2 * linux/fs/ext4/ialloc.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
9 * BSD ufs-inspired inode and directory allocation by
10 * Stephen Tweedie (sct@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
15 #include <linux/time.h>
17 #include <linux/jbd2.h>
18 #include <linux/stat.h>
19 #include <linux/string.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
22 #include <linux/random.h>
23 #include <linux/bitops.h>
24 #include <linux/blkdev.h>
25 #include <asm/byteorder.h>
28 #include "ext4_jbd2.h"
32 #include <trace/events/ext4.h>
35 * ialloc.c contains the inodes allocation and deallocation routines
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
53 void mark_bitmap_end(int start_bit
, int end_bit
, char *bitmap
)
57 if (start_bit
>= end_bit
)
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit
, end_bit
);
61 for (i
= start_bit
; i
< ((start_bit
+ 7) & ~7UL); i
++)
62 ext4_set_bit(i
, bitmap
);
64 memset(bitmap
+ (i
>> 3), 0xff, (end_bit
- i
) >> 3);
67 /* Initializes an uninitialized inode bitmap */
68 unsigned ext4_init_inode_bitmap(struct super_block
*sb
, struct buffer_head
*bh
,
69 ext4_group_t block_group
,
70 struct ext4_group_desc
*gdp
)
72 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
74 J_ASSERT_BH(bh
, buffer_locked(bh
));
76 /* If checksum is bad mark all blocks and inodes use to prevent
77 * allocation, essentially implementing a per-group read-only flag. */
78 if (!ext4_group_desc_csum_verify(sbi
, block_group
, gdp
)) {
79 ext4_error(sb
, __func__
, "Checksum bad for group %u",
81 ext4_free_blks_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 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
92 return EXT4_INODES_PER_GROUP(sb
);
96 * Read the inode allocation bitmap for a given block_group, reading
97 * into the specified slot in the superblock's bitmap cache.
99 * Return buffer_head of bitmap on success or NULL.
101 static struct buffer_head
*
102 ext4_read_inode_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
104 struct ext4_group_desc
*desc
;
105 struct buffer_head
*bh
= NULL
;
106 ext4_fsblk_t bitmap_blk
;
108 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
111 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
112 bh
= sb_getblk(sb
, bitmap_blk
);
114 ext4_error(sb
, __func__
,
115 "Cannot read inode bitmap - "
116 "block_group = %u, inode_bitmap = %llu",
117 block_group
, bitmap_blk
);
120 if (bitmap_uptodate(bh
))
124 if (bitmap_uptodate(bh
)) {
128 ext4_lock_group(sb
, block_group
);
129 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
130 ext4_init_inode_bitmap(sb
, bh
, block_group
, desc
);
131 set_bitmap_uptodate(bh
);
132 set_buffer_uptodate(bh
);
133 ext4_unlock_group(sb
, block_group
);
137 ext4_unlock_group(sb
, block_group
);
138 if (buffer_uptodate(bh
)) {
140 * if not uninit if bh is uptodate,
141 * bitmap is also uptodate
143 set_bitmap_uptodate(bh
);
148 * submit the buffer_head for read. We can
149 * safely mark the bitmap as uptodate now.
150 * We do it here so the bitmap uptodate bit
151 * get set with buffer lock held.
153 set_bitmap_uptodate(bh
);
154 if (bh_submit_read(bh
) < 0) {
156 ext4_error(sb
, __func__
,
157 "Cannot read inode bitmap - "
158 "block_group = %u, inode_bitmap = %llu",
159 block_group
, bitmap_blk
);
166 * NOTE! When we get the inode, we're the only people
167 * that have access to it, and as such there are no
168 * race conditions we have to worry about. The inode
169 * is not on the hash-lists, and it cannot be reached
170 * through the filesystem because the directory entry
171 * has been deleted earlier.
173 * HOWEVER: we must make sure that we get no aliases,
174 * which means that we have to call "clear_inode()"
175 * _before_ we mark the inode not in use in the inode
176 * bitmaps. Otherwise a newly created file might use
177 * the same inode number (not actually the same pointer
178 * though), and then we'd have two inodes sharing the
179 * same inode number and space on the harddisk.
181 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
183 struct super_block
*sb
= inode
->i_sb
;
186 struct buffer_head
*bitmap_bh
= NULL
;
187 struct buffer_head
*bh2
;
188 ext4_group_t block_group
;
190 struct ext4_group_desc
*gdp
;
191 struct ext4_super_block
*es
;
192 struct ext4_sb_info
*sbi
;
193 int fatal
= 0, err
, count
, cleared
;
195 if (atomic_read(&inode
->i_count
) > 1) {
196 printk(KERN_ERR
"ext4_free_inode: inode has count=%d\n",
197 atomic_read(&inode
->i_count
));
200 if (inode
->i_nlink
) {
201 printk(KERN_ERR
"ext4_free_inode: inode has nlink=%d\n",
206 printk(KERN_ERR
"ext4_free_inode: inode on "
207 "nonexistent device\n");
213 ext4_debug("freeing inode %lu\n", ino
);
214 trace_ext4_free_inode(inode
);
217 * Note: we must free any quota before locking the superblock,
218 * as writing the quota to disk may need the lock as well.
221 ext4_xattr_delete_inode(handle
, inode
);
222 vfs_dq_free_inode(inode
);
225 is_directory
= S_ISDIR(inode
->i_mode
);
227 /* Do this BEFORE marking the inode not in use or returning an error */
230 es
= EXT4_SB(sb
)->s_es
;
231 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
232 ext4_error(sb
, "ext4_free_inode",
233 "reserved or nonexistent inode %lu", ino
);
236 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
237 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
238 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
242 BUFFER_TRACE(bitmap_bh
, "get_write_access");
243 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
247 /* Ok, now we can actually update the inode bitmaps.. */
248 cleared
= ext4_clear_bit_atomic(ext4_group_lock_ptr(sb
, block_group
),
249 bit
, bitmap_bh
->b_data
);
251 ext4_error(sb
, "ext4_free_inode",
252 "bit already cleared for inode %lu", ino
);
254 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
256 BUFFER_TRACE(bh2
, "get_write_access");
257 fatal
= ext4_journal_get_write_access(handle
, bh2
);
258 if (fatal
) goto error_return
;
261 ext4_lock_group(sb
, block_group
);
262 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
263 ext4_free_inodes_set(sb
, gdp
, count
);
265 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
266 ext4_used_dirs_set(sb
, gdp
, count
);
267 if (sbi
->s_log_groups_per_flex
) {
270 f
= ext4_flex_group(sbi
, block_group
);
271 atomic_dec(&sbi
->s_flex_groups
[f
].free_inodes
);
275 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
,
277 ext4_unlock_group(sb
, block_group
);
278 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
280 percpu_counter_dec(&sbi
->s_dirs_counter
);
282 if (sbi
->s_log_groups_per_flex
) {
285 f
= ext4_flex_group(sbi
, block_group
);
286 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
289 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
290 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
291 if (!fatal
) fatal
= err
;
293 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
294 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
300 ext4_std_error(sb
, fatal
);
304 * There are two policies for allocating an inode. If the new inode is
305 * a directory, then a forward search is made for a block group with both
306 * free space and a low directory-to-inode ratio; if that fails, then of
307 * the groups with above-average free space, that group with the fewest
308 * directories already is chosen.
310 * For other inodes, search forward from the parent directory\'s block
311 * group to find a free inode.
313 static int find_group_dir(struct super_block
*sb
, struct inode
*parent
,
314 ext4_group_t
*best_group
)
316 ext4_group_t ngroups
= ext4_get_groups_count(sb
);
317 unsigned int freei
, avefreei
;
318 struct ext4_group_desc
*desc
, *best_desc
= NULL
;
322 freei
= percpu_counter_read_positive(&EXT4_SB(sb
)->s_freeinodes_counter
);
323 avefreei
= freei
/ ngroups
;
325 for (group
= 0; group
< ngroups
; group
++) {
326 desc
= ext4_get_group_desc(sb
, group
, NULL
);
327 if (!desc
|| !ext4_free_inodes_count(sb
, desc
))
329 if (ext4_free_inodes_count(sb
, desc
) < avefreei
)
332 (ext4_free_blks_count(sb
, desc
) >
333 ext4_free_blks_count(sb
, best_desc
))) {
342 #define free_block_ratio 10
344 static int find_group_flex(struct super_block
*sb
, struct inode
*parent
,
345 ext4_group_t
*best_group
)
347 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
348 struct ext4_group_desc
*desc
;
349 struct flex_groups
*flex_group
= sbi
->s_flex_groups
;
350 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
351 ext4_group_t parent_fbg_group
= ext4_flex_group(sbi
, parent_group
);
352 ext4_group_t ngroups
= ext4_get_groups_count(sb
);
353 int flex_size
= ext4_flex_bg_size(sbi
);
354 ext4_group_t best_flex
= parent_fbg_group
;
355 int blocks_per_flex
= sbi
->s_blocks_per_group
* flex_size
;
356 int flexbg_free_blocks
;
357 int flex_freeb_ratio
;
358 ext4_group_t n_fbg_groups
;
361 n_fbg_groups
= (ngroups
+ flex_size
- 1) >>
362 sbi
->s_log_groups_per_flex
;
364 find_close_to_parent
:
365 flexbg_free_blocks
= atomic_read(&flex_group
[best_flex
].free_blocks
);
366 flex_freeb_ratio
= flexbg_free_blocks
* 100 / blocks_per_flex
;
367 if (atomic_read(&flex_group
[best_flex
].free_inodes
) &&
368 flex_freeb_ratio
> free_block_ratio
)
371 if (best_flex
&& best_flex
== parent_fbg_group
) {
373 goto find_close_to_parent
;
376 for (i
= 0; i
< n_fbg_groups
; i
++) {
377 if (i
== parent_fbg_group
|| i
== parent_fbg_group
- 1)
380 flexbg_free_blocks
= atomic_read(&flex_group
[i
].free_blocks
);
381 flex_freeb_ratio
= flexbg_free_blocks
* 100 / blocks_per_flex
;
383 if (flex_freeb_ratio
> free_block_ratio
&&
384 (atomic_read(&flex_group
[i
].free_inodes
))) {
389 if ((atomic_read(&flex_group
[best_flex
].free_inodes
) == 0) ||
390 ((atomic_read(&flex_group
[i
].free_blocks
) >
391 atomic_read(&flex_group
[best_flex
].free_blocks
)) &&
392 atomic_read(&flex_group
[i
].free_inodes
)))
396 if (!atomic_read(&flex_group
[best_flex
].free_inodes
) ||
397 !atomic_read(&flex_group
[best_flex
].free_blocks
))
401 for (i
= best_flex
* flex_size
; i
< ngroups
&&
402 i
< (best_flex
+ 1) * flex_size
; i
++) {
403 desc
= ext4_get_group_desc(sb
, i
, NULL
);
404 if (ext4_free_inodes_count(sb
, desc
)) {
422 * Helper function for Orlov's allocator; returns critical information
423 * for a particular block group or flex_bg. If flex_size is 1, then g
424 * is a block group number; otherwise it is flex_bg number.
426 void get_orlov_stats(struct super_block
*sb
, ext4_group_t g
,
427 int flex_size
, struct orlov_stats
*stats
)
429 struct ext4_group_desc
*desc
;
430 struct flex_groups
*flex_group
= EXT4_SB(sb
)->s_flex_groups
;
433 stats
->free_inodes
= atomic_read(&flex_group
[g
].free_inodes
);
434 stats
->free_blocks
= atomic_read(&flex_group
[g
].free_blocks
);
435 stats
->used_dirs
= atomic_read(&flex_group
[g
].used_dirs
);
439 desc
= ext4_get_group_desc(sb
, g
, NULL
);
441 stats
->free_inodes
= ext4_free_inodes_count(sb
, desc
);
442 stats
->free_blocks
= ext4_free_blks_count(sb
, desc
);
443 stats
->used_dirs
= ext4_used_dirs_count(sb
, desc
);
445 stats
->free_inodes
= 0;
446 stats
->free_blocks
= 0;
447 stats
->used_dirs
= 0;
452 * Orlov's allocator for directories.
454 * We always try to spread first-level directories.
456 * If there are blockgroups with both free inodes and free blocks counts
457 * not worse than average we return one with smallest directory count.
458 * Otherwise we simply return a random group.
460 * For the rest rules look so:
462 * It's OK to put directory into a group unless
463 * it has too many directories already (max_dirs) or
464 * it has too few free inodes left (min_inodes) or
465 * it has too few free blocks left (min_blocks) or
466 * Parent's group is preferred, if it doesn't satisfy these
467 * conditions we search cyclically through the rest. If none
468 * of the groups look good we just look for a group with more
469 * free inodes than average (starting at parent's group).
472 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
473 ext4_group_t
*group
, int mode
)
475 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
476 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
477 ext4_group_t real_ngroups
= ext4_get_groups_count(sb
);
478 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
479 unsigned int freei
, avefreei
;
480 ext4_fsblk_t freeb
, avefreeb
;
482 int max_dirs
, min_inodes
;
483 ext4_grpblk_t min_blocks
;
484 ext4_group_t i
, grp
, g
, ngroups
;
485 struct ext4_group_desc
*desc
;
486 struct orlov_stats stats
;
487 int flex_size
= ext4_flex_bg_size(sbi
);
489 ngroups
= real_ngroups
;
491 ngroups
= (real_ngroups
+ flex_size
- 1) >>
492 sbi
->s_log_groups_per_flex
;
493 parent_group
>>= sbi
->s_log_groups_per_flex
;
496 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
497 avefreei
= freei
/ ngroups
;
498 freeb
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
500 do_div(avefreeb
, ngroups
);
501 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
504 ((parent
== sb
->s_root
->d_inode
) ||
505 (EXT4_I(parent
)->i_flags
& EXT4_TOPDIR_FL
))) {
506 int best_ndir
= inodes_per_group
;
509 get_random_bytes(&grp
, sizeof(grp
));
510 parent_group
= (unsigned)grp
% ngroups
;
511 for (i
= 0; i
< ngroups
; i
++) {
512 g
= (parent_group
+ i
) % ngroups
;
513 get_orlov_stats(sb
, g
, flex_size
, &stats
);
514 if (!stats
.free_inodes
)
516 if (stats
.used_dirs
>= best_ndir
)
518 if (stats
.free_inodes
< avefreei
)
520 if (stats
.free_blocks
< avefreeb
)
524 best_ndir
= stats
.used_dirs
;
529 if (flex_size
== 1) {
535 * We pack inodes at the beginning of the flexgroup's
536 * inode tables. Block allocation decisions will do
537 * something similar, although regular files will
538 * start at 2nd block group of the flexgroup. See
539 * ext4_ext_find_goal() and ext4_find_near().
542 for (i
= 0; i
< flex_size
; i
++) {
543 if (grp
+i
>= real_ngroups
)
545 desc
= ext4_get_group_desc(sb
, grp
+i
, NULL
);
546 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
554 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
555 min_inodes
= avefreei
- inodes_per_group
*flex_size
/ 4;
558 min_blocks
= avefreeb
- EXT4_BLOCKS_PER_GROUP(sb
)*flex_size
/ 4;
561 * Start looking in the flex group where we last allocated an
562 * inode for this parent directory
564 if (EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
565 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
567 parent_group
>>= sbi
->s_log_groups_per_flex
;
570 for (i
= 0; i
< ngroups
; i
++) {
571 grp
= (parent_group
+ i
) % ngroups
;
572 get_orlov_stats(sb
, grp
, flex_size
, &stats
);
573 if (stats
.used_dirs
>= max_dirs
)
575 if (stats
.free_inodes
< min_inodes
)
577 if (stats
.free_blocks
< min_blocks
)
583 ngroups
= real_ngroups
;
584 avefreei
= freei
/ ngroups
;
586 parent_group
= EXT4_I(parent
)->i_block_group
;
587 for (i
= 0; i
< ngroups
; i
++) {
588 grp
= (parent_group
+ i
) % ngroups
;
589 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
590 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
591 ext4_free_inodes_count(sb
, desc
) >= avefreei
) {
599 * The free-inodes counter is approximate, and for really small
600 * filesystems the above test can fail to find any blockgroups
609 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
610 ext4_group_t
*group
, int mode
)
612 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
613 ext4_group_t i
, last
, ngroups
= ext4_get_groups_count(sb
);
614 struct ext4_group_desc
*desc
;
615 int flex_size
= ext4_flex_bg_size(EXT4_SB(sb
));
618 * Try to place the inode is the same flex group as its
619 * parent. If we can't find space, use the Orlov algorithm to
620 * find another flex group, and store that information in the
621 * parent directory's inode information so that use that flex
622 * group for future allocations.
628 parent_group
&= ~(flex_size
-1);
629 last
= parent_group
+ flex_size
;
632 for (i
= parent_group
; i
< last
; i
++) {
633 desc
= ext4_get_group_desc(sb
, i
, NULL
);
634 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
639 if (!retry
&& EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
641 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
645 * If this didn't work, use the Orlov search algorithm
646 * to find a new flex group; we pass in the mode to
647 * avoid the topdir algorithms.
649 *group
= parent_group
+ flex_size
;
650 if (*group
> ngroups
)
652 return find_group_orlov(sb
, parent
, group
, mode
);
656 * Try to place the inode in its parent directory
658 *group
= parent_group
;
659 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
660 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
661 ext4_free_blks_count(sb
, desc
))
665 * We're going to place this inode in a different blockgroup from its
666 * parent. We want to cause files in a common directory to all land in
667 * the same blockgroup. But we want files which are in a different
668 * directory which shares a blockgroup with our parent to land in a
669 * different blockgroup.
671 * So add our directory's i_ino into the starting point for the hash.
673 *group
= (*group
+ parent
->i_ino
) % ngroups
;
676 * Use a quadratic hash to find a group with a free inode and some free
679 for (i
= 1; i
< ngroups
; i
<<= 1) {
681 if (*group
>= ngroups
)
683 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
684 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
685 ext4_free_blks_count(sb
, desc
))
690 * That failed: try linear search for a free inode, even if that group
691 * has no free blocks.
693 *group
= parent_group
;
694 for (i
= 0; i
< ngroups
; i
++) {
695 if (++*group
>= ngroups
)
697 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
698 if (desc
&& ext4_free_inodes_count(sb
, desc
))
706 * claim the inode from the inode bitmap. If the group
707 * is uninit we need to take the groups's ext4_group_lock
708 * and clear the uninit flag. The inode bitmap update
709 * and group desc uninit flag clear should be done
710 * after holding ext4_group_lock so that ext4_read_inode_bitmap
711 * doesn't race with the ext4_claim_inode
713 static int ext4_claim_inode(struct super_block
*sb
,
714 struct buffer_head
*inode_bitmap_bh
,
715 unsigned long ino
, ext4_group_t group
, int mode
)
717 int free
= 0, retval
= 0, count
;
718 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
719 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, group
, NULL
);
721 ext4_lock_group(sb
, group
);
722 if (ext4_set_bit(ino
, inode_bitmap_bh
->b_data
)) {
723 /* not a free inode */
728 if ((group
== 0 && ino
< EXT4_FIRST_INO(sb
)) ||
729 ino
> EXT4_INODES_PER_GROUP(sb
)) {
730 ext4_unlock_group(sb
, group
);
731 ext4_error(sb
, __func__
,
732 "reserved inode or inode > inodes count - "
733 "block_group = %u, inode=%lu", group
,
734 ino
+ group
* EXT4_INODES_PER_GROUP(sb
));
737 /* If we didn't allocate from within the initialized part of the inode
738 * table then we need to initialize up to this inode. */
739 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
)) {
741 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
742 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
743 /* When marking the block group with
744 * ~EXT4_BG_INODE_UNINIT we don't want to depend
745 * on the value of bg_itable_unused even though
746 * mke2fs could have initialized the same for us.
747 * Instead we calculated the value below
752 free
= EXT4_INODES_PER_GROUP(sb
) -
753 ext4_itable_unused_count(sb
, gdp
);
757 * Check the relative inode number against the last used
758 * relative inode number in this group. if it is greater
759 * we need to update the bg_itable_unused count
763 ext4_itable_unused_set(sb
, gdp
,
764 (EXT4_INODES_PER_GROUP(sb
) - ino
));
766 count
= ext4_free_inodes_count(sb
, gdp
) - 1;
767 ext4_free_inodes_set(sb
, gdp
, count
);
769 count
= ext4_used_dirs_count(sb
, gdp
) + 1;
770 ext4_used_dirs_set(sb
, gdp
, count
);
771 if (sbi
->s_log_groups_per_flex
) {
772 ext4_group_t f
= ext4_flex_group(sbi
, group
);
774 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
777 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
, gdp
);
779 ext4_unlock_group(sb
, group
);
784 * There are two policies for allocating an inode. If the new inode is
785 * a directory, then a forward search is made for a block group with both
786 * free space and a low directory-to-inode ratio; if that fails, then of
787 * the groups with above-average free space, that group with the fewest
788 * directories already is chosen.
790 * For other inodes, search forward from the parent directory's block
791 * group to find a free inode.
793 struct inode
*ext4_new_inode(handle_t
*handle
, struct inode
*dir
, int mode
)
795 struct super_block
*sb
;
796 struct buffer_head
*inode_bitmap_bh
= NULL
;
797 struct buffer_head
*group_desc_bh
;
798 ext4_group_t ngroups
, group
= 0;
799 unsigned long ino
= 0;
801 struct ext4_group_desc
*gdp
= NULL
;
802 struct ext4_inode_info
*ei
;
803 struct ext4_sb_info
*sbi
;
809 ext4_group_t flex_group
;
811 /* Cannot create files in a deleted directory */
812 if (!dir
|| !dir
->i_nlink
)
813 return ERR_PTR(-EPERM
);
816 ngroups
= ext4_get_groups_count(sb
);
817 trace_ext4_request_inode(dir
, mode
);
818 inode
= new_inode(sb
);
820 return ERR_PTR(-ENOMEM
);
824 if (sbi
->s_log_groups_per_flex
&& test_opt(sb
, OLDALLOC
)) {
825 ret2
= find_group_flex(sb
, dir
, &group
);
827 ret2
= find_group_other(sb
, dir
, &group
, mode
);
828 if (ret2
== 0 && once
) {
830 printk(KERN_NOTICE
"ext4: find_group_flex "
831 "failed, fallback succeeded dir %lu\n",
839 if (test_opt(sb
, OLDALLOC
))
840 ret2
= find_group_dir(sb
, dir
, &group
);
842 ret2
= find_group_orlov(sb
, dir
, &group
, mode
);
844 ret2
= find_group_other(sb
, dir
, &group
, mode
);
847 EXT4_I(dir
)->i_last_alloc_group
= group
;
852 for (i
= 0; i
< ngroups
; i
++) {
855 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
859 brelse(inode_bitmap_bh
);
860 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
861 if (!inode_bitmap_bh
)
866 repeat_in_this_group
:
867 ino
= ext4_find_next_zero_bit((unsigned long *)
868 inode_bitmap_bh
->b_data
,
869 EXT4_INODES_PER_GROUP(sb
), ino
);
871 if (ino
< EXT4_INODES_PER_GROUP(sb
)) {
873 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
874 err
= ext4_journal_get_write_access(handle
,
879 BUFFER_TRACE(group_desc_bh
, "get_write_access");
880 err
= ext4_journal_get_write_access(handle
,
884 if (!ext4_claim_inode(sb
, inode_bitmap_bh
,
887 BUFFER_TRACE(inode_bitmap_bh
,
888 "call ext4_handle_dirty_metadata");
889 err
= ext4_handle_dirty_metadata(handle
,
894 /* zero bit is inode number 1*/
899 ext4_handle_release_buffer(handle
, inode_bitmap_bh
);
900 ext4_handle_release_buffer(handle
, group_desc_bh
);
902 if (++ino
< EXT4_INODES_PER_GROUP(sb
))
903 goto repeat_in_this_group
;
907 * This case is possible in concurrent environment. It is very
908 * rare. We cannot repeat the find_group_xxx() call because
909 * that will simply return the same blockgroup, because the
910 * group descriptor metadata has not yet been updated.
911 * So we just go onto the next blockgroup.
913 if (++group
== ngroups
)
920 /* We may have to initialize the block bitmap if it isn't already */
921 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_GDT_CSUM
) &&
922 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
923 struct buffer_head
*block_bitmap_bh
;
925 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
926 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
927 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
929 brelse(block_bitmap_bh
);
934 ext4_lock_group(sb
, group
);
935 /* recheck and clear flag under lock if we still need to */
936 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
937 free
= ext4_free_blocks_after_init(sb
, group
, gdp
);
938 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
939 ext4_free_blks_set(sb
, gdp
, free
);
940 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
,
943 ext4_unlock_group(sb
, group
);
945 /* Don't need to dirty bitmap block if we didn't change it */
947 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
948 err
= ext4_handle_dirty_metadata(handle
,
949 NULL
, block_bitmap_bh
);
952 brelse(block_bitmap_bh
);
956 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
957 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
961 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
963 percpu_counter_inc(&sbi
->s_dirs_counter
);
966 if (sbi
->s_log_groups_per_flex
) {
967 flex_group
= ext4_flex_group(sbi
, group
);
968 atomic_dec(&sbi
->s_flex_groups
[flex_group
].free_inodes
);
971 inode
->i_uid
= current_fsuid();
972 if (test_opt(sb
, GRPID
))
973 inode
->i_gid
= dir
->i_gid
;
974 else if (dir
->i_mode
& S_ISGID
) {
975 inode
->i_gid
= dir
->i_gid
;
979 inode
->i_gid
= current_fsgid();
980 inode
->i_mode
= mode
;
982 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
983 /* This is the optimal IO size (for stat), not the fs block size */
985 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
986 ext4_current_time(inode
);
988 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
989 ei
->i_dir_start_lookup
= 0;
993 * Don't inherit extent flag from directory, amongst others. We set
994 * extent flag on newly created directory and file only if -o extent
995 * mount option is specified
998 ext4_mask_flags(mode
, EXT4_I(dir
)->i_flags
& EXT4_FL_INHERITED
);
1001 ei
->i_block_group
= group
;
1002 ei
->i_last_alloc_group
= ~0;
1004 ext4_set_inode_flags(inode
);
1005 if (IS_DIRSYNC(inode
))
1006 ext4_handle_sync(handle
);
1007 if (insert_inode_locked(inode
) < 0) {
1011 spin_lock(&sbi
->s_next_gen_lock
);
1012 inode
->i_generation
= sbi
->s_next_generation
++;
1013 spin_unlock(&sbi
->s_next_gen_lock
);
1015 ei
->i_state
= EXT4_STATE_NEW
;
1017 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
1020 if (vfs_dq_alloc_inode(inode
)) {
1025 err
= ext4_init_acl(handle
, inode
, dir
);
1027 goto fail_free_drop
;
1029 err
= ext4_init_security(handle
, inode
, dir
);
1031 goto fail_free_drop
;
1033 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
1034 /* set extent flag only for directory, file and normal symlink*/
1035 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
1036 EXT4_I(inode
)->i_flags
|= EXT4_EXTENTS_FL
;
1037 ext4_ext_tree_init(handle
, inode
);
1041 err
= ext4_mark_inode_dirty(handle
, inode
);
1043 ext4_std_error(sb
, err
);
1044 goto fail_free_drop
;
1047 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
1048 trace_ext4_allocate_inode(inode
, dir
, mode
);
1051 ext4_std_error(sb
, err
);
1056 brelse(inode_bitmap_bh
);
1060 vfs_dq_free_inode(inode
);
1064 inode
->i_flags
|= S_NOQUOTA
;
1066 unlock_new_inode(inode
);
1068 brelse(inode_bitmap_bh
);
1069 return ERR_PTR(err
);
1072 /* Verify that we are loading a valid orphan from disk */
1073 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
1075 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
1076 ext4_group_t block_group
;
1078 struct buffer_head
*bitmap_bh
;
1079 struct inode
*inode
= NULL
;
1082 /* Error cases - e2fsck has already cleaned up for us */
1083 if (ino
> max_ino
) {
1084 ext4_warning(sb
, __func__
,
1085 "bad orphan ino %lu! e2fsck was run?", ino
);
1089 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
1090 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
1091 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
1093 ext4_warning(sb
, __func__
,
1094 "inode bitmap error for orphan %lu", ino
);
1098 /* Having the inode bit set should be a 100% indicator that this
1099 * is a valid orphan (no e2fsck run on fs). Orphans also include
1100 * inodes that were being truncated, so we can't check i_nlink==0.
1102 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
1105 inode
= ext4_iget(sb
, ino
);
1110 * If the orphans has i_nlinks > 0 then it should be able to be
1111 * truncated, otherwise it won't be removed from the orphan list
1112 * during processing and an infinite loop will result.
1114 if (inode
->i_nlink
&& !ext4_can_truncate(inode
))
1117 if (NEXT_ORPHAN(inode
) > max_ino
)
1123 err
= PTR_ERR(inode
);
1126 ext4_warning(sb
, __func__
,
1127 "bad orphan inode %lu! e2fsck was run?", ino
);
1128 printk(KERN_NOTICE
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
1129 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
1130 ext4_test_bit(bit
, bitmap_bh
->b_data
));
1131 printk(KERN_NOTICE
"inode=%p\n", inode
);
1133 printk(KERN_NOTICE
"is_bad_inode(inode)=%d\n",
1134 is_bad_inode(inode
));
1135 printk(KERN_NOTICE
"NEXT_ORPHAN(inode)=%u\n",
1136 NEXT_ORPHAN(inode
));
1137 printk(KERN_NOTICE
"max_ino=%lu\n", max_ino
);
1138 printk(KERN_NOTICE
"i_nlink=%u\n", inode
->i_nlink
);
1139 /* Avoid freeing blocks if we got a bad deleted inode */
1140 if (inode
->i_nlink
== 0)
1141 inode
->i_blocks
= 0;
1146 return ERR_PTR(err
);
1149 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
1151 unsigned long desc_count
;
1152 struct ext4_group_desc
*gdp
;
1153 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1155 struct ext4_super_block
*es
;
1156 unsigned long bitmap_count
, x
;
1157 struct buffer_head
*bitmap_bh
= NULL
;
1159 es
= EXT4_SB(sb
)->s_es
;
1163 for (i
= 0; i
< ngroups
; i
++) {
1164 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1167 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1169 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1173 x
= ext4_count_free(bitmap_bh
, EXT4_INODES_PER_GROUP(sb
) / 8);
1174 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1175 i
, ext4_free_inodes_count(sb
, gdp
), x
);
1179 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1180 "stored = %u, computed = %lu, %lu\n",
1181 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1185 for (i
= 0; i
< ngroups
; i
++) {
1186 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1189 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1196 /* Called at mount-time, super-block is locked */
1197 unsigned long ext4_count_dirs(struct super_block
* sb
)
1199 unsigned long count
= 0;
1200 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1202 for (i
= 0; i
< ngroups
; i
++) {
1203 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1206 count
+= ext4_used_dirs_count(sb
, gdp
);