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 <linux/cred.h>
26 #include <asm/byteorder.h>
29 #include "ext4_jbd2.h"
33 #include <trace/events/ext4.h>
36 * ialloc.c contains the inodes allocation and deallocation routines
40 * The free inodes are managed by bitmaps. A file system contains several
41 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
42 * block for inodes, N blocks for the inode table and data blocks.
44 * The file system contains group descriptors which are located after the
45 * super block. Each descriptor contains the number of the bitmap block and
46 * the free blocks count in the block.
50 * To avoid calling the atomic setbit hundreds or thousands of times, we only
51 * need to use it within a single byte (to ensure we get endianness right).
52 * We can use memset for the rest of the bitmap as there are no other users.
54 void ext4_mark_bitmap_end(int start_bit
, int end_bit
, char *bitmap
)
58 if (start_bit
>= end_bit
)
61 ext4_debug("mark end bits +%d through +%d used\n", start_bit
, end_bit
);
62 for (i
= start_bit
; i
< ((start_bit
+ 7) & ~7UL); i
++)
63 ext4_set_bit(i
, bitmap
);
65 memset(bitmap
+ (i
>> 3), 0xff, (end_bit
- i
) >> 3);
68 /* Initializes an uninitialized inode bitmap */
69 static int ext4_init_inode_bitmap(struct super_block
*sb
,
70 struct buffer_head
*bh
,
71 ext4_group_t block_group
,
72 struct ext4_group_desc
*gdp
)
74 struct ext4_group_info
*grp
;
75 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
76 J_ASSERT_BH(bh
, buffer_locked(bh
));
78 /* If checksum is bad mark all blocks and inodes use to prevent
79 * allocation, essentially implementing a per-group read-only flag. */
80 if (!ext4_group_desc_csum_verify(sb
, block_group
, gdp
)) {
81 grp
= ext4_get_group_info(sb
, block_group
);
82 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp
))
83 percpu_counter_sub(&sbi
->s_freeclusters_counter
,
85 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
86 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
88 count
= ext4_free_inodes_count(sb
, gdp
);
89 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
92 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
96 memset(bh
->b_data
, 0, (EXT4_INODES_PER_GROUP(sb
) + 7) / 8);
97 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
99 ext4_inode_bitmap_csum_set(sb
, block_group
, gdp
, bh
,
100 EXT4_INODES_PER_GROUP(sb
) / 8);
101 ext4_group_desc_csum_set(sb
, block_group
, gdp
);
106 void ext4_end_bitmap_read(struct buffer_head
*bh
, int uptodate
)
109 set_buffer_uptodate(bh
);
110 set_bitmap_uptodate(bh
);
116 static int ext4_validate_inode_bitmap(struct super_block
*sb
,
117 struct ext4_group_desc
*desc
,
118 ext4_group_t block_group
,
119 struct buffer_head
*bh
)
122 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, block_group
);
123 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
125 if (buffer_verified(bh
))
127 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))
128 return -EFSCORRUPTED
;
130 ext4_lock_group(sb
, block_group
);
131 blk
= ext4_inode_bitmap(sb
, desc
);
132 if (!ext4_inode_bitmap_csum_verify(sb
, block_group
, desc
, bh
,
133 EXT4_INODES_PER_GROUP(sb
) / 8)) {
134 ext4_unlock_group(sb
, block_group
);
135 ext4_error(sb
, "Corrupt inode bitmap - block_group = %u, "
136 "inode_bitmap = %llu", block_group
, blk
);
137 grp
= ext4_get_group_info(sb
, block_group
);
138 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
140 count
= ext4_free_inodes_count(sb
, desc
);
141 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
144 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
147 set_buffer_verified(bh
);
148 ext4_unlock_group(sb
, block_group
);
153 * Read the inode allocation bitmap for a given block_group, reading
154 * into the specified slot in the superblock's bitmap cache.
156 * Return buffer_head of bitmap on success or NULL.
158 static struct buffer_head
*
159 ext4_read_inode_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
161 struct ext4_group_desc
*desc
;
162 struct buffer_head
*bh
= NULL
;
163 ext4_fsblk_t bitmap_blk
;
166 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
168 return ERR_PTR(-EFSCORRUPTED
);
170 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
171 bh
= sb_getblk(sb
, bitmap_blk
);
173 ext4_error(sb
, "Cannot read inode bitmap - "
174 "block_group = %u, inode_bitmap = %llu",
175 block_group
, bitmap_blk
);
176 return ERR_PTR(-EIO
);
178 if (bitmap_uptodate(bh
))
182 if (bitmap_uptodate(bh
)) {
187 ext4_lock_group(sb
, block_group
);
188 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
189 err
= ext4_init_inode_bitmap(sb
, bh
, block_group
, desc
);
190 set_bitmap_uptodate(bh
);
191 set_buffer_uptodate(bh
);
192 set_buffer_verified(bh
);
193 ext4_unlock_group(sb
, block_group
);
196 ext4_error(sb
, "Failed to init inode bitmap for group "
197 "%u: %d", block_group
, err
);
202 ext4_unlock_group(sb
, block_group
);
204 if (buffer_uptodate(bh
)) {
206 * if not uninit if bh is uptodate,
207 * bitmap is also uptodate
209 set_bitmap_uptodate(bh
);
214 * submit the buffer_head for reading
216 trace_ext4_load_inode_bitmap(sb
, block_group
);
217 bh
->b_end_io
= ext4_end_bitmap_read
;
219 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
221 if (!buffer_uptodate(bh
)) {
223 ext4_error(sb
, "Cannot read inode bitmap - "
224 "block_group = %u, inode_bitmap = %llu",
225 block_group
, bitmap_blk
);
226 return ERR_PTR(-EIO
);
230 err
= ext4_validate_inode_bitmap(sb
, desc
, block_group
, bh
);
240 * NOTE! When we get the inode, we're the only people
241 * that have access to it, and as such there are no
242 * race conditions we have to worry about. The inode
243 * is not on the hash-lists, and it cannot be reached
244 * through the filesystem because the directory entry
245 * has been deleted earlier.
247 * HOWEVER: we must make sure that we get no aliases,
248 * which means that we have to call "clear_inode()"
249 * _before_ we mark the inode not in use in the inode
250 * bitmaps. Otherwise a newly created file might use
251 * the same inode number (not actually the same pointer
252 * though), and then we'd have two inodes sharing the
253 * same inode number and space on the harddisk.
255 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
257 struct super_block
*sb
= inode
->i_sb
;
260 struct buffer_head
*bitmap_bh
= NULL
;
261 struct buffer_head
*bh2
;
262 ext4_group_t block_group
;
264 struct ext4_group_desc
*gdp
;
265 struct ext4_super_block
*es
;
266 struct ext4_sb_info
*sbi
;
267 int fatal
= 0, err
, count
, cleared
;
268 struct ext4_group_info
*grp
;
271 printk(KERN_ERR
"EXT4-fs: %s:%d: inode on "
272 "nonexistent device\n", __func__
, __LINE__
);
275 if (atomic_read(&inode
->i_count
) > 1) {
276 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: count=%d",
277 __func__
, __LINE__
, inode
->i_ino
,
278 atomic_read(&inode
->i_count
));
281 if (inode
->i_nlink
) {
282 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: nlink=%d\n",
283 __func__
, __LINE__
, inode
->i_ino
, inode
->i_nlink
);
289 ext4_debug("freeing inode %lu\n", ino
);
290 trace_ext4_free_inode(inode
);
293 * Note: we must free any quota before locking the superblock,
294 * as writing the quota to disk may need the lock as well.
296 dquot_initialize(inode
);
297 ext4_xattr_delete_inode(handle
, inode
);
298 dquot_free_inode(inode
);
301 is_directory
= S_ISDIR(inode
->i_mode
);
303 /* Do this BEFORE marking the inode not in use or returning an error */
304 ext4_clear_inode(inode
);
306 es
= EXT4_SB(sb
)->s_es
;
307 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
308 ext4_error(sb
, "reserved or nonexistent inode %lu", ino
);
311 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
312 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
313 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
314 /* Don't bother if the inode bitmap is corrupt. */
315 grp
= ext4_get_group_info(sb
, block_group
);
316 if (IS_ERR(bitmap_bh
)) {
317 fatal
= PTR_ERR(bitmap_bh
);
321 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))) {
322 fatal
= -EFSCORRUPTED
;
326 BUFFER_TRACE(bitmap_bh
, "get_write_access");
327 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
332 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
334 BUFFER_TRACE(bh2
, "get_write_access");
335 fatal
= ext4_journal_get_write_access(handle
, bh2
);
337 ext4_lock_group(sb
, block_group
);
338 cleared
= ext4_test_and_clear_bit(bit
, bitmap_bh
->b_data
);
339 if (fatal
|| !cleared
) {
340 ext4_unlock_group(sb
, block_group
);
344 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
345 ext4_free_inodes_set(sb
, gdp
, count
);
347 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
348 ext4_used_dirs_set(sb
, gdp
, count
);
349 percpu_counter_dec(&sbi
->s_dirs_counter
);
351 ext4_inode_bitmap_csum_set(sb
, block_group
, gdp
, bitmap_bh
,
352 EXT4_INODES_PER_GROUP(sb
) / 8);
353 ext4_group_desc_csum_set(sb
, block_group
, gdp
);
354 ext4_unlock_group(sb
, block_group
);
356 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
357 if (sbi
->s_log_groups_per_flex
) {
358 ext4_group_t f
= ext4_flex_group(sbi
, block_group
);
360 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
362 atomic_dec(&sbi
->s_flex_groups
[f
].used_dirs
);
364 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
365 fatal
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
368 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
369 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
373 ext4_error(sb
, "bit already cleared for inode %lu", ino
);
374 if (gdp
&& !EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
376 count
= ext4_free_inodes_count(sb
, gdp
);
377 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
380 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
385 ext4_std_error(sb
, fatal
);
395 * Helper function for Orlov's allocator; returns critical information
396 * for a particular block group or flex_bg. If flex_size is 1, then g
397 * is a block group number; otherwise it is flex_bg number.
399 static void get_orlov_stats(struct super_block
*sb
, ext4_group_t g
,
400 int flex_size
, struct orlov_stats
*stats
)
402 struct ext4_group_desc
*desc
;
403 struct flex_groups
*flex_group
= EXT4_SB(sb
)->s_flex_groups
;
406 stats
->free_inodes
= atomic_read(&flex_group
[g
].free_inodes
);
407 stats
->free_clusters
= atomic64_read(&flex_group
[g
].free_clusters
);
408 stats
->used_dirs
= atomic_read(&flex_group
[g
].used_dirs
);
412 desc
= ext4_get_group_desc(sb
, g
, NULL
);
414 stats
->free_inodes
= ext4_free_inodes_count(sb
, desc
);
415 stats
->free_clusters
= ext4_free_group_clusters(sb
, desc
);
416 stats
->used_dirs
= ext4_used_dirs_count(sb
, desc
);
418 stats
->free_inodes
= 0;
419 stats
->free_clusters
= 0;
420 stats
->used_dirs
= 0;
425 * Orlov's allocator for directories.
427 * We always try to spread first-level directories.
429 * If there are blockgroups with both free inodes and free blocks counts
430 * not worse than average we return one with smallest directory count.
431 * Otherwise we simply return a random group.
433 * For the rest rules look so:
435 * It's OK to put directory into a group unless
436 * it has too many directories already (max_dirs) or
437 * it has too few free inodes left (min_inodes) or
438 * it has too few free blocks left (min_blocks) or
439 * Parent's group is preferred, if it doesn't satisfy these
440 * conditions we search cyclically through the rest. If none
441 * of the groups look good we just look for a group with more
442 * free inodes than average (starting at parent's group).
445 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
446 ext4_group_t
*group
, umode_t mode
,
447 const struct qstr
*qstr
)
449 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
450 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
451 ext4_group_t real_ngroups
= ext4_get_groups_count(sb
);
452 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
453 unsigned int freei
, avefreei
, grp_free
;
454 ext4_fsblk_t freeb
, avefreec
;
456 int max_dirs
, min_inodes
;
457 ext4_grpblk_t min_clusters
;
458 ext4_group_t i
, grp
, g
, ngroups
;
459 struct ext4_group_desc
*desc
;
460 struct orlov_stats stats
;
461 int flex_size
= ext4_flex_bg_size(sbi
);
462 struct dx_hash_info hinfo
;
464 ngroups
= real_ngroups
;
466 ngroups
= (real_ngroups
+ flex_size
- 1) >>
467 sbi
->s_log_groups_per_flex
;
468 parent_group
>>= sbi
->s_log_groups_per_flex
;
471 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
472 avefreei
= freei
/ ngroups
;
473 freeb
= EXT4_C2B(sbi
,
474 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
476 do_div(avefreec
, ngroups
);
477 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
480 ((parent
== d_inode(sb
->s_root
)) ||
481 (ext4_test_inode_flag(parent
, EXT4_INODE_TOPDIR
)))) {
482 int best_ndir
= inodes_per_group
;
486 hinfo
.hash_version
= DX_HASH_HALF_MD4
;
487 hinfo
.seed
= sbi
->s_hash_seed
;
488 ext4fs_dirhash(qstr
->name
, qstr
->len
, &hinfo
);
492 parent_group
= (unsigned)grp
% ngroups
;
493 for (i
= 0; i
< ngroups
; i
++) {
494 g
= (parent_group
+ i
) % ngroups
;
495 get_orlov_stats(sb
, g
, flex_size
, &stats
);
496 if (!stats
.free_inodes
)
498 if (stats
.used_dirs
>= best_ndir
)
500 if (stats
.free_inodes
< avefreei
)
502 if (stats
.free_clusters
< avefreec
)
506 best_ndir
= stats
.used_dirs
;
511 if (flex_size
== 1) {
517 * We pack inodes at the beginning of the flexgroup's
518 * inode tables. Block allocation decisions will do
519 * something similar, although regular files will
520 * start at 2nd block group of the flexgroup. See
521 * ext4_ext_find_goal() and ext4_find_near().
524 for (i
= 0; i
< flex_size
; i
++) {
525 if (grp
+i
>= real_ngroups
)
527 desc
= ext4_get_group_desc(sb
, grp
+i
, NULL
);
528 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
536 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
537 min_inodes
= avefreei
- inodes_per_group
*flex_size
/ 4;
540 min_clusters
= avefreec
- EXT4_CLUSTERS_PER_GROUP(sb
)*flex_size
/ 4;
543 * Start looking in the flex group where we last allocated an
544 * inode for this parent directory
546 if (EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
547 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
549 parent_group
>>= sbi
->s_log_groups_per_flex
;
552 for (i
= 0; i
< ngroups
; i
++) {
553 grp
= (parent_group
+ i
) % ngroups
;
554 get_orlov_stats(sb
, grp
, flex_size
, &stats
);
555 if (stats
.used_dirs
>= max_dirs
)
557 if (stats
.free_inodes
< min_inodes
)
559 if (stats
.free_clusters
< min_clusters
)
565 ngroups
= real_ngroups
;
566 avefreei
= freei
/ ngroups
;
568 parent_group
= EXT4_I(parent
)->i_block_group
;
569 for (i
= 0; i
< ngroups
; i
++) {
570 grp
= (parent_group
+ i
) % ngroups
;
571 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
573 grp_free
= ext4_free_inodes_count(sb
, desc
);
574 if (grp_free
&& grp_free
>= avefreei
) {
583 * The free-inodes counter is approximate, and for really small
584 * filesystems the above test can fail to find any blockgroups
593 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
594 ext4_group_t
*group
, umode_t mode
)
596 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
597 ext4_group_t i
, last
, ngroups
= ext4_get_groups_count(sb
);
598 struct ext4_group_desc
*desc
;
599 int flex_size
= ext4_flex_bg_size(EXT4_SB(sb
));
602 * Try to place the inode is the same flex group as its
603 * parent. If we can't find space, use the Orlov algorithm to
604 * find another flex group, and store that information in the
605 * parent directory's inode information so that use that flex
606 * group for future allocations.
612 parent_group
&= ~(flex_size
-1);
613 last
= parent_group
+ flex_size
;
616 for (i
= parent_group
; i
< last
; i
++) {
617 desc
= ext4_get_group_desc(sb
, i
, NULL
);
618 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
623 if (!retry
&& EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
625 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
629 * If this didn't work, use the Orlov search algorithm
630 * to find a new flex group; we pass in the mode to
631 * avoid the topdir algorithms.
633 *group
= parent_group
+ flex_size
;
634 if (*group
> ngroups
)
636 return find_group_orlov(sb
, parent
, group
, mode
, NULL
);
640 * Try to place the inode in its parent directory
642 *group
= parent_group
;
643 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
644 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
645 ext4_free_group_clusters(sb
, desc
))
649 * We're going to place this inode in a different blockgroup from its
650 * parent. We want to cause files in a common directory to all land in
651 * the same blockgroup. But we want files which are in a different
652 * directory which shares a blockgroup with our parent to land in a
653 * different blockgroup.
655 * So add our directory's i_ino into the starting point for the hash.
657 *group
= (*group
+ parent
->i_ino
) % ngroups
;
660 * Use a quadratic hash to find a group with a free inode and some free
663 for (i
= 1; i
< ngroups
; i
<<= 1) {
665 if (*group
>= ngroups
)
667 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
668 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
669 ext4_free_group_clusters(sb
, desc
))
674 * That failed: try linear search for a free inode, even if that group
675 * has no free blocks.
677 *group
= parent_group
;
678 for (i
= 0; i
< ngroups
; i
++) {
679 if (++*group
>= ngroups
)
681 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
682 if (desc
&& ext4_free_inodes_count(sb
, desc
))
690 * In no journal mode, if an inode has recently been deleted, we want
691 * to avoid reusing it until we're reasonably sure the inode table
692 * block has been written back to disk. (Yes, these values are
693 * somewhat arbitrary...)
695 #define RECENTCY_MIN 5
696 #define RECENTCY_DIRTY 30
698 static int recently_deleted(struct super_block
*sb
, ext4_group_t group
, int ino
)
700 struct ext4_group_desc
*gdp
;
701 struct ext4_inode
*raw_inode
;
702 struct buffer_head
*bh
;
703 unsigned long dtime
, now
;
704 int inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
705 int offset
, ret
= 0, recentcy
= RECENTCY_MIN
;
707 gdp
= ext4_get_group_desc(sb
, group
, NULL
);
711 bh
= sb_getblk(sb
, ext4_inode_table(sb
, gdp
) +
712 (ino
/ inodes_per_block
));
713 if (unlikely(!bh
) || !buffer_uptodate(bh
))
715 * If the block is not in the buffer cache, then it
716 * must have been written out.
720 offset
= (ino
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
721 raw_inode
= (struct ext4_inode
*) (bh
->b_data
+ offset
);
722 dtime
= le32_to_cpu(raw_inode
->i_dtime
);
724 if (buffer_dirty(bh
))
725 recentcy
+= RECENTCY_DIRTY
;
727 if (dtime
&& (dtime
< now
) && (now
< dtime
+ recentcy
))
735 * There are two policies for allocating an inode. If the new inode is
736 * a directory, then a forward search is made for a block group with both
737 * free space and a low directory-to-inode ratio; if that fails, then of
738 * the groups with above-average free space, that group with the fewest
739 * directories already is chosen.
741 * For other inodes, search forward from the parent directory's block
742 * group to find a free inode.
744 struct inode
*__ext4_new_inode(handle_t
*handle
, struct inode
*dir
,
745 umode_t mode
, const struct qstr
*qstr
,
746 __u32 goal
, uid_t
*owner
, int handle_type
,
747 unsigned int line_no
, int nblocks
)
749 struct super_block
*sb
;
750 struct buffer_head
*inode_bitmap_bh
= NULL
;
751 struct buffer_head
*group_desc_bh
;
752 ext4_group_t ngroups
, group
= 0;
753 unsigned long ino
= 0;
755 struct ext4_group_desc
*gdp
= NULL
;
756 struct ext4_inode_info
*ei
;
757 struct ext4_sb_info
*sbi
;
761 ext4_group_t flex_group
;
762 struct ext4_group_info
*grp
;
765 /* Cannot create files in a deleted directory */
766 if (!dir
|| !dir
->i_nlink
)
767 return ERR_PTR(-EPERM
);
769 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
770 return ERR_PTR(-EIO
);
772 if ((ext4_encrypted_inode(dir
) ||
773 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir
->i_sb
))) &&
774 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
775 err
= fscrypt_get_encryption_info(dir
);
778 if (!fscrypt_has_encryption_key(dir
))
779 return ERR_PTR(-ENOKEY
);
781 nblocks
+= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
);
786 ngroups
= ext4_get_groups_count(sb
);
787 trace_ext4_request_inode(dir
, mode
);
788 inode
= new_inode(sb
);
790 return ERR_PTR(-ENOMEM
);
795 * Initialize owners and quota early so that we don't have to account
796 * for quota initialization worst case in standard inode creating
800 inode
->i_mode
= mode
;
801 i_uid_write(inode
, owner
[0]);
802 i_gid_write(inode
, owner
[1]);
803 } else if (test_opt(sb
, GRPID
)) {
804 inode
->i_mode
= mode
;
805 inode
->i_uid
= current_fsuid();
806 inode
->i_gid
= dir
->i_gid
;
808 inode_init_owner(inode
, dir
, mode
);
810 if (ext4_has_feature_project(sb
) &&
811 ext4_test_inode_flag(dir
, EXT4_INODE_PROJINHERIT
))
812 ei
->i_projid
= EXT4_I(dir
)->i_projid
;
814 ei
->i_projid
= make_kprojid(&init_user_ns
, EXT4_DEF_PROJID
);
816 err
= dquot_initialize(inode
);
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
);
831 ret2
= find_group_orlov(sb
, dir
, &group
, mode
, qstr
);
833 ret2
= find_group_other(sb
, dir
, &group
, mode
);
836 EXT4_I(dir
)->i_last_alloc_group
= group
;
842 * Normally we will only go through one pass of this loop,
843 * unless we get unlucky and it turns out the group we selected
844 * had its last inode grabbed by someone else.
846 for (i
= 0; i
< ngroups
; i
++, ino
= 0) {
849 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
854 * Check free inodes count before loading bitmap.
856 if (ext4_free_inodes_count(sb
, gdp
) == 0) {
857 if (++group
== ngroups
)
862 grp
= ext4_get_group_info(sb
, group
);
863 /* Skip groups with already-known suspicious inode tables */
864 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
865 if (++group
== ngroups
)
870 brelse(inode_bitmap_bh
);
871 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
872 /* Skip groups with suspicious inode tables */
873 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
) ||
874 IS_ERR(inode_bitmap_bh
)) {
875 inode_bitmap_bh
= NULL
;
876 if (++group
== ngroups
)
881 repeat_in_this_group
:
882 ino
= ext4_find_next_zero_bit((unsigned long *)
883 inode_bitmap_bh
->b_data
,
884 EXT4_INODES_PER_GROUP(sb
), ino
);
885 if (ino
>= EXT4_INODES_PER_GROUP(sb
))
887 if (group
== 0 && (ino
+1) < EXT4_FIRST_INO(sb
)) {
888 ext4_error(sb
, "reserved inode found cleared - "
889 "inode=%lu", ino
+ 1);
892 if ((EXT4_SB(sb
)->s_journal
== NULL
) &&
893 recently_deleted(sb
, group
, ino
)) {
898 BUG_ON(nblocks
<= 0);
899 handle
= __ext4_journal_start_sb(dir
->i_sb
, line_no
,
900 handle_type
, nblocks
,
902 if (IS_ERR(handle
)) {
903 err
= PTR_ERR(handle
);
904 ext4_std_error(sb
, err
);
908 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
909 err
= ext4_journal_get_write_access(handle
, inode_bitmap_bh
);
911 ext4_std_error(sb
, err
);
914 ext4_lock_group(sb
, group
);
915 ret2
= ext4_test_and_set_bit(ino
, inode_bitmap_bh
->b_data
);
916 ext4_unlock_group(sb
, group
);
917 ino
++; /* the inode bitmap is zero-based */
919 goto got
; /* we grabbed the inode! */
921 if (ino
< EXT4_INODES_PER_GROUP(sb
))
922 goto repeat_in_this_group
;
924 if (++group
== ngroups
)
931 BUFFER_TRACE(inode_bitmap_bh
, "call ext4_handle_dirty_metadata");
932 err
= ext4_handle_dirty_metadata(handle
, NULL
, inode_bitmap_bh
);
934 ext4_std_error(sb
, err
);
938 BUFFER_TRACE(group_desc_bh
, "get_write_access");
939 err
= ext4_journal_get_write_access(handle
, group_desc_bh
);
941 ext4_std_error(sb
, err
);
945 /* We may have to initialize the block bitmap if it isn't already */
946 if (ext4_has_group_desc_csum(sb
) &&
947 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
948 struct buffer_head
*block_bitmap_bh
;
950 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
951 if (IS_ERR(block_bitmap_bh
)) {
952 err
= PTR_ERR(block_bitmap_bh
);
955 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
956 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
958 brelse(block_bitmap_bh
);
959 ext4_std_error(sb
, err
);
963 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
964 err
= ext4_handle_dirty_metadata(handle
, NULL
, block_bitmap_bh
);
966 /* recheck and clear flag under lock if we still need to */
967 ext4_lock_group(sb
, group
);
968 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
969 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
970 ext4_free_group_clusters_set(sb
, gdp
,
971 ext4_free_clusters_after_init(sb
, group
, gdp
));
972 ext4_block_bitmap_csum_set(sb
, group
, gdp
,
974 ext4_group_desc_csum_set(sb
, group
, gdp
);
976 ext4_unlock_group(sb
, group
);
977 brelse(block_bitmap_bh
);
980 ext4_std_error(sb
, err
);
985 /* Update the relevant bg descriptor fields */
986 if (ext4_has_group_desc_csum(sb
)) {
988 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
990 down_read(&grp
->alloc_sem
); /* protect vs itable lazyinit */
991 ext4_lock_group(sb
, group
); /* while we modify the bg desc */
992 free
= EXT4_INODES_PER_GROUP(sb
) -
993 ext4_itable_unused_count(sb
, gdp
);
994 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
995 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
999 * Check the relative inode number against the last used
1000 * relative inode number in this group. if it is greater
1001 * we need to update the bg_itable_unused count
1004 ext4_itable_unused_set(sb
, gdp
,
1005 (EXT4_INODES_PER_GROUP(sb
) - ino
));
1006 up_read(&grp
->alloc_sem
);
1008 ext4_lock_group(sb
, group
);
1011 ext4_free_inodes_set(sb
, gdp
, ext4_free_inodes_count(sb
, gdp
) - 1);
1012 if (S_ISDIR(mode
)) {
1013 ext4_used_dirs_set(sb
, gdp
, ext4_used_dirs_count(sb
, gdp
) + 1);
1014 if (sbi
->s_log_groups_per_flex
) {
1015 ext4_group_t f
= ext4_flex_group(sbi
, group
);
1017 atomic_inc(&sbi
->s_flex_groups
[f
].used_dirs
);
1020 if (ext4_has_group_desc_csum(sb
)) {
1021 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, inode_bitmap_bh
,
1022 EXT4_INODES_PER_GROUP(sb
) / 8);
1023 ext4_group_desc_csum_set(sb
, group
, gdp
);
1025 ext4_unlock_group(sb
, group
);
1027 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
1028 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
1030 ext4_std_error(sb
, err
);
1034 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
1036 percpu_counter_inc(&sbi
->s_dirs_counter
);
1038 if (sbi
->s_log_groups_per_flex
) {
1039 flex_group
= ext4_flex_group(sbi
, group
);
1040 atomic_dec(&sbi
->s_flex_groups
[flex_group
].free_inodes
);
1043 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
1044 /* This is the optimal IO size (for stat), not the fs block size */
1045 inode
->i_blocks
= 0;
1046 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
1047 current_time(inode
);
1049 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
1050 ei
->i_dir_start_lookup
= 0;
1053 /* Don't inherit extent flag from directory, amongst others. */
1055 ext4_mask_flags(mode
, EXT4_I(dir
)->i_flags
& EXT4_FL_INHERITED
);
1058 ei
->i_block_group
= group
;
1059 ei
->i_last_alloc_group
= ~0;
1061 ext4_set_inode_flags(inode
);
1062 if (IS_DIRSYNC(inode
))
1063 ext4_handle_sync(handle
);
1064 if (insert_inode_locked(inode
) < 0) {
1066 * Likely a bitmap corruption causing inode to be allocated
1070 ext4_error(sb
, "failed to insert inode %lu: doubly allocated?",
1074 spin_lock(&sbi
->s_next_gen_lock
);
1075 inode
->i_generation
= sbi
->s_next_generation
++;
1076 spin_unlock(&sbi
->s_next_gen_lock
);
1078 /* Precompute checksum seed for inode metadata */
1079 if (ext4_has_metadata_csum(sb
)) {
1081 __le32 inum
= cpu_to_le32(inode
->i_ino
);
1082 __le32 gen
= cpu_to_le32(inode
->i_generation
);
1083 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
1085 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
1089 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
1090 ext4_set_inode_state(inode
, EXT4_STATE_NEW
);
1092 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
1093 ei
->i_inline_off
= 0;
1094 if (ext4_has_feature_inline_data(sb
))
1095 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1097 err
= dquot_alloc_inode(inode
);
1102 * Since the encryption xattr will always be unique, create it first so
1103 * that it's less likely to end up in an external xattr block and
1104 * prevent its deduplication.
1107 err
= fscrypt_inherit_context(dir
, inode
, handle
, true);
1109 goto fail_free_drop
;
1112 err
= ext4_init_acl(handle
, inode
, dir
);
1114 goto fail_free_drop
;
1116 err
= ext4_init_security(handle
, inode
, dir
, qstr
);
1118 goto fail_free_drop
;
1120 if (ext4_has_feature_extents(sb
)) {
1121 /* set extent flag only for directory, file and normal symlink*/
1122 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
1123 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
1124 ext4_ext_tree_init(handle
, inode
);
1128 if (ext4_handle_valid(handle
)) {
1129 ei
->i_sync_tid
= handle
->h_transaction
->t_tid
;
1130 ei
->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1133 err
= ext4_mark_inode_dirty(handle
, inode
);
1135 ext4_std_error(sb
, err
);
1136 goto fail_free_drop
;
1139 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
1140 trace_ext4_allocate_inode(inode
, dir
, mode
);
1141 brelse(inode_bitmap_bh
);
1145 dquot_free_inode(inode
);
1148 unlock_new_inode(inode
);
1151 inode
->i_flags
|= S_NOQUOTA
;
1153 brelse(inode_bitmap_bh
);
1154 return ERR_PTR(err
);
1157 /* Verify that we are loading a valid orphan from disk */
1158 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
1160 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
1161 ext4_group_t block_group
;
1163 struct buffer_head
*bitmap_bh
= NULL
;
1164 struct inode
*inode
= NULL
;
1165 int err
= -EFSCORRUPTED
;
1167 if (ino
< EXT4_FIRST_INO(sb
) || ino
> max_ino
)
1170 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
1171 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
1172 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
1173 if (IS_ERR(bitmap_bh
)) {
1174 ext4_error(sb
, "inode bitmap error %ld for orphan %lu",
1175 ino
, PTR_ERR(bitmap_bh
));
1176 return (struct inode
*) bitmap_bh
;
1179 /* Having the inode bit set should be a 100% indicator that this
1180 * is a valid orphan (no e2fsck run on fs). Orphans also include
1181 * inodes that were being truncated, so we can't check i_nlink==0.
1183 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
1186 inode
= ext4_iget(sb
, ino
);
1187 if (IS_ERR(inode
)) {
1188 err
= PTR_ERR(inode
);
1189 ext4_error(sb
, "couldn't read orphan inode %lu (err %d)",
1195 * If the orphans has i_nlinks > 0 then it should be able to
1196 * be truncated, otherwise it won't be removed from the orphan
1197 * list during processing and an infinite loop will result.
1198 * Similarly, it must not be a bad inode.
1200 if ((inode
->i_nlink
&& !ext4_can_truncate(inode
)) ||
1201 is_bad_inode(inode
))
1204 if (NEXT_ORPHAN(inode
) > max_ino
)
1210 ext4_error(sb
, "bad orphan inode %lu", ino
);
1212 printk(KERN_ERR
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
1213 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
1214 ext4_test_bit(bit
, bitmap_bh
->b_data
));
1216 printk(KERN_ERR
"is_bad_inode(inode)=%d\n",
1217 is_bad_inode(inode
));
1218 printk(KERN_ERR
"NEXT_ORPHAN(inode)=%u\n",
1219 NEXT_ORPHAN(inode
));
1220 printk(KERN_ERR
"max_ino=%lu\n", max_ino
);
1221 printk(KERN_ERR
"i_nlink=%u\n", inode
->i_nlink
);
1222 /* Avoid freeing blocks if we got a bad deleted inode */
1223 if (inode
->i_nlink
== 0)
1224 inode
->i_blocks
= 0;
1228 return ERR_PTR(err
);
1231 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
1233 unsigned long desc_count
;
1234 struct ext4_group_desc
*gdp
;
1235 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1237 struct ext4_super_block
*es
;
1238 unsigned long bitmap_count
, x
;
1239 struct buffer_head
*bitmap_bh
= NULL
;
1241 es
= EXT4_SB(sb
)->s_es
;
1245 for (i
= 0; i
< ngroups
; i
++) {
1246 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1249 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1251 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1252 if (IS_ERR(bitmap_bh
)) {
1257 x
= ext4_count_free(bitmap_bh
->b_data
,
1258 EXT4_INODES_PER_GROUP(sb
) / 8);
1259 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1260 (unsigned long) i
, ext4_free_inodes_count(sb
, gdp
), x
);
1264 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1265 "stored = %u, computed = %lu, %lu\n",
1266 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1270 for (i
= 0; i
< ngroups
; i
++) {
1271 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1274 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1281 /* Called at mount-time, super-block is locked */
1282 unsigned long ext4_count_dirs(struct super_block
* sb
)
1284 unsigned long count
= 0;
1285 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1287 for (i
= 0; i
< ngroups
; i
++) {
1288 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1291 count
+= ext4_used_dirs_count(sb
, gdp
);
1297 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1298 * inode table. Must be called without any spinlock held. The only place
1299 * where it is called from on active part of filesystem is ext4lazyinit
1300 * thread, so we do not need any special locks, however we have to prevent
1301 * inode allocation from the current group, so we take alloc_sem lock, to
1302 * block ext4_new_inode() until we are finished.
1304 int ext4_init_inode_table(struct super_block
*sb
, ext4_group_t group
,
1307 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
1308 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1309 struct ext4_group_desc
*gdp
= NULL
;
1310 struct buffer_head
*group_desc_bh
;
1313 int num
, ret
= 0, used_blks
= 0;
1315 /* This should not happen, but just to be sure check this */
1316 if (sb
->s_flags
& MS_RDONLY
) {
1321 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
1326 * We do not need to lock this, because we are the only one
1327 * handling this flag.
1329 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_ZEROED
))
1332 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
1333 if (IS_ERR(handle
)) {
1334 ret
= PTR_ERR(handle
);
1338 down_write(&grp
->alloc_sem
);
1340 * If inode bitmap was already initialized there may be some
1341 * used inodes so we need to skip blocks with used inodes in
1344 if (!(gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)))
1345 used_blks
= DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb
) -
1346 ext4_itable_unused_count(sb
, gdp
)),
1347 sbi
->s_inodes_per_block
);
1349 if ((used_blks
< 0) || (used_blks
> sbi
->s_itb_per_group
)) {
1350 ext4_error(sb
, "Something is wrong with group %u: "
1351 "used itable blocks: %d; "
1352 "itable unused count: %u",
1354 ext4_itable_unused_count(sb
, gdp
));
1359 blk
= ext4_inode_table(sb
, gdp
) + used_blks
;
1360 num
= sbi
->s_itb_per_group
- used_blks
;
1362 BUFFER_TRACE(group_desc_bh
, "get_write_access");
1363 ret
= ext4_journal_get_write_access(handle
,
1369 * Skip zeroout if the inode table is full. But we set the ZEROED
1370 * flag anyway, because obviously, when it is full it does not need
1373 if (unlikely(num
== 0))
1376 ext4_debug("going to zero out inode table in group %d\n",
1378 ret
= sb_issue_zeroout(sb
, blk
, num
, GFP_NOFS
);
1382 blkdev_issue_flush(sb
->s_bdev
, GFP_NOFS
, NULL
);
1385 ext4_lock_group(sb
, group
);
1386 gdp
->bg_flags
|= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
1387 ext4_group_desc_csum_set(sb
, group
, gdp
);
1388 ext4_unlock_group(sb
, group
);
1390 BUFFER_TRACE(group_desc_bh
,
1391 "call ext4_handle_dirty_metadata");
1392 ret
= ext4_handle_dirty_metadata(handle
, NULL
,
1396 up_write(&grp
->alloc_sem
);
1397 ext4_journal_stop(handle
);