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 dentry
*dentry
, 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 block
= dx_get_block(frame
->at
);
1110 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
1111 start_hash
, start_minor_hash
);
1118 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
1119 frame
, frames
, &hashval
);
1120 *next_hash
= hashval
;
1126 * Stop if: (a) there are no more entries, or
1127 * (b) we have inserted at least one entry and the
1128 * next hash value is not a continuation
1131 (count
&& ((hashval
& 1) == 0)))
1135 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1136 "next hash: %x\n", count
, *next_hash
));
1143 static inline int search_dirblock(struct buffer_head
*bh
,
1145 struct ext4_filename
*fname
,
1146 const struct qstr
*d_name
,
1147 unsigned int offset
,
1148 struct ext4_dir_entry_2
**res_dir
)
1150 return ext4_search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1151 fname
, d_name
, offset
, res_dir
);
1155 * Directory block splitting, compacting
1159 * Create map of hash values, offsets, and sizes, stored at end of block.
1160 * Returns number of entries mapped.
1162 static int dx_make_map(struct inode
*dir
, struct ext4_dir_entry_2
*de
,
1163 unsigned blocksize
, struct dx_hash_info
*hinfo
,
1164 struct dx_map_entry
*map_tail
)
1167 char *base
= (char *) de
;
1168 struct dx_hash_info h
= *hinfo
;
1170 while ((char *) de
< base
+ blocksize
) {
1171 if (de
->name_len
&& de
->inode
) {
1172 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1174 map_tail
->hash
= h
.hash
;
1175 map_tail
->offs
= ((char *) de
- base
)>>2;
1176 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1180 /* XXX: do we need to check rec_len == 0 case? -Chris */
1181 de
= ext4_next_entry(de
, blocksize
);
1186 /* Sort map by hash value */
1187 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1189 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1191 /* Combsort until bubble sort doesn't suck */
1193 count
= count
*10/13;
1194 if (count
- 9 < 2) /* 9, 10 -> 11 */
1196 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1197 if (p
->hash
< q
->hash
)
1200 /* Garden variety bubble sort */
1205 if (q
[1].hash
>= q
[0].hash
)
1213 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1215 struct dx_entry
*entries
= frame
->entries
;
1216 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1217 int count
= dx_get_count(entries
);
1219 assert(count
< dx_get_limit(entries
));
1220 assert(old
< entries
+ count
);
1221 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1222 dx_set_hash(new, hash
);
1223 dx_set_block(new, block
);
1224 dx_set_count(entries
, count
+ 1);
1228 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1230 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1231 * `de != NULL' is guaranteed by caller.
1233 static inline int ext4_match(struct ext4_filename
*fname
,
1234 struct ext4_dir_entry_2
*de
)
1236 const void *name
= fname_name(fname
);
1237 u32 len
= fname_len(fname
);
1242 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1243 if (unlikely(!name
)) {
1244 if (fname
->usr_fname
->name
[0] == '_') {
1246 if (de
->name_len
<= 32)
1248 ret
= memcmp(de
->name
+ ((de
->name_len
- 17) & ~15),
1249 fname
->crypto_buf
.name
+ 8, 16);
1250 return (ret
== 0) ? 1 : 0;
1252 name
= fname
->crypto_buf
.name
;
1253 len
= fname
->crypto_buf
.len
;
1256 if (de
->name_len
!= len
)
1258 return (memcmp(de
->name
, name
, len
) == 0) ? 1 : 0;
1262 * Returns 0 if not found, -1 on failure, and 1 on success
1264 int ext4_search_dir(struct buffer_head
*bh
, char *search_buf
, int buf_size
,
1265 struct inode
*dir
, struct ext4_filename
*fname
,
1266 const struct qstr
*d_name
,
1267 unsigned int offset
, struct ext4_dir_entry_2
**res_dir
)
1269 struct ext4_dir_entry_2
* de
;
1274 de
= (struct ext4_dir_entry_2
*)search_buf
;
1275 dlimit
= search_buf
+ buf_size
;
1276 while ((char *) de
< dlimit
) {
1277 /* this code is executed quadratically often */
1278 /* do minimal checking `by hand' */
1279 if ((char *) de
+ de
->name_len
<= dlimit
) {
1280 res
= ext4_match(fname
, de
);
1286 /* found a match - just to be sure, do
1288 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1290 bh
->b_size
, offset
)) {
1300 /* prevent looping on a bad block */
1301 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1302 dir
->i_sb
->s_blocksize
);
1308 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1316 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1317 struct ext4_dir_entry
*de
)
1319 struct super_block
*sb
= dir
->i_sb
;
1325 if (de
->inode
== 0 &&
1326 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1335 * finds an entry in the specified directory with the wanted name. It
1336 * returns the cache buffer in which the entry was found, and the entry
1337 * itself (as a parameter - res_dir). It does NOT read the inode of the
1338 * entry - you'll have to do that yourself if you want to.
1340 * The returned buffer_head has ->b_count elevated. The caller is expected
1341 * to brelse() it when appropriate.
1343 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1344 const struct qstr
*d_name
,
1345 struct ext4_dir_entry_2
**res_dir
,
1348 struct super_block
*sb
;
1349 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1350 struct buffer_head
*bh
, *ret
= NULL
;
1351 ext4_lblk_t start
, block
, b
;
1352 const u8
*name
= d_name
->name
;
1353 int ra_max
= 0; /* Number of bh's in the readahead
1355 int ra_ptr
= 0; /* Current index into readahead
1358 ext4_lblk_t nblocks
;
1359 int i
, namelen
, retval
;
1360 struct ext4_filename fname
;
1364 namelen
= d_name
->len
;
1365 if (namelen
> EXT4_NAME_LEN
)
1368 retval
= ext4_fname_setup_filename(dir
, d_name
, 1, &fname
);
1370 return ERR_PTR(retval
);
1372 if (ext4_has_inline_data(dir
)) {
1373 int has_inline_data
= 1;
1374 ret
= ext4_find_inline_entry(dir
, &fname
, d_name
, res_dir
,
1376 if (has_inline_data
) {
1379 goto cleanup_and_exit
;
1383 if ((namelen
<= 2) && (name
[0] == '.') &&
1384 (name
[1] == '.' || name
[1] == '\0')) {
1386 * "." or ".." will only be in the first block
1387 * NFS may look up ".."; "." should be handled by the VFS
1394 ret
= ext4_dx_find_entry(dir
, &fname
, res_dir
);
1396 * On success, or if the error was file not found,
1397 * return. Otherwise, fall back to doing a search the
1398 * old fashioned way.
1400 if (!IS_ERR(ret
) || PTR_ERR(ret
) != ERR_BAD_DX_DIR
)
1401 goto cleanup_and_exit
;
1402 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1406 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1409 goto cleanup_and_exit
;
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_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1611 S_ISLNK(inode
->i_mode
)) &&
1612 !ext4_is_child_context_consistent_with_parent(dir
,
1615 ext4_warning(inode
->i_sb
,
1616 "Inconsistent encryption contexts: %lu/%lu\n",
1617 (unsigned long) dir
->i_ino
,
1618 (unsigned long) inode
->i_ino
);
1619 return ERR_PTR(-EPERM
);
1622 return d_splice_alias(inode
, dentry
);
1626 struct dentry
*ext4_get_parent(struct dentry
*child
)
1629 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1630 struct ext4_dir_entry_2
* de
;
1631 struct buffer_head
*bh
;
1633 bh
= ext4_find_entry(d_inode(child
), &dotdot
, &de
, NULL
);
1635 return (struct dentry
*) bh
;
1637 return ERR_PTR(-ENOENT
);
1638 ino
= le32_to_cpu(de
->inode
);
1641 if (!ext4_valid_inum(d_inode(child
)->i_sb
, ino
)) {
1642 EXT4_ERROR_INODE(d_inode(child
),
1643 "bad parent inode number: %u", ino
);
1644 return ERR_PTR(-EFSCORRUPTED
);
1647 return d_obtain_alias(ext4_iget_normal(d_inode(child
)->i_sb
, ino
));
1651 * Move count entries from end of map between two memory locations.
1652 * Returns pointer to last entry moved.
1654 static struct ext4_dir_entry_2
*
1655 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1658 unsigned rec_len
= 0;
1661 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1662 (from
+ (map
->offs
<<2));
1663 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1664 memcpy (to
, de
, rec_len
);
1665 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1666 ext4_rec_len_to_disk(rec_len
, blocksize
);
1671 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1675 * Compact each dir entry in the range to the minimal rec_len.
1676 * Returns pointer to last entry in range.
1678 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1680 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1681 unsigned rec_len
= 0;
1684 while ((char*)de
< base
+ blocksize
) {
1685 next
= ext4_next_entry(de
, blocksize
);
1686 if (de
->inode
&& de
->name_len
) {
1687 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1689 memmove(to
, de
, rec_len
);
1690 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1692 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1700 * Split a full leaf block to make room for a new dir entry.
1701 * Allocate a new block, and move entries so that they are approx. equally full.
1702 * Returns pointer to de in block into which the new entry will be inserted.
1704 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1705 struct buffer_head
**bh
,struct dx_frame
*frame
,
1706 struct dx_hash_info
*hinfo
)
1708 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1709 unsigned count
, continued
;
1710 struct buffer_head
*bh2
;
1711 ext4_lblk_t newblock
;
1713 struct dx_map_entry
*map
;
1714 char *data1
= (*bh
)->b_data
, *data2
;
1715 unsigned split
, move
, size
;
1716 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1717 struct ext4_dir_entry_tail
*t
;
1721 if (ext4_has_metadata_csum(dir
->i_sb
))
1722 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1724 bh2
= ext4_append(handle
, dir
, &newblock
);
1728 return (struct ext4_dir_entry_2
*) bh2
;
1731 BUFFER_TRACE(*bh
, "get_write_access");
1732 err
= ext4_journal_get_write_access(handle
, *bh
);
1736 BUFFER_TRACE(frame
->bh
, "get_write_access");
1737 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1741 data2
= bh2
->b_data
;
1743 /* create map in the end of data2 block */
1744 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1745 count
= dx_make_map(dir
, (struct ext4_dir_entry_2
*) data1
,
1746 blocksize
, hinfo
, map
);
1748 dx_sort_map(map
, count
);
1749 /* Split the existing block in the middle, size-wise */
1752 for (i
= count
-1; i
>= 0; i
--) {
1753 /* is more than half of this entry in 2nd half of the block? */
1754 if (size
+ map
[i
].size
/2 > blocksize
/2)
1756 size
+= map
[i
].size
;
1759 /* map index at which we will split */
1760 split
= count
- move
;
1761 hash2
= map
[split
].hash
;
1762 continued
= hash2
== map
[split
- 1].hash
;
1763 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1764 (unsigned long)dx_get_block(frame
->at
),
1765 hash2
, split
, count
-split
));
1767 /* Fancy dance to stay within two buffers */
1768 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
,
1770 de
= dx_pack_dirents(data1
, blocksize
);
1771 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1774 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1778 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1779 initialize_dirent_tail(t
, blocksize
);
1781 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1782 initialize_dirent_tail(t
, blocksize
);
1785 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data1
,
1787 dxtrace(dx_show_leaf(dir
, hinfo
, (struct ext4_dir_entry_2
*) data2
,
1790 /* Which block gets the new entry? */
1791 if (hinfo
->hash
>= hash2
) {
1795 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1796 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1799 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1803 dxtrace(dx_show_index("frame", frame
->entries
));
1810 ext4_std_error(dir
->i_sb
, err
);
1811 return ERR_PTR(err
);
1814 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1815 struct buffer_head
*bh
,
1816 void *buf
, int buf_size
,
1817 struct ext4_filename
*fname
,
1818 struct ext4_dir_entry_2
**dest_de
)
1820 struct ext4_dir_entry_2
*de
;
1821 unsigned short reclen
= EXT4_DIR_REC_LEN(fname_len(fname
));
1823 unsigned int offset
= 0;
1827 de
= (struct ext4_dir_entry_2
*)buf
;
1828 top
= buf
+ buf_size
- reclen
;
1829 while ((char *) de
<= top
) {
1830 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1831 buf
, buf_size
, offset
)) {
1832 res
= -EFSCORRUPTED
;
1835 /* Provide crypto context and crypto buffer to ext4 match */
1836 res
= ext4_match(fname
, de
);
1843 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1844 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1845 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1847 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1851 if ((char *) de
> top
)
1861 int ext4_insert_dentry(struct inode
*dir
,
1862 struct inode
*inode
,
1863 struct ext4_dir_entry_2
*de
,
1865 struct ext4_filename
*fname
)
1870 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1871 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1873 struct ext4_dir_entry_2
*de1
=
1874 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1875 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1876 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1879 de
->file_type
= EXT4_FT_UNKNOWN
;
1880 de
->inode
= cpu_to_le32(inode
->i_ino
);
1881 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1882 de
->name_len
= fname_len(fname
);
1883 memcpy(de
->name
, fname_name(fname
), fname_len(fname
));
1888 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1889 * it points to a directory entry which is guaranteed to be large
1890 * enough for new directory entry. If de is NULL, then
1891 * add_dirent_to_buf will attempt search the directory block for
1892 * space. It will return -ENOSPC if no space is available, and -EIO
1893 * and -EEXIST if directory entry already exists.
1895 static int add_dirent_to_buf(handle_t
*handle
, struct ext4_filename
*fname
,
1897 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1898 struct buffer_head
*bh
)
1900 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1904 if (ext4_has_metadata_csum(inode
->i_sb
))
1905 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1908 err
= ext4_find_dest_de(dir
, inode
, bh
, bh
->b_data
,
1909 blocksize
- csum_size
, fname
, &de
);
1913 BUFFER_TRACE(bh
, "get_write_access");
1914 err
= ext4_journal_get_write_access(handle
, bh
);
1916 ext4_std_error(dir
->i_sb
, err
);
1920 /* By now the buffer is marked for journaling. Due to crypto operations,
1921 * the following function call may fail */
1922 err
= ext4_insert_dentry(dir
, inode
, de
, blocksize
, fname
);
1927 * XXX shouldn't update any times until successful
1928 * completion of syscall, but too many callers depend
1931 * XXX similarly, too many callers depend on
1932 * ext4_new_inode() setting the times, but error
1933 * recovery deletes the inode, so the worst that can
1934 * happen is that the times are slightly out of date
1935 * and/or different from the directory change time.
1937 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1938 ext4_update_dx_flag(dir
);
1940 ext4_mark_inode_dirty(handle
, dir
);
1941 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1942 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1944 ext4_std_error(dir
->i_sb
, err
);
1949 * This converts a one block unindexed directory to a 3 block indexed
1950 * directory, and adds the dentry to the indexed directory.
1952 static int make_indexed_dir(handle_t
*handle
, struct ext4_filename
*fname
,
1953 struct dentry
*dentry
,
1954 struct inode
*inode
, struct buffer_head
*bh
)
1956 struct inode
*dir
= d_inode(dentry
->d_parent
);
1957 struct buffer_head
*bh2
;
1958 struct dx_root
*root
;
1959 struct dx_frame frames
[2], *frame
;
1960 struct dx_entry
*entries
;
1961 struct ext4_dir_entry_2
*de
, *de2
;
1962 struct ext4_dir_entry_tail
*t
;
1968 struct fake_dirent
*fde
;
1971 if (ext4_has_metadata_csum(inode
->i_sb
))
1972 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1974 blocksize
= dir
->i_sb
->s_blocksize
;
1975 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1976 BUFFER_TRACE(bh
, "get_write_access");
1977 retval
= ext4_journal_get_write_access(handle
, bh
);
1979 ext4_std_error(dir
->i_sb
, retval
);
1983 root
= (struct dx_root
*) bh
->b_data
;
1985 /* The 0th block becomes the root, move the dirents out */
1986 fde
= &root
->dotdot
;
1987 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1988 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1989 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1990 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1992 return -EFSCORRUPTED
;
1994 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1996 /* Allocate new block for the 0th block's dirents */
1997 bh2
= ext4_append(handle
, dir
, &block
);
2000 return PTR_ERR(bh2
);
2002 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
2003 data1
= bh2
->b_data
;
2005 memcpy (data1
, de
, len
);
2006 de
= (struct ext4_dir_entry_2
*) data1
;
2008 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
2010 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
2015 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
2016 initialize_dirent_tail(t
, blocksize
);
2019 /* Initialize the root; the dot dirents already exist */
2020 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
2021 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
2023 memset (&root
->info
, 0, sizeof(root
->info
));
2024 root
->info
.info_length
= sizeof(root
->info
);
2025 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
2026 entries
= root
->entries
;
2027 dx_set_block(entries
, 1);
2028 dx_set_count(entries
, 1);
2029 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
2031 /* Initialize as for dx_probe */
2032 fname
->hinfo
.hash_version
= root
->info
.hash_version
;
2033 if (fname
->hinfo
.hash_version
<= DX_HASH_TEA
)
2034 fname
->hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
2035 fname
->hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
2036 ext4fs_dirhash(fname_name(fname
), fname_len(fname
), &fname
->hinfo
);
2038 memset(frames
, 0, sizeof(frames
));
2040 frame
->entries
= entries
;
2041 frame
->at
= entries
;
2044 retval
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
2047 retval
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
2051 de
= do_split(handle
,dir
, &bh2
, frame
, &fname
->hinfo
);
2053 retval
= PTR_ERR(de
);
2057 retval
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh2
);
2060 * Even if the block split failed, we have to properly write
2061 * out all the changes we did so far. Otherwise we can end up
2062 * with corrupted filesystem.
2065 ext4_mark_inode_dirty(handle
, dir
);
2074 * adds a file entry to the specified directory, using the same
2075 * semantics as ext4_find_entry(). It returns NULL if it failed.
2077 * NOTE!! The inode part of 'de' is left at 0 - which means you
2078 * may not sleep between calling this and putting something into
2079 * the entry, as someone else might have used it while you slept.
2081 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
2082 struct inode
*inode
)
2084 struct inode
*dir
= d_inode(dentry
->d_parent
);
2085 struct buffer_head
*bh
= NULL
;
2086 struct ext4_dir_entry_2
*de
;
2087 struct ext4_dir_entry_tail
*t
;
2088 struct super_block
*sb
;
2089 struct ext4_filename fname
;
2093 ext4_lblk_t block
, blocks
;
2096 if (ext4_has_metadata_csum(inode
->i_sb
))
2097 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2100 blocksize
= sb
->s_blocksize
;
2101 if (!dentry
->d_name
.len
)
2104 retval
= ext4_fname_setup_filename(dir
, &dentry
->d_name
, 0, &fname
);
2108 if (ext4_has_inline_data(dir
)) {
2109 retval
= ext4_try_add_inline_entry(handle
, &fname
,
2120 retval
= ext4_dx_add_entry(handle
, &fname
, dentry
, inode
);
2121 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
2123 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
2125 ext4_mark_inode_dirty(handle
, dir
);
2127 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
2128 for (block
= 0; block
< blocks
; block
++) {
2129 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
2131 retval
= PTR_ERR(bh
);
2135 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
,
2137 if (retval
!= -ENOSPC
)
2140 if (blocks
== 1 && !dx_fallback
&&
2141 ext4_has_feature_dir_index(sb
)) {
2142 retval
= make_indexed_dir(handle
, &fname
, dentry
,
2144 bh
= NULL
; /* make_indexed_dir releases bh */
2149 bh
= ext4_append(handle
, dir
, &block
);
2151 retval
= PTR_ERR(bh
);
2155 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2157 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
2160 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
2161 initialize_dirent_tail(t
, blocksize
);
2164 retval
= add_dirent_to_buf(handle
, &fname
, dir
, inode
, de
, bh
);
2166 ext4_fname_free_filename(&fname
);
2169 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
2174 * Returns 0 for success, or a negative error value
2176 static int ext4_dx_add_entry(handle_t
*handle
, struct ext4_filename
*fname
,
2177 struct dentry
*dentry
, struct inode
*inode
)
2179 struct dx_frame frames
[2], *frame
;
2180 struct dx_entry
*entries
, *at
;
2181 struct buffer_head
*bh
;
2182 struct inode
*dir
= d_inode(dentry
->d_parent
);
2183 struct super_block
*sb
= dir
->i_sb
;
2184 struct ext4_dir_entry_2
*de
;
2187 frame
= dx_probe(fname
, dir
, NULL
, frames
);
2189 return PTR_ERR(frame
);
2190 entries
= frame
->entries
;
2192 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
2199 BUFFER_TRACE(bh
, "get_write_access");
2200 err
= ext4_journal_get_write_access(handle
, bh
);
2204 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, NULL
, bh
);
2208 /* Block full, should compress but for now just split */
2209 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
2210 dx_get_count(entries
), dx_get_limit(entries
)));
2211 /* Need to split index? */
2212 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
2213 ext4_lblk_t newblock
;
2214 unsigned icount
= dx_get_count(entries
);
2215 int levels
= frame
- frames
;
2216 struct dx_entry
*entries2
;
2217 struct dx_node
*node2
;
2218 struct buffer_head
*bh2
;
2220 if (levels
&& (dx_get_count(frames
->entries
) ==
2221 dx_get_limit(frames
->entries
))) {
2222 ext4_warning_inode(dir
, "Directory index full!");
2226 bh2
= ext4_append(handle
, dir
, &newblock
);
2231 node2
= (struct dx_node
*)(bh2
->b_data
);
2232 entries2
= node2
->entries
;
2233 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2234 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2236 BUFFER_TRACE(frame
->bh
, "get_write_access");
2237 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2241 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2242 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2243 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2246 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2247 err
= ext4_journal_get_write_access(handle
,
2252 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2253 icount2
* sizeof(struct dx_entry
));
2254 dx_set_count(entries
, icount1
);
2255 dx_set_count(entries2
, icount2
);
2256 dx_set_limit(entries2
, dx_node_limit(dir
));
2258 /* Which index block gets the new entry? */
2259 if (at
- entries
>= icount1
) {
2260 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2261 frame
->entries
= entries
= entries2
;
2262 swap(frame
->bh
, bh2
);
2264 dx_insert_block(frames
+ 0, hash2
, newblock
);
2265 dxtrace(dx_show_index("node", frames
[1].entries
));
2266 dxtrace(dx_show_index("node",
2267 ((struct dx_node
*) bh2
->b_data
)->entries
));
2268 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2273 dxtrace(printk(KERN_DEBUG
2274 "Creating second level index...\n"));
2275 memcpy((char *) entries2
, (char *) entries
,
2276 icount
* sizeof(struct dx_entry
));
2277 dx_set_limit(entries2
, dx_node_limit(dir
));
2280 dx_set_count(entries
, 1);
2281 dx_set_block(entries
+ 0, newblock
);
2282 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2284 /* Add new access path frame */
2286 frame
->at
= at
= at
- entries
+ entries2
;
2287 frame
->entries
= entries
= entries2
;
2289 err
= ext4_journal_get_write_access(handle
,
2294 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2296 ext4_std_error(inode
->i_sb
, err
);
2300 de
= do_split(handle
, dir
, &bh
, frame
, &fname
->hinfo
);
2305 err
= add_dirent_to_buf(handle
, fname
, dir
, inode
, de
, bh
);
2309 ext4_std_error(dir
->i_sb
, err
);
2317 * ext4_generic_delete_entry deletes a directory entry by merging it
2318 * with the previous entry
2320 int ext4_generic_delete_entry(handle_t
*handle
,
2322 struct ext4_dir_entry_2
*de_del
,
2323 struct buffer_head
*bh
,
2328 struct ext4_dir_entry_2
*de
, *pde
;
2329 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2334 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2335 while (i
< buf_size
- csum_size
) {
2336 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2337 bh
->b_data
, bh
->b_size
, i
))
2338 return -EFSCORRUPTED
;
2341 pde
->rec_len
= ext4_rec_len_to_disk(
2342 ext4_rec_len_from_disk(pde
->rec_len
,
2344 ext4_rec_len_from_disk(de
->rec_len
,
2352 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2354 de
= ext4_next_entry(de
, blocksize
);
2359 static int ext4_delete_entry(handle_t
*handle
,
2361 struct ext4_dir_entry_2
*de_del
,
2362 struct buffer_head
*bh
)
2364 int err
, csum_size
= 0;
2366 if (ext4_has_inline_data(dir
)) {
2367 int has_inline_data
= 1;
2368 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2370 if (has_inline_data
)
2374 if (ext4_has_metadata_csum(dir
->i_sb
))
2375 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2377 BUFFER_TRACE(bh
, "get_write_access");
2378 err
= ext4_journal_get_write_access(handle
, bh
);
2382 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2384 dir
->i_sb
->s_blocksize
, csum_size
);
2388 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2389 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2396 ext4_std_error(dir
->i_sb
, err
);
2401 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2402 * since this indicates that nlinks count was previously 1.
2404 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2407 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2408 /* limit is 16-bit i_links_count */
2409 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2410 set_nlink(inode
, 1);
2411 ext4_set_feature_dir_nlink(inode
->i_sb
);
2417 * If a directory had nlink == 1, then we should let it be 1. This indicates
2418 * directory has >EXT4_LINK_MAX subdirs.
2420 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2422 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2427 static int ext4_add_nondir(handle_t
*handle
,
2428 struct dentry
*dentry
, struct inode
*inode
)
2430 int err
= ext4_add_entry(handle
, dentry
, inode
);
2432 ext4_mark_inode_dirty(handle
, inode
);
2433 d_instantiate_new(dentry
, inode
);
2437 unlock_new_inode(inode
);
2443 * By the time this is called, we already have created
2444 * the directory cache entry for the new file, but it
2445 * is so far negative - it has no inode.
2447 * If the create succeeds, we fill in the inode information
2448 * with d_instantiate().
2450 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2454 struct inode
*inode
;
2455 int err
, credits
, retries
= 0;
2457 err
= dquot_initialize(dir
);
2461 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2462 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2464 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2465 NULL
, EXT4_HT_DIR
, credits
);
2466 handle
= ext4_journal_current_handle();
2467 err
= PTR_ERR(inode
);
2468 if (!IS_ERR(inode
)) {
2469 inode
->i_op
= &ext4_file_inode_operations
;
2470 inode
->i_fop
= &ext4_file_operations
;
2471 ext4_set_aops(inode
);
2472 err
= ext4_add_nondir(handle
, dentry
, inode
);
2473 if (!err
&& IS_DIRSYNC(dir
))
2474 ext4_handle_sync(handle
);
2477 ext4_journal_stop(handle
);
2478 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2483 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2484 umode_t mode
, dev_t rdev
)
2487 struct inode
*inode
;
2488 int err
, credits
, retries
= 0;
2490 err
= dquot_initialize(dir
);
2494 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2495 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2497 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2498 NULL
, EXT4_HT_DIR
, credits
);
2499 handle
= ext4_journal_current_handle();
2500 err
= PTR_ERR(inode
);
2501 if (!IS_ERR(inode
)) {
2502 init_special_inode(inode
, inode
->i_mode
, rdev
);
2503 inode
->i_op
= &ext4_special_inode_operations
;
2504 err
= ext4_add_nondir(handle
, dentry
, inode
);
2505 if (!err
&& IS_DIRSYNC(dir
))
2506 ext4_handle_sync(handle
);
2509 ext4_journal_stop(handle
);
2510 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2515 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2518 struct inode
*inode
;
2519 int err
, retries
= 0;
2521 err
= dquot_initialize(dir
);
2526 inode
= ext4_new_inode_start_handle(dir
, mode
,
2529 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2530 4 + EXT4_XATTR_TRANS_BLOCKS
);
2531 handle
= ext4_journal_current_handle();
2532 err
= PTR_ERR(inode
);
2533 if (!IS_ERR(inode
)) {
2534 inode
->i_op
= &ext4_file_inode_operations
;
2535 inode
->i_fop
= &ext4_file_operations
;
2536 ext4_set_aops(inode
);
2537 d_tmpfile(dentry
, inode
);
2538 err
= ext4_orphan_add(handle
, inode
);
2540 goto err_unlock_inode
;
2541 mark_inode_dirty(inode
);
2542 unlock_new_inode(inode
);
2545 ext4_journal_stop(handle
);
2546 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2550 ext4_journal_stop(handle
);
2551 unlock_new_inode(inode
);
2555 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2556 struct ext4_dir_entry_2
*de
,
2557 int blocksize
, int csum_size
,
2558 unsigned int parent_ino
, int dotdot_real_len
)
2560 de
->inode
= cpu_to_le32(inode
->i_ino
);
2562 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2564 strcpy(de
->name
, ".");
2565 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2567 de
= ext4_next_entry(de
, blocksize
);
2568 de
->inode
= cpu_to_le32(parent_ino
);
2570 if (!dotdot_real_len
)
2571 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2572 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2575 de
->rec_len
= ext4_rec_len_to_disk(
2576 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2577 strcpy(de
->name
, "..");
2578 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2580 return ext4_next_entry(de
, blocksize
);
2583 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2584 struct inode
*inode
)
2586 struct buffer_head
*dir_block
= NULL
;
2587 struct ext4_dir_entry_2
*de
;
2588 struct ext4_dir_entry_tail
*t
;
2589 ext4_lblk_t block
= 0;
2590 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2594 if (ext4_has_metadata_csum(dir
->i_sb
))
2595 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2597 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2598 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2599 if (err
< 0 && err
!= -ENOSPC
)
2606 dir_block
= ext4_append(handle
, inode
, &block
);
2607 if (IS_ERR(dir_block
))
2608 return PTR_ERR(dir_block
);
2609 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2610 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2611 set_nlink(inode
, 2);
2613 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2614 initialize_dirent_tail(t
, blocksize
);
2617 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2618 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2621 set_buffer_verified(dir_block
);
2627 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2630 struct inode
*inode
;
2631 int err
, credits
, retries
= 0;
2633 if (EXT4_DIR_LINK_MAX(dir
))
2636 err
= dquot_initialize(dir
);
2640 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2641 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2643 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2645 0, NULL
, EXT4_HT_DIR
, credits
);
2646 handle
= ext4_journal_current_handle();
2647 err
= PTR_ERR(inode
);
2651 inode
->i_op
= &ext4_dir_inode_operations
;
2652 inode
->i_fop
= &ext4_dir_operations
;
2653 err
= ext4_init_new_dir(handle
, dir
, inode
);
2655 goto out_clear_inode
;
2656 err
= ext4_mark_inode_dirty(handle
, inode
);
2658 err
= ext4_add_entry(handle
, dentry
, inode
);
2662 unlock_new_inode(inode
);
2663 ext4_mark_inode_dirty(handle
, inode
);
2667 ext4_inc_count(handle
, dir
);
2668 ext4_update_dx_flag(dir
);
2669 err
= ext4_mark_inode_dirty(handle
, dir
);
2671 goto out_clear_inode
;
2672 d_instantiate_new(dentry
, inode
);
2673 if (IS_DIRSYNC(dir
))
2674 ext4_handle_sync(handle
);
2678 ext4_journal_stop(handle
);
2679 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2685 * routine to check that the specified directory is empty (for rmdir)
2687 int ext4_empty_dir(struct inode
*inode
)
2689 unsigned int offset
;
2690 struct buffer_head
*bh
;
2691 struct ext4_dir_entry_2
*de
, *de1
;
2692 struct super_block
*sb
;
2695 if (ext4_has_inline_data(inode
)) {
2696 int has_inline_data
= 1;
2698 err
= empty_inline_dir(inode
, &has_inline_data
);
2699 if (has_inline_data
)
2704 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2705 EXT4_ERROR_INODE(inode
, "invalid size");
2708 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2712 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2713 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2714 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2715 le32_to_cpu(de1
->inode
) == 0 ||
2716 strcmp(".", de
->name
) || strcmp("..", de1
->name
)) {
2717 ext4_warning_inode(inode
, "directory missing '.' and/or '..'");
2721 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2722 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2723 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2724 while (offset
< inode
->i_size
) {
2725 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2726 unsigned int lblock
;
2729 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2730 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2733 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2735 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2736 bh
->b_data
, bh
->b_size
, offset
)) {
2737 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2739 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2742 if (le32_to_cpu(de
->inode
)) {
2746 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2747 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2754 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2755 * such inodes, starting at the superblock, in case we crash before the
2756 * file is closed/deleted, or in case the inode truncate spans multiple
2757 * transactions and the last transaction is not recovered after a crash.
2759 * At filesystem recovery time, we walk this list deleting unlinked
2760 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2762 * Orphan list manipulation functions must be called under i_mutex unless
2763 * we are just creating the inode or deleting it.
2765 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2767 struct super_block
*sb
= inode
->i_sb
;
2768 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2769 struct ext4_iloc iloc
;
2773 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2776 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2777 !mutex_is_locked(&inode
->i_mutex
));
2779 * Exit early if inode already is on orphan list. This is a big speedup
2780 * since we don't have to contend on the global s_orphan_lock.
2782 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2786 * Orphan handling is only valid for files with data blocks
2787 * being truncated, or files being unlinked. Note that we either
2788 * hold i_mutex, or the inode can not be referenced from outside,
2789 * so i_nlink should not be bumped due to race
2791 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2792 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2794 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2795 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2799 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2803 mutex_lock(&sbi
->s_orphan_lock
);
2805 * Due to previous errors inode may be already a part of on-disk
2806 * orphan list. If so skip on-disk list modification.
2808 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2809 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2810 /* Insert this inode at the head of the on-disk orphan list */
2811 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2812 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2815 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2816 mutex_unlock(&sbi
->s_orphan_lock
);
2819 err
= ext4_handle_dirty_super(handle
, sb
);
2820 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2825 * We have to remove inode from in-memory list if
2826 * addition to on disk orphan list failed. Stray orphan
2827 * list entries can cause panics at unmount time.
2829 mutex_lock(&sbi
->s_orphan_lock
);
2830 list_del_init(&EXT4_I(inode
)->i_orphan
);
2831 mutex_unlock(&sbi
->s_orphan_lock
);
2834 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2835 jbd_debug(4, "orphan inode %lu will point to %d\n",
2836 inode
->i_ino
, NEXT_ORPHAN(inode
));
2838 ext4_std_error(sb
, err
);
2843 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2844 * of such inodes stored on disk, because it is finally being cleaned up.
2846 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2848 struct list_head
*prev
;
2849 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2850 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2852 struct ext4_iloc iloc
;
2855 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2858 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2859 !mutex_is_locked(&inode
->i_mutex
));
2860 /* Do this quick check before taking global s_orphan_lock. */
2861 if (list_empty(&ei
->i_orphan
))
2865 /* Grab inode buffer early before taking global s_orphan_lock */
2866 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2869 mutex_lock(&sbi
->s_orphan_lock
);
2870 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2872 prev
= ei
->i_orphan
.prev
;
2873 list_del_init(&ei
->i_orphan
);
2875 /* If we're on an error path, we may not have a valid
2876 * transaction handle with which to update the orphan list on
2877 * disk, but we still need to remove the inode from the linked
2878 * list in memory. */
2879 if (!handle
|| err
) {
2880 mutex_unlock(&sbi
->s_orphan_lock
);
2884 ino_next
= NEXT_ORPHAN(inode
);
2885 if (prev
== &sbi
->s_orphan
) {
2886 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2887 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2888 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2890 mutex_unlock(&sbi
->s_orphan_lock
);
2893 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2894 mutex_unlock(&sbi
->s_orphan_lock
);
2895 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2897 struct ext4_iloc iloc2
;
2898 struct inode
*i_prev
=
2899 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2901 jbd_debug(4, "orphan inode %lu will point to %u\n",
2902 i_prev
->i_ino
, ino_next
);
2903 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2905 mutex_unlock(&sbi
->s_orphan_lock
);
2908 NEXT_ORPHAN(i_prev
) = ino_next
;
2909 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2910 mutex_unlock(&sbi
->s_orphan_lock
);
2914 NEXT_ORPHAN(inode
) = 0;
2915 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2917 ext4_std_error(inode
->i_sb
, err
);
2925 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2928 struct inode
*inode
;
2929 struct buffer_head
*bh
;
2930 struct ext4_dir_entry_2
*de
;
2931 handle_t
*handle
= NULL
;
2933 /* Initialize quotas before so that eventual writes go in
2934 * separate transaction */
2935 retval
= dquot_initialize(dir
);
2938 retval
= dquot_initialize(d_inode(dentry
));
2943 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2949 inode
= d_inode(dentry
);
2951 retval
= -EFSCORRUPTED
;
2952 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2955 retval
= -ENOTEMPTY
;
2956 if (!ext4_empty_dir(inode
))
2959 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2960 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2961 if (IS_ERR(handle
)) {
2962 retval
= PTR_ERR(handle
);
2967 if (IS_DIRSYNC(dir
))
2968 ext4_handle_sync(handle
);
2970 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2973 if (!EXT4_DIR_LINK_EMPTY(inode
))
2974 ext4_warning_inode(inode
,
2975 "empty directory '%.*s' has too many links (%u)",
2976 dentry
->d_name
.len
, dentry
->d_name
.name
,
2980 /* There's no need to set i_disksize: the fact that i_nlink is
2981 * zero will ensure that the right thing happens during any
2984 ext4_orphan_add(handle
, inode
);
2985 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2986 ext4_mark_inode_dirty(handle
, inode
);
2987 ext4_dec_count(handle
, dir
);
2988 ext4_update_dx_flag(dir
);
2989 ext4_mark_inode_dirty(handle
, dir
);
2994 ext4_journal_stop(handle
);
2998 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
3001 struct inode
*inode
;
3002 struct buffer_head
*bh
;
3003 struct ext4_dir_entry_2
*de
;
3004 handle_t
*handle
= NULL
;
3006 trace_ext4_unlink_enter(dir
, dentry
);
3007 /* Initialize quotas before so that eventual writes go
3008 * in separate transaction */
3009 retval
= dquot_initialize(dir
);
3012 retval
= dquot_initialize(d_inode(dentry
));
3017 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
3023 inode
= d_inode(dentry
);
3025 retval
= -EFSCORRUPTED
;
3026 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
3029 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3030 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
3031 if (IS_ERR(handle
)) {
3032 retval
= PTR_ERR(handle
);
3037 if (IS_DIRSYNC(dir
))
3038 ext4_handle_sync(handle
);
3040 if (inode
->i_nlink
== 0) {
3041 ext4_warning_inode(inode
, "Deleting file '%.*s' with no links",
3042 dentry
->d_name
.len
, dentry
->d_name
.name
);
3043 set_nlink(inode
, 1);
3045 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3048 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
3049 ext4_update_dx_flag(dir
);
3050 ext4_mark_inode_dirty(handle
, dir
);
3052 if (!inode
->i_nlink
)
3053 ext4_orphan_add(handle
, inode
);
3054 inode
->i_ctime
= ext4_current_time(inode
);
3055 ext4_mark_inode_dirty(handle
, inode
);
3060 ext4_journal_stop(handle
);
3061 trace_ext4_unlink_exit(dentry
, retval
);
3065 static int ext4_symlink(struct inode
*dir
,
3066 struct dentry
*dentry
, const char *symname
)
3069 struct inode
*inode
;
3070 int err
, len
= strlen(symname
);
3072 bool encryption_required
;
3073 struct ext4_str disk_link
;
3074 struct ext4_encrypted_symlink_data
*sd
= NULL
;
3076 disk_link
.len
= len
+ 1;
3077 disk_link
.name
= (char *) symname
;
3079 encryption_required
= (ext4_encrypted_inode(dir
) ||
3080 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir
->i_sb
)));
3081 if (encryption_required
) {
3082 err
= ext4_get_encryption_info(dir
);
3085 if (ext4_encryption_info(dir
) == NULL
)
3087 disk_link
.len
= (ext4_fname_encrypted_size(dir
, len
) +
3088 sizeof(struct ext4_encrypted_symlink_data
));
3089 sd
= kzalloc(disk_link
.len
, GFP_KERNEL
);
3094 if (disk_link
.len
> dir
->i_sb
->s_blocksize
) {
3095 err
= -ENAMETOOLONG
;
3099 err
= dquot_initialize(dir
);
3103 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3105 * For non-fast symlinks, we just allocate inode and put it on
3106 * orphan list in the first transaction => we need bitmap,
3107 * group descriptor, sb, inode block, quota blocks, and
3108 * possibly selinux xattr blocks.
3110 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
3111 EXT4_XATTR_TRANS_BLOCKS
;
3114 * Fast symlink. We have to add entry to directory
3115 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3116 * allocate new inode (bitmap, group descriptor, inode block,
3117 * quota blocks, sb is already counted in previous macros).
3119 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3120 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
3123 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
3124 &dentry
->d_name
, 0, NULL
,
3125 EXT4_HT_DIR
, credits
);
3126 handle
= ext4_journal_current_handle();
3127 if (IS_ERR(inode
)) {
3129 ext4_journal_stop(handle
);
3130 err
= PTR_ERR(inode
);
3134 if (encryption_required
) {
3136 struct ext4_str ostr
;
3138 istr
.name
= (const unsigned char *) symname
;
3140 ostr
.name
= sd
->encrypted_path
;
3141 ostr
.len
= disk_link
.len
;
3142 err
= ext4_fname_usr_to_disk(inode
, &istr
, &ostr
);
3144 goto err_drop_inode
;
3145 sd
->len
= cpu_to_le16(ostr
.len
);
3146 disk_link
.name
= (char *) sd
;
3147 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
3150 if ((disk_link
.len
> EXT4_N_BLOCKS
* 4)) {
3151 if (!encryption_required
)
3152 inode
->i_op
= &ext4_symlink_inode_operations
;
3153 inode_nohighmem(inode
);
3154 ext4_set_aops(inode
);
3156 * We cannot call page_symlink() with transaction started
3157 * because it calls into ext4_write_begin() which can wait
3158 * for transaction commit if we are running out of space
3159 * and thus we deadlock. So we have to stop transaction now
3160 * and restart it when symlink contents is written.
3162 * To keep fs consistent in case of crash, we have to put inode
3163 * to orphan list in the mean time.
3166 err
= ext4_orphan_add(handle
, inode
);
3167 ext4_journal_stop(handle
);
3170 goto err_drop_inode
;
3171 err
= __page_symlink(inode
, disk_link
.name
, disk_link
.len
, 1);
3173 goto err_drop_inode
;
3175 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3176 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3178 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3179 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3180 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
3181 if (IS_ERR(handle
)) {
3182 err
= PTR_ERR(handle
);
3184 goto err_drop_inode
;
3186 set_nlink(inode
, 1);
3187 err
= ext4_orphan_del(handle
, inode
);
3189 goto err_drop_inode
;
3191 /* clear the extent format for fast symlink */
3192 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
3193 if (!encryption_required
) {
3194 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3195 inode
->i_link
= (char *)&EXT4_I(inode
)->i_data
;
3197 memcpy((char *)&EXT4_I(inode
)->i_data
, disk_link
.name
,
3199 inode
->i_size
= disk_link
.len
- 1;
3201 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3202 err
= ext4_add_nondir(handle
, dentry
, inode
);
3203 if (!err
&& IS_DIRSYNC(dir
))
3204 ext4_handle_sync(handle
);
3207 ext4_journal_stop(handle
);
3212 ext4_journal_stop(handle
);
3214 unlock_new_inode(inode
);
3221 static int ext4_link(struct dentry
*old_dentry
,
3222 struct inode
*dir
, struct dentry
*dentry
)
3225 struct inode
*inode
= d_inode(old_dentry
);
3226 int err
, retries
= 0;
3228 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
3230 if (ext4_encrypted_inode(dir
) &&
3231 !ext4_is_child_context_consistent_with_parent(dir
, inode
))
3233 err
= dquot_initialize(dir
);
3238 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3239 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3240 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
3242 return PTR_ERR(handle
);
3244 if (IS_DIRSYNC(dir
))
3245 ext4_handle_sync(handle
);
3247 inode
->i_ctime
= ext4_current_time(inode
);
3248 ext4_inc_count(handle
, inode
);
3251 err
= ext4_add_entry(handle
, dentry
, inode
);
3253 ext4_mark_inode_dirty(handle
, inode
);
3254 /* this can happen only for tmpfile being
3255 * linked the first time
3257 if (inode
->i_nlink
== 1)
3258 ext4_orphan_del(handle
, inode
);
3259 d_instantiate(dentry
, inode
);
3264 ext4_journal_stop(handle
);
3265 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3272 * Try to find buffer head where contains the parent block.
3273 * It should be the inode block if it is inlined or the 1st block
3274 * if it is a normal dir.
3276 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
3277 struct inode
*inode
,
3279 struct ext4_dir_entry_2
**parent_de
,
3282 struct buffer_head
*bh
;
3284 if (!ext4_has_inline_data(inode
)) {
3285 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
3287 *retval
= PTR_ERR(bh
);
3290 *parent_de
= ext4_next_entry(
3291 (struct ext4_dir_entry_2
*)bh
->b_data
,
3292 inode
->i_sb
->s_blocksize
);
3297 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3300 struct ext4_renament
{
3302 struct dentry
*dentry
;
3303 struct inode
*inode
;
3305 int dir_nlink_delta
;
3307 /* entry for "dentry" */
3308 struct buffer_head
*bh
;
3309 struct ext4_dir_entry_2
*de
;
3312 /* entry for ".." in inode if it's a directory */
3313 struct buffer_head
*dir_bh
;
3314 struct ext4_dir_entry_2
*parent_de
;
3318 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3322 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3323 &retval
, &ent
->parent_de
,
3327 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3328 return -EFSCORRUPTED
;
3329 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3330 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3333 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3338 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3339 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3340 if (!ent
->dir_inlined
) {
3341 if (is_dx(ent
->inode
)) {
3342 retval
= ext4_handle_dirty_dx_node(handle
,
3346 retval
= ext4_handle_dirty_dirent_node(handle
,
3351 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3354 ext4_std_error(ent
->dir
->i_sb
, retval
);
3360 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3361 unsigned ino
, unsigned file_type
)
3365 BUFFER_TRACE(ent
->bh
, "get write access");
3366 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3369 ent
->de
->inode
= cpu_to_le32(ino
);
3370 if (ext4_has_feature_filetype(ent
->dir
->i_sb
))
3371 ent
->de
->file_type
= file_type
;
3372 ent
->dir
->i_version
++;
3373 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3374 ext4_current_time(ent
->dir
);
3375 ext4_mark_inode_dirty(handle
, ent
->dir
);
3376 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3377 if (!ent
->inlined
) {
3378 retval
= ext4_handle_dirty_dirent_node(handle
,
3380 if (unlikely(retval
)) {
3381 ext4_std_error(ent
->dir
->i_sb
, retval
);
3391 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3392 const struct qstr
*d_name
)
3394 int retval
= -ENOENT
;
3395 struct buffer_head
*bh
;
3396 struct ext4_dir_entry_2
*de
;
3398 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3402 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3408 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3413 * ent->de could have moved from under us during htree split, so make
3414 * sure that we are deleting the right entry. We might also be pointing
3415 * to a stale entry in the unused part of ent->bh so just checking inum
3416 * and the name isn't enough.
3418 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3419 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3420 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3421 ent
->de
->name_len
) ||
3423 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3424 &ent
->dentry
->d_name
);
3426 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3427 if (retval
== -ENOENT
) {
3428 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3429 &ent
->dentry
->d_name
);
3434 ext4_warning_inode(ent
->dir
,
3435 "Deleting old file: nlink %d, error=%d",
3436 ent
->dir
->i_nlink
, retval
);
3440 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3442 if (ent
->dir_nlink_delta
) {
3443 if (ent
->dir_nlink_delta
== -1)
3444 ext4_dec_count(handle
, ent
->dir
);
3446 ext4_inc_count(handle
, ent
->dir
);
3447 ext4_mark_inode_dirty(handle
, ent
->dir
);
3451 static struct inode
*ext4_whiteout_for_rename(struct ext4_renament
*ent
,
3452 int credits
, handle_t
**h
)
3459 * for inode block, sb block, group summaries,
3462 credits
+= (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent
->dir
->i_sb
) +
3463 EXT4_XATTR_TRANS_BLOCKS
+ 4);
3465 wh
= ext4_new_inode_start_handle(ent
->dir
, S_IFCHR
| WHITEOUT_MODE
,
3466 &ent
->dentry
->d_name
, 0, NULL
,
3467 EXT4_HT_DIR
, credits
);
3469 handle
= ext4_journal_current_handle();
3472 ext4_journal_stop(handle
);
3473 if (PTR_ERR(wh
) == -ENOSPC
&&
3474 ext4_should_retry_alloc(ent
->dir
->i_sb
, &retries
))
3478 init_special_inode(wh
, wh
->i_mode
, WHITEOUT_DEV
);
3479 wh
->i_op
= &ext4_special_inode_operations
;
3485 * Anybody can rename anything with this: the permission checks are left to the
3486 * higher-level routines.
3488 * n.b. old_{dentry,inode) refers to the source dentry/inode
3489 * while new_{dentry,inode) refers to the destination dentry/inode
3490 * This comes from rename(const char *oldpath, const char *newpath)
3492 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3493 struct inode
*new_dir
, struct dentry
*new_dentry
,
3496 handle_t
*handle
= NULL
;
3497 struct ext4_renament old
= {
3499 .dentry
= old_dentry
,
3500 .inode
= d_inode(old_dentry
),
3502 struct ext4_renament
new = {
3504 .dentry
= new_dentry
,
3505 .inode
= d_inode(new_dentry
),
3509 struct inode
*whiteout
= NULL
;
3513 if ((ext4_encrypted_inode(old_dir
) &&
3514 !ext4_has_encryption_key(old_dir
)) ||
3515 (ext4_encrypted_inode(new_dir
) &&
3516 !ext4_has_encryption_key(new_dir
)))
3519 retval
= dquot_initialize(old
.dir
);
3522 retval
= dquot_initialize(new.dir
);
3526 /* Initialize quotas before so that eventual writes go
3527 * in separate transaction */
3529 retval
= dquot_initialize(new.inode
);
3534 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3536 return PTR_ERR(old
.bh
);
3538 * Check for inode number is _not_ due to possible IO errors.
3539 * We might rmdir the source, keep it as pwd of some process
3540 * and merrily kill the link to whatever was created under the
3541 * same name. Goodbye sticky bit ;-<
3544 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3547 if ((old
.dir
!= new.dir
) &&
3548 ext4_encrypted_inode(new.dir
) &&
3549 !ext4_is_child_context_consistent_with_parent(new.dir
,
3555 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3556 &new.de
, &new.inlined
);
3557 if (IS_ERR(new.bh
)) {
3558 retval
= PTR_ERR(new.bh
);
3568 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3569 ext4_alloc_da_blocks(old
.inode
);
3571 credits
= (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3572 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
3573 if (!(flags
& RENAME_WHITEOUT
)) {
3574 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
, credits
);
3575 if (IS_ERR(handle
)) {
3576 retval
= PTR_ERR(handle
);
3581 whiteout
= ext4_whiteout_for_rename(&old
, credits
, &handle
);
3582 if (IS_ERR(whiteout
)) {
3583 retval
= PTR_ERR(whiteout
);
3589 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3590 ext4_handle_sync(handle
);
3592 if (S_ISDIR(old
.inode
->i_mode
)) {
3594 retval
= -ENOTEMPTY
;
3595 if (!ext4_empty_dir(new.inode
))
3599 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3602 retval
= ext4_rename_dir_prepare(handle
, &old
);
3607 * If we're renaming a file within an inline_data dir and adding or
3608 * setting the new dirent causes a conversion from inline_data to
3609 * extents/blockmap, we need to force the dirent delete code to
3610 * re-read the directory, or else we end up trying to delete a dirent
3611 * from what is now the extent tree root (or a block map).
3613 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3614 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3616 old_file_type
= old
.de
->file_type
;
3619 * Do this before adding a new entry, so the old entry is sure
3620 * to be still pointing to the valid old entry.
3622 retval
= ext4_setent(handle
, &old
, whiteout
->i_ino
,
3626 ext4_mark_inode_dirty(handle
, whiteout
);
3629 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3633 retval
= ext4_setent(handle
, &new,
3634 old
.inode
->i_ino
, old_file_type
);
3639 force_reread
= !ext4_test_inode_flag(new.dir
,
3640 EXT4_INODE_INLINE_DATA
);
3643 * Like most other Unix systems, set the ctime for inodes on a
3646 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3647 ext4_mark_inode_dirty(handle
, old
.inode
);
3653 ext4_rename_delete(handle
, &old
, force_reread
);
3657 ext4_dec_count(handle
, new.inode
);
3658 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3660 old
.dir
->i_ctime
= old
.dir
->i_mtime
= ext4_current_time(old
.dir
);
3661 ext4_update_dx_flag(old
.dir
);
3663 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3667 ext4_dec_count(handle
, old
.dir
);
3669 /* checked ext4_empty_dir above, can't have another
3670 * parent, ext4_dec_count() won't work for many-linked
3672 clear_nlink(new.inode
);
3674 ext4_inc_count(handle
, new.dir
);
3675 ext4_update_dx_flag(new.dir
);
3676 ext4_mark_inode_dirty(handle
, new.dir
);
3679 ext4_mark_inode_dirty(handle
, old
.dir
);
3681 ext4_mark_inode_dirty(handle
, new.inode
);
3682 if (!new.inode
->i_nlink
)
3683 ext4_orphan_add(handle
, new.inode
);
3693 drop_nlink(whiteout
);
3694 unlock_new_inode(whiteout
);
3698 ext4_journal_stop(handle
);
3702 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3703 struct inode
*new_dir
, struct dentry
*new_dentry
)
3705 handle_t
*handle
= NULL
;
3706 struct ext4_renament old
= {
3708 .dentry
= old_dentry
,
3709 .inode
= d_inode(old_dentry
),
3711 struct ext4_renament
new = {
3713 .dentry
= new_dentry
,
3714 .inode
= d_inode(new_dentry
),
3719 if ((ext4_encrypted_inode(old_dir
) &&
3720 !ext4_has_encryption_key(old_dir
)) ||
3721 (ext4_encrypted_inode(new_dir
) &&
3722 !ext4_has_encryption_key(new_dir
)))
3725 if ((ext4_encrypted_inode(old_dir
) ||
3726 ext4_encrypted_inode(new_dir
)) &&
3727 (old_dir
!= new_dir
) &&
3728 (!ext4_is_child_context_consistent_with_parent(new_dir
,
3730 !ext4_is_child_context_consistent_with_parent(old_dir
,
3734 retval
= dquot_initialize(old
.dir
);
3737 retval
= dquot_initialize(new.dir
);
3741 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3742 &old
.de
, &old
.inlined
);
3744 return PTR_ERR(old
.bh
);
3746 * Check for inode number is _not_ due to possible IO errors.
3747 * We might rmdir the source, keep it as pwd of some process
3748 * and merrily kill the link to whatever was created under the
3749 * same name. Goodbye sticky bit ;-<
3752 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3755 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3756 &new.de
, &new.inlined
);
3757 if (IS_ERR(new.bh
)) {
3758 retval
= PTR_ERR(new.bh
);
3763 /* RENAME_EXCHANGE case: old *and* new must both exist */
3764 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3767 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3768 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3769 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3770 if (IS_ERR(handle
)) {
3771 retval
= PTR_ERR(handle
);
3776 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3777 ext4_handle_sync(handle
);
3779 if (S_ISDIR(old
.inode
->i_mode
)) {
3781 retval
= ext4_rename_dir_prepare(handle
, &old
);
3785 if (S_ISDIR(new.inode
->i_mode
)) {
3787 retval
= ext4_rename_dir_prepare(handle
, &new);
3793 * Other than the special case of overwriting a directory, parents'
3794 * nlink only needs to be modified if this is a cross directory rename.
3796 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3797 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3798 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3800 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3801 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3805 new_file_type
= new.de
->file_type
;
3806 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3810 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3815 * Like most other Unix systems, set the ctime for inodes on a
3818 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3819 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3820 ext4_mark_inode_dirty(handle
, old
.inode
);
3821 ext4_mark_inode_dirty(handle
, new.inode
);
3824 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3829 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3833 ext4_update_dir_count(handle
, &old
);
3834 ext4_update_dir_count(handle
, &new);
3843 ext4_journal_stop(handle
);
3847 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3848 struct inode
*new_dir
, struct dentry
*new_dentry
,
3851 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3854 if (flags
& RENAME_EXCHANGE
) {
3855 return ext4_cross_rename(old_dir
, old_dentry
,
3856 new_dir
, new_dentry
);
3859 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
3863 * directories can handle most operations...
3865 const struct inode_operations ext4_dir_inode_operations
= {
3866 .create
= ext4_create
,
3867 .lookup
= ext4_lookup
,
3869 .unlink
= ext4_unlink
,
3870 .symlink
= ext4_symlink
,
3871 .mkdir
= ext4_mkdir
,
3872 .rmdir
= ext4_rmdir
,
3873 .mknod
= ext4_mknod
,
3874 .tmpfile
= ext4_tmpfile
,
3875 .rename2
= ext4_rename2
,
3876 .setattr
= ext4_setattr
,
3877 .setxattr
= generic_setxattr
,
3878 .getxattr
= generic_getxattr
,
3879 .listxattr
= ext4_listxattr
,
3880 .removexattr
= generic_removexattr
,
3881 .get_acl
= ext4_get_acl
,
3882 .set_acl
= ext4_set_acl
,
3883 .fiemap
= ext4_fiemap
,
3886 const struct inode_operations ext4_special_inode_operations
= {
3887 .setattr
= ext4_setattr
,
3888 .setxattr
= generic_setxattr
,
3889 .getxattr
= generic_getxattr
,
3890 .listxattr
= ext4_listxattr
,
3891 .removexattr
= generic_removexattr
,
3892 .get_acl
= ext4_get_acl
,
3893 .set_acl
= ext4_set_acl
,