1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/namei.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)
12 * linux/fs/minix/namei.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
38 #include "ext4_jbd2.h"
43 #include <trace/events/ext4.h>
45 * define how far ahead to read directories while searching them.
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51 static struct buffer_head
*ext4_append(handle_t
*handle
,
55 struct buffer_head
*bh
;
58 if (unlikely(EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
&&
59 ((inode
->i_size
>> 10) >=
60 EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
)))
61 return ERR_PTR(-ENOSPC
);
63 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
65 bh
= ext4_bread(handle
, inode
, *block
, EXT4_GET_BLOCKS_CREATE
);
68 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
69 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
70 BUFFER_TRACE(bh
, "get_write_access");
71 err
= ext4_journal_get_write_access(handle
, bh
);
74 ext4_std_error(inode
->i_sb
, err
);
80 static int ext4_dx_csum_verify(struct inode
*inode
,
81 struct ext4_dir_entry
*dirent
);
84 * Hints to ext4_read_dirblock regarding whether we expect a directory
85 * block being read to be an index block, or a block containing
86 * directory entries (and if the latter, whether it was found via a
87 * logical block in an htree index block). This is used to control
88 * what sort of sanity checkinig ext4_read_dirblock() will do on the
89 * directory block read from the storage device. EITHER will means
90 * the caller doesn't know what kind of directory block will be read,
91 * so no specific verification will be done.
94 EITHER
, INDEX
, DIRENT
, DIRENT_HTREE
97 #define ext4_read_dirblock(inode, block, type) \
98 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
100 static struct buffer_head
*__ext4_read_dirblock(struct inode
*inode
,
102 dirblock_type_t type
,
106 struct buffer_head
*bh
;
107 struct ext4_dir_entry
*dirent
;
110 bh
= ext4_bread(NULL
, inode
, block
, 0);
112 __ext4_warning(inode
->i_sb
, func
, line
,
113 "inode #%lu: lblock %lu: comm %s: "
114 "error %ld reading directory block",
115 inode
->i_ino
, (unsigned long)block
,
116 current
->comm
, PTR_ERR(bh
));
120 if (!bh
&& (type
== INDEX
|| type
== DIRENT_HTREE
)) {
121 ext4_error_inode(inode
, func
, line
, block
,
122 "Directory hole found for htree %s block",
123 (type
== INDEX
) ? "index" : "leaf");
124 return ERR_PTR(-EFSCORRUPTED
);
128 dirent
= (struct ext4_dir_entry
*) bh
->b_data
;
129 /* Determine whether or not we have an index block */
133 else if (ext4_rec_len_from_disk(dirent
->rec_len
,
134 inode
->i_sb
->s_blocksize
) ==
135 inode
->i_sb
->s_blocksize
)
138 if (!is_dx_block
&& type
== INDEX
) {
139 ext4_error_inode(inode
, func
, line
, block
,
140 "directory leaf block found instead of index block");
142 return ERR_PTR(-EFSCORRUPTED
);
144 if (!ext4_has_metadata_csum(inode
->i_sb
) ||
149 * An empty leaf block can get mistaken for a index block; for
150 * this reason, we can only check the index checksum when the
151 * caller is sure it should be an index block.
153 if (is_dx_block
&& type
== INDEX
) {
154 if (ext4_dx_csum_verify(inode
, dirent
))
155 set_buffer_verified(bh
);
157 ext4_error_inode(inode
, func
, line
, block
,
158 "Directory index failed checksum");
160 return ERR_PTR(-EFSBADCRC
);
164 if (ext4_dirent_csum_verify(inode
, dirent
))
165 set_buffer_verified(bh
);
167 ext4_error_inode(inode
, func
, line
, block
,
168 "Directory block failed checksum");
170 return ERR_PTR(-EFSBADCRC
);
177 #define assert(test) J_ASSERT(test)
181 #define dxtrace(command) command
183 #define dxtrace(command)
207 * dx_root_info is laid out so that if it should somehow get overlaid by a
208 * dirent the two low bits of the hash version will be zero. Therefore, the
209 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
214 struct fake_dirent dot
;
216 struct fake_dirent dotdot
;
220 __le32 reserved_zero
;
222 u8 info_length
; /* 8 */
227 struct dx_entry entries
[0];
232 struct fake_dirent fake
;
233 struct dx_entry entries
[0];
239 struct buffer_head
*bh
;
240 struct dx_entry
*entries
;
252 * This goes at the end of each htree block.
256 __le32 dt_checksum
; /* crc32c(uuid+inum+dirblock) */
259 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
260 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
261 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
262 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
263 static unsigned dx_get_count(struct dx_entry
*entries
);
264 static unsigned dx_get_limit(struct dx_entry
*entries
);
265 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
266 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
267 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
268 static unsigned dx_node_limit(struct inode
*dir
);
269 static struct dx_frame
*dx_probe(struct ext4_filename
*fname
,
271 struct dx_hash_info
*hinfo
,
272 struct dx_frame
*frame
);
273 static void dx_release(struct dx_frame
*frames
);
274 static int dx_make_map(struct inode
*dir
, struct ext4_dir_entry_2
*de
,
275 unsigned blocksize
, struct dx_hash_info
*hinfo
,
276 struct dx_map_entry map
[]);
277 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
278 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
279 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
280 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
281 static void dx_insert_block(struct dx_frame
*frame
,
282 u32 hash
, ext4_lblk_t block
);
283 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
284 struct dx_frame
*frame
,
285 struct dx_frame
*frames
,
287 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
288 struct ext4_filename
*fname
,
289 struct ext4_dir_entry_2
**res_dir
);
290 static int ext4_dx_add_entry(handle_t
*handle
, struct ext4_filename
*fname
,
291 struct inode
*dir
, struct inode
*inode
);
293 /* checksumming functions */
294 void initialize_dirent_tail(struct ext4_dir_entry_tail
*t
,
295 unsigned int blocksize
)
297 memset(t
, 0, sizeof(struct ext4_dir_entry_tail
));
298 t
->det_rec_len
= ext4_rec_len_to_disk(
299 sizeof(struct ext4_dir_entry_tail
), blocksize
);
300 t
->det_reserved_ft
= EXT4_FT_DIR_CSUM
;
303 /* Walk through a dirent block to find a checksum "dirent" at the tail */
304 static struct ext4_dir_entry_tail
*get_dirent_tail(struct inode
*inode
,
305 struct ext4_dir_entry
*de
)
307 struct ext4_dir_entry_tail
*t
;
310 struct ext4_dir_entry
*d
, *top
;
313 top
= (struct ext4_dir_entry
*)(((void *)de
) +
314 (EXT4_BLOCK_SIZE(inode
->i_sb
) -
315 sizeof(struct ext4_dir_entry_tail
)));
316 while (d
< top
&& d
->rec_len
)
317 d
= (struct ext4_dir_entry
*)(((void *)d
) +
318 le16_to_cpu(d
->rec_len
));
323 t
= (struct ext4_dir_entry_tail
*)d
;
325 t
= EXT4_DIRENT_TAIL(de
, EXT4_BLOCK_SIZE(inode
->i_sb
));
328 if (t
->det_reserved_zero1
||
329 le16_to_cpu(t
->det_rec_len
) != sizeof(struct ext4_dir_entry_tail
) ||
330 t
->det_reserved_zero2
||
331 t
->det_reserved_ft
!= EXT4_FT_DIR_CSUM
)
337 static __le32
ext4_dirent_csum(struct inode
*inode
,
338 struct ext4_dir_entry
*dirent
, int size
)
340 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
341 struct ext4_inode_info
*ei
= EXT4_I(inode
);
344 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
345 return cpu_to_le32(csum
);
348 #define warn_no_space_for_csum(inode) \
349 __warn_no_space_for_csum((inode), __func__, __LINE__)
351 static void __warn_no_space_for_csum(struct inode
*inode
, const char *func
,
354 __ext4_warning_inode(inode
, func
, line
,
355 "No space for directory leaf checksum. Please run e2fsck -D.");
358 int ext4_dirent_csum_verify(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
360 struct ext4_dir_entry_tail
*t
;
362 if (!ext4_has_metadata_csum(inode
->i_sb
))
365 t
= get_dirent_tail(inode
, dirent
);
367 warn_no_space_for_csum(inode
);
371 if (t
->det_checksum
!= ext4_dirent_csum(inode
, dirent
,
372 (void *)t
- (void *)dirent
))
378 static void ext4_dirent_csum_set(struct inode
*inode
,
379 struct ext4_dir_entry
*dirent
)
381 struct ext4_dir_entry_tail
*t
;
383 if (!ext4_has_metadata_csum(inode
->i_sb
))
386 t
= get_dirent_tail(inode
, dirent
);
388 warn_no_space_for_csum(inode
);
392 t
->det_checksum
= ext4_dirent_csum(inode
, dirent
,
393 (void *)t
- (void *)dirent
);
396 int ext4_handle_dirty_dirent_node(handle_t
*handle
,
398 struct buffer_head
*bh
)
400 ext4_dirent_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
401 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
404 static struct dx_countlimit
*get_dx_countlimit(struct inode
*inode
,
405 struct ext4_dir_entry
*dirent
,
408 struct ext4_dir_entry
*dp
;
409 struct dx_root_info
*root
;
412 if (le16_to_cpu(dirent
->rec_len
) == EXT4_BLOCK_SIZE(inode
->i_sb
))
414 else if (le16_to_cpu(dirent
->rec_len
) == 12) {
415 dp
= (struct ext4_dir_entry
*)(((void *)dirent
) + 12);
416 if (le16_to_cpu(dp
->rec_len
) !=
417 EXT4_BLOCK_SIZE(inode
->i_sb
) - 12)
419 root
= (struct dx_root_info
*)(((void *)dp
+ 12));
420 if (root
->reserved_zero
||
421 root
->info_length
!= sizeof(struct dx_root_info
))
428 *offset
= count_offset
;
429 return (struct dx_countlimit
*)(((void *)dirent
) + count_offset
);
432 static __le32
ext4_dx_csum(struct inode
*inode
, struct ext4_dir_entry
*dirent
,
433 int count_offset
, int count
, struct dx_tail
*t
)
435 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
436 struct ext4_inode_info
*ei
= EXT4_I(inode
);
439 __u32 dummy_csum
= 0;
440 int offset
= offsetof(struct dx_tail
, dt_checksum
);
442 size
= count_offset
+ (count
* sizeof(struct dx_entry
));
443 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
444 csum
= ext4_chksum(sbi
, csum
, (__u8
*)t
, offset
);
445 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, sizeof(dummy_csum
));
447 return cpu_to_le32(csum
);
450 static int ext4_dx_csum_verify(struct inode
*inode
,
451 struct ext4_dir_entry
*dirent
)
453 struct dx_countlimit
*c
;
455 int count_offset
, limit
, count
;
457 if (!ext4_has_metadata_csum(inode
->i_sb
))
460 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
462 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
465 limit
= le16_to_cpu(c
->limit
);
466 count
= le16_to_cpu(c
->count
);
467 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
468 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
469 warn_no_space_for_csum(inode
);
472 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
474 if (t
->dt_checksum
!= ext4_dx_csum(inode
, dirent
, count_offset
,
480 static void ext4_dx_csum_set(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
482 struct dx_countlimit
*c
;
484 int count_offset
, limit
, count
;
486 if (!ext4_has_metadata_csum(inode
->i_sb
))
489 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
491 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
494 limit
= le16_to_cpu(c
->limit
);
495 count
= le16_to_cpu(c
->count
);
496 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
497 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
498 warn_no_space_for_csum(inode
);
501 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
503 t
->dt_checksum
= ext4_dx_csum(inode
, dirent
, count_offset
, count
, t
);
506 static inline int ext4_handle_dirty_dx_node(handle_t
*handle
,
508 struct buffer_head
*bh
)
510 ext4_dx_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
511 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
515 * p is at least 6 bytes before the end of page
517 static inline struct ext4_dir_entry_2
*
518 ext4_next_entry(struct ext4_dir_entry_2
*p
, unsigned long blocksize
)
520 return (struct ext4_dir_entry_2
*)((char *)p
+
521 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
525 * Future: use high four bits of block for coalesce-on-delete flags
526 * Mask them off for now.
529 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
531 return le32_to_cpu(entry
->block
) & 0x0fffffff;
534 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
536 entry
->block
= cpu_to_le32(value
);
539 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
541 return le32_to_cpu(entry
->hash
);
544 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
546 entry
->hash
= cpu_to_le32(value
);
549 static inline unsigned dx_get_count(struct dx_entry
*entries
)
551 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
554 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
556 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
559 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
561 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
564 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
566 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
569 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
571 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
572 EXT4_DIR_REC_LEN(2) - infosize
;
574 if (ext4_has_metadata_csum(dir
->i_sb
))
575 entry_space
-= sizeof(struct dx_tail
);
576 return entry_space
/ sizeof(struct dx_entry
);
579 static inline unsigned dx_node_limit(struct inode
*dir
)
581 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
583 if (ext4_has_metadata_csum(dir
->i_sb
))
584 entry_space
-= sizeof(struct dx_tail
);
585 return entry_space
/ sizeof(struct dx_entry
);
592 static void dx_show_index(char * label
, struct dx_entry
*entries
)
594 int i
, n
= dx_get_count (entries
);
595 printk(KERN_DEBUG
"%s index", label
);
596 for (i
= 0; i
< n
; i
++) {
597 printk(KERN_CONT
" %x->%lu",
598 i
? dx_get_hash(entries
+ i
) : 0,
599 (unsigned long)dx_get_block(entries
+ i
));
601 printk(KERN_CONT
"\n");
611 static struct stats
dx_show_leaf(struct inode
*dir
,
612 struct dx_hash_info
*hinfo
,
613 struct ext4_dir_entry_2
*de
,
614 int size
, int show_names
)
616 unsigned names
= 0, space
= 0;
617 char *base
= (char *) de
;
618 struct dx_hash_info h
= *hinfo
;
621 while ((char *) de
< base
+ size
)
627 #ifdef CONFIG_EXT4_FS_ENCRYPTION
630 struct fscrypt_str fname_crypto_str
=
636 if (ext4_encrypted_inode(dir
))
637 res
= fscrypt_get_encryption_info(dir
);
639 printk(KERN_WARNING
"Error setting up"
640 " fname crypto: %d\n", res
);
642 if (!fscrypt_has_encryption_key(dir
)) {
643 /* Directory is not encrypted */
644 ext4fs_dirhash(de
->name
,
646 printk("%*.s:(U)%x.%u ", len
,
648 (unsigned) ((char *) de
651 struct fscrypt_str de_name
=
652 FSTR_INIT(name
, len
);
654 /* Directory is encrypted */
655 res
= fscrypt_fname_alloc_buffer(
659 printk(KERN_WARNING
"Error "
663 res
= fscrypt_fname_disk_to_usr(dir
,
667 printk(KERN_WARNING
"Error "
668 "converting filename "
674 name
= fname_crypto_str
.name
;
675 len
= fname_crypto_str
.len
;
677 ext4fs_dirhash(de
->name
, de
->name_len
,
679 printk("%*.s:(E)%x.%u ", len
, name
,
680 h
.hash
, (unsigned) ((char *) de
682 fscrypt_fname_free_buffer(
686 int len
= de
->name_len
;
687 char *name
= de
->name
;
688 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
689 printk("%*.s:%x.%u ", len
, name
, h
.hash
,
690 (unsigned) ((char *) de
- base
));
693 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
696 de
= ext4_next_entry(de
, size
);
698 printk(KERN_CONT
"(%i)\n", names
);
699 return (struct stats
) { names
, space
, 1 };
702 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
703 struct dx_entry
*entries
, int levels
)
705 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
706 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
708 struct buffer_head
*bh
;
709 printk("%i indexed blocks...\n", count
);
710 for (i
= 0; i
< count
; i
++, entries
++)
712 ext4_lblk_t block
= dx_get_block(entries
);
713 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
714 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
716 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
717 bh
= ext4_bread(NULL
,dir
, block
, 0);
718 if (!bh
|| IS_ERR(bh
))
721 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
722 dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*)
723 bh
->b_data
, blocksize
, 0);
724 names
+= stats
.names
;
725 space
+= stats
.space
;
726 bcount
+= stats
.bcount
;
730 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
731 levels
? "" : " ", names
, space
/bcount
,
732 (space
/bcount
)*100/blocksize
);
733 return (struct stats
) { names
, space
, bcount
};
735 #endif /* DX_DEBUG */
738 * Probe for a directory leaf block to search.
740 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
741 * error in the directory index, and the caller should fall back to
742 * searching the directory normally. The callers of dx_probe **MUST**
743 * check for this error code, and make sure it never gets reflected
746 static struct dx_frame
*
747 dx_probe(struct ext4_filename
*fname
, struct inode
*dir
,
748 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
)
750 unsigned count
, indirect
;
751 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
752 struct dx_root
*root
;
753 struct dx_frame
*frame
= frame_in
;
754 struct dx_frame
*ret_err
= ERR_PTR(ERR_BAD_DX_DIR
);
757 memset(frame_in
, 0, EXT4_HTREE_LEVEL
* sizeof(frame_in
[0]));
758 frame
->bh
= ext4_read_dirblock(dir
, 0, INDEX
);
759 if (IS_ERR(frame
->bh
))
760 return (struct dx_frame
*) frame
->bh
;
762 root
= (struct dx_root
*) frame
->bh
->b_data
;
763 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
764 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
765 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
766 ext4_warning_inode(dir
, "Unrecognised inode hash code %u",
767 root
->info
.hash_version
);
771 hinfo
= &fname
->hinfo
;
772 hinfo
->hash_version
= root
->info
.hash_version
;
773 if (hinfo
->hash_version
<= DX_HASH_TEA
)
774 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
775 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
776 if (fname
&& fname_name(fname
))
777 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), hinfo
);
780 if (root
->info
.unused_flags
& 1) {
781 ext4_warning_inode(dir
, "Unimplemented hash flags: %#06x",
782 root
->info
.unused_flags
);
786 indirect
= root
->info
.indirect_levels
;
787 if (indirect
>= ext4_dir_htree_level(dir
->i_sb
)) {
788 ext4_warning(dir
->i_sb
,
789 "Directory (ino: %lu) htree depth %#06x exceed"
790 "supported value", dir
->i_ino
,
791 ext4_dir_htree_level(dir
->i_sb
));
792 if (ext4_dir_htree_level(dir
->i_sb
) < EXT4_HTREE_LEVEL
) {
793 ext4_warning(dir
->i_sb
, "Enable large directory "
794 "feature to access it");
799 entries
= (struct dx_entry
*)(((char *)&root
->info
) +
800 root
->info
.info_length
);
802 if (dx_get_limit(entries
) != dx_root_limit(dir
,
803 root
->info
.info_length
)) {
804 ext4_warning_inode(dir
, "dx entry: limit %u != root limit %u",
805 dx_get_limit(entries
),
806 dx_root_limit(dir
, root
->info
.info_length
));
810 dxtrace(printk("Look up %x", hash
));
812 count
= dx_get_count(entries
);
813 if (!count
|| count
> dx_get_limit(entries
)) {
814 ext4_warning_inode(dir
,
815 "dx entry: count %u beyond limit %u",
816 count
, dx_get_limit(entries
));
821 q
= entries
+ count
- 1;
824 dxtrace(printk(KERN_CONT
"."));
825 if (dx_get_hash(m
) > hash
)
831 if (0) { // linear search cross check
832 unsigned n
= count
- 1;
836 dxtrace(printk(KERN_CONT
","));
837 if (dx_get_hash(++at
) > hash
)
843 assert (at
== p
- 1);
847 dxtrace(printk(KERN_CONT
" %x->%u\n",
848 at
== entries
? 0 : dx_get_hash(at
),
850 frame
->entries
= entries
;
855 frame
->bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
856 if (IS_ERR(frame
->bh
)) {
857 ret_err
= (struct dx_frame
*) frame
->bh
;
861 entries
= ((struct dx_node
*) frame
->bh
->b_data
)->entries
;
863 if (dx_get_limit(entries
) != dx_node_limit(dir
)) {
864 ext4_warning_inode(dir
,
865 "dx entry: limit %u != node limit %u",
866 dx_get_limit(entries
), dx_node_limit(dir
));
871 while (frame
>= frame_in
) {
876 if (ret_err
== ERR_PTR(ERR_BAD_DX_DIR
))
877 ext4_warning_inode(dir
,
878 "Corrupt directory, running e2fsck is recommended");
882 static void dx_release(struct dx_frame
*frames
)
884 struct dx_root_info
*info
;
886 unsigned int indirect_levels
;
888 if (frames
[0].bh
== NULL
)
891 info
= &((struct dx_root
*)frames
[0].bh
->b_data
)->info
;
892 /* save local copy, "info" may be freed after brelse() */
893 indirect_levels
= info
->indirect_levels
;
894 for (i
= 0; i
<= indirect_levels
; i
++) {
895 if (frames
[i
].bh
== NULL
)
897 brelse(frames
[i
].bh
);
903 * This function increments the frame pointer to search the next leaf
904 * block, and reads in the necessary intervening nodes if the search
905 * should be necessary. Whether or not the search is necessary is
906 * controlled by the hash parameter. If the hash value is even, then
907 * the search is only continued if the next block starts with that
908 * hash value. This is used if we are searching for a specific file.
910 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
912 * This function returns 1 if the caller should continue to search,
913 * or 0 if it should not. If there is an error reading one of the
914 * index blocks, it will a negative error code.
916 * If start_hash is non-null, it will be filled in with the starting
917 * hash of the next page.
919 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
920 struct dx_frame
*frame
,
921 struct dx_frame
*frames
,
925 struct buffer_head
*bh
;
931 * Find the next leaf page by incrementing the frame pointer.
932 * If we run out of entries in the interior node, loop around and
933 * increment pointer in the parent node. When we break out of
934 * this loop, num_frames indicates the number of interior
935 * nodes need to be read.
938 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
947 * If the hash is 1, then continue only if the next page has a
948 * continuation hash of any value. This is used for readdir
949 * handling. Otherwise, check to see if the hash matches the
950 * desired contiuation hash. If it doesn't, return since
951 * there's no point to read in the successive index pages.
953 bhash
= dx_get_hash(p
->at
);
956 if ((hash
& 1) == 0) {
957 if ((bhash
& ~1) != hash
)
961 * If the hash is HASH_NB_ALWAYS, we always go to the next
962 * block so no check is necessary
964 while (num_frames
--) {
965 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
971 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
978 * This function fills a red-black tree with information from a
979 * directory block. It returns the number directory entries loaded
980 * into the tree. If there is an error it is returned in err.
982 static int htree_dirblock_to_tree(struct file
*dir_file
,
983 struct inode
*dir
, ext4_lblk_t block
,
984 struct dx_hash_info
*hinfo
,
985 __u32 start_hash
, __u32 start_minor_hash
)
987 struct buffer_head
*bh
;
988 struct ext4_dir_entry_2
*de
, *top
;
989 int err
= 0, count
= 0;
990 struct fscrypt_str fname_crypto_str
= FSTR_INIT(NULL
, 0), tmp_str
;
992 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
993 (unsigned long)block
));
994 bh
= ext4_read_dirblock(dir
, block
, DIRENT_HTREE
);
998 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
999 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
1000 dir
->i_sb
->s_blocksize
-
1001 EXT4_DIR_REC_LEN(0));
1002 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1003 /* Check if the directory is encrypted */
1004 if (ext4_encrypted_inode(dir
)) {
1005 err
= fscrypt_get_encryption_info(dir
);
1010 err
= fscrypt_fname_alloc_buffer(dir
, EXT4_NAME_LEN
,
1018 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
1019 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1020 bh
->b_data
, bh
->b_size
,
1021 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
1022 + ((char *)de
- bh
->b_data
))) {
1023 /* silently ignore the rest of the block */
1026 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
1027 if ((hinfo
->hash
< start_hash
) ||
1028 ((hinfo
->hash
== start_hash
) &&
1029 (hinfo
->minor_hash
< start_minor_hash
)))
1033 if (!ext4_encrypted_inode(dir
)) {
1034 tmp_str
.name
= de
->name
;
1035 tmp_str
.len
= de
->name_len
;
1036 err
= ext4_htree_store_dirent(dir_file
,
1037 hinfo
->hash
, hinfo
->minor_hash
, de
,
1040 int save_len
= fname_crypto_str
.len
;
1041 struct fscrypt_str de_name
= FSTR_INIT(de
->name
,
1044 /* Directory is encrypted */
1045 err
= fscrypt_fname_disk_to_usr(dir
, hinfo
->hash
,
1046 hinfo
->minor_hash
, &de_name
,
1052 err
= ext4_htree_store_dirent(dir_file
,
1053 hinfo
->hash
, hinfo
->minor_hash
, de
,
1055 fname_crypto_str
.len
= save_len
;
1065 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1066 fscrypt_fname_free_buffer(&fname_crypto_str
);
1073 * This function fills a red-black tree with information from a
1074 * directory. We start scanning the directory in hash order, starting
1075 * at start_hash and start_minor_hash.
1077 * This function returns the number of entries inserted into the tree,
1078 * or a negative error code.
1080 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
1081 __u32 start_minor_hash
, __u32
*next_hash
)
1083 struct dx_hash_info hinfo
;
1084 struct ext4_dir_entry_2
*de
;
1085 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
1091 struct fscrypt_str tmp_str
;
1093 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
1094 start_hash
, start_minor_hash
));
1095 dir
= file_inode(dir_file
);
1096 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
1097 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1098 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1099 hinfo
.hash_version
+=
1100 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1101 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1102 if (ext4_has_inline_data(dir
)) {
1103 int has_inline_data
= 1;
1104 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
1108 if (has_inline_data
) {
1113 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
1114 start_hash
, start_minor_hash
);
1118 hinfo
.hash
= start_hash
;
1119 hinfo
.minor_hash
= 0;
1120 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
);
1122 return PTR_ERR(frame
);
1124 /* Add '.' and '..' from the htree header */
1125 if (!start_hash
&& !start_minor_hash
) {
1126 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1127 tmp_str
.name
= de
->name
;
1128 tmp_str
.len
= de
->name_len
;
1129 err
= ext4_htree_store_dirent(dir_file
, 0, 0,
1135 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
1136 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1137 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
1138 tmp_str
.name
= de
->name
;
1139 tmp_str
.len
= de
->name_len
;
1140 err
= ext4_htree_store_dirent(dir_file
, 2, 0,
1148 if (fatal_signal_pending(current
)) {
1153 block
= dx_get_block(frame
->at
);
1154 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
1155 start_hash
, start_minor_hash
);
1162 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
1163 frame
, frames
, &hashval
);
1164 *next_hash
= hashval
;
1170 * Stop if: (a) there are no more entries, or
1171 * (b) we have inserted at least one entry and the
1172 * next hash value is not a continuation
1175 (count
&& ((hashval
& 1) == 0)))
1179 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1180 "next hash: %x\n", count
, *next_hash
));
1187 static inline int search_dirblock(struct buffer_head
*bh
,
1189 struct ext4_filename
*fname
,
1190 unsigned int offset
,
1191 struct ext4_dir_entry_2
**res_dir
)
1193 return ext4_search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1194 fname
, offset
, res_dir
);
1198 * Directory block splitting, compacting
1202 * Create map of hash values, offsets, and sizes, stored at end of block.
1203 * Returns number of entries mapped.
1205 static int dx_make_map(struct inode
*dir
, struct ext4_dir_entry_2
*de
,
1206 unsigned blocksize
, struct dx_hash_info
*hinfo
,
1207 struct dx_map_entry
*map_tail
)
1210 char *base
= (char *) de
;
1211 struct dx_hash_info h
= *hinfo
;
1213 while ((char *) de
< base
+ blocksize
) {
1214 if (de
->name_len
&& de
->inode
) {
1215 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1217 map_tail
->hash
= h
.hash
;
1218 map_tail
->offs
= ((char *) de
- base
)>>2;
1219 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1223 /* XXX: do we need to check rec_len == 0 case? -Chris */
1224 de
= ext4_next_entry(de
, blocksize
);
1229 /* Sort map by hash value */
1230 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1232 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1234 /* Combsort until bubble sort doesn't suck */
1236 count
= count
*10/13;
1237 if (count
- 9 < 2) /* 9, 10 -> 11 */
1239 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1240 if (p
->hash
< q
->hash
)
1243 /* Garden variety bubble sort */
1248 if (q
[1].hash
>= q
[0].hash
)
1256 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1258 struct dx_entry
*entries
= frame
->entries
;
1259 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1260 int count
= dx_get_count(entries
);
1262 assert(count
< dx_get_limit(entries
));
1263 assert(old
< entries
+ count
);
1264 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1265 dx_set_hash(new, hash
);
1266 dx_set_block(new, block
);
1267 dx_set_count(entries
, count
+ 1);
1271 * Test whether a directory entry matches the filename being searched for.
1273 * Return: %true if the directory entry matches, otherwise %false.
1275 static inline bool ext4_match(const struct ext4_filename
*fname
,
1276 const struct ext4_dir_entry_2
*de
)
1278 struct fscrypt_name f
;
1283 f
.usr_fname
= fname
->usr_fname
;
1284 f
.disk_name
= fname
->disk_name
;
1285 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1286 f
.crypto_buf
= fname
->crypto_buf
;
1288 return fscrypt_match_name(&f
, de
->name
, de
->name_len
);
1292 * Returns 0 if not found, -1 on failure, and 1 on success
1294 int ext4_search_dir(struct buffer_head
*bh
, char *search_buf
, int buf_size
,
1295 struct inode
*dir
, struct ext4_filename
*fname
,
1296 unsigned int offset
, struct ext4_dir_entry_2
**res_dir
)
1298 struct ext4_dir_entry_2
* de
;
1302 de
= (struct ext4_dir_entry_2
*)search_buf
;
1303 dlimit
= search_buf
+ buf_size
;
1304 while ((char *) de
< dlimit
) {
1305 /* this code is executed quadratically often */
1306 /* do minimal checking `by hand' */
1307 if ((char *) de
+ de
->name_len
<= dlimit
&&
1308 ext4_match(fname
, de
)) {
1309 /* found a match - just to be sure, do
1311 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, bh
->b_data
,
1312 bh
->b_size
, offset
))
1317 /* prevent looping on a bad block */
1318 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1319 dir
->i_sb
->s_blocksize
);
1323 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1328 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1329 struct ext4_dir_entry
*de
)
1331 struct super_block
*sb
= dir
->i_sb
;
1337 if (de
->inode
== 0 &&
1338 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1347 * finds an entry in the specified directory with the wanted name. It
1348 * returns the cache buffer in which the entry was found, and the entry
1349 * itself (as a parameter - res_dir). It does NOT read the inode of the
1350 * entry - you'll have to do that yourself if you want to.
1352 * The returned buffer_head has ->b_count elevated. The caller is expected
1353 * to brelse() it when appropriate.
1355 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1356 const struct qstr
*d_name
,
1357 struct ext4_dir_entry_2
**res_dir
,
1360 struct super_block
*sb
;
1361 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1362 struct buffer_head
*bh
, *ret
= NULL
;
1363 ext4_lblk_t start
, block
;
1364 const u8
*name
= d_name
->name
;
1365 size_t ra_max
= 0; /* Number of bh's in the readahead
1367 size_t ra_ptr
= 0; /* Current index into readahead
1369 ext4_lblk_t nblocks
;
1370 int i
, namelen
, retval
;
1371 struct ext4_filename fname
;
1375 namelen
= d_name
->len
;
1376 if (namelen
> EXT4_NAME_LEN
)
1379 retval
= ext4_fname_setup_filename(dir
, d_name
, 1, &fname
);
1380 if (retval
== -ENOENT
)
1383 return ERR_PTR(retval
);
1385 if (ext4_has_inline_data(dir
)) {
1386 int has_inline_data
= 1;
1387 ret
= ext4_find_inline_entry(dir
, &fname
, res_dir
,
1389 if (has_inline_data
) {
1392 goto cleanup_and_exit
;
1396 if ((namelen
<= 2) && (name
[0] == '.') &&
1397 (name
[1] == '.' || name
[1] == '\0')) {
1399 * "." or ".." will only be in the first block
1400 * NFS may look up ".."; "." should be handled by the VFS
1407 ret
= ext4_dx_find_entry(dir
, &fname
, res_dir
);
1409 * On success, or if the error was file not found,
1410 * return. Otherwise, fall back to doing a search the
1411 * old fashioned way.
1413 if (!IS_ERR(ret
) || PTR_ERR(ret
) != ERR_BAD_DX_DIR
)
1414 goto cleanup_and_exit
;
1415 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1419 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1422 goto cleanup_and_exit
;
1424 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1425 if (start
>= nblocks
)
1431 * We deal with the read-ahead logic here.
1433 if (ra_ptr
>= ra_max
) {
1434 /* Refill the readahead buffer */
1437 ra_max
= start
- block
;
1439 ra_max
= nblocks
- block
;
1440 ra_max
= min(ra_max
, ARRAY_SIZE(bh_use
));
1441 retval
= ext4_bread_batch(dir
, block
, ra_max
,
1442 false /* wait */, bh_use
);
1444 ret
= ERR_PTR(retval
);
1446 goto cleanup_and_exit
;
1449 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1452 if (!buffer_uptodate(bh
)) {
1453 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1454 (unsigned long) block
);
1456 ret
= ERR_PTR(-EIO
);
1457 goto cleanup_and_exit
;
1459 if (!buffer_verified(bh
) &&
1460 !is_dx_internal_node(dir
, block
,
1461 (struct ext4_dir_entry
*)bh
->b_data
) &&
1462 !ext4_dirent_csum_verify(dir
,
1463 (struct ext4_dir_entry
*)bh
->b_data
)) {
1464 EXT4_ERROR_INODE(dir
, "checksumming directory "
1465 "block %lu", (unsigned long)block
);
1467 ret
= ERR_PTR(-EFSBADCRC
);
1468 goto cleanup_and_exit
;
1470 set_buffer_verified(bh
);
1471 i
= search_dirblock(bh
, dir
, &fname
,
1472 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1474 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1476 goto cleanup_and_exit
;
1480 goto cleanup_and_exit
;
1483 if (++block
>= nblocks
)
1485 } while (block
!= start
);
1488 * If the directory has grown while we were searching, then
1489 * search the last part of the directory before giving up.
1492 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1493 if (block
< nblocks
) {
1499 /* Clean up the read-ahead blocks */
1500 for (; ra_ptr
< ra_max
; ra_ptr
++)
1501 brelse(bh_use
[ra_ptr
]);
1502 ext4_fname_free_filename(&fname
);
1506 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
1507 struct ext4_filename
*fname
,
1508 struct ext4_dir_entry_2
**res_dir
)
1510 struct super_block
* sb
= dir
->i_sb
;
1511 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
1512 struct buffer_head
*bh
;
1516 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1519 frame
= dx_probe(fname
, dir
, NULL
, frames
);
1521 return (struct buffer_head
*) frame
;
1523 block
= dx_get_block(frame
->at
);
1524 bh
= ext4_read_dirblock(dir
, block
, DIRENT_HTREE
);
1528 retval
= search_dirblock(bh
, dir
, fname
,
1529 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1535 bh
= ERR_PTR(ERR_BAD_DX_DIR
);
1539 /* Check to see if we should continue to search */
1540 retval
= ext4_htree_next_block(dir
, fname
->hinfo
.hash
, frame
,
1543 ext4_warning_inode(dir
,
1544 "error %d reading directory index block",
1546 bh
= ERR_PTR(retval
);
1549 } while (retval
== 1);
1553 dxtrace(printk(KERN_DEBUG
"%s not found\n", fname
->usr_fname
->name
));
1559 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1561 struct inode
*inode
;
1562 struct ext4_dir_entry_2
*de
;
1563 struct buffer_head
*bh
;
1566 err
= fscrypt_prepare_lookup(dir
, dentry
, flags
);
1568 return ERR_PTR(err
);
1570 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1571 return ERR_PTR(-ENAMETOOLONG
);
1573 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1575 return (struct dentry
*) bh
;
1578 __u32 ino
= le32_to_cpu(de
->inode
);
1580 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1581 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1582 return ERR_PTR(-EFSCORRUPTED
);
1584 if (unlikely(ino
== dir
->i_ino
)) {
1585 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1587 return ERR_PTR(-EFSCORRUPTED
);
1589 inode
= ext4_iget(dir
->i_sb
, ino
, EXT4_IGET_NORMAL
);
1590 if (inode
== ERR_PTR(-ESTALE
)) {
1591 EXT4_ERROR_INODE(dir
,
1592 "deleted inode referenced: %u",
1594 return ERR_PTR(-EFSCORRUPTED
);
1596 if (!IS_ERR(inode
) && ext4_encrypted_inode(dir
) &&
1597 (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
)) &&
1598 !fscrypt_has_permitted_context(dir
, inode
)) {
1599 ext4_warning(inode
->i_sb
,
1600 "Inconsistent encryption contexts: %lu/%lu",
1601 dir
->i_ino
, inode
->i_ino
);
1603 return ERR_PTR(-EPERM
);
1606 return d_splice_alias(inode
, dentry
);
1610 struct dentry
*ext4_get_parent(struct dentry
*child
)
1613 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1614 struct ext4_dir_entry_2
* de
;
1615 struct buffer_head
*bh
;
1617 bh
= ext4_find_entry(d_inode(child
), &dotdot
, &de
, NULL
);
1619 return (struct dentry
*) bh
;
1621 return ERR_PTR(-ENOENT
);
1622 ino
= le32_to_cpu(de
->inode
);
1625 if (!ext4_valid_inum(child
->d_sb
, ino
)) {
1626 EXT4_ERROR_INODE(d_inode(child
),
1627 "bad parent inode number: %u", ino
);
1628 return ERR_PTR(-EFSCORRUPTED
);
1631 return d_obtain_alias(ext4_iget(child
->d_sb
, ino
, EXT4_IGET_NORMAL
));
1635 * Move count entries from end of map between two memory locations.
1636 * Returns pointer to last entry moved.
1638 static struct ext4_dir_entry_2
*
1639 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1642 unsigned rec_len
= 0;
1645 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1646 (from
+ (map
->offs
<<2));
1647 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1648 memcpy (to
, de
, rec_len
);
1649 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1650 ext4_rec_len_to_disk(rec_len
, blocksize
);
1655 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1659 * Compact each dir entry in the range to the minimal rec_len.
1660 * Returns pointer to last entry in range.
1662 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1664 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1665 unsigned rec_len
= 0;
1668 while ((char*)de
< base
+ blocksize
) {
1669 next
= ext4_next_entry(de
, blocksize
);
1670 if (de
->inode
&& de
->name_len
) {
1671 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1673 memmove(to
, de
, rec_len
);
1674 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1676 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1684 * Split a full leaf block to make room for a new dir entry.
1685 * Allocate a new block, and move entries so that they are approx. equally full.
1686 * Returns pointer to de in block into which the new entry will be inserted.
1688 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1689 struct buffer_head
**bh
,struct dx_frame
*frame
,
1690 struct dx_hash_info
*hinfo
)
1692 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1693 unsigned count
, continued
;
1694 struct buffer_head
*bh2
;
1695 ext4_lblk_t newblock
;
1697 struct dx_map_entry
*map
;
1698 char *data1
= (*bh
)->b_data
, *data2
;
1699 unsigned split
, move
, size
;
1700 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1701 struct ext4_dir_entry_tail
*t
;
1705 if (ext4_has_metadata_csum(dir
->i_sb
))
1706 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1708 bh2
= ext4_append(handle
, dir
, &newblock
);
1712 return (struct ext4_dir_entry_2
*) bh2
;
1715 BUFFER_TRACE(*bh
, "get_write_access");
1716 err
= ext4_journal_get_write_access(handle
, *bh
);
1720 BUFFER_TRACE(frame
->bh
, "get_write_access");
1721 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1725 data2
= bh2
->b_data
;
1727 /* create map in the end of data2 block */
1728 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1729 count
= dx_make_map(dir
, (struct ext4_dir_entry_2
*) data1
,
1730 blocksize
, hinfo
, map
);
1732 dx_sort_map(map
, count
);
1733 /* Split the existing block in the middle, size-wise */
1736 for (i
= count
-1; i
>= 0; i
--) {
1737 /* is more than half of this entry in 2nd half of the block? */
1738 if (size
+ map
[i
].size
/2 > blocksize
/2)
1740 size
+= map
[i
].size
;
1743 /* map index at which we will split */
1744 split
= count
- move
;
1745 hash2
= map
[split
].hash
;
1746 continued
= hash2
== map
[split
- 1].hash
;
1747 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1748 (unsigned long)dx_get_block(frame
->at
),
1749 hash2
, split
, count
-split
));
1751 /* Fancy dance to stay within two buffers */
1752 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
,
1754 de
= dx_pack_dirents(data1
, blocksize
);
1755 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1758 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1762 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1763 initialize_dirent_tail(t
, blocksize
);
1765 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1766 initialize_dirent_tail(t
, blocksize
);
1769 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data1
,
1771 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data2
,
1774 /* Which block gets the new entry? */
1775 if (hinfo
->hash
>= hash2
) {
1779 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1780 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1783 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1787 dxtrace(dx_show_index("frame", frame
->entries
));
1794 ext4_std_error(dir
->i_sb
, err
);
1795 return ERR_PTR(err
);
1798 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1799 struct buffer_head
*bh
,
1800 void *buf
, int buf_size
,
1801 struct ext4_filename
*fname
,
1802 struct ext4_dir_entry_2
**dest_de
)
1804 struct ext4_dir_entry_2
*de
;
1805 unsigned short reclen
= EXT4_DIR_REC_LEN(fname_len(fname
));
1807 unsigned int offset
= 0;
1810 de
= (struct ext4_dir_entry_2
*)buf
;
1811 top
= buf
+ buf_size
- reclen
;
1812 while ((char *) de
<= top
) {
1813 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1814 buf
, buf_size
, offset
))
1815 return -EFSCORRUPTED
;
1816 if (ext4_match(fname
, de
))
1818 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1819 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1820 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1822 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1825 if ((char *) de
> top
)
1832 void ext4_insert_dentry(struct inode
*inode
,
1833 struct ext4_dir_entry_2
*de
,
1835 struct ext4_filename
*fname
)
1840 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1841 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1843 struct ext4_dir_entry_2
*de1
=
1844 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1845 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1846 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1849 de
->file_type
= EXT4_FT_UNKNOWN
;
1850 de
->inode
= cpu_to_le32(inode
->i_ino
);
1851 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1852 de
->name_len
= fname_len(fname
);
1853 memcpy(de
->name
, fname_name(fname
), fname_len(fname
));
1857 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1858 * it points to a directory entry which is guaranteed to be large
1859 * enough for new directory entry. If de is NULL, then
1860 * add_dirent_to_buf will attempt search the directory block for
1861 * space. It will return -ENOSPC if no space is available, and -EIO
1862 * and -EEXIST if directory entry already exists.
1864 static int add_dirent_to_buf(handle_t
*handle
, struct ext4_filename
*fname
,
1866 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1867 struct buffer_head
*bh
)
1869 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1873 if (ext4_has_metadata_csum(inode
->i_sb
))
1874 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1877 err
= ext4_find_dest_de(dir
, inode
, bh
, bh
->b_data
,
1878 blocksize
- csum_size
, fname
, &de
);
1882 BUFFER_TRACE(bh
, "get_write_access");
1883 err
= ext4_journal_get_write_access(handle
, bh
);
1885 ext4_std_error(dir
->i_sb
, err
);
1889 /* By now the buffer is marked for journaling */
1890 ext4_insert_dentry(inode
, de
, blocksize
, fname
);
1893 * XXX shouldn't update any times until successful
1894 * completion of syscall, but too many callers depend
1897 * XXX similarly, too many callers depend on
1898 * ext4_new_inode() setting the times, but error
1899 * recovery deletes the inode, so the worst that can
1900 * happen is that the times are slightly out of date
1901 * and/or different from the directory change time.
1903 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
1904 ext4_update_dx_flag(dir
);
1905 inode_inc_iversion(dir
);
1906 ext4_mark_inode_dirty(handle
, dir
);
1907 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1908 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1910 ext4_std_error(dir
->i_sb
, err
);
1915 * This converts a one block unindexed directory to a 3 block indexed
1916 * directory, and adds the dentry to the indexed directory.
1918 static int make_indexed_dir(handle_t
*handle
, struct ext4_filename
*fname
,
1920 struct inode
*inode
, struct buffer_head
*bh
)
1922 struct buffer_head
*bh2
;
1923 struct dx_root
*root
;
1924 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
1925 struct dx_entry
*entries
;
1926 struct ext4_dir_entry_2
*de
, *de2
;
1927 struct ext4_dir_entry_tail
*t
;
1933 struct fake_dirent
*fde
;
1936 if (ext4_has_metadata_csum(inode
->i_sb
))
1937 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1939 blocksize
= dir
->i_sb
->s_blocksize
;
1940 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1941 BUFFER_TRACE(bh
, "get_write_access");
1942 retval
= ext4_journal_get_write_access(handle
, bh
);
1944 ext4_std_error(dir
->i_sb
, retval
);
1948 root
= (struct dx_root
*) bh
->b_data
;
1950 /* The 0th block becomes the root, move the dirents out */
1951 fde
= &root
->dotdot
;
1952 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1953 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1954 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1955 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1957 return -EFSCORRUPTED
;
1959 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1961 /* Allocate new block for the 0th block's dirents */
1962 bh2
= ext4_append(handle
, dir
, &block
);
1965 return PTR_ERR(bh2
);
1967 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1968 data1
= bh2
->b_data
;
1970 memcpy (data1
, de
, len
);
1971 de
= (struct ext4_dir_entry_2
*) data1
;
1973 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1975 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1980 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1981 initialize_dirent_tail(t
, blocksize
);
1984 /* Initialize the root; the dot dirents already exist */
1985 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1986 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1988 memset (&root
->info
, 0, sizeof(root
->info
));
1989 root
->info
.info_length
= sizeof(root
->info
);
1990 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1991 entries
= root
->entries
;
1992 dx_set_block(entries
, 1);
1993 dx_set_count(entries
, 1);
1994 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1996 /* Initialize as for dx_probe */
1997 fname
->hinfo
.hash_version
= root
->info
.hash_version
;
1998 if (fname
->hinfo
.hash_version
<= DX_HASH_TEA
)
1999 fname
->hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
2000 fname
->hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
2001 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), &fname
->hinfo
);
2003 memset(frames
, 0, sizeof(frames
));
2005 frame
->entries
= entries
;
2006 frame
->at
= entries
;
2009 retval
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
2012 retval
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
2016 de
= do_split(handle
,dir
, &bh2
, frame
, &fname
->hinfo
);
2018 retval
= PTR_ERR(de
);
2022 retval
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh2
);
2025 * Even if the block split failed, we have to properly write
2026 * out all the changes we did so far. Otherwise we can end up
2027 * with corrupted filesystem.
2030 ext4_mark_inode_dirty(handle
, dir
);
2039 * adds a file entry to the specified directory, using the same
2040 * semantics as ext4_find_entry(). It returns NULL if it failed.
2042 * NOTE!! The inode part of 'de' is left at 0 - which means you
2043 * may not sleep between calling this and putting something into
2044 * the entry, as someone else might have used it while you slept.
2046 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
2047 struct inode
*inode
)
2049 struct inode
*dir
= d_inode(dentry
->d_parent
);
2050 struct buffer_head
*bh
= NULL
;
2051 struct ext4_dir_entry_2
*de
;
2052 struct ext4_dir_entry_tail
*t
;
2053 struct super_block
*sb
;
2054 struct ext4_filename fname
;
2058 ext4_lblk_t block
, blocks
;
2061 if (ext4_has_metadata_csum(inode
->i_sb
))
2062 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2065 blocksize
= sb
->s_blocksize
;
2066 if (!dentry
->d_name
.len
)
2069 retval
= ext4_fname_setup_filename(dir
, &dentry
->d_name
, 0, &fname
);
2073 if (ext4_has_inline_data(dir
)) {
2074 retval
= ext4_try_add_inline_entry(handle
, &fname
, dir
, inode
);
2084 retval
= ext4_dx_add_entry(handle
, &fname
, dir
, inode
);
2085 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
2087 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
2089 ext4_mark_inode_dirty(handle
, dir
);
2091 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
2092 for (block
= 0; block
< blocks
; block
++) {
2093 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
2095 bh
= ext4_bread(handle
, dir
, block
,
2096 EXT4_GET_BLOCKS_CREATE
);
2097 goto add_to_new_block
;
2100 retval
= PTR_ERR(bh
);
2104 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
,
2106 if (retval
!= -ENOSPC
)
2109 if (blocks
== 1 && !dx_fallback
&&
2110 ext4_has_feature_dir_index(sb
)) {
2111 retval
= make_indexed_dir(handle
, &fname
, dir
,
2113 bh
= NULL
; /* make_indexed_dir releases bh */
2118 bh
= ext4_append(handle
, dir
, &block
);
2121 retval
= PTR_ERR(bh
);
2125 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2127 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
2130 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
2131 initialize_dirent_tail(t
, blocksize
);
2134 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
, de
, bh
);
2136 ext4_fname_free_filename(&fname
);
2139 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
2144 * Returns 0 for success, or a negative error value
2146 static int ext4_dx_add_entry(handle_t
*handle
, struct ext4_filename
*fname
,
2147 struct inode
*dir
, struct inode
*inode
)
2149 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
2150 struct dx_entry
*entries
, *at
;
2151 struct buffer_head
*bh
;
2152 struct super_block
*sb
= dir
->i_sb
;
2153 struct ext4_dir_entry_2
*de
;
2159 frame
= dx_probe(fname
, dir
, NULL
, frames
);
2161 return PTR_ERR(frame
);
2162 entries
= frame
->entries
;
2164 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT_HTREE
);
2171 BUFFER_TRACE(bh
, "get_write_access");
2172 err
= ext4_journal_get_write_access(handle
, bh
);
2176 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, NULL
, bh
);
2181 /* Block full, should compress but for now just split */
2182 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
2183 dx_get_count(entries
), dx_get_limit(entries
)));
2184 /* Need to split index? */
2185 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
2186 ext4_lblk_t newblock
;
2187 int levels
= frame
- frames
+ 1;
2188 unsigned int icount
;
2190 struct dx_entry
*entries2
;
2191 struct dx_node
*node2
;
2192 struct buffer_head
*bh2
;
2194 while (frame
> frames
) {
2195 if (dx_get_count((frame
- 1)->entries
) <
2196 dx_get_limit((frame
- 1)->entries
)) {
2200 frame
--; /* split higher index block */
2202 entries
= frame
->entries
;
2205 if (add_level
&& levels
== ext4_dir_htree_level(sb
)) {
2206 ext4_warning(sb
, "Directory (ino: %lu) index full, "
2207 "reach max htree level :%d",
2208 dir
->i_ino
, levels
);
2209 if (ext4_dir_htree_level(sb
) < EXT4_HTREE_LEVEL
) {
2210 ext4_warning(sb
, "Large directory feature is "
2211 "not enabled on this "
2217 icount
= dx_get_count(entries
);
2218 bh2
= ext4_append(handle
, dir
, &newblock
);
2223 node2
= (struct dx_node
*)(bh2
->b_data
);
2224 entries2
= node2
->entries
;
2225 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2226 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2228 BUFFER_TRACE(frame
->bh
, "get_write_access");
2229 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2233 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2234 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2235 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2238 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2239 err
= ext4_journal_get_write_access(handle
,
2244 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2245 icount2
* sizeof(struct dx_entry
));
2246 dx_set_count(entries
, icount1
);
2247 dx_set_count(entries2
, icount2
);
2248 dx_set_limit(entries2
, dx_node_limit(dir
));
2250 /* Which index block gets the new entry? */
2251 if (at
- entries
>= icount1
) {
2252 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2253 frame
->entries
= entries
= entries2
;
2254 swap(frame
->bh
, bh2
);
2256 dx_insert_block((frame
- 1), hash2
, newblock
);
2257 dxtrace(dx_show_index("node", frame
->entries
));
2258 dxtrace(dx_show_index("node",
2259 ((struct dx_node
*) bh2
->b_data
)->entries
));
2260 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2264 err
= ext4_handle_dirty_dx_node(handle
, dir
,
2269 err
= ext4_handle_dirty_dx_node(handle
, dir
,
2274 struct dx_root
*dxroot
;
2275 memcpy((char *) entries2
, (char *) entries
,
2276 icount
* sizeof(struct dx_entry
));
2277 dx_set_limit(entries2
, dx_node_limit(dir
));
2280 dx_set_count(entries
, 1);
2281 dx_set_block(entries
+ 0, newblock
);
2282 dxroot
= (struct dx_root
*)frames
[0].bh
->b_data
;
2283 dxroot
->info
.indirect_levels
+= 1;
2284 dxtrace(printk(KERN_DEBUG
2285 "Creating %d level index...\n",
2286 dxroot
->info
.indirect_levels
));
2287 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
2290 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2296 de
= do_split(handle
, dir
, &bh
, frame
, &fname
->hinfo
);
2301 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh
);
2305 ext4_std_error(dir
->i_sb
, err
); /* this is a no-op if err == 0 */
2309 /* @restart is true means htree-path has been changed, we need to
2310 * repeat dx_probe() to find out valid htree-path
2312 if (restart
&& err
== 0)
2318 * ext4_generic_delete_entry deletes a directory entry by merging it
2319 * with the previous entry
2321 int ext4_generic_delete_entry(handle_t
*handle
,
2323 struct ext4_dir_entry_2
*de_del
,
2324 struct buffer_head
*bh
,
2329 struct ext4_dir_entry_2
*de
, *pde
;
2330 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2335 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2336 while (i
< buf_size
- csum_size
) {
2337 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2338 bh
->b_data
, bh
->b_size
, i
))
2339 return -EFSCORRUPTED
;
2342 pde
->rec_len
= ext4_rec_len_to_disk(
2343 ext4_rec_len_from_disk(pde
->rec_len
,
2345 ext4_rec_len_from_disk(de
->rec_len
,
2350 inode_inc_iversion(dir
);
2353 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2355 de
= ext4_next_entry(de
, blocksize
);
2360 static int ext4_delete_entry(handle_t
*handle
,
2362 struct ext4_dir_entry_2
*de_del
,
2363 struct buffer_head
*bh
)
2365 int err
, csum_size
= 0;
2367 if (ext4_has_inline_data(dir
)) {
2368 int has_inline_data
= 1;
2369 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2371 if (has_inline_data
)
2375 if (ext4_has_metadata_csum(dir
->i_sb
))
2376 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2378 BUFFER_TRACE(bh
, "get_write_access");
2379 err
= ext4_journal_get_write_access(handle
, bh
);
2383 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2385 dir
->i_sb
->s_blocksize
, csum_size
);
2389 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2390 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2397 ext4_std_error(dir
->i_sb
, err
);
2402 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2403 * since this indicates that nlinks count was previously 1 to avoid overflowing
2404 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2405 * that subdirectory link counts are not being maintained accurately.
2407 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2408 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2409 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2410 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2412 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2416 (inode
->i_nlink
> EXT4_LINK_MAX
|| inode
->i_nlink
== 2))
2417 set_nlink(inode
, 1);
2421 * If a directory had nlink == 1, then we should let it be 1. This indicates
2422 * directory has >EXT4_LINK_MAX subdirs.
2424 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2426 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2431 static int ext4_add_nondir(handle_t
*handle
,
2432 struct dentry
*dentry
, struct inode
*inode
)
2434 int err
= ext4_add_entry(handle
, dentry
, inode
);
2436 ext4_mark_inode_dirty(handle
, inode
);
2437 d_instantiate_new(dentry
, inode
);
2441 unlock_new_inode(inode
);
2447 * By the time this is called, we already have created
2448 * the directory cache entry for the new file, but it
2449 * is so far negative - it has no inode.
2451 * If the create succeeds, we fill in the inode information
2452 * with d_instantiate().
2454 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2458 struct inode
*inode
;
2459 int err
, credits
, retries
= 0;
2461 err
= dquot_initialize(dir
);
2465 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2466 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2468 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2469 NULL
, EXT4_HT_DIR
, credits
);
2470 handle
= ext4_journal_current_handle();
2471 err
= PTR_ERR(inode
);
2472 if (!IS_ERR(inode
)) {
2473 inode
->i_op
= &ext4_file_inode_operations
;
2474 inode
->i_fop
= &ext4_file_operations
;
2475 ext4_set_aops(inode
);
2476 err
= ext4_add_nondir(handle
, dentry
, inode
);
2477 if (!err
&& IS_DIRSYNC(dir
))
2478 ext4_handle_sync(handle
);
2481 ext4_journal_stop(handle
);
2482 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2487 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2488 umode_t mode
, dev_t rdev
)
2491 struct inode
*inode
;
2492 int err
, credits
, retries
= 0;
2494 err
= dquot_initialize(dir
);
2498 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2499 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2501 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2502 NULL
, EXT4_HT_DIR
, credits
);
2503 handle
= ext4_journal_current_handle();
2504 err
= PTR_ERR(inode
);
2505 if (!IS_ERR(inode
)) {
2506 init_special_inode(inode
, inode
->i_mode
, rdev
);
2507 inode
->i_op
= &ext4_special_inode_operations
;
2508 err
= ext4_add_nondir(handle
, dentry
, inode
);
2509 if (!err
&& IS_DIRSYNC(dir
))
2510 ext4_handle_sync(handle
);
2513 ext4_journal_stop(handle
);
2514 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2519 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2522 struct inode
*inode
;
2523 int err
, retries
= 0;
2525 err
= dquot_initialize(dir
);
2530 inode
= ext4_new_inode_start_handle(dir
, mode
,
2533 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2534 4 + EXT4_XATTR_TRANS_BLOCKS
);
2535 handle
= ext4_journal_current_handle();
2536 err
= PTR_ERR(inode
);
2537 if (!IS_ERR(inode
)) {
2538 inode
->i_op
= &ext4_file_inode_operations
;
2539 inode
->i_fop
= &ext4_file_operations
;
2540 ext4_set_aops(inode
);
2541 d_tmpfile(dentry
, inode
);
2542 err
= ext4_orphan_add(handle
, inode
);
2544 goto err_unlock_inode
;
2545 mark_inode_dirty(inode
);
2546 unlock_new_inode(inode
);
2549 ext4_journal_stop(handle
);
2550 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2554 ext4_journal_stop(handle
);
2555 unlock_new_inode(inode
);
2559 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2560 struct ext4_dir_entry_2
*de
,
2561 int blocksize
, int csum_size
,
2562 unsigned int parent_ino
, int dotdot_real_len
)
2564 de
->inode
= cpu_to_le32(inode
->i_ino
);
2566 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2568 strcpy(de
->name
, ".");
2569 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2571 de
= ext4_next_entry(de
, blocksize
);
2572 de
->inode
= cpu_to_le32(parent_ino
);
2574 if (!dotdot_real_len
)
2575 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2576 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2579 de
->rec_len
= ext4_rec_len_to_disk(
2580 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2581 strcpy(de
->name
, "..");
2582 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2584 return ext4_next_entry(de
, blocksize
);
2587 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2588 struct inode
*inode
)
2590 struct buffer_head
*dir_block
= NULL
;
2591 struct ext4_dir_entry_2
*de
;
2592 struct ext4_dir_entry_tail
*t
;
2593 ext4_lblk_t block
= 0;
2594 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2598 if (ext4_has_metadata_csum(dir
->i_sb
))
2599 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2601 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2602 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2603 if (err
< 0 && err
!= -ENOSPC
)
2610 dir_block
= ext4_append(handle
, inode
, &block
);
2611 if (IS_ERR(dir_block
))
2612 return PTR_ERR(dir_block
);
2613 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2614 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2615 set_nlink(inode
, 2);
2617 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2618 initialize_dirent_tail(t
, blocksize
);
2621 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2622 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2625 set_buffer_verified(dir_block
);
2631 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2634 struct inode
*inode
;
2635 int err
, credits
, retries
= 0;
2637 if (EXT4_DIR_LINK_MAX(dir
))
2640 err
= dquot_initialize(dir
);
2644 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2645 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2647 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2649 0, NULL
, EXT4_HT_DIR
, credits
);
2650 handle
= ext4_journal_current_handle();
2651 err
= PTR_ERR(inode
);
2655 inode
->i_op
= &ext4_dir_inode_operations
;
2656 inode
->i_fop
= &ext4_dir_operations
;
2657 err
= ext4_init_new_dir(handle
, dir
, inode
);
2659 goto out_clear_inode
;
2660 err
= ext4_mark_inode_dirty(handle
, inode
);
2662 err
= ext4_add_entry(handle
, dentry
, inode
);
2666 unlock_new_inode(inode
);
2667 ext4_mark_inode_dirty(handle
, inode
);
2671 ext4_inc_count(handle
, dir
);
2672 ext4_update_dx_flag(dir
);
2673 err
= ext4_mark_inode_dirty(handle
, dir
);
2675 goto out_clear_inode
;
2676 d_instantiate_new(dentry
, inode
);
2677 if (IS_DIRSYNC(dir
))
2678 ext4_handle_sync(handle
);
2682 ext4_journal_stop(handle
);
2683 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2689 * routine to check that the specified directory is empty (for rmdir)
2691 bool ext4_empty_dir(struct inode
*inode
)
2693 unsigned int offset
;
2694 struct buffer_head
*bh
;
2695 struct ext4_dir_entry_2
*de
;
2696 struct super_block
*sb
;
2698 if (ext4_has_inline_data(inode
)) {
2699 int has_inline_data
= 1;
2702 ret
= empty_inline_dir(inode
, &has_inline_data
);
2703 if (has_inline_data
)
2708 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2709 EXT4_ERROR_INODE(inode
, "invalid size");
2712 /* The first directory block must not be a hole,
2713 * so treat it as DIRENT_HTREE
2715 bh
= ext4_read_dirblock(inode
, 0, DIRENT_HTREE
);
2719 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2720 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
, bh
->b_data
, bh
->b_size
,
2722 le32_to_cpu(de
->inode
) != inode
->i_ino
|| strcmp(".", de
->name
)) {
2723 ext4_warning_inode(inode
, "directory missing '.'");
2727 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2728 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2729 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
, bh
->b_data
, bh
->b_size
,
2731 le32_to_cpu(de
->inode
) == 0 || strcmp("..", de
->name
)) {
2732 ext4_warning_inode(inode
, "directory missing '..'");
2736 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2737 while (offset
< inode
->i_size
) {
2738 if (!(offset
& (sb
->s_blocksize
- 1))) {
2739 unsigned int lblock
;
2741 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2742 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2744 offset
+= sb
->s_blocksize
;
2750 de
= (struct ext4_dir_entry_2
*) (bh
->b_data
+
2751 (offset
& (sb
->s_blocksize
- 1)));
2752 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2753 bh
->b_data
, bh
->b_size
, offset
)) {
2754 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2757 if (le32_to_cpu(de
->inode
)) {
2761 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2768 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2769 * such inodes, starting at the superblock, in case we crash before the
2770 * file is closed/deleted, or in case the inode truncate spans multiple
2771 * transactions and the last transaction is not recovered after a crash.
2773 * At filesystem recovery time, we walk this list deleting unlinked
2774 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2776 * Orphan list manipulation functions must be called under i_mutex unless
2777 * we are just creating the inode or deleting it.
2779 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2781 struct super_block
*sb
= inode
->i_sb
;
2782 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2783 struct ext4_iloc iloc
;
2787 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2790 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2791 !inode_is_locked(inode
));
2793 * Exit early if inode already is on orphan list. This is a big speedup
2794 * since we don't have to contend on the global s_orphan_lock.
2796 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2800 * Orphan handling is only valid for files with data blocks
2801 * being truncated, or files being unlinked. Note that we either
2802 * hold i_mutex, or the inode can not be referenced from outside,
2803 * so i_nlink should not be bumped due to race
2805 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2806 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2808 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2809 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2813 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2817 mutex_lock(&sbi
->s_orphan_lock
);
2819 * Due to previous errors inode may be already a part of on-disk
2820 * orphan list. If so skip on-disk list modification.
2822 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2823 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2824 /* Insert this inode at the head of the on-disk orphan list */
2825 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2826 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2829 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2830 mutex_unlock(&sbi
->s_orphan_lock
);
2833 err
= ext4_handle_dirty_super(handle
, sb
);
2834 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2839 * We have to remove inode from in-memory list if
2840 * addition to on disk orphan list failed. Stray orphan
2841 * list entries can cause panics at unmount time.
2843 mutex_lock(&sbi
->s_orphan_lock
);
2844 list_del_init(&EXT4_I(inode
)->i_orphan
);
2845 mutex_unlock(&sbi
->s_orphan_lock
);
2850 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2851 jbd_debug(4, "orphan inode %lu will point to %d\n",
2852 inode
->i_ino
, NEXT_ORPHAN(inode
));
2854 ext4_std_error(sb
, err
);
2859 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2860 * of such inodes stored on disk, because it is finally being cleaned up.
2862 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2864 struct list_head
*prev
;
2865 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2866 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2868 struct ext4_iloc iloc
;
2871 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2874 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2875 !inode_is_locked(inode
));
2876 /* Do this quick check before taking global s_orphan_lock. */
2877 if (list_empty(&ei
->i_orphan
))
2881 /* Grab inode buffer early before taking global s_orphan_lock */
2882 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2885 mutex_lock(&sbi
->s_orphan_lock
);
2886 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2888 prev
= ei
->i_orphan
.prev
;
2889 list_del_init(&ei
->i_orphan
);
2891 /* If we're on an error path, we may not have a valid
2892 * transaction handle with which to update the orphan list on
2893 * disk, but we still need to remove the inode from the linked
2894 * list in memory. */
2895 if (!handle
|| err
) {
2896 mutex_unlock(&sbi
->s_orphan_lock
);
2900 ino_next
= NEXT_ORPHAN(inode
);
2901 if (prev
== &sbi
->s_orphan
) {
2902 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2903 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2904 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2906 mutex_unlock(&sbi
->s_orphan_lock
);
2909 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2910 mutex_unlock(&sbi
->s_orphan_lock
);
2911 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2913 struct ext4_iloc iloc2
;
2914 struct inode
*i_prev
=
2915 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2917 jbd_debug(4, "orphan inode %lu will point to %u\n",
2918 i_prev
->i_ino
, ino_next
);
2919 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2921 mutex_unlock(&sbi
->s_orphan_lock
);
2924 NEXT_ORPHAN(i_prev
) = ino_next
;
2925 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2926 mutex_unlock(&sbi
->s_orphan_lock
);
2930 NEXT_ORPHAN(inode
) = 0;
2931 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2933 ext4_std_error(inode
->i_sb
, err
);
2941 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2944 struct inode
*inode
;
2945 struct buffer_head
*bh
;
2946 struct ext4_dir_entry_2
*de
;
2947 handle_t
*handle
= NULL
;
2949 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
2952 /* Initialize quotas before so that eventual writes go in
2953 * separate transaction */
2954 retval
= dquot_initialize(dir
);
2957 retval
= dquot_initialize(d_inode(dentry
));
2962 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2968 inode
= d_inode(dentry
);
2970 retval
= -EFSCORRUPTED
;
2971 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2974 retval
= -ENOTEMPTY
;
2975 if (!ext4_empty_dir(inode
))
2978 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2979 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2980 if (IS_ERR(handle
)) {
2981 retval
= PTR_ERR(handle
);
2986 if (IS_DIRSYNC(dir
))
2987 ext4_handle_sync(handle
);
2989 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2992 if (!EXT4_DIR_LINK_EMPTY(inode
))
2993 ext4_warning_inode(inode
,
2994 "empty directory '%.*s' has too many links (%u)",
2995 dentry
->d_name
.len
, dentry
->d_name
.name
,
2999 /* There's no need to set i_disksize: the fact that i_nlink is
3000 * zero will ensure that the right thing happens during any
3003 ext4_orphan_add(handle
, inode
);
3004 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3005 ext4_mark_inode_dirty(handle
, inode
);
3006 ext4_dec_count(handle
, dir
);
3007 ext4_update_dx_flag(dir
);
3008 ext4_mark_inode_dirty(handle
, dir
);
3013 ext4_journal_stop(handle
);
3017 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
3020 struct inode
*inode
;
3021 struct buffer_head
*bh
;
3022 struct ext4_dir_entry_2
*de
;
3023 handle_t
*handle
= NULL
;
3025 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
3028 trace_ext4_unlink_enter(dir
, dentry
);
3029 /* Initialize quotas before so that eventual writes go
3030 * in separate transaction */
3031 retval
= dquot_initialize(dir
);
3034 retval
= dquot_initialize(d_inode(dentry
));
3039 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
3045 inode
= d_inode(dentry
);
3047 retval
= -EFSCORRUPTED
;
3048 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
3051 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3052 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
3053 if (IS_ERR(handle
)) {
3054 retval
= PTR_ERR(handle
);
3059 if (IS_DIRSYNC(dir
))
3060 ext4_handle_sync(handle
);
3062 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3065 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3066 ext4_update_dx_flag(dir
);
3067 ext4_mark_inode_dirty(handle
, dir
);
3068 if (inode
->i_nlink
== 0)
3069 ext4_warning_inode(inode
, "Deleting file '%.*s' with no links",
3070 dentry
->d_name
.len
, dentry
->d_name
.name
);
3073 if (!inode
->i_nlink
)
3074 ext4_orphan_add(handle
, inode
);
3075 inode
->i_ctime
= current_time(inode
);
3076 ext4_mark_inode_dirty(handle
, inode
);
3081 ext4_journal_stop(handle
);
3082 trace_ext4_unlink_exit(dentry
, retval
);
3086 static int ext4_symlink(struct inode
*dir
,
3087 struct dentry
*dentry
, const char *symname
)
3090 struct inode
*inode
;
3091 int err
, len
= strlen(symname
);
3093 bool encryption_required
;
3094 struct fscrypt_str disk_link
;
3095 struct fscrypt_symlink_data
*sd
= NULL
;
3097 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
3100 disk_link
.len
= len
+ 1;
3101 disk_link
.name
= (char *) symname
;
3103 encryption_required
= (ext4_encrypted_inode(dir
) ||
3104 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir
->i_sb
)));
3105 if (encryption_required
) {
3106 err
= fscrypt_get_encryption_info(dir
);
3109 if (!fscrypt_has_encryption_key(dir
))
3111 disk_link
.len
= (fscrypt_fname_encrypted_size(dir
, len
) +
3112 sizeof(struct fscrypt_symlink_data
));
3113 sd
= kzalloc(disk_link
.len
, GFP_KERNEL
);
3118 if (disk_link
.len
> dir
->i_sb
->s_blocksize
) {
3119 err
= -ENAMETOOLONG
;
3123 err
= dquot_initialize(dir
);
3127 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3129 * For non-fast symlinks, we just allocate inode and put it on
3130 * orphan list in the first transaction => we need bitmap,
3131 * group descriptor, sb, inode block, quota blocks, and
3132 * possibly selinux xattr blocks.
3134 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
3135 EXT4_XATTR_TRANS_BLOCKS
;
3138 * Fast symlink. We have to add entry to directory
3139 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3140 * allocate new inode (bitmap, group descriptor, inode block,
3141 * quota blocks, sb is already counted in previous macros).
3143 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3144 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
3147 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
3148 &dentry
->d_name
, 0, NULL
,
3149 EXT4_HT_DIR
, credits
);
3150 handle
= ext4_journal_current_handle();
3151 if (IS_ERR(inode
)) {
3153 ext4_journal_stop(handle
);
3154 err
= PTR_ERR(inode
);
3158 if (encryption_required
) {
3160 struct fscrypt_str ostr
=
3161 FSTR_INIT(sd
->encrypted_path
, disk_link
.len
);
3163 istr
.name
= (const unsigned char *) symname
;
3165 err
= fscrypt_fname_usr_to_disk(inode
, &istr
, &ostr
);
3167 goto err_drop_inode
;
3168 sd
->len
= cpu_to_le16(ostr
.len
);
3169 disk_link
.name
= (char *) sd
;
3170 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
3173 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3174 if (!encryption_required
)
3175 inode
->i_op
= &ext4_symlink_inode_operations
;
3176 inode_nohighmem(inode
);
3177 ext4_set_aops(inode
);
3179 * We cannot call page_symlink() with transaction started
3180 * because it calls into ext4_write_begin() which can wait
3181 * for transaction commit if we are running out of space
3182 * and thus we deadlock. So we have to stop transaction now
3183 * and restart it when symlink contents is written.
3185 * To keep fs consistent in case of crash, we have to put inode
3186 * to orphan list in the mean time.
3189 err
= ext4_orphan_add(handle
, inode
);
3190 ext4_journal_stop(handle
);
3193 goto err_drop_inode
;
3194 err
= __page_symlink(inode
, disk_link
.name
, disk_link
.len
, 1);
3196 goto err_drop_inode
;
3198 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3199 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3201 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3202 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3203 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
3204 if (IS_ERR(handle
)) {
3205 err
= PTR_ERR(handle
);
3207 goto err_drop_inode
;
3209 set_nlink(inode
, 1);
3210 err
= ext4_orphan_del(handle
, inode
);
3212 goto err_drop_inode
;
3214 /* clear the extent format for fast symlink */
3215 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
3216 if (!encryption_required
) {
3217 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3218 inode
->i_link
= (char *)&EXT4_I(inode
)->i_data
;
3220 memcpy((char *)&EXT4_I(inode
)->i_data
, disk_link
.name
,
3222 inode
->i_size
= disk_link
.len
- 1;
3224 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3225 err
= ext4_add_nondir(handle
, dentry
, inode
);
3226 if (!err
&& IS_DIRSYNC(dir
))
3227 ext4_handle_sync(handle
);
3230 ext4_journal_stop(handle
);
3235 ext4_journal_stop(handle
);
3237 unlock_new_inode(inode
);
3244 static int ext4_link(struct dentry
*old_dentry
,
3245 struct inode
*dir
, struct dentry
*dentry
)
3248 struct inode
*inode
= d_inode(old_dentry
);
3249 int err
, retries
= 0;
3251 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
3254 err
= fscrypt_prepare_link(old_dentry
, dir
, dentry
);
3258 if ((ext4_test_inode_flag(dir
, EXT4_INODE_PROJINHERIT
)) &&
3259 (!projid_valid_eq(EXT4_I(dir
)->i_projid
,
3260 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3263 err
= dquot_initialize(dir
);
3268 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3269 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3270 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
3272 return PTR_ERR(handle
);
3274 if (IS_DIRSYNC(dir
))
3275 ext4_handle_sync(handle
);
3277 inode
->i_ctime
= current_time(inode
);
3278 ext4_inc_count(handle
, inode
);
3281 err
= ext4_add_entry(handle
, dentry
, inode
);
3283 ext4_mark_inode_dirty(handle
, inode
);
3284 /* this can happen only for tmpfile being
3285 * linked the first time
3287 if (inode
->i_nlink
== 1)
3288 ext4_orphan_del(handle
, inode
);
3289 d_instantiate(dentry
, inode
);
3294 ext4_journal_stop(handle
);
3295 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3302 * Try to find buffer head where contains the parent block.
3303 * It should be the inode block if it is inlined or the 1st block
3304 * if it is a normal dir.
3306 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
3307 struct inode
*inode
,
3309 struct ext4_dir_entry_2
**parent_de
,
3312 struct buffer_head
*bh
;
3314 if (!ext4_has_inline_data(inode
)) {
3315 /* The first directory block must not be a hole, so
3316 * treat it as DIRENT_HTREE
3318 bh
= ext4_read_dirblock(inode
, 0, DIRENT_HTREE
);
3320 *retval
= PTR_ERR(bh
);
3323 *parent_de
= ext4_next_entry(
3324 (struct ext4_dir_entry_2
*)bh
->b_data
,
3325 inode
->i_sb
->s_blocksize
);
3330 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3333 struct ext4_renament
{
3335 struct dentry
*dentry
;
3336 struct inode
*inode
;
3338 int dir_nlink_delta
;
3340 /* entry for "dentry" */
3341 struct buffer_head
*bh
;
3342 struct ext4_dir_entry_2
*de
;
3345 /* entry for ".." in inode if it's a directory */
3346 struct buffer_head
*dir_bh
;
3347 struct ext4_dir_entry_2
*parent_de
;
3351 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3355 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3356 &retval
, &ent
->parent_de
,
3360 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3361 return -EFSCORRUPTED
;
3362 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3363 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3366 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3371 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3372 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3373 if (!ent
->dir_inlined
) {
3374 if (is_dx(ent
->inode
)) {
3375 retval
= ext4_handle_dirty_dx_node(handle
,
3379 retval
= ext4_handle_dirty_dirent_node(handle
,
3384 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3387 ext4_std_error(ent
->dir
->i_sb
, retval
);
3393 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3394 unsigned ino
, unsigned file_type
)
3398 BUFFER_TRACE(ent
->bh
, "get write access");
3399 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3402 ent
->de
->inode
= cpu_to_le32(ino
);
3403 if (ext4_has_feature_filetype(ent
->dir
->i_sb
))
3404 ent
->de
->file_type
= file_type
;
3405 ent
->dir
->i_version
++;
3406 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3407 current_time(ent
->dir
);
3408 ext4_mark_inode_dirty(handle
, ent
->dir
);
3409 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3410 if (!ent
->inlined
) {
3411 retval
= ext4_handle_dirty_dirent_node(handle
,
3413 if (unlikely(retval
)) {
3414 ext4_std_error(ent
->dir
->i_sb
, retval
);
3424 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3425 const struct qstr
*d_name
)
3427 int retval
= -ENOENT
;
3428 struct buffer_head
*bh
;
3429 struct ext4_dir_entry_2
*de
;
3431 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3435 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3441 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3446 * ent->de could have moved from under us during htree split, so make
3447 * sure that we are deleting the right entry. We might also be pointing
3448 * to a stale entry in the unused part of ent->bh so just checking inum
3449 * and the name isn't enough.
3451 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3452 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3453 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3454 ent
->de
->name_len
) ||
3456 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3457 &ent
->dentry
->d_name
);
3459 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3460 if (retval
== -ENOENT
) {
3461 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3462 &ent
->dentry
->d_name
);
3467 ext4_warning_inode(ent
->dir
,
3468 "Deleting old file: nlink %d, error=%d",
3469 ent
->dir
->i_nlink
, retval
);
3473 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3475 if (ent
->dir_nlink_delta
) {
3476 if (ent
->dir_nlink_delta
== -1)
3477 ext4_dec_count(handle
, ent
->dir
);
3479 ext4_inc_count(handle
, ent
->dir
);
3480 ext4_mark_inode_dirty(handle
, ent
->dir
);
3484 static struct inode
*ext4_whiteout_for_rename(struct ext4_renament
*ent
,
3485 int credits
, handle_t
**h
)
3492 * for inode block, sb block, group summaries,
3495 credits
+= (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent
->dir
->i_sb
) +
3496 EXT4_XATTR_TRANS_BLOCKS
+ 4);
3498 wh
= ext4_new_inode_start_handle(ent
->dir
, S_IFCHR
| WHITEOUT_MODE
,
3499 &ent
->dentry
->d_name
, 0, NULL
,
3500 EXT4_HT_DIR
, credits
);
3502 handle
= ext4_journal_current_handle();
3505 ext4_journal_stop(handle
);
3506 if (PTR_ERR(wh
) == -ENOSPC
&&
3507 ext4_should_retry_alloc(ent
->dir
->i_sb
, &retries
))
3511 init_special_inode(wh
, wh
->i_mode
, WHITEOUT_DEV
);
3512 wh
->i_op
= &ext4_special_inode_operations
;
3518 * Anybody can rename anything with this: the permission checks are left to the
3519 * higher-level routines.
3521 * n.b. old_{dentry,inode) refers to the source dentry/inode
3522 * while new_{dentry,inode) refers to the destination dentry/inode
3523 * This comes from rename(const char *oldpath, const char *newpath)
3525 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3526 struct inode
*new_dir
, struct dentry
*new_dentry
,
3529 handle_t
*handle
= NULL
;
3530 struct ext4_renament old
= {
3532 .dentry
= old_dentry
,
3533 .inode
= d_inode(old_dentry
),
3535 struct ext4_renament
new = {
3537 .dentry
= new_dentry
,
3538 .inode
= d_inode(new_dentry
),
3542 struct inode
*whiteout
= NULL
;
3546 if (new.inode
&& new.inode
->i_nlink
== 0) {
3547 EXT4_ERROR_INODE(new.inode
,
3548 "target of rename is already freed");
3549 return -EFSCORRUPTED
;
3552 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
)) &&
3553 (!projid_valid_eq(EXT4_I(new_dir
)->i_projid
,
3554 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3557 retval
= dquot_initialize(old
.dir
);
3560 retval
= dquot_initialize(new.dir
);
3564 /* Initialize quotas before so that eventual writes go
3565 * in separate transaction */
3567 retval
= dquot_initialize(new.inode
);
3572 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3574 return PTR_ERR(old
.bh
);
3576 * Check for inode number is _not_ due to possible IO errors.
3577 * We might rmdir the source, keep it as pwd of some process
3578 * and merrily kill the link to whatever was created under the
3579 * same name. Goodbye sticky bit ;-<
3582 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3585 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3586 &new.de
, &new.inlined
);
3587 if (IS_ERR(new.bh
)) {
3588 retval
= PTR_ERR(new.bh
);
3598 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3599 ext4_alloc_da_blocks(old
.inode
);
3601 credits
= (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3602 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
3603 if (!(flags
& RENAME_WHITEOUT
)) {
3604 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
, credits
);
3605 if (IS_ERR(handle
)) {
3606 retval
= PTR_ERR(handle
);
3611 whiteout
= ext4_whiteout_for_rename(&old
, credits
, &handle
);
3612 if (IS_ERR(whiteout
)) {
3613 retval
= PTR_ERR(whiteout
);
3619 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3620 ext4_handle_sync(handle
);
3622 if (S_ISDIR(old
.inode
->i_mode
)) {
3624 retval
= -ENOTEMPTY
;
3625 if (!ext4_empty_dir(new.inode
))
3629 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3632 retval
= ext4_rename_dir_prepare(handle
, &old
);
3637 * If we're renaming a file within an inline_data dir and adding or
3638 * setting the new dirent causes a conversion from inline_data to
3639 * extents/blockmap, we need to force the dirent delete code to
3640 * re-read the directory, or else we end up trying to delete a dirent
3641 * from what is now the extent tree root (or a block map).
3643 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3644 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3646 old_file_type
= old
.de
->file_type
;
3649 * Do this before adding a new entry, so the old entry is sure
3650 * to be still pointing to the valid old entry.
3652 retval
= ext4_setent(handle
, &old
, whiteout
->i_ino
,
3656 ext4_mark_inode_dirty(handle
, whiteout
);
3659 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3663 retval
= ext4_setent(handle
, &new,
3664 old
.inode
->i_ino
, old_file_type
);
3669 force_reread
= !ext4_test_inode_flag(new.dir
,
3670 EXT4_INODE_INLINE_DATA
);
3673 * Like most other Unix systems, set the ctime for inodes on a
3676 old
.inode
->i_ctime
= current_time(old
.inode
);
3677 ext4_mark_inode_dirty(handle
, old
.inode
);
3683 ext4_rename_delete(handle
, &old
, force_reread
);
3687 ext4_dec_count(handle
, new.inode
);
3688 new.inode
->i_ctime
= current_time(new.inode
);
3690 old
.dir
->i_ctime
= old
.dir
->i_mtime
= current_time(old
.dir
);
3691 ext4_update_dx_flag(old
.dir
);
3693 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3697 ext4_dec_count(handle
, old
.dir
);
3699 /* checked ext4_empty_dir above, can't have another
3700 * parent, ext4_dec_count() won't work for many-linked
3702 clear_nlink(new.inode
);
3704 ext4_inc_count(handle
, new.dir
);
3705 ext4_update_dx_flag(new.dir
);
3706 ext4_mark_inode_dirty(handle
, new.dir
);
3709 ext4_mark_inode_dirty(handle
, old
.dir
);
3711 ext4_mark_inode_dirty(handle
, new.inode
);
3712 if (!new.inode
->i_nlink
)
3713 ext4_orphan_add(handle
, new.inode
);
3723 drop_nlink(whiteout
);
3724 unlock_new_inode(whiteout
);
3728 ext4_journal_stop(handle
);
3732 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3733 struct inode
*new_dir
, struct dentry
*new_dentry
)
3735 handle_t
*handle
= NULL
;
3736 struct ext4_renament old
= {
3738 .dentry
= old_dentry
,
3739 .inode
= d_inode(old_dentry
),
3741 struct ext4_renament
new = {
3743 .dentry
= new_dentry
,
3744 .inode
= d_inode(new_dentry
),
3748 struct timespec ctime
;
3750 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
) &&
3751 !projid_valid_eq(EXT4_I(new_dir
)->i_projid
,
3752 EXT4_I(old_dentry
->d_inode
)->i_projid
)) ||
3753 (ext4_test_inode_flag(old_dir
, EXT4_INODE_PROJINHERIT
) &&
3754 !projid_valid_eq(EXT4_I(old_dir
)->i_projid
,
3755 EXT4_I(new_dentry
->d_inode
)->i_projid
)))
3758 retval
= dquot_initialize(old
.dir
);
3761 retval
= dquot_initialize(new.dir
);
3765 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3766 &old
.de
, &old
.inlined
);
3768 return PTR_ERR(old
.bh
);
3770 * Check for inode number is _not_ due to possible IO errors.
3771 * We might rmdir the source, keep it as pwd of some process
3772 * and merrily kill the link to whatever was created under the
3773 * same name. Goodbye sticky bit ;-<
3776 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3779 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3780 &new.de
, &new.inlined
);
3781 if (IS_ERR(new.bh
)) {
3782 retval
= PTR_ERR(new.bh
);
3787 /* RENAME_EXCHANGE case: old *and* new must both exist */
3788 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3791 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3792 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3793 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3794 if (IS_ERR(handle
)) {
3795 retval
= PTR_ERR(handle
);
3800 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3801 ext4_handle_sync(handle
);
3803 if (S_ISDIR(old
.inode
->i_mode
)) {
3805 retval
= ext4_rename_dir_prepare(handle
, &old
);
3809 if (S_ISDIR(new.inode
->i_mode
)) {
3811 retval
= ext4_rename_dir_prepare(handle
, &new);
3817 * Other than the special case of overwriting a directory, parents'
3818 * nlink only needs to be modified if this is a cross directory rename.
3820 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3821 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3822 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3824 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3825 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3829 new_file_type
= new.de
->file_type
;
3830 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3834 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3839 * Like most other Unix systems, set the ctime for inodes on a
3842 ctime
= current_time(old
.inode
);
3843 old
.inode
->i_ctime
= ctime
;
3844 new.inode
->i_ctime
= ctime
;
3845 ext4_mark_inode_dirty(handle
, old
.inode
);
3846 ext4_mark_inode_dirty(handle
, new.inode
);
3849 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3854 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3858 ext4_update_dir_count(handle
, &old
);
3859 ext4_update_dir_count(handle
, &new);
3868 ext4_journal_stop(handle
);
3872 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3873 struct inode
*new_dir
, struct dentry
*new_dentry
,
3878 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir
->i_sb
))))
3881 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3884 err
= fscrypt_prepare_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
3889 if (flags
& RENAME_EXCHANGE
) {
3890 return ext4_cross_rename(old_dir
, old_dentry
,
3891 new_dir
, new_dentry
);
3894 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
3898 * directories can handle most operations...
3900 const struct inode_operations ext4_dir_inode_operations
= {
3901 .create
= ext4_create
,
3902 .lookup
= ext4_lookup
,
3904 .unlink
= ext4_unlink
,
3905 .symlink
= ext4_symlink
,
3906 .mkdir
= ext4_mkdir
,
3907 .rmdir
= ext4_rmdir
,
3908 .mknod
= ext4_mknod
,
3909 .tmpfile
= ext4_tmpfile
,
3910 .rename
= ext4_rename2
,
3911 .setattr
= ext4_setattr
,
3912 .getattr
= ext4_getattr
,
3913 .listxattr
= ext4_listxattr
,
3914 .get_acl
= ext4_get_acl
,
3915 .set_acl
= ext4_set_acl
,
3916 .fiemap
= ext4_fiemap
,
3919 const struct inode_operations ext4_special_inode_operations
= {
3920 .setattr
= ext4_setattr
,
3921 .getattr
= ext4_getattr
,
3922 .listxattr
= ext4_listxattr
,
3923 .get_acl
= ext4_get_acl
,
3924 .set_acl
= ext4_set_acl
,