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
) & 0x00ffffff;
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 frame
->bh
= ext4_read_dirblock(dir
, 0, INDEX
);
743 if (IS_ERR(frame
->bh
))
744 return (struct dx_frame
*) frame
->bh
;
746 root
= (struct dx_root
*) frame
->bh
->b_data
;
747 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
748 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
749 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
750 ext4_warning_inode(dir
, "Unrecognised inode hash code %u",
751 root
->info
.hash_version
);
755 hinfo
= &fname
->hinfo
;
756 hinfo
->hash_version
= root
->info
.hash_version
;
757 if (hinfo
->hash_version
<= DX_HASH_TEA
)
758 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
759 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
760 if (fname
&& fname_name(fname
))
761 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), hinfo
);
764 if (root
->info
.unused_flags
& 1) {
765 ext4_warning_inode(dir
, "Unimplemented hash flags: %#06x",
766 root
->info
.unused_flags
);
770 indirect
= root
->info
.indirect_levels
;
772 ext4_warning_inode(dir
, "Unimplemented hash depth: %#06x",
773 root
->info
.indirect_levels
);
777 entries
= (struct dx_entry
*)(((char *)&root
->info
) +
778 root
->info
.info_length
);
780 if (dx_get_limit(entries
) != dx_root_limit(dir
,
781 root
->info
.info_length
)) {
782 ext4_warning_inode(dir
, "dx entry: limit %u != root limit %u",
783 dx_get_limit(entries
),
784 dx_root_limit(dir
, root
->info
.info_length
));
788 dxtrace(printk("Look up %x", hash
));
790 count
= dx_get_count(entries
);
791 if (!count
|| count
> dx_get_limit(entries
)) {
792 ext4_warning_inode(dir
,
793 "dx entry: count %u beyond limit %u",
794 count
, dx_get_limit(entries
));
799 q
= entries
+ count
- 1;
802 dxtrace(printk(KERN_CONT
"."));
803 if (dx_get_hash(m
) > hash
)
809 if (0) { // linear search cross check
810 unsigned n
= count
- 1;
814 dxtrace(printk(KERN_CONT
","));
815 if (dx_get_hash(++at
) > hash
)
821 assert (at
== p
- 1);
825 dxtrace(printk(KERN_CONT
" %x->%u\n",
826 at
== entries
? 0 : dx_get_hash(at
),
828 frame
->entries
= entries
;
833 frame
->bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
834 if (IS_ERR(frame
->bh
)) {
835 ret_err
= (struct dx_frame
*) frame
->bh
;
839 entries
= ((struct dx_node
*) frame
->bh
->b_data
)->entries
;
841 if (dx_get_limit(entries
) != dx_node_limit(dir
)) {
842 ext4_warning_inode(dir
,
843 "dx entry: limit %u != node limit %u",
844 dx_get_limit(entries
), dx_node_limit(dir
));
849 while (frame
>= frame_in
) {
854 if (ret_err
== ERR_PTR(ERR_BAD_DX_DIR
))
855 ext4_warning_inode(dir
,
856 "Corrupt directory, running e2fsck is recommended");
860 static void dx_release(struct dx_frame
*frames
)
862 if (frames
[0].bh
== NULL
)
865 if (((struct dx_root
*)frames
[0].bh
->b_data
)->info
.indirect_levels
)
866 brelse(frames
[1].bh
);
867 brelse(frames
[0].bh
);
871 * This function increments the frame pointer to search the next leaf
872 * block, and reads in the necessary intervening nodes if the search
873 * should be necessary. Whether or not the search is necessary is
874 * controlled by the hash parameter. If the hash value is even, then
875 * the search is only continued if the next block starts with that
876 * hash value. This is used if we are searching for a specific file.
878 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
880 * This function returns 1 if the caller should continue to search,
881 * or 0 if it should not. If there is an error reading one of the
882 * index blocks, it will a negative error code.
884 * If start_hash is non-null, it will be filled in with the starting
885 * hash of the next page.
887 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
888 struct dx_frame
*frame
,
889 struct dx_frame
*frames
,
893 struct buffer_head
*bh
;
899 * Find the next leaf page by incrementing the frame pointer.
900 * If we run out of entries in the interior node, loop around and
901 * increment pointer in the parent node. When we break out of
902 * this loop, num_frames indicates the number of interior
903 * nodes need to be read.
906 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
915 * If the hash is 1, then continue only if the next page has a
916 * continuation hash of any value. This is used for readdir
917 * handling. Otherwise, check to see if the hash matches the
918 * desired contiuation hash. If it doesn't, return since
919 * there's no point to read in the successive index pages.
921 bhash
= dx_get_hash(p
->at
);
924 if ((hash
& 1) == 0) {
925 if ((bhash
& ~1) != hash
)
929 * If the hash is HASH_NB_ALWAYS, we always go to the next
930 * block so no check is necessary
932 while (num_frames
--) {
933 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
939 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
946 * This function fills a red-black tree with information from a
947 * directory block. It returns the number directory entries loaded
948 * into the tree. If there is an error it is returned in err.
950 static int htree_dirblock_to_tree(struct file
*dir_file
,
951 struct inode
*dir
, ext4_lblk_t block
,
952 struct dx_hash_info
*hinfo
,
953 __u32 start_hash
, __u32 start_minor_hash
)
955 struct buffer_head
*bh
;
956 struct ext4_dir_entry_2
*de
, *top
;
957 int err
= 0, count
= 0;
958 struct fscrypt_str fname_crypto_str
= FSTR_INIT(NULL
, 0), tmp_str
;
960 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
961 (unsigned long)block
));
962 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
966 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
967 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
968 dir
->i_sb
->s_blocksize
-
969 EXT4_DIR_REC_LEN(0));
970 #ifdef CONFIG_EXT4_FS_ENCRYPTION
971 /* Check if the directory is encrypted */
972 if (ext4_encrypted_inode(dir
)) {
973 err
= fscrypt_get_encryption_info(dir
);
978 err
= fscrypt_fname_alloc_buffer(dir
, EXT4_NAME_LEN
,
986 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
987 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
988 bh
->b_data
, bh
->b_size
,
989 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
990 + ((char *)de
- bh
->b_data
))) {
991 /* silently ignore the rest of the block */
994 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
995 if ((hinfo
->hash
< start_hash
) ||
996 ((hinfo
->hash
== start_hash
) &&
997 (hinfo
->minor_hash
< start_minor_hash
)))
1001 if (!ext4_encrypted_inode(dir
)) {
1002 tmp_str
.name
= de
->name
;
1003 tmp_str
.len
= de
->name_len
;
1004 err
= ext4_htree_store_dirent(dir_file
,
1005 hinfo
->hash
, hinfo
->minor_hash
, de
,
1008 int save_len
= fname_crypto_str
.len
;
1009 struct fscrypt_str de_name
= FSTR_INIT(de
->name
,
1012 /* Directory is encrypted */
1013 err
= fscrypt_fname_disk_to_usr(dir
, hinfo
->hash
,
1014 hinfo
->minor_hash
, &de_name
,
1020 err
= ext4_htree_store_dirent(dir_file
,
1021 hinfo
->hash
, hinfo
->minor_hash
, de
,
1023 fname_crypto_str
.len
= save_len
;
1033 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1034 fscrypt_fname_free_buffer(&fname_crypto_str
);
1041 * This function fills a red-black tree with information from a
1042 * directory. We start scanning the directory in hash order, starting
1043 * at start_hash and start_minor_hash.
1045 * This function returns the number of entries inserted into the tree,
1046 * or a negative error code.
1048 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
1049 __u32 start_minor_hash
, __u32
*next_hash
)
1051 struct dx_hash_info hinfo
;
1052 struct ext4_dir_entry_2
*de
;
1053 struct dx_frame frames
[2], *frame
;
1059 struct fscrypt_str tmp_str
;
1061 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
1062 start_hash
, start_minor_hash
));
1063 dir
= file_inode(dir_file
);
1064 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
1065 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1066 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1067 hinfo
.hash_version
+=
1068 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1069 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1070 if (ext4_has_inline_data(dir
)) {
1071 int has_inline_data
= 1;
1072 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
1076 if (has_inline_data
) {
1081 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
1082 start_hash
, start_minor_hash
);
1086 hinfo
.hash
= start_hash
;
1087 hinfo
.minor_hash
= 0;
1088 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
);
1090 return PTR_ERR(frame
);
1092 /* Add '.' and '..' from the htree header */
1093 if (!start_hash
&& !start_minor_hash
) {
1094 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1095 tmp_str
.name
= de
->name
;
1096 tmp_str
.len
= de
->name_len
;
1097 err
= ext4_htree_store_dirent(dir_file
, 0, 0,
1103 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
1104 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1105 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
1106 tmp_str
.name
= de
->name
;
1107 tmp_str
.len
= de
->name_len
;
1108 err
= ext4_htree_store_dirent(dir_file
, 2, 0,
1116 if (fatal_signal_pending(current
)) {
1121 block
= dx_get_block(frame
->at
);
1122 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
1123 start_hash
, start_minor_hash
);
1130 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
1131 frame
, frames
, &hashval
);
1132 *next_hash
= hashval
;
1138 * Stop if: (a) there are no more entries, or
1139 * (b) we have inserted at least one entry and the
1140 * next hash value is not a continuation
1143 (count
&& ((hashval
& 1) == 0)))
1147 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1148 "next hash: %x\n", count
, *next_hash
));
1155 static inline int search_dirblock(struct buffer_head
*bh
,
1157 struct ext4_filename
*fname
,
1158 const struct qstr
*d_name
,
1159 unsigned int offset
,
1160 struct ext4_dir_entry_2
**res_dir
)
1162 return ext4_search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1163 fname
, d_name
, offset
, res_dir
);
1167 * Directory block splitting, compacting
1171 * Create map of hash values, offsets, and sizes, stored at end of block.
1172 * Returns number of entries mapped.
1174 static int dx_make_map(struct inode
*dir
, struct ext4_dir_entry_2
*de
,
1175 unsigned blocksize
, struct dx_hash_info
*hinfo
,
1176 struct dx_map_entry
*map_tail
)
1179 char *base
= (char *) de
;
1180 struct dx_hash_info h
= *hinfo
;
1182 while ((char *) de
< base
+ blocksize
) {
1183 if (de
->name_len
&& de
->inode
) {
1184 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1186 map_tail
->hash
= h
.hash
;
1187 map_tail
->offs
= ((char *) de
- base
)>>2;
1188 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1192 /* XXX: do we need to check rec_len == 0 case? -Chris */
1193 de
= ext4_next_entry(de
, blocksize
);
1198 /* Sort map by hash value */
1199 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1201 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1203 /* Combsort until bubble sort doesn't suck */
1205 count
= count
*10/13;
1206 if (count
- 9 < 2) /* 9, 10 -> 11 */
1208 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1209 if (p
->hash
< q
->hash
)
1212 /* Garden variety bubble sort */
1217 if (q
[1].hash
>= q
[0].hash
)
1225 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1227 struct dx_entry
*entries
= frame
->entries
;
1228 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1229 int count
= dx_get_count(entries
);
1231 assert(count
< dx_get_limit(entries
));
1232 assert(old
< entries
+ count
);
1233 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1234 dx_set_hash(new, hash
);
1235 dx_set_block(new, block
);
1236 dx_set_count(entries
, count
+ 1);
1240 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1242 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1243 * `de != NULL' is guaranteed by caller.
1245 static inline int ext4_match(struct ext4_filename
*fname
,
1246 struct ext4_dir_entry_2
*de
)
1248 const void *name
= fname_name(fname
);
1249 u32 len
= fname_len(fname
);
1254 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1255 if (unlikely(!name
)) {
1256 if (fname
->usr_fname
->name
[0] == '_') {
1258 if (de
->name_len
< 16)
1260 ret
= memcmp(de
->name
+ de
->name_len
- 16,
1261 fname
->crypto_buf
.name
+ 8, 16);
1262 return (ret
== 0) ? 1 : 0;
1264 name
= fname
->crypto_buf
.name
;
1265 len
= fname
->crypto_buf
.len
;
1268 if (de
->name_len
!= len
)
1270 return (memcmp(de
->name
, name
, len
) == 0) ? 1 : 0;
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 const struct qstr
*d_name
,
1279 unsigned int offset
, struct ext4_dir_entry_2
**res_dir
)
1281 struct ext4_dir_entry_2
* de
;
1286 de
= (struct ext4_dir_entry_2
*)search_buf
;
1287 dlimit
= search_buf
+ buf_size
;
1288 while ((char *) de
< dlimit
) {
1289 /* this code is executed quadratically often */
1290 /* do minimal checking `by hand' */
1291 if ((char *) de
+ de
->name_len
<= dlimit
) {
1292 res
= ext4_match(fname
, de
);
1298 /* found a match - just to be sure, do
1300 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1302 bh
->b_size
, offset
)) {
1312 /* prevent looping on a bad block */
1313 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1314 dir
->i_sb
->s_blocksize
);
1320 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1328 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1329 struct ext4_dir_entry
*de
)
1331 struct super_block
*sb
= dir
->i_sb
;
1337 if (de
->inode
== 0 &&
1338 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1347 * finds an entry in the specified directory with the wanted name. It
1348 * returns the cache buffer in which the entry was found, and the entry
1349 * itself (as a parameter - res_dir). It does NOT read the inode of the
1350 * entry - you'll have to do that yourself if you want to.
1352 * The returned buffer_head has ->b_count elevated. The caller is expected
1353 * to brelse() it when appropriate.
1355 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1356 const struct qstr
*d_name
,
1357 struct ext4_dir_entry_2
**res_dir
,
1360 struct super_block
*sb
;
1361 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1362 struct buffer_head
*bh
, *ret
= NULL
;
1363 ext4_lblk_t start
, block
, b
;
1364 const u8
*name
= d_name
->name
;
1365 int ra_max
= 0; /* Number of bh's in the readahead
1367 int ra_ptr
= 0; /* Current index into readahead
1370 ext4_lblk_t nblocks
;
1371 int i
, namelen
, retval
;
1372 struct ext4_filename fname
;
1376 namelen
= d_name
->len
;
1377 if (namelen
> EXT4_NAME_LEN
)
1380 retval
= ext4_fname_setup_filename(dir
, d_name
, 1, &fname
);
1381 if (retval
== -ENOENT
)
1384 return ERR_PTR(retval
);
1386 if (ext4_has_inline_data(dir
)) {
1387 int has_inline_data
= 1;
1388 ret
= ext4_find_inline_entry(dir
, &fname
, d_name
, res_dir
,
1390 if (has_inline_data
) {
1393 goto cleanup_and_exit
;
1397 if ((namelen
<= 2) && (name
[0] == '.') &&
1398 (name
[1] == '.' || name
[1] == '\0')) {
1400 * "." or ".." will only be in the first block
1401 * NFS may look up ".."; "." should be handled by the VFS
1408 ret
= ext4_dx_find_entry(dir
, &fname
, res_dir
);
1410 * On success, or if the error was file not found,
1411 * return. Otherwise, fall back to doing a search the
1412 * old fashioned way.
1414 if (!IS_ERR(ret
) || PTR_ERR(ret
) != ERR_BAD_DX_DIR
)
1415 goto cleanup_and_exit
;
1416 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1419 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1420 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1421 if (start
>= nblocks
)
1427 * We deal with the read-ahead logic here.
1429 if (ra_ptr
>= ra_max
) {
1430 /* Refill the readahead buffer */
1433 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1435 * Terminate if we reach the end of the
1436 * directory and must wrap, or if our
1437 * search has finished at this block.
1439 if (b
>= nblocks
|| (num
&& block
== start
)) {
1440 bh_use
[ra_max
] = NULL
;
1444 bh
= ext4_getblk(NULL
, dir
, b
++, 0);
1448 goto cleanup_and_exit
;
1452 bh_use
[ra_max
] = bh
;
1454 ll_rw_block(REQ_OP_READ
,
1455 REQ_META
| REQ_PRIO
,
1459 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1462 if (!buffer_uptodate(bh
)) {
1463 /* read error, skip block & hope for the best */
1464 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1465 (unsigned long) block
);
1469 if (!buffer_verified(bh
) &&
1470 !is_dx_internal_node(dir
, block
,
1471 (struct ext4_dir_entry
*)bh
->b_data
) &&
1472 !ext4_dirent_csum_verify(dir
,
1473 (struct ext4_dir_entry
*)bh
->b_data
)) {
1474 EXT4_ERROR_INODE(dir
, "checksumming directory "
1475 "block %lu", (unsigned long)block
);
1479 set_buffer_verified(bh
);
1480 i
= search_dirblock(bh
, dir
, &fname
, d_name
,
1481 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1483 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1485 goto cleanup_and_exit
;
1489 goto cleanup_and_exit
;
1492 if (++block
>= nblocks
)
1494 } while (block
!= start
);
1497 * If the directory has grown while we were searching, then
1498 * search the last part of the directory before giving up.
1501 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1502 if (block
< nblocks
) {
1508 /* Clean up the read-ahead blocks */
1509 for (; ra_ptr
< ra_max
; ra_ptr
++)
1510 brelse(bh_use
[ra_ptr
]);
1511 ext4_fname_free_filename(&fname
);
1515 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
1516 struct ext4_filename
*fname
,
1517 struct ext4_dir_entry_2
**res_dir
)
1519 struct super_block
* sb
= dir
->i_sb
;
1520 struct dx_frame frames
[2], *frame
;
1521 const struct qstr
*d_name
= fname
->usr_fname
;
1522 struct buffer_head
*bh
;
1526 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1529 frame
= dx_probe(fname
, dir
, NULL
, frames
);
1531 return (struct buffer_head
*) frame
;
1533 block
= dx_get_block(frame
->at
);
1534 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1538 retval
= search_dirblock(bh
, dir
, fname
, d_name
,
1539 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1545 bh
= ERR_PTR(ERR_BAD_DX_DIR
);
1549 /* Check to see if we should continue to search */
1550 retval
= ext4_htree_next_block(dir
, fname
->hinfo
.hash
, frame
,
1553 ext4_warning_inode(dir
,
1554 "error %d reading directory index block",
1556 bh
= ERR_PTR(retval
);
1559 } while (retval
== 1);
1563 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1569 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1571 struct inode
*inode
;
1572 struct ext4_dir_entry_2
*de
;
1573 struct buffer_head
*bh
;
1575 if (ext4_encrypted_inode(dir
)) {
1576 int res
= fscrypt_get_encryption_info(dir
);
1579 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1580 * created while the directory was encrypted and we
1581 * have access to the key.
1583 if (fscrypt_has_encryption_key(dir
))
1584 fscrypt_set_encrypted_dentry(dentry
);
1585 fscrypt_set_d_op(dentry
);
1586 if (res
&& res
!= -ENOKEY
)
1587 return ERR_PTR(res
);
1590 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1591 return ERR_PTR(-ENAMETOOLONG
);
1593 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1595 return (struct dentry
*) bh
;
1598 __u32 ino
= le32_to_cpu(de
->inode
);
1600 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1601 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1602 return ERR_PTR(-EFSCORRUPTED
);
1604 if (unlikely(ino
== dir
->i_ino
)) {
1605 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1607 return ERR_PTR(-EFSCORRUPTED
);
1609 inode
= ext4_iget_normal(dir
->i_sb
, ino
);
1610 if (inode
== ERR_PTR(-ESTALE
)) {
1611 EXT4_ERROR_INODE(dir
,
1612 "deleted inode referenced: %u",
1614 return ERR_PTR(-EFSCORRUPTED
);
1616 if (!IS_ERR(inode
) && ext4_encrypted_inode(dir
) &&
1617 (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
)) &&
1618 !fscrypt_has_permitted_context(dir
, inode
)) {
1619 int nokey
= ext4_encrypted_inode(inode
) &&
1620 !fscrypt_has_encryption_key(inode
);
1623 return ERR_PTR(-ENOKEY
);
1625 ext4_warning(inode
->i_sb
,
1626 "Inconsistent encryption contexts: %lu/%lu",
1627 (unsigned long) dir
->i_ino
,
1628 (unsigned long) inode
->i_ino
);
1630 return ERR_PTR(-EPERM
);
1633 return d_splice_alias(inode
, dentry
);
1637 struct dentry
*ext4_get_parent(struct dentry
*child
)
1640 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1641 struct ext4_dir_entry_2
* de
;
1642 struct buffer_head
*bh
;
1644 bh
= ext4_find_entry(d_inode(child
), &dotdot
, &de
, NULL
);
1646 return (struct dentry
*) bh
;
1648 return ERR_PTR(-ENOENT
);
1649 ino
= le32_to_cpu(de
->inode
);
1652 if (!ext4_valid_inum(child
->d_sb
, ino
)) {
1653 EXT4_ERROR_INODE(d_inode(child
),
1654 "bad parent inode number: %u", ino
);
1655 return ERR_PTR(-EFSCORRUPTED
);
1658 return d_obtain_alias(ext4_iget_normal(child
->d_sb
, ino
));
1662 * Move count entries from end of map between two memory locations.
1663 * Returns pointer to last entry moved.
1665 static struct ext4_dir_entry_2
*
1666 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1669 unsigned rec_len
= 0;
1672 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1673 (from
+ (map
->offs
<<2));
1674 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1675 memcpy (to
, de
, rec_len
);
1676 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1677 ext4_rec_len_to_disk(rec_len
, blocksize
);
1682 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1686 * Compact each dir entry in the range to the minimal rec_len.
1687 * Returns pointer to last entry in range.
1689 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1691 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1692 unsigned rec_len
= 0;
1695 while ((char*)de
< base
+ blocksize
) {
1696 next
= ext4_next_entry(de
, blocksize
);
1697 if (de
->inode
&& de
->name_len
) {
1698 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1700 memmove(to
, de
, rec_len
);
1701 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1703 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1711 * Split a full leaf block to make room for a new dir entry.
1712 * Allocate a new block, and move entries so that they are approx. equally full.
1713 * Returns pointer to de in block into which the new entry will be inserted.
1715 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1716 struct buffer_head
**bh
,struct dx_frame
*frame
,
1717 struct dx_hash_info
*hinfo
)
1719 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1720 unsigned count
, continued
;
1721 struct buffer_head
*bh2
;
1722 ext4_lblk_t newblock
;
1724 struct dx_map_entry
*map
;
1725 char *data1
= (*bh
)->b_data
, *data2
;
1726 unsigned split
, move
, size
;
1727 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1728 struct ext4_dir_entry_tail
*t
;
1732 if (ext4_has_metadata_csum(dir
->i_sb
))
1733 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1735 bh2
= ext4_append(handle
, dir
, &newblock
);
1739 return (struct ext4_dir_entry_2
*) bh2
;
1742 BUFFER_TRACE(*bh
, "get_write_access");
1743 err
= ext4_journal_get_write_access(handle
, *bh
);
1747 BUFFER_TRACE(frame
->bh
, "get_write_access");
1748 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1752 data2
= bh2
->b_data
;
1754 /* create map in the end of data2 block */
1755 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1756 count
= dx_make_map(dir
, (struct ext4_dir_entry_2
*) data1
,
1757 blocksize
, hinfo
, map
);
1759 dx_sort_map(map
, count
);
1760 /* Split the existing block in the middle, size-wise */
1763 for (i
= count
-1; i
>= 0; i
--) {
1764 /* is more than half of this entry in 2nd half of the block? */
1765 if (size
+ map
[i
].size
/2 > blocksize
/2)
1767 size
+= map
[i
].size
;
1770 /* map index at which we will split */
1771 split
= count
- move
;
1772 hash2
= map
[split
].hash
;
1773 continued
= hash2
== map
[split
- 1].hash
;
1774 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1775 (unsigned long)dx_get_block(frame
->at
),
1776 hash2
, split
, count
-split
));
1778 /* Fancy dance to stay within two buffers */
1779 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
,
1781 de
= dx_pack_dirents(data1
, blocksize
);
1782 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1785 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1789 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1790 initialize_dirent_tail(t
, blocksize
);
1792 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1793 initialize_dirent_tail(t
, blocksize
);
1796 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data1
,
1798 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data2
,
1801 /* Which block gets the new entry? */
1802 if (hinfo
->hash
>= hash2
) {
1806 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1807 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1810 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1814 dxtrace(dx_show_index("frame", frame
->entries
));
1821 ext4_std_error(dir
->i_sb
, err
);
1822 return ERR_PTR(err
);
1825 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1826 struct buffer_head
*bh
,
1827 void *buf
, int buf_size
,
1828 struct ext4_filename
*fname
,
1829 struct ext4_dir_entry_2
**dest_de
)
1831 struct ext4_dir_entry_2
*de
;
1832 unsigned short reclen
= EXT4_DIR_REC_LEN(fname_len(fname
));
1834 unsigned int offset
= 0;
1838 de
= (struct ext4_dir_entry_2
*)buf
;
1839 top
= buf
+ buf_size
- reclen
;
1840 while ((char *) de
<= top
) {
1841 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1842 buf
, buf_size
, offset
)) {
1843 res
= -EFSCORRUPTED
;
1846 /* Provide crypto context and crypto buffer to ext4 match */
1847 res
= ext4_match(fname
, de
);
1854 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1855 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1856 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1858 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1862 if ((char *) de
> top
)
1872 int ext4_insert_dentry(struct inode
*dir
,
1873 struct inode
*inode
,
1874 struct ext4_dir_entry_2
*de
,
1876 struct ext4_filename
*fname
)
1881 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1882 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1884 struct ext4_dir_entry_2
*de1
=
1885 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1886 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1887 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1890 de
->file_type
= EXT4_FT_UNKNOWN
;
1891 de
->inode
= cpu_to_le32(inode
->i_ino
);
1892 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1893 de
->name_len
= fname_len(fname
);
1894 memcpy(de
->name
, fname_name(fname
), fname_len(fname
));
1899 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1900 * it points to a directory entry which is guaranteed to be large
1901 * enough for new directory entry. If de is NULL, then
1902 * add_dirent_to_buf will attempt search the directory block for
1903 * space. It will return -ENOSPC if no space is available, and -EIO
1904 * and -EEXIST if directory entry already exists.
1906 static int add_dirent_to_buf(handle_t
*handle
, struct ext4_filename
*fname
,
1908 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1909 struct buffer_head
*bh
)
1911 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1915 if (ext4_has_metadata_csum(inode
->i_sb
))
1916 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1919 err
= ext4_find_dest_de(dir
, inode
, bh
, bh
->b_data
,
1920 blocksize
- csum_size
, fname
, &de
);
1924 BUFFER_TRACE(bh
, "get_write_access");
1925 err
= ext4_journal_get_write_access(handle
, bh
);
1927 ext4_std_error(dir
->i_sb
, err
);
1931 /* By now the buffer is marked for journaling. Due to crypto operations,
1932 * the following function call may fail */
1933 err
= ext4_insert_dentry(dir
, inode
, de
, blocksize
, fname
);
1938 * XXX shouldn't update any times until successful
1939 * completion of syscall, but too many callers depend
1942 * XXX similarly, too many callers depend on
1943 * ext4_new_inode() setting the times, but error
1944 * recovery deletes the inode, so the worst that can
1945 * happen is that the times are slightly out of date
1946 * and/or different from the directory change time.
1948 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
1949 ext4_update_dx_flag(dir
);
1951 ext4_mark_inode_dirty(handle
, dir
);
1952 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1953 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1955 ext4_std_error(dir
->i_sb
, err
);
1960 * This converts a one block unindexed directory to a 3 block indexed
1961 * directory, and adds the dentry to the indexed directory.
1963 static int make_indexed_dir(handle_t
*handle
, struct ext4_filename
*fname
,
1965 struct inode
*inode
, struct buffer_head
*bh
)
1967 struct buffer_head
*bh2
;
1968 struct dx_root
*root
;
1969 struct dx_frame frames
[2], *frame
;
1970 struct dx_entry
*entries
;
1971 struct ext4_dir_entry_2
*de
, *de2
;
1972 struct ext4_dir_entry_tail
*t
;
1978 struct fake_dirent
*fde
;
1981 if (ext4_has_metadata_csum(inode
->i_sb
))
1982 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1984 blocksize
= dir
->i_sb
->s_blocksize
;
1985 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1986 BUFFER_TRACE(bh
, "get_write_access");
1987 retval
= ext4_journal_get_write_access(handle
, bh
);
1989 ext4_std_error(dir
->i_sb
, retval
);
1993 root
= (struct dx_root
*) bh
->b_data
;
1995 /* The 0th block becomes the root, move the dirents out */
1996 fde
= &root
->dotdot
;
1997 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1998 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1999 if ((char *) de
>= (((char *) root
) + blocksize
)) {
2000 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
2002 return -EFSCORRUPTED
;
2004 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
2006 /* Allocate new block for the 0th block's dirents */
2007 bh2
= ext4_append(handle
, dir
, &block
);
2010 return PTR_ERR(bh2
);
2012 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
2013 data1
= bh2
->b_data
;
2015 memcpy (data1
, de
, len
);
2016 de
= (struct ext4_dir_entry_2
*) data1
;
2018 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
2020 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
2025 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
2026 initialize_dirent_tail(t
, blocksize
);
2029 /* Initialize the root; the dot dirents already exist */
2030 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
2031 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
2033 memset (&root
->info
, 0, sizeof(root
->info
));
2034 root
->info
.info_length
= sizeof(root
->info
);
2035 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
2036 entries
= root
->entries
;
2037 dx_set_block(entries
, 1);
2038 dx_set_count(entries
, 1);
2039 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
2041 /* Initialize as for dx_probe */
2042 fname
->hinfo
.hash_version
= root
->info
.hash_version
;
2043 if (fname
->hinfo
.hash_version
<= DX_HASH_TEA
)
2044 fname
->hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
2045 fname
->hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
2046 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), &fname
->hinfo
);
2048 memset(frames
, 0, sizeof(frames
));
2050 frame
->entries
= entries
;
2051 frame
->at
= entries
;
2054 retval
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
2057 retval
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
2061 de
= do_split(handle
,dir
, &bh2
, frame
, &fname
->hinfo
);
2063 retval
= PTR_ERR(de
);
2067 retval
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh2
);
2070 * Even if the block split failed, we have to properly write
2071 * out all the changes we did so far. Otherwise we can end up
2072 * with corrupted filesystem.
2075 ext4_mark_inode_dirty(handle
, dir
);
2084 * adds a file entry to the specified directory, using the same
2085 * semantics as ext4_find_entry(). It returns NULL if it failed.
2087 * NOTE!! The inode part of 'de' is left at 0 - which means you
2088 * may not sleep between calling this and putting something into
2089 * the entry, as someone else might have used it while you slept.
2091 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
2092 struct inode
*inode
)
2094 struct inode
*dir
= d_inode(dentry
->d_parent
);
2095 struct buffer_head
*bh
= NULL
;
2096 struct ext4_dir_entry_2
*de
;
2097 struct ext4_dir_entry_tail
*t
;
2098 struct super_block
*sb
;
2099 struct ext4_filename fname
;
2103 ext4_lblk_t block
, blocks
;
2106 if (ext4_has_metadata_csum(inode
->i_sb
))
2107 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2110 blocksize
= sb
->s_blocksize
;
2111 if (!dentry
->d_name
.len
)
2114 retval
= ext4_fname_setup_filename(dir
, &dentry
->d_name
, 0, &fname
);
2118 if (ext4_has_inline_data(dir
)) {
2119 retval
= ext4_try_add_inline_entry(handle
, &fname
, dir
, inode
);
2129 retval
= ext4_dx_add_entry(handle
, &fname
, dir
, inode
);
2130 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
2132 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
2134 ext4_mark_inode_dirty(handle
, dir
);
2136 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
2137 for (block
= 0; block
< blocks
; block
++) {
2138 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
2140 retval
= PTR_ERR(bh
);
2144 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
,
2146 if (retval
!= -ENOSPC
)
2149 if (blocks
== 1 && !dx_fallback
&&
2150 ext4_has_feature_dir_index(sb
)) {
2151 retval
= make_indexed_dir(handle
, &fname
, dir
,
2153 bh
= NULL
; /* make_indexed_dir releases bh */
2158 bh
= ext4_append(handle
, dir
, &block
);
2160 retval
= PTR_ERR(bh
);
2164 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2166 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
2169 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
2170 initialize_dirent_tail(t
, blocksize
);
2173 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
, de
, bh
);
2175 ext4_fname_free_filename(&fname
);
2178 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
2183 * Returns 0 for success, or a negative error value
2185 static int ext4_dx_add_entry(handle_t
*handle
, struct ext4_filename
*fname
,
2186 struct inode
*dir
, struct inode
*inode
)
2188 struct dx_frame frames
[2], *frame
;
2189 struct dx_entry
*entries
, *at
;
2190 struct buffer_head
*bh
;
2191 struct super_block
*sb
= dir
->i_sb
;
2192 struct ext4_dir_entry_2
*de
;
2195 frame
= dx_probe(fname
, dir
, NULL
, frames
);
2197 return PTR_ERR(frame
);
2198 entries
= frame
->entries
;
2200 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
2207 BUFFER_TRACE(bh
, "get_write_access");
2208 err
= ext4_journal_get_write_access(handle
, bh
);
2212 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, NULL
, bh
);
2216 /* Block full, should compress but for now just split */
2217 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
2218 dx_get_count(entries
), dx_get_limit(entries
)));
2219 /* Need to split index? */
2220 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
2221 ext4_lblk_t newblock
;
2222 unsigned icount
= dx_get_count(entries
);
2223 int levels
= frame
- frames
;
2224 struct dx_entry
*entries2
;
2225 struct dx_node
*node2
;
2226 struct buffer_head
*bh2
;
2228 if (levels
&& (dx_get_count(frames
->entries
) ==
2229 dx_get_limit(frames
->entries
))) {
2230 ext4_warning_inode(dir
, "Directory index full!");
2234 bh2
= ext4_append(handle
, dir
, &newblock
);
2239 node2
= (struct dx_node
*)(bh2
->b_data
);
2240 entries2
= node2
->entries
;
2241 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2242 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2244 BUFFER_TRACE(frame
->bh
, "get_write_access");
2245 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2249 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2250 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2251 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2254 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2255 err
= ext4_journal_get_write_access(handle
,
2260 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2261 icount2
* sizeof(struct dx_entry
));
2262 dx_set_count(entries
, icount1
);
2263 dx_set_count(entries2
, icount2
);
2264 dx_set_limit(entries2
, dx_node_limit(dir
));
2266 /* Which index block gets the new entry? */
2267 if (at
- entries
>= icount1
) {
2268 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2269 frame
->entries
= entries
= entries2
;
2270 swap(frame
->bh
, bh2
);
2272 dx_insert_block(frames
+ 0, hash2
, newblock
);
2273 dxtrace(dx_show_index("node", frames
[1].entries
));
2274 dxtrace(dx_show_index("node",
2275 ((struct dx_node
*) bh2
->b_data
)->entries
));
2276 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2281 dxtrace(printk(KERN_DEBUG
2282 "Creating second level index...\n"));
2283 memcpy((char *) entries2
, (char *) entries
,
2284 icount
* sizeof(struct dx_entry
));
2285 dx_set_limit(entries2
, dx_node_limit(dir
));
2288 dx_set_count(entries
, 1);
2289 dx_set_block(entries
+ 0, newblock
);
2290 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2292 /* Add new access path frame */
2294 frame
->at
= at
= at
- entries
+ entries2
;
2295 frame
->entries
= entries
= entries2
;
2297 err
= ext4_journal_get_write_access(handle
,
2302 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2304 ext4_std_error(inode
->i_sb
, err
);
2308 de
= do_split(handle
, dir
, &bh
, frame
, &fname
->hinfo
);
2313 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh
);
2317 ext4_std_error(dir
->i_sb
, err
);
2325 * ext4_generic_delete_entry deletes a directory entry by merging it
2326 * with the previous entry
2328 int ext4_generic_delete_entry(handle_t
*handle
,
2330 struct ext4_dir_entry_2
*de_del
,
2331 struct buffer_head
*bh
,
2336 struct ext4_dir_entry_2
*de
, *pde
;
2337 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2342 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2343 while (i
< buf_size
- csum_size
) {
2344 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2345 bh
->b_data
, bh
->b_size
, i
))
2346 return -EFSCORRUPTED
;
2349 pde
->rec_len
= ext4_rec_len_to_disk(
2350 ext4_rec_len_from_disk(pde
->rec_len
,
2352 ext4_rec_len_from_disk(de
->rec_len
,
2360 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2362 de
= ext4_next_entry(de
, blocksize
);
2367 static int ext4_delete_entry(handle_t
*handle
,
2369 struct ext4_dir_entry_2
*de_del
,
2370 struct buffer_head
*bh
)
2372 int err
, csum_size
= 0;
2374 if (ext4_has_inline_data(dir
)) {
2375 int has_inline_data
= 1;
2376 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2378 if (has_inline_data
)
2382 if (ext4_has_metadata_csum(dir
->i_sb
))
2383 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2385 BUFFER_TRACE(bh
, "get_write_access");
2386 err
= ext4_journal_get_write_access(handle
, bh
);
2390 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2392 dir
->i_sb
->s_blocksize
, csum_size
);
2396 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2397 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2404 ext4_std_error(dir
->i_sb
, err
);
2409 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2410 * since this indicates that nlinks count was previously 1.
2412 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2415 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2416 /* limit is 16-bit i_links_count */
2417 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2418 set_nlink(inode
, 1);
2419 ext4_set_feature_dir_nlink(inode
->i_sb
);
2425 * If a directory had nlink == 1, then we should let it be 1. This indicates
2426 * directory has >EXT4_LINK_MAX subdirs.
2428 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2430 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2435 static int ext4_add_nondir(handle_t
*handle
,
2436 struct dentry
*dentry
, struct inode
*inode
)
2438 int err
= ext4_add_entry(handle
, dentry
, inode
);
2440 ext4_mark_inode_dirty(handle
, inode
);
2441 unlock_new_inode(inode
);
2442 d_instantiate(dentry
, inode
);
2446 unlock_new_inode(inode
);
2452 * By the time this is called, we already have created
2453 * the directory cache entry for the new file, but it
2454 * is so far negative - it has no inode.
2456 * If the create succeeds, we fill in the inode information
2457 * with d_instantiate().
2459 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2463 struct inode
*inode
;
2464 int err
, credits
, retries
= 0;
2466 err
= dquot_initialize(dir
);
2470 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2471 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2473 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2474 NULL
, EXT4_HT_DIR
, credits
);
2475 handle
= ext4_journal_current_handle();
2476 err
= PTR_ERR(inode
);
2477 if (!IS_ERR(inode
)) {
2478 inode
->i_op
= &ext4_file_inode_operations
;
2479 inode
->i_fop
= &ext4_file_operations
;
2480 ext4_set_aops(inode
);
2481 err
= ext4_add_nondir(handle
, dentry
, inode
);
2482 if (!err
&& IS_DIRSYNC(dir
))
2483 ext4_handle_sync(handle
);
2486 ext4_journal_stop(handle
);
2487 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2492 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2493 umode_t mode
, dev_t rdev
)
2496 struct inode
*inode
;
2497 int err
, credits
, retries
= 0;
2499 err
= dquot_initialize(dir
);
2503 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2504 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2506 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2507 NULL
, EXT4_HT_DIR
, credits
);
2508 handle
= ext4_journal_current_handle();
2509 err
= PTR_ERR(inode
);
2510 if (!IS_ERR(inode
)) {
2511 init_special_inode(inode
, inode
->i_mode
, rdev
);
2512 inode
->i_op
= &ext4_special_inode_operations
;
2513 err
= ext4_add_nondir(handle
, dentry
, inode
);
2514 if (!err
&& IS_DIRSYNC(dir
))
2515 ext4_handle_sync(handle
);
2518 ext4_journal_stop(handle
);
2519 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2524 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2527 struct inode
*inode
;
2528 int err
, retries
= 0;
2530 err
= dquot_initialize(dir
);
2535 inode
= ext4_new_inode_start_handle(dir
, mode
,
2538 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2539 4 + EXT4_XATTR_TRANS_BLOCKS
);
2540 handle
= ext4_journal_current_handle();
2541 err
= PTR_ERR(inode
);
2542 if (!IS_ERR(inode
)) {
2543 inode
->i_op
= &ext4_file_inode_operations
;
2544 inode
->i_fop
= &ext4_file_operations
;
2545 ext4_set_aops(inode
);
2546 d_tmpfile(dentry
, inode
);
2547 err
= ext4_orphan_add(handle
, inode
);
2549 goto err_unlock_inode
;
2550 mark_inode_dirty(inode
);
2551 unlock_new_inode(inode
);
2554 ext4_journal_stop(handle
);
2555 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2559 ext4_journal_stop(handle
);
2560 unlock_new_inode(inode
);
2564 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2565 struct ext4_dir_entry_2
*de
,
2566 int blocksize
, int csum_size
,
2567 unsigned int parent_ino
, int dotdot_real_len
)
2569 de
->inode
= cpu_to_le32(inode
->i_ino
);
2571 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2573 strcpy(de
->name
, ".");
2574 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2576 de
= ext4_next_entry(de
, blocksize
);
2577 de
->inode
= cpu_to_le32(parent_ino
);
2579 if (!dotdot_real_len
)
2580 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2581 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2584 de
->rec_len
= ext4_rec_len_to_disk(
2585 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2586 strcpy(de
->name
, "..");
2587 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2589 return ext4_next_entry(de
, blocksize
);
2592 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2593 struct inode
*inode
)
2595 struct buffer_head
*dir_block
= NULL
;
2596 struct ext4_dir_entry_2
*de
;
2597 struct ext4_dir_entry_tail
*t
;
2598 ext4_lblk_t block
= 0;
2599 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2603 if (ext4_has_metadata_csum(dir
->i_sb
))
2604 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2606 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2607 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2608 if (err
< 0 && err
!= -ENOSPC
)
2615 dir_block
= ext4_append(handle
, inode
, &block
);
2616 if (IS_ERR(dir_block
))
2617 return PTR_ERR(dir_block
);
2618 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2619 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2620 set_nlink(inode
, 2);
2622 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2623 initialize_dirent_tail(t
, blocksize
);
2626 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2627 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2630 set_buffer_verified(dir_block
);
2636 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2639 struct inode
*inode
;
2640 int err
, credits
, retries
= 0;
2642 if (EXT4_DIR_LINK_MAX(dir
))
2645 err
= dquot_initialize(dir
);
2649 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2650 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2652 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2654 0, NULL
, EXT4_HT_DIR
, credits
);
2655 handle
= ext4_journal_current_handle();
2656 err
= PTR_ERR(inode
);
2660 inode
->i_op
= &ext4_dir_inode_operations
;
2661 inode
->i_fop
= &ext4_dir_operations
;
2662 err
= ext4_init_new_dir(handle
, dir
, inode
);
2664 goto out_clear_inode
;
2665 err
= ext4_mark_inode_dirty(handle
, inode
);
2667 err
= ext4_add_entry(handle
, dentry
, inode
);
2671 unlock_new_inode(inode
);
2672 ext4_mark_inode_dirty(handle
, inode
);
2676 ext4_inc_count(handle
, dir
);
2677 ext4_update_dx_flag(dir
);
2678 err
= ext4_mark_inode_dirty(handle
, dir
);
2680 goto out_clear_inode
;
2681 unlock_new_inode(inode
);
2682 d_instantiate(dentry
, inode
);
2683 if (IS_DIRSYNC(dir
))
2684 ext4_handle_sync(handle
);
2688 ext4_journal_stop(handle
);
2689 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2695 * routine to check that the specified directory is empty (for rmdir)
2697 bool ext4_empty_dir(struct inode
*inode
)
2699 unsigned int offset
;
2700 struct buffer_head
*bh
;
2701 struct ext4_dir_entry_2
*de
, *de1
;
2702 struct super_block
*sb
;
2704 if (ext4_has_inline_data(inode
)) {
2705 int has_inline_data
= 1;
2708 ret
= empty_inline_dir(inode
, &has_inline_data
);
2709 if (has_inline_data
)
2714 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2715 EXT4_ERROR_INODE(inode
, "invalid size");
2718 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2722 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2723 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2724 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2725 le32_to_cpu(de1
->inode
) == 0 ||
2726 strcmp(".", de
->name
) || strcmp("..", de1
->name
)) {
2727 ext4_warning_inode(inode
, "directory missing '.' and/or '..'");
2731 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2732 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2733 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2734 while (offset
< inode
->i_size
) {
2735 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2736 unsigned int lblock
;
2738 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2739 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2742 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2744 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2745 bh
->b_data
, bh
->b_size
, offset
)) {
2746 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2748 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2751 if (le32_to_cpu(de
->inode
)) {
2755 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2756 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2763 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2764 * such inodes, starting at the superblock, in case we crash before the
2765 * file is closed/deleted, or in case the inode truncate spans multiple
2766 * transactions and the last transaction is not recovered after a crash.
2768 * At filesystem recovery time, we walk this list deleting unlinked
2769 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2771 * Orphan list manipulation functions must be called under i_mutex unless
2772 * we are just creating the inode or deleting it.
2774 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2776 struct super_block
*sb
= inode
->i_sb
;
2777 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2778 struct ext4_iloc iloc
;
2782 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2785 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2786 !inode_is_locked(inode
));
2788 * Exit early if inode already is on orphan list. This is a big speedup
2789 * since we don't have to contend on the global s_orphan_lock.
2791 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2795 * Orphan handling is only valid for files with data blocks
2796 * being truncated, or files being unlinked. Note that we either
2797 * hold i_mutex, or the inode can not be referenced from outside,
2798 * so i_nlink should not be bumped due to race
2800 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2801 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2803 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2804 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2808 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2812 mutex_lock(&sbi
->s_orphan_lock
);
2814 * Due to previous errors inode may be already a part of on-disk
2815 * orphan list. If so skip on-disk list modification.
2817 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2818 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2819 /* Insert this inode at the head of the on-disk orphan list */
2820 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2821 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2824 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2825 mutex_unlock(&sbi
->s_orphan_lock
);
2828 err
= ext4_handle_dirty_super(handle
, sb
);
2829 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2834 * We have to remove inode from in-memory list if
2835 * addition to on disk orphan list failed. Stray orphan
2836 * list entries can cause panics at unmount time.
2838 mutex_lock(&sbi
->s_orphan_lock
);
2839 list_del_init(&EXT4_I(inode
)->i_orphan
);
2840 mutex_unlock(&sbi
->s_orphan_lock
);
2843 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2844 jbd_debug(4, "orphan inode %lu will point to %d\n",
2845 inode
->i_ino
, NEXT_ORPHAN(inode
));
2847 ext4_std_error(sb
, err
);
2852 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2853 * of such inodes stored on disk, because it is finally being cleaned up.
2855 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2857 struct list_head
*prev
;
2858 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2859 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2861 struct ext4_iloc iloc
;
2864 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2867 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2868 !inode_is_locked(inode
));
2869 /* Do this quick check before taking global s_orphan_lock. */
2870 if (list_empty(&ei
->i_orphan
))
2874 /* Grab inode buffer early before taking global s_orphan_lock */
2875 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2878 mutex_lock(&sbi
->s_orphan_lock
);
2879 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2881 prev
= ei
->i_orphan
.prev
;
2882 list_del_init(&ei
->i_orphan
);
2884 /* If we're on an error path, we may not have a valid
2885 * transaction handle with which to update the orphan list on
2886 * disk, but we still need to remove the inode from the linked
2887 * list in memory. */
2888 if (!handle
|| err
) {
2889 mutex_unlock(&sbi
->s_orphan_lock
);
2893 ino_next
= NEXT_ORPHAN(inode
);
2894 if (prev
== &sbi
->s_orphan
) {
2895 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2896 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2897 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2899 mutex_unlock(&sbi
->s_orphan_lock
);
2902 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2903 mutex_unlock(&sbi
->s_orphan_lock
);
2904 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2906 struct ext4_iloc iloc2
;
2907 struct inode
*i_prev
=
2908 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2910 jbd_debug(4, "orphan inode %lu will point to %u\n",
2911 i_prev
->i_ino
, ino_next
);
2912 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2914 mutex_unlock(&sbi
->s_orphan_lock
);
2917 NEXT_ORPHAN(i_prev
) = ino_next
;
2918 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2919 mutex_unlock(&sbi
->s_orphan_lock
);
2923 NEXT_ORPHAN(inode
) = 0;
2924 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2926 ext4_std_error(inode
->i_sb
, err
);
2934 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2937 struct inode
*inode
;
2938 struct buffer_head
*bh
;
2939 struct ext4_dir_entry_2
*de
;
2940 handle_t
*handle
= NULL
;
2942 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
2945 /* Initialize quotas before so that eventual writes go in
2946 * separate transaction */
2947 retval
= dquot_initialize(dir
);
2950 retval
= dquot_initialize(d_inode(dentry
));
2955 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2961 inode
= d_inode(dentry
);
2963 retval
= -EFSCORRUPTED
;
2964 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2967 retval
= -ENOTEMPTY
;
2968 if (!ext4_empty_dir(inode
))
2971 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2972 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2973 if (IS_ERR(handle
)) {
2974 retval
= PTR_ERR(handle
);
2979 if (IS_DIRSYNC(dir
))
2980 ext4_handle_sync(handle
);
2982 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2985 if (!EXT4_DIR_LINK_EMPTY(inode
))
2986 ext4_warning_inode(inode
,
2987 "empty directory '%.*s' has too many links (%u)",
2988 dentry
->d_name
.len
, dentry
->d_name
.name
,
2992 /* There's no need to set i_disksize: the fact that i_nlink is
2993 * zero will ensure that the right thing happens during any
2996 ext4_orphan_add(handle
, inode
);
2997 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2998 ext4_mark_inode_dirty(handle
, inode
);
2999 ext4_dec_count(handle
, dir
);
3000 ext4_update_dx_flag(dir
);
3001 ext4_mark_inode_dirty(handle
, dir
);
3006 ext4_journal_stop(handle
);
3010 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
3013 struct inode
*inode
;
3014 struct buffer_head
*bh
;
3015 struct ext4_dir_entry_2
*de
;
3016 handle_t
*handle
= NULL
;
3018 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
3021 trace_ext4_unlink_enter(dir
, dentry
);
3022 /* Initialize quotas before so that eventual writes go
3023 * in separate transaction */
3024 retval
= dquot_initialize(dir
);
3027 retval
= dquot_initialize(d_inode(dentry
));
3032 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
3038 inode
= d_inode(dentry
);
3040 retval
= -EFSCORRUPTED
;
3041 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
3044 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3045 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
3046 if (IS_ERR(handle
)) {
3047 retval
= PTR_ERR(handle
);
3052 if (IS_DIRSYNC(dir
))
3053 ext4_handle_sync(handle
);
3055 if (inode
->i_nlink
== 0) {
3056 ext4_warning_inode(inode
, "Deleting file '%.*s' with no links",
3057 dentry
->d_name
.len
, dentry
->d_name
.name
);
3058 set_nlink(inode
, 1);
3060 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3063 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3064 ext4_update_dx_flag(dir
);
3065 ext4_mark_inode_dirty(handle
, dir
);
3067 if (!inode
->i_nlink
)
3068 ext4_orphan_add(handle
, inode
);
3069 inode
->i_ctime
= current_time(inode
);
3070 ext4_mark_inode_dirty(handle
, inode
);
3075 ext4_journal_stop(handle
);
3076 trace_ext4_unlink_exit(dentry
, retval
);
3080 static int ext4_symlink(struct inode
*dir
,
3081 struct dentry
*dentry
, const char *symname
)
3084 struct inode
*inode
;
3085 int err
, len
= strlen(symname
);
3087 bool encryption_required
;
3088 struct fscrypt_str disk_link
;
3089 struct fscrypt_symlink_data
*sd
= NULL
;
3091 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir
->i_sb
))))
3094 disk_link
.len
= len
+ 1;
3095 disk_link
.name
= (char *) symname
;
3097 encryption_required
= (ext4_encrypted_inode(dir
) ||
3098 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir
->i_sb
)));
3099 if (encryption_required
) {
3100 err
= fscrypt_get_encryption_info(dir
);
3103 if (!fscrypt_has_encryption_key(dir
))
3105 disk_link
.len
= (fscrypt_fname_encrypted_size(dir
, len
) +
3106 sizeof(struct fscrypt_symlink_data
));
3107 sd
= kzalloc(disk_link
.len
, GFP_KERNEL
);
3112 if (disk_link
.len
> dir
->i_sb
->s_blocksize
) {
3113 err
= -ENAMETOOLONG
;
3117 err
= dquot_initialize(dir
);
3121 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3123 * For non-fast symlinks, we just allocate inode and put it on
3124 * orphan list in the first transaction => we need bitmap,
3125 * group descriptor, sb, inode block, quota blocks, and
3126 * possibly selinux xattr blocks.
3128 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
3129 EXT4_XATTR_TRANS_BLOCKS
;
3132 * Fast symlink. We have to add entry to directory
3133 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3134 * allocate new inode (bitmap, group descriptor, inode block,
3135 * quota blocks, sb is already counted in previous macros).
3137 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3138 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
3141 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
3142 &dentry
->d_name
, 0, NULL
,
3143 EXT4_HT_DIR
, credits
);
3144 handle
= ext4_journal_current_handle();
3145 if (IS_ERR(inode
)) {
3147 ext4_journal_stop(handle
);
3148 err
= PTR_ERR(inode
);
3152 if (encryption_required
) {
3154 struct fscrypt_str ostr
=
3155 FSTR_INIT(sd
->encrypted_path
, disk_link
.len
);
3157 istr
.name
= (const unsigned char *) symname
;
3159 err
= fscrypt_fname_usr_to_disk(inode
, &istr
, &ostr
);
3161 goto err_drop_inode
;
3162 sd
->len
= cpu_to_le16(ostr
.len
);
3163 disk_link
.name
= (char *) sd
;
3164 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
3167 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3168 if (!encryption_required
)
3169 inode
->i_op
= &ext4_symlink_inode_operations
;
3170 inode_nohighmem(inode
);
3171 ext4_set_aops(inode
);
3173 * We cannot call page_symlink() with transaction started
3174 * because it calls into ext4_write_begin() which can wait
3175 * for transaction commit if we are running out of space
3176 * and thus we deadlock. So we have to stop transaction now
3177 * and restart it when symlink contents is written.
3179 * To keep fs consistent in case of crash, we have to put inode
3180 * to orphan list in the mean time.
3183 err
= ext4_orphan_add(handle
, inode
);
3184 ext4_journal_stop(handle
);
3187 goto err_drop_inode
;
3188 err
= __page_symlink(inode
, disk_link
.name
, disk_link
.len
, 1);
3190 goto err_drop_inode
;
3192 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3193 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3195 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3196 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3197 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
3198 if (IS_ERR(handle
)) {
3199 err
= PTR_ERR(handle
);
3201 goto err_drop_inode
;
3203 set_nlink(inode
, 1);
3204 err
= ext4_orphan_del(handle
, inode
);
3206 goto err_drop_inode
;
3208 /* clear the extent format for fast symlink */
3209 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
3210 if (!encryption_required
) {
3211 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3212 inode
->i_link
= (char *)&EXT4_I(inode
)->i_data
;
3214 memcpy((char *)&EXT4_I(inode
)->i_data
, disk_link
.name
,
3216 inode
->i_size
= disk_link
.len
- 1;
3218 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3219 err
= ext4_add_nondir(handle
, dentry
, inode
);
3220 if (!err
&& IS_DIRSYNC(dir
))
3221 ext4_handle_sync(handle
);
3224 ext4_journal_stop(handle
);
3229 ext4_journal_stop(handle
);
3231 unlock_new_inode(inode
);
3238 static int ext4_link(struct dentry
*old_dentry
,
3239 struct inode
*dir
, struct dentry
*dentry
)
3242 struct inode
*inode
= d_inode(old_dentry
);
3243 int err
, retries
= 0;
3245 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
3247 if (ext4_encrypted_inode(dir
) &&
3248 !fscrypt_has_permitted_context(dir
, inode
))
3251 if ((ext4_test_inode_flag(dir
, EXT4_INODE_PROJINHERIT
)) &&
3252 (!projid_eq(EXT4_I(dir
)->i_projid
,
3253 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3256 err
= dquot_initialize(dir
);
3261 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3262 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3263 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
3265 return PTR_ERR(handle
);
3267 if (IS_DIRSYNC(dir
))
3268 ext4_handle_sync(handle
);
3270 inode
->i_ctime
= current_time(inode
);
3271 ext4_inc_count(handle
, inode
);
3274 err
= ext4_add_entry(handle
, dentry
, inode
);
3276 ext4_mark_inode_dirty(handle
, inode
);
3277 /* this can happen only for tmpfile being
3278 * linked the first time
3280 if (inode
->i_nlink
== 1)
3281 ext4_orphan_del(handle
, inode
);
3282 d_instantiate(dentry
, inode
);
3287 ext4_journal_stop(handle
);
3288 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3295 * Try to find buffer head where contains the parent block.
3296 * It should be the inode block if it is inlined or the 1st block
3297 * if it is a normal dir.
3299 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
3300 struct inode
*inode
,
3302 struct ext4_dir_entry_2
**parent_de
,
3305 struct buffer_head
*bh
;
3307 if (!ext4_has_inline_data(inode
)) {
3308 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
3310 *retval
= PTR_ERR(bh
);
3313 *parent_de
= ext4_next_entry(
3314 (struct ext4_dir_entry_2
*)bh
->b_data
,
3315 inode
->i_sb
->s_blocksize
);
3320 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3323 struct ext4_renament
{
3325 struct dentry
*dentry
;
3326 struct inode
*inode
;
3328 int dir_nlink_delta
;
3330 /* entry for "dentry" */
3331 struct buffer_head
*bh
;
3332 struct ext4_dir_entry_2
*de
;
3335 /* entry for ".." in inode if it's a directory */
3336 struct buffer_head
*dir_bh
;
3337 struct ext4_dir_entry_2
*parent_de
;
3341 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3345 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3346 &retval
, &ent
->parent_de
,
3350 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3351 return -EFSCORRUPTED
;
3352 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3353 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3356 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3361 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3362 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3363 if (!ent
->dir_inlined
) {
3364 if (is_dx(ent
->inode
)) {
3365 retval
= ext4_handle_dirty_dx_node(handle
,
3369 retval
= ext4_handle_dirty_dirent_node(handle
,
3374 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3377 ext4_std_error(ent
->dir
->i_sb
, retval
);
3383 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3384 unsigned ino
, unsigned file_type
)
3388 BUFFER_TRACE(ent
->bh
, "get write access");
3389 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3392 ent
->de
->inode
= cpu_to_le32(ino
);
3393 if (ext4_has_feature_filetype(ent
->dir
->i_sb
))
3394 ent
->de
->file_type
= file_type
;
3395 ent
->dir
->i_version
++;
3396 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3397 current_time(ent
->dir
);
3398 ext4_mark_inode_dirty(handle
, ent
->dir
);
3399 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3400 if (!ent
->inlined
) {
3401 retval
= ext4_handle_dirty_dirent_node(handle
,
3403 if (unlikely(retval
)) {
3404 ext4_std_error(ent
->dir
->i_sb
, retval
);
3414 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3415 const struct qstr
*d_name
)
3417 int retval
= -ENOENT
;
3418 struct buffer_head
*bh
;
3419 struct ext4_dir_entry_2
*de
;
3421 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3425 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3431 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3436 * ent->de could have moved from under us during htree split, so make
3437 * sure that we are deleting the right entry. We might also be pointing
3438 * to a stale entry in the unused part of ent->bh so just checking inum
3439 * and the name isn't enough.
3441 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3442 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3443 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3444 ent
->de
->name_len
) ||
3446 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3447 &ent
->dentry
->d_name
);
3449 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3450 if (retval
== -ENOENT
) {
3451 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3452 &ent
->dentry
->d_name
);
3457 ext4_warning_inode(ent
->dir
,
3458 "Deleting old file: nlink %d, error=%d",
3459 ent
->dir
->i_nlink
, retval
);
3463 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3465 if (ent
->dir_nlink_delta
) {
3466 if (ent
->dir_nlink_delta
== -1)
3467 ext4_dec_count(handle
, ent
->dir
);
3469 ext4_inc_count(handle
, ent
->dir
);
3470 ext4_mark_inode_dirty(handle
, ent
->dir
);
3474 static struct inode
*ext4_whiteout_for_rename(struct ext4_renament
*ent
,
3475 int credits
, handle_t
**h
)
3482 * for inode block, sb block, group summaries,
3485 credits
+= (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent
->dir
->i_sb
) +
3486 EXT4_XATTR_TRANS_BLOCKS
+ 4);
3488 wh
= ext4_new_inode_start_handle(ent
->dir
, S_IFCHR
| WHITEOUT_MODE
,
3489 &ent
->dentry
->d_name
, 0, NULL
,
3490 EXT4_HT_DIR
, credits
);
3492 handle
= ext4_journal_current_handle();
3495 ext4_journal_stop(handle
);
3496 if (PTR_ERR(wh
) == -ENOSPC
&&
3497 ext4_should_retry_alloc(ent
->dir
->i_sb
, &retries
))
3501 init_special_inode(wh
, wh
->i_mode
, WHITEOUT_DEV
);
3502 wh
->i_op
= &ext4_special_inode_operations
;
3508 * Anybody can rename anything with this: the permission checks are left to the
3509 * higher-level routines.
3511 * n.b. old_{dentry,inode) refers to the source dentry/inode
3512 * while new_{dentry,inode) refers to the destination dentry/inode
3513 * This comes from rename(const char *oldpath, const char *newpath)
3515 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3516 struct inode
*new_dir
, struct dentry
*new_dentry
,
3519 handle_t
*handle
= NULL
;
3520 struct ext4_renament old
= {
3522 .dentry
= old_dentry
,
3523 .inode
= d_inode(old_dentry
),
3525 struct ext4_renament
new = {
3527 .dentry
= new_dentry
,
3528 .inode
= d_inode(new_dentry
),
3532 struct inode
*whiteout
= NULL
;
3536 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
)) &&
3537 (!projid_eq(EXT4_I(new_dir
)->i_projid
,
3538 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3541 if ((ext4_encrypted_inode(old_dir
) &&
3542 !fscrypt_has_encryption_key(old_dir
)) ||
3543 (ext4_encrypted_inode(new_dir
) &&
3544 !fscrypt_has_encryption_key(new_dir
)))
3547 retval
= dquot_initialize(old
.dir
);
3550 retval
= dquot_initialize(new.dir
);
3554 /* Initialize quotas before so that eventual writes go
3555 * in separate transaction */
3557 retval
= dquot_initialize(new.inode
);
3562 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3564 return PTR_ERR(old
.bh
);
3566 * Check for inode number is _not_ due to possible IO errors.
3567 * We might rmdir the source, keep it as pwd of some process
3568 * and merrily kill the link to whatever was created under the
3569 * same name. Goodbye sticky bit ;-<
3572 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3575 if ((old
.dir
!= new.dir
) &&
3576 ext4_encrypted_inode(new.dir
) &&
3577 !fscrypt_has_permitted_context(new.dir
, old
.inode
)) {
3582 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3583 &new.de
, &new.inlined
);
3584 if (IS_ERR(new.bh
)) {
3585 retval
= PTR_ERR(new.bh
);
3595 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3596 ext4_alloc_da_blocks(old
.inode
);
3598 credits
= (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3599 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
3600 if (!(flags
& RENAME_WHITEOUT
)) {
3601 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
, credits
);
3602 if (IS_ERR(handle
)) {
3603 retval
= PTR_ERR(handle
);
3608 whiteout
= ext4_whiteout_for_rename(&old
, credits
, &handle
);
3609 if (IS_ERR(whiteout
)) {
3610 retval
= PTR_ERR(whiteout
);
3616 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3617 ext4_handle_sync(handle
);
3619 if (S_ISDIR(old
.inode
->i_mode
)) {
3621 retval
= -ENOTEMPTY
;
3622 if (!ext4_empty_dir(new.inode
))
3626 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3629 retval
= ext4_rename_dir_prepare(handle
, &old
);
3634 * If we're renaming a file within an inline_data dir and adding or
3635 * setting the new dirent causes a conversion from inline_data to
3636 * extents/blockmap, we need to force the dirent delete code to
3637 * re-read the directory, or else we end up trying to delete a dirent
3638 * from what is now the extent tree root (or a block map).
3640 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3641 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3643 old_file_type
= old
.de
->file_type
;
3646 * Do this before adding a new entry, so the old entry is sure
3647 * to be still pointing to the valid old entry.
3649 retval
= ext4_setent(handle
, &old
, whiteout
->i_ino
,
3653 ext4_mark_inode_dirty(handle
, whiteout
);
3656 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3660 retval
= ext4_setent(handle
, &new,
3661 old
.inode
->i_ino
, old_file_type
);
3666 force_reread
= !ext4_test_inode_flag(new.dir
,
3667 EXT4_INODE_INLINE_DATA
);
3670 * Like most other Unix systems, set the ctime for inodes on a
3673 old
.inode
->i_ctime
= current_time(old
.inode
);
3674 ext4_mark_inode_dirty(handle
, old
.inode
);
3680 ext4_rename_delete(handle
, &old
, force_reread
);
3684 ext4_dec_count(handle
, new.inode
);
3685 new.inode
->i_ctime
= current_time(new.inode
);
3687 old
.dir
->i_ctime
= old
.dir
->i_mtime
= current_time(old
.dir
);
3688 ext4_update_dx_flag(old
.dir
);
3690 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3694 ext4_dec_count(handle
, old
.dir
);
3696 /* checked ext4_empty_dir above, can't have another
3697 * parent, ext4_dec_count() won't work for many-linked
3699 clear_nlink(new.inode
);
3701 ext4_inc_count(handle
, new.dir
);
3702 ext4_update_dx_flag(new.dir
);
3703 ext4_mark_inode_dirty(handle
, new.dir
);
3706 ext4_mark_inode_dirty(handle
, old
.dir
);
3708 ext4_mark_inode_dirty(handle
, new.inode
);
3709 if (!new.inode
->i_nlink
)
3710 ext4_orphan_add(handle
, new.inode
);
3720 drop_nlink(whiteout
);
3721 unlock_new_inode(whiteout
);
3725 ext4_journal_stop(handle
);
3729 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3730 struct inode
*new_dir
, struct dentry
*new_dentry
)
3732 handle_t
*handle
= NULL
;
3733 struct ext4_renament old
= {
3735 .dentry
= old_dentry
,
3736 .inode
= d_inode(old_dentry
),
3738 struct ext4_renament
new = {
3740 .dentry
= new_dentry
,
3741 .inode
= d_inode(new_dentry
),
3745 struct timespec ctime
;
3747 if ((ext4_encrypted_inode(old_dir
) &&
3748 !fscrypt_has_encryption_key(old_dir
)) ||
3749 (ext4_encrypted_inode(new_dir
) &&
3750 !fscrypt_has_encryption_key(new_dir
)))
3753 if ((ext4_encrypted_inode(old_dir
) ||
3754 ext4_encrypted_inode(new_dir
)) &&
3755 (old_dir
!= new_dir
) &&
3756 (!fscrypt_has_permitted_context(new_dir
, old
.inode
) ||
3757 !fscrypt_has_permitted_context(old_dir
, new.inode
)))
3760 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
) &&
3761 !projid_eq(EXT4_I(new_dir
)->i_projid
,
3762 EXT4_I(old_dentry
->d_inode
)->i_projid
)) ||
3763 (ext4_test_inode_flag(old_dir
, EXT4_INODE_PROJINHERIT
) &&
3764 !projid_eq(EXT4_I(old_dir
)->i_projid
,
3765 EXT4_I(new_dentry
->d_inode
)->i_projid
)))
3768 retval
= dquot_initialize(old
.dir
);
3771 retval
= dquot_initialize(new.dir
);
3775 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3776 &old
.de
, &old
.inlined
);
3778 return PTR_ERR(old
.bh
);
3780 * Check for inode number is _not_ due to possible IO errors.
3781 * We might rmdir the source, keep it as pwd of some process
3782 * and merrily kill the link to whatever was created under the
3783 * same name. Goodbye sticky bit ;-<
3786 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3789 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3790 &new.de
, &new.inlined
);
3791 if (IS_ERR(new.bh
)) {
3792 retval
= PTR_ERR(new.bh
);
3797 /* RENAME_EXCHANGE case: old *and* new must both exist */
3798 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3801 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3802 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3803 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3804 if (IS_ERR(handle
)) {
3805 retval
= PTR_ERR(handle
);
3810 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3811 ext4_handle_sync(handle
);
3813 if (S_ISDIR(old
.inode
->i_mode
)) {
3815 retval
= ext4_rename_dir_prepare(handle
, &old
);
3819 if (S_ISDIR(new.inode
->i_mode
)) {
3821 retval
= ext4_rename_dir_prepare(handle
, &new);
3827 * Other than the special case of overwriting a directory, parents'
3828 * nlink only needs to be modified if this is a cross directory rename.
3830 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3831 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3832 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3834 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3835 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3839 new_file_type
= new.de
->file_type
;
3840 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3844 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3849 * Like most other Unix systems, set the ctime for inodes on a
3852 ctime
= current_time(old
.inode
);
3853 old
.inode
->i_ctime
= ctime
;
3854 new.inode
->i_ctime
= ctime
;
3855 ext4_mark_inode_dirty(handle
, old
.inode
);
3856 ext4_mark_inode_dirty(handle
, new.inode
);
3859 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3864 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3868 ext4_update_dir_count(handle
, &old
);
3869 ext4_update_dir_count(handle
, &new);
3878 ext4_journal_stop(handle
);
3882 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3883 struct inode
*new_dir
, struct dentry
*new_dentry
,
3886 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir
->i_sb
))))
3889 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3892 if (flags
& RENAME_EXCHANGE
) {
3893 return ext4_cross_rename(old_dir
, old_dentry
,
3894 new_dir
, new_dentry
);
3897 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
3901 * directories can handle most operations...
3903 const struct inode_operations ext4_dir_inode_operations
= {
3904 .create
= ext4_create
,
3905 .lookup
= ext4_lookup
,
3907 .unlink
= ext4_unlink
,
3908 .symlink
= ext4_symlink
,
3909 .mkdir
= ext4_mkdir
,
3910 .rmdir
= ext4_rmdir
,
3911 .mknod
= ext4_mknod
,
3912 .tmpfile
= ext4_tmpfile
,
3913 .rename
= ext4_rename2
,
3914 .setattr
= ext4_setattr
,
3915 .getattr
= ext4_getattr
,
3916 .listxattr
= ext4_listxattr
,
3917 .get_acl
= ext4_get_acl
,
3918 .set_acl
= ext4_set_acl
,
3919 .fiemap
= ext4_fiemap
,
3922 const struct inode_operations ext4_special_inode_operations
= {
3923 .setattr
= ext4_setattr
,
3924 .getattr
= ext4_getattr
,
3925 .listxattr
= ext4_listxattr
,
3926 .get_acl
= ext4_get_acl
,
3927 .set_acl
= ext4_set_acl
,