1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/ialloc.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@redhat.com), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
16 #include <linux/time.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 <linux/cred.h>
27 #include <asm/byteorder.h>
30 #include "ext4_jbd2.h"
34 #include <trace/events/ext4.h>
37 * ialloc.c contains the inodes allocation and deallocation routines
41 * The free inodes are managed by bitmaps. A file system contains several
42 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
43 * block for inodes, N blocks for the inode table and data blocks.
45 * The file system contains group descriptors which are located after the
46 * super block. Each descriptor contains the number of the bitmap block and
47 * the free blocks count in the block.
51 * To avoid calling the atomic setbit hundreds or thousands of times, we only
52 * need to use it within a single byte (to ensure we get endianness right).
53 * We can use memset for the rest of the bitmap as there are no other users.
55 void ext4_mark_bitmap_end(int start_bit
, int end_bit
, char *bitmap
)
59 if (start_bit
>= end_bit
)
62 ext4_debug("mark end bits +%d through +%d used\n", start_bit
, end_bit
);
63 for (i
= start_bit
; i
< ((start_bit
+ 7) & ~7UL); i
++)
64 ext4_set_bit(i
, bitmap
);
66 memset(bitmap
+ (i
>> 3), 0xff, (end_bit
- i
) >> 3);
69 void ext4_end_bitmap_read(struct buffer_head
*bh
, int uptodate
)
72 set_buffer_uptodate(bh
);
73 set_bitmap_uptodate(bh
);
79 static int ext4_validate_inode_bitmap(struct super_block
*sb
,
80 struct ext4_group_desc
*desc
,
81 ext4_group_t block_group
,
82 struct buffer_head
*bh
)
85 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, block_group
);
86 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
88 if (buffer_verified(bh
))
90 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))
93 ext4_lock_group(sb
, block_group
);
94 if (buffer_verified(bh
))
96 blk
= ext4_inode_bitmap(sb
, desc
);
97 if (!ext4_inode_bitmap_csum_verify(sb
, block_group
, desc
, bh
,
98 EXT4_INODES_PER_GROUP(sb
) / 8)) {
99 ext4_unlock_group(sb
, block_group
);
100 ext4_error(sb
, "Corrupt inode bitmap - block_group = %u, "
101 "inode_bitmap = %llu", block_group
, blk
);
102 grp
= ext4_get_group_info(sb
, block_group
);
103 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
105 count
= ext4_free_inodes_count(sb
, desc
);
106 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
109 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
112 set_buffer_verified(bh
);
114 ext4_unlock_group(sb
, block_group
);
119 * Read the inode allocation bitmap for a given block_group, reading
120 * into the specified slot in the superblock's bitmap cache.
122 * Return buffer_head of bitmap on success or NULL.
124 static struct buffer_head
*
125 ext4_read_inode_bitmap(struct super_block
*sb
, ext4_group_t block_group
)
127 struct ext4_group_desc
*desc
;
128 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
129 struct buffer_head
*bh
= NULL
;
130 ext4_fsblk_t bitmap_blk
;
133 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
135 return ERR_PTR(-EFSCORRUPTED
);
137 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
138 if ((bitmap_blk
<= le32_to_cpu(sbi
->s_es
->s_first_data_block
)) ||
139 (bitmap_blk
>= ext4_blocks_count(sbi
->s_es
))) {
140 ext4_error(sb
, "Invalid inode bitmap blk %llu in "
141 "block_group %u", bitmap_blk
, block_group
);
142 return ERR_PTR(-EFSCORRUPTED
);
144 bh
= sb_getblk(sb
, bitmap_blk
);
146 ext4_warning(sb
, "Cannot read inode bitmap - "
147 "block_group = %u, inode_bitmap = %llu",
148 block_group
, bitmap_blk
);
149 return ERR_PTR(-EIO
);
151 if (bitmap_uptodate(bh
))
155 if (bitmap_uptodate(bh
)) {
160 ext4_lock_group(sb
, block_group
);
161 if (ext4_has_group_desc_csum(sb
) &&
162 (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
))) {
163 if (block_group
== 0) {
164 ext4_unlock_group(sb
, block_group
);
166 ext4_error(sb
, "Inode bitmap for bg 0 marked "
171 memset(bh
->b_data
, 0, (EXT4_INODES_PER_GROUP(sb
) + 7) / 8);
172 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
173 sb
->s_blocksize
* 8, bh
->b_data
);
174 set_bitmap_uptodate(bh
);
175 set_buffer_uptodate(bh
);
176 set_buffer_verified(bh
);
177 ext4_unlock_group(sb
, block_group
);
181 ext4_unlock_group(sb
, block_group
);
183 if (buffer_uptodate(bh
)) {
185 * if not uninit if bh is uptodate,
186 * bitmap is also uptodate
188 set_bitmap_uptodate(bh
);
193 * submit the buffer_head for reading
195 trace_ext4_load_inode_bitmap(sb
, block_group
);
196 bh
->b_end_io
= ext4_end_bitmap_read
;
198 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
200 if (!buffer_uptodate(bh
)) {
202 ext4_error(sb
, "Cannot read inode bitmap - "
203 "block_group = %u, inode_bitmap = %llu",
204 block_group
, bitmap_blk
);
205 return ERR_PTR(-EIO
);
209 err
= ext4_validate_inode_bitmap(sb
, desc
, block_group
, bh
);
219 * NOTE! When we get the inode, we're the only people
220 * that have access to it, and as such there are no
221 * race conditions we have to worry about. The inode
222 * is not on the hash-lists, and it cannot be reached
223 * through the filesystem because the directory entry
224 * has been deleted earlier.
226 * HOWEVER: we must make sure that we get no aliases,
227 * which means that we have to call "clear_inode()"
228 * _before_ we mark the inode not in use in the inode
229 * bitmaps. Otherwise a newly created file might use
230 * the same inode number (not actually the same pointer
231 * though), and then we'd have two inodes sharing the
232 * same inode number and space on the harddisk.
234 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
236 struct super_block
*sb
= inode
->i_sb
;
239 struct buffer_head
*bitmap_bh
= NULL
;
240 struct buffer_head
*bh2
;
241 ext4_group_t block_group
;
243 struct ext4_group_desc
*gdp
;
244 struct ext4_super_block
*es
;
245 struct ext4_sb_info
*sbi
;
246 int fatal
= 0, err
, count
, cleared
;
247 struct ext4_group_info
*grp
;
250 printk(KERN_ERR
"EXT4-fs: %s:%d: inode on "
251 "nonexistent device\n", __func__
, __LINE__
);
254 if (atomic_read(&inode
->i_count
) > 1) {
255 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: count=%d",
256 __func__
, __LINE__
, inode
->i_ino
,
257 atomic_read(&inode
->i_count
));
260 if (inode
->i_nlink
) {
261 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: nlink=%d\n",
262 __func__
, __LINE__
, inode
->i_ino
, inode
->i_nlink
);
268 ext4_debug("freeing inode %lu\n", ino
);
269 trace_ext4_free_inode(inode
);
272 * Note: we must free any quota before locking the superblock,
273 * as writing the quota to disk may need the lock as well.
275 dquot_initialize(inode
);
276 dquot_free_inode(inode
);
279 is_directory
= S_ISDIR(inode
->i_mode
);
281 /* Do this BEFORE marking the inode not in use or returning an error */
282 ext4_clear_inode(inode
);
284 es
= EXT4_SB(sb
)->s_es
;
285 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
286 ext4_error(sb
, "reserved or nonexistent inode %lu", ino
);
289 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
290 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
291 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
292 /* Don't bother if the inode bitmap is corrupt. */
293 grp
= ext4_get_group_info(sb
, block_group
);
294 if (IS_ERR(bitmap_bh
)) {
295 fatal
= PTR_ERR(bitmap_bh
);
299 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))) {
300 fatal
= -EFSCORRUPTED
;
304 BUFFER_TRACE(bitmap_bh
, "get_write_access");
305 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
310 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
312 BUFFER_TRACE(bh2
, "get_write_access");
313 fatal
= ext4_journal_get_write_access(handle
, bh2
);
315 ext4_lock_group(sb
, block_group
);
316 cleared
= ext4_test_and_clear_bit(bit
, bitmap_bh
->b_data
);
317 if (fatal
|| !cleared
) {
318 ext4_unlock_group(sb
, block_group
);
322 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
323 ext4_free_inodes_set(sb
, gdp
, count
);
325 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
326 ext4_used_dirs_set(sb
, gdp
, count
);
327 percpu_counter_dec(&sbi
->s_dirs_counter
);
329 ext4_inode_bitmap_csum_set(sb
, block_group
, gdp
, bitmap_bh
,
330 EXT4_INODES_PER_GROUP(sb
) / 8);
331 ext4_group_desc_csum_set(sb
, block_group
, gdp
);
332 ext4_unlock_group(sb
, block_group
);
334 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
335 if (sbi
->s_log_groups_per_flex
) {
336 ext4_group_t f
= ext4_flex_group(sbi
, block_group
);
338 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
340 atomic_dec(&sbi
->s_flex_groups
[f
].used_dirs
);
342 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
343 fatal
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
346 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
347 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
351 ext4_error(sb
, "bit already cleared for inode %lu", ino
);
352 if (gdp
&& !EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
354 count
= ext4_free_inodes_count(sb
, gdp
);
355 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
358 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
363 ext4_std_error(sb
, fatal
);
373 * Helper function for Orlov's allocator; returns critical information
374 * for a particular block group or flex_bg. If flex_size is 1, then g
375 * is a block group number; otherwise it is flex_bg number.
377 static void get_orlov_stats(struct super_block
*sb
, ext4_group_t g
,
378 int flex_size
, struct orlov_stats
*stats
)
380 struct ext4_group_desc
*desc
;
381 struct flex_groups
*flex_group
= EXT4_SB(sb
)->s_flex_groups
;
384 stats
->free_inodes
= atomic_read(&flex_group
[g
].free_inodes
);
385 stats
->free_clusters
= atomic64_read(&flex_group
[g
].free_clusters
);
386 stats
->used_dirs
= atomic_read(&flex_group
[g
].used_dirs
);
390 desc
= ext4_get_group_desc(sb
, g
, NULL
);
392 stats
->free_inodes
= ext4_free_inodes_count(sb
, desc
);
393 stats
->free_clusters
= ext4_free_group_clusters(sb
, desc
);
394 stats
->used_dirs
= ext4_used_dirs_count(sb
, desc
);
396 stats
->free_inodes
= 0;
397 stats
->free_clusters
= 0;
398 stats
->used_dirs
= 0;
403 * Orlov's allocator for directories.
405 * We always try to spread first-level directories.
407 * If there are blockgroups with both free inodes and free blocks counts
408 * not worse than average we return one with smallest directory count.
409 * Otherwise we simply return a random group.
411 * For the rest rules look so:
413 * It's OK to put directory into a group unless
414 * it has too many directories already (max_dirs) or
415 * it has too few free inodes left (min_inodes) or
416 * it has too few free blocks left (min_blocks) or
417 * Parent's group is preferred, if it doesn't satisfy these
418 * conditions we search cyclically through the rest. If none
419 * of the groups look good we just look for a group with more
420 * free inodes than average (starting at parent's group).
423 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
424 ext4_group_t
*group
, umode_t mode
,
425 const struct qstr
*qstr
)
427 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
428 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
429 ext4_group_t real_ngroups
= ext4_get_groups_count(sb
);
430 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
431 unsigned int freei
, avefreei
, grp_free
;
432 ext4_fsblk_t freeb
, avefreec
;
434 int max_dirs
, min_inodes
;
435 ext4_grpblk_t min_clusters
;
436 ext4_group_t i
, grp
, g
, ngroups
;
437 struct ext4_group_desc
*desc
;
438 struct orlov_stats stats
;
439 int flex_size
= ext4_flex_bg_size(sbi
);
440 struct dx_hash_info hinfo
;
442 ngroups
= real_ngroups
;
444 ngroups
= (real_ngroups
+ flex_size
- 1) >>
445 sbi
->s_log_groups_per_flex
;
446 parent_group
>>= sbi
->s_log_groups_per_flex
;
449 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
450 avefreei
= freei
/ ngroups
;
451 freeb
= EXT4_C2B(sbi
,
452 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
454 do_div(avefreec
, ngroups
);
455 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
458 ((parent
== d_inode(sb
->s_root
)) ||
459 (ext4_test_inode_flag(parent
, EXT4_INODE_TOPDIR
)))) {
460 int best_ndir
= inodes_per_group
;
464 hinfo
.hash_version
= DX_HASH_HALF_MD4
;
465 hinfo
.seed
= sbi
->s_hash_seed
;
466 ext4fs_dirhash(qstr
->name
, qstr
->len
, &hinfo
);
470 parent_group
= (unsigned)grp
% ngroups
;
471 for (i
= 0; i
< ngroups
; i
++) {
472 g
= (parent_group
+ i
) % ngroups
;
473 get_orlov_stats(sb
, g
, flex_size
, &stats
);
474 if (!stats
.free_inodes
)
476 if (stats
.used_dirs
>= best_ndir
)
478 if (stats
.free_inodes
< avefreei
)
480 if (stats
.free_clusters
< avefreec
)
484 best_ndir
= stats
.used_dirs
;
489 if (flex_size
== 1) {
495 * We pack inodes at the beginning of the flexgroup's
496 * inode tables. Block allocation decisions will do
497 * something similar, although regular files will
498 * start at 2nd block group of the flexgroup. See
499 * ext4_ext_find_goal() and ext4_find_near().
502 for (i
= 0; i
< flex_size
; i
++) {
503 if (grp
+i
>= real_ngroups
)
505 desc
= ext4_get_group_desc(sb
, grp
+i
, NULL
);
506 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
514 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
515 min_inodes
= avefreei
- inodes_per_group
*flex_size
/ 4;
518 min_clusters
= avefreec
- EXT4_CLUSTERS_PER_GROUP(sb
)*flex_size
/ 4;
521 * Start looking in the flex group where we last allocated an
522 * inode for this parent directory
524 if (EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
525 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
527 parent_group
>>= sbi
->s_log_groups_per_flex
;
530 for (i
= 0; i
< ngroups
; i
++) {
531 grp
= (parent_group
+ i
) % ngroups
;
532 get_orlov_stats(sb
, grp
, flex_size
, &stats
);
533 if (stats
.used_dirs
>= max_dirs
)
535 if (stats
.free_inodes
< min_inodes
)
537 if (stats
.free_clusters
< min_clusters
)
543 ngroups
= real_ngroups
;
544 avefreei
= freei
/ ngroups
;
546 parent_group
= EXT4_I(parent
)->i_block_group
;
547 for (i
= 0; i
< ngroups
; i
++) {
548 grp
= (parent_group
+ i
) % ngroups
;
549 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
551 grp_free
= ext4_free_inodes_count(sb
, desc
);
552 if (grp_free
&& grp_free
>= avefreei
) {
561 * The free-inodes counter is approximate, and for really small
562 * filesystems the above test can fail to find any blockgroups
571 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
572 ext4_group_t
*group
, umode_t mode
)
574 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
575 ext4_group_t i
, last
, ngroups
= ext4_get_groups_count(sb
);
576 struct ext4_group_desc
*desc
;
577 int flex_size
= ext4_flex_bg_size(EXT4_SB(sb
));
580 * Try to place the inode is the same flex group as its
581 * parent. If we can't find space, use the Orlov algorithm to
582 * find another flex group, and store that information in the
583 * parent directory's inode information so that use that flex
584 * group for future allocations.
590 parent_group
&= ~(flex_size
-1);
591 last
= parent_group
+ flex_size
;
594 for (i
= parent_group
; i
< last
; i
++) {
595 desc
= ext4_get_group_desc(sb
, i
, NULL
);
596 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
601 if (!retry
&& EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
603 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
607 * If this didn't work, use the Orlov search algorithm
608 * to find a new flex group; we pass in the mode to
609 * avoid the topdir algorithms.
611 *group
= parent_group
+ flex_size
;
612 if (*group
> ngroups
)
614 return find_group_orlov(sb
, parent
, group
, mode
, NULL
);
618 * Try to place the inode in its parent directory
620 *group
= parent_group
;
621 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
622 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
623 ext4_free_group_clusters(sb
, desc
))
627 * We're going to place this inode in a different blockgroup from its
628 * parent. We want to cause files in a common directory to all land in
629 * the same blockgroup. But we want files which are in a different
630 * directory which shares a blockgroup with our parent to land in a
631 * different blockgroup.
633 * So add our directory's i_ino into the starting point for the hash.
635 *group
= (*group
+ parent
->i_ino
) % ngroups
;
638 * Use a quadratic hash to find a group with a free inode and some free
641 for (i
= 1; i
< ngroups
; i
<<= 1) {
643 if (*group
>= ngroups
)
645 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
646 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
647 ext4_free_group_clusters(sb
, desc
))
652 * That failed: try linear search for a free inode, even if that group
653 * has no free blocks.
655 *group
= parent_group
;
656 for (i
= 0; i
< ngroups
; i
++) {
657 if (++*group
>= ngroups
)
659 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
660 if (desc
&& ext4_free_inodes_count(sb
, desc
))
668 * In no journal mode, if an inode has recently been deleted, we want
669 * to avoid reusing it until we're reasonably sure the inode table
670 * block has been written back to disk. (Yes, these values are
671 * somewhat arbitrary...)
673 #define RECENTCY_MIN 5
674 #define RECENTCY_DIRTY 300
676 static int recently_deleted(struct super_block
*sb
, ext4_group_t group
, int ino
)
678 struct ext4_group_desc
*gdp
;
679 struct ext4_inode
*raw_inode
;
680 struct buffer_head
*bh
;
681 int inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
683 int recentcy
= RECENTCY_MIN
;
686 gdp
= ext4_get_group_desc(sb
, group
, NULL
);
690 bh
= sb_find_get_block(sb
, ext4_inode_table(sb
, gdp
) +
691 (ino
/ inodes_per_block
));
692 if (!bh
|| !buffer_uptodate(bh
))
694 * If the block is not in the buffer cache, then it
695 * must have been written out.
699 offset
= (ino
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
700 raw_inode
= (struct ext4_inode
*) (bh
->b_data
+ offset
);
702 /* i_dtime is only 32 bits on disk, but we only care about relative
703 * times in the range of a few minutes (i.e. long enough to sync a
704 * recently-deleted inode to disk), so using the low 32 bits of the
705 * clock (a 68 year range) is enough, see time_before32() */
706 dtime
= le32_to_cpu(raw_inode
->i_dtime
);
707 now
= ktime_get_real_seconds();
708 if (buffer_dirty(bh
))
709 recentcy
+= RECENTCY_DIRTY
;
711 if (dtime
&& time_before32(dtime
, now
) &&
712 time_before32(now
, dtime
+ recentcy
))
719 static int find_inode_bit(struct super_block
*sb
, ext4_group_t group
,
720 struct buffer_head
*bitmap
, unsigned long *ino
)
723 *ino
= ext4_find_next_zero_bit((unsigned long *)
725 EXT4_INODES_PER_GROUP(sb
), *ino
);
726 if (*ino
>= EXT4_INODES_PER_GROUP(sb
))
729 if ((EXT4_SB(sb
)->s_journal
== NULL
) &&
730 recently_deleted(sb
, group
, *ino
)) {
732 if (*ino
< EXT4_INODES_PER_GROUP(sb
))
741 * There are two policies for allocating an inode. If the new inode is
742 * a directory, then a forward search is made for a block group with both
743 * free space and a low directory-to-inode ratio; if that fails, then of
744 * the groups with above-average free space, that group with the fewest
745 * directories already is chosen.
747 * For other inodes, search forward from the parent directory's block
748 * group to find a free inode.
750 struct inode
*__ext4_new_inode(handle_t
*handle
, struct inode
*dir
,
751 umode_t mode
, const struct qstr
*qstr
,
752 __u32 goal
, uid_t
*owner
, __u32 i_flags
,
753 int handle_type
, unsigned int line_no
,
756 struct super_block
*sb
;
757 struct buffer_head
*inode_bitmap_bh
= NULL
;
758 struct buffer_head
*group_desc_bh
;
759 ext4_group_t ngroups
, group
= 0;
760 unsigned long ino
= 0;
762 struct ext4_group_desc
*gdp
= NULL
;
763 struct ext4_inode_info
*ei
;
764 struct ext4_sb_info
*sbi
;
768 ext4_group_t flex_group
;
769 struct ext4_group_info
*grp
;
772 /* Cannot create files in a deleted directory */
773 if (!dir
|| !dir
->i_nlink
)
774 return ERR_PTR(-EPERM
);
779 if (unlikely(ext4_forced_shutdown(sbi
)))
780 return ERR_PTR(-EIO
);
782 /* Supplied owner must be valid */
783 if (owner
&& (owner
[0] == (uid_t
)-1 || owner
[1] == (uid_t
)-1))
784 return ERR_PTR(-EOVERFLOW
);
786 if ((ext4_encrypted_inode(dir
) || DUMMY_ENCRYPTION_ENABLED(sbi
)) &&
787 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) &&
788 !(i_flags
& EXT4_EA_INODE_FL
)) {
789 err
= fscrypt_get_encryption_info(dir
);
792 if (!fscrypt_has_encryption_key(dir
))
793 return ERR_PTR(-ENOKEY
);
797 if (!handle
&& sbi
->s_journal
&& !(i_flags
& EXT4_EA_INODE_FL
)) {
798 #ifdef CONFIG_EXT4_FS_POSIX_ACL
799 struct posix_acl
*p
= get_acl(dir
, ACL_TYPE_DEFAULT
);
804 int acl_size
= p
->a_count
* sizeof(ext4_acl_entry
);
806 nblocks
+= (S_ISDIR(mode
) ? 2 : 1) *
807 __ext4_xattr_set_credits(sb
, NULL
/* inode */,
808 NULL
/* block_bh */, acl_size
,
809 true /* is_create */);
810 posix_acl_release(p
);
814 #ifdef CONFIG_SECURITY
816 int num_security_xattrs
= 1;
818 #ifdef CONFIG_INTEGRITY
819 num_security_xattrs
++;
822 * We assume that security xattrs are never
823 * more than 1k. In practice they are under
826 nblocks
+= num_security_xattrs
*
827 __ext4_xattr_set_credits(sb
, NULL
/* inode */,
828 NULL
/* block_bh */, 1024,
829 true /* is_create */);
833 nblocks
+= __ext4_xattr_set_credits(sb
,
834 NULL
/* inode */, NULL
/* block_bh */,
835 FSCRYPT_SET_CONTEXT_MAX_SIZE
,
836 true /* is_create */);
839 ngroups
= ext4_get_groups_count(sb
);
840 trace_ext4_request_inode(dir
, mode
);
841 inode
= new_inode(sb
);
843 return ERR_PTR(-ENOMEM
);
847 * Initialize owners and quota early so that we don't have to account
848 * for quota initialization worst case in standard inode creating
852 inode
->i_mode
= mode
;
853 i_uid_write(inode
, owner
[0]);
854 i_gid_write(inode
, owner
[1]);
855 } else if (test_opt(sb
, GRPID
)) {
856 inode
->i_mode
= mode
;
857 inode
->i_uid
= current_fsuid();
858 inode
->i_gid
= dir
->i_gid
;
860 inode_init_owner(inode
, dir
, mode
);
862 if (ext4_has_feature_project(sb
) &&
863 ext4_test_inode_flag(dir
, EXT4_INODE_PROJINHERIT
))
864 ei
->i_projid
= EXT4_I(dir
)->i_projid
;
866 ei
->i_projid
= make_kprojid(sb
->s_user_ns
, EXT4_DEF_PROJID
);
868 err
= dquot_initialize(inode
);
873 goal
= sbi
->s_inode_goal
;
875 if (goal
&& goal
<= le32_to_cpu(sbi
->s_es
->s_inodes_count
)) {
876 group
= (goal
- 1) / EXT4_INODES_PER_GROUP(sb
);
877 ino
= (goal
- 1) % EXT4_INODES_PER_GROUP(sb
);
883 ret2
= find_group_orlov(sb
, dir
, &group
, mode
, qstr
);
885 ret2
= find_group_other(sb
, dir
, &group
, mode
);
888 EXT4_I(dir
)->i_last_alloc_group
= group
;
894 * Normally we will only go through one pass of this loop,
895 * unless we get unlucky and it turns out the group we selected
896 * had its last inode grabbed by someone else.
898 for (i
= 0; i
< ngroups
; i
++, ino
= 0) {
901 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
906 * Check free inodes count before loading bitmap.
908 if (ext4_free_inodes_count(sb
, gdp
) == 0)
911 grp
= ext4_get_group_info(sb
, group
);
912 /* Skip groups with already-known suspicious inode tables */
913 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))
916 brelse(inode_bitmap_bh
);
917 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
918 /* Skip groups with suspicious inode tables */
919 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
) ||
920 IS_ERR(inode_bitmap_bh
)) {
921 inode_bitmap_bh
= NULL
;
925 repeat_in_this_group
:
926 ret2
= find_inode_bit(sb
, group
, inode_bitmap_bh
, &ino
);
930 if (group
== 0 && (ino
+ 1) < EXT4_FIRST_INO(sb
)) {
931 ext4_error(sb
, "reserved inode found cleared - "
932 "inode=%lu", ino
+ 1);
937 BUG_ON(nblocks
<= 0);
938 handle
= __ext4_journal_start_sb(dir
->i_sb
, line_no
,
939 handle_type
, nblocks
,
941 if (IS_ERR(handle
)) {
942 err
= PTR_ERR(handle
);
943 ext4_std_error(sb
, err
);
947 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
948 err
= ext4_journal_get_write_access(handle
, inode_bitmap_bh
);
950 ext4_std_error(sb
, err
);
953 ext4_lock_group(sb
, group
);
954 ret2
= ext4_test_and_set_bit(ino
, inode_bitmap_bh
->b_data
);
956 /* Someone already took the bit. Repeat the search
959 ret2
= find_inode_bit(sb
, group
, inode_bitmap_bh
, &ino
);
961 ext4_set_bit(ino
, inode_bitmap_bh
->b_data
);
964 ret2
= 1; /* we didn't grab the inode */
967 ext4_unlock_group(sb
, group
);
968 ino
++; /* the inode bitmap is zero-based */
970 goto got
; /* we grabbed the inode! */
972 if (ino
< EXT4_INODES_PER_GROUP(sb
))
973 goto repeat_in_this_group
;
975 if (++group
== ngroups
)
982 BUFFER_TRACE(inode_bitmap_bh
, "call ext4_handle_dirty_metadata");
983 err
= ext4_handle_dirty_metadata(handle
, NULL
, inode_bitmap_bh
);
985 ext4_std_error(sb
, err
);
989 BUFFER_TRACE(group_desc_bh
, "get_write_access");
990 err
= ext4_journal_get_write_access(handle
, group_desc_bh
);
992 ext4_std_error(sb
, err
);
996 /* We may have to initialize the block bitmap if it isn't already */
997 if (ext4_has_group_desc_csum(sb
) &&
998 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
999 struct buffer_head
*block_bitmap_bh
;
1001 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
1002 if (IS_ERR(block_bitmap_bh
)) {
1003 err
= PTR_ERR(block_bitmap_bh
);
1006 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
1007 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
1009 brelse(block_bitmap_bh
);
1010 ext4_std_error(sb
, err
);
1014 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
1015 err
= ext4_handle_dirty_metadata(handle
, NULL
, block_bitmap_bh
);
1017 /* recheck and clear flag under lock if we still need to */
1018 ext4_lock_group(sb
, group
);
1019 if (ext4_has_group_desc_csum(sb
) &&
1020 (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
))) {
1021 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
1022 ext4_free_group_clusters_set(sb
, gdp
,
1023 ext4_free_clusters_after_init(sb
, group
, gdp
));
1024 ext4_block_bitmap_csum_set(sb
, group
, gdp
,
1026 ext4_group_desc_csum_set(sb
, group
, gdp
);
1028 ext4_unlock_group(sb
, group
);
1029 brelse(block_bitmap_bh
);
1032 ext4_std_error(sb
, err
);
1037 /* Update the relevant bg descriptor fields */
1038 if (ext4_has_group_desc_csum(sb
)) {
1040 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
1042 down_read(&grp
->alloc_sem
); /* protect vs itable lazyinit */
1043 ext4_lock_group(sb
, group
); /* while we modify the bg desc */
1044 free
= EXT4_INODES_PER_GROUP(sb
) -
1045 ext4_itable_unused_count(sb
, gdp
);
1046 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
1047 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
1051 * Check the relative inode number against the last used
1052 * relative inode number in this group. if it is greater
1053 * we need to update the bg_itable_unused count
1056 ext4_itable_unused_set(sb
, gdp
,
1057 (EXT4_INODES_PER_GROUP(sb
) - ino
));
1058 up_read(&grp
->alloc_sem
);
1060 ext4_lock_group(sb
, group
);
1063 ext4_free_inodes_set(sb
, gdp
, ext4_free_inodes_count(sb
, gdp
) - 1);
1064 if (S_ISDIR(mode
)) {
1065 ext4_used_dirs_set(sb
, gdp
, ext4_used_dirs_count(sb
, gdp
) + 1);
1066 if (sbi
->s_log_groups_per_flex
) {
1067 ext4_group_t f
= ext4_flex_group(sbi
, group
);
1069 atomic_inc(&sbi
->s_flex_groups
[f
].used_dirs
);
1072 if (ext4_has_group_desc_csum(sb
)) {
1073 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, inode_bitmap_bh
,
1074 EXT4_INODES_PER_GROUP(sb
) / 8);
1075 ext4_group_desc_csum_set(sb
, group
, gdp
);
1077 ext4_unlock_group(sb
, group
);
1079 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
1080 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
1082 ext4_std_error(sb
, err
);
1086 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
1088 percpu_counter_inc(&sbi
->s_dirs_counter
);
1090 if (sbi
->s_log_groups_per_flex
) {
1091 flex_group
= ext4_flex_group(sbi
, group
);
1092 atomic_dec(&sbi
->s_flex_groups
[flex_group
].free_inodes
);
1095 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
1096 /* This is the optimal IO size (for stat), not the fs block size */
1097 inode
->i_blocks
= 0;
1098 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
1099 current_time(inode
);
1101 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
1102 ei
->i_dir_start_lookup
= 0;
1105 /* Don't inherit extent flag from directory, amongst others. */
1107 ext4_mask_flags(mode
, EXT4_I(dir
)->i_flags
& EXT4_FL_INHERITED
);
1108 ei
->i_flags
|= i_flags
;
1111 ei
->i_block_group
= group
;
1112 ei
->i_last_alloc_group
= ~0;
1114 ext4_set_inode_flags(inode
);
1115 if (IS_DIRSYNC(inode
))
1116 ext4_handle_sync(handle
);
1117 if (insert_inode_locked(inode
) < 0) {
1119 * Likely a bitmap corruption causing inode to be allocated
1123 ext4_error(sb
, "failed to insert inode %lu: doubly allocated?",
1127 inode
->i_generation
= prandom_u32();
1129 /* Precompute checksum seed for inode metadata */
1130 if (ext4_has_metadata_csum(sb
)) {
1132 __le32 inum
= cpu_to_le32(inode
->i_ino
);
1133 __le32 gen
= cpu_to_le32(inode
->i_generation
);
1134 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
1136 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
1140 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
1141 ext4_set_inode_state(inode
, EXT4_STATE_NEW
);
1143 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
1144 ei
->i_inline_off
= 0;
1145 if (ext4_has_feature_inline_data(sb
))
1146 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1148 err
= dquot_alloc_inode(inode
);
1153 * Since the encryption xattr will always be unique, create it first so
1154 * that it's less likely to end up in an external xattr block and
1155 * prevent its deduplication.
1158 err
= fscrypt_inherit_context(dir
, inode
, handle
, true);
1160 goto fail_free_drop
;
1163 if (!(ei
->i_flags
& EXT4_EA_INODE_FL
)) {
1164 err
= ext4_init_acl(handle
, inode
, dir
);
1166 goto fail_free_drop
;
1168 err
= ext4_init_security(handle
, inode
, dir
, qstr
);
1170 goto fail_free_drop
;
1173 if (ext4_has_feature_extents(sb
)) {
1174 /* set extent flag only for directory, file and normal symlink*/
1175 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
1176 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
1177 ext4_ext_tree_init(handle
, inode
);
1181 if (ext4_handle_valid(handle
)) {
1182 ei
->i_sync_tid
= handle
->h_transaction
->t_tid
;
1183 ei
->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1186 err
= ext4_mark_inode_dirty(handle
, inode
);
1188 ext4_std_error(sb
, err
);
1189 goto fail_free_drop
;
1192 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
1193 trace_ext4_allocate_inode(inode
, dir
, mode
);
1194 brelse(inode_bitmap_bh
);
1198 dquot_free_inode(inode
);
1201 unlock_new_inode(inode
);
1204 inode
->i_flags
|= S_NOQUOTA
;
1206 brelse(inode_bitmap_bh
);
1207 return ERR_PTR(err
);
1210 /* Verify that we are loading a valid orphan from disk */
1211 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
1213 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
1214 ext4_group_t block_group
;
1216 struct buffer_head
*bitmap_bh
= NULL
;
1217 struct inode
*inode
= NULL
;
1218 int err
= -EFSCORRUPTED
;
1220 if (ino
< EXT4_FIRST_INO(sb
) || ino
> max_ino
)
1223 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
1224 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
1225 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
1226 if (IS_ERR(bitmap_bh
)) {
1227 ext4_error(sb
, "inode bitmap error %ld for orphan %lu",
1228 ino
, PTR_ERR(bitmap_bh
));
1229 return (struct inode
*) bitmap_bh
;
1232 /* Having the inode bit set should be a 100% indicator that this
1233 * is a valid orphan (no e2fsck run on fs). Orphans also include
1234 * inodes that were being truncated, so we can't check i_nlink==0.
1236 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
1239 inode
= ext4_iget(sb
, ino
, EXT4_IGET_NORMAL
);
1240 if (IS_ERR(inode
)) {
1241 err
= PTR_ERR(inode
);
1242 ext4_error(sb
, "couldn't read orphan inode %lu (err %d)",
1248 * If the orphans has i_nlinks > 0 then it should be able to
1249 * be truncated, otherwise it won't be removed from the orphan
1250 * list during processing and an infinite loop will result.
1251 * Similarly, it must not be a bad inode.
1253 if ((inode
->i_nlink
&& !ext4_can_truncate(inode
)) ||
1254 is_bad_inode(inode
))
1257 if (NEXT_ORPHAN(inode
) > max_ino
)
1263 ext4_error(sb
, "bad orphan inode %lu", ino
);
1265 printk(KERN_ERR
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
1266 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
1267 ext4_test_bit(bit
, bitmap_bh
->b_data
));
1269 printk(KERN_ERR
"is_bad_inode(inode)=%d\n",
1270 is_bad_inode(inode
));
1271 printk(KERN_ERR
"NEXT_ORPHAN(inode)=%u\n",
1272 NEXT_ORPHAN(inode
));
1273 printk(KERN_ERR
"max_ino=%lu\n", max_ino
);
1274 printk(KERN_ERR
"i_nlink=%u\n", inode
->i_nlink
);
1275 /* Avoid freeing blocks if we got a bad deleted inode */
1276 if (inode
->i_nlink
== 0)
1277 inode
->i_blocks
= 0;
1281 return ERR_PTR(err
);
1284 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
1286 unsigned long desc_count
;
1287 struct ext4_group_desc
*gdp
;
1288 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1290 struct ext4_super_block
*es
;
1291 unsigned long bitmap_count
, x
;
1292 struct buffer_head
*bitmap_bh
= NULL
;
1294 es
= EXT4_SB(sb
)->s_es
;
1298 for (i
= 0; i
< ngroups
; i
++) {
1299 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1302 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1304 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1305 if (IS_ERR(bitmap_bh
)) {
1310 x
= ext4_count_free(bitmap_bh
->b_data
,
1311 EXT4_INODES_PER_GROUP(sb
) / 8);
1312 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1313 (unsigned long) i
, ext4_free_inodes_count(sb
, gdp
), x
);
1317 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1318 "stored = %u, computed = %lu, %lu\n",
1319 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1323 for (i
= 0; i
< ngroups
; i
++) {
1324 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1327 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1334 /* Called at mount-time, super-block is locked */
1335 unsigned long ext4_count_dirs(struct super_block
* sb
)
1337 unsigned long count
= 0;
1338 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1340 for (i
= 0; i
< ngroups
; i
++) {
1341 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1344 count
+= ext4_used_dirs_count(sb
, gdp
);
1350 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1351 * inode table. Must be called without any spinlock held. The only place
1352 * where it is called from on active part of filesystem is ext4lazyinit
1353 * thread, so we do not need any special locks, however we have to prevent
1354 * inode allocation from the current group, so we take alloc_sem lock, to
1355 * block ext4_new_inode() until we are finished.
1357 int ext4_init_inode_table(struct super_block
*sb
, ext4_group_t group
,
1360 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
1361 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1362 struct ext4_group_desc
*gdp
= NULL
;
1363 struct buffer_head
*group_desc_bh
;
1366 int num
, ret
= 0, used_blks
= 0;
1368 /* This should not happen, but just to be sure check this */
1369 if (sb_rdonly(sb
)) {
1374 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
1379 * We do not need to lock this, because we are the only one
1380 * handling this flag.
1382 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_ZEROED
))
1385 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
1386 if (IS_ERR(handle
)) {
1387 ret
= PTR_ERR(handle
);
1391 down_write(&grp
->alloc_sem
);
1393 * If inode bitmap was already initialized there may be some
1394 * used inodes so we need to skip blocks with used inodes in
1397 if (!(gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)))
1398 used_blks
= DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb
) -
1399 ext4_itable_unused_count(sb
, gdp
)),
1400 sbi
->s_inodes_per_block
);
1402 if ((used_blks
< 0) || (used_blks
> sbi
->s_itb_per_group
) ||
1403 ((group
== 0) && ((EXT4_INODES_PER_GROUP(sb
) -
1404 ext4_itable_unused_count(sb
, gdp
)) <
1405 EXT4_FIRST_INO(sb
)))) {
1406 ext4_error(sb
, "Something is wrong with group %u: "
1407 "used itable blocks: %d; "
1408 "itable unused count: %u",
1410 ext4_itable_unused_count(sb
, gdp
));
1415 blk
= ext4_inode_table(sb
, gdp
) + used_blks
;
1416 num
= sbi
->s_itb_per_group
- used_blks
;
1418 BUFFER_TRACE(group_desc_bh
, "get_write_access");
1419 ret
= ext4_journal_get_write_access(handle
,
1425 * Skip zeroout if the inode table is full. But we set the ZEROED
1426 * flag anyway, because obviously, when it is full it does not need
1429 if (unlikely(num
== 0))
1432 ext4_debug("going to zero out inode table in group %d\n",
1434 ret
= sb_issue_zeroout(sb
, blk
, num
, GFP_NOFS
);
1438 blkdev_issue_flush(sb
->s_bdev
, GFP_NOFS
, NULL
);
1441 ext4_lock_group(sb
, group
);
1442 gdp
->bg_flags
|= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
1443 ext4_group_desc_csum_set(sb
, group
, gdp
);
1444 ext4_unlock_group(sb
, group
);
1446 BUFFER_TRACE(group_desc_bh
,
1447 "call ext4_handle_dirty_metadata");
1448 ret
= ext4_handle_dirty_metadata(handle
, NULL
,
1452 up_write(&grp
->alloc_sem
);
1453 ext4_journal_stop(handle
);