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 buffer_head
*bh
= NULL
;
129 ext4_fsblk_t bitmap_blk
;
132 desc
= ext4_get_group_desc(sb
, block_group
, NULL
);
134 return ERR_PTR(-EFSCORRUPTED
);
136 bitmap_blk
= ext4_inode_bitmap(sb
, desc
);
137 bh
= sb_getblk(sb
, bitmap_blk
);
139 ext4_error(sb
, "Cannot read inode bitmap - "
140 "block_group = %u, inode_bitmap = %llu",
141 block_group
, bitmap_blk
);
142 return ERR_PTR(-EIO
);
144 if (bitmap_uptodate(bh
))
148 if (bitmap_uptodate(bh
)) {
153 ext4_lock_group(sb
, block_group
);
154 if (desc
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
155 memset(bh
->b_data
, 0, (EXT4_INODES_PER_GROUP(sb
) + 7) / 8);
156 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
157 sb
->s_blocksize
* 8, bh
->b_data
);
158 set_bitmap_uptodate(bh
);
159 set_buffer_uptodate(bh
);
160 set_buffer_verified(bh
);
161 ext4_unlock_group(sb
, block_group
);
165 ext4_unlock_group(sb
, block_group
);
167 if (buffer_uptodate(bh
)) {
169 * if not uninit if bh is uptodate,
170 * bitmap is also uptodate
172 set_bitmap_uptodate(bh
);
177 * submit the buffer_head for reading
179 trace_ext4_load_inode_bitmap(sb
, block_group
);
180 bh
->b_end_io
= ext4_end_bitmap_read
;
182 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
184 if (!buffer_uptodate(bh
)) {
186 ext4_error(sb
, "Cannot read inode bitmap - "
187 "block_group = %u, inode_bitmap = %llu",
188 block_group
, bitmap_blk
);
189 return ERR_PTR(-EIO
);
193 err
= ext4_validate_inode_bitmap(sb
, desc
, block_group
, bh
);
203 * NOTE! When we get the inode, we're the only people
204 * that have access to it, and as such there are no
205 * race conditions we have to worry about. The inode
206 * is not on the hash-lists, and it cannot be reached
207 * through the filesystem because the directory entry
208 * has been deleted earlier.
210 * HOWEVER: we must make sure that we get no aliases,
211 * which means that we have to call "clear_inode()"
212 * _before_ we mark the inode not in use in the inode
213 * bitmaps. Otherwise a newly created file might use
214 * the same inode number (not actually the same pointer
215 * though), and then we'd have two inodes sharing the
216 * same inode number and space on the harddisk.
218 void ext4_free_inode(handle_t
*handle
, struct inode
*inode
)
220 struct super_block
*sb
= inode
->i_sb
;
223 struct buffer_head
*bitmap_bh
= NULL
;
224 struct buffer_head
*bh2
;
225 ext4_group_t block_group
;
227 struct ext4_group_desc
*gdp
;
228 struct ext4_super_block
*es
;
229 struct ext4_sb_info
*sbi
;
230 int fatal
= 0, err
, count
, cleared
;
231 struct ext4_group_info
*grp
;
234 printk(KERN_ERR
"EXT4-fs: %s:%d: inode on "
235 "nonexistent device\n", __func__
, __LINE__
);
238 if (atomic_read(&inode
->i_count
) > 1) {
239 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: count=%d",
240 __func__
, __LINE__
, inode
->i_ino
,
241 atomic_read(&inode
->i_count
));
244 if (inode
->i_nlink
) {
245 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: nlink=%d\n",
246 __func__
, __LINE__
, inode
->i_ino
, inode
->i_nlink
);
252 ext4_debug("freeing inode %lu\n", ino
);
253 trace_ext4_free_inode(inode
);
256 * Note: we must free any quota before locking the superblock,
257 * as writing the quota to disk may need the lock as well.
259 dquot_initialize(inode
);
260 dquot_free_inode(inode
);
263 is_directory
= S_ISDIR(inode
->i_mode
);
265 /* Do this BEFORE marking the inode not in use or returning an error */
266 ext4_clear_inode(inode
);
268 es
= EXT4_SB(sb
)->s_es
;
269 if (ino
< EXT4_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
270 ext4_error(sb
, "reserved or nonexistent inode %lu", ino
);
273 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
274 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
275 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
276 /* Don't bother if the inode bitmap is corrupt. */
277 grp
= ext4_get_group_info(sb
, block_group
);
278 if (IS_ERR(bitmap_bh
)) {
279 fatal
= PTR_ERR(bitmap_bh
);
283 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))) {
284 fatal
= -EFSCORRUPTED
;
288 BUFFER_TRACE(bitmap_bh
, "get_write_access");
289 fatal
= ext4_journal_get_write_access(handle
, bitmap_bh
);
294 gdp
= ext4_get_group_desc(sb
, block_group
, &bh2
);
296 BUFFER_TRACE(bh2
, "get_write_access");
297 fatal
= ext4_journal_get_write_access(handle
, bh2
);
299 ext4_lock_group(sb
, block_group
);
300 cleared
= ext4_test_and_clear_bit(bit
, bitmap_bh
->b_data
);
301 if (fatal
|| !cleared
) {
302 ext4_unlock_group(sb
, block_group
);
306 count
= ext4_free_inodes_count(sb
, gdp
) + 1;
307 ext4_free_inodes_set(sb
, gdp
, count
);
309 count
= ext4_used_dirs_count(sb
, gdp
) - 1;
310 ext4_used_dirs_set(sb
, gdp
, count
);
311 percpu_counter_dec(&sbi
->s_dirs_counter
);
313 ext4_inode_bitmap_csum_set(sb
, block_group
, gdp
, bitmap_bh
,
314 EXT4_INODES_PER_GROUP(sb
) / 8);
315 ext4_group_desc_csum_set(sb
, block_group
, gdp
);
316 ext4_unlock_group(sb
, block_group
);
318 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
319 if (sbi
->s_log_groups_per_flex
) {
320 ext4_group_t f
= ext4_flex_group(sbi
, block_group
);
322 atomic_inc(&sbi
->s_flex_groups
[f
].free_inodes
);
324 atomic_dec(&sbi
->s_flex_groups
[f
].used_dirs
);
326 BUFFER_TRACE(bh2
, "call ext4_handle_dirty_metadata");
327 fatal
= ext4_handle_dirty_metadata(handle
, NULL
, bh2
);
330 BUFFER_TRACE(bitmap_bh
, "call ext4_handle_dirty_metadata");
331 err
= ext4_handle_dirty_metadata(handle
, NULL
, bitmap_bh
);
335 ext4_error(sb
, "bit already cleared for inode %lu", ino
);
336 if (gdp
&& !EXT4_MB_GRP_IBITMAP_CORRUPT(grp
)) {
338 count
= ext4_free_inodes_count(sb
, gdp
);
339 percpu_counter_sub(&sbi
->s_freeinodes_counter
,
342 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT
, &grp
->bb_state
);
347 ext4_std_error(sb
, fatal
);
357 * Helper function for Orlov's allocator; returns critical information
358 * for a particular block group or flex_bg. If flex_size is 1, then g
359 * is a block group number; otherwise it is flex_bg number.
361 static void get_orlov_stats(struct super_block
*sb
, ext4_group_t g
,
362 int flex_size
, struct orlov_stats
*stats
)
364 struct ext4_group_desc
*desc
;
365 struct flex_groups
*flex_group
= EXT4_SB(sb
)->s_flex_groups
;
368 stats
->free_inodes
= atomic_read(&flex_group
[g
].free_inodes
);
369 stats
->free_clusters
= atomic64_read(&flex_group
[g
].free_clusters
);
370 stats
->used_dirs
= atomic_read(&flex_group
[g
].used_dirs
);
374 desc
= ext4_get_group_desc(sb
, g
, NULL
);
376 stats
->free_inodes
= ext4_free_inodes_count(sb
, desc
);
377 stats
->free_clusters
= ext4_free_group_clusters(sb
, desc
);
378 stats
->used_dirs
= ext4_used_dirs_count(sb
, desc
);
380 stats
->free_inodes
= 0;
381 stats
->free_clusters
= 0;
382 stats
->used_dirs
= 0;
387 * Orlov's allocator for directories.
389 * We always try to spread first-level directories.
391 * If there are blockgroups with both free inodes and free blocks counts
392 * not worse than average we return one with smallest directory count.
393 * Otherwise we simply return a random group.
395 * For the rest rules look so:
397 * It's OK to put directory into a group unless
398 * it has too many directories already (max_dirs) or
399 * it has too few free inodes left (min_inodes) or
400 * it has too few free blocks left (min_blocks) or
401 * Parent's group is preferred, if it doesn't satisfy these
402 * conditions we search cyclically through the rest. If none
403 * of the groups look good we just look for a group with more
404 * free inodes than average (starting at parent's group).
407 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
,
408 ext4_group_t
*group
, umode_t mode
,
409 const struct qstr
*qstr
)
411 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
412 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
413 ext4_group_t real_ngroups
= ext4_get_groups_count(sb
);
414 int inodes_per_group
= EXT4_INODES_PER_GROUP(sb
);
415 unsigned int freei
, avefreei
, grp_free
;
416 ext4_fsblk_t freeb
, avefreec
;
418 int max_dirs
, min_inodes
;
419 ext4_grpblk_t min_clusters
;
420 ext4_group_t i
, grp
, g
, ngroups
;
421 struct ext4_group_desc
*desc
;
422 struct orlov_stats stats
;
423 int flex_size
= ext4_flex_bg_size(sbi
);
424 struct dx_hash_info hinfo
;
426 ngroups
= real_ngroups
;
428 ngroups
= (real_ngroups
+ flex_size
- 1) >>
429 sbi
->s_log_groups_per_flex
;
430 parent_group
>>= sbi
->s_log_groups_per_flex
;
433 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
434 avefreei
= freei
/ ngroups
;
435 freeb
= EXT4_C2B(sbi
,
436 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
438 do_div(avefreec
, ngroups
);
439 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
442 ((parent
== d_inode(sb
->s_root
)) ||
443 (ext4_test_inode_flag(parent
, EXT4_INODE_TOPDIR
)))) {
444 int best_ndir
= inodes_per_group
;
448 hinfo
.hash_version
= DX_HASH_HALF_MD4
;
449 hinfo
.seed
= sbi
->s_hash_seed
;
450 ext4fs_dirhash(qstr
->name
, qstr
->len
, &hinfo
);
454 parent_group
= (unsigned)grp
% ngroups
;
455 for (i
= 0; i
< ngroups
; i
++) {
456 g
= (parent_group
+ i
) % ngroups
;
457 get_orlov_stats(sb
, g
, flex_size
, &stats
);
458 if (!stats
.free_inodes
)
460 if (stats
.used_dirs
>= best_ndir
)
462 if (stats
.free_inodes
< avefreei
)
464 if (stats
.free_clusters
< avefreec
)
468 best_ndir
= stats
.used_dirs
;
473 if (flex_size
== 1) {
479 * We pack inodes at the beginning of the flexgroup's
480 * inode tables. Block allocation decisions will do
481 * something similar, although regular files will
482 * start at 2nd block group of the flexgroup. See
483 * ext4_ext_find_goal() and ext4_find_near().
486 for (i
= 0; i
< flex_size
; i
++) {
487 if (grp
+i
>= real_ngroups
)
489 desc
= ext4_get_group_desc(sb
, grp
+i
, NULL
);
490 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
498 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
499 min_inodes
= avefreei
- inodes_per_group
*flex_size
/ 4;
502 min_clusters
= avefreec
- EXT4_CLUSTERS_PER_GROUP(sb
)*flex_size
/ 4;
505 * Start looking in the flex group where we last allocated an
506 * inode for this parent directory
508 if (EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
509 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
511 parent_group
>>= sbi
->s_log_groups_per_flex
;
514 for (i
= 0; i
< ngroups
; i
++) {
515 grp
= (parent_group
+ i
) % ngroups
;
516 get_orlov_stats(sb
, grp
, flex_size
, &stats
);
517 if (stats
.used_dirs
>= max_dirs
)
519 if (stats
.free_inodes
< min_inodes
)
521 if (stats
.free_clusters
< min_clusters
)
527 ngroups
= real_ngroups
;
528 avefreei
= freei
/ ngroups
;
530 parent_group
= EXT4_I(parent
)->i_block_group
;
531 for (i
= 0; i
< ngroups
; i
++) {
532 grp
= (parent_group
+ i
) % ngroups
;
533 desc
= ext4_get_group_desc(sb
, grp
, NULL
);
535 grp_free
= ext4_free_inodes_count(sb
, desc
);
536 if (grp_free
&& grp_free
>= avefreei
) {
545 * The free-inodes counter is approximate, and for really small
546 * filesystems the above test can fail to find any blockgroups
555 static int find_group_other(struct super_block
*sb
, struct inode
*parent
,
556 ext4_group_t
*group
, umode_t mode
)
558 ext4_group_t parent_group
= EXT4_I(parent
)->i_block_group
;
559 ext4_group_t i
, last
, ngroups
= ext4_get_groups_count(sb
);
560 struct ext4_group_desc
*desc
;
561 int flex_size
= ext4_flex_bg_size(EXT4_SB(sb
));
564 * Try to place the inode is the same flex group as its
565 * parent. If we can't find space, use the Orlov algorithm to
566 * find another flex group, and store that information in the
567 * parent directory's inode information so that use that flex
568 * group for future allocations.
574 parent_group
&= ~(flex_size
-1);
575 last
= parent_group
+ flex_size
;
578 for (i
= parent_group
; i
< last
; i
++) {
579 desc
= ext4_get_group_desc(sb
, i
, NULL
);
580 if (desc
&& ext4_free_inodes_count(sb
, desc
)) {
585 if (!retry
&& EXT4_I(parent
)->i_last_alloc_group
!= ~0) {
587 parent_group
= EXT4_I(parent
)->i_last_alloc_group
;
591 * If this didn't work, use the Orlov search algorithm
592 * to find a new flex group; we pass in the mode to
593 * avoid the topdir algorithms.
595 *group
= parent_group
+ flex_size
;
596 if (*group
> ngroups
)
598 return find_group_orlov(sb
, parent
, group
, mode
, NULL
);
602 * Try to place the inode in its parent directory
604 *group
= parent_group
;
605 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
606 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
607 ext4_free_group_clusters(sb
, desc
))
611 * We're going to place this inode in a different blockgroup from its
612 * parent. We want to cause files in a common directory to all land in
613 * the same blockgroup. But we want files which are in a different
614 * directory which shares a blockgroup with our parent to land in a
615 * different blockgroup.
617 * So add our directory's i_ino into the starting point for the hash.
619 *group
= (*group
+ parent
->i_ino
) % ngroups
;
622 * Use a quadratic hash to find a group with a free inode and some free
625 for (i
= 1; i
< ngroups
; i
<<= 1) {
627 if (*group
>= ngroups
)
629 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
630 if (desc
&& ext4_free_inodes_count(sb
, desc
) &&
631 ext4_free_group_clusters(sb
, desc
))
636 * That failed: try linear search for a free inode, even if that group
637 * has no free blocks.
639 *group
= parent_group
;
640 for (i
= 0; i
< ngroups
; i
++) {
641 if (++*group
>= ngroups
)
643 desc
= ext4_get_group_desc(sb
, *group
, NULL
);
644 if (desc
&& ext4_free_inodes_count(sb
, desc
))
652 * In no journal mode, if an inode has recently been deleted, we want
653 * to avoid reusing it until we're reasonably sure the inode table
654 * block has been written back to disk. (Yes, these values are
655 * somewhat arbitrary...)
657 #define RECENTCY_MIN 5
658 #define RECENTCY_DIRTY 300
660 static int recently_deleted(struct super_block
*sb
, ext4_group_t group
, int ino
)
662 struct ext4_group_desc
*gdp
;
663 struct ext4_inode
*raw_inode
;
664 struct buffer_head
*bh
;
665 int inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
667 int recentcy
= RECENTCY_MIN
;
670 gdp
= ext4_get_group_desc(sb
, group
, NULL
);
674 bh
= sb_find_get_block(sb
, ext4_inode_table(sb
, gdp
) +
675 (ino
/ inodes_per_block
));
676 if (!bh
|| !buffer_uptodate(bh
))
678 * If the block is not in the buffer cache, then it
679 * must have been written out.
683 offset
= (ino
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
684 raw_inode
= (struct ext4_inode
*) (bh
->b_data
+ offset
);
686 /* i_dtime is only 32 bits on disk, but we only care about relative
687 * times in the range of a few minutes (i.e. long enough to sync a
688 * recently-deleted inode to disk), so using the low 32 bits of the
689 * clock (a 68 year range) is enough, see time_before32() */
690 dtime
= le32_to_cpu(raw_inode
->i_dtime
);
691 now
= ktime_get_real_seconds();
692 if (buffer_dirty(bh
))
693 recentcy
+= RECENTCY_DIRTY
;
695 if (dtime
&& time_before32(dtime
, now
) &&
696 time_before32(now
, dtime
+ recentcy
))
703 static int find_inode_bit(struct super_block
*sb
, ext4_group_t group
,
704 struct buffer_head
*bitmap
, unsigned long *ino
)
707 *ino
= ext4_find_next_zero_bit((unsigned long *)
709 EXT4_INODES_PER_GROUP(sb
), *ino
);
710 if (*ino
>= EXT4_INODES_PER_GROUP(sb
))
713 if ((EXT4_SB(sb
)->s_journal
== NULL
) &&
714 recently_deleted(sb
, group
, *ino
)) {
716 if (*ino
< EXT4_INODES_PER_GROUP(sb
))
725 * There are two policies for allocating an inode. If the new inode is
726 * a directory, then a forward search is made for a block group with both
727 * free space and a low directory-to-inode ratio; if that fails, then of
728 * the groups with above-average free space, that group with the fewest
729 * directories already is chosen.
731 * For other inodes, search forward from the parent directory's block
732 * group to find a free inode.
734 struct inode
*__ext4_new_inode(handle_t
*handle
, struct inode
*dir
,
735 umode_t mode
, const struct qstr
*qstr
,
736 __u32 goal
, uid_t
*owner
, __u32 i_flags
,
737 int handle_type
, unsigned int line_no
,
740 struct super_block
*sb
;
741 struct buffer_head
*inode_bitmap_bh
= NULL
;
742 struct buffer_head
*group_desc_bh
;
743 ext4_group_t ngroups
, group
= 0;
744 unsigned long ino
= 0;
746 struct ext4_group_desc
*gdp
= NULL
;
747 struct ext4_inode_info
*ei
;
748 struct ext4_sb_info
*sbi
;
752 ext4_group_t flex_group
;
753 struct ext4_group_info
*grp
;
756 /* Cannot create files in a deleted directory */
757 if (!dir
|| !dir
->i_nlink
)
758 return ERR_PTR(-EPERM
);
763 if (unlikely(ext4_forced_shutdown(sbi
)))
764 return ERR_PTR(-EIO
);
766 /* Supplied owner must be valid */
767 if (owner
&& (owner
[0] == (uid_t
)-1 || owner
[1] == (uid_t
)-1))
768 return ERR_PTR(-EOVERFLOW
);
770 if ((ext4_encrypted_inode(dir
) || DUMMY_ENCRYPTION_ENABLED(sbi
)) &&
771 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) &&
772 !(i_flags
& EXT4_EA_INODE_FL
)) {
773 err
= fscrypt_get_encryption_info(dir
);
776 if (!fscrypt_has_encryption_key(dir
))
777 return ERR_PTR(-ENOKEY
);
781 if (!handle
&& sbi
->s_journal
&& !(i_flags
& EXT4_EA_INODE_FL
)) {
782 #ifdef CONFIG_EXT4_FS_POSIX_ACL
783 struct posix_acl
*p
= get_acl(dir
, ACL_TYPE_DEFAULT
);
788 int acl_size
= p
->a_count
* sizeof(ext4_acl_entry
);
790 nblocks
+= (S_ISDIR(mode
) ? 2 : 1) *
791 __ext4_xattr_set_credits(sb
, NULL
/* inode */,
792 NULL
/* block_bh */, acl_size
,
793 true /* is_create */);
794 posix_acl_release(p
);
798 #ifdef CONFIG_SECURITY
800 int num_security_xattrs
= 1;
802 #ifdef CONFIG_INTEGRITY
803 num_security_xattrs
++;
806 * We assume that security xattrs are never
807 * more than 1k. In practice they are under
810 nblocks
+= num_security_xattrs
*
811 __ext4_xattr_set_credits(sb
, NULL
/* inode */,
812 NULL
/* block_bh */, 1024,
813 true /* is_create */);
817 nblocks
+= __ext4_xattr_set_credits(sb
,
818 NULL
/* inode */, NULL
/* block_bh */,
819 FSCRYPT_SET_CONTEXT_MAX_SIZE
,
820 true /* is_create */);
823 ngroups
= ext4_get_groups_count(sb
);
824 trace_ext4_request_inode(dir
, mode
);
825 inode
= new_inode(sb
);
827 return ERR_PTR(-ENOMEM
);
831 * Initialize owners and quota early so that we don't have to account
832 * for quota initialization worst case in standard inode creating
836 inode
->i_mode
= mode
;
837 i_uid_write(inode
, owner
[0]);
838 i_gid_write(inode
, owner
[1]);
839 } else if (test_opt(sb
, GRPID
)) {
840 inode
->i_mode
= mode
;
841 inode
->i_uid
= current_fsuid();
842 inode
->i_gid
= dir
->i_gid
;
844 inode_init_owner(inode
, dir
, mode
);
846 if (ext4_has_feature_project(sb
) &&
847 ext4_test_inode_flag(dir
, EXT4_INODE_PROJINHERIT
))
848 ei
->i_projid
= EXT4_I(dir
)->i_projid
;
850 ei
->i_projid
= make_kprojid(sb
->s_user_ns
, EXT4_DEF_PROJID
);
852 err
= dquot_initialize(inode
);
857 goal
= sbi
->s_inode_goal
;
859 if (goal
&& goal
<= le32_to_cpu(sbi
->s_es
->s_inodes_count
)) {
860 group
= (goal
- 1) / EXT4_INODES_PER_GROUP(sb
);
861 ino
= (goal
- 1) % EXT4_INODES_PER_GROUP(sb
);
867 ret2
= find_group_orlov(sb
, dir
, &group
, mode
, qstr
);
869 ret2
= find_group_other(sb
, dir
, &group
, mode
);
872 EXT4_I(dir
)->i_last_alloc_group
= group
;
878 * Normally we will only go through one pass of this loop,
879 * unless we get unlucky and it turns out the group we selected
880 * had its last inode grabbed by someone else.
882 for (i
= 0; i
< ngroups
; i
++, ino
= 0) {
885 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
890 * Check free inodes count before loading bitmap.
892 if (ext4_free_inodes_count(sb
, gdp
) == 0)
895 grp
= ext4_get_group_info(sb
, group
);
896 /* Skip groups with already-known suspicious inode tables */
897 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
))
900 brelse(inode_bitmap_bh
);
901 inode_bitmap_bh
= ext4_read_inode_bitmap(sb
, group
);
902 /* Skip groups with suspicious inode tables */
903 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp
) ||
904 IS_ERR(inode_bitmap_bh
)) {
905 inode_bitmap_bh
= NULL
;
909 repeat_in_this_group
:
910 ret2
= find_inode_bit(sb
, group
, inode_bitmap_bh
, &ino
);
914 if (group
== 0 && (ino
+ 1) < EXT4_FIRST_INO(sb
)) {
915 ext4_error(sb
, "reserved inode found cleared - "
916 "inode=%lu", ino
+ 1);
921 BUG_ON(nblocks
<= 0);
922 handle
= __ext4_journal_start_sb(dir
->i_sb
, line_no
,
923 handle_type
, nblocks
,
925 if (IS_ERR(handle
)) {
926 err
= PTR_ERR(handle
);
927 ext4_std_error(sb
, err
);
931 BUFFER_TRACE(inode_bitmap_bh
, "get_write_access");
932 err
= ext4_journal_get_write_access(handle
, inode_bitmap_bh
);
934 ext4_std_error(sb
, err
);
937 ext4_lock_group(sb
, group
);
938 ret2
= ext4_test_and_set_bit(ino
, inode_bitmap_bh
->b_data
);
940 /* Someone already took the bit. Repeat the search
943 ret2
= find_inode_bit(sb
, group
, inode_bitmap_bh
, &ino
);
945 ext4_set_bit(ino
, inode_bitmap_bh
->b_data
);
948 ret2
= 1; /* we didn't grab the inode */
951 ext4_unlock_group(sb
, group
);
952 ino
++; /* the inode bitmap is zero-based */
954 goto got
; /* we grabbed the inode! */
956 if (ino
< EXT4_INODES_PER_GROUP(sb
))
957 goto repeat_in_this_group
;
959 if (++group
== ngroups
)
966 BUFFER_TRACE(inode_bitmap_bh
, "call ext4_handle_dirty_metadata");
967 err
= ext4_handle_dirty_metadata(handle
, NULL
, inode_bitmap_bh
);
969 ext4_std_error(sb
, err
);
973 BUFFER_TRACE(group_desc_bh
, "get_write_access");
974 err
= ext4_journal_get_write_access(handle
, group_desc_bh
);
976 ext4_std_error(sb
, err
);
980 /* We may have to initialize the block bitmap if it isn't already */
981 if (ext4_has_group_desc_csum(sb
) &&
982 gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
983 struct buffer_head
*block_bitmap_bh
;
985 block_bitmap_bh
= ext4_read_block_bitmap(sb
, group
);
986 if (IS_ERR(block_bitmap_bh
)) {
987 err
= PTR_ERR(block_bitmap_bh
);
990 BUFFER_TRACE(block_bitmap_bh
, "get block bitmap access");
991 err
= ext4_journal_get_write_access(handle
, block_bitmap_bh
);
993 brelse(block_bitmap_bh
);
994 ext4_std_error(sb
, err
);
998 BUFFER_TRACE(block_bitmap_bh
, "dirty block bitmap");
999 err
= ext4_handle_dirty_metadata(handle
, NULL
, block_bitmap_bh
);
1001 /* recheck and clear flag under lock if we still need to */
1002 ext4_lock_group(sb
, group
);
1003 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_BLOCK_UNINIT
)) {
1004 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT
);
1005 ext4_free_group_clusters_set(sb
, gdp
,
1006 ext4_free_clusters_after_init(sb
, group
, gdp
));
1007 ext4_block_bitmap_csum_set(sb
, group
, gdp
,
1009 ext4_group_desc_csum_set(sb
, group
, gdp
);
1011 ext4_unlock_group(sb
, group
);
1012 brelse(block_bitmap_bh
);
1015 ext4_std_error(sb
, err
);
1020 /* Update the relevant bg descriptor fields */
1021 if (ext4_has_group_desc_csum(sb
)) {
1023 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
1025 down_read(&grp
->alloc_sem
); /* protect vs itable lazyinit */
1026 ext4_lock_group(sb
, group
); /* while we modify the bg desc */
1027 free
= EXT4_INODES_PER_GROUP(sb
) -
1028 ext4_itable_unused_count(sb
, gdp
);
1029 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)) {
1030 gdp
->bg_flags
&= cpu_to_le16(~EXT4_BG_INODE_UNINIT
);
1034 * Check the relative inode number against the last used
1035 * relative inode number in this group. if it is greater
1036 * we need to update the bg_itable_unused count
1039 ext4_itable_unused_set(sb
, gdp
,
1040 (EXT4_INODES_PER_GROUP(sb
) - ino
));
1041 up_read(&grp
->alloc_sem
);
1043 ext4_lock_group(sb
, group
);
1046 ext4_free_inodes_set(sb
, gdp
, ext4_free_inodes_count(sb
, gdp
) - 1);
1047 if (S_ISDIR(mode
)) {
1048 ext4_used_dirs_set(sb
, gdp
, ext4_used_dirs_count(sb
, gdp
) + 1);
1049 if (sbi
->s_log_groups_per_flex
) {
1050 ext4_group_t f
= ext4_flex_group(sbi
, group
);
1052 atomic_inc(&sbi
->s_flex_groups
[f
].used_dirs
);
1055 if (ext4_has_group_desc_csum(sb
)) {
1056 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, inode_bitmap_bh
,
1057 EXT4_INODES_PER_GROUP(sb
) / 8);
1058 ext4_group_desc_csum_set(sb
, group
, gdp
);
1060 ext4_unlock_group(sb
, group
);
1062 BUFFER_TRACE(group_desc_bh
, "call ext4_handle_dirty_metadata");
1063 err
= ext4_handle_dirty_metadata(handle
, NULL
, group_desc_bh
);
1065 ext4_std_error(sb
, err
);
1069 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
1071 percpu_counter_inc(&sbi
->s_dirs_counter
);
1073 if (sbi
->s_log_groups_per_flex
) {
1074 flex_group
= ext4_flex_group(sbi
, group
);
1075 atomic_dec(&sbi
->s_flex_groups
[flex_group
].free_inodes
);
1078 inode
->i_ino
= ino
+ group
* EXT4_INODES_PER_GROUP(sb
);
1079 /* This is the optimal IO size (for stat), not the fs block size */
1080 inode
->i_blocks
= 0;
1081 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= ei
->i_crtime
=
1082 current_time(inode
);
1084 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
1085 ei
->i_dir_start_lookup
= 0;
1088 /* Don't inherit extent flag from directory, amongst others. */
1090 ext4_mask_flags(mode
, EXT4_I(dir
)->i_flags
& EXT4_FL_INHERITED
);
1091 ei
->i_flags
|= i_flags
;
1094 ei
->i_block_group
= group
;
1095 ei
->i_last_alloc_group
= ~0;
1097 ext4_set_inode_flags(inode
);
1098 if (IS_DIRSYNC(inode
))
1099 ext4_handle_sync(handle
);
1100 if (insert_inode_locked(inode
) < 0) {
1102 * Likely a bitmap corruption causing inode to be allocated
1106 ext4_error(sb
, "failed to insert inode %lu: doubly allocated?",
1110 inode
->i_generation
= prandom_u32();
1112 /* Precompute checksum seed for inode metadata */
1113 if (ext4_has_metadata_csum(sb
)) {
1115 __le32 inum
= cpu_to_le32(inode
->i_ino
);
1116 __le32 gen
= cpu_to_le32(inode
->i_generation
);
1117 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
1119 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
1123 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
1124 ext4_set_inode_state(inode
, EXT4_STATE_NEW
);
1126 ei
->i_extra_isize
= EXT4_SB(sb
)->s_want_extra_isize
;
1127 ei
->i_inline_off
= 0;
1128 if (ext4_has_feature_inline_data(sb
))
1129 ext4_set_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
1131 err
= dquot_alloc_inode(inode
);
1136 * Since the encryption xattr will always be unique, create it first so
1137 * that it's less likely to end up in an external xattr block and
1138 * prevent its deduplication.
1141 err
= fscrypt_inherit_context(dir
, inode
, handle
, true);
1143 goto fail_free_drop
;
1146 if (!(ei
->i_flags
& EXT4_EA_INODE_FL
)) {
1147 err
= ext4_init_acl(handle
, inode
, dir
);
1149 goto fail_free_drop
;
1151 err
= ext4_init_security(handle
, inode
, dir
, qstr
);
1153 goto fail_free_drop
;
1156 if (ext4_has_feature_extents(sb
)) {
1157 /* set extent flag only for directory, file and normal symlink*/
1158 if (S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
)) {
1159 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
1160 ext4_ext_tree_init(handle
, inode
);
1164 if (ext4_handle_valid(handle
)) {
1165 ei
->i_sync_tid
= handle
->h_transaction
->t_tid
;
1166 ei
->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1169 err
= ext4_mark_inode_dirty(handle
, inode
);
1171 ext4_std_error(sb
, err
);
1172 goto fail_free_drop
;
1175 ext4_debug("allocating inode %lu\n", inode
->i_ino
);
1176 trace_ext4_allocate_inode(inode
, dir
, mode
);
1177 brelse(inode_bitmap_bh
);
1181 dquot_free_inode(inode
);
1184 unlock_new_inode(inode
);
1187 inode
->i_flags
|= S_NOQUOTA
;
1189 brelse(inode_bitmap_bh
);
1190 return ERR_PTR(err
);
1193 /* Verify that we are loading a valid orphan from disk */
1194 struct inode
*ext4_orphan_get(struct super_block
*sb
, unsigned long ino
)
1196 unsigned long max_ino
= le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
);
1197 ext4_group_t block_group
;
1199 struct buffer_head
*bitmap_bh
= NULL
;
1200 struct inode
*inode
= NULL
;
1201 int err
= -EFSCORRUPTED
;
1203 if (ino
< EXT4_FIRST_INO(sb
) || ino
> max_ino
)
1206 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
1207 bit
= (ino
- 1) % EXT4_INODES_PER_GROUP(sb
);
1208 bitmap_bh
= ext4_read_inode_bitmap(sb
, block_group
);
1209 if (IS_ERR(bitmap_bh
)) {
1210 ext4_error(sb
, "inode bitmap error %ld for orphan %lu",
1211 ino
, PTR_ERR(bitmap_bh
));
1212 return (struct inode
*) bitmap_bh
;
1215 /* Having the inode bit set should be a 100% indicator that this
1216 * is a valid orphan (no e2fsck run on fs). Orphans also include
1217 * inodes that were being truncated, so we can't check i_nlink==0.
1219 if (!ext4_test_bit(bit
, bitmap_bh
->b_data
))
1222 inode
= ext4_iget(sb
, ino
);
1223 if (IS_ERR(inode
)) {
1224 err
= PTR_ERR(inode
);
1225 ext4_error(sb
, "couldn't read orphan inode %lu (err %d)",
1231 * If the orphans has i_nlinks > 0 then it should be able to
1232 * be truncated, otherwise it won't be removed from the orphan
1233 * list during processing and an infinite loop will result.
1234 * Similarly, it must not be a bad inode.
1236 if ((inode
->i_nlink
&& !ext4_can_truncate(inode
)) ||
1237 is_bad_inode(inode
))
1240 if (NEXT_ORPHAN(inode
) > max_ino
)
1246 ext4_error(sb
, "bad orphan inode %lu", ino
);
1248 printk(KERN_ERR
"ext4_test_bit(bit=%d, block=%llu) = %d\n",
1249 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
1250 ext4_test_bit(bit
, bitmap_bh
->b_data
));
1252 printk(KERN_ERR
"is_bad_inode(inode)=%d\n",
1253 is_bad_inode(inode
));
1254 printk(KERN_ERR
"NEXT_ORPHAN(inode)=%u\n",
1255 NEXT_ORPHAN(inode
));
1256 printk(KERN_ERR
"max_ino=%lu\n", max_ino
);
1257 printk(KERN_ERR
"i_nlink=%u\n", inode
->i_nlink
);
1258 /* Avoid freeing blocks if we got a bad deleted inode */
1259 if (inode
->i_nlink
== 0)
1260 inode
->i_blocks
= 0;
1264 return ERR_PTR(err
);
1267 unsigned long ext4_count_free_inodes(struct super_block
*sb
)
1269 unsigned long desc_count
;
1270 struct ext4_group_desc
*gdp
;
1271 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1273 struct ext4_super_block
*es
;
1274 unsigned long bitmap_count
, x
;
1275 struct buffer_head
*bitmap_bh
= NULL
;
1277 es
= EXT4_SB(sb
)->s_es
;
1281 for (i
= 0; i
< ngroups
; i
++) {
1282 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1285 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1287 bitmap_bh
= ext4_read_inode_bitmap(sb
, i
);
1288 if (IS_ERR(bitmap_bh
)) {
1293 x
= ext4_count_free(bitmap_bh
->b_data
,
1294 EXT4_INODES_PER_GROUP(sb
) / 8);
1295 printk(KERN_DEBUG
"group %lu: stored = %d, counted = %lu\n",
1296 (unsigned long) i
, ext4_free_inodes_count(sb
, gdp
), x
);
1300 printk(KERN_DEBUG
"ext4_count_free_inodes: "
1301 "stored = %u, computed = %lu, %lu\n",
1302 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
1306 for (i
= 0; i
< ngroups
; i
++) {
1307 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1310 desc_count
+= ext4_free_inodes_count(sb
, gdp
);
1317 /* Called at mount-time, super-block is locked */
1318 unsigned long ext4_count_dirs(struct super_block
* sb
)
1320 unsigned long count
= 0;
1321 ext4_group_t i
, ngroups
= ext4_get_groups_count(sb
);
1323 for (i
= 0; i
< ngroups
; i
++) {
1324 struct ext4_group_desc
*gdp
= ext4_get_group_desc(sb
, i
, NULL
);
1327 count
+= ext4_used_dirs_count(sb
, gdp
);
1333 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1334 * inode table. Must be called without any spinlock held. The only place
1335 * where it is called from on active part of filesystem is ext4lazyinit
1336 * thread, so we do not need any special locks, however we have to prevent
1337 * inode allocation from the current group, so we take alloc_sem lock, to
1338 * block ext4_new_inode() until we are finished.
1340 int ext4_init_inode_table(struct super_block
*sb
, ext4_group_t group
,
1343 struct ext4_group_info
*grp
= ext4_get_group_info(sb
, group
);
1344 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1345 struct ext4_group_desc
*gdp
= NULL
;
1346 struct buffer_head
*group_desc_bh
;
1349 int num
, ret
= 0, used_blks
= 0;
1351 /* This should not happen, but just to be sure check this */
1352 if (sb_rdonly(sb
)) {
1357 gdp
= ext4_get_group_desc(sb
, group
, &group_desc_bh
);
1362 * We do not need to lock this, because we are the only one
1363 * handling this flag.
1365 if (gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_ZEROED
))
1368 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
1369 if (IS_ERR(handle
)) {
1370 ret
= PTR_ERR(handle
);
1374 down_write(&grp
->alloc_sem
);
1376 * If inode bitmap was already initialized there may be some
1377 * used inodes so we need to skip blocks with used inodes in
1380 if (!(gdp
->bg_flags
& cpu_to_le16(EXT4_BG_INODE_UNINIT
)))
1381 used_blks
= DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb
) -
1382 ext4_itable_unused_count(sb
, gdp
)),
1383 sbi
->s_inodes_per_block
);
1385 if ((used_blks
< 0) || (used_blks
> sbi
->s_itb_per_group
)) {
1386 ext4_error(sb
, "Something is wrong with group %u: "
1387 "used itable blocks: %d; "
1388 "itable unused count: %u",
1390 ext4_itable_unused_count(sb
, gdp
));
1395 blk
= ext4_inode_table(sb
, gdp
) + used_blks
;
1396 num
= sbi
->s_itb_per_group
- used_blks
;
1398 BUFFER_TRACE(group_desc_bh
, "get_write_access");
1399 ret
= ext4_journal_get_write_access(handle
,
1405 * Skip zeroout if the inode table is full. But we set the ZEROED
1406 * flag anyway, because obviously, when it is full it does not need
1409 if (unlikely(num
== 0))
1412 ext4_debug("going to zero out inode table in group %d\n",
1414 ret
= sb_issue_zeroout(sb
, blk
, num
, GFP_NOFS
);
1418 blkdev_issue_flush(sb
->s_bdev
, GFP_NOFS
, NULL
);
1421 ext4_lock_group(sb
, group
);
1422 gdp
->bg_flags
|= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
1423 ext4_group_desc_csum_set(sb
, group
, gdp
);
1424 ext4_unlock_group(sb
, group
);
1426 BUFFER_TRACE(group_desc_bh
,
1427 "call ext4_handle_dirty_metadata");
1428 ret
= ext4_handle_dirty_metadata(handle
, NULL
,
1432 up_write(&grp
->alloc_sem
);
1433 ext4_journal_stop(handle
);