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>
27 #include "ext4_jbd2.h"
33 * ialloc.c contains the inodes allocation and deallocation routines
37 * The free inodes are managed by bitmaps. A file system contains several
38 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
39 * block for inodes, N blocks for the inode table and data blocks.
41 * The file system contains group descriptors which are located after the
42 * super block. Each descriptor contains the number of the bitmap block and
43 * the free blocks count in the block.
47 * To avoid calling the atomic setbit hundreds or thousands of times, we only
48 * need to use it within a single byte (to ensure we get endianness right).
49 * We can use memset for the rest of the bitmap as there are no other users.
51 void mark_bitmap_end(int start_bit
, int end_bit
, char *bitmap
)
55 if (start_bit
>= end_bit
)
58 ext4_debug("mark end bits +%d through +%d used\n", start_bit
, end_bit
);
59 for (i
= start_bit
; i
< ((start_bit
+ 7) & ~7UL); i
++)
60 ext4_set_bit(i
, bitmap
);
62 memset(bitmap
+ (i
>> 3), 0xff, (end_bit
- i
) >> 3);
65 /* Initializes an uninitialized inode bitmap */
66 unsigned ext4_init_inode_bitmap(struct super_block
*sb
, struct buffer_head
*bh
,
67 ext4_group_t block_group
,
68 struct ext4_group_desc
*gdp
)
70 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
72 J_ASSERT_BH(bh
, buffer_locked(bh
));
74 /* If checksum is bad mark all blocks and inodes use to prevent
75 * allocation, essentially implementing a per-group read-only flag. */
76 if (!ext4_group_desc_csum_verify(sbi
, block_group
, gdp
)) {
77 ext4_error(sb
, __func__
, "Checksum bad for group %u",
79 ext4_free_blks_set(sb
, gdp
, 0);
80 ext4_free_inodes_set(sb
, gdp
, 0);
81 ext4_itable_unused_set(sb
, gdp
, 0);
82 memset(bh
->b_data
, 0xff, sb
->s_blocksize
);
86 memset(bh
->b_data
, 0, (EXT4_INODES_PER_GROUP(sb
) + 7) / 8);
87 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
90 return EXT4_INODES_PER_GROUP(sb
);
94 * Read the inode allocation bitmap for a given block_group, reading
95 * into the specified slot in the superblock's bitmap cache.
97 * Return buffer_head of bitmap on success or NULL.
99 static struct buffer_head
*
100 ext4_read_inode_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
102 struct ext4_group_desc
*desc
;
103 struct buffer_head
*bh
= NULL
;
104 ext4_fsblk_t bitmap_blk
;
106 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
109 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
110 bh
= sb_getblk(sb
, bitmap_blk
);
112 ext4_error(sb
, __func__
,
113 "Cannot read inode bitmap - "
114 "block_group = %u, inode_bitmap = %llu",
115 block_group
, bitmap_blk
);
118 if (bitmap_uptodate(bh
))
122 if (bitmap_uptodate(bh
)) {
126 spin_lock(sb_bgl_lock(EXT4_SB(sb
), block_group
));
127 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
128 ext4_init_inode_bitmap(sb
, bh
, block_group
, desc
);
129 set_bitmap_uptodate(bh
);
130 set_buffer_uptodate(bh
);
131 spin_unlock(sb_bgl_lock(EXT4_SB(sb
), block_group
));
135 spin_unlock(sb_bgl_lock(EXT4_SB(sb
), block_group
));
136 if (buffer_uptodate(bh
)) {
138 * if not uninit if bh is uptodate,
139 * bitmap is also uptodate
141 set_bitmap_uptodate(bh
);
146 * submit the buffer_head for read. We can
147 * safely mark the bitmap as uptodate now.
148 * We do it here so the bitmap uptodate bit
149 * get set with buffer lock held.
151 set_bitmap_uptodate(bh
);
152 if (bh_submit_read(bh
) < 0) {
154 ext4_error(sb
, __func__
,
155 "Cannot read inode bitmap - "
156 "block_group = %u, inode_bitmap = %llu",
157 block_group
, bitmap_blk
);
164 * NOTE! When we get the inode, we're the only people
165 * that have access to it, and as such there are no
166 * race conditions we have to worry about. The inode
167 * is not on the hash-lists, and it cannot be reached
168 * through the filesystem because the directory entry
169 * has been deleted earlier.
171 * HOWEVER: we must make sure that we get no aliases,
172 * which means that we have to call "clear_inode()"
173 * _before_ we mark the inode not in use in the inode
174 * bitmaps. Otherwise a newly created file might use
175 * the same inode number (not actually the same pointer
176 * though), and then we'd have two inodes sharing the
177 * same inode number and space on the harddisk.
179 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
181 struct super_block
*sb
= inode
->i_sb
;
184 struct buffer_head
*bitmap_bh
= NULL
;
185 struct buffer_head
*bh2
;
186 ext4_group_t block_group
;
188 struct ext4_group_desc
*gdp
;
189 struct ext4_super_block
*es
;
190 struct ext4_sb_info
*sbi
;
191 int fatal
= 0, err
, count
;
192 ext4_group_t flex_group
;
194 if (atomic_read(&inode
->i_count
) > 1) {
195 printk(KERN_ERR
"ext4_free_inode: inode has count=%d\n",
196 atomic_read(&inode
->i_count
));
199 if (inode
->i_nlink
) {
200 printk(KERN_ERR
"ext4_free_inode: inode has nlink=%d\n",
205 printk(KERN_ERR
"ext4_free_inode: inode on "
206 "nonexistent device\n");
212 ext4_debug("freeing inode %lu\n", ino
);
213 trace_mark(ext4_free_inode
,
214 "dev %s ino %lu mode %d uid %lu gid %lu bocks %llu",
215 sb
->s_id
, inode
->i_ino
, inode
->i_mode
,
216 (unsigned long) inode
->i_uid
, (unsigned long) inode
->i_gid
,
217 (unsigned long long) inode
->i_blocks
);
220 * Note: we must free any quota before locking the superblock,
221 * as writing the quota to disk may need the lock as well.
224 ext4_xattr_delete_inode(handle
, inode
);
225 DQUOT_FREE_INODE(inode
);
228 is_directory
= S_ISDIR(inode
->i_mode
);
230 /* Do this BEFORE marking the inode not in use or returning an error */
233 es
= EXT4_SB(sb
)->s_es
;
234 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
235 ext4_error(sb
, "ext4_free_inode",
236 "reserved or nonexistent inode %lu", ino
);
239 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
240 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
241 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
245 BUFFER_TRACE(bitmap_bh
, "get_write_access");
246 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
250 /* Ok, now we can actually update the inode bitmaps.. */
251 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
252 bit
, bitmap_bh
->b_data
))
253 ext4_error(sb
, "ext4_free_inode",
254 "bit already cleared for inode %lu", ino
);
256 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
258 BUFFER_TRACE(bh2
, "get_write_access");
259 fatal
= ext4_journal_get_write_access(handle
, bh2
);
260 if (fatal
) goto error_return
;
263 spin_lock(sb_bgl_lock(sbi
, block_group
));
264 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
265 ext4_free_inodes_set(sb
, gdp
, count
);
267 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
268 ext4_used_dirs_set(sb
, gdp
, count
);
270 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
,
272 spin_unlock(sb_bgl_lock(sbi
, block_group
));
273 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
275 percpu_counter_dec(&sbi
->s_dirs_counter
);
277 if (sbi
->s_log_groups_per_flex
) {
278 flex_group
= ext4_flex_group(sbi
, block_group
);
279 spin_lock(sb_bgl_lock(sbi
, flex_group
));
280 sbi
->s_flex_groups
[flex_group
].free_inodes
++;
281 spin_unlock(sb_bgl_lock(sbi
, flex_group
));
284 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
285 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
286 if (!fatal
) fatal
= err
;
288 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
289 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
295 ext4_std_error(sb
, fatal
);
299 * There are two policies for allocating an inode. If the new inode is
300 * a directory, then a forward search is made for a block group with both
301 * free space and a low directory-to-inode ratio; if that fails, then of
302 * the groups with above-average free space, that group with the fewest
303 * directories already is chosen.
305 * For other inodes, search forward from the parent directory\'s block
306 * group to find a free inode.
308 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
,
309 ext4_group_t
*best_group
)
311 ext4_group_t ngroups
= EXT4_SB(sb
)->s_groups_count
;
312 unsigned int freei
, avefreei
;
313 struct ext4_group_desc
*desc
, *best_desc
= NULL
;
317 freei
= percpu_counter_read_positive(&EXT4_SB(sb
)->s_freeinodes_counter
);
318 avefreei
= freei
/ ngroups
;
320 for (group
= 0; group
< ngroups
; group
++) {
321 desc
= ext4_get_group_desc(sb
, group
, NULL
);
322 if (!desc
|| !ext4_free_inodes_count(sb
, desc
))
324 if (ext4_free_inodes_count(sb
, desc
) < avefreei
)
327 (ext4_free_blks_count(sb
, desc
) >
328 ext4_free_blks_count(sb
, best_desc
))) {
337 #define free_block_ratio 10
339 static int find_group_flex(struct super_block
*sb
, struct inode
*parent
,
340 ext4_group_t
*best_group
)
342 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
343 struct ext4_group_desc
*desc
;
344 struct buffer_head
*bh
;
345 struct flex_groups
*flex_group
= sbi
->s_flex_groups
;
346 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
347 ext4_group_t parent_fbg_group
= ext4_flex_group(sbi
, parent_group
);
348 ext4_group_t ngroups
= sbi
->s_groups_count
;
349 int flex_size
= ext4_flex_bg_size(sbi
);
350 ext4_group_t best_flex
= parent_fbg_group
;
351 int blocks_per_flex
= sbi
->s_blocks_per_group
* flex_size
;
352 int flexbg_free_blocks
;
353 int flex_freeb_ratio
;
354 ext4_group_t n_fbg_groups
;
357 n_fbg_groups
= (sbi
->s_groups_count
+ flex_size
- 1) >>
358 sbi
->s_log_groups_per_flex
;
360 find_close_to_parent
:
361 flexbg_free_blocks
= flex_group
[best_flex
].free_blocks
;
362 flex_freeb_ratio
= flexbg_free_blocks
* 100 / blocks_per_flex
;
363 if (flex_group
[best_flex
].free_inodes
&&
364 flex_freeb_ratio
> free_block_ratio
)
367 if (best_flex
&& best_flex
== parent_fbg_group
) {
369 goto find_close_to_parent
;
372 for (i
= 0; i
< n_fbg_groups
; i
++) {
373 if (i
== parent_fbg_group
|| i
== parent_fbg_group
- 1)
376 flexbg_free_blocks
= flex_group
[i
].free_blocks
;
377 flex_freeb_ratio
= flexbg_free_blocks
* 100 / blocks_per_flex
;
379 if (flex_freeb_ratio
> free_block_ratio
&&
380 flex_group
[i
].free_inodes
) {
385 if (flex_group
[best_flex
].free_inodes
== 0 ||
386 (flex_group
[i
].free_blocks
>
387 flex_group
[best_flex
].free_blocks
&&
388 flex_group
[i
].free_inodes
))
392 if (!flex_group
[best_flex
].free_inodes
||
393 !flex_group
[best_flex
].free_blocks
)
397 for (i
= best_flex
* flex_size
; i
< ngroups
&&
398 i
< (best_flex
+ 1) * flex_size
; i
++) {
399 desc
= ext4_get_group_desc(sb
, i
, &bh
);
400 if (ext4_free_inodes_count(sb
, desc
)) {
412 * Orlov's allocator for directories.
414 * We always try to spread first-level directories.
416 * If there are blockgroups with both free inodes and free blocks counts
417 * not worse than average we return one with smallest directory count.
418 * Otherwise we simply return a random group.
420 * For the rest rules look so:
422 * It's OK to put directory into a group unless
423 * it has too many directories already (max_dirs) or
424 * it has too few free inodes left (min_inodes) or
425 * it has too few free blocks left (min_blocks) or
426 * it's already running too large debt (max_debt).
427 * Parent's group is preferred, if it doesn't satisfy these
428 * conditions we search cyclically through the rest. If none
429 * of the groups look good we just look for a group with more
430 * free inodes than average (starting at parent's group).
432 * Debt is incremented each time we allocate a directory and decremented
433 * when we allocate an inode, within 0--255.
436 #define INODE_COST 64
437 #define BLOCK_COST 256
439 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
442 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
443 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
444 struct ext4_super_block
*es
= sbi
->s_es
;
445 ext4_group_t ngroups
= sbi
->s_groups_count
;
446 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
447 unsigned int freei
, avefreei
;
448 ext4_fsblk_t freeb
, avefreeb
;
449 ext4_fsblk_t blocks_per_dir
;
451 int max_debt
, max_dirs
, min_inodes
;
452 ext4_grpblk_t min_blocks
;
454 struct ext4_group_desc
*desc
;
456 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
457 avefreei
= freei
/ ngroups
;
458 freeb
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
460 do_div(avefreeb
, ngroups
);
461 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
463 if ((parent
== sb
->s_root
->d_inode
) ||
464 (EXT4_I(parent
)->i_flags
& EXT4_TOPDIR_FL
)) {
465 int best_ndir
= inodes_per_group
;
469 get_random_bytes(&grp
, sizeof(grp
));
470 parent_group
= (unsigned)grp
% ngroups
;
471 for (i
= 0; i
< ngroups
; i
++) {
472 grp
= (parent_group
+ i
) % ngroups
;
473 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
474 if (!desc
|| !ext4_free_inodes_count(sb
, desc
))
476 if (ext4_used_dirs_count(sb
, desc
) >= best_ndir
)
478 if (ext4_free_inodes_count(sb
, desc
) < avefreei
)
480 if (ext4_free_blks_count(sb
, desc
) < avefreeb
)
484 best_ndir
= ext4_used_dirs_count(sb
, desc
);
491 blocks_per_dir
= ext4_blocks_count(es
) - freeb
;
492 do_div(blocks_per_dir
, ndirs
);
494 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
495 min_inodes
= avefreei
- inodes_per_group
/ 4;
496 min_blocks
= avefreeb
- EXT4_BLOCKS_PER_GROUP(sb
) / 4;
498 max_debt
= EXT4_BLOCKS_PER_GROUP(sb
);
499 max_debt
/= max_t(int, blocks_per_dir
, BLOCK_COST
);
500 if (max_debt
* INODE_COST
> inodes_per_group
)
501 max_debt
= inodes_per_group
/ INODE_COST
;
507 for (i
= 0; i
< ngroups
; i
++) {
508 *group
= (parent_group
+ i
) % ngroups
;
509 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
510 if (!desc
|| !ext4_free_inodes_count(sb
, desc
))
512 if (ext4_used_dirs_count(sb
, desc
) >= max_dirs
)
514 if (ext4_free_inodes_count(sb
, desc
) < min_inodes
)
516 if (ext4_free_blks_count(sb
, desc
) < min_blocks
)
522 for (i
= 0; i
< ngroups
; i
++) {
523 *group
= (parent_group
+ i
) % ngroups
;
524 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
525 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
526 ext4_free_inodes_count(sb
, desc
) >= avefreei
)
532 * The free-inodes counter is approximate, and for really small
533 * filesystems the above test can fail to find any blockgroups
542 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
545 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
546 ext4_group_t ngroups
= EXT4_SB(sb
)->s_groups_count
;
547 struct ext4_group_desc
*desc
;
551 * Try to place the inode in its parent directory
553 *group
= parent_group
;
554 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
555 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
556 ext4_free_blks_count(sb
, desc
))
560 * We're going to place this inode in a different blockgroup from its
561 * parent. We want to cause files in a common directory to all land in
562 * the same blockgroup. But we want files which are in a different
563 * directory which shares a blockgroup with our parent to land in a
564 * different blockgroup.
566 * So add our directory's i_ino into the starting point for the hash.
568 *group
= (*group
+ parent
->i_ino
) % ngroups
;
571 * Use a quadratic hash to find a group with a free inode and some free
574 for (i
= 1; i
< ngroups
; i
<<= 1) {
576 if (*group
>= ngroups
)
578 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
579 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
580 ext4_free_blks_count(sb
, desc
))
585 * That failed: try linear search for a free inode, even if that group
586 * has no free blocks.
588 *group
= parent_group
;
589 for (i
= 0; i
< ngroups
; i
++) {
590 if (++*group
>= ngroups
)
592 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
593 if (desc
&& ext4_free_inodes_count(sb
, desc
))
601 * claim the inode from the inode bitmap. If the group
602 * is uninit we need to take the groups's sb_bgl_lock
603 * and clear the uninit flag. The inode bitmap update
604 * and group desc uninit flag clear should be done
605 * after holding sb_bgl_lock so that ext4_read_inode_bitmap
606 * doesn't race with the ext4_claim_inode
608 static int ext4_claim_inode(struct super_block
*sb
,
609 struct buffer_head
*inode_bitmap_bh
,
610 unsigned long ino
, ext4_group_t group
, int mode
)
612 int free
= 0, retval
= 0, count
;
613 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
614 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, group
, NULL
);
616 spin_lock(sb_bgl_lock(sbi
, group
));
617 if (ext4_set_bit(ino
, inode_bitmap_bh
->b_data
)) {
618 /* not a free inode */
623 if ((group
== 0 && ino
< EXT4_FIRST_INO(sb
)) ||
624 ino
> EXT4_INODES_PER_GROUP(sb
)) {
625 spin_unlock(sb_bgl_lock(sbi
, group
));
626 ext4_error(sb
, __func__
,
627 "reserved inode or inode > inodes count - "
628 "block_group = %u, inode=%lu", group
,
629 ino
+ group
* EXT4_INODES_PER_GROUP(sb
));
632 /* If we didn't allocate from within the initialized part of the inode
633 * table then we need to initialize up to this inode. */
634 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
)) {
636 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
637 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
638 /* When marking the block group with
639 * ~EXT4_BG_INODE_UNINIT we don't want to depend
640 * on the value of bg_itable_unused even though
641 * mke2fs could have initialized the same for us.
642 * Instead we calculated the value below
647 free
= EXT4_INODES_PER_GROUP(sb
) -
648 ext4_itable_unused_count(sb
, gdp
);
652 * Check the relative inode number against the last used
653 * relative inode number in this group. if it is greater
654 * we need to update the bg_itable_unused count
658 ext4_itable_unused_set(sb
, gdp
,
659 (EXT4_INODES_PER_GROUP(sb
) - ino
));
661 count
= ext4_free_inodes_count(sb
, gdp
) - 1;
662 ext4_free_inodes_set(sb
, gdp
, count
);
664 count
= ext4_used_dirs_count(sb
, gdp
) + 1;
665 ext4_used_dirs_set(sb
, gdp
, count
);
667 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
, gdp
);
669 spin_unlock(sb_bgl_lock(sbi
, group
));
674 * There are two policies for allocating an inode. If the new inode is
675 * a directory, then a forward search is made for a block group with both
676 * free space and a low directory-to-inode ratio; if that fails, then of
677 * the groups with above-average free space, that group with the fewest
678 * directories already is chosen.
680 * For other inodes, search forward from the parent directory's block
681 * group to find a free inode.
683 struct inode
*ext4_new_inode(handle_t
*handle
, struct inode
*dir
, int mode
)
685 struct super_block
*sb
;
686 struct buffer_head
*inode_bitmap_bh
= NULL
;
687 struct buffer_head
*group_desc_bh
;
688 ext4_group_t group
= 0;
689 unsigned long ino
= 0;
691 struct ext4_group_desc
*gdp
= NULL
;
692 struct ext4_super_block
*es
;
693 struct ext4_inode_info
*ei
;
694 struct ext4_sb_info
*sbi
;
699 ext4_group_t flex_group
;
701 /* Cannot create files in a deleted directory */
702 if (!dir
|| !dir
->i_nlink
)
703 return ERR_PTR(-EPERM
);
706 trace_mark(ext4_request_inode
, "dev %s dir %lu mode %d", sb
->s_id
,
708 inode
= new_inode(sb
);
710 return ERR_PTR(-ENOMEM
);
716 if (sbi
->s_log_groups_per_flex
) {
717 ret2
= find_group_flex(sb
, dir
, &group
);
719 ret2
= find_group_other(sb
, dir
, &group
);
720 if (ret2
== 0 && printk_ratelimit())
721 printk(KERN_NOTICE
"ext4: find_group_flex "
722 "failed, fallback succeeded dir %lu\n",
729 if (test_opt(sb
, OLDALLOC
))
730 ret2
= find_group_dir(sb
, dir
, &group
);
732 ret2
= find_group_orlov(sb
, dir
, &group
);
734 ret2
= find_group_other(sb
, dir
, &group
);
741 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
744 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
748 brelse(inode_bitmap_bh
);
749 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
750 if (!inode_bitmap_bh
)
755 repeat_in_this_group
:
756 ino
= ext4_find_next_zero_bit((unsigned long *)
757 inode_bitmap_bh
->b_data
,
758 EXT4_INODES_PER_GROUP(sb
), ino
);
760 if (ino
< EXT4_INODES_PER_GROUP(sb
)) {
762 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
763 err
= ext4_journal_get_write_access(handle
,
768 BUFFER_TRACE(group_desc_bh
, "get_write_access");
769 err
= ext4_journal_get_write_access(handle
,
773 if (!ext4_claim_inode(sb
, inode_bitmap_bh
,
776 BUFFER_TRACE(inode_bitmap_bh
,
777 "call ext4_handle_dirty_metadata");
778 err
= ext4_handle_dirty_metadata(handle
,
783 /* zero bit is inode number 1*/
788 ext4_handle_release_buffer(handle
, inode_bitmap_bh
);
789 ext4_handle_release_buffer(handle
, group_desc_bh
);
791 if (++ino
< EXT4_INODES_PER_GROUP(sb
))
792 goto repeat_in_this_group
;
796 * This case is possible in concurrent environment. It is very
797 * rare. We cannot repeat the find_group_xxx() call because
798 * that will simply return the same blockgroup, because the
799 * group descriptor metadata has not yet been updated.
800 * So we just go onto the next blockgroup.
802 if (++group
== sbi
->s_groups_count
)
809 /* We may have to initialize the block bitmap if it isn't already */
810 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
) &&
811 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
812 struct buffer_head
*block_bitmap_bh
;
814 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
815 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
816 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
818 brelse(block_bitmap_bh
);
823 spin_lock(sb_bgl_lock(sbi
, group
));
824 /* recheck and clear flag under lock if we still need to */
825 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
826 free
= ext4_free_blocks_after_init(sb
, group
, gdp
);
827 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
828 ext4_free_blks_set(sb
, gdp
, free
);
829 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
,
832 spin_unlock(sb_bgl_lock(sbi
, group
));
834 /* Don't need to dirty bitmap block if we didn't change it */
836 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
837 err
= ext4_handle_dirty_metadata(handle
,
838 NULL
, block_bitmap_bh
);
841 brelse(block_bitmap_bh
);
845 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
846 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
850 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
852 percpu_counter_inc(&sbi
->s_dirs_counter
);
855 if (sbi
->s_log_groups_per_flex
) {
856 flex_group
= ext4_flex_group(sbi
, group
);
857 spin_lock(sb_bgl_lock(sbi
, flex_group
));
858 sbi
->s_flex_groups
[flex_group
].free_inodes
--;
859 spin_unlock(sb_bgl_lock(sbi
, flex_group
));
862 inode
->i_uid
= current_fsuid();
863 if (test_opt(sb
, GRPID
))
864 inode
->i_gid
= dir
->i_gid
;
865 else if (dir
->i_mode
& S_ISGID
) {
866 inode
->i_gid
= dir
->i_gid
;
870 inode
->i_gid
= current_fsgid();
871 inode
->i_mode
= mode
;
873 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
874 /* This is the optimal IO size (for stat), not the fs block size */
876 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
877 ext4_current_time(inode
);
879 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
880 ei
->i_dir_start_lookup
= 0;
884 * Don't inherit extent flag from directory. We set extent flag on
885 * newly created directory and file only if -o extent mount option is
888 ei
->i_flags
= EXT4_I(dir
)->i_flags
& ~(EXT4_INDEX_FL
|EXT4_EXTENTS_FL
);
890 ei
->i_flags
&= ~(EXT4_IMMUTABLE_FL
|EXT4_APPEND_FL
);
891 /* dirsync only applies to directories */
893 ei
->i_flags
&= ~EXT4_DIRSYNC_FL
;
896 ei
->i_block_group
= group
;
898 ext4_set_inode_flags(inode
);
899 if (IS_DIRSYNC(inode
))
900 ext4_handle_sync(handle
);
901 if (insert_inode_locked(inode
) < 0) {
905 spin_lock(&sbi
->s_next_gen_lock
);
906 inode
->i_generation
= sbi
->s_next_generation
++;
907 spin_unlock(&sbi
->s_next_gen_lock
);
909 ei
->i_state
= EXT4_STATE_NEW
;
911 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
914 if (DQUOT_ALLOC_INODE(inode
)) {
919 err
= ext4_init_acl(handle
, inode
, dir
);
923 err
= ext4_init_security(handle
, inode
, dir
);
927 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
928 /* set extent flag only for directory, file and normal symlink*/
929 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
930 EXT4_I(inode
)->i_flags
|= EXT4_EXTENTS_FL
;
931 ext4_ext_tree_init(handle
, inode
);
935 err
= ext4_mark_inode_dirty(handle
, inode
);
937 ext4_std_error(sb
, err
);
941 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
942 trace_mark(ext4_allocate_inode
, "dev %s ino %lu dir %lu mode %d",
943 sb
->s_id
, inode
->i_ino
, dir
->i_ino
, mode
);
946 ext4_std_error(sb
, err
);
951 brelse(inode_bitmap_bh
);
955 DQUOT_FREE_INODE(inode
);
959 inode
->i_flags
|= S_NOQUOTA
;
961 unlock_new_inode(inode
);
963 brelse(inode_bitmap_bh
);
967 /* Verify that we are loading a valid orphan from disk */
968 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
970 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
971 ext4_group_t block_group
;
973 struct buffer_head
*bitmap_bh
;
974 struct inode
*inode
= NULL
;
977 /* Error cases - e2fsck has already cleaned up for us */
979 ext4_warning(sb
, __func__
,
980 "bad orphan ino %lu! e2fsck was run?", ino
);
984 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
985 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
986 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
988 ext4_warning(sb
, __func__
,
989 "inode bitmap error for orphan %lu", ino
);
993 /* Having the inode bit set should be a 100% indicator that this
994 * is a valid orphan (no e2fsck run on fs). Orphans also include
995 * inodes that were being truncated, so we can't check i_nlink==0.
997 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
1000 inode
= ext4_iget(sb
, ino
);
1005 * If the orphans has i_nlinks > 0 then it should be able to be
1006 * truncated, otherwise it won't be removed from the orphan list
1007 * during processing and an infinite loop will result.
1009 if (inode
->i_nlink
&& !ext4_can_truncate(inode
))
1012 if (NEXT_ORPHAN(inode
) > max_ino
)
1018 err
= PTR_ERR(inode
);
1021 ext4_warning(sb
, __func__
,
1022 "bad orphan inode %lu! e2fsck was run?", ino
);
1023 printk(KERN_NOTICE
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
1024 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
1025 ext4_test_bit(bit
, bitmap_bh
->b_data
));
1026 printk(KERN_NOTICE
"inode=%p\n", inode
);
1028 printk(KERN_NOTICE
"is_bad_inode(inode)=%d\n",
1029 is_bad_inode(inode
));
1030 printk(KERN_NOTICE
"NEXT_ORPHAN(inode)=%u\n",
1031 NEXT_ORPHAN(inode
));
1032 printk(KERN_NOTICE
"max_ino=%lu\n", max_ino
);
1033 printk(KERN_NOTICE
"i_nlink=%u\n", inode
->i_nlink
);
1034 /* Avoid freeing blocks if we got a bad deleted inode */
1035 if (inode
->i_nlink
== 0)
1036 inode
->i_blocks
= 0;
1041 return ERR_PTR(err
);
1044 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
1046 unsigned long desc_count
;
1047 struct ext4_group_desc
*gdp
;
1050 struct ext4_super_block
*es
;
1051 unsigned long bitmap_count
, x
;
1052 struct buffer_head
*bitmap_bh
= NULL
;
1054 es
= EXT4_SB(sb
)->s_es
;
1058 for (i
= 0; i
< EXT4_SB(sb
)->s_groups_count
; i
++) {
1059 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1062 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1064 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1068 x
= ext4_count_free(bitmap_bh
, EXT4_INODES_PER_GROUP(sb
) / 8);
1069 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1070 i
, ext4_free_inodes_count(sb
, gdp
), x
);
1074 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1075 "stored = %u, computed = %lu, %lu\n",
1076 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1080 for (i
= 0; i
< EXT4_SB(sb
)->s_groups_count
; i
++) {
1081 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1084 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1091 /* Called at mount-time, super-block is locked */
1092 unsigned long ext4_count_dirs(struct super_block
* sb
)
1094 unsigned long count
= 0;
1097 for (i
= 0; i
< EXT4_SB(sb
)->s_groups_count
; i
++) {
1098 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1101 count
+= ext4_used_dirs_count(sb
, gdp
);