2 * linux/fs/ext4/namei.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
37 #include "ext4_jbd2.h"
42 #include <trace/events/ext4.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 static struct buffer_head
*ext4_append(handle_t
*handle
,
54 struct buffer_head
*bh
;
57 if (unlikely(EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
&&
58 ((inode
->i_size
>> 10) >=
59 EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
)))
60 return ERR_PTR(-ENOSPC
);
62 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
64 bh
= ext4_bread(handle
, inode
, *block
, EXT4_GET_BLOCKS_CREATE
);
67 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
68 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
69 BUFFER_TRACE(bh
, "get_write_access");
70 err
= ext4_journal_get_write_access(handle
, bh
);
73 ext4_std_error(inode
->i_sb
, err
);
79 static int ext4_dx_csum_verify(struct inode
*inode
,
80 struct ext4_dir_entry
*dirent
);
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89 static struct buffer_head
*__ext4_read_dirblock(struct inode
*inode
,
95 struct buffer_head
*bh
;
96 struct ext4_dir_entry
*dirent
;
99 bh
= ext4_bread(NULL
, inode
, block
, 0);
101 __ext4_warning(inode
->i_sb
, func
, line
,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode
->i_ino
, (unsigned long)block
,
105 current
->comm
, PTR_ERR(bh
));
110 ext4_error_inode(inode
, func
, line
, block
,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED
);
114 dirent
= (struct ext4_dir_entry
*) bh
->b_data
;
115 /* Determine whether or not we have an index block */
119 else if (ext4_rec_len_from_disk(dirent
->rec_len
,
120 inode
->i_sb
->s_blocksize
) ==
121 inode
->i_sb
->s_blocksize
)
124 if (!is_dx_block
&& type
== INDEX
) {
125 ext4_error_inode(inode
, func
, line
, block
,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED
);
129 if (!ext4_has_metadata_csum(inode
->i_sb
) ||
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
138 if (is_dx_block
&& type
== INDEX
) {
139 if (ext4_dx_csum_verify(inode
, dirent
))
140 set_buffer_verified(bh
);
142 ext4_error_inode(inode
, func
, line
, block
,
143 "Directory index failed checksum");
145 return ERR_PTR(-EFSBADCRC
);
149 if (ext4_dirent_csum_verify(inode
, dirent
))
150 set_buffer_verified(bh
);
152 ext4_error_inode(inode
, func
, line
, block
,
153 "Directory block failed checksum");
155 return ERR_PTR(-EFSBADCRC
);
162 #define assert(test) J_ASSERT(test)
166 #define dxtrace(command) command
168 #define dxtrace(command)
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
199 struct fake_dirent dot
;
201 struct fake_dirent dotdot
;
205 __le32 reserved_zero
;
207 u8 info_length
; /* 8 */
212 struct dx_entry entries
[0];
217 struct fake_dirent fake
;
218 struct dx_entry entries
[0];
224 struct buffer_head
*bh
;
225 struct dx_entry
*entries
;
237 * This goes at the end of each htree block.
241 __le32 dt_checksum
; /* crc32c(uuid+inum+dirblock) */
244 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
245 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
246 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
247 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
248 static unsigned dx_get_count(struct dx_entry
*entries
);
249 static unsigned dx_get_limit(struct dx_entry
*entries
);
250 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
251 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
252 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
253 static unsigned dx_node_limit(struct inode
*dir
);
254 static struct dx_frame
*dx_probe(struct ext4_filename
*fname
,
256 struct dx_hash_info
*hinfo
,
257 struct dx_frame
*frame
);
258 static void dx_release(struct dx_frame
*frames
);
259 static int dx_make_map(struct inode
*dir
, struct ext4_dir_entry_2
*de
,
260 unsigned blocksize
, struct dx_hash_info
*hinfo
,
261 struct dx_map_entry map
[]);
262 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
263 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
264 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
265 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
266 static void dx_insert_block(struct dx_frame
*frame
,
267 u32 hash
, ext4_lblk_t block
);
268 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
269 struct dx_frame
*frame
,
270 struct dx_frame
*frames
,
272 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
273 struct ext4_filename
*fname
,
274 struct ext4_dir_entry_2
**res_dir
);
275 static int ext4_dx_add_entry(handle_t
*handle
, struct ext4_filename
*fname
,
276 struct inode
*dir
, struct inode
*inode
);
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail
*t
,
280 unsigned int blocksize
)
282 memset(t
, 0, sizeof(struct ext4_dir_entry_tail
));
283 t
->det_rec_len
= ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail
), blocksize
);
285 t
->det_reserved_ft
= EXT4_FT_DIR_CSUM
;
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail
*get_dirent_tail(struct inode
*inode
,
290 struct ext4_dir_entry
*de
)
292 struct ext4_dir_entry_tail
*t
;
295 struct ext4_dir_entry
*d
, *top
;
298 top
= (struct ext4_dir_entry
*)(((void *)de
) +
299 (EXT4_BLOCK_SIZE(inode
->i_sb
) -
300 sizeof(struct ext4_dir_entry_tail
)));
301 while (d
< top
&& d
->rec_len
)
302 d
= (struct ext4_dir_entry
*)(((void *)d
) +
303 le16_to_cpu(d
->rec_len
));
308 t
= (struct ext4_dir_entry_tail
*)d
;
310 t
= EXT4_DIRENT_TAIL(de
, EXT4_BLOCK_SIZE(inode
->i_sb
));
313 if (t
->det_reserved_zero1
||
314 le16_to_cpu(t
->det_rec_len
) != sizeof(struct ext4_dir_entry_tail
) ||
315 t
->det_reserved_zero2
||
316 t
->det_reserved_ft
!= EXT4_FT_DIR_CSUM
)
322 static __le32
ext4_dirent_csum(struct inode
*inode
,
323 struct ext4_dir_entry
*dirent
, int size
)
325 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
326 struct ext4_inode_info
*ei
= EXT4_I(inode
);
329 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
330 return cpu_to_le32(csum
);
333 #define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
336 static void __warn_no_space_for_csum(struct inode
*inode
, const char *func
,
339 __ext4_warning_inode(inode
, func
, line
,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
343 int ext4_dirent_csum_verify(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
345 struct ext4_dir_entry_tail
*t
;
347 if (!ext4_has_metadata_csum(inode
->i_sb
))
350 t
= get_dirent_tail(inode
, dirent
);
352 warn_no_space_for_csum(inode
);
356 if (t
->det_checksum
!= ext4_dirent_csum(inode
, dirent
,
357 (void *)t
- (void *)dirent
))
363 static void ext4_dirent_csum_set(struct inode
*inode
,
364 struct ext4_dir_entry
*dirent
)
366 struct ext4_dir_entry_tail
*t
;
368 if (!ext4_has_metadata_csum(inode
->i_sb
))
371 t
= get_dirent_tail(inode
, dirent
);
373 warn_no_space_for_csum(inode
);
377 t
->det_checksum
= ext4_dirent_csum(inode
, dirent
,
378 (void *)t
- (void *)dirent
);
381 int ext4_handle_dirty_dirent_node(handle_t
*handle
,
383 struct buffer_head
*bh
)
385 ext4_dirent_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
386 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
389 static struct dx_countlimit
*get_dx_countlimit(struct inode
*inode
,
390 struct ext4_dir_entry
*dirent
,
393 struct ext4_dir_entry
*dp
;
394 struct dx_root_info
*root
;
397 if (le16_to_cpu(dirent
->rec_len
) == EXT4_BLOCK_SIZE(inode
->i_sb
))
399 else if (le16_to_cpu(dirent
->rec_len
) == 12) {
400 dp
= (struct ext4_dir_entry
*)(((void *)dirent
) + 12);
401 if (le16_to_cpu(dp
->rec_len
) !=
402 EXT4_BLOCK_SIZE(inode
->i_sb
) - 12)
404 root
= (struct dx_root_info
*)(((void *)dp
+ 12));
405 if (root
->reserved_zero
||
406 root
->info_length
!= sizeof(struct dx_root_info
))
413 *offset
= count_offset
;
414 return (struct dx_countlimit
*)(((void *)dirent
) + count_offset
);
417 static __le32
ext4_dx_csum(struct inode
*inode
, struct ext4_dir_entry
*dirent
,
418 int count_offset
, int count
, struct dx_tail
*t
)
420 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
421 struct ext4_inode_info
*ei
= EXT4_I(inode
);
424 __u32 dummy_csum
= 0;
425 int offset
= offsetof(struct dx_tail
, dt_checksum
);
427 size
= count_offset
+ (count
* sizeof(struct dx_entry
));
428 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
429 csum
= ext4_chksum(sbi
, csum
, (__u8
*)t
, offset
);
430 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, sizeof(dummy_csum
));
432 return cpu_to_le32(csum
);
435 static int ext4_dx_csum_verify(struct inode
*inode
,
436 struct ext4_dir_entry
*dirent
)
438 struct dx_countlimit
*c
;
440 int count_offset
, limit
, count
;
442 if (!ext4_has_metadata_csum(inode
->i_sb
))
445 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
447 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
450 limit
= le16_to_cpu(c
->limit
);
451 count
= le16_to_cpu(c
->count
);
452 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
453 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
454 warn_no_space_for_csum(inode
);
457 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
459 if (t
->dt_checksum
!= ext4_dx_csum(inode
, dirent
, count_offset
,
465 static void ext4_dx_csum_set(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
467 struct dx_countlimit
*c
;
469 int count_offset
, limit
, count
;
471 if (!ext4_has_metadata_csum(inode
->i_sb
))
474 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
476 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
479 limit
= le16_to_cpu(c
->limit
);
480 count
= le16_to_cpu(c
->count
);
481 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
482 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
483 warn_no_space_for_csum(inode
);
486 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
488 t
->dt_checksum
= ext4_dx_csum(inode
, dirent
, count_offset
, count
, t
);
491 static inline int ext4_handle_dirty_dx_node(handle_t
*handle
,
493 struct buffer_head
*bh
)
495 ext4_dx_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
496 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
500 * p is at least 6 bytes before the end of page
502 static inline struct ext4_dir_entry_2
*
503 ext4_next_entry(struct ext4_dir_entry_2
*p
, unsigned long blocksize
)
505 return (struct ext4_dir_entry_2
*)((char *)p
+
506 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
514 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
516 return le32_to_cpu(entry
->block
) & 0x0fffffff;
519 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
521 entry
->block
= cpu_to_le32(value
);
524 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
526 return le32_to_cpu(entry
->hash
);
529 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
531 entry
->hash
= cpu_to_le32(value
);
534 static inline unsigned dx_get_count(struct dx_entry
*entries
)
536 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
539 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
541 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
544 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
546 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
549 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
551 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
554 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
556 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize
;
559 if (ext4_has_metadata_csum(dir
->i_sb
))
560 entry_space
-= sizeof(struct dx_tail
);
561 return entry_space
/ sizeof(struct dx_entry
);
564 static inline unsigned dx_node_limit(struct inode
*dir
)
566 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
568 if (ext4_has_metadata_csum(dir
->i_sb
))
569 entry_space
-= sizeof(struct dx_tail
);
570 return entry_space
/ sizeof(struct dx_entry
);
577 static void dx_show_index(char * label
, struct dx_entry
*entries
)
579 int i
, n
= dx_get_count (entries
);
580 printk(KERN_DEBUG
"%s index", label
);
581 for (i
= 0; i
< n
; i
++) {
582 printk(KERN_CONT
" %x->%lu",
583 i
? dx_get_hash(entries
+ i
) : 0,
584 (unsigned long)dx_get_block(entries
+ i
));
586 printk(KERN_CONT
"\n");
596 static struct stats
dx_show_leaf(struct inode
*dir
,
597 struct dx_hash_info
*hinfo
,
598 struct ext4_dir_entry_2
*de
,
599 int size
, int show_names
)
601 unsigned names
= 0, space
= 0;
602 char *base
= (char *) de
;
603 struct dx_hash_info h
= *hinfo
;
606 while ((char *) de
< base
+ size
)
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
615 struct fscrypt_str fname_crypto_str
=
621 if (ext4_encrypted_inode(dir
))
622 res
= fscrypt_get_encryption_info(dir
);
624 printk(KERN_WARNING
"Error setting up"
625 " fname crypto: %d\n", res
);
627 if (!fscrypt_has_encryption_key(dir
)) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de
->name
,
631 printk("%*.s:(U)%x.%u ", len
,
633 (unsigned) ((char *) de
636 struct fscrypt_str de_name
=
637 FSTR_INIT(name
, len
);
639 /* Directory is encrypted */
640 res
= fscrypt_fname_alloc_buffer(
644 printk(KERN_WARNING
"Error "
648 res
= fscrypt_fname_disk_to_usr(dir
,
652 printk(KERN_WARNING
"Error "
653 "converting filename "
659 name
= fname_crypto_str
.name
;
660 len
= fname_crypto_str
.len
;
662 ext4fs_dirhash(de
->name
, de
->name_len
,
664 printk("%*.s:(E)%x.%u ", len
, name
,
665 h
.hash
, (unsigned) ((char *) de
667 fscrypt_fname_free_buffer(
671 int len
= de
->name_len
;
672 char *name
= de
->name
;
673 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
674 printk("%*.s:%x.%u ", len
, name
, h
.hash
,
675 (unsigned) ((char *) de
- base
));
678 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
681 de
= ext4_next_entry(de
, size
);
683 printk(KERN_CONT
"(%i)\n", names
);
684 return (struct stats
) { names
, space
, 1 };
687 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
688 struct dx_entry
*entries
, int levels
)
690 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
691 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
693 struct buffer_head
*bh
;
694 printk("%i indexed blocks...\n", count
);
695 for (i
= 0; i
< count
; i
++, entries
++)
697 ext4_lblk_t block
= dx_get_block(entries
);
698 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
699 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
701 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
702 bh
= ext4_bread(NULL
,dir
, block
, 0);
703 if (!bh
|| IS_ERR(bh
))
706 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
707 dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*)
708 bh
->b_data
, blocksize
, 0);
709 names
+= stats
.names
;
710 space
+= stats
.space
;
711 bcount
+= stats
.bcount
;
715 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
716 levels
? "" : " ", names
, space
/bcount
,
717 (space
/bcount
)*100/blocksize
);
718 return (struct stats
) { names
, space
, bcount
};
720 #endif /* DX_DEBUG */
723 * Probe for a directory leaf block to search.
725 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
726 * error in the directory index, and the caller should fall back to
727 * searching the directory normally. The callers of dx_probe **MUST**
728 * check for this error code, and make sure it never gets reflected
731 static struct dx_frame
*
732 dx_probe(struct ext4_filename
*fname
, struct inode
*dir
,
733 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
)
735 unsigned count
, indirect
;
736 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
737 struct dx_root
*root
;
738 struct dx_frame
*frame
= frame_in
;
739 struct dx_frame
*ret_err
= ERR_PTR(ERR_BAD_DX_DIR
);
742 memset(frame_in
, 0, EXT4_HTREE_LEVEL
* sizeof(frame_in
[0]));
743 frame
->bh
= ext4_read_dirblock(dir
, 0, INDEX
);
744 if (IS_ERR(frame
->bh
))
745 return (struct dx_frame
*) frame
->bh
;
747 root
= (struct dx_root
*) frame
->bh
->b_data
;
748 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
749 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
750 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
751 ext4_warning_inode(dir
, "Unrecognised inode hash code %u",
752 root
->info
.hash_version
);
756 hinfo
= &fname
->hinfo
;
757 hinfo
->hash_version
= root
->info
.hash_version
;
758 if (hinfo
->hash_version
<= DX_HASH_TEA
)
759 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
760 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
761 if (fname
&& fname_name(fname
))
762 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), hinfo
);
765 if (root
->info
.unused_flags
& 1) {
766 ext4_warning_inode(dir
, "Unimplemented hash flags: %#06x",
767 root
->info
.unused_flags
);
771 indirect
= root
->info
.indirect_levels
;
772 if (indirect
>= ext4_dir_htree_level(dir
->i_sb
)) {
773 ext4_warning(dir
->i_sb
,
774 "Directory (ino: %lu) htree depth %#06x exceed"
775 "supported value", dir
->i_ino
,
776 ext4_dir_htree_level(dir
->i_sb
));
777 if (ext4_dir_htree_level(dir
->i_sb
) < EXT4_HTREE_LEVEL
) {
778 ext4_warning(dir
->i_sb
, "Enable large directory "
779 "feature to access it");
784 entries
= (struct dx_entry
*)(((char *)&root
->info
) +
785 root
->info
.info_length
);
787 if (dx_get_limit(entries
) != dx_root_limit(dir
,
788 root
->info
.info_length
)) {
789 ext4_warning_inode(dir
, "dx entry: limit %u != root limit %u",
790 dx_get_limit(entries
),
791 dx_root_limit(dir
, root
->info
.info_length
));
795 dxtrace(printk("Look up %x", hash
));
797 count
= dx_get_count(entries
);
798 if (!count
|| count
> dx_get_limit(entries
)) {
799 ext4_warning_inode(dir
,
800 "dx entry: count %u beyond limit %u",
801 count
, dx_get_limit(entries
));
806 q
= entries
+ count
- 1;
809 dxtrace(printk(KERN_CONT
"."));
810 if (dx_get_hash(m
) > hash
)
816 if (0) { // linear search cross check
817 unsigned n
= count
- 1;
821 dxtrace(printk(KERN_CONT
","));
822 if (dx_get_hash(++at
) > hash
)
828 assert (at
== p
- 1);
832 dxtrace(printk(KERN_CONT
" %x->%u\n",
833 at
== entries
? 0 : dx_get_hash(at
),
835 frame
->entries
= entries
;
840 frame
->bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
841 if (IS_ERR(frame
->bh
)) {
842 ret_err
= (struct dx_frame
*) frame
->bh
;
846 entries
= ((struct dx_node
*) frame
->bh
->b_data
)->entries
;
848 if (dx_get_limit(entries
) != dx_node_limit(dir
)) {
849 ext4_warning_inode(dir
,
850 "dx entry: limit %u != node limit %u",
851 dx_get_limit(entries
), dx_node_limit(dir
));
856 while (frame
>= frame_in
) {
861 if (ret_err
== ERR_PTR(ERR_BAD_DX_DIR
))
862 ext4_warning_inode(dir
,
863 "Corrupt directory, running e2fsck is recommended");
867 static void dx_release(struct dx_frame
*frames
)
869 struct dx_root_info
*info
;
872 if (frames
[0].bh
== NULL
)
875 info
= &((struct dx_root
*)frames
[0].bh
->b_data
)->info
;
876 for (i
= 0; i
<= info
->indirect_levels
; i
++) {
877 if (frames
[i
].bh
== NULL
)
879 brelse(frames
[i
].bh
);
885 * This function increments the frame pointer to search the next leaf
886 * block, and reads in the necessary intervening nodes if the search
887 * should be necessary. Whether or not the search is necessary is
888 * controlled by the hash parameter. If the hash value is even, then
889 * the search is only continued if the next block starts with that
890 * hash value. This is used if we are searching for a specific file.
892 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
894 * This function returns 1 if the caller should continue to search,
895 * or 0 if it should not. If there is an error reading one of the
896 * index blocks, it will a negative error code.
898 * If start_hash is non-null, it will be filled in with the starting
899 * hash of the next page.
901 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
902 struct dx_frame
*frame
,
903 struct dx_frame
*frames
,
907 struct buffer_head
*bh
;
913 * Find the next leaf page by incrementing the frame pointer.
914 * If we run out of entries in the interior node, loop around and
915 * increment pointer in the parent node. When we break out of
916 * this loop, num_frames indicates the number of interior
917 * nodes need to be read.
920 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
929 * If the hash is 1, then continue only if the next page has a
930 * continuation hash of any value. This is used for readdir
931 * handling. Otherwise, check to see if the hash matches the
932 * desired contiuation hash. If it doesn't, return since
933 * there's no point to read in the successive index pages.
935 bhash
= dx_get_hash(p
->at
);
938 if ((hash
& 1) == 0) {
939 if ((bhash
& ~1) != hash
)
943 * If the hash is HASH_NB_ALWAYS, we always go to the next
944 * block so no check is necessary
946 while (num_frames
--) {
947 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
953 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
960 * This function fills a red-black tree with information from a
961 * directory block. It returns the number directory entries loaded
962 * into the tree. If there is an error it is returned in err.
964 static int htree_dirblock_to_tree(struct file
*dir_file
,
965 struct inode
*dir
, ext4_lblk_t block
,
966 struct dx_hash_info
*hinfo
,
967 __u32 start_hash
, __u32 start_minor_hash
)
969 struct buffer_head
*bh
;
970 struct ext4_dir_entry_2
*de
, *top
;
971 int err
= 0, count
= 0;
972 struct fscrypt_str fname_crypto_str
= FSTR_INIT(NULL
, 0), tmp_str
;
974 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
975 (unsigned long)block
));
976 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
980 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
981 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
982 dir
->i_sb
->s_blocksize
-
983 EXT4_DIR_REC_LEN(0));
984 #ifdef CONFIG_EXT4_FS_ENCRYPTION
985 /* Check if the directory is encrypted */
986 if (ext4_encrypted_inode(dir
)) {
987 err
= fscrypt_get_encryption_info(dir
);
992 err
= fscrypt_fname_alloc_buffer(dir
, EXT4_NAME_LEN
,
1000 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
1001 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1002 bh
->b_data
, bh
->b_size
,
1003 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
1004 + ((char *)de
- bh
->b_data
))) {
1005 /* silently ignore the rest of the block */
1008 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
1009 if ((hinfo
->hash
< start_hash
) ||
1010 ((hinfo
->hash
== start_hash
) &&
1011 (hinfo
->minor_hash
< start_minor_hash
)))
1015 if (!ext4_encrypted_inode(dir
)) {
1016 tmp_str
.name
= de
->name
;
1017 tmp_str
.len
= de
->name_len
;
1018 err
= ext4_htree_store_dirent(dir_file
,
1019 hinfo
->hash
, hinfo
->minor_hash
, de
,
1022 int save_len
= fname_crypto_str
.len
;
1023 struct fscrypt_str de_name
= FSTR_INIT(de
->name
,
1026 /* Directory is encrypted */
1027 err
= fscrypt_fname_disk_to_usr(dir
, hinfo
->hash
,
1028 hinfo
->minor_hash
, &de_name
,
1034 err
= ext4_htree_store_dirent(dir_file
,
1035 hinfo
->hash
, hinfo
->minor_hash
, de
,
1037 fname_crypto_str
.len
= save_len
;
1047 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1048 fscrypt_fname_free_buffer(&fname_crypto_str
);
1055 * This function fills a red-black tree with information from a
1056 * directory. We start scanning the directory in hash order, starting
1057 * at start_hash and start_minor_hash.
1059 * This function returns the number of entries inserted into the tree,
1060 * or a negative error code.
1062 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
1063 __u32 start_minor_hash
, __u32
*next_hash
)
1065 struct dx_hash_info hinfo
;
1066 struct ext4_dir_entry_2
*de
;
1067 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
1073 struct fscrypt_str tmp_str
;
1075 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
1076 start_hash
, start_minor_hash
));
1077 dir
= file_inode(dir_file
);
1078 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
1079 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1080 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1081 hinfo
.hash_version
+=
1082 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1083 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1084 if (ext4_has_inline_data(dir
)) {
1085 int has_inline_data
= 1;
1086 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
1090 if (has_inline_data
) {
1095 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
1096 start_hash
, start_minor_hash
);
1100 hinfo
.hash
= start_hash
;
1101 hinfo
.minor_hash
= 0;
1102 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
);
1104 return PTR_ERR(frame
);
1106 /* Add '.' and '..' from the htree header */
1107 if (!start_hash
&& !start_minor_hash
) {
1108 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1109 tmp_str
.name
= de
->name
;
1110 tmp_str
.len
= de
->name_len
;
1111 err
= ext4_htree_store_dirent(dir_file
, 0, 0,
1117 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
1118 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1119 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
1120 tmp_str
.name
= de
->name
;
1121 tmp_str
.len
= de
->name_len
;
1122 err
= ext4_htree_store_dirent(dir_file
, 2, 0,
1130 if (fatal_signal_pending(current
)) {
1135 block
= dx_get_block(frame
->at
);
1136 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
1137 start_hash
, start_minor_hash
);
1144 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
1145 frame
, frames
, &hashval
);
1146 *next_hash
= hashval
;
1152 * Stop if: (a) there are no more entries, or
1153 * (b) we have inserted at least one entry and the
1154 * next hash value is not a continuation
1157 (count
&& ((hashval
& 1) == 0)))
1161 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1162 "next hash: %x\n", count
, *next_hash
));
1169 static inline int search_dirblock(struct buffer_head
*bh
,
1171 struct ext4_filename
*fname
,
1172 unsigned int offset
,
1173 struct ext4_dir_entry_2
**res_dir
)
1175 return ext4_search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1176 fname
, offset
, res_dir
);
1180 * Directory block splitting, compacting
1184 * Create map of hash values, offsets, and sizes, stored at end of block.
1185 * Returns number of entries mapped.
1187 static int dx_make_map(struct inode
*dir
, struct ext4_dir_entry_2
*de
,
1188 unsigned blocksize
, struct dx_hash_info
*hinfo
,
1189 struct dx_map_entry
*map_tail
)
1192 char *base
= (char *) de
;
1193 struct dx_hash_info h
= *hinfo
;
1195 while ((char *) de
< base
+ blocksize
) {
1196 if (de
->name_len
&& de
->inode
) {
1197 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1199 map_tail
->hash
= h
.hash
;
1200 map_tail
->offs
= ((char *) de
- base
)>>2;
1201 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1205 /* XXX: do we need to check rec_len == 0 case? -Chris */
1206 de
= ext4_next_entry(de
, blocksize
);
1211 /* Sort map by hash value */
1212 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1214 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1216 /* Combsort until bubble sort doesn't suck */
1218 count
= count
*10/13;
1219 if (count
- 9 < 2) /* 9, 10 -> 11 */
1221 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1222 if (p
->hash
< q
->hash
)
1225 /* Garden variety bubble sort */
1230 if (q
[1].hash
>= q
[0].hash
)
1238 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1240 struct dx_entry
*entries
= frame
->entries
;
1241 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1242 int count
= dx_get_count(entries
);
1244 assert(count
< dx_get_limit(entries
));
1245 assert(old
< entries
+ count
);
1246 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1247 dx_set_hash(new, hash
);
1248 dx_set_block(new, block
);
1249 dx_set_count(entries
, count
+ 1);
1253 * Test whether a directory entry matches the filename being searched for.
1255 * Return: %true if the directory entry matches, otherwise %false.
1257 static inline bool ext4_match(const struct ext4_filename
*fname
,
1258 const struct ext4_dir_entry_2
*de
)
1260 struct fscrypt_name f
;
1265 f
.usr_fname
= fname
->usr_fname
;
1266 f
.disk_name
= fname
->disk_name
;
1267 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1268 f
.crypto_buf
= fname
->crypto_buf
;
1270 return fscrypt_match_name(&f
, de
->name
, de
->name_len
);
1274 * Returns 0 if not found, -1 on failure, and 1 on success
1276 int ext4_search_dir(struct buffer_head
*bh
, char *search_buf
, int buf_size
,
1277 struct inode
*dir
, struct ext4_filename
*fname
,
1278 unsigned int offset
, struct ext4_dir_entry_2
**res_dir
)
1280 struct ext4_dir_entry_2
* de
;
1284 de
= (struct ext4_dir_entry_2
*)search_buf
;
1285 dlimit
= search_buf
+ buf_size
;
1286 while ((char *) de
< dlimit
) {
1287 /* this code is executed quadratically often */
1288 /* do minimal checking `by hand' */
1289 if ((char *) de
+ de
->name_len
<= dlimit
&&
1290 ext4_match(fname
, de
)) {
1291 /* found a match - just to be sure, do
1293 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, bh
->b_data
,
1294 bh
->b_size
, offset
))
1299 /* prevent looping on a bad block */
1300 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1301 dir
->i_sb
->s_blocksize
);
1305 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1310 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1311 struct ext4_dir_entry
*de
)
1313 struct super_block
*sb
= dir
->i_sb
;
1319 if (de
->inode
== 0 &&
1320 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1329 * finds an entry in the specified directory with the wanted name. It
1330 * returns the cache buffer in which the entry was found, and the entry
1331 * itself (as a parameter - res_dir). It does NOT read the inode of the
1332 * entry - you'll have to do that yourself if you want to.
1334 * The returned buffer_head has ->b_count elevated. The caller is expected
1335 * to brelse() it when appropriate.
1337 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1338 const struct qstr
*d_name
,
1339 struct ext4_dir_entry_2
**res_dir
,
1342 struct super_block
*sb
;
1343 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1344 struct buffer_head
*bh
, *ret
= NULL
;
1345 ext4_lblk_t start
, block
;
1346 const u8
*name
= d_name
->name
;
1347 size_t ra_max
= 0; /* Number of bh's in the readahead
1349 size_t ra_ptr
= 0; /* Current index into readahead
1351 ext4_lblk_t nblocks
;
1352 int i
, namelen
, retval
;
1353 struct ext4_filename fname
;
1357 namelen
= d_name
->len
;
1358 if (namelen
> EXT4_NAME_LEN
)
1361 retval
= ext4_fname_setup_filename(dir
, d_name
, 1, &fname
);
1362 if (retval
== -ENOENT
)
1365 return ERR_PTR(retval
);
1367 if (ext4_has_inline_data(dir
)) {
1368 int has_inline_data
= 1;
1369 ret
= ext4_find_inline_entry(dir
, &fname
, res_dir
,
1371 if (has_inline_data
) {
1374 goto cleanup_and_exit
;
1378 if ((namelen
<= 2) && (name
[0] == '.') &&
1379 (name
[1] == '.' || name
[1] == '\0')) {
1381 * "." or ".." will only be in the first block
1382 * NFS may look up ".."; "." should be handled by the VFS
1389 ret
= ext4_dx_find_entry(dir
, &fname
, res_dir
);
1391 * On success, or if the error was file not found,
1392 * return. Otherwise, fall back to doing a search the
1393 * old fashioned way.
1395 if (!IS_ERR(ret
) || PTR_ERR(ret
) != ERR_BAD_DX_DIR
)
1396 goto cleanup_and_exit
;
1397 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1400 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1401 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1402 if (start
>= nblocks
)
1408 * We deal with the read-ahead logic here.
1410 if (ra_ptr
>= ra_max
) {
1411 /* Refill the readahead buffer */
1414 ra_max
= start
- block
;
1416 ra_max
= nblocks
- block
;
1417 ra_max
= min(ra_max
, ARRAY_SIZE(bh_use
));
1418 retval
= ext4_bread_batch(dir
, block
, ra_max
,
1419 false /* wait */, bh_use
);
1421 ret
= ERR_PTR(retval
);
1423 goto cleanup_and_exit
;
1426 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1429 if (!buffer_uptodate(bh
)) {
1430 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1431 (unsigned long) block
);
1433 ret
= ERR_PTR(-EIO
);
1434 goto cleanup_and_exit
;
1436 if (!buffer_verified(bh
) &&
1437 !is_dx_internal_node(dir
, block
,
1438 (struct ext4_dir_entry
*)bh
->b_data
) &&
1439 !ext4_dirent_csum_verify(dir
,
1440 (struct ext4_dir_entry
*)bh
->b_data
)) {
1441 EXT4_ERROR_INODE(dir
, "checksumming directory "
1442 "block %lu", (unsigned long)block
);
1444 ret
= ERR_PTR(-EFSBADCRC
);
1445 goto cleanup_and_exit
;
1447 set_buffer_verified(bh
);
1448 i
= search_dirblock(bh
, dir
, &fname
,
1449 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1451 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1453 goto cleanup_and_exit
;
1457 goto cleanup_and_exit
;
1460 if (++block
>= nblocks
)
1462 } while (block
!= start
);
1465 * If the directory has grown while we were searching, then
1466 * search the last part of the directory before giving up.
1469 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1470 if (block
< nblocks
) {
1476 /* Clean up the read-ahead blocks */
1477 for (; ra_ptr
< ra_max
; ra_ptr
++)
1478 brelse(bh_use
[ra_ptr
]);
1479 ext4_fname_free_filename(&fname
);
1483 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
1484 struct ext4_filename
*fname
,
1485 struct ext4_dir_entry_2
**res_dir
)
1487 struct super_block
* sb
= dir
->i_sb
;
1488 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
1489 struct buffer_head
*bh
;
1493 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1496 frame
= dx_probe(fname
, dir
, NULL
, frames
);
1498 return (struct buffer_head
*) frame
;
1500 block
= dx_get_block(frame
->at
);
1501 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1505 retval
= search_dirblock(bh
, dir
, fname
,
1506 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1512 bh
= ERR_PTR(ERR_BAD_DX_DIR
);
1516 /* Check to see if we should continue to search */
1517 retval
= ext4_htree_next_block(dir
, fname
->hinfo
.hash
, frame
,
1520 ext4_warning_inode(dir
,
1521 "error %d reading directory index block",
1523 bh
= ERR_PTR(retval
);
1526 } while (retval
== 1);
1530 dxtrace(printk(KERN_DEBUG
"%s not found\n", fname
->usr_fname
->name
));
1536 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1538 struct inode
*inode
;
1539 struct ext4_dir_entry_2
*de
;
1540 struct buffer_head
*bh
;
1542 if (ext4_encrypted_inode(dir
)) {
1543 int res
= fscrypt_get_encryption_info(dir
);
1546 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1547 * created while the directory was encrypted and we
1548 * have access to the key.
1550 if (fscrypt_has_encryption_key(dir
))
1551 fscrypt_set_encrypted_dentry(dentry
);
1552 fscrypt_set_d_op(dentry
);
1553 if (res
&& res
!= -ENOKEY
)
1554 return ERR_PTR(res
);
1557 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1558 return ERR_PTR(-ENAMETOOLONG
);
1560 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1562 return (struct dentry
*) bh
;
1565 __u32 ino
= le32_to_cpu(de
->inode
);
1567 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1568 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1569 return ERR_PTR(-EFSCORRUPTED
);
1571 if (unlikely(ino
== dir
->i_ino
)) {
1572 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1574 return ERR_PTR(-EFSCORRUPTED
);
1576 inode
= ext4_iget_normal(dir
->i_sb
, ino
);
1577 if (inode
== ERR_PTR(-ESTALE
)) {
1578 EXT4_ERROR_INODE(dir
,
1579 "deleted inode referenced: %u",
1581 return ERR_PTR(-EFSCORRUPTED
);
1583 if (!IS_ERR(inode
) && ext4_encrypted_inode(dir
) &&
1584 (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
)) &&
1585 !fscrypt_has_permitted_context(dir
, inode
)) {
1586 ext4_warning(inode
->i_sb
,
1587 "Inconsistent encryption contexts: %lu/%lu",
1588 dir
->i_ino
, inode
->i_ino
);
1590 return ERR_PTR(-EPERM
);
1593 return d_splice_alias(inode
, dentry
);
1597 struct dentry
*ext4_get_parent(struct dentry
*child
)
1600 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1601 struct ext4_dir_entry_2
* de
;
1602 struct buffer_head
*bh
;
1604 bh
= ext4_find_entry(d_inode(child
), &dotdot
, &de
, NULL
);
1606 return (struct dentry
*) bh
;
1608 return ERR_PTR(-ENOENT
);
1609 ino
= le32_to_cpu(de
->inode
);
1612 if (!ext4_valid_inum(child
->d_sb
, ino
)) {
1613 EXT4_ERROR_INODE(d_inode(child
),
1614 "bad parent inode number: %u", ino
);
1615 return ERR_PTR(-EFSCORRUPTED
);
1618 return d_obtain_alias(ext4_iget_normal(child
->d_sb
, ino
));
1622 * Move count entries from end of map between two memory locations.
1623 * Returns pointer to last entry moved.
1625 static struct ext4_dir_entry_2
*
1626 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1629 unsigned rec_len
= 0;
1632 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1633 (from
+ (map
->offs
<<2));
1634 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1635 memcpy (to
, de
, rec_len
);
1636 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1637 ext4_rec_len_to_disk(rec_len
, blocksize
);
1642 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1646 * Compact each dir entry in the range to the minimal rec_len.
1647 * Returns pointer to last entry in range.
1649 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1651 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1652 unsigned rec_len
= 0;
1655 while ((char*)de
< base
+ blocksize
) {
1656 next
= ext4_next_entry(de
, blocksize
);
1657 if (de
->inode
&& de
->name_len
) {
1658 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1660 memmove(to
, de
, rec_len
);
1661 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1663 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1671 * Split a full leaf block to make room for a new dir entry.
1672 * Allocate a new block, and move entries so that they are approx. equally full.
1673 * Returns pointer to de in block into which the new entry will be inserted.
1675 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1676 struct buffer_head
**bh
,struct dx_frame
*frame
,
1677 struct dx_hash_info
*hinfo
)
1679 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1680 unsigned count
, continued
;
1681 struct buffer_head
*bh2
;
1682 ext4_lblk_t newblock
;
1684 struct dx_map_entry
*map
;
1685 char *data1
= (*bh
)->b_data
, *data2
;
1686 unsigned split
, move
, size
;
1687 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1688 struct ext4_dir_entry_tail
*t
;
1692 if (ext4_has_metadata_csum(dir
->i_sb
))
1693 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1695 bh2
= ext4_append(handle
, dir
, &newblock
);
1699 return (struct ext4_dir_entry_2
*) bh2
;
1702 BUFFER_TRACE(*bh
, "get_write_access");
1703 err
= ext4_journal_get_write_access(handle
, *bh
);
1707 BUFFER_TRACE(frame
->bh
, "get_write_access");
1708 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1712 data2
= bh2
->b_data
;
1714 /* create map in the end of data2 block */
1715 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1716 count
= dx_make_map(dir
, (struct ext4_dir_entry_2
*) data1
,
1717 blocksize
, hinfo
, map
);
1719 dx_sort_map(map
, count
);
1720 /* Split the existing block in the middle, size-wise */
1723 for (i
= count
-1; i
>= 0; i
--) {
1724 /* is more than half of this entry in 2nd half of the block? */
1725 if (size
+ map
[i
].size
/2 > blocksize
/2)
1727 size
+= map
[i
].size
;
1730 /* map index at which we will split */
1731 split
= count
- move
;
1732 hash2
= map
[split
].hash
;
1733 continued
= hash2
== map
[split
- 1].hash
;
1734 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1735 (unsigned long)dx_get_block(frame
->at
),
1736 hash2
, split
, count
-split
));
1738 /* Fancy dance to stay within two buffers */
1739 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
,
1741 de
= dx_pack_dirents(data1
, blocksize
);
1742 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1745 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1749 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1750 initialize_dirent_tail(t
, blocksize
);
1752 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1753 initialize_dirent_tail(t
, blocksize
);
1756 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data1
,
1758 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data2
,
1761 /* Which block gets the new entry? */
1762 if (hinfo
->hash
>= hash2
) {
1766 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1767 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1770 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1774 dxtrace(dx_show_index("frame", frame
->entries
));
1781 ext4_std_error(dir
->i_sb
, err
);
1782 return ERR_PTR(err
);
1785 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1786 struct buffer_head
*bh
,
1787 void *buf
, int buf_size
,
1788 struct ext4_filename
*fname
,
1789 struct ext4_dir_entry_2
**dest_de
)
1791 struct ext4_dir_entry_2
*de
;
1792 unsigned short reclen
= EXT4_DIR_REC_LEN(fname_len(fname
));
1794 unsigned int offset
= 0;
1797 de
= (struct ext4_dir_entry_2
*)buf
;
1798 top
= buf
+ buf_size
- reclen
;
1799 while ((char *) de
<= top
) {
1800 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1801 buf
, buf_size
, offset
))
1802 return -EFSCORRUPTED
;
1803 if (ext4_match(fname
, de
))
1805 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1806 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1807 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1809 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1812 if ((char *) de
> top
)
1819 void ext4_insert_dentry(struct inode
*inode
,
1820 struct ext4_dir_entry_2
*de
,
1822 struct ext4_filename
*fname
)
1827 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1828 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1830 struct ext4_dir_entry_2
*de1
=
1831 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1832 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1833 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1836 de
->file_type
= EXT4_FT_UNKNOWN
;
1837 de
->inode
= cpu_to_le32(inode
->i_ino
);
1838 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1839 de
->name_len
= fname_len(fname
);
1840 memcpy(de
->name
, fname_name(fname
), fname_len(fname
));
1844 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1845 * it points to a directory entry which is guaranteed to be large
1846 * enough for new directory entry. If de is NULL, then
1847 * add_dirent_to_buf will attempt search the directory block for
1848 * space. It will return -ENOSPC if no space is available, and -EIO
1849 * and -EEXIST if directory entry already exists.
1851 static int add_dirent_to_buf(handle_t
*handle
, struct ext4_filename
*fname
,
1853 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1854 struct buffer_head
*bh
)
1856 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1860 if (ext4_has_metadata_csum(inode
->i_sb
))
1861 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1864 err
= ext4_find_dest_de(dir
, inode
, bh
, bh
->b_data
,
1865 blocksize
- csum_size
, fname
, &de
);
1869 BUFFER_TRACE(bh
, "get_write_access");
1870 err
= ext4_journal_get_write_access(handle
, bh
);
1872 ext4_std_error(dir
->i_sb
, err
);
1876 /* By now the buffer is marked for journaling */
1877 ext4_insert_dentry(inode
, de
, blocksize
, fname
);
1880 * XXX shouldn't update any times until successful
1881 * completion of syscall, but too many callers depend
1884 * XXX similarly, too many callers depend on
1885 * ext4_new_inode() setting the times, but error
1886 * recovery deletes the inode, so the worst that can
1887 * happen is that the times are slightly out of date
1888 * and/or different from the directory change time.
1890 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
1891 ext4_update_dx_flag(dir
);
1892 inode_inc_iversion(dir
);
1893 ext4_mark_inode_dirty(handle
, dir
);
1894 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1895 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1897 ext4_std_error(dir
->i_sb
, err
);
1902 * This converts a one block unindexed directory to a 3 block indexed
1903 * directory, and adds the dentry to the indexed directory.
1905 static int make_indexed_dir(handle_t
*handle
, struct ext4_filename
*fname
,
1907 struct inode
*inode
, struct buffer_head
*bh
)
1909 struct buffer_head
*bh2
;
1910 struct dx_root
*root
;
1911 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
1912 struct dx_entry
*entries
;
1913 struct ext4_dir_entry_2
*de
, *de2
;
1914 struct ext4_dir_entry_tail
*t
;
1920 struct fake_dirent
*fde
;
1923 if (ext4_has_metadata_csum(inode
->i_sb
))
1924 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1926 blocksize
= dir
->i_sb
->s_blocksize
;
1927 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1928 BUFFER_TRACE(bh
, "get_write_access");
1929 retval
= ext4_journal_get_write_access(handle
, bh
);
1931 ext4_std_error(dir
->i_sb
, retval
);
1935 root
= (struct dx_root
*) bh
->b_data
;
1937 /* The 0th block becomes the root, move the dirents out */
1938 fde
= &root
->dotdot
;
1939 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1940 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1941 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1942 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1944 return -EFSCORRUPTED
;
1946 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1948 /* Allocate new block for the 0th block's dirents */
1949 bh2
= ext4_append(handle
, dir
, &block
);
1952 return PTR_ERR(bh2
);
1954 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1955 data1
= bh2
->b_data
;
1957 memcpy (data1
, de
, len
);
1958 de
= (struct ext4_dir_entry_2
*) data1
;
1960 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1962 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1967 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1968 initialize_dirent_tail(t
, blocksize
);
1971 /* Initialize the root; the dot dirents already exist */
1972 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1973 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1975 memset (&root
->info
, 0, sizeof(root
->info
));
1976 root
->info
.info_length
= sizeof(root
->info
);
1977 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1978 entries
= root
->entries
;
1979 dx_set_block(entries
, 1);
1980 dx_set_count(entries
, 1);
1981 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1983 /* Initialize as for dx_probe */
1984 fname
->hinfo
.hash_version
= root
->info
.hash_version
;
1985 if (fname
->hinfo
.hash_version
<= DX_HASH_TEA
)
1986 fname
->hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1987 fname
->hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1988 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), &fname
->hinfo
);
1990 memset(frames
, 0, sizeof(frames
));
1992 frame
->entries
= entries
;
1993 frame
->at
= entries
;
1996 retval
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1999 retval
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
2003 de
= do_split(handle
,dir
, &bh2
, frame
, &fname
->hinfo
);
2005 retval
= PTR_ERR(de
);
2009 retval
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh2
);
2012 * Even if the block split failed, we have to properly write
2013 * out all the changes we did so far. Otherwise we can end up
2014 * with corrupted filesystem.
2017 ext4_mark_inode_dirty(handle
, dir
);
2026 * adds a file entry to the specified directory, using the same
2027 * semantics as ext4_find_entry(). It returns NULL if it failed.
2029 * NOTE!! The inode part of 'de' is left at 0 - which means you
2030 * may not sleep between calling this and putting something into
2031 * the entry, as someone else might have used it while you slept.
2033 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
2034 struct inode
*inode
)
2036 struct inode
*dir
= d_inode(dentry
->d_parent
);
2037 struct buffer_head
*bh
= NULL
;
2038 struct ext4_dir_entry_2
*de
;
2039 struct ext4_dir_entry_tail
*t
;
2040 struct super_block
*sb
;
2041 struct ext4_filename fname
;
2045 ext4_lblk_t block
, blocks
;
2048 if (ext4_has_metadata_csum(inode
->i_sb
))
2049 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2052 blocksize
= sb
->s_blocksize
;
2053 if (!dentry
->d_name
.len
)
2056 retval
= ext4_fname_setup_filename(dir
, &dentry
->d_name
, 0, &fname
);
2060 if (ext4_has_inline_data(dir
)) {
2061 retval
= ext4_try_add_inline_entry(handle
, &fname
, dir
, inode
);
2071 retval
= ext4_dx_add_entry(handle
, &fname
, dir
, inode
);
2072 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
2074 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
2076 ext4_mark_inode_dirty(handle
, dir
);
2078 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
2079 for (block
= 0; block
< blocks
; block
++) {
2080 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
2082 retval
= PTR_ERR(bh
);
2086 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
,
2088 if (retval
!= -ENOSPC
)
2091 if (blocks
== 1 && !dx_fallback
&&
2092 ext4_has_feature_dir_index(sb
)) {
2093 retval
= make_indexed_dir(handle
, &fname
, dir
,
2095 bh
= NULL
; /* make_indexed_dir releases bh */
2100 bh
= ext4_append(handle
, dir
, &block
);
2102 retval
= PTR_ERR(bh
);
2106 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2108 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
2111 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
2112 initialize_dirent_tail(t
, blocksize
);
2115 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
, de
, bh
);
2117 ext4_fname_free_filename(&fname
);
2120 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
2125 * Returns 0 for success, or a negative error value
2127 static int ext4_dx_add_entry(handle_t
*handle
, struct ext4_filename
*fname
,
2128 struct inode
*dir
, struct inode
*inode
)
2130 struct dx_frame frames
[EXT4_HTREE_LEVEL
], *frame
;
2131 struct dx_entry
*entries
, *at
;
2132 struct buffer_head
*bh
;
2133 struct super_block
*sb
= dir
->i_sb
;
2134 struct ext4_dir_entry_2
*de
;
2140 frame
= dx_probe(fname
, dir
, NULL
, frames
);
2142 return PTR_ERR(frame
);
2143 entries
= frame
->entries
;
2145 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
2152 BUFFER_TRACE(bh
, "get_write_access");
2153 err
= ext4_journal_get_write_access(handle
, bh
);
2157 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, NULL
, bh
);
2162 /* Block full, should compress but for now just split */
2163 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
2164 dx_get_count(entries
), dx_get_limit(entries
)));
2165 /* Need to split index? */
2166 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
2167 ext4_lblk_t newblock
;
2168 int levels
= frame
- frames
+ 1;
2169 unsigned int icount
;
2171 struct dx_entry
*entries2
;
2172 struct dx_node
*node2
;
2173 struct buffer_head
*bh2
;
2175 while (frame
> frames
) {
2176 if (dx_get_count((frame
- 1)->entries
) <
2177 dx_get_limit((frame
- 1)->entries
)) {
2181 frame
--; /* split higher index block */
2183 entries
= frame
->entries
;
2186 if (add_level
&& levels
== ext4_dir_htree_level(sb
)) {
2187 ext4_warning(sb
, "Directory (ino: %lu) index full, "
2188 "reach max htree level :%d",
2189 dir
->i_ino
, levels
);
2190 if (ext4_dir_htree_level(sb
) < EXT4_HTREE_LEVEL
) {
2191 ext4_warning(sb
, "Large directory feature is "
2192 "not enabled on this "
2198 icount
= dx_get_count(entries
);
2199 bh2
= ext4_append(handle
, dir
, &newblock
);
2204 node2
= (struct dx_node
*)(bh2
->b_data
);
2205 entries2
= node2
->entries
;
2206 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2207 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2209 BUFFER_TRACE(frame
->bh
, "get_write_access");
2210 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2214 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2215 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2216 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2219 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2220 err
= ext4_journal_get_write_access(handle
,
2225 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2226 icount2
* sizeof(struct dx_entry
));
2227 dx_set_count(entries
, icount1
);
2228 dx_set_count(entries2
, icount2
);
2229 dx_set_limit(entries2
, dx_node_limit(dir
));
2231 /* Which index block gets the new entry? */
2232 if (at
- entries
>= icount1
) {
2233 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2234 frame
->entries
= entries
= entries2
;
2235 swap(frame
->bh
, bh2
);
2237 dx_insert_block((frame
- 1), hash2
, newblock
);
2238 dxtrace(dx_show_index("node", frame
->entries
));
2239 dxtrace(dx_show_index("node",
2240 ((struct dx_node
*) bh2
->b_data
)->entries
));
2241 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2245 err
= ext4_handle_dirty_dx_node(handle
, dir
,
2250 err
= ext4_handle_dirty_dx_node(handle
, dir
,
2255 struct dx_root
*dxroot
;
2256 memcpy((char *) entries2
, (char *) entries
,
2257 icount
* sizeof(struct dx_entry
));
2258 dx_set_limit(entries2
, dx_node_limit(dir
));
2261 dx_set_count(entries
, 1);
2262 dx_set_block(entries
+ 0, newblock
);
2263 dxroot
= (struct dx_root
*)frames
[0].bh
->b_data
;
2264 dxroot
->info
.indirect_levels
+= 1;
2265 dxtrace(printk(KERN_DEBUG
2266 "Creating %d level index...\n",
2267 info
->indirect_levels
));
2268 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
2271 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2277 de
= do_split(handle
, dir
, &bh
, frame
, &fname
->hinfo
);
2282 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh
);
2286 ext4_std_error(dir
->i_sb
, err
); /* this is a no-op if err == 0 */
2290 /* @restart is true means htree-path has been changed, we need to
2291 * repeat dx_probe() to find out valid htree-path
2293 if (restart
&& err
== 0)
2299 * ext4_generic_delete_entry deletes a directory entry by merging it
2300 * with the previous entry
2302 int ext4_generic_delete_entry(handle_t
*handle
,
2304 struct ext4_dir_entry_2
*de_del
,
2305 struct buffer_head
*bh
,
2310 struct ext4_dir_entry_2
*de
, *pde
;
2311 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2316 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2317 while (i
< buf_size
- csum_size
) {
2318 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2319 bh
->b_data
, bh
->b_size
, i
))
2320 return -EFSCORRUPTED
;
2323 pde
->rec_len
= ext4_rec_len_to_disk(
2324 ext4_rec_len_from_disk(pde
->rec_len
,
2326 ext4_rec_len_from_disk(de
->rec_len
,
2331 inode_inc_iversion(dir
);
2334 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2336 de
= ext4_next_entry(de
, blocksize
);
2341 static int ext4_delete_entry(handle_t
*handle
,
2343 struct ext4_dir_entry_2
*de_del
,
2344 struct buffer_head
*bh
)
2346 int err
, csum_size
= 0;
2348 if (ext4_has_inline_data(dir
)) {
2349 int has_inline_data
= 1;
2350 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2352 if (has_inline_data
)
2356 if (ext4_has_metadata_csum(dir
->i_sb
))
2357 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2359 BUFFER_TRACE(bh
, "get_write_access");
2360 err
= ext4_journal_get_write_access(handle
, bh
);
2364 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2366 dir
->i_sb
->s_blocksize
, csum_size
);
2370 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2371 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2378 ext4_std_error(dir
->i_sb
, err
);
2383 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2384 * since this indicates that nlinks count was previously 1 to avoid overflowing
2385 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2386 * that subdirectory link counts are not being maintained accurately.
2388 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2389 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2390 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2391 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2393 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2397 (inode
->i_nlink
> EXT4_LINK_MAX
|| inode
->i_nlink
== 2))
2398 set_nlink(inode
, 1);
2402 * If a directory had nlink == 1, then we should let it be 1. This indicates
2403 * directory has >EXT4_LINK_MAX subdirs.
2405 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2407 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2412 static int ext4_add_nondir(handle_t
*handle
,
2413 struct dentry
*dentry
, struct inode
*inode
)
2415 int err
= ext4_add_entry(handle
, dentry
, inode
);
2417 ext4_mark_inode_dirty(handle
, inode
);
2418 unlock_new_inode(inode
);
2419 d_instantiate(dentry
, inode
);
2423 unlock_new_inode(inode
);
2429 * By the time this is called, we already have created
2430 * the directory cache entry for the new file, but it
2431 * is so far negative - it has no inode.
2433 * If the create succeeds, we fill in the inode information
2434 * with d_instantiate().
2436 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2440 struct inode
*inode
;
2441 int err
, credits
, retries
= 0;
2443 err
= dquot_initialize(dir
);
2447 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2448 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2450 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2451 NULL
, EXT4_HT_DIR
, credits
);
2452 handle
= ext4_journal_current_handle();
2453 err
= PTR_ERR(inode
);
2454 if (!IS_ERR(inode
)) {
2455 inode
->i_op
= &ext4_file_inode_operations
;
2456 inode
->i_fop
= &ext4_file_operations
;
2457 ext4_set_aops(inode
);
2458 err
= ext4_add_nondir(handle
, dentry
, inode
);
2459 if (!err
&& IS_DIRSYNC(dir
))
2460 ext4_handle_sync(handle
);
2463 ext4_journal_stop(handle
);
2464 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2469 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2470 umode_t mode
, dev_t rdev
)
2473 struct inode
*inode
;
2474 int err
, credits
, retries
= 0;
2476 err
= dquot_initialize(dir
);
2480 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2481 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2483 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2484 NULL
, EXT4_HT_DIR
, credits
);
2485 handle
= ext4_journal_current_handle();
2486 err
= PTR_ERR(inode
);
2487 if (!IS_ERR(inode
)) {
2488 init_special_inode(inode
, inode
->i_mode
, rdev
);
2489 inode
->i_op
= &ext4_special_inode_operations
;
2490 err
= ext4_add_nondir(handle
, dentry
, inode
);
2491 if (!err
&& IS_DIRSYNC(dir
))
2492 ext4_handle_sync(handle
);
2495 ext4_journal_stop(handle
);
2496 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2501 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2504 struct inode
*inode
;
2505 int err
, retries
= 0;
2507 err
= dquot_initialize(dir
);
2512 inode
= ext4_new_inode_start_handle(dir
, mode
,
2515 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2516 4 + EXT4_XATTR_TRANS_BLOCKS
);
2517 handle
= ext4_journal_current_handle();
2518 err
= PTR_ERR(inode
);
2519 if (!IS_ERR(inode
)) {
2520 inode
->i_op
= &ext4_file_inode_operations
;
2521 inode
->i_fop
= &ext4_file_operations
;
2522 ext4_set_aops(inode
);
2523 d_tmpfile(dentry
, inode
);
2524 err
= ext4_orphan_add(handle
, inode
);
2526 goto err_unlock_inode
;
2527 mark_inode_dirty(inode
);
2528 unlock_new_inode(inode
);
2531 ext4_journal_stop(handle
);
2532 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2536 ext4_journal_stop(handle
);
2537 unlock_new_inode(inode
);
2541 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2542 struct ext4_dir_entry_2
*de
,
2543 int blocksize
, int csum_size
,
2544 unsigned int parent_ino
, int dotdot_real_len
)
2546 de
->inode
= cpu_to_le32(inode
->i_ino
);
2548 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2550 strcpy(de
->name
, ".");
2551 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2553 de
= ext4_next_entry(de
, blocksize
);
2554 de
->inode
= cpu_to_le32(parent_ino
);
2556 if (!dotdot_real_len
)
2557 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2558 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2561 de
->rec_len
= ext4_rec_len_to_disk(
2562 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2563 strcpy(de
->name
, "..");
2564 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2566 return ext4_next_entry(de
, blocksize
);
2569 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2570 struct inode
*inode
)
2572 struct buffer_head
*dir_block
= NULL
;
2573 struct ext4_dir_entry_2
*de
;
2574 struct ext4_dir_entry_tail
*t
;
2575 ext4_lblk_t block
= 0;
2576 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2580 if (ext4_has_metadata_csum(dir
->i_sb
))
2581 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2583 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2584 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2585 if (err
< 0 && err
!= -ENOSPC
)
2592 dir_block
= ext4_append(handle
, inode
, &block
);
2593 if (IS_ERR(dir_block
))
2594 return PTR_ERR(dir_block
);
2595 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2596 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2597 set_nlink(inode
, 2);
2599 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2600 initialize_dirent_tail(t
, blocksize
);
2603 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2604 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2607 set_buffer_verified(dir_block
);
2613 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2616 struct inode
*inode
;
2617 int err
, credits
, retries
= 0;
2619 if (EXT4_DIR_LINK_MAX(dir
))
2622 err
= dquot_initialize(dir
);
2626 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2627 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2629 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2631 0, NULL
, EXT4_HT_DIR
, credits
);
2632 handle
= ext4_journal_current_handle();
2633 err
= PTR_ERR(inode
);
2637 inode
->i_op
= &ext4_dir_inode_operations
;
2638 inode
->i_fop
= &ext4_dir_operations
;
2639 err
= ext4_init_new_dir(handle
, dir
, inode
);
2641 goto out_clear_inode
;
2642 err
= ext4_mark_inode_dirty(handle
, inode
);
2644 err
= ext4_add_entry(handle
, dentry
, inode
);
2648 unlock_new_inode(inode
);
2649 ext4_mark_inode_dirty(handle
, inode
);
2653 ext4_inc_count(handle
, dir
);
2654 ext4_update_dx_flag(dir
);
2655 err
= ext4_mark_inode_dirty(handle
, dir
);
2657 goto out_clear_inode
;
2658 unlock_new_inode(inode
);
2659 d_instantiate(dentry
, inode
);
2660 if (IS_DIRSYNC(dir
))
2661 ext4_handle_sync(handle
);
2665 ext4_journal_stop(handle
);
2666 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2672 * routine to check that the specified directory is empty (for rmdir)
2674 bool ext4_empty_dir(struct inode
*inode
)
2676 unsigned int offset
;
2677 struct buffer_head
*bh
;
2678 struct ext4_dir_entry_2
*de
, *de1
;
2679 struct super_block
*sb
;
2681 if (ext4_has_inline_data(inode
)) {
2682 int has_inline_data
= 1;
2685 ret
= empty_inline_dir(inode
, &has_inline_data
);
2686 if (has_inline_data
)
2691 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2692 EXT4_ERROR_INODE(inode
, "invalid size");
2695 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2699 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2700 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2701 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2702 le32_to_cpu(de1
->inode
) == 0 ||
2703 strcmp(".", de
->name
) || strcmp("..", de1
->name
)) {
2704 ext4_warning_inode(inode
, "directory missing '.' and/or '..'");
2708 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2709 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2710 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2711 while (offset
< inode
->i_size
) {
2712 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2713 unsigned int lblock
;
2715 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2716 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2719 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2721 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2722 bh
->b_data
, bh
->b_size
, offset
)) {
2723 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2725 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2728 if (le32_to_cpu(de
->inode
)) {
2732 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2733 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2740 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2741 * such inodes, starting at the superblock, in case we crash before the
2742 * file is closed/deleted, or in case the inode truncate spans multiple
2743 * transactions and the last transaction is not recovered after a crash.
2745 * At filesystem recovery time, we walk this list deleting unlinked
2746 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2748 * Orphan list manipulation functions must be called under i_mutex unless
2749 * we are just creating the inode or deleting it.
2751 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2753 struct super_block
*sb
= inode
->i_sb
;
2754 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2755 struct ext4_iloc iloc
;
2759 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2762 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2763 !inode_is_locked(inode
));
2765 * Exit early if inode already is on orphan list. This is a big speedup
2766 * since we don't have to contend on the global s_orphan_lock.
2768 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2772 * Orphan handling is only valid for files with data blocks
2773 * being truncated, or files being unlinked. Note that we either
2774 * hold i_mutex, or the inode can not be referenced from outside,
2775 * so i_nlink should not be bumped due to race
2777 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2778 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2780 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2781 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2785 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2789 mutex_lock(&sbi
->s_orphan_lock
);
2791 * Due to previous errors inode may be already a part of on-disk
2792 * orphan list. If so skip on-disk list modification.
2794 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2795 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2796 /* Insert this inode at the head of the on-disk orphan list */
2797 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2798 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2801 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2802 mutex_unlock(&sbi
->s_orphan_lock
);
2805 err
= ext4_handle_dirty_super(handle
, sb
);
2806 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2811 * We have to remove inode from in-memory list if
2812 * addition to on disk orphan list failed. Stray orphan
2813 * list entries can cause panics at unmount time.
2815 mutex_lock(&sbi
->s_orphan_lock
);
2816 list_del_init(&EXT4_I(inode
)->i_orphan
);
2817 mutex_unlock(&sbi
->s_orphan_lock
);
2820 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2821 jbd_debug(4, "orphan inode %lu will point to %d\n",
2822 inode
->i_ino
, NEXT_ORPHAN(inode
));
2824 ext4_std_error(sb
, err
);
2829 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2830 * of such inodes stored on disk, because it is finally being cleaned up.
2832 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2834 struct list_head
*prev
;
2835 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2836 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2838 struct ext4_iloc iloc
;
2841 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2844 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2845 !inode_is_locked(inode
));
2846 /* Do this quick check before taking global s_orphan_lock. */
2847 if (list_empty(&ei
->i_orphan
))
2851 /* Grab inode buffer early before taking global s_orphan_lock */
2852 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2855 mutex_lock(&sbi
->s_orphan_lock
);
2856 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2858 prev
= ei
->i_orphan
.prev
;
2859 list_del_init(&ei
->i_orphan
);
2861 /* If we're on an error path, we may not have a valid
2862 * transaction handle with which to update the orphan list on
2863 * disk, but we still need to remove the inode from the linked
2864 * list in memory. */
2865 if (!handle
|| err
) {
2866 mutex_unlock(&sbi
->s_orphan_lock
);
2870 ino_next
= NEXT_ORPHAN(inode
);
2871 if (prev
== &sbi
->s_orphan
) {
2872 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2873 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2874 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2876 mutex_unlock(&sbi
->s_orphan_lock
);
2879 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2880 mutex_unlock(&sbi
->s_orphan_lock
);
2881 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2883 struct ext4_iloc iloc2
;
2884 struct inode
*i_prev
=
2885 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2887 jbd_debug(4, "orphan inode %lu will point to %u\n",
2888 i_prev
->i_ino
, ino_next
);
2889 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2891 mutex_unlock(&sbi
->s_orphan_lock
);
2894 NEXT_ORPHAN(i_prev
) = ino_next
;
2895 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2896 mutex_unlock(&sbi
->s_orphan_lock
);
2900 NEXT_ORPHAN(inode
) = 0;
2901 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2903 ext4_std_error(inode
->i_sb
, err
);
2911 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2914 struct inode
*inode
;
2915 struct buffer_head
*bh
;
2916 struct ext4_dir_entry_2
*de
;
2917 handle_t
*handle
= NULL
;
2919 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
2922 /* Initialize quotas before so that eventual writes go in
2923 * separate transaction */
2924 retval
= dquot_initialize(dir
);
2927 retval
= dquot_initialize(d_inode(dentry
));
2932 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2938 inode
= d_inode(dentry
);
2940 retval
= -EFSCORRUPTED
;
2941 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2944 retval
= -ENOTEMPTY
;
2945 if (!ext4_empty_dir(inode
))
2948 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2949 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2950 if (IS_ERR(handle
)) {
2951 retval
= PTR_ERR(handle
);
2956 if (IS_DIRSYNC(dir
))
2957 ext4_handle_sync(handle
);
2959 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2962 if (!EXT4_DIR_LINK_EMPTY(inode
))
2963 ext4_warning_inode(inode
,
2964 "empty directory '%.*s' has too many links (%u)",
2965 dentry
->d_name
.len
, dentry
->d_name
.name
,
2969 /* There's no need to set i_disksize: the fact that i_nlink is
2970 * zero will ensure that the right thing happens during any
2973 ext4_orphan_add(handle
, inode
);
2974 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2975 ext4_mark_inode_dirty(handle
, inode
);
2976 ext4_dec_count(handle
, dir
);
2977 ext4_update_dx_flag(dir
);
2978 ext4_mark_inode_dirty(handle
, dir
);
2983 ext4_journal_stop(handle
);
2987 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2990 struct inode
*inode
;
2991 struct buffer_head
*bh
;
2992 struct ext4_dir_entry_2
*de
;
2993 handle_t
*handle
= NULL
;
2995 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
2998 trace_ext4_unlink_enter(dir
, dentry
);
2999 /* Initialize quotas before so that eventual writes go
3000 * in separate transaction */
3001 retval
= dquot_initialize(dir
);
3004 retval
= dquot_initialize(d_inode(dentry
));
3009 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
3015 inode
= d_inode(dentry
);
3017 retval
= -EFSCORRUPTED
;
3018 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
3021 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3022 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
3023 if (IS_ERR(handle
)) {
3024 retval
= PTR_ERR(handle
);
3029 if (IS_DIRSYNC(dir
))
3030 ext4_handle_sync(handle
);
3032 if (inode
->i_nlink
== 0) {
3033 ext4_warning_inode(inode
, "Deleting file '%.*s' with no links",
3034 dentry
->d_name
.len
, dentry
->d_name
.name
);
3035 set_nlink(inode
, 1);
3037 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3040 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3041 ext4_update_dx_flag(dir
);
3042 ext4_mark_inode_dirty(handle
, dir
);
3044 if (!inode
->i_nlink
)
3045 ext4_orphan_add(handle
, inode
);
3046 inode
->i_ctime
= current_time(inode
);
3047 ext4_mark_inode_dirty(handle
, inode
);
3052 ext4_journal_stop(handle
);
3053 trace_ext4_unlink_exit(dentry
, retval
);
3057 static int ext4_symlink(struct inode
*dir
,
3058 struct dentry
*dentry
, const char *symname
)
3061 struct inode
*inode
;
3062 int err
, len
= strlen(symname
);
3064 bool encryption_required
;
3065 struct fscrypt_str disk_link
;
3066 struct fscrypt_symlink_data
*sd
= NULL
;
3068 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
3071 disk_link
.len
= len
+ 1;
3072 disk_link
.name
= (char *) symname
;
3074 encryption_required
= (ext4_encrypted_inode(dir
) ||
3075 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir
->i_sb
)));
3076 if (encryption_required
) {
3077 err
= fscrypt_get_encryption_info(dir
);
3080 if (!fscrypt_has_encryption_key(dir
))
3082 disk_link
.len
= (fscrypt_fname_encrypted_size(dir
, len
) +
3083 sizeof(struct fscrypt_symlink_data
));
3084 sd
= kzalloc(disk_link
.len
, GFP_KERNEL
);
3089 if (disk_link
.len
> dir
->i_sb
->s_blocksize
) {
3090 err
= -ENAMETOOLONG
;
3094 err
= dquot_initialize(dir
);
3098 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3100 * For non-fast symlinks, we just allocate inode and put it on
3101 * orphan list in the first transaction => we need bitmap,
3102 * group descriptor, sb, inode block, quota blocks, and
3103 * possibly selinux xattr blocks.
3105 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
3106 EXT4_XATTR_TRANS_BLOCKS
;
3109 * Fast symlink. We have to add entry to directory
3110 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3111 * allocate new inode (bitmap, group descriptor, inode block,
3112 * quota blocks, sb is already counted in previous macros).
3114 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3115 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
3118 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
3119 &dentry
->d_name
, 0, NULL
,
3120 EXT4_HT_DIR
, credits
);
3121 handle
= ext4_journal_current_handle();
3122 if (IS_ERR(inode
)) {
3124 ext4_journal_stop(handle
);
3125 err
= PTR_ERR(inode
);
3129 if (encryption_required
) {
3131 struct fscrypt_str ostr
=
3132 FSTR_INIT(sd
->encrypted_path
, disk_link
.len
);
3134 istr
.name
= (const unsigned char *) symname
;
3136 err
= fscrypt_fname_usr_to_disk(inode
, &istr
, &ostr
);
3138 goto err_drop_inode
;
3139 sd
->len
= cpu_to_le16(ostr
.len
);
3140 disk_link
.name
= (char *) sd
;
3141 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
3144 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3145 if (!encryption_required
)
3146 inode
->i_op
= &ext4_symlink_inode_operations
;
3147 inode_nohighmem(inode
);
3148 ext4_set_aops(inode
);
3150 * We cannot call page_symlink() with transaction started
3151 * because it calls into ext4_write_begin() which can wait
3152 * for transaction commit if we are running out of space
3153 * and thus we deadlock. So we have to stop transaction now
3154 * and restart it when symlink contents is written.
3156 * To keep fs consistent in case of crash, we have to put inode
3157 * to orphan list in the mean time.
3160 err
= ext4_orphan_add(handle
, inode
);
3161 ext4_journal_stop(handle
);
3164 goto err_drop_inode
;
3165 err
= __page_symlink(inode
, disk_link
.name
, disk_link
.len
, 1);
3167 goto err_drop_inode
;
3169 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3170 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3172 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3173 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3174 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
3175 if (IS_ERR(handle
)) {
3176 err
= PTR_ERR(handle
);
3178 goto err_drop_inode
;
3180 set_nlink(inode
, 1);
3181 err
= ext4_orphan_del(handle
, inode
);
3183 goto err_drop_inode
;
3185 /* clear the extent format for fast symlink */
3186 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
3187 if (!encryption_required
) {
3188 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3189 inode
->i_link
= (char *)&EXT4_I(inode
)->i_data
;
3191 memcpy((char *)&EXT4_I(inode
)->i_data
, disk_link
.name
,
3193 inode
->i_size
= disk_link
.len
- 1;
3195 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3196 err
= ext4_add_nondir(handle
, dentry
, inode
);
3197 if (!err
&& IS_DIRSYNC(dir
))
3198 ext4_handle_sync(handle
);
3201 ext4_journal_stop(handle
);
3206 ext4_journal_stop(handle
);
3208 unlock_new_inode(inode
);
3215 static int ext4_link(struct dentry
*old_dentry
,
3216 struct inode
*dir
, struct dentry
*dentry
)
3219 struct inode
*inode
= d_inode(old_dentry
);
3220 int err
, retries
= 0;
3222 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
3224 if (ext4_encrypted_inode(dir
) &&
3225 !fscrypt_has_permitted_context(dir
, inode
))
3228 if ((ext4_test_inode_flag(dir
, EXT4_INODE_PROJINHERIT
)) &&
3229 (!projid_eq(EXT4_I(dir
)->i_projid
,
3230 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3233 err
= dquot_initialize(dir
);
3238 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3239 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3240 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
3242 return PTR_ERR(handle
);
3244 if (IS_DIRSYNC(dir
))
3245 ext4_handle_sync(handle
);
3247 inode
->i_ctime
= current_time(inode
);
3248 ext4_inc_count(handle
, inode
);
3251 err
= ext4_add_entry(handle
, dentry
, inode
);
3253 ext4_mark_inode_dirty(handle
, inode
);
3254 /* this can happen only for tmpfile being
3255 * linked the first time
3257 if (inode
->i_nlink
== 1)
3258 ext4_orphan_del(handle
, inode
);
3259 d_instantiate(dentry
, inode
);
3264 ext4_journal_stop(handle
);
3265 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3272 * Try to find buffer head where contains the parent block.
3273 * It should be the inode block if it is inlined or the 1st block
3274 * if it is a normal dir.
3276 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
3277 struct inode
*inode
,
3279 struct ext4_dir_entry_2
**parent_de
,
3282 struct buffer_head
*bh
;
3284 if (!ext4_has_inline_data(inode
)) {
3285 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
3287 *retval
= PTR_ERR(bh
);
3290 *parent_de
= ext4_next_entry(
3291 (struct ext4_dir_entry_2
*)bh
->b_data
,
3292 inode
->i_sb
->s_blocksize
);
3297 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3300 struct ext4_renament
{
3302 struct dentry
*dentry
;
3303 struct inode
*inode
;
3305 int dir_nlink_delta
;
3307 /* entry for "dentry" */
3308 struct buffer_head
*bh
;
3309 struct ext4_dir_entry_2
*de
;
3312 /* entry for ".." in inode if it's a directory */
3313 struct buffer_head
*dir_bh
;
3314 struct ext4_dir_entry_2
*parent_de
;
3318 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3322 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3323 &retval
, &ent
->parent_de
,
3327 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3328 return -EFSCORRUPTED
;
3329 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3330 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3333 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3338 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3339 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3340 if (!ent
->dir_inlined
) {
3341 if (is_dx(ent
->inode
)) {
3342 retval
= ext4_handle_dirty_dx_node(handle
,
3346 retval
= ext4_handle_dirty_dirent_node(handle
,
3351 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3354 ext4_std_error(ent
->dir
->i_sb
, retval
);
3360 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3361 unsigned ino
, unsigned file_type
)
3365 BUFFER_TRACE(ent
->bh
, "get write access");
3366 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3369 ent
->de
->inode
= cpu_to_le32(ino
);
3370 if (ext4_has_feature_filetype(ent
->dir
->i_sb
))
3371 ent
->de
->file_type
= file_type
;
3372 ent
->dir
->i_version
++;
3373 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3374 current_time(ent
->dir
);
3375 ext4_mark_inode_dirty(handle
, ent
->dir
);
3376 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3377 if (!ent
->inlined
) {
3378 retval
= ext4_handle_dirty_dirent_node(handle
,
3380 if (unlikely(retval
)) {
3381 ext4_std_error(ent
->dir
->i_sb
, retval
);
3391 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3392 const struct qstr
*d_name
)
3394 int retval
= -ENOENT
;
3395 struct buffer_head
*bh
;
3396 struct ext4_dir_entry_2
*de
;
3398 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3402 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3408 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3413 * ent->de could have moved from under us during htree split, so make
3414 * sure that we are deleting the right entry. We might also be pointing
3415 * to a stale entry in the unused part of ent->bh so just checking inum
3416 * and the name isn't enough.
3418 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3419 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3420 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3421 ent
->de
->name_len
) ||
3423 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3424 &ent
->dentry
->d_name
);
3426 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3427 if (retval
== -ENOENT
) {
3428 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3429 &ent
->dentry
->d_name
);
3434 ext4_warning_inode(ent
->dir
,
3435 "Deleting old file: nlink %d, error=%d",
3436 ent
->dir
->i_nlink
, retval
);
3440 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3442 if (ent
->dir_nlink_delta
) {
3443 if (ent
->dir_nlink_delta
== -1)
3444 ext4_dec_count(handle
, ent
->dir
);
3446 ext4_inc_count(handle
, ent
->dir
);
3447 ext4_mark_inode_dirty(handle
, ent
->dir
);
3451 static struct inode
*ext4_whiteout_for_rename(struct ext4_renament
*ent
,
3452 int credits
, handle_t
**h
)
3459 * for inode block, sb block, group summaries,
3462 credits
+= (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent
->dir
->i_sb
) +
3463 EXT4_XATTR_TRANS_BLOCKS
+ 4);
3465 wh
= ext4_new_inode_start_handle(ent
->dir
, S_IFCHR
| WHITEOUT_MODE
,
3466 &ent
->dentry
->d_name
, 0, NULL
,
3467 EXT4_HT_DIR
, credits
);
3469 handle
= ext4_journal_current_handle();
3472 ext4_journal_stop(handle
);
3473 if (PTR_ERR(wh
) == -ENOSPC
&&
3474 ext4_should_retry_alloc(ent
->dir
->i_sb
, &retries
))
3478 init_special_inode(wh
, wh
->i_mode
, WHITEOUT_DEV
);
3479 wh
->i_op
= &ext4_special_inode_operations
;
3485 * Anybody can rename anything with this: the permission checks are left to the
3486 * higher-level routines.
3488 * n.b. old_{dentry,inode) refers to the source dentry/inode
3489 * while new_{dentry,inode) refers to the destination dentry/inode
3490 * This comes from rename(const char *oldpath, const char *newpath)
3492 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3493 struct inode
*new_dir
, struct dentry
*new_dentry
,
3496 handle_t
*handle
= NULL
;
3497 struct ext4_renament old
= {
3499 .dentry
= old_dentry
,
3500 .inode
= d_inode(old_dentry
),
3502 struct ext4_renament
new = {
3504 .dentry
= new_dentry
,
3505 .inode
= d_inode(new_dentry
),
3509 struct inode
*whiteout
= NULL
;
3513 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
)) &&
3514 (!projid_eq(EXT4_I(new_dir
)->i_projid
,
3515 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3518 if ((ext4_encrypted_inode(old_dir
) &&
3519 !fscrypt_has_encryption_key(old_dir
)) ||
3520 (ext4_encrypted_inode(new_dir
) &&
3521 !fscrypt_has_encryption_key(new_dir
)))
3524 retval
= dquot_initialize(old
.dir
);
3527 retval
= dquot_initialize(new.dir
);
3531 /* Initialize quotas before so that eventual writes go
3532 * in separate transaction */
3534 retval
= dquot_initialize(new.inode
);
3539 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3541 return PTR_ERR(old
.bh
);
3543 * Check for inode number is _not_ due to possible IO errors.
3544 * We might rmdir the source, keep it as pwd of some process
3545 * and merrily kill the link to whatever was created under the
3546 * same name. Goodbye sticky bit ;-<
3549 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3552 if ((old
.dir
!= new.dir
) &&
3553 ext4_encrypted_inode(new.dir
) &&
3554 !fscrypt_has_permitted_context(new.dir
, old
.inode
)) {
3559 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3560 &new.de
, &new.inlined
);
3561 if (IS_ERR(new.bh
)) {
3562 retval
= PTR_ERR(new.bh
);
3572 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3573 ext4_alloc_da_blocks(old
.inode
);
3575 credits
= (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3576 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
3577 if (!(flags
& RENAME_WHITEOUT
)) {
3578 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
, credits
);
3579 if (IS_ERR(handle
)) {
3580 retval
= PTR_ERR(handle
);
3585 whiteout
= ext4_whiteout_for_rename(&old
, credits
, &handle
);
3586 if (IS_ERR(whiteout
)) {
3587 retval
= PTR_ERR(whiteout
);
3593 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3594 ext4_handle_sync(handle
);
3596 if (S_ISDIR(old
.inode
->i_mode
)) {
3598 retval
= -ENOTEMPTY
;
3599 if (!ext4_empty_dir(new.inode
))
3603 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3606 retval
= ext4_rename_dir_prepare(handle
, &old
);
3611 * If we're renaming a file within an inline_data dir and adding or
3612 * setting the new dirent causes a conversion from inline_data to
3613 * extents/blockmap, we need to force the dirent delete code to
3614 * re-read the directory, or else we end up trying to delete a dirent
3615 * from what is now the extent tree root (or a block map).
3617 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3618 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3620 old_file_type
= old
.de
->file_type
;
3623 * Do this before adding a new entry, so the old entry is sure
3624 * to be still pointing to the valid old entry.
3626 retval
= ext4_setent(handle
, &old
, whiteout
->i_ino
,
3630 ext4_mark_inode_dirty(handle
, whiteout
);
3633 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3637 retval
= ext4_setent(handle
, &new,
3638 old
.inode
->i_ino
, old_file_type
);
3643 force_reread
= !ext4_test_inode_flag(new.dir
,
3644 EXT4_INODE_INLINE_DATA
);
3647 * Like most other Unix systems, set the ctime for inodes on a
3650 old
.inode
->i_ctime
= current_time(old
.inode
);
3651 ext4_mark_inode_dirty(handle
, old
.inode
);
3657 ext4_rename_delete(handle
, &old
, force_reread
);
3661 ext4_dec_count(handle
, new.inode
);
3662 new.inode
->i_ctime
= current_time(new.inode
);
3664 old
.dir
->i_ctime
= old
.dir
->i_mtime
= current_time(old
.dir
);
3665 ext4_update_dx_flag(old
.dir
);
3667 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3671 ext4_dec_count(handle
, old
.dir
);
3673 /* checked ext4_empty_dir above, can't have another
3674 * parent, ext4_dec_count() won't work for many-linked
3676 clear_nlink(new.inode
);
3678 ext4_inc_count(handle
, new.dir
);
3679 ext4_update_dx_flag(new.dir
);
3680 ext4_mark_inode_dirty(handle
, new.dir
);
3683 ext4_mark_inode_dirty(handle
, old
.dir
);
3685 ext4_mark_inode_dirty(handle
, new.inode
);
3686 if (!new.inode
->i_nlink
)
3687 ext4_orphan_add(handle
, new.inode
);
3697 drop_nlink(whiteout
);
3698 unlock_new_inode(whiteout
);
3702 ext4_journal_stop(handle
);
3706 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3707 struct inode
*new_dir
, struct dentry
*new_dentry
)
3709 handle_t
*handle
= NULL
;
3710 struct ext4_renament old
= {
3712 .dentry
= old_dentry
,
3713 .inode
= d_inode(old_dentry
),
3715 struct ext4_renament
new = {
3717 .dentry
= new_dentry
,
3718 .inode
= d_inode(new_dentry
),
3722 struct timespec ctime
;
3724 if ((ext4_encrypted_inode(old_dir
) &&
3725 !fscrypt_has_encryption_key(old_dir
)) ||
3726 (ext4_encrypted_inode(new_dir
) &&
3727 !fscrypt_has_encryption_key(new_dir
)))
3730 if ((ext4_encrypted_inode(old_dir
) ||
3731 ext4_encrypted_inode(new_dir
)) &&
3732 (old_dir
!= new_dir
) &&
3733 (!fscrypt_has_permitted_context(new_dir
, old
.inode
) ||
3734 !fscrypt_has_permitted_context(old_dir
, new.inode
)))
3737 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
) &&
3738 !projid_eq(EXT4_I(new_dir
)->i_projid
,
3739 EXT4_I(old_dentry
->d_inode
)->i_projid
)) ||
3740 (ext4_test_inode_flag(old_dir
, EXT4_INODE_PROJINHERIT
) &&
3741 !projid_eq(EXT4_I(old_dir
)->i_projid
,
3742 EXT4_I(new_dentry
->d_inode
)->i_projid
)))
3745 retval
= dquot_initialize(old
.dir
);
3748 retval
= dquot_initialize(new.dir
);
3752 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3753 &old
.de
, &old
.inlined
);
3755 return PTR_ERR(old
.bh
);
3757 * Check for inode number is _not_ due to possible IO errors.
3758 * We might rmdir the source, keep it as pwd of some process
3759 * and merrily kill the link to whatever was created under the
3760 * same name. Goodbye sticky bit ;-<
3763 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3766 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3767 &new.de
, &new.inlined
);
3768 if (IS_ERR(new.bh
)) {
3769 retval
= PTR_ERR(new.bh
);
3774 /* RENAME_EXCHANGE case: old *and* new must both exist */
3775 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3778 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3779 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3780 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3781 if (IS_ERR(handle
)) {
3782 retval
= PTR_ERR(handle
);
3787 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3788 ext4_handle_sync(handle
);
3790 if (S_ISDIR(old
.inode
->i_mode
)) {
3792 retval
= ext4_rename_dir_prepare(handle
, &old
);
3796 if (S_ISDIR(new.inode
->i_mode
)) {
3798 retval
= ext4_rename_dir_prepare(handle
, &new);
3804 * Other than the special case of overwriting a directory, parents'
3805 * nlink only needs to be modified if this is a cross directory rename.
3807 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3808 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3809 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3811 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3812 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3816 new_file_type
= new.de
->file_type
;
3817 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3821 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3826 * Like most other Unix systems, set the ctime for inodes on a
3829 ctime
= current_time(old
.inode
);
3830 old
.inode
->i_ctime
= ctime
;
3831 new.inode
->i_ctime
= ctime
;
3832 ext4_mark_inode_dirty(handle
, old
.inode
);
3833 ext4_mark_inode_dirty(handle
, new.inode
);
3836 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3841 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3845 ext4_update_dir_count(handle
, &old
);
3846 ext4_update_dir_count(handle
, &new);
3855 ext4_journal_stop(handle
);
3859 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3860 struct inode
*new_dir
, struct dentry
*new_dentry
,
3863 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir
->i_sb
))))
3866 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3869 if (flags
& RENAME_EXCHANGE
) {
3870 return ext4_cross_rename(old_dir
, old_dentry
,
3871 new_dir
, new_dentry
);
3874 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
3878 * directories can handle most operations...
3880 const struct inode_operations ext4_dir_inode_operations
= {
3881 .create
= ext4_create
,
3882 .lookup
= ext4_lookup
,
3884 .unlink
= ext4_unlink
,
3885 .symlink
= ext4_symlink
,
3886 .mkdir
= ext4_mkdir
,
3887 .rmdir
= ext4_rmdir
,
3888 .mknod
= ext4_mknod
,
3889 .tmpfile
= ext4_tmpfile
,
3890 .rename
= ext4_rename2
,
3891 .setattr
= ext4_setattr
,
3892 .getattr
= ext4_getattr
,
3893 .listxattr
= ext4_listxattr
,
3894 .get_acl
= ext4_get_acl
,
3895 .set_acl
= ext4_set_acl
,
3896 .fiemap
= ext4_fiemap
,
3899 const struct inode_operations ext4_special_inode_operations
= {
3900 .setattr
= ext4_setattr
,
3901 .getattr
= ext4_getattr
,
3902 .listxattr
= ext4_listxattr
,
3903 .get_acl
= ext4_get_acl
,
3904 .set_acl
= ext4_set_acl
,