1286 * block ext4_new_inode() until we are finished.
1288 int ext4_init_inode_table(struct super_block
*sb
, ext4_group_t group
,
1291 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
1292 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1293 struct ext4_group_desc
*gdp
= NULL
;
1294 struct buffer_head
*group_desc_bh
;
1297 int num
, ret
= 0, used_blks
= 0;
1299 /* This should not happen, but just to be sure check this */
1300 if (sb
->s_flags
& MS_RDONLY
) {
1305 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
1310 * We do not need to lock this, because we are the only one
1311 * handling this flag.
1313 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_ZEROED
))
1316 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
1317 if (IS_ERR(handle
)) {
1318 ret
= PTR_ERR(handle
);
1322 down_write(&grp
->alloc_sem
);
1324 * If inode bitmap was already initialized there may be some
1325 * used inodes so we need to skip blocks with used inodes in
1328 if (!(gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)))
1329 used_blks
= DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb
) -
1330 ext4_itable_unused_count(sb
, gdp
)),
1331 sbi
->s_inodes_per_block
);
1333 if ((used_blks
< 0) || (used_blks
> sbi
->s_itb_per_group
)) {
1334 ext4_error(sb
, "Something is wrong with group %u: "
1335 "used itable blocks: %d; "
1336 "itable unused count: %u",
1338 ext4_itable_unused_count(sb
, gdp
));
1343 blk
= ext4_inode_table(sb
, gdp
) + used_blks
;
1344 num
= sbi
->s_itb_per_group
- used_blks
;
1346 BUFFER_TRACE(group_desc_bh
, "get_write_access");
1347 ret
= ext4_journal_get_write_access(handle
,
1353 * Skip zeroout if the inode table is full. But we set the ZEROED
1354 * flag anyway, because obviously, when it is full it does not need
1357 if (unlikely(num
== 0))
1360 ext4_debug("going to zero out inode table in group %d\n",
1362 ret
= sb_issue_zeroout(sb
, blk
, num
, GFP_NOFS
);
1366 blkdev_issue_flush(sb
->s_bdev
, GFP_NOFS
, NULL
);
1369 ext4_lock_group(sb
, group
);
1370 gdp
->bg_flags
|= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
1371 ext4_group_desc_csum_set(sb
, group
, gdp
);
1372 ext4_unlock_group(sb
, group
);
1374 BUFFER_TRACE(group_desc_bh
,
1375 "call ext4_handle_dirty_metadata");
1376 ret
= ext4_handle_dirty_metadata(handle
, NULL
,
1380 up_write(&grp
->alloc_sem
);
1381 ext4_journal_stop(handle
);