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 ext4_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 static unsigned ext4_init_inode_bitmap(struct super_block
*sb
,
69 struct buffer_head
*bh
,
70 ext4_group_t block_group
,
71 struct ext4_group_desc
*gdp
)
73 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
75 J_ASSERT_BH(bh
, buffer_locked(bh
));
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
79 if (!ext4_group_desc_csum_verify(sbi
, block_group
, gdp
)) {
80 ext4_error(sb
, "Checksum bad for group %u", block_group
);
81 ext4_free_group_clusters_set(sb
, gdp
, 0);
82 ext4_free_inodes_set(sb
, gdp
, 0);
83 ext4_itable_unused_set(sb
, gdp
, 0);
84 memset(bh
->b_data
, 0xff, sb
->s_blocksize
);
88 memset(bh
->b_data
, 0, (EXT4_INODES_PER_GROUP(sb
) + 7) / 8);
89 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
92 return EXT4_INODES_PER_GROUP(sb
);
95 void ext4_end_bitmap_read(struct buffer_head
*bh
, int uptodate
)
98 set_buffer_uptodate(bh
);
99 set_bitmap_uptodate(bh
);
106 * Read the inode allocation bitmap for a given block_group, reading
107 * into the specified slot in the superblock's bitmap cache.
109 * Return buffer_head of bitmap on success or NULL.
111 static struct buffer_head
*
112 ext4_read_inode_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
114 struct ext4_group_desc
*desc
;
115 struct buffer_head
*bh
= NULL
;
116 ext4_fsblk_t bitmap_blk
;
118 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
122 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
123 bh
= sb_getblk(sb
, bitmap_blk
);
125 ext4_error(sb
, "Cannot read inode bitmap - "
126 "block_group = %u, inode_bitmap = %llu",
127 block_group
, bitmap_blk
);
130 if (bitmap_uptodate(bh
))
134 if (bitmap_uptodate(bh
)) {
139 ext4_lock_group(sb
, block_group
);
140 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
141 ext4_init_inode_bitmap(sb
, bh
, block_group
, desc
);
142 set_bitmap_uptodate(bh
);
143 set_buffer_uptodate(bh
);
144 ext4_unlock_group(sb
, block_group
);
148 ext4_unlock_group(sb
, block_group
);
150 if (buffer_uptodate(bh
)) {
152 * if not uninit if bh is uptodate,
153 * bitmap is also uptodate
155 set_bitmap_uptodate(bh
);
160 * submit the buffer_head for reading
162 trace_ext4_load_inode_bitmap(sb
, block_group
);
163 bh
->b_end_io
= ext4_end_bitmap_read
;
167 if (!buffer_uptodate(bh
)) {
169 ext4_error(sb
, "Cannot read inode bitmap - "
170 "block_group = %u, inode_bitmap = %llu",
171 block_group
, bitmap_blk
);
178 * NOTE! When we get the inode, we're the only people
179 * that have access to it, and as such there are no
180 * race conditions we have to worry about. The inode
181 * is not on the hash-lists, and it cannot be reached
182 * through the filesystem because the directory entry
183 * has been deleted earlier.
185 * HOWEVER: we must make sure that we get no aliases,
186 * which means that we have to call "clear_inode()"
187 * _before_ we mark the inode not in use in the inode
188 * bitmaps. Otherwise a newly created file might use
189 * the same inode number (not actually the same pointer
190 * though), and then we'd have two inodes sharing the
191 * same inode number and space on the harddisk.
193 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
195 struct super_block
*sb
= inode
->i_sb
;
198 struct buffer_head
*bitmap_bh
= NULL
;
199 struct buffer_head
*bh2
;
200 ext4_group_t block_group
;
202 struct ext4_group_desc
*gdp
;
203 struct ext4_super_block
*es
;
204 struct ext4_sb_info
*sbi
;
205 int fatal
= 0, err
, count
, cleared
;
208 printk(KERN_ERR
"EXT4-fs: %s:%d: inode on "
209 "nonexistent device\n", __func__
, __LINE__
);
212 if (atomic_read(&inode
->i_count
) > 1) {
213 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: count=%d",
214 __func__
, __LINE__
, inode
->i_ino
,
215 atomic_read(&inode
->i_count
));
218 if (inode
->i_nlink
) {
219 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: nlink=%d\n",
220 __func__
, __LINE__
, inode
->i_ino
, inode
->i_nlink
);
226 ext4_debug("freeing inode %lu\n", ino
);
227 trace_ext4_free_inode(inode
);
230 * Note: we must free any quota before locking the superblock,
231 * as writing the quota to disk may need the lock as well.
233 dquot_initialize(inode
);
234 ext4_xattr_delete_inode(handle
, inode
);
235 dquot_free_inode(inode
);
238 is_directory
= S_ISDIR(inode
->i_mode
);
240 /* Do this BEFORE marking the inode not in use or returning an error */
241 ext4_clear_inode(inode
);
243 es
= EXT4_SB(sb
)->s_es
;
244 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
245 ext4_error(sb
, "reserved or nonexistent inode %lu", ino
);
248 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
249 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
250 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
254 BUFFER_TRACE(bitmap_bh
, "get_write_access");
255 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
260 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
262 BUFFER_TRACE(bh2
, "get_write_access");
263 fatal
= ext4_journal_get_write_access(handle
, bh2
);
265 ext4_lock_group(sb
, block_group
);
266 cleared
= ext4_test_and_clear_bit(bit
, bitmap_bh
->b_data
);
267 if (fatal
|| !cleared
) {
268 ext4_unlock_group(sb
, block_group
);
272 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
273 ext4_free_inodes_set(sb
, gdp
, count
);
275 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
276 ext4_used_dirs_set(sb
, gdp
, count
);
277 percpu_counter_dec(&sbi
->s_dirs_counter
);
279 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, block_group
, gdp
);
280 ext4_unlock_group(sb
, block_group
);
282 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
283 if (sbi
->s_log_groups_per_flex
) {
284 ext4_group_t f
= ext4_flex_group(sbi
, block_group
);
286 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
288 atomic_dec(&sbi
->s_flex_groups
[f
].used_dirs
);
290 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
291 fatal
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
294 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
295 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
298 ext4_mark_super_dirty(sb
);
300 ext4_error(sb
, "bit already cleared for inode %lu", ino
);
304 ext4_std_error(sb
, fatal
);
314 * Helper function for Orlov's allocator; returns critical information
315 * for a particular block group or flex_bg. If flex_size is 1, then g
316 * is a block group number; otherwise it is flex_bg number.
318 static void get_orlov_stats(struct super_block
*sb
, ext4_group_t g
,
319 int flex_size
, struct orlov_stats
*stats
)
321 struct ext4_group_desc
*desc
;
322 struct flex_groups
*flex_group
= EXT4_SB(sb
)->s_flex_groups
;
325 stats
->free_inodes
= atomic_read(&flex_group
[g
].free_inodes
);
326 stats
->free_clusters
= atomic_read(&flex_group
[g
].free_clusters
);
327 stats
->used_dirs
= atomic_read(&flex_group
[g
].used_dirs
);
331 desc
= ext4_get_group_desc(sb
, g
, NULL
);
333 stats
->free_inodes
= ext4_free_inodes_count(sb
, desc
);
334 stats
->free_clusters
= ext4_free_group_clusters(sb
, desc
);
335 stats
->used_dirs
= ext4_used_dirs_count(sb
, desc
);
337 stats
->free_inodes
= 0;
338 stats
->free_clusters
= 0;
339 stats
->used_dirs
= 0;
344 * Orlov's allocator for directories.
346 * We always try to spread first-level directories.
348 * If there are blockgroups with both free inodes and free blocks counts
349 * not worse than average we return one with smallest directory count.
350 * Otherwise we simply return a random group.
352 * For the rest rules look so:
354 * It's OK to put directory into a group unless
355 * it has too many directories already (max_dirs) or
356 * it has too few free inodes left (min_inodes) or
357 * it has too few free blocks left (min_blocks) or
358 * Parent's group is preferred, if it doesn't satisfy these
359 * conditions we search cyclically through the rest. If none
360 * of the groups look good we just look for a group with more
361 * free inodes than average (starting at parent's group).
364 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
365 ext4_group_t
*group
, umode_t mode
,
366 const struct qstr
*qstr
)
368 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
369 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
370 ext4_group_t real_ngroups
= ext4_get_groups_count(sb
);
371 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
372 unsigned int freei
, avefreei
, grp_free
;
373 ext4_fsblk_t freeb
, avefreec
;
375 int max_dirs
, min_inodes
;
376 ext4_grpblk_t min_clusters
;
377 ext4_group_t i
, grp
, g
, ngroups
;
378 struct ext4_group_desc
*desc
;
379 struct orlov_stats stats
;
380 int flex_size
= ext4_flex_bg_size(sbi
);
381 struct dx_hash_info hinfo
;
383 ngroups
= real_ngroups
;
385 ngroups
= (real_ngroups
+ flex_size
- 1) >>
386 sbi
->s_log_groups_per_flex
;
387 parent_group
>>= sbi
->s_log_groups_per_flex
;
390 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
391 avefreei
= freei
/ ngroups
;
392 freeb
= EXT4_C2B(sbi
,
393 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
395 do_div(avefreec
, ngroups
);
396 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
399 ((parent
== sb
->s_root
->d_inode
) ||
400 (ext4_test_inode_flag(parent
, EXT4_INODE_TOPDIR
)))) {
401 int best_ndir
= inodes_per_group
;
405 hinfo
.hash_version
= DX_HASH_HALF_MD4
;
406 hinfo
.seed
= sbi
->s_hash_seed
;
407 ext4fs_dirhash(qstr
->name
, qstr
->len
, &hinfo
);
410 get_random_bytes(&grp
, sizeof(grp
));
411 parent_group
= (unsigned)grp
% ngroups
;
412 for (i
= 0; i
< ngroups
; i
++) {
413 g
= (parent_group
+ i
) % ngroups
;
414 get_orlov_stats(sb
, g
, flex_size
, &stats
);
415 if (!stats
.free_inodes
)
417 if (stats
.used_dirs
>= best_ndir
)
419 if (stats
.free_inodes
< avefreei
)
421 if (stats
.free_clusters
< avefreec
)
425 best_ndir
= stats
.used_dirs
;
430 if (flex_size
== 1) {
436 * We pack inodes at the beginning of the flexgroup's
437 * inode tables. Block allocation decisions will do
438 * something similar, although regular files will
439 * start at 2nd block group of the flexgroup. See
440 * ext4_ext_find_goal() and ext4_find_near().
443 for (i
= 0; i
< flex_size
; i
++) {
444 if (grp
+i
>= real_ngroups
)
446 desc
= ext4_get_group_desc(sb
, grp
+i
, NULL
);
447 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
455 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
456 min_inodes
= avefreei
- inodes_per_group
*flex_size
/ 4;
459 min_clusters
= avefreec
- EXT4_CLUSTERS_PER_GROUP(sb
)*flex_size
/ 4;
462 * Start looking in the flex group where we last allocated an
463 * inode for this parent directory
465 if (EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
466 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
468 parent_group
>>= sbi
->s_log_groups_per_flex
;
471 for (i
= 0; i
< ngroups
; i
++) {
472 grp
= (parent_group
+ i
) % ngroups
;
473 get_orlov_stats(sb
, grp
, flex_size
, &stats
);
474 if (stats
.used_dirs
>= max_dirs
)
476 if (stats
.free_inodes
< min_inodes
)
478 if (stats
.free_clusters
< min_clusters
)
484 ngroups
= real_ngroups
;
485 avefreei
= freei
/ ngroups
;
487 parent_group
= EXT4_I(parent
)->i_block_group
;
488 for (i
= 0; i
< ngroups
; i
++) {
489 grp
= (parent_group
+ i
) % ngroups
;
490 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
491 grp_free
= ext4_free_inodes_count(sb
, desc
);
492 if (desc
&& grp_free
&& grp_free
>= avefreei
) {
500 * The free-inodes counter is approximate, and for really small
501 * filesystems the above test can fail to find any blockgroups
510 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
511 ext4_group_t
*group
, umode_t mode
)
513 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
514 ext4_group_t i
, last
, ngroups
= ext4_get_groups_count(sb
);
515 struct ext4_group_desc
*desc
;
516 int flex_size
= ext4_flex_bg_size(EXT4_SB(sb
));
519 * Try to place the inode is the same flex group as its
520 * parent. If we can't find space, use the Orlov algorithm to
521 * find another flex group, and store that information in the
522 * parent directory's inode information so that use that flex
523 * group for future allocations.
529 parent_group
&= ~(flex_size
-1);
530 last
= parent_group
+ flex_size
;
533 for (i
= parent_group
; i
< last
; i
++) {
534 desc
= ext4_get_group_desc(sb
, i
, NULL
);
535 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
540 if (!retry
&& EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
542 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
546 * If this didn't work, use the Orlov search algorithm
547 * to find a new flex group; we pass in the mode to
548 * avoid the topdir algorithms.
550 *group
= parent_group
+ flex_size
;
551 if (*group
> ngroups
)
553 return find_group_orlov(sb
, parent
, group
, mode
, NULL
);
557 * Try to place the inode in its parent directory
559 *group
= parent_group
;
560 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
561 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
562 ext4_free_group_clusters(sb
, desc
))
566 * We're going to place this inode in a different blockgroup from its
567 * parent. We want to cause files in a common directory to all land in
568 * the same blockgroup. But we want files which are in a different
569 * directory which shares a blockgroup with our parent to land in a
570 * different blockgroup.
572 * So add our directory's i_ino into the starting point for the hash.
574 *group
= (*group
+ parent
->i_ino
) % ngroups
;
577 * Use a quadratic hash to find a group with a free inode and some free
580 for (i
= 1; i
< ngroups
; i
<<= 1) {
582 if (*group
>= ngroups
)
584 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
585 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
586 ext4_free_group_clusters(sb
, desc
))
591 * That failed: try linear search for a free inode, even if that group
592 * has no free blocks.
594 *group
= parent_group
;
595 for (i
= 0; i
< ngroups
; i
++) {
596 if (++*group
>= ngroups
)
598 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
599 if (desc
&& ext4_free_inodes_count(sb
, desc
))
607 * There are two policies for allocating an inode. If the new inode is
608 * a directory, then a forward search is made for a block group with both
609 * free space and a low directory-to-inode ratio; if that fails, then of
610 * the groups with above-average free space, that group with the fewest
611 * directories already is chosen.
613 * For other inodes, search forward from the parent directory's block
614 * group to find a free inode.
616 struct inode
*ext4_new_inode(handle_t
*handle
, struct inode
*dir
, umode_t mode
,
617 const struct qstr
*qstr
, __u32 goal
, uid_t
*owner
)
619 struct super_block
*sb
;
620 struct buffer_head
*inode_bitmap_bh
= NULL
;
621 struct buffer_head
*group_desc_bh
;
622 ext4_group_t ngroups
, group
= 0;
623 unsigned long ino
= 0;
625 struct ext4_group_desc
*gdp
= NULL
;
626 struct ext4_inode_info
*ei
;
627 struct ext4_sb_info
*sbi
;
631 ext4_group_t flex_group
;
633 /* Cannot create files in a deleted directory */
634 if (!dir
|| !dir
->i_nlink
)
635 return ERR_PTR(-EPERM
);
638 ngroups
= ext4_get_groups_count(sb
);
639 trace_ext4_request_inode(dir
, mode
);
640 inode
= new_inode(sb
);
642 return ERR_PTR(-ENOMEM
);
647 goal
= sbi
->s_inode_goal
;
649 if (goal
&& goal
<= le32_to_cpu(sbi
->s_es
->s_inodes_count
)) {
650 group
= (goal
- 1) / EXT4_INODES_PER_GROUP(sb
);
651 ino
= (goal
- 1) % EXT4_INODES_PER_GROUP(sb
);
657 ret2
= find_group_orlov(sb
, dir
, &group
, mode
, qstr
);
659 ret2
= find_group_other(sb
, dir
, &group
, mode
);
662 EXT4_I(dir
)->i_last_alloc_group
= group
;
668 * Normally we will only go through one pass of this loop,
669 * unless we get unlucky and it turns out the group we selected
670 * had its last inode grabbed by someone else.
672 for (i
= 0; i
< ngroups
; i
++, ino
= 0) {
675 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
679 brelse(inode_bitmap_bh
);
680 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
681 if (!inode_bitmap_bh
)
684 repeat_in_this_group
:
685 ino
= ext4_find_next_zero_bit((unsigned long *)
686 inode_bitmap_bh
->b_data
,
687 EXT4_INODES_PER_GROUP(sb
), ino
);
688 if (ino
>= EXT4_INODES_PER_GROUP(sb
)) {
689 if (++group
== ngroups
)
693 if (group
== 0 && (ino
+1) < EXT4_FIRST_INO(sb
)) {
694 ext4_error(sb
, "reserved inode found cleared - "
695 "inode=%lu", ino
+ 1);
698 ext4_lock_group(sb
, group
);
699 ret2
= ext4_test_and_set_bit(ino
, inode_bitmap_bh
->b_data
);
700 ext4_unlock_group(sb
, group
);
701 ino
++; /* the inode bitmap is zero-based */
703 goto got
; /* we grabbed the inode! */
704 if (ino
< EXT4_INODES_PER_GROUP(sb
))
705 goto repeat_in_this_group
;
711 /* We may have to initialize the block bitmap if it isn't already */
712 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
) &&
713 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
714 struct buffer_head
*block_bitmap_bh
;
716 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
717 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
718 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
720 brelse(block_bitmap_bh
);
724 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
725 err
= ext4_handle_dirty_metadata(handle
, NULL
, block_bitmap_bh
);
726 brelse(block_bitmap_bh
);
728 /* recheck and clear flag under lock if we still need to */
729 ext4_lock_group(sb
, group
);
730 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
731 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
732 ext4_free_group_clusters_set(sb
, gdp
,
733 ext4_free_clusters_after_init(sb
, group
, gdp
));
734 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
,
737 ext4_unlock_group(sb
, group
);
743 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
744 err
= ext4_journal_get_write_access(handle
, inode_bitmap_bh
);
748 BUFFER_TRACE(group_desc_bh
, "get_write_access");
749 err
= ext4_journal_get_write_access(handle
, group_desc_bh
);
753 /* Update the relevant bg descriptor fields */
754 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
)) {
756 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
758 down_read(&grp
->alloc_sem
); /* protect vs itable lazyinit */
759 ext4_lock_group(sb
, group
); /* while we modify the bg desc */
760 free
= EXT4_INODES_PER_GROUP(sb
) -
761 ext4_itable_unused_count(sb
, gdp
);
762 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
763 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
767 * Check the relative inode number against the last used
768 * relative inode number in this group. if it is greater
769 * we need to update the bg_itable_unused count
772 ext4_itable_unused_set(sb
, gdp
,
773 (EXT4_INODES_PER_GROUP(sb
) - ino
));
774 up_read(&grp
->alloc_sem
);
776 ext4_free_inodes_set(sb
, gdp
, ext4_free_inodes_count(sb
, gdp
) - 1);
778 ext4_used_dirs_set(sb
, gdp
, ext4_used_dirs_count(sb
, gdp
) + 1);
779 if (sbi
->s_log_groups_per_flex
) {
780 ext4_group_t f
= ext4_flex_group(sbi
, group
);
782 atomic_inc(&sbi
->s_flex_groups
[f
].used_dirs
);
785 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
)) {
786 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
, gdp
);
787 ext4_unlock_group(sb
, group
);
790 BUFFER_TRACE(inode_bitmap_bh
, "call ext4_handle_dirty_metadata");
791 err
= ext4_handle_dirty_metadata(handle
, NULL
, inode_bitmap_bh
);
795 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
796 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
800 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
802 percpu_counter_inc(&sbi
->s_dirs_counter
);
803 ext4_mark_super_dirty(sb
);
805 if (sbi
->s_log_groups_per_flex
) {
806 flex_group
= ext4_flex_group(sbi
, group
);
807 atomic_dec(&sbi
->s_flex_groups
[flex_group
].free_inodes
);
810 inode
->i_mode
= mode
;
811 inode
->i_uid
= owner
[0];
812 inode
->i_gid
= owner
[1];
813 } else if (test_opt(sb
, GRPID
)) {
814 inode
->i_mode
= mode
;
815 inode
->i_uid
= current_fsuid();
816 inode
->i_gid
= dir
->i_gid
;
818 inode_init_owner(inode
, dir
, mode
);
820 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
821 /* This is the optimal IO size (for stat), not the fs block size */
823 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
824 ext4_current_time(inode
);
826 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
827 ei
->i_dir_start_lookup
= 0;
830 /* Don't inherit extent flag from directory, amongst others. */
832 ext4_mask_flags(mode
, EXT4_I(dir
)->i_flags
& EXT4_FL_INHERITED
);
835 ei
->i_block_group
= group
;
836 ei
->i_last_alloc_group
= ~0;
838 ext4_set_inode_flags(inode
);
839 if (IS_DIRSYNC(inode
))
840 ext4_handle_sync(handle
);
841 if (insert_inode_locked(inode
) < 0) {
843 * Likely a bitmap corruption causing inode to be allocated
849 spin_lock(&sbi
->s_next_gen_lock
);
850 inode
->i_generation
= sbi
->s_next_generation
++;
851 spin_unlock(&sbi
->s_next_gen_lock
);
853 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
854 ext4_set_inode_state(inode
, EXT4_STATE_NEW
);
856 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
859 dquot_initialize(inode
);
860 err
= dquot_alloc_inode(inode
);
864 err
= ext4_init_acl(handle
, inode
, dir
);
868 err
= ext4_init_security(handle
, inode
, dir
, qstr
);
872 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
873 /* set extent flag only for directory, file and normal symlink*/
874 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
875 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
876 ext4_ext_tree_init(handle
, inode
);
880 if (ext4_handle_valid(handle
)) {
881 ei
->i_sync_tid
= handle
->h_transaction
->t_tid
;
882 ei
->i_datasync_tid
= handle
->h_transaction
->t_tid
;
885 err
= ext4_mark_inode_dirty(handle
, inode
);
887 ext4_std_error(sb
, err
);
891 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
892 trace_ext4_allocate_inode(inode
, dir
, mode
);
895 ext4_std_error(sb
, err
);
900 brelse(inode_bitmap_bh
);
904 dquot_free_inode(inode
);
908 inode
->i_flags
|= S_NOQUOTA
;
910 unlock_new_inode(inode
);
912 brelse(inode_bitmap_bh
);
916 /* Verify that we are loading a valid orphan from disk */
917 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
919 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
920 ext4_group_t block_group
;
922 struct buffer_head
*bitmap_bh
;
923 struct inode
*inode
= NULL
;
926 /* Error cases - e2fsck has already cleaned up for us */
928 ext4_warning(sb
, "bad orphan ino %lu! e2fsck was run?", ino
);
932 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
933 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
934 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
936 ext4_warning(sb
, "inode bitmap error for orphan %lu", ino
);
940 /* Having the inode bit set should be a 100% indicator that this
941 * is a valid orphan (no e2fsck run on fs). Orphans also include
942 * inodes that were being truncated, so we can't check i_nlink==0.
944 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
947 inode
= ext4_iget(sb
, ino
);
952 * If the orphans has i_nlinks > 0 then it should be able to be
953 * truncated, otherwise it won't be removed from the orphan list
954 * during processing and an infinite loop will result.
956 if (inode
->i_nlink
&& !ext4_can_truncate(inode
))
959 if (NEXT_ORPHAN(inode
) > max_ino
)
965 err
= PTR_ERR(inode
);
968 ext4_warning(sb
, "bad orphan inode %lu! e2fsck was run?", ino
);
969 printk(KERN_NOTICE
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
970 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
971 ext4_test_bit(bit
, bitmap_bh
->b_data
));
972 printk(KERN_NOTICE
"inode=%p\n", inode
);
974 printk(KERN_NOTICE
"is_bad_inode(inode)=%d\n",
975 is_bad_inode(inode
));
976 printk(KERN_NOTICE
"NEXT_ORPHAN(inode)=%u\n",
978 printk(KERN_NOTICE
"max_ino=%lu\n", max_ino
);
979 printk(KERN_NOTICE
"i_nlink=%u\n", inode
->i_nlink
);
980 /* Avoid freeing blocks if we got a bad deleted inode */
981 if (inode
->i_nlink
== 0)
990 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
992 unsigned long desc_count
;
993 struct ext4_group_desc
*gdp
;
994 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
996 struct ext4_super_block
*es
;
997 unsigned long bitmap_count
, x
;
998 struct buffer_head
*bitmap_bh
= NULL
;
1000 es
= EXT4_SB(sb
)->s_es
;
1004 for (i
= 0; i
< ngroups
; i
++) {
1005 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1008 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1010 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1014 x
= ext4_count_free(bitmap_bh
, EXT4_INODES_PER_GROUP(sb
) / 8);
1015 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1016 (unsigned long) i
, ext4_free_inodes_count(sb
, gdp
), x
);
1020 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1021 "stored = %u, computed = %lu, %lu\n",
1022 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1026 for (i
= 0; i
< ngroups
; i
++) {
1027 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1030 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1037 /* Called at mount-time, super-block is locked */
1038 unsigned long ext4_count_dirs(struct super_block
* sb
)
1040 unsigned long count
= 0;
1041 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1043 for (i
= 0; i
< ngroups
; i
++) {
1044 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1047 count
+= ext4_used_dirs_count(sb
, gdp
);
1053 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1054 * inode table. Must be called without any spinlock held. The only place
1055 * where it is called from on active part of filesystem is ext4lazyinit
1056 * thread, so we do not need any special locks, however we have to prevent
1057 * inode allocation from the current group, so we take alloc_sem lock, to
1058 * block ext4_new_inode() until we are finished.
1060 int ext4_init_inode_table(struct super_block
*sb
, ext4_group_t group
,
1063 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
1064 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1065 struct ext4_group_desc
*gdp
= NULL
;
1066 struct buffer_head
*group_desc_bh
;
1069 int num
, ret
= 0, used_blks
= 0;
1071 /* This should not happen, but just to be sure check this */
1072 if (sb
->s_flags
& MS_RDONLY
) {
1077 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
1082 * We do not need to lock this, because we are the only one
1083 * handling this flag.
1085 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_ZEROED
))
1088 handle
= ext4_journal_start_sb(sb
, 1);
1089 if (IS_ERR(handle
)) {
1090 ret
= PTR_ERR(handle
);
1094 down_write(&grp
->alloc_sem
);
1096 * If inode bitmap was already initialized there may be some
1097 * used inodes so we need to skip blocks with used inodes in
1100 if (!(gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)))
1101 used_blks
= DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb
) -
1102 ext4_itable_unused_count(sb
, gdp
)),
1103 sbi
->s_inodes_per_block
);
1105 if ((used_blks
< 0) || (used_blks
> sbi
->s_itb_per_group
)) {
1106 ext4_error(sb
, "Something is wrong with group %u: "
1107 "used itable blocks: %d; "
1108 "itable unused count: %u",
1110 ext4_itable_unused_count(sb
, gdp
));
1115 blk
= ext4_inode_table(sb
, gdp
) + used_blks
;
1116 num
= sbi
->s_itb_per_group
- used_blks
;
1118 BUFFER_TRACE(group_desc_bh
, "get_write_access");
1119 ret
= ext4_journal_get_write_access(handle
,
1125 * Skip zeroout if the inode table is full. But we set the ZEROED
1126 * flag anyway, because obviously, when it is full it does not need
1129 if (unlikely(num
== 0))
1132 ext4_debug("going to zero out inode table in group %d\n",
1134 ret
= sb_issue_zeroout(sb
, blk
, num
, GFP_NOFS
);
1138 blkdev_issue_flush(sb
->s_bdev
, GFP_NOFS
, NULL
);
1141 ext4_lock_group(sb
, group
);
1142 gdp
->bg_flags
|= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
1143 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
, gdp
);
1144 ext4_unlock_group(sb
, group
);
1146 BUFFER_TRACE(group_desc_bh
,
1147 "call ext4_handle_dirty_metadata");
1148 ret
= ext4_handle_dirty_metadata(handle
, NULL
,
1152 up_write(&grp
->alloc_sem
);
1153 ext4_journal_stop(handle
);