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("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
583 0, (unsigned long)dx_get_block(entries
+ i
));
595 static struct stats
dx_show_leaf(struct inode
*dir
,
596 struct dx_hash_info
*hinfo
,
597 struct ext4_dir_entry_2
*de
,
598 int size
, int show_names
)
600 unsigned names
= 0, space
= 0;
601 char *base
= (char *) de
;
602 struct dx_hash_info h
= *hinfo
;
605 while ((char *) de
< base
+ size
)
611 #ifdef CONFIG_EXT4_FS_ENCRYPTION
614 struct ext4_str fname_crypto_str
615 = {.name
= NULL
, .len
= 0};
620 if (ext4_encrypted_inode(inode
))
621 res
= ext4_get_encryption_info(dir
);
623 printk(KERN_WARNING
"Error setting up"
624 " fname crypto: %d\n", res
);
627 /* Directory is not encrypted */
628 ext4fs_dirhash(de
->name
,
630 printk("%*.s:(U)%x.%u ", len
,
632 (unsigned) ((char *) de
635 /* Directory is encrypted */
636 res
= ext4_fname_crypto_alloc_buffer(
640 printk(KERN_WARNING
"Error "
646 res
= ext4_fname_disk_to_usr(ctx
, NULL
, de
,
649 printk(KERN_WARNING
"Error "
650 "converting filename "
656 name
= fname_crypto_str
.name
;
657 len
= fname_crypto_str
.len
;
659 ext4fs_dirhash(de
->name
, de
->name_len
,
661 printk("%*.s:(E)%x.%u ", len
, name
,
662 h
.hash
, (unsigned) ((char *) de
664 ext4_fname_crypto_free_buffer(
668 int len
= de
->name_len
;
669 char *name
= de
->name
;
670 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
671 printk("%*.s:%x.%u ", len
, name
, h
.hash
,
672 (unsigned) ((char *) de
- base
));
675 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
678 de
= ext4_next_entry(de
, size
);
680 printk("(%i)\n", names
);
681 return (struct stats
) { names
, space
, 1 };
684 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
685 struct dx_entry
*entries
, int levels
)
687 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
688 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
690 struct buffer_head
*bh
;
691 printk("%i indexed blocks...\n", count
);
692 for (i
= 0; i
< count
; i
++, entries
++)
694 ext4_lblk_t block
= dx_get_block(entries
);
695 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
696 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
698 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
699 bh
= ext4_bread(NULL
,dir
, block
, 0);
700 if (!bh
|| IS_ERR(bh
))
703 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
704 dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*)
705 bh
->b_data
, blocksize
, 0);
706 names
+= stats
.names
;
707 space
+= stats
.space
;
708 bcount
+= stats
.bcount
;
712 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
713 levels
? "" : " ", names
, space
/bcount
,
714 (space
/bcount
)*100/blocksize
);
715 return (struct stats
) { names
, space
, bcount
};
717 #endif /* DX_DEBUG */
720 * Probe for a directory leaf block to search.
722 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
723 * error in the directory index, and the caller should fall back to
724 * searching the directory normally. The callers of dx_probe **MUST**
725 * check for this error code, and make sure it never gets reflected
728 static struct dx_frame
*
729 dx_probe(struct ext4_filename
*fname
, struct inode
*dir
,
730 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
)
732 unsigned count
, indirect
;
733 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
734 struct dx_root
*root
;
735 struct dx_frame
*frame
= frame_in
;
736 struct dx_frame
*ret_err
= ERR_PTR(ERR_BAD_DX_DIR
);
739 frame
->bh
= ext4_read_dirblock(dir
, 0, INDEX
);
740 if (IS_ERR(frame
->bh
))
741 return (struct dx_frame
*) frame
->bh
;
743 root
= (struct dx_root
*) frame
->bh
->b_data
;
744 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
745 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
746 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
747 ext4_warning_inode(dir
, "Unrecognised inode hash code %u",
748 root
->info
.hash_version
);
752 hinfo
= &fname
->hinfo
;
753 hinfo
->hash_version
= root
->info
.hash_version
;
754 if (hinfo
->hash_version
<= DX_HASH_TEA
)
755 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
756 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
757 if (fname
&& fname_name(fname
))
758 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), hinfo
);
761 if (root
->info
.unused_flags
& 1) {
762 ext4_warning_inode(dir
, "Unimplemented hash flags: %#06x",
763 root
->info
.unused_flags
);
767 indirect
= root
->info
.indirect_levels
;
769 ext4_warning_inode(dir
, "Unimplemented hash depth: %#06x",
770 root
->info
.indirect_levels
);
774 entries
= (struct dx_entry
*)(((char *)&root
->info
) +
775 root
->info
.info_length
);
777 if (dx_get_limit(entries
) != dx_root_limit(dir
,
778 root
->info
.info_length
)) {
779 ext4_warning_inode(dir
, "dx entry: limit %u != root limit %u",
780 dx_get_limit(entries
),
781 dx_root_limit(dir
, root
->info
.info_length
));
785 dxtrace(printk("Look up %x", hash
));
787 count
= dx_get_count(entries
);
788 if (!count
|| count
> dx_get_limit(entries
)) {
789 ext4_warning_inode(dir
,
790 "dx entry: count %u beyond limit %u",
791 count
, dx_get_limit(entries
));
796 q
= entries
+ count
- 1;
799 dxtrace(printk("."));
800 if (dx_get_hash(m
) > hash
)
806 if (0) { // linear search cross check
807 unsigned n
= count
- 1;
811 dxtrace(printk(","));
812 if (dx_get_hash(++at
) > hash
)
818 assert (at
== p
- 1);
822 dxtrace(printk(" %x->%u\n", at
== entries
? 0 : dx_get_hash(at
),
824 frame
->entries
= entries
;
829 frame
->bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
830 if (IS_ERR(frame
->bh
)) {
831 ret_err
= (struct dx_frame
*) frame
->bh
;
835 entries
= ((struct dx_node
*) frame
->bh
->b_data
)->entries
;
837 if (dx_get_limit(entries
) != dx_node_limit(dir
)) {
838 ext4_warning_inode(dir
,
839 "dx entry: limit %u != node limit %u",
840 dx_get_limit(entries
), dx_node_limit(dir
));
845 while (frame
>= frame_in
) {
850 if (ret_err
== ERR_PTR(ERR_BAD_DX_DIR
))
851 ext4_warning_inode(dir
,
852 "Corrupt directory, running e2fsck is recommended");
856 static void dx_release(struct dx_frame
*frames
)
858 if (frames
[0].bh
== NULL
)
861 if (((struct dx_root
*)frames
[0].bh
->b_data
)->info
.indirect_levels
)
862 brelse(frames
[1].bh
);
863 brelse(frames
[0].bh
);
867 * This function increments the frame pointer to search the next leaf
868 * block, and reads in the necessary intervening nodes if the search
869 * should be necessary. Whether or not the search is necessary is
870 * controlled by the hash parameter. If the hash value is even, then
871 * the search is only continued if the next block starts with that
872 * hash value. This is used if we are searching for a specific file.
874 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
876 * This function returns 1 if the caller should continue to search,
877 * or 0 if it should not. If there is an error reading one of the
878 * index blocks, it will a negative error code.
880 * If start_hash is non-null, it will be filled in with the starting
881 * hash of the next page.
883 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
884 struct dx_frame
*frame
,
885 struct dx_frame
*frames
,
889 struct buffer_head
*bh
;
895 * Find the next leaf page by incrementing the frame pointer.
896 * If we run out of entries in the interior node, loop around and
897 * increment pointer in the parent node. When we break out of
898 * this loop, num_frames indicates the number of interior
899 * nodes need to be read.
902 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
911 * If the hash is 1, then continue only if the next page has a
912 * continuation hash of any value. This is used for readdir
913 * handling. Otherwise, check to see if the hash matches the
914 * desired contiuation hash. If it doesn't, return since
915 * there's no point to read in the successive index pages.
917 bhash
= dx_get_hash(p
->at
);
920 if ((hash
& 1) == 0) {
921 if ((bhash
& ~1) != hash
)
925 * If the hash is HASH_NB_ALWAYS, we always go to the next
926 * block so no check is necessary
928 while (num_frames
--) {
929 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
935 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
942 * This function fills a red-black tree with information from a
943 * directory block. It returns the number directory entries loaded
944 * into the tree. If there is an error it is returned in err.
946 static int htree_dirblock_to_tree(struct file
*dir_file
,
947 struct inode
*dir
, ext4_lblk_t block
,
948 struct dx_hash_info
*hinfo
,
949 __u32 start_hash
, __u32 start_minor_hash
)
951 struct buffer_head
*bh
;
952 struct ext4_dir_entry_2
*de
, *top
;
953 int err
= 0, count
= 0;
954 struct ext4_str fname_crypto_str
= {.name
= NULL
, .len
= 0}, tmp_str
;
956 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
957 (unsigned long)block
));
958 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
962 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
963 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
964 dir
->i_sb
->s_blocksize
-
965 EXT4_DIR_REC_LEN(0));
966 #ifdef CONFIG_EXT4_FS_ENCRYPTION
967 /* Check if the directory is encrypted */
968 if (ext4_encrypted_inode(dir
)) {
969 err
= ext4_get_encryption_info(dir
);
974 err
= ext4_fname_crypto_alloc_buffer(dir
, EXT4_NAME_LEN
,
982 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
983 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
984 bh
->b_data
, bh
->b_size
,
985 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
986 + ((char *)de
- bh
->b_data
))) {
987 /* silently ignore the rest of the block */
990 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
991 if ((hinfo
->hash
< start_hash
) ||
992 ((hinfo
->hash
== start_hash
) &&
993 (hinfo
->minor_hash
< start_minor_hash
)))
997 if (!ext4_encrypted_inode(dir
)) {
998 tmp_str
.name
= de
->name
;
999 tmp_str
.len
= de
->name_len
;
1000 err
= ext4_htree_store_dirent(dir_file
,
1001 hinfo
->hash
, hinfo
->minor_hash
, de
,
1004 int save_len
= fname_crypto_str
.len
;
1006 /* Directory is encrypted */
1007 err
= ext4_fname_disk_to_usr(dir
, hinfo
, de
,
1013 err
= ext4_htree_store_dirent(dir_file
,
1014 hinfo
->hash
, hinfo
->minor_hash
, de
,
1016 fname_crypto_str
.len
= save_len
;
1026 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1027 ext4_fname_crypto_free_buffer(&fname_crypto_str
);
1034 * This function fills a red-black tree with information from a
1035 * directory. We start scanning the directory in hash order, starting
1036 * at start_hash and start_minor_hash.
1038 * This function returns the number of entries inserted into the tree,
1039 * or a negative error code.
1041 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
1042 __u32 start_minor_hash
, __u32
*next_hash
)
1044 struct dx_hash_info hinfo
;
1045 struct ext4_dir_entry_2
*de
;
1046 struct dx_frame frames
[2], *frame
;
1052 struct ext4_str tmp_str
;
1054 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
1055 start_hash
, start_minor_hash
));
1056 dir
= file_inode(dir_file
);
1057 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
1058 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1059 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1060 hinfo
.hash_version
+=
1061 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1062 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1063 if (ext4_has_inline_data(dir
)) {
1064 int has_inline_data
= 1;
1065 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
1069 if (has_inline_data
) {
1074 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
1075 start_hash
, start_minor_hash
);
1079 hinfo
.hash
= start_hash
;
1080 hinfo
.minor_hash
= 0;
1081 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
);
1083 return PTR_ERR(frame
);
1085 /* Add '.' and '..' from the htree header */
1086 if (!start_hash
&& !start_minor_hash
) {
1087 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1088 tmp_str
.name
= de
->name
;
1089 tmp_str
.len
= de
->name_len
;
1090 err
= ext4_htree_store_dirent(dir_file
, 0, 0,
1096 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
1097 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1098 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
1099 tmp_str
.name
= de
->name
;
1100 tmp_str
.len
= de
->name_len
;
1101 err
= ext4_htree_store_dirent(dir_file
, 2, 0,
1109 if (fatal_signal_pending(current
)) {
1114 block
= dx_get_block(frame
->at
);
1115 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
1116 start_hash
, start_minor_hash
);
1123 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
1124 frame
, frames
, &hashval
);
1125 *next_hash
= hashval
;
1131 * Stop if: (a) there are no more entries, or
1132 * (b) we have inserted at least one entry and the
1133 * next hash value is not a continuation
1136 (count
&& ((hashval
& 1) == 0)))
1140 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1141 "next hash: %x\n", count
, *next_hash
));
1148 static inline int search_dirblock(struct buffer_head
*bh
,
1150 struct ext4_filename
*fname
,
1151 const struct qstr
*d_name
,
1152 unsigned int offset
,
1153 struct ext4_dir_entry_2
**res_dir
)
1155 return ext4_search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1156 fname
, d_name
, offset
, res_dir
);
1160 * Directory block splitting, compacting
1164 * Create map of hash values, offsets, and sizes, stored at end of block.
1165 * Returns number of entries mapped.
1167 static int dx_make_map(struct inode
*dir
, struct ext4_dir_entry_2
*de
,
1168 unsigned blocksize
, struct dx_hash_info
*hinfo
,
1169 struct dx_map_entry
*map_tail
)
1172 char *base
= (char *) de
;
1173 struct dx_hash_info h
= *hinfo
;
1175 while ((char *) de
< base
+ blocksize
) {
1176 if (de
->name_len
&& de
->inode
) {
1177 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1179 map_tail
->hash
= h
.hash
;
1180 map_tail
->offs
= ((char *) de
- base
)>>2;
1181 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1185 /* XXX: do we need to check rec_len == 0 case? -Chris */
1186 de
= ext4_next_entry(de
, blocksize
);
1191 /* Sort map by hash value */
1192 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1194 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1196 /* Combsort until bubble sort doesn't suck */
1198 count
= count
*10/13;
1199 if (count
- 9 < 2) /* 9, 10 -> 11 */
1201 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1202 if (p
->hash
< q
->hash
)
1205 /* Garden variety bubble sort */
1210 if (q
[1].hash
>= q
[0].hash
)
1218 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1220 struct dx_entry
*entries
= frame
->entries
;
1221 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1222 int count
= dx_get_count(entries
);
1224 assert(count
< dx_get_limit(entries
));
1225 assert(old
< entries
+ count
);
1226 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1227 dx_set_hash(new, hash
);
1228 dx_set_block(new, block
);
1229 dx_set_count(entries
, count
+ 1);
1233 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1235 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1236 * `de != NULL' is guaranteed by caller.
1238 static inline int ext4_match(struct ext4_filename
*fname
,
1239 struct ext4_dir_entry_2
*de
)
1241 const void *name
= fname_name(fname
);
1242 u32 len
= fname_len(fname
);
1247 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1248 if (unlikely(!name
)) {
1249 if (fname
->usr_fname
->name
[0] == '_') {
1251 if (de
->name_len
< 16)
1253 ret
= memcmp(de
->name
+ de
->name_len
- 16,
1254 fname
->crypto_buf
.name
+ 8, 16);
1255 return (ret
== 0) ? 1 : 0;
1257 name
= fname
->crypto_buf
.name
;
1258 len
= fname
->crypto_buf
.len
;
1261 if (de
->name_len
!= len
)
1263 return (memcmp(de
->name
, name
, len
) == 0) ? 1 : 0;
1267 * Returns 0 if not found, -1 on failure, and 1 on success
1269 int ext4_search_dir(struct buffer_head
*bh
, char *search_buf
, int buf_size
,
1270 struct inode
*dir
, struct ext4_filename
*fname
,
1271 const struct qstr
*d_name
,
1272 unsigned int offset
, struct ext4_dir_entry_2
**res_dir
)
1274 struct ext4_dir_entry_2
* de
;
1279 de
= (struct ext4_dir_entry_2
*)search_buf
;
1280 dlimit
= search_buf
+ buf_size
;
1281 while ((char *) de
< dlimit
) {
1282 /* this code is executed quadratically often */
1283 /* do minimal checking `by hand' */
1284 if ((char *) de
+ de
->name_len
<= dlimit
) {
1285 res
= ext4_match(fname
, de
);
1291 /* found a match - just to be sure, do
1293 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1295 bh
->b_size
, offset
)) {
1305 /* prevent looping on a bad block */
1306 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1307 dir
->i_sb
->s_blocksize
);
1313 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1321 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1322 struct ext4_dir_entry
*de
)
1324 struct super_block
*sb
= dir
->i_sb
;
1330 if (de
->inode
== 0 &&
1331 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1340 * finds an entry in the specified directory with the wanted name. It
1341 * returns the cache buffer in which the entry was found, and the entry
1342 * itself (as a parameter - res_dir). It does NOT read the inode of the
1343 * entry - you'll have to do that yourself if you want to.
1345 * The returned buffer_head has ->b_count elevated. The caller is expected
1346 * to brelse() it when appropriate.
1348 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1349 const struct qstr
*d_name
,
1350 struct ext4_dir_entry_2
**res_dir
,
1353 struct super_block
*sb
;
1354 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1355 struct buffer_head
*bh
, *ret
= NULL
;
1356 ext4_lblk_t start
, block
, b
;
1357 const u8
*name
= d_name
->name
;
1358 int ra_max
= 0; /* Number of bh's in the readahead
1360 int ra_ptr
= 0; /* Current index into readahead
1363 ext4_lblk_t nblocks
;
1364 int i
, namelen
, retval
;
1365 struct ext4_filename fname
;
1369 namelen
= d_name
->len
;
1370 if (namelen
> EXT4_NAME_LEN
)
1373 retval
= ext4_fname_setup_filename(dir
, d_name
, 1, &fname
);
1375 return ERR_PTR(retval
);
1377 if (ext4_has_inline_data(dir
)) {
1378 int has_inline_data
= 1;
1379 ret
= ext4_find_inline_entry(dir
, &fname
, d_name
, res_dir
,
1381 if (has_inline_data
) {
1384 goto cleanup_and_exit
;
1388 if ((namelen
<= 2) && (name
[0] == '.') &&
1389 (name
[1] == '.' || name
[1] == '\0')) {
1391 * "." or ".." will only be in the first block
1392 * NFS may look up ".."; "." should be handled by the VFS
1399 ret
= ext4_dx_find_entry(dir
, &fname
, res_dir
);
1401 * On success, or if the error was file not found,
1402 * return. Otherwise, fall back to doing a search the
1403 * old fashioned way.
1405 if (!IS_ERR(ret
) || PTR_ERR(ret
) != ERR_BAD_DX_DIR
)
1406 goto cleanup_and_exit
;
1407 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1410 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1411 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1412 if (start
>= nblocks
)
1418 * We deal with the read-ahead logic here.
1420 if (ra_ptr
>= ra_max
) {
1421 /* Refill the readahead buffer */
1424 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1426 * Terminate if we reach the end of the
1427 * directory and must wrap, or if our
1428 * search has finished at this block.
1430 if (b
>= nblocks
|| (num
&& block
== start
)) {
1431 bh_use
[ra_max
] = NULL
;
1435 bh
= ext4_getblk(NULL
, dir
, b
++, 0);
1439 goto cleanup_and_exit
;
1443 bh_use
[ra_max
] = bh
;
1445 ll_rw_block(READ
| REQ_META
| REQ_PRIO
,
1449 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1452 if (!buffer_uptodate(bh
)) {
1453 /* read error, skip block & hope for the best */
1454 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1455 (unsigned long) block
);
1459 if (!buffer_verified(bh
) &&
1460 !is_dx_internal_node(dir
, block
,
1461 (struct ext4_dir_entry
*)bh
->b_data
) &&
1462 !ext4_dirent_csum_verify(dir
,
1463 (struct ext4_dir_entry
*)bh
->b_data
)) {
1464 EXT4_ERROR_INODE(dir
, "checksumming directory "
1465 "block %lu", (unsigned long)block
);
1469 set_buffer_verified(bh
);
1470 i
= search_dirblock(bh
, dir
, &fname
, d_name
,
1471 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1473 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1475 goto cleanup_and_exit
;
1479 goto cleanup_and_exit
;
1482 if (++block
>= nblocks
)
1484 } while (block
!= start
);
1487 * If the directory has grown while we were searching, then
1488 * search the last part of the directory before giving up.
1491 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1492 if (block
< nblocks
) {
1498 /* Clean up the read-ahead blocks */
1499 for (; ra_ptr
< ra_max
; ra_ptr
++)
1500 brelse(bh_use
[ra_ptr
]);
1501 ext4_fname_free_filename(&fname
);
1505 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
1506 struct ext4_filename
*fname
,
1507 struct ext4_dir_entry_2
**res_dir
)
1509 struct super_block
* sb
= dir
->i_sb
;
1510 struct dx_frame frames
[2], *frame
;
1511 const struct qstr
*d_name
= fname
->usr_fname
;
1512 struct buffer_head
*bh
;
1516 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1519 frame
= dx_probe(fname
, dir
, NULL
, frames
);
1521 return (struct buffer_head
*) frame
;
1523 block
= dx_get_block(frame
->at
);
1524 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1528 retval
= search_dirblock(bh
, dir
, fname
, d_name
,
1529 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1535 bh
= ERR_PTR(ERR_BAD_DX_DIR
);
1539 /* Check to see if we should continue to search */
1540 retval
= ext4_htree_next_block(dir
, fname
->hinfo
.hash
, frame
,
1543 ext4_warning_inode(dir
,
1544 "error %d reading directory index block",
1546 bh
= ERR_PTR(retval
);
1549 } while (retval
== 1);
1553 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1559 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1561 struct inode
*inode
;
1562 struct ext4_dir_entry_2
*de
;
1563 struct buffer_head
*bh
;
1565 if (ext4_encrypted_inode(dir
)) {
1566 int res
= ext4_get_encryption_info(dir
);
1569 * This should be a properly defined flag for
1570 * dentry->d_flags when we uplift this to the VFS.
1571 * d_fsdata is set to (void *) 1 if if the dentry is
1572 * created while the directory was encrypted and we
1573 * don't have access to the key.
1575 dentry
->d_fsdata
= NULL
;
1576 if (ext4_encryption_info(dir
))
1577 dentry
->d_fsdata
= (void *) 1;
1578 d_set_d_op(dentry
, &ext4_encrypted_d_ops
);
1579 if (res
&& res
!= -ENOKEY
)
1580 return ERR_PTR(res
);
1583 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1584 return ERR_PTR(-ENAMETOOLONG
);
1586 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1588 return (struct dentry
*) bh
;
1591 __u32 ino
= le32_to_cpu(de
->inode
);
1593 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1594 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1595 return ERR_PTR(-EFSCORRUPTED
);
1597 if (unlikely(ino
== dir
->i_ino
)) {
1598 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1600 return ERR_PTR(-EFSCORRUPTED
);
1602 inode
= ext4_iget_normal(dir
->i_sb
, ino
);
1603 if (inode
== ERR_PTR(-ESTALE
)) {
1604 EXT4_ERROR_INODE(dir
,
1605 "deleted inode referenced: %u",
1607 return ERR_PTR(-EFSCORRUPTED
);
1609 if (!IS_ERR(inode
) && ext4_encrypted_inode(dir
) &&
1610 (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
)) &&
1611 !ext4_is_child_context_consistent_with_parent(dir
,
1613 int nokey
= ext4_encrypted_inode(inode
) &&
1614 !ext4_encryption_info(inode
);
1618 return ERR_PTR(-ENOKEY
);
1619 ext4_warning(inode
->i_sb
,
1620 "Inconsistent encryption contexts: %lu/%lu",
1621 (unsigned long) dir
->i_ino
,
1622 (unsigned long) inode
->i_ino
);
1623 return ERR_PTR(-EPERM
);
1626 return d_splice_alias(inode
, dentry
);
1630 struct dentry
*ext4_get_parent(struct dentry
*child
)
1633 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1634 struct ext4_dir_entry_2
* de
;
1635 struct buffer_head
*bh
;
1637 bh
= ext4_find_entry(d_inode(child
), &dotdot
, &de
, NULL
);
1639 return (struct dentry
*) bh
;
1641 return ERR_PTR(-ENOENT
);
1642 ino
= le32_to_cpu(de
->inode
);
1645 if (!ext4_valid_inum(child
->d_sb
, ino
)) {
1646 EXT4_ERROR_INODE(d_inode(child
),
1647 "bad parent inode number: %u", ino
);
1648 return ERR_PTR(-EFSCORRUPTED
);
1651 return d_obtain_alias(ext4_iget_normal(child
->d_sb
, ino
));
1655 * Move count entries from end of map between two memory locations.
1656 * Returns pointer to last entry moved.
1658 static struct ext4_dir_entry_2
*
1659 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1662 unsigned rec_len
= 0;
1665 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1666 (from
+ (map
->offs
<<2));
1667 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1668 memcpy (to
, de
, rec_len
);
1669 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1670 ext4_rec_len_to_disk(rec_len
, blocksize
);
1675 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1679 * Compact each dir entry in the range to the minimal rec_len.
1680 * Returns pointer to last entry in range.
1682 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1684 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1685 unsigned rec_len
= 0;
1688 while ((char*)de
< base
+ blocksize
) {
1689 next
= ext4_next_entry(de
, blocksize
);
1690 if (de
->inode
&& de
->name_len
) {
1691 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1693 memmove(to
, de
, rec_len
);
1694 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1696 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1704 * Split a full leaf block to make room for a new dir entry.
1705 * Allocate a new block, and move entries so that they are approx. equally full.
1706 * Returns pointer to de in block into which the new entry will be inserted.
1708 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1709 struct buffer_head
**bh
,struct dx_frame
*frame
,
1710 struct dx_hash_info
*hinfo
)
1712 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1713 unsigned count
, continued
;
1714 struct buffer_head
*bh2
;
1715 ext4_lblk_t newblock
;
1717 struct dx_map_entry
*map
;
1718 char *data1
= (*bh
)->b_data
, *data2
;
1719 unsigned split
, move
, size
;
1720 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1721 struct ext4_dir_entry_tail
*t
;
1725 if (ext4_has_metadata_csum(dir
->i_sb
))
1726 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1728 bh2
= ext4_append(handle
, dir
, &newblock
);
1732 return (struct ext4_dir_entry_2
*) bh2
;
1735 BUFFER_TRACE(*bh
, "get_write_access");
1736 err
= ext4_journal_get_write_access(handle
, *bh
);
1740 BUFFER_TRACE(frame
->bh
, "get_write_access");
1741 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1745 data2
= bh2
->b_data
;
1747 /* create map in the end of data2 block */
1748 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1749 count
= dx_make_map(dir
, (struct ext4_dir_entry_2
*) data1
,
1750 blocksize
, hinfo
, map
);
1752 dx_sort_map(map
, count
);
1753 /* Split the existing block in the middle, size-wise */
1756 for (i
= count
-1; i
>= 0; i
--) {
1757 /* is more than half of this entry in 2nd half of the block? */
1758 if (size
+ map
[i
].size
/2 > blocksize
/2)
1760 size
+= map
[i
].size
;
1763 /* map index at which we will split */
1764 split
= count
- move
;
1765 hash2
= map
[split
].hash
;
1766 continued
= hash2
== map
[split
- 1].hash
;
1767 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1768 (unsigned long)dx_get_block(frame
->at
),
1769 hash2
, split
, count
-split
));
1771 /* Fancy dance to stay within two buffers */
1772 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
,
1774 de
= dx_pack_dirents(data1
, blocksize
);
1775 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1778 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1782 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1783 initialize_dirent_tail(t
, blocksize
);
1785 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1786 initialize_dirent_tail(t
, blocksize
);
1789 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data1
,
1791 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data2
,
1794 /* Which block gets the new entry? */
1795 if (hinfo
->hash
>= hash2
) {
1799 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1800 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1803 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1807 dxtrace(dx_show_index("frame", frame
->entries
));
1814 ext4_std_error(dir
->i_sb
, err
);
1815 return ERR_PTR(err
);
1818 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1819 struct buffer_head
*bh
,
1820 void *buf
, int buf_size
,
1821 struct ext4_filename
*fname
,
1822 struct ext4_dir_entry_2
**dest_de
)
1824 struct ext4_dir_entry_2
*de
;
1825 unsigned short reclen
= EXT4_DIR_REC_LEN(fname_len(fname
));
1827 unsigned int offset
= 0;
1831 de
= (struct ext4_dir_entry_2
*)buf
;
1832 top
= buf
+ buf_size
- reclen
;
1833 while ((char *) de
<= top
) {
1834 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1835 buf
, buf_size
, offset
)) {
1836 res
= -EFSCORRUPTED
;
1839 /* Provide crypto context and crypto buffer to ext4 match */
1840 res
= ext4_match(fname
, de
);
1847 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1848 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1849 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1851 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1855 if ((char *) de
> top
)
1865 int ext4_insert_dentry(struct inode
*dir
,
1866 struct inode
*inode
,
1867 struct ext4_dir_entry_2
*de
,
1869 struct ext4_filename
*fname
)
1874 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1875 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1877 struct ext4_dir_entry_2
*de1
=
1878 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1879 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1880 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1883 de
->file_type
= EXT4_FT_UNKNOWN
;
1884 de
->inode
= cpu_to_le32(inode
->i_ino
);
1885 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1886 de
->name_len
= fname_len(fname
);
1887 memcpy(de
->name
, fname_name(fname
), fname_len(fname
));
1892 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1893 * it points to a directory entry which is guaranteed to be large
1894 * enough for new directory entry. If de is NULL, then
1895 * add_dirent_to_buf will attempt search the directory block for
1896 * space. It will return -ENOSPC if no space is available, and -EIO
1897 * and -EEXIST if directory entry already exists.
1899 static int add_dirent_to_buf(handle_t
*handle
, struct ext4_filename
*fname
,
1901 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1902 struct buffer_head
*bh
)
1904 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1908 if (ext4_has_metadata_csum(inode
->i_sb
))
1909 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1912 err
= ext4_find_dest_de(dir
, inode
, bh
, bh
->b_data
,
1913 blocksize
- csum_size
, fname
, &de
);
1917 BUFFER_TRACE(bh
, "get_write_access");
1918 err
= ext4_journal_get_write_access(handle
, bh
);
1920 ext4_std_error(dir
->i_sb
, err
);
1924 /* By now the buffer is marked for journaling. Due to crypto operations,
1925 * the following function call may fail */
1926 err
= ext4_insert_dentry(dir
, inode
, de
, blocksize
, fname
);
1931 * XXX shouldn't update any times until successful
1932 * completion of syscall, but too many callers depend
1935 * XXX similarly, too many callers depend on
1936 * ext4_new_inode() setting the times, but error
1937 * recovery deletes the inode, so the worst that can
1938 * happen is that the times are slightly out of date
1939 * and/or different from the directory change time.
1941 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1942 ext4_update_dx_flag(dir
);
1944 ext4_mark_inode_dirty(handle
, dir
);
1945 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1946 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1948 ext4_std_error(dir
->i_sb
, err
);
1953 * This converts a one block unindexed directory to a 3 block indexed
1954 * directory, and adds the dentry to the indexed directory.
1956 static int make_indexed_dir(handle_t
*handle
, struct ext4_filename
*fname
,
1958 struct inode
*inode
, struct buffer_head
*bh
)
1960 struct buffer_head
*bh2
;
1961 struct dx_root
*root
;
1962 struct dx_frame frames
[2], *frame
;
1963 struct dx_entry
*entries
;
1964 struct ext4_dir_entry_2
*de
, *de2
;
1965 struct ext4_dir_entry_tail
*t
;
1971 struct fake_dirent
*fde
;
1974 if (ext4_has_metadata_csum(inode
->i_sb
))
1975 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1977 blocksize
= dir
->i_sb
->s_blocksize
;
1978 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1979 BUFFER_TRACE(bh
, "get_write_access");
1980 retval
= ext4_journal_get_write_access(handle
, bh
);
1982 ext4_std_error(dir
->i_sb
, retval
);
1986 root
= (struct dx_root
*) bh
->b_data
;
1988 /* The 0th block becomes the root, move the dirents out */
1989 fde
= &root
->dotdot
;
1990 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1991 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1992 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1993 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1995 return -EFSCORRUPTED
;
1997 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1999 /* Allocate new block for the 0th block's dirents */
2000 bh2
= ext4_append(handle
, dir
, &block
);
2003 return PTR_ERR(bh2
);
2005 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
2006 data1
= bh2
->b_data
;
2008 memcpy (data1
, de
, len
);
2009 de
= (struct ext4_dir_entry_2
*) data1
;
2011 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
2013 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
2018 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
2019 initialize_dirent_tail(t
, blocksize
);
2022 /* Initialize the root; the dot dirents already exist */
2023 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
2024 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
2026 memset (&root
->info
, 0, sizeof(root
->info
));
2027 root
->info
.info_length
= sizeof(root
->info
);
2028 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
2029 entries
= root
->entries
;
2030 dx_set_block(entries
, 1);
2031 dx_set_count(entries
, 1);
2032 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
2034 /* Initialize as for dx_probe */
2035 fname
->hinfo
.hash_version
= root
->info
.hash_version
;
2036 if (fname
->hinfo
.hash_version
<= DX_HASH_TEA
)
2037 fname
->hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
2038 fname
->hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
2039 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), &fname
->hinfo
);
2041 memset(frames
, 0, sizeof(frames
));
2043 frame
->entries
= entries
;
2044 frame
->at
= entries
;
2048 retval
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
2051 retval
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2055 de
= do_split(handle
,dir
, &bh
, frame
, &fname
->hinfo
);
2057 retval
= PTR_ERR(de
);
2062 retval
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh
);
2067 * Even if the block split failed, we have to properly write
2068 * out all the changes we did so far. Otherwise we can end up
2069 * with corrupted filesystem.
2071 ext4_mark_inode_dirty(handle
, dir
);
2079 * adds a file entry to the specified directory, using the same
2080 * semantics as ext4_find_entry(). It returns NULL if it failed.
2082 * NOTE!! The inode part of 'de' is left at 0 - which means you
2083 * may not sleep between calling this and putting something into
2084 * the entry, as someone else might have used it while you slept.
2086 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
2087 struct inode
*inode
)
2089 struct inode
*dir
= d_inode(dentry
->d_parent
);
2090 struct buffer_head
*bh
= NULL
;
2091 struct ext4_dir_entry_2
*de
;
2092 struct ext4_dir_entry_tail
*t
;
2093 struct super_block
*sb
;
2094 struct ext4_filename fname
;
2098 ext4_lblk_t block
, blocks
;
2101 if (ext4_has_metadata_csum(inode
->i_sb
))
2102 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2105 blocksize
= sb
->s_blocksize
;
2106 if (!dentry
->d_name
.len
)
2109 retval
= ext4_fname_setup_filename(dir
, &dentry
->d_name
, 0, &fname
);
2113 if (ext4_has_inline_data(dir
)) {
2114 retval
= ext4_try_add_inline_entry(handle
, &fname
, dir
, inode
);
2124 retval
= ext4_dx_add_entry(handle
, &fname
, dir
, inode
);
2125 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
2127 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
2129 ext4_mark_inode_dirty(handle
, dir
);
2131 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
2132 for (block
= 0; block
< blocks
; block
++) {
2133 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
2135 retval
= PTR_ERR(bh
);
2139 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
,
2141 if (retval
!= -ENOSPC
)
2144 if (blocks
== 1 && !dx_fallback
&&
2145 ext4_has_feature_dir_index(sb
)) {
2146 retval
= make_indexed_dir(handle
, &fname
, dir
,
2148 bh
= NULL
; /* make_indexed_dir releases bh */
2153 bh
= ext4_append(handle
, dir
, &block
);
2155 retval
= PTR_ERR(bh
);
2159 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2161 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
2164 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
2165 initialize_dirent_tail(t
, blocksize
);
2168 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
, de
, bh
);
2170 ext4_fname_free_filename(&fname
);
2173 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
2178 * Returns 0 for success, or a negative error value
2180 static int ext4_dx_add_entry(handle_t
*handle
, struct ext4_filename
*fname
,
2181 struct inode
*dir
, struct inode
*inode
)
2183 struct dx_frame frames
[2], *frame
;
2184 struct dx_entry
*entries
, *at
;
2185 struct buffer_head
*bh
;
2186 struct super_block
*sb
= dir
->i_sb
;
2187 struct ext4_dir_entry_2
*de
;
2190 frame
= dx_probe(fname
, dir
, NULL
, frames
);
2192 return PTR_ERR(frame
);
2193 entries
= frame
->entries
;
2195 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
2202 BUFFER_TRACE(bh
, "get_write_access");
2203 err
= ext4_journal_get_write_access(handle
, bh
);
2207 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, NULL
, bh
);
2211 /* Block full, should compress but for now just split */
2212 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
2213 dx_get_count(entries
), dx_get_limit(entries
)));
2214 /* Need to split index? */
2215 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
2216 ext4_lblk_t newblock
;
2217 unsigned icount
= dx_get_count(entries
);
2218 int levels
= frame
- frames
;
2219 struct dx_entry
*entries2
;
2220 struct dx_node
*node2
;
2221 struct buffer_head
*bh2
;
2223 if (levels
&& (dx_get_count(frames
->entries
) ==
2224 dx_get_limit(frames
->entries
))) {
2225 ext4_warning_inode(dir
, "Directory index full!");
2229 bh2
= ext4_append(handle
, dir
, &newblock
);
2234 node2
= (struct dx_node
*)(bh2
->b_data
);
2235 entries2
= node2
->entries
;
2236 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2237 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2239 BUFFER_TRACE(frame
->bh
, "get_write_access");
2240 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2244 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2245 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2246 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2249 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2250 err
= ext4_journal_get_write_access(handle
,
2255 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2256 icount2
* sizeof(struct dx_entry
));
2257 dx_set_count(entries
, icount1
);
2258 dx_set_count(entries2
, icount2
);
2259 dx_set_limit(entries2
, dx_node_limit(dir
));
2261 /* Which index block gets the new entry? */
2262 if (at
- entries
>= icount1
) {
2263 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2264 frame
->entries
= entries
= entries2
;
2265 swap(frame
->bh
, bh2
);
2267 dx_insert_block(frames
+ 0, hash2
, newblock
);
2268 dxtrace(dx_show_index("node", frames
[1].entries
));
2269 dxtrace(dx_show_index("node",
2270 ((struct dx_node
*) bh2
->b_data
)->entries
));
2271 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2276 dxtrace(printk(KERN_DEBUG
2277 "Creating second level index...\n"));
2278 memcpy((char *) entries2
, (char *) entries
,
2279 icount
* sizeof(struct dx_entry
));
2280 dx_set_limit(entries2
, dx_node_limit(dir
));
2283 dx_set_count(entries
, 1);
2284 dx_set_block(entries
+ 0, newblock
);
2285 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2287 /* Add new access path frame */
2289 frame
->at
= at
= at
- entries
+ entries2
;
2290 frame
->entries
= entries
= entries2
;
2292 err
= ext4_journal_get_write_access(handle
,
2297 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2299 ext4_std_error(inode
->i_sb
, err
);
2303 de
= do_split(handle
, dir
, &bh
, frame
, &fname
->hinfo
);
2308 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh
);
2312 ext4_std_error(dir
->i_sb
, err
);
2320 * ext4_generic_delete_entry deletes a directory entry by merging it
2321 * with the previous entry
2323 int ext4_generic_delete_entry(handle_t
*handle
,
2325 struct ext4_dir_entry_2
*de_del
,
2326 struct buffer_head
*bh
,
2331 struct ext4_dir_entry_2
*de
, *pde
;
2332 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2337 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2338 while (i
< buf_size
- csum_size
) {
2339 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2340 bh
->b_data
, bh
->b_size
, i
))
2341 return -EFSCORRUPTED
;
2344 pde
->rec_len
= ext4_rec_len_to_disk(
2345 ext4_rec_len_from_disk(pde
->rec_len
,
2347 ext4_rec_len_from_disk(de
->rec_len
,
2355 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2357 de
= ext4_next_entry(de
, blocksize
);
2362 static int ext4_delete_entry(handle_t
*handle
,
2364 struct ext4_dir_entry_2
*de_del
,
2365 struct buffer_head
*bh
)
2367 int err
, csum_size
= 0;
2369 if (ext4_has_inline_data(dir
)) {
2370 int has_inline_data
= 1;
2371 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2373 if (has_inline_data
)
2377 if (ext4_has_metadata_csum(dir
->i_sb
))
2378 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2380 BUFFER_TRACE(bh
, "get_write_access");
2381 err
= ext4_journal_get_write_access(handle
, bh
);
2385 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2387 dir
->i_sb
->s_blocksize
, csum_size
);
2391 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2392 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2399 ext4_std_error(dir
->i_sb
, err
);
2404 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2405 * since this indicates that nlinks count was previously 1.
2407 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2410 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2411 /* limit is 16-bit i_links_count */
2412 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2413 set_nlink(inode
, 1);
2414 ext4_set_feature_dir_nlink(inode
->i_sb
);
2420 * If a directory had nlink == 1, then we should let it be 1. This indicates
2421 * directory has >EXT4_LINK_MAX subdirs.
2423 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2425 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2430 static int ext4_add_nondir(handle_t
*handle
,
2431 struct dentry
*dentry
, struct inode
*inode
)
2433 int err
= ext4_add_entry(handle
, dentry
, inode
);
2435 ext4_mark_inode_dirty(handle
, inode
);
2436 unlock_new_inode(inode
);
2437 d_instantiate(dentry
, inode
);
2441 unlock_new_inode(inode
);
2447 * By the time this is called, we already have created
2448 * the directory cache entry for the new file, but it
2449 * is so far negative - it has no inode.
2451 * If the create succeeds, we fill in the inode information
2452 * with d_instantiate().
2454 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2458 struct inode
*inode
;
2459 int err
, credits
, retries
= 0;
2461 err
= dquot_initialize(dir
);
2465 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2466 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2468 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2469 NULL
, EXT4_HT_DIR
, credits
);
2470 handle
= ext4_journal_current_handle();
2471 err
= PTR_ERR(inode
);
2472 if (!IS_ERR(inode
)) {
2473 inode
->i_op
= &ext4_file_inode_operations
;
2474 inode
->i_fop
= &ext4_file_operations
;
2475 ext4_set_aops(inode
);
2476 err
= ext4_add_nondir(handle
, dentry
, inode
);
2477 if (!err
&& IS_DIRSYNC(dir
))
2478 ext4_handle_sync(handle
);
2481 ext4_journal_stop(handle
);
2482 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2487 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2488 umode_t mode
, dev_t rdev
)
2491 struct inode
*inode
;
2492 int err
, credits
, retries
= 0;
2494 err
= dquot_initialize(dir
);
2498 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2499 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2501 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2502 NULL
, EXT4_HT_DIR
, credits
);
2503 handle
= ext4_journal_current_handle();
2504 err
= PTR_ERR(inode
);
2505 if (!IS_ERR(inode
)) {
2506 init_special_inode(inode
, inode
->i_mode
, rdev
);
2507 inode
->i_op
= &ext4_special_inode_operations
;
2508 err
= ext4_add_nondir(handle
, dentry
, inode
);
2509 if (!err
&& IS_DIRSYNC(dir
))
2510 ext4_handle_sync(handle
);
2513 ext4_journal_stop(handle
);
2514 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2519 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2522 struct inode
*inode
;
2523 int err
, retries
= 0;
2525 err
= dquot_initialize(dir
);
2530 inode
= ext4_new_inode_start_handle(dir
, mode
,
2533 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2534 4 + EXT4_XATTR_TRANS_BLOCKS
);
2535 handle
= ext4_journal_current_handle();
2536 err
= PTR_ERR(inode
);
2537 if (!IS_ERR(inode
)) {
2538 inode
->i_op
= &ext4_file_inode_operations
;
2539 inode
->i_fop
= &ext4_file_operations
;
2540 ext4_set_aops(inode
);
2541 d_tmpfile(dentry
, inode
);
2542 err
= ext4_orphan_add(handle
, inode
);
2544 goto err_unlock_inode
;
2545 mark_inode_dirty(inode
);
2546 unlock_new_inode(inode
);
2549 ext4_journal_stop(handle
);
2550 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2554 ext4_journal_stop(handle
);
2555 unlock_new_inode(inode
);
2559 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2560 struct ext4_dir_entry_2
*de
,
2561 int blocksize
, int csum_size
,
2562 unsigned int parent_ino
, int dotdot_real_len
)
2564 de
->inode
= cpu_to_le32(inode
->i_ino
);
2566 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2568 strcpy(de
->name
, ".");
2569 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2571 de
= ext4_next_entry(de
, blocksize
);
2572 de
->inode
= cpu_to_le32(parent_ino
);
2574 if (!dotdot_real_len
)
2575 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2576 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2579 de
->rec_len
= ext4_rec_len_to_disk(
2580 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2581 strcpy(de
->name
, "..");
2582 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2584 return ext4_next_entry(de
, blocksize
);
2587 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2588 struct inode
*inode
)
2590 struct buffer_head
*dir_block
= NULL
;
2591 struct ext4_dir_entry_2
*de
;
2592 struct ext4_dir_entry_tail
*t
;
2593 ext4_lblk_t block
= 0;
2594 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2598 if (ext4_has_metadata_csum(dir
->i_sb
))
2599 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2601 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2602 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2603 if (err
< 0 && err
!= -ENOSPC
)
2610 dir_block
= ext4_append(handle
, inode
, &block
);
2611 if (IS_ERR(dir_block
))
2612 return PTR_ERR(dir_block
);
2613 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2614 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2615 set_nlink(inode
, 2);
2617 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2618 initialize_dirent_tail(t
, blocksize
);
2621 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2622 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2625 set_buffer_verified(dir_block
);
2631 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2634 struct inode
*inode
;
2635 int err
, credits
, retries
= 0;
2637 if (EXT4_DIR_LINK_MAX(dir
))
2640 err
= dquot_initialize(dir
);
2644 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2645 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2647 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2649 0, NULL
, EXT4_HT_DIR
, credits
);
2650 handle
= ext4_journal_current_handle();
2651 err
= PTR_ERR(inode
);
2655 inode
->i_op
= &ext4_dir_inode_operations
;
2656 inode
->i_fop
= &ext4_dir_operations
;
2657 err
= ext4_init_new_dir(handle
, dir
, inode
);
2659 goto out_clear_inode
;
2660 err
= ext4_mark_inode_dirty(handle
, inode
);
2662 err
= ext4_add_entry(handle
, dentry
, inode
);
2666 unlock_new_inode(inode
);
2667 ext4_mark_inode_dirty(handle
, inode
);
2671 ext4_inc_count(handle
, dir
);
2672 ext4_update_dx_flag(dir
);
2673 err
= ext4_mark_inode_dirty(handle
, dir
);
2675 goto out_clear_inode
;
2676 unlock_new_inode(inode
);
2677 d_instantiate(dentry
, inode
);
2678 if (IS_DIRSYNC(dir
))
2679 ext4_handle_sync(handle
);
2683 ext4_journal_stop(handle
);
2684 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2690 * routine to check that the specified directory is empty (for rmdir)
2692 int ext4_empty_dir(struct inode
*inode
)
2694 unsigned int offset
;
2695 struct buffer_head
*bh
;
2696 struct ext4_dir_entry_2
*de
, *de1
;
2697 struct super_block
*sb
;
2700 if (ext4_has_inline_data(inode
)) {
2701 int has_inline_data
= 1;
2703 err
= empty_inline_dir(inode
, &has_inline_data
);
2704 if (has_inline_data
)
2709 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2710 EXT4_ERROR_INODE(inode
, "invalid size");
2713 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2717 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2718 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2719 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2720 le32_to_cpu(de1
->inode
) == 0 ||
2721 strcmp(".", de
->name
) || strcmp("..", de1
->name
)) {
2722 ext4_warning_inode(inode
, "directory missing '.' and/or '..'");
2726 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2727 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2728 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2729 while (offset
< inode
->i_size
) {
2730 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2731 unsigned int lblock
;
2734 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2735 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2738 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2740 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2741 bh
->b_data
, bh
->b_size
, offset
)) {
2742 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2744 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2747 if (le32_to_cpu(de
->inode
)) {
2751 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2752 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2759 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2760 * such inodes, starting at the superblock, in case we crash before the
2761 * file is closed/deleted, or in case the inode truncate spans multiple
2762 * transactions and the last transaction is not recovered after a crash.
2764 * At filesystem recovery time, we walk this list deleting unlinked
2765 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2767 * Orphan list manipulation functions must be called under i_mutex unless
2768 * we are just creating the inode or deleting it.
2770 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2772 struct super_block
*sb
= inode
->i_sb
;
2773 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2774 struct ext4_iloc iloc
;
2778 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2781 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2782 !inode_is_locked(inode
));
2784 * Exit early if inode already is on orphan list. This is a big speedup
2785 * since we don't have to contend on the global s_orphan_lock.
2787 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2791 * Orphan handling is only valid for files with data blocks
2792 * being truncated, or files being unlinked. Note that we either
2793 * hold i_mutex, or the inode can not be referenced from outside,
2794 * so i_nlink should not be bumped due to race
2796 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2797 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2799 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2800 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2804 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2808 mutex_lock(&sbi
->s_orphan_lock
);
2810 * Due to previous errors inode may be already a part of on-disk
2811 * orphan list. If so skip on-disk list modification.
2813 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2814 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2815 /* Insert this inode at the head of the on-disk orphan list */
2816 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2817 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2820 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2821 mutex_unlock(&sbi
->s_orphan_lock
);
2824 err
= ext4_handle_dirty_super(handle
, sb
);
2825 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2830 * We have to remove inode from in-memory list if
2831 * addition to on disk orphan list failed. Stray orphan
2832 * list entries can cause panics at unmount time.
2834 mutex_lock(&sbi
->s_orphan_lock
);
2835 list_del_init(&EXT4_I(inode
)->i_orphan
);
2836 mutex_unlock(&sbi
->s_orphan_lock
);
2839 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2840 jbd_debug(4, "orphan inode %lu will point to %d\n",
2841 inode
->i_ino
, NEXT_ORPHAN(inode
));
2843 ext4_std_error(sb
, err
);
2848 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2849 * of such inodes stored on disk, because it is finally being cleaned up.
2851 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2853 struct list_head
*prev
;
2854 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2855 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2857 struct ext4_iloc iloc
;
2860 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2863 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2864 !inode_is_locked(inode
));
2865 /* Do this quick check before taking global s_orphan_lock. */
2866 if (list_empty(&ei
->i_orphan
))
2870 /* Grab inode buffer early before taking global s_orphan_lock */
2871 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2874 mutex_lock(&sbi
->s_orphan_lock
);
2875 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2877 prev
= ei
->i_orphan
.prev
;
2878 list_del_init(&ei
->i_orphan
);
2880 /* If we're on an error path, we may not have a valid
2881 * transaction handle with which to update the orphan list on
2882 * disk, but we still need to remove the inode from the linked
2883 * list in memory. */
2884 if (!handle
|| err
) {
2885 mutex_unlock(&sbi
->s_orphan_lock
);
2889 ino_next
= NEXT_ORPHAN(inode
);
2890 if (prev
== &sbi
->s_orphan
) {
2891 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2892 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2893 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2895 mutex_unlock(&sbi
->s_orphan_lock
);
2898 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2899 mutex_unlock(&sbi
->s_orphan_lock
);
2900 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2902 struct ext4_iloc iloc2
;
2903 struct inode
*i_prev
=
2904 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2906 jbd_debug(4, "orphan inode %lu will point to %u\n",
2907 i_prev
->i_ino
, ino_next
);
2908 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2910 mutex_unlock(&sbi
->s_orphan_lock
);
2913 NEXT_ORPHAN(i_prev
) = ino_next
;
2914 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2915 mutex_unlock(&sbi
->s_orphan_lock
);
2919 NEXT_ORPHAN(inode
) = 0;
2920 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2922 ext4_std_error(inode
->i_sb
, err
);
2930 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2933 struct inode
*inode
;
2934 struct buffer_head
*bh
;
2935 struct ext4_dir_entry_2
*de
;
2936 handle_t
*handle
= NULL
;
2938 /* Initialize quotas before so that eventual writes go in
2939 * separate transaction */
2940 retval
= dquot_initialize(dir
);
2943 retval
= dquot_initialize(d_inode(dentry
));
2948 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2954 inode
= d_inode(dentry
);
2956 retval
= -EFSCORRUPTED
;
2957 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2960 retval
= -ENOTEMPTY
;
2961 if (!ext4_empty_dir(inode
))
2964 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2965 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2966 if (IS_ERR(handle
)) {
2967 retval
= PTR_ERR(handle
);
2972 if (IS_DIRSYNC(dir
))
2973 ext4_handle_sync(handle
);
2975 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2978 if (!EXT4_DIR_LINK_EMPTY(inode
))
2979 ext4_warning_inode(inode
,
2980 "empty directory '%.*s' has too many links (%u)",
2981 dentry
->d_name
.len
, dentry
->d_name
.name
,
2985 /* There's no need to set i_disksize: the fact that i_nlink is
2986 * zero will ensure that the right thing happens during any
2989 ext4_orphan_add(handle
, inode
);
2990 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2991 ext4_mark_inode_dirty(handle
, inode
);
2992 ext4_dec_count(handle
, dir
);
2993 ext4_update_dx_flag(dir
);
2994 ext4_mark_inode_dirty(handle
, dir
);
2999 ext4_journal_stop(handle
);
3003 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
3006 struct inode
*inode
;
3007 struct buffer_head
*bh
;
3008 struct ext4_dir_entry_2
*de
;
3009 handle_t
*handle
= NULL
;
3011 trace_ext4_unlink_enter(dir
, dentry
);
3012 /* Initialize quotas before so that eventual writes go
3013 * in separate transaction */
3014 retval
= dquot_initialize(dir
);
3017 retval
= dquot_initialize(d_inode(dentry
));
3022 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
3028 inode
= d_inode(dentry
);
3030 retval
= -EFSCORRUPTED
;
3031 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
3034 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3035 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
3036 if (IS_ERR(handle
)) {
3037 retval
= PTR_ERR(handle
);
3042 if (IS_DIRSYNC(dir
))
3043 ext4_handle_sync(handle
);
3045 if (inode
->i_nlink
== 0) {
3046 ext4_warning_inode(inode
, "Deleting file '%.*s' with no links",
3047 dentry
->d_name
.len
, dentry
->d_name
.name
);
3048 set_nlink(inode
, 1);
3050 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3053 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
3054 ext4_update_dx_flag(dir
);
3055 ext4_mark_inode_dirty(handle
, dir
);
3057 if (!inode
->i_nlink
)
3058 ext4_orphan_add(handle
, inode
);
3059 inode
->i_ctime
= ext4_current_time(inode
);
3060 ext4_mark_inode_dirty(handle
, inode
);
3065 ext4_journal_stop(handle
);
3066 trace_ext4_unlink_exit(dentry
, retval
);
3070 static int ext4_symlink(struct inode
*dir
,
3071 struct dentry
*dentry
, const char *symname
)
3074 struct inode
*inode
;
3075 int err
, len
= strlen(symname
);
3077 bool encryption_required
;
3078 struct ext4_str disk_link
;
3079 struct ext4_encrypted_symlink_data
*sd
= NULL
;
3081 disk_link
.len
= len
+ 1;
3082 disk_link
.name
= (char *) symname
;
3084 encryption_required
= (ext4_encrypted_inode(dir
) ||
3085 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir
->i_sb
)));
3086 if (encryption_required
) {
3087 err
= ext4_get_encryption_info(dir
);
3090 if (ext4_encryption_info(dir
) == NULL
)
3092 disk_link
.len
= (ext4_fname_encrypted_size(dir
, len
) +
3093 sizeof(struct ext4_encrypted_symlink_data
));
3094 sd
= kzalloc(disk_link
.len
, GFP_KERNEL
);
3099 if (disk_link
.len
> dir
->i_sb
->s_blocksize
) {
3100 err
= -ENAMETOOLONG
;
3104 err
= dquot_initialize(dir
);
3108 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3110 * For non-fast symlinks, we just allocate inode and put it on
3111 * orphan list in the first transaction => we need bitmap,
3112 * group descriptor, sb, inode block, quota blocks, and
3113 * possibly selinux xattr blocks.
3115 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
3116 EXT4_XATTR_TRANS_BLOCKS
;
3119 * Fast symlink. We have to add entry to directory
3120 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3121 * allocate new inode (bitmap, group descriptor, inode block,
3122 * quota blocks, sb is already counted in previous macros).
3124 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3125 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
3128 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
3129 &dentry
->d_name
, 0, NULL
,
3130 EXT4_HT_DIR
, credits
);
3131 handle
= ext4_journal_current_handle();
3132 if (IS_ERR(inode
)) {
3134 ext4_journal_stop(handle
);
3135 err
= PTR_ERR(inode
);
3139 if (encryption_required
) {
3141 struct ext4_str ostr
;
3143 istr
.name
= (const unsigned char *) symname
;
3145 ostr
.name
= sd
->encrypted_path
;
3146 ostr
.len
= disk_link
.len
;
3147 err
= ext4_fname_usr_to_disk(inode
, &istr
, &ostr
);
3149 goto err_drop_inode
;
3150 sd
->len
= cpu_to_le16(ostr
.len
);
3151 disk_link
.name
= (char *) sd
;
3152 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
3155 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3156 if (!encryption_required
)
3157 inode
->i_op
= &ext4_symlink_inode_operations
;
3158 inode_nohighmem(inode
);
3159 ext4_set_aops(inode
);
3161 * We cannot call page_symlink() with transaction started
3162 * because it calls into ext4_write_begin() which can wait
3163 * for transaction commit if we are running out of space
3164 * and thus we deadlock. So we have to stop transaction now
3165 * and restart it when symlink contents is written.
3167 * To keep fs consistent in case of crash, we have to put inode
3168 * to orphan list in the mean time.
3171 err
= ext4_orphan_add(handle
, inode
);
3172 ext4_journal_stop(handle
);
3175 goto err_drop_inode
;
3176 err
= __page_symlink(inode
, disk_link
.name
, disk_link
.len
, 1);
3178 goto err_drop_inode
;
3180 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3181 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3183 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3184 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3185 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
3186 if (IS_ERR(handle
)) {
3187 err
= PTR_ERR(handle
);
3189 goto err_drop_inode
;
3191 set_nlink(inode
, 1);
3192 err
= ext4_orphan_del(handle
, inode
);
3194 goto err_drop_inode
;
3196 /* clear the extent format for fast symlink */
3197 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
3198 if (!encryption_required
) {
3199 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3200 inode
->i_link
= (char *)&EXT4_I(inode
)->i_data
;
3202 memcpy((char *)&EXT4_I(inode
)->i_data
, disk_link
.name
,
3204 inode
->i_size
= disk_link
.len
- 1;
3206 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3207 err
= ext4_add_nondir(handle
, dentry
, inode
);
3208 if (!err
&& IS_DIRSYNC(dir
))
3209 ext4_handle_sync(handle
);
3212 ext4_journal_stop(handle
);
3217 ext4_journal_stop(handle
);
3219 unlock_new_inode(inode
);
3226 static int ext4_link(struct dentry
*old_dentry
,
3227 struct inode
*dir
, struct dentry
*dentry
)
3230 struct inode
*inode
= d_inode(old_dentry
);
3231 int err
, retries
= 0;
3233 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
3235 if (ext4_encrypted_inode(dir
) &&
3236 !ext4_is_child_context_consistent_with_parent(dir
, inode
))
3239 if ((ext4_test_inode_flag(dir
, EXT4_INODE_PROJINHERIT
)) &&
3240 (!projid_eq(EXT4_I(dir
)->i_projid
,
3241 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3244 err
= dquot_initialize(dir
);
3249 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3250 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3251 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
3253 return PTR_ERR(handle
);
3255 if (IS_DIRSYNC(dir
))
3256 ext4_handle_sync(handle
);
3258 inode
->i_ctime
= ext4_current_time(inode
);
3259 ext4_inc_count(handle
, inode
);
3262 err
= ext4_add_entry(handle
, dentry
, inode
);
3264 ext4_mark_inode_dirty(handle
, inode
);
3265 /* this can happen only for tmpfile being
3266 * linked the first time
3268 if (inode
->i_nlink
== 1)
3269 ext4_orphan_del(handle
, inode
);
3270 d_instantiate(dentry
, inode
);
3275 ext4_journal_stop(handle
);
3276 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3283 * Try to find buffer head where contains the parent block.
3284 * It should be the inode block if it is inlined or the 1st block
3285 * if it is a normal dir.
3287 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
3288 struct inode
*inode
,
3290 struct ext4_dir_entry_2
**parent_de
,
3293 struct buffer_head
*bh
;
3295 if (!ext4_has_inline_data(inode
)) {
3296 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
3298 *retval
= PTR_ERR(bh
);
3301 *parent_de
= ext4_next_entry(
3302 (struct ext4_dir_entry_2
*)bh
->b_data
,
3303 inode
->i_sb
->s_blocksize
);
3308 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3311 struct ext4_renament
{
3313 struct dentry
*dentry
;
3314 struct inode
*inode
;
3316 int dir_nlink_delta
;
3318 /* entry for "dentry" */
3319 struct buffer_head
*bh
;
3320 struct ext4_dir_entry_2
*de
;
3323 /* entry for ".." in inode if it's a directory */
3324 struct buffer_head
*dir_bh
;
3325 struct ext4_dir_entry_2
*parent_de
;
3329 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3333 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3334 &retval
, &ent
->parent_de
,
3338 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3339 return -EFSCORRUPTED
;
3340 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3341 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3344 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3349 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3350 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3351 if (!ent
->dir_inlined
) {
3352 if (is_dx(ent
->inode
)) {
3353 retval
= ext4_handle_dirty_dx_node(handle
,
3357 retval
= ext4_handle_dirty_dirent_node(handle
,
3362 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3365 ext4_std_error(ent
->dir
->i_sb
, retval
);
3371 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3372 unsigned ino
, unsigned file_type
)
3376 BUFFER_TRACE(ent
->bh
, "get write access");
3377 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3380 ent
->de
->inode
= cpu_to_le32(ino
);
3381 if (ext4_has_feature_filetype(ent
->dir
->i_sb
))
3382 ent
->de
->file_type
= file_type
;
3383 ent
->dir
->i_version
++;
3384 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3385 ext4_current_time(ent
->dir
);
3386 ext4_mark_inode_dirty(handle
, ent
->dir
);
3387 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3388 if (!ent
->inlined
) {
3389 retval
= ext4_handle_dirty_dirent_node(handle
,
3391 if (unlikely(retval
)) {
3392 ext4_std_error(ent
->dir
->i_sb
, retval
);
3402 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3403 const struct qstr
*d_name
)
3405 int retval
= -ENOENT
;
3406 struct buffer_head
*bh
;
3407 struct ext4_dir_entry_2
*de
;
3409 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3413 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3419 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3424 * ent->de could have moved from under us during htree split, so make
3425 * sure that we are deleting the right entry. We might also be pointing
3426 * to a stale entry in the unused part of ent->bh so just checking inum
3427 * and the name isn't enough.
3429 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3430 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3431 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3432 ent
->de
->name_len
) ||
3434 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3435 &ent
->dentry
->d_name
);
3437 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3438 if (retval
== -ENOENT
) {
3439 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3440 &ent
->dentry
->d_name
);
3445 ext4_warning_inode(ent
->dir
,
3446 "Deleting old file: nlink %d, error=%d",
3447 ent
->dir
->i_nlink
, retval
);
3451 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3453 if (ent
->dir_nlink_delta
) {
3454 if (ent
->dir_nlink_delta
== -1)
3455 ext4_dec_count(handle
, ent
->dir
);
3457 ext4_inc_count(handle
, ent
->dir
);
3458 ext4_mark_inode_dirty(handle
, ent
->dir
);
3462 static struct inode
*ext4_whiteout_for_rename(struct ext4_renament
*ent
,
3463 int credits
, handle_t
**h
)
3470 * for inode block, sb block, group summaries,
3473 credits
+= (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent
->dir
->i_sb
) +
3474 EXT4_XATTR_TRANS_BLOCKS
+ 4);
3476 wh
= ext4_new_inode_start_handle(ent
->dir
, S_IFCHR
| WHITEOUT_MODE
,
3477 &ent
->dentry
->d_name
, 0, NULL
,
3478 EXT4_HT_DIR
, credits
);
3480 handle
= ext4_journal_current_handle();
3483 ext4_journal_stop(handle
);
3484 if (PTR_ERR(wh
) == -ENOSPC
&&
3485 ext4_should_retry_alloc(ent
->dir
->i_sb
, &retries
))
3489 init_special_inode(wh
, wh
->i_mode
, WHITEOUT_DEV
);
3490 wh
->i_op
= &ext4_special_inode_operations
;
3496 * Anybody can rename anything with this: the permission checks are left to the
3497 * higher-level routines.
3499 * n.b. old_{dentry,inode) refers to the source dentry/inode
3500 * while new_{dentry,inode) refers to the destination dentry/inode
3501 * This comes from rename(const char *oldpath, const char *newpath)
3503 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3504 struct inode
*new_dir
, struct dentry
*new_dentry
,
3507 handle_t
*handle
= NULL
;
3508 struct ext4_renament old
= {
3510 .dentry
= old_dentry
,
3511 .inode
= d_inode(old_dentry
),
3513 struct ext4_renament
new = {
3515 .dentry
= new_dentry
,
3516 .inode
= d_inode(new_dentry
),
3520 struct inode
*whiteout
= NULL
;
3524 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
)) &&
3525 (!projid_eq(EXT4_I(new_dir
)->i_projid
,
3526 EXT4_I(old_dentry
->d_inode
)->i_projid
)))
3529 retval
= dquot_initialize(old
.dir
);
3532 retval
= dquot_initialize(new.dir
);
3536 /* Initialize quotas before so that eventual writes go
3537 * in separate transaction */
3539 retval
= dquot_initialize(new.inode
);
3544 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3546 return PTR_ERR(old
.bh
);
3548 * Check for inode number is _not_ due to possible IO errors.
3549 * We might rmdir the source, keep it as pwd of some process
3550 * and merrily kill the link to whatever was created under the
3551 * same name. Goodbye sticky bit ;-<
3554 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3557 if ((old
.dir
!= new.dir
) &&
3558 ext4_encrypted_inode(new.dir
) &&
3559 !ext4_is_child_context_consistent_with_parent(new.dir
,
3565 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3566 &new.de
, &new.inlined
);
3567 if (IS_ERR(new.bh
)) {
3568 retval
= PTR_ERR(new.bh
);
3578 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3579 ext4_alloc_da_blocks(old
.inode
);
3581 credits
= (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3582 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
3583 if (!(flags
& RENAME_WHITEOUT
)) {
3584 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
, credits
);
3585 if (IS_ERR(handle
)) {
3586 retval
= PTR_ERR(handle
);
3591 whiteout
= ext4_whiteout_for_rename(&old
, credits
, &handle
);
3592 if (IS_ERR(whiteout
)) {
3593 retval
= PTR_ERR(whiteout
);
3599 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3600 ext4_handle_sync(handle
);
3602 if (S_ISDIR(old
.inode
->i_mode
)) {
3604 retval
= -ENOTEMPTY
;
3605 if (!ext4_empty_dir(new.inode
))
3609 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3612 retval
= ext4_rename_dir_prepare(handle
, &old
);
3617 * If we're renaming a file within an inline_data dir and adding or
3618 * setting the new dirent causes a conversion from inline_data to
3619 * extents/blockmap, we need to force the dirent delete code to
3620 * re-read the directory, or else we end up trying to delete a dirent
3621 * from what is now the extent tree root (or a block map).
3623 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3624 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3626 old_file_type
= old
.de
->file_type
;
3629 * Do this before adding a new entry, so the old entry is sure
3630 * to be still pointing to the valid old entry.
3632 retval
= ext4_setent(handle
, &old
, whiteout
->i_ino
,
3636 ext4_mark_inode_dirty(handle
, whiteout
);
3639 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3643 retval
= ext4_setent(handle
, &new,
3644 old
.inode
->i_ino
, old_file_type
);
3649 force_reread
= !ext4_test_inode_flag(new.dir
,
3650 EXT4_INODE_INLINE_DATA
);
3653 * Like most other Unix systems, set the ctime for inodes on a
3656 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3657 ext4_mark_inode_dirty(handle
, old
.inode
);
3663 ext4_rename_delete(handle
, &old
, force_reread
);
3667 ext4_dec_count(handle
, new.inode
);
3668 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3670 old
.dir
->i_ctime
= old
.dir
->i_mtime
= ext4_current_time(old
.dir
);
3671 ext4_update_dx_flag(old
.dir
);
3673 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3677 ext4_dec_count(handle
, old
.dir
);
3679 /* checked ext4_empty_dir above, can't have another
3680 * parent, ext4_dec_count() won't work for many-linked
3682 clear_nlink(new.inode
);
3684 ext4_inc_count(handle
, new.dir
);
3685 ext4_update_dx_flag(new.dir
);
3686 ext4_mark_inode_dirty(handle
, new.dir
);
3689 ext4_mark_inode_dirty(handle
, old
.dir
);
3691 ext4_mark_inode_dirty(handle
, new.inode
);
3692 if (!new.inode
->i_nlink
)
3693 ext4_orphan_add(handle
, new.inode
);
3703 drop_nlink(whiteout
);
3704 unlock_new_inode(whiteout
);
3708 ext4_journal_stop(handle
);
3712 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3713 struct inode
*new_dir
, struct dentry
*new_dentry
)
3715 handle_t
*handle
= NULL
;
3716 struct ext4_renament old
= {
3718 .dentry
= old_dentry
,
3719 .inode
= d_inode(old_dentry
),
3721 struct ext4_renament
new = {
3723 .dentry
= new_dentry
,
3724 .inode
= d_inode(new_dentry
),
3729 if ((ext4_encrypted_inode(old_dir
) ||
3730 ext4_encrypted_inode(new_dir
)) &&
3731 (old_dir
!= new_dir
) &&
3732 (!ext4_is_child_context_consistent_with_parent(new_dir
,
3734 !ext4_is_child_context_consistent_with_parent(old_dir
,
3738 if ((ext4_test_inode_flag(new_dir
, EXT4_INODE_PROJINHERIT
) &&
3739 !projid_eq(EXT4_I(new_dir
)->i_projid
,
3740 EXT4_I(old_dentry
->d_inode
)->i_projid
)) ||
3741 (ext4_test_inode_flag(old_dir
, EXT4_INODE_PROJINHERIT
) &&
3742 !projid_eq(EXT4_I(old_dir
)->i_projid
,
3743 EXT4_I(new_dentry
->d_inode
)->i_projid
)))
3746 retval
= dquot_initialize(old
.dir
);
3749 retval
= dquot_initialize(new.dir
);
3753 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3754 &old
.de
, &old
.inlined
);
3756 return PTR_ERR(old
.bh
);
3758 * Check for inode number is _not_ due to possible IO errors.
3759 * We might rmdir the source, keep it as pwd of some process
3760 * and merrily kill the link to whatever was created under the
3761 * same name. Goodbye sticky bit ;-<
3764 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3767 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3768 &new.de
, &new.inlined
);
3769 if (IS_ERR(new.bh
)) {
3770 retval
= PTR_ERR(new.bh
);
3775 /* RENAME_EXCHANGE case: old *and* new must both exist */
3776 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3779 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3780 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3781 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3782 if (IS_ERR(handle
)) {
3783 retval
= PTR_ERR(handle
);
3788 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3789 ext4_handle_sync(handle
);
3791 if (S_ISDIR(old
.inode
->i_mode
)) {
3793 retval
= ext4_rename_dir_prepare(handle
, &old
);
3797 if (S_ISDIR(new.inode
->i_mode
)) {
3799 retval
= ext4_rename_dir_prepare(handle
, &new);
3805 * Other than the special case of overwriting a directory, parents'
3806 * nlink only needs to be modified if this is a cross directory rename.
3808 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3809 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3810 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3812 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3813 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3817 new_file_type
= new.de
->file_type
;
3818 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3822 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3827 * Like most other Unix systems, set the ctime for inodes on a
3830 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3831 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3832 ext4_mark_inode_dirty(handle
, old
.inode
);
3833 ext4_mark_inode_dirty(handle
, new.inode
);
3836 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3841 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3845 ext4_update_dir_count(handle
, &old
);
3846 ext4_update_dir_count(handle
, &new);
3855 ext4_journal_stop(handle
);
3859 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3860 struct inode
*new_dir
, struct dentry
*new_dentry
,
3863 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3866 if (flags
& RENAME_EXCHANGE
) {
3867 return ext4_cross_rename(old_dir
, old_dentry
,
3868 new_dir
, new_dentry
);
3871 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
3875 * directories can handle most operations...
3877 const struct inode_operations ext4_dir_inode_operations
= {
3878 .create
= ext4_create
,
3879 .lookup
= ext4_lookup
,
3881 .unlink
= ext4_unlink
,
3882 .symlink
= ext4_symlink
,
3883 .mkdir
= ext4_mkdir
,
3884 .rmdir
= ext4_rmdir
,
3885 .mknod
= ext4_mknod
,
3886 .tmpfile
= ext4_tmpfile
,
3887 .rename2
= ext4_rename2
,
3888 .setattr
= ext4_setattr
,
3889 .setxattr
= generic_setxattr
,
3890 .getxattr
= generic_getxattr
,
3891 .listxattr
= ext4_listxattr
,
3892 .removexattr
= generic_removexattr
,
3893 .get_acl
= ext4_get_acl
,
3894 .set_acl
= ext4_set_acl
,
3895 .fiemap
= ext4_fiemap
,
3898 const struct inode_operations ext4_special_inode_operations
= {
3899 .setattr
= ext4_setattr
,
3900 .setxattr
= generic_setxattr
,
3901 .getxattr
= generic_getxattr
,
3902 .listxattr
= ext4_listxattr
,
3903 .removexattr
= generic_removexattr
,
3904 .get_acl
= ext4_get_acl
,
3905 .set_acl
= ext4_set_acl
,