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/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
38 #include "ext4_jbd2.h"
43 #include <trace/events/ext4.h>
45 * define how far ahead to read directories while searching them.
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51 static struct buffer_head
*ext4_append(handle_t
*handle
,
55 struct buffer_head
*bh
;
58 if (unlikely(EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
&&
59 ((inode
->i_size
>> 10) >=
60 EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
)))
61 return ERR_PTR(-ENOSPC
);
63 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
65 bh
= ext4_bread(handle
, inode
, *block
, 1);
68 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
69 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
70 BUFFER_TRACE(bh
, "get_write_access");
71 err
= ext4_journal_get_write_access(handle
, bh
);
74 ext4_std_error(inode
->i_sb
, err
);
80 static int ext4_dx_csum_verify(struct inode
*inode
,
81 struct ext4_dir_entry
*dirent
);
87 #define ext4_read_dirblock(inode, block, type) \
88 __ext4_read_dirblock((inode), (block), (type), __LINE__)
90 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 "error %ld reading directory block "
103 "(ino %lu, block %lu)", PTR_ERR(bh
), inode
->i_ino
,
104 (unsigned long) block
);
109 ext4_error_inode(inode
, __func__
, line
, block
, "Directory hole found");
110 return ERR_PTR(-EIO
);
112 dirent
= (struct ext4_dir_entry
*) bh
->b_data
;
113 /* Determine whether or not we have an index block */
117 else if (ext4_rec_len_from_disk(dirent
->rec_len
,
118 inode
->i_sb
->s_blocksize
) ==
119 inode
->i_sb
->s_blocksize
)
122 if (!is_dx_block
&& type
== INDEX
) {
123 ext4_error_inode(inode
, __func__
, line
, block
,
124 "directory leaf block found instead of index block");
125 return ERR_PTR(-EIO
);
127 if (!ext4_has_metadata_csum(inode
->i_sb
) ||
132 * An empty leaf block can get mistaken for a index block; for
133 * this reason, we can only check the index checksum when the
134 * caller is sure it should be an index block.
136 if (is_dx_block
&& type
== INDEX
) {
137 if (ext4_dx_csum_verify(inode
, dirent
))
138 set_buffer_verified(bh
);
140 ext4_error_inode(inode
, __func__
, line
, block
,
141 "Directory index failed checksum");
143 return ERR_PTR(-EIO
);
147 if (ext4_dirent_csum_verify(inode
, dirent
))
148 set_buffer_verified(bh
);
150 ext4_error_inode(inode
, __func__
, line
, block
,
151 "Directory block failed checksum");
153 return ERR_PTR(-EIO
);
160 #define assert(test) J_ASSERT(test)
164 #define dxtrace(command) command
166 #define dxtrace(command)
190 * dx_root_info is laid out so that if it should somehow get overlaid by a
191 * dirent the two low bits of the hash version will be zero. Therefore, the
192 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
197 struct fake_dirent dot
;
199 struct fake_dirent dotdot
;
203 __le32 reserved_zero
;
205 u8 info_length
; /* 8 */
210 struct dx_entry entries
[0];
215 struct fake_dirent fake
;
216 struct dx_entry entries
[0];
222 struct buffer_head
*bh
;
223 struct dx_entry
*entries
;
235 * This goes at the end of each htree block.
239 __le32 dt_checksum
; /* crc32c(uuid+inum+dirblock) */
242 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
243 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
244 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
245 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
246 static unsigned dx_get_count(struct dx_entry
*entries
);
247 static unsigned dx_get_limit(struct dx_entry
*entries
);
248 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
249 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
250 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
251 static unsigned dx_node_limit(struct inode
*dir
);
252 static struct dx_frame
*dx_probe(const struct qstr
*d_name
,
254 struct dx_hash_info
*hinfo
,
255 struct dx_frame
*frame
);
256 static void dx_release(struct dx_frame
*frames
);
257 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
258 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
259 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
260 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
261 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
262 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
263 static void dx_insert_block(struct dx_frame
*frame
,
264 u32 hash
, ext4_lblk_t block
);
265 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
266 struct dx_frame
*frame
,
267 struct dx_frame
*frames
,
269 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
270 const struct qstr
*d_name
,
271 struct ext4_dir_entry_2
**res_dir
);
272 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
273 struct inode
*inode
);
275 /* checksumming functions */
276 void initialize_dirent_tail(struct ext4_dir_entry_tail
*t
,
277 unsigned int blocksize
)
279 memset(t
, 0, sizeof(struct ext4_dir_entry_tail
));
280 t
->det_rec_len
= ext4_rec_len_to_disk(
281 sizeof(struct ext4_dir_entry_tail
), blocksize
);
282 t
->det_reserved_ft
= EXT4_FT_DIR_CSUM
;
285 /* Walk through a dirent block to find a checksum "dirent" at the tail */
286 static struct ext4_dir_entry_tail
*get_dirent_tail(struct inode
*inode
,
287 struct ext4_dir_entry
*de
)
289 struct ext4_dir_entry_tail
*t
;
292 struct ext4_dir_entry
*d
, *top
;
295 top
= (struct ext4_dir_entry
*)(((void *)de
) +
296 (EXT4_BLOCK_SIZE(inode
->i_sb
) -
297 sizeof(struct ext4_dir_entry_tail
)));
298 while (d
< top
&& d
->rec_len
)
299 d
= (struct ext4_dir_entry
*)(((void *)d
) +
300 le16_to_cpu(d
->rec_len
));
305 t
= (struct ext4_dir_entry_tail
*)d
;
307 t
= EXT4_DIRENT_TAIL(de
, EXT4_BLOCK_SIZE(inode
->i_sb
));
310 if (t
->det_reserved_zero1
||
311 le16_to_cpu(t
->det_rec_len
) != sizeof(struct ext4_dir_entry_tail
) ||
312 t
->det_reserved_zero2
||
313 t
->det_reserved_ft
!= EXT4_FT_DIR_CSUM
)
319 static __le32
ext4_dirent_csum(struct inode
*inode
,
320 struct ext4_dir_entry
*dirent
, int size
)
322 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
323 struct ext4_inode_info
*ei
= EXT4_I(inode
);
326 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
327 return cpu_to_le32(csum
);
330 static void warn_no_space_for_csum(struct inode
*inode
)
332 ext4_warning(inode
->i_sb
, "no space in directory inode %lu leaf for "
333 "checksum. Please run e2fsck -D.", inode
->i_ino
);
336 int ext4_dirent_csum_verify(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
338 struct ext4_dir_entry_tail
*t
;
340 if (!ext4_has_metadata_csum(inode
->i_sb
))
343 t
= get_dirent_tail(inode
, dirent
);
345 warn_no_space_for_csum(inode
);
349 if (t
->det_checksum
!= ext4_dirent_csum(inode
, dirent
,
350 (void *)t
- (void *)dirent
))
356 static void ext4_dirent_csum_set(struct inode
*inode
,
357 struct ext4_dir_entry
*dirent
)
359 struct ext4_dir_entry_tail
*t
;
361 if (!ext4_has_metadata_csum(inode
->i_sb
))
364 t
= get_dirent_tail(inode
, dirent
);
366 warn_no_space_for_csum(inode
);
370 t
->det_checksum
= ext4_dirent_csum(inode
, dirent
,
371 (void *)t
- (void *)dirent
);
374 int ext4_handle_dirty_dirent_node(handle_t
*handle
,
376 struct buffer_head
*bh
)
378 ext4_dirent_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
379 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
382 static struct dx_countlimit
*get_dx_countlimit(struct inode
*inode
,
383 struct ext4_dir_entry
*dirent
,
386 struct ext4_dir_entry
*dp
;
387 struct dx_root_info
*root
;
390 if (le16_to_cpu(dirent
->rec_len
) == EXT4_BLOCK_SIZE(inode
->i_sb
))
392 else if (le16_to_cpu(dirent
->rec_len
) == 12) {
393 dp
= (struct ext4_dir_entry
*)(((void *)dirent
) + 12);
394 if (le16_to_cpu(dp
->rec_len
) !=
395 EXT4_BLOCK_SIZE(inode
->i_sb
) - 12)
397 root
= (struct dx_root_info
*)(((void *)dp
+ 12));
398 if (root
->reserved_zero
||
399 root
->info_length
!= sizeof(struct dx_root_info
))
406 *offset
= count_offset
;
407 return (struct dx_countlimit
*)(((void *)dirent
) + count_offset
);
410 static __le32
ext4_dx_csum(struct inode
*inode
, struct ext4_dir_entry
*dirent
,
411 int count_offset
, int count
, struct dx_tail
*t
)
413 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
414 struct ext4_inode_info
*ei
= EXT4_I(inode
);
419 size
= count_offset
+ (count
* sizeof(struct dx_entry
));
420 save_csum
= t
->dt_checksum
;
422 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
423 csum
= ext4_chksum(sbi
, csum
, (__u8
*)t
, sizeof(struct dx_tail
));
424 t
->dt_checksum
= save_csum
;
426 return cpu_to_le32(csum
);
429 static int ext4_dx_csum_verify(struct inode
*inode
,
430 struct ext4_dir_entry
*dirent
)
432 struct dx_countlimit
*c
;
434 int count_offset
, limit
, count
;
436 if (!ext4_has_metadata_csum(inode
->i_sb
))
439 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
441 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
444 limit
= le16_to_cpu(c
->limit
);
445 count
= le16_to_cpu(c
->count
);
446 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
447 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
448 warn_no_space_for_csum(inode
);
451 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
453 if (t
->dt_checksum
!= ext4_dx_csum(inode
, dirent
, count_offset
,
459 static void ext4_dx_csum_set(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
461 struct dx_countlimit
*c
;
463 int count_offset
, limit
, count
;
465 if (!ext4_has_metadata_csum(inode
->i_sb
))
468 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
470 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
473 limit
= le16_to_cpu(c
->limit
);
474 count
= le16_to_cpu(c
->count
);
475 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
476 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
477 warn_no_space_for_csum(inode
);
480 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
482 t
->dt_checksum
= ext4_dx_csum(inode
, dirent
, count_offset
, count
, t
);
485 static inline int ext4_handle_dirty_dx_node(handle_t
*handle
,
487 struct buffer_head
*bh
)
489 ext4_dx_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
490 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
494 * p is at least 6 bytes before the end of page
496 static inline struct ext4_dir_entry_2
*
497 ext4_next_entry(struct ext4_dir_entry_2
*p
, unsigned long blocksize
)
499 return (struct ext4_dir_entry_2
*)((char *)p
+
500 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
504 * Future: use high four bits of block for coalesce-on-delete flags
505 * Mask them off for now.
508 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
510 return le32_to_cpu(entry
->block
) & 0x00ffffff;
513 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
515 entry
->block
= cpu_to_le32(value
);
518 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
520 return le32_to_cpu(entry
->hash
);
523 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
525 entry
->hash
= cpu_to_le32(value
);
528 static inline unsigned dx_get_count(struct dx_entry
*entries
)
530 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
533 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
535 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
538 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
540 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
543 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
545 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
548 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
550 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
551 EXT4_DIR_REC_LEN(2) - infosize
;
553 if (ext4_has_metadata_csum(dir
->i_sb
))
554 entry_space
-= sizeof(struct dx_tail
);
555 return entry_space
/ sizeof(struct dx_entry
);
558 static inline unsigned dx_node_limit(struct inode
*dir
)
560 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
562 if (ext4_has_metadata_csum(dir
->i_sb
))
563 entry_space
-= sizeof(struct dx_tail
);
564 return entry_space
/ sizeof(struct dx_entry
);
571 static void dx_show_index(char * label
, struct dx_entry
*entries
)
573 int i
, n
= dx_get_count (entries
);
574 printk(KERN_DEBUG
"%s index ", label
);
575 for (i
= 0; i
< n
; i
++) {
576 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
577 0, (unsigned long)dx_get_block(entries
+ i
));
589 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
590 int size
, int show_names
)
592 unsigned names
= 0, space
= 0;
593 char *base
= (char *) de
;
594 struct dx_hash_info h
= *hinfo
;
597 while ((char *) de
< base
+ size
)
603 int len
= de
->name_len
;
604 char *name
= de
->name
;
605 while (len
--) printk("%c", *name
++);
606 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
607 printk(":%x.%u ", h
.hash
,
608 (unsigned) ((char *) de
- base
));
610 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
613 de
= ext4_next_entry(de
, size
);
615 printk("(%i)\n", names
);
616 return (struct stats
) { names
, space
, 1 };
619 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
620 struct dx_entry
*entries
, int levels
)
622 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
623 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
625 struct buffer_head
*bh
;
627 printk("%i indexed blocks...\n", count
);
628 for (i
= 0; i
< count
; i
++, entries
++)
630 ext4_lblk_t block
= dx_get_block(entries
);
631 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
632 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
634 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
635 bh
= ext4_bread(NULL
,dir
, block
, 0);
636 if (!bh
|| IS_ERR(bh
))
639 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
640 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
641 names
+= stats
.names
;
642 space
+= stats
.space
;
643 bcount
+= stats
.bcount
;
647 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
648 levels
? "" : " ", names
, space
/bcount
,
649 (space
/bcount
)*100/blocksize
);
650 return (struct stats
) { names
, space
, bcount
};
652 #endif /* DX_DEBUG */
655 * Probe for a directory leaf block to search.
657 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
658 * error in the directory index, and the caller should fall back to
659 * searching the directory normally. The callers of dx_probe **MUST**
660 * check for this error code, and make sure it never gets reflected
663 static struct dx_frame
*
664 dx_probe(const struct qstr
*d_name
, struct inode
*dir
,
665 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
)
667 unsigned count
, indirect
;
668 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
669 struct dx_root
*root
;
670 struct dx_frame
*frame
= frame_in
;
671 struct dx_frame
*ret_err
= ERR_PTR(ERR_BAD_DX_DIR
);
674 frame
->bh
= ext4_read_dirblock(dir
, 0, INDEX
);
675 if (IS_ERR(frame
->bh
))
676 return (struct dx_frame
*) frame
->bh
;
678 root
= (struct dx_root
*) frame
->bh
->b_data
;
679 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
680 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
681 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
682 ext4_warning(dir
->i_sb
, "Unrecognised inode hash code %d",
683 root
->info
.hash_version
);
686 hinfo
->hash_version
= root
->info
.hash_version
;
687 if (hinfo
->hash_version
<= DX_HASH_TEA
)
688 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
689 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
691 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
694 if (root
->info
.unused_flags
& 1) {
695 ext4_warning(dir
->i_sb
, "Unimplemented inode hash flags: %#06x",
696 root
->info
.unused_flags
);
700 if ((indirect
= root
->info
.indirect_levels
) > 1) {
701 ext4_warning(dir
->i_sb
, "Unimplemented inode hash depth: %#06x",
702 root
->info
.indirect_levels
);
706 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
707 root
->info
.info_length
);
709 if (dx_get_limit(entries
) != dx_root_limit(dir
,
710 root
->info
.info_length
)) {
711 ext4_warning(dir
->i_sb
, "dx entry: limit != root limit");
715 dxtrace(printk("Look up %x", hash
));
717 count
= dx_get_count(entries
);
718 if (!count
|| count
> dx_get_limit(entries
)) {
719 ext4_warning(dir
->i_sb
,
720 "dx entry: no count or count > limit");
725 q
= entries
+ count
- 1;
728 dxtrace(printk("."));
729 if (dx_get_hash(m
) > hash
)
735 if (0) { // linear search cross check
736 unsigned n
= count
- 1;
740 dxtrace(printk(","));
741 if (dx_get_hash(++at
) > hash
)
747 assert (at
== p
- 1);
751 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
752 frame
->entries
= entries
;
757 frame
->bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
758 if (IS_ERR(frame
->bh
)) {
759 ret_err
= (struct dx_frame
*) frame
->bh
;
763 entries
= ((struct dx_node
*) frame
->bh
->b_data
)->entries
;
765 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
766 ext4_warning(dir
->i_sb
,
767 "dx entry: limit != node limit");
772 while (frame
>= frame_in
) {
776 if (ret_err
== ERR_PTR(ERR_BAD_DX_DIR
))
777 ext4_warning(dir
->i_sb
,
778 "Corrupt dir inode %lu, running e2fsck is "
779 "recommended.", dir
->i_ino
);
783 static void dx_release (struct dx_frame
*frames
)
785 if (frames
[0].bh
== NULL
)
788 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
789 brelse(frames
[1].bh
);
790 brelse(frames
[0].bh
);
794 * This function increments the frame pointer to search the next leaf
795 * block, and reads in the necessary intervening nodes if the search
796 * should be necessary. Whether or not the search is necessary is
797 * controlled by the hash parameter. If the hash value is even, then
798 * the search is only continued if the next block starts with that
799 * hash value. This is used if we are searching for a specific file.
801 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
803 * This function returns 1 if the caller should continue to search,
804 * or 0 if it should not. If there is an error reading one of the
805 * index blocks, it will a negative error code.
807 * If start_hash is non-null, it will be filled in with the starting
808 * hash of the next page.
810 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
811 struct dx_frame
*frame
,
812 struct dx_frame
*frames
,
816 struct buffer_head
*bh
;
822 * Find the next leaf page by incrementing the frame pointer.
823 * If we run out of entries in the interior node, loop around and
824 * increment pointer in the parent node. When we break out of
825 * this loop, num_frames indicates the number of interior
826 * nodes need to be read.
829 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
838 * If the hash is 1, then continue only if the next page has a
839 * continuation hash of any value. This is used for readdir
840 * handling. Otherwise, check to see if the hash matches the
841 * desired contiuation hash. If it doesn't, return since
842 * there's no point to read in the successive index pages.
844 bhash
= dx_get_hash(p
->at
);
847 if ((hash
& 1) == 0) {
848 if ((bhash
& ~1) != hash
)
852 * If the hash is HASH_NB_ALWAYS, we always go to the next
853 * block so no check is necessary
855 while (num_frames
--) {
856 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
862 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
869 * This function fills a red-black tree with information from a
870 * directory block. It returns the number directory entries loaded
871 * into the tree. If there is an error it is returned in err.
873 static int htree_dirblock_to_tree(struct file
*dir_file
,
874 struct inode
*dir
, ext4_lblk_t block
,
875 struct dx_hash_info
*hinfo
,
876 __u32 start_hash
, __u32 start_minor_hash
)
878 struct buffer_head
*bh
;
879 struct ext4_dir_entry_2
*de
, *top
;
880 int err
= 0, count
= 0;
882 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
883 (unsigned long)block
));
884 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
888 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
889 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
890 dir
->i_sb
->s_blocksize
-
891 EXT4_DIR_REC_LEN(0));
892 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
893 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
894 bh
->b_data
, bh
->b_size
,
895 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
896 + ((char *)de
- bh
->b_data
))) {
897 /* silently ignore the rest of the block */
900 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
901 if ((hinfo
->hash
< start_hash
) ||
902 ((hinfo
->hash
== start_hash
) &&
903 (hinfo
->minor_hash
< start_minor_hash
)))
907 if ((err
= ext4_htree_store_dirent(dir_file
,
908 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
920 * This function fills a red-black tree with information from a
921 * directory. We start scanning the directory in hash order, starting
922 * at start_hash and start_minor_hash.
924 * This function returns the number of entries inserted into the tree,
925 * or a negative error code.
927 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
928 __u32 start_minor_hash
, __u32
*next_hash
)
930 struct dx_hash_info hinfo
;
931 struct ext4_dir_entry_2
*de
;
932 struct dx_frame frames
[2], *frame
;
939 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
940 start_hash
, start_minor_hash
));
941 dir
= file_inode(dir_file
);
942 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
943 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
944 if (hinfo
.hash_version
<= DX_HASH_TEA
)
945 hinfo
.hash_version
+=
946 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
947 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
948 if (ext4_has_inline_data(dir
)) {
949 int has_inline_data
= 1;
950 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
954 if (has_inline_data
) {
959 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
960 start_hash
, start_minor_hash
);
964 hinfo
.hash
= start_hash
;
965 hinfo
.minor_hash
= 0;
966 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
);
968 return PTR_ERR(frame
);
970 /* Add '.' and '..' from the htree header */
971 if (!start_hash
&& !start_minor_hash
) {
972 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
973 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
977 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
978 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
979 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
980 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
986 block
= dx_get_block(frame
->at
);
987 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
988 start_hash
, start_minor_hash
);
995 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
996 frame
, frames
, &hashval
);
997 *next_hash
= hashval
;
1003 * Stop if: (a) there are no more entries, or
1004 * (b) we have inserted at least one entry and the
1005 * next hash value is not a continuation
1008 (count
&& ((hashval
& 1) == 0)))
1012 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1013 "next hash: %x\n", count
, *next_hash
));
1020 static inline int search_dirblock(struct buffer_head
*bh
,
1022 const struct qstr
*d_name
,
1023 unsigned int offset
,
1024 struct ext4_dir_entry_2
**res_dir
)
1026 return search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1027 d_name
, offset
, res_dir
);
1031 * Directory block splitting, compacting
1035 * Create map of hash values, offsets, and sizes, stored at end of block.
1036 * Returns number of entries mapped.
1038 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
1039 struct dx_hash_info
*hinfo
,
1040 struct dx_map_entry
*map_tail
)
1043 char *base
= (char *) de
;
1044 struct dx_hash_info h
= *hinfo
;
1046 while ((char *) de
< base
+ blocksize
) {
1047 if (de
->name_len
&& de
->inode
) {
1048 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1050 map_tail
->hash
= h
.hash
;
1051 map_tail
->offs
= ((char *) de
- base
)>>2;
1052 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1056 /* XXX: do we need to check rec_len == 0 case? -Chris */
1057 de
= ext4_next_entry(de
, blocksize
);
1062 /* Sort map by hash value */
1063 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1065 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1067 /* Combsort until bubble sort doesn't suck */
1069 count
= count
*10/13;
1070 if (count
- 9 < 2) /* 9, 10 -> 11 */
1072 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1073 if (p
->hash
< q
->hash
)
1076 /* Garden variety bubble sort */
1081 if (q
[1].hash
>= q
[0].hash
)
1089 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1091 struct dx_entry
*entries
= frame
->entries
;
1092 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1093 int count
= dx_get_count(entries
);
1095 assert(count
< dx_get_limit(entries
));
1096 assert(old
< entries
+ count
);
1097 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1098 dx_set_hash(new, hash
);
1099 dx_set_block(new, block
);
1100 dx_set_count(entries
, count
+ 1);
1104 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1106 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1107 * `de != NULL' is guaranteed by caller.
1109 static inline int ext4_match (int len
, const char * const name
,
1110 struct ext4_dir_entry_2
* de
)
1112 if (len
!= de
->name_len
)
1116 return !memcmp(name
, de
->name
, len
);
1120 * Returns 0 if not found, -1 on failure, and 1 on success
1122 int search_dir(struct buffer_head
*bh
,
1126 const struct qstr
*d_name
,
1127 unsigned int offset
,
1128 struct ext4_dir_entry_2
**res_dir
)
1130 struct ext4_dir_entry_2
* de
;
1133 const char *name
= d_name
->name
;
1134 int namelen
= d_name
->len
;
1136 de
= (struct ext4_dir_entry_2
*)search_buf
;
1137 dlimit
= search_buf
+ buf_size
;
1138 while ((char *) de
< dlimit
) {
1139 /* this code is executed quadratically often */
1140 /* do minimal checking `by hand' */
1142 if ((char *) de
+ namelen
<= dlimit
&&
1143 ext4_match (namelen
, name
, de
)) {
1144 /* found a match - just to be sure, do a full check */
1145 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, bh
->b_data
,
1146 bh
->b_size
, offset
))
1151 /* prevent looping on a bad block */
1152 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1153 dir
->i_sb
->s_blocksize
);
1157 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1162 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1163 struct ext4_dir_entry
*de
)
1165 struct super_block
*sb
= dir
->i_sb
;
1171 if (de
->inode
== 0 &&
1172 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1181 * finds an entry in the specified directory with the wanted name. It
1182 * returns the cache buffer in which the entry was found, and the entry
1183 * itself (as a parameter - res_dir). It does NOT read the inode of the
1184 * entry - you'll have to do that yourself if you want to.
1186 * The returned buffer_head has ->b_count elevated. The caller is expected
1187 * to brelse() it when appropriate.
1189 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1190 const struct qstr
*d_name
,
1191 struct ext4_dir_entry_2
**res_dir
,
1194 struct super_block
*sb
;
1195 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1196 struct buffer_head
*bh
, *ret
= NULL
;
1197 ext4_lblk_t start
, block
, b
;
1198 const u8
*name
= d_name
->name
;
1199 int ra_max
= 0; /* Number of bh's in the readahead
1201 int ra_ptr
= 0; /* Current index into readahead
1204 ext4_lblk_t nblocks
;
1209 namelen
= d_name
->len
;
1210 if (namelen
> EXT4_NAME_LEN
)
1213 if (ext4_has_inline_data(dir
)) {
1214 int has_inline_data
= 1;
1215 ret
= ext4_find_inline_entry(dir
, d_name
, res_dir
,
1217 if (has_inline_data
) {
1224 if ((namelen
<= 2) && (name
[0] == '.') &&
1225 (name
[1] == '.' || name
[1] == '\0')) {
1227 * "." or ".." will only be in the first block
1228 * NFS may look up ".."; "." should be handled by the VFS
1235 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
);
1237 * On success, or if the error was file not found,
1238 * return. Otherwise, fall back to doing a search the
1239 * old fashioned way.
1241 if (!IS_ERR(bh
) || PTR_ERR(bh
) != ERR_BAD_DX_DIR
)
1243 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1246 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1247 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1248 if (start
>= nblocks
)
1254 * We deal with the read-ahead logic here.
1256 if (ra_ptr
>= ra_max
) {
1257 /* Refill the readahead buffer */
1260 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1262 * Terminate if we reach the end of the
1263 * directory and must wrap, or if our
1264 * search has finished at this block.
1266 if (b
>= nblocks
|| (num
&& block
== start
)) {
1267 bh_use
[ra_max
] = NULL
;
1271 bh
= ext4_getblk(NULL
, dir
, b
++, 0);
1272 if (unlikely(IS_ERR(bh
))) {
1277 bh_use
[ra_max
] = bh
;
1279 ll_rw_block(READ
| REQ_META
| REQ_PRIO
,
1283 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1286 if (!buffer_uptodate(bh
)) {
1287 /* read error, skip block & hope for the best */
1288 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1289 (unsigned long) block
);
1293 if (!buffer_verified(bh
) &&
1294 !is_dx_internal_node(dir
, block
,
1295 (struct ext4_dir_entry
*)bh
->b_data
) &&
1296 !ext4_dirent_csum_verify(dir
,
1297 (struct ext4_dir_entry
*)bh
->b_data
)) {
1298 EXT4_ERROR_INODE(dir
, "checksumming directory "
1299 "block %lu", (unsigned long)block
);
1303 set_buffer_verified(bh
);
1304 i
= search_dirblock(bh
, dir
, d_name
,
1305 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1307 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1309 goto cleanup_and_exit
;
1313 goto cleanup_and_exit
;
1316 if (++block
>= nblocks
)
1318 } while (block
!= start
);
1321 * If the directory has grown while we were searching, then
1322 * search the last part of the directory before giving up.
1325 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1326 if (block
< nblocks
) {
1332 /* Clean up the read-ahead blocks */
1333 for (; ra_ptr
< ra_max
; ra_ptr
++)
1334 brelse(bh_use
[ra_ptr
]);
1338 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
1339 struct ext4_dir_entry_2
**res_dir
)
1341 struct super_block
* sb
= dir
->i_sb
;
1342 struct dx_hash_info hinfo
;
1343 struct dx_frame frames
[2], *frame
;
1344 struct buffer_head
*bh
;
1348 frame
= dx_probe(d_name
, dir
, &hinfo
, frames
);
1350 return (struct buffer_head
*) frame
;
1352 block
= dx_get_block(frame
->at
);
1353 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1357 retval
= search_dirblock(bh
, dir
, d_name
,
1358 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1364 bh
= ERR_PTR(ERR_BAD_DX_DIR
);
1368 /* Check to see if we should continue to search */
1369 retval
= ext4_htree_next_block(dir
, hinfo
.hash
, frame
,
1373 "error %d reading index page in directory #%lu",
1374 retval
, dir
->i_ino
);
1375 bh
= ERR_PTR(retval
);
1378 } while (retval
== 1);
1382 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1388 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1390 struct inode
*inode
;
1391 struct ext4_dir_entry_2
*de
;
1392 struct buffer_head
*bh
;
1394 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1395 return ERR_PTR(-ENAMETOOLONG
);
1397 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1399 return (struct dentry
*) bh
;
1402 __u32 ino
= le32_to_cpu(de
->inode
);
1404 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1405 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1406 return ERR_PTR(-EIO
);
1408 if (unlikely(ino
== dir
->i_ino
)) {
1409 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1411 return ERR_PTR(-EIO
);
1413 inode
= ext4_iget_normal(dir
->i_sb
, ino
);
1414 if (inode
== ERR_PTR(-ESTALE
)) {
1415 EXT4_ERROR_INODE(dir
,
1416 "deleted inode referenced: %u",
1418 return ERR_PTR(-EIO
);
1421 return d_splice_alias(inode
, dentry
);
1425 struct dentry
*ext4_get_parent(struct dentry
*child
)
1428 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1429 struct ext4_dir_entry_2
* de
;
1430 struct buffer_head
*bh
;
1432 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
, NULL
);
1434 return (struct dentry
*) bh
;
1436 return ERR_PTR(-ENOENT
);
1437 ino
= le32_to_cpu(de
->inode
);
1440 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1441 EXT4_ERROR_INODE(child
->d_inode
,
1442 "bad parent inode number: %u", ino
);
1443 return ERR_PTR(-EIO
);
1446 return d_obtain_alias(ext4_iget_normal(child
->d_inode
->i_sb
, ino
));
1450 * Move count entries from end of map between two memory locations.
1451 * Returns pointer to last entry moved.
1453 static struct ext4_dir_entry_2
*
1454 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1457 unsigned rec_len
= 0;
1460 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1461 (from
+ (map
->offs
<<2));
1462 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1463 memcpy (to
, de
, rec_len
);
1464 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1465 ext4_rec_len_to_disk(rec_len
, blocksize
);
1470 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1474 * Compact each dir entry in the range to the minimal rec_len.
1475 * Returns pointer to last entry in range.
1477 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1479 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1480 unsigned rec_len
= 0;
1483 while ((char*)de
< base
+ blocksize
) {
1484 next
= ext4_next_entry(de
, blocksize
);
1485 if (de
->inode
&& de
->name_len
) {
1486 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1488 memmove(to
, de
, rec_len
);
1489 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1491 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1499 * Split a full leaf block to make room for a new dir entry.
1500 * Allocate a new block, and move entries so that they are approx. equally full.
1501 * Returns pointer to de in block into which the new entry will be inserted.
1503 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1504 struct buffer_head
**bh
,struct dx_frame
*frame
,
1505 struct dx_hash_info
*hinfo
)
1507 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1508 unsigned count
, continued
;
1509 struct buffer_head
*bh2
;
1510 ext4_lblk_t newblock
;
1512 struct dx_map_entry
*map
;
1513 char *data1
= (*bh
)->b_data
, *data2
;
1514 unsigned split
, move
, size
;
1515 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1516 struct ext4_dir_entry_tail
*t
;
1520 if (ext4_has_metadata_csum(dir
->i_sb
))
1521 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1523 bh2
= ext4_append(handle
, dir
, &newblock
);
1527 return (struct ext4_dir_entry_2
*) bh2
;
1530 BUFFER_TRACE(*bh
, "get_write_access");
1531 err
= ext4_journal_get_write_access(handle
, *bh
);
1535 BUFFER_TRACE(frame
->bh
, "get_write_access");
1536 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1540 data2
= bh2
->b_data
;
1542 /* create map in the end of data2 block */
1543 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1544 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1545 blocksize
, hinfo
, map
);
1547 dx_sort_map(map
, count
);
1548 /* Split the existing block in the middle, size-wise */
1551 for (i
= count
-1; i
>= 0; i
--) {
1552 /* is more than half of this entry in 2nd half of the block? */
1553 if (size
+ map
[i
].size
/2 > blocksize
/2)
1555 size
+= map
[i
].size
;
1558 /* map index at which we will split */
1559 split
= count
- move
;
1560 hash2
= map
[split
].hash
;
1561 continued
= hash2
== map
[split
- 1].hash
;
1562 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1563 (unsigned long)dx_get_block(frame
->at
),
1564 hash2
, split
, count
-split
));
1566 /* Fancy dance to stay within two buffers */
1567 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
, blocksize
);
1568 de
= dx_pack_dirents(data1
, blocksize
);
1569 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1572 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1576 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1577 initialize_dirent_tail(t
, blocksize
);
1579 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1580 initialize_dirent_tail(t
, blocksize
);
1583 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1584 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1586 /* Which block gets the new entry? */
1587 if (hinfo
->hash
>= hash2
) {
1591 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1592 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1595 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1599 dxtrace(dx_show_index("frame", frame
->entries
));
1606 ext4_std_error(dir
->i_sb
, err
);
1607 return ERR_PTR(err
);
1610 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1611 struct buffer_head
*bh
,
1612 void *buf
, int buf_size
,
1613 const char *name
, int namelen
,
1614 struct ext4_dir_entry_2
**dest_de
)
1616 struct ext4_dir_entry_2
*de
;
1617 unsigned short reclen
= EXT4_DIR_REC_LEN(namelen
);
1619 unsigned int offset
= 0;
1622 de
= (struct ext4_dir_entry_2
*)buf
;
1623 top
= buf
+ buf_size
- reclen
;
1624 while ((char *) de
<= top
) {
1625 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1626 buf
, buf_size
, offset
))
1628 if (ext4_match(namelen
, name
, de
))
1630 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1631 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1632 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1634 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1637 if ((char *) de
> top
)
1644 void ext4_insert_dentry(struct inode
*inode
,
1645 struct ext4_dir_entry_2
*de
,
1647 const char *name
, int namelen
)
1652 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1653 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1655 struct ext4_dir_entry_2
*de1
=
1656 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1657 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1658 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1661 de
->file_type
= EXT4_FT_UNKNOWN
;
1662 de
->inode
= cpu_to_le32(inode
->i_ino
);
1663 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1664 de
->name_len
= namelen
;
1665 memcpy(de
->name
, name
, namelen
);
1668 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1669 * it points to a directory entry which is guaranteed to be large
1670 * enough for new directory entry. If de is NULL, then
1671 * add_dirent_to_buf will attempt search the directory block for
1672 * space. It will return -ENOSPC if no space is available, and -EIO
1673 * and -EEXIST if directory entry already exists.
1675 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1676 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1677 struct buffer_head
*bh
)
1679 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1680 const char *name
= dentry
->d_name
.name
;
1681 int namelen
= dentry
->d_name
.len
;
1682 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1686 if (ext4_has_metadata_csum(inode
->i_sb
))
1687 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1690 err
= ext4_find_dest_de(dir
, inode
,
1691 bh
, bh
->b_data
, blocksize
- csum_size
,
1692 name
, namelen
, &de
);
1696 BUFFER_TRACE(bh
, "get_write_access");
1697 err
= ext4_journal_get_write_access(handle
, bh
);
1699 ext4_std_error(dir
->i_sb
, err
);
1703 /* By now the buffer is marked for journaling */
1704 ext4_insert_dentry(inode
, de
, blocksize
, name
, namelen
);
1707 * XXX shouldn't update any times until successful
1708 * completion of syscall, but too many callers depend
1711 * XXX similarly, too many callers depend on
1712 * ext4_new_inode() setting the times, but error
1713 * recovery deletes the inode, so the worst that can
1714 * happen is that the times are slightly out of date
1715 * and/or different from the directory change time.
1717 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1718 ext4_update_dx_flag(dir
);
1720 ext4_mark_inode_dirty(handle
, dir
);
1721 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1722 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1724 ext4_std_error(dir
->i_sb
, err
);
1729 * This converts a one block unindexed directory to a 3 block indexed
1730 * directory, and adds the dentry to the indexed directory.
1732 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1733 struct inode
*inode
, struct buffer_head
*bh
)
1735 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1736 const char *name
= dentry
->d_name
.name
;
1737 int namelen
= dentry
->d_name
.len
;
1738 struct buffer_head
*bh2
;
1739 struct dx_root
*root
;
1740 struct dx_frame frames
[2], *frame
;
1741 struct dx_entry
*entries
;
1742 struct ext4_dir_entry_2
*de
, *de2
;
1743 struct ext4_dir_entry_tail
*t
;
1748 struct dx_hash_info hinfo
;
1750 struct fake_dirent
*fde
;
1753 if (ext4_has_metadata_csum(inode
->i_sb
))
1754 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1756 blocksize
= dir
->i_sb
->s_blocksize
;
1757 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1758 BUFFER_TRACE(bh
, "get_write_access");
1759 retval
= ext4_journal_get_write_access(handle
, bh
);
1761 ext4_std_error(dir
->i_sb
, retval
);
1765 root
= (struct dx_root
*) bh
->b_data
;
1767 /* The 0th block becomes the root, move the dirents out */
1768 fde
= &root
->dotdot
;
1769 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1770 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1771 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1772 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1776 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1778 /* Allocate new block for the 0th block's dirents */
1779 bh2
= ext4_append(handle
, dir
, &block
);
1782 return PTR_ERR(bh2
);
1784 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1785 data1
= bh2
->b_data
;
1787 memcpy (data1
, de
, len
);
1788 de
= (struct ext4_dir_entry_2
*) data1
;
1790 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1792 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1797 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1798 initialize_dirent_tail(t
, blocksize
);
1801 /* Initialize the root; the dot dirents already exist */
1802 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1803 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1805 memset (&root
->info
, 0, sizeof(root
->info
));
1806 root
->info
.info_length
= sizeof(root
->info
);
1807 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1808 entries
= root
->entries
;
1809 dx_set_block(entries
, 1);
1810 dx_set_count(entries
, 1);
1811 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1813 /* Initialize as for dx_probe */
1814 hinfo
.hash_version
= root
->info
.hash_version
;
1815 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1816 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1817 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1818 ext4fs_dirhash(name
, namelen
, &hinfo
);
1820 frame
->entries
= entries
;
1821 frame
->at
= entries
;
1825 ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1826 ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1828 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
);
1831 * Even if the block split failed, we have to properly write
1832 * out all the changes we did so far. Otherwise we can end up
1833 * with corrupted filesystem.
1835 ext4_mark_inode_dirty(handle
, dir
);
1841 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1849 * adds a file entry to the specified directory, using the same
1850 * semantics as ext4_find_entry(). It returns NULL if it failed.
1852 * NOTE!! The inode part of 'de' is left at 0 - which means you
1853 * may not sleep between calling this and putting something into
1854 * the entry, as someone else might have used it while you slept.
1856 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1857 struct inode
*inode
)
1859 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1860 struct buffer_head
*bh
;
1861 struct ext4_dir_entry_2
*de
;
1862 struct ext4_dir_entry_tail
*t
;
1863 struct super_block
*sb
;
1867 ext4_lblk_t block
, blocks
;
1870 if (ext4_has_metadata_csum(inode
->i_sb
))
1871 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1874 blocksize
= sb
->s_blocksize
;
1875 if (!dentry
->d_name
.len
)
1878 if (ext4_has_inline_data(dir
)) {
1879 retval
= ext4_try_add_inline_entry(handle
, dentry
, inode
);
1889 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1890 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1892 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
1894 ext4_mark_inode_dirty(handle
, dir
);
1896 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1897 for (block
= 0; block
< blocks
; block
++) {
1898 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1902 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1903 if (retval
!= -ENOSPC
) {
1908 if (blocks
== 1 && !dx_fallback
&&
1909 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1910 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1913 bh
= ext4_append(handle
, dir
, &block
);
1916 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1918 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
1921 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
1922 initialize_dirent_tail(t
, blocksize
);
1925 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1928 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
1933 * Returns 0 for success, or a negative error value
1935 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1936 struct inode
*inode
)
1938 struct dx_frame frames
[2], *frame
;
1939 struct dx_entry
*entries
, *at
;
1940 struct dx_hash_info hinfo
;
1941 struct buffer_head
*bh
;
1942 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1943 struct super_block
*sb
= dir
->i_sb
;
1944 struct ext4_dir_entry_2
*de
;
1947 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
);
1949 return PTR_ERR(frame
);
1950 entries
= frame
->entries
;
1952 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
1959 BUFFER_TRACE(bh
, "get_write_access");
1960 err
= ext4_journal_get_write_access(handle
, bh
);
1964 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1968 /* Block full, should compress but for now just split */
1969 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
1970 dx_get_count(entries
), dx_get_limit(entries
)));
1971 /* Need to split index? */
1972 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1973 ext4_lblk_t newblock
;
1974 unsigned icount
= dx_get_count(entries
);
1975 int levels
= frame
- frames
;
1976 struct dx_entry
*entries2
;
1977 struct dx_node
*node2
;
1978 struct buffer_head
*bh2
;
1980 if (levels
&& (dx_get_count(frames
->entries
) ==
1981 dx_get_limit(frames
->entries
))) {
1982 ext4_warning(sb
, "Directory index full!");
1986 bh2
= ext4_append(handle
, dir
, &newblock
);
1991 node2
= (struct dx_node
*)(bh2
->b_data
);
1992 entries2
= node2
->entries
;
1993 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
1994 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
1996 BUFFER_TRACE(frame
->bh
, "get_write_access");
1997 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2001 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2002 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2003 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2006 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2007 err
= ext4_journal_get_write_access(handle
,
2012 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2013 icount2
* sizeof(struct dx_entry
));
2014 dx_set_count(entries
, icount1
);
2015 dx_set_count(entries2
, icount2
);
2016 dx_set_limit(entries2
, dx_node_limit(dir
));
2018 /* Which index block gets the new entry? */
2019 if (at
- entries
>= icount1
) {
2020 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2021 frame
->entries
= entries
= entries2
;
2022 swap(frame
->bh
, bh2
);
2024 dx_insert_block(frames
+ 0, hash2
, newblock
);
2025 dxtrace(dx_show_index("node", frames
[1].entries
));
2026 dxtrace(dx_show_index("node",
2027 ((struct dx_node
*) bh2
->b_data
)->entries
));
2028 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2033 dxtrace(printk(KERN_DEBUG
2034 "Creating second level index...\n"));
2035 memcpy((char *) entries2
, (char *) entries
,
2036 icount
* sizeof(struct dx_entry
));
2037 dx_set_limit(entries2
, dx_node_limit(dir
));
2040 dx_set_count(entries
, 1);
2041 dx_set_block(entries
+ 0, newblock
);
2042 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2044 /* Add new access path frame */
2046 frame
->at
= at
= at
- entries
+ entries2
;
2047 frame
->entries
= entries
= entries2
;
2049 err
= ext4_journal_get_write_access(handle
,
2054 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2056 ext4_std_error(inode
->i_sb
, err
);
2060 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
);
2065 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
2069 ext4_std_error(dir
->i_sb
, err
);
2077 * ext4_generic_delete_entry deletes a directory entry by merging it
2078 * with the previous entry
2080 int ext4_generic_delete_entry(handle_t
*handle
,
2082 struct ext4_dir_entry_2
*de_del
,
2083 struct buffer_head
*bh
,
2088 struct ext4_dir_entry_2
*de
, *pde
;
2089 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2094 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2095 while (i
< buf_size
- csum_size
) {
2096 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2097 bh
->b_data
, bh
->b_size
, i
))
2101 pde
->rec_len
= ext4_rec_len_to_disk(
2102 ext4_rec_len_from_disk(pde
->rec_len
,
2104 ext4_rec_len_from_disk(de
->rec_len
,
2112 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2114 de
= ext4_next_entry(de
, blocksize
);
2119 static int ext4_delete_entry(handle_t
*handle
,
2121 struct ext4_dir_entry_2
*de_del
,
2122 struct buffer_head
*bh
)
2124 int err
, csum_size
= 0;
2126 if (ext4_has_inline_data(dir
)) {
2127 int has_inline_data
= 1;
2128 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2130 if (has_inline_data
)
2134 if (ext4_has_metadata_csum(dir
->i_sb
))
2135 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2137 BUFFER_TRACE(bh
, "get_write_access");
2138 err
= ext4_journal_get_write_access(handle
, bh
);
2142 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2144 dir
->i_sb
->s_blocksize
, csum_size
);
2148 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2149 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2156 ext4_std_error(dir
->i_sb
, err
);
2161 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2162 * since this indicates that nlinks count was previously 1.
2164 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2167 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2168 /* limit is 16-bit i_links_count */
2169 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2170 set_nlink(inode
, 1);
2171 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
2172 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
2178 * If a directory had nlink == 1, then we should let it be 1. This indicates
2179 * directory has >EXT4_LINK_MAX subdirs.
2181 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2183 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2188 static int ext4_add_nondir(handle_t
*handle
,
2189 struct dentry
*dentry
, struct inode
*inode
)
2191 int err
= ext4_add_entry(handle
, dentry
, inode
);
2193 ext4_mark_inode_dirty(handle
, inode
);
2194 unlock_new_inode(inode
);
2195 d_instantiate(dentry
, inode
);
2199 unlock_new_inode(inode
);
2205 * By the time this is called, we already have created
2206 * the directory cache entry for the new file, but it
2207 * is so far negative - it has no inode.
2209 * If the create succeeds, we fill in the inode information
2210 * with d_instantiate().
2212 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2216 struct inode
*inode
;
2217 int err
, credits
, retries
= 0;
2219 dquot_initialize(dir
);
2221 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2222 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2224 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2225 NULL
, EXT4_HT_DIR
, credits
);
2226 handle
= ext4_journal_current_handle();
2227 err
= PTR_ERR(inode
);
2228 if (!IS_ERR(inode
)) {
2229 inode
->i_op
= &ext4_file_inode_operations
;
2230 inode
->i_fop
= &ext4_file_operations
;
2231 ext4_set_aops(inode
);
2232 err
= ext4_add_nondir(handle
, dentry
, inode
);
2233 if (!err
&& IS_DIRSYNC(dir
))
2234 ext4_handle_sync(handle
);
2237 ext4_journal_stop(handle
);
2238 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2243 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2244 umode_t mode
, dev_t rdev
)
2247 struct inode
*inode
;
2248 int err
, credits
, retries
= 0;
2250 if (!new_valid_dev(rdev
))
2253 dquot_initialize(dir
);
2255 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2256 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2258 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2259 NULL
, EXT4_HT_DIR
, credits
);
2260 handle
= ext4_journal_current_handle();
2261 err
= PTR_ERR(inode
);
2262 if (!IS_ERR(inode
)) {
2263 init_special_inode(inode
, inode
->i_mode
, rdev
);
2264 inode
->i_op
= &ext4_special_inode_operations
;
2265 err
= ext4_add_nondir(handle
, dentry
, inode
);
2266 if (!err
&& IS_DIRSYNC(dir
))
2267 ext4_handle_sync(handle
);
2270 ext4_journal_stop(handle
);
2271 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2276 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2279 struct inode
*inode
;
2280 int err
, retries
= 0;
2282 dquot_initialize(dir
);
2285 inode
= ext4_new_inode_start_handle(dir
, mode
,
2288 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2289 4 + EXT4_XATTR_TRANS_BLOCKS
);
2290 handle
= ext4_journal_current_handle();
2291 err
= PTR_ERR(inode
);
2292 if (!IS_ERR(inode
)) {
2293 inode
->i_op
= &ext4_file_inode_operations
;
2294 inode
->i_fop
= &ext4_file_operations
;
2295 ext4_set_aops(inode
);
2296 d_tmpfile(dentry
, inode
);
2297 err
= ext4_orphan_add(handle
, inode
);
2299 goto err_unlock_inode
;
2300 mark_inode_dirty(inode
);
2301 unlock_new_inode(inode
);
2304 ext4_journal_stop(handle
);
2305 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2309 ext4_journal_stop(handle
);
2310 unlock_new_inode(inode
);
2314 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2315 struct ext4_dir_entry_2
*de
,
2316 int blocksize
, int csum_size
,
2317 unsigned int parent_ino
, int dotdot_real_len
)
2319 de
->inode
= cpu_to_le32(inode
->i_ino
);
2321 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2323 strcpy(de
->name
, ".");
2324 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2326 de
= ext4_next_entry(de
, blocksize
);
2327 de
->inode
= cpu_to_le32(parent_ino
);
2329 if (!dotdot_real_len
)
2330 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2331 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2334 de
->rec_len
= ext4_rec_len_to_disk(
2335 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2336 strcpy(de
->name
, "..");
2337 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2339 return ext4_next_entry(de
, blocksize
);
2342 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2343 struct inode
*inode
)
2345 struct buffer_head
*dir_block
= NULL
;
2346 struct ext4_dir_entry_2
*de
;
2347 struct ext4_dir_entry_tail
*t
;
2348 ext4_lblk_t block
= 0;
2349 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2353 if (ext4_has_metadata_csum(dir
->i_sb
))
2354 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2356 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2357 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2358 if (err
< 0 && err
!= -ENOSPC
)
2365 dir_block
= ext4_append(handle
, inode
, &block
);
2366 if (IS_ERR(dir_block
))
2367 return PTR_ERR(dir_block
);
2368 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2369 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2370 set_nlink(inode
, 2);
2372 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2373 initialize_dirent_tail(t
, blocksize
);
2376 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2377 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2380 set_buffer_verified(dir_block
);
2386 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2389 struct inode
*inode
;
2390 int err
, credits
, retries
= 0;
2392 if (EXT4_DIR_LINK_MAX(dir
))
2395 dquot_initialize(dir
);
2397 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2398 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2400 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2402 0, NULL
, EXT4_HT_DIR
, credits
);
2403 handle
= ext4_journal_current_handle();
2404 err
= PTR_ERR(inode
);
2408 inode
->i_op
= &ext4_dir_inode_operations
;
2409 inode
->i_fop
= &ext4_dir_operations
;
2410 err
= ext4_init_new_dir(handle
, dir
, inode
);
2412 goto out_clear_inode
;
2413 err
= ext4_mark_inode_dirty(handle
, inode
);
2415 err
= ext4_add_entry(handle
, dentry
, inode
);
2419 unlock_new_inode(inode
);
2420 ext4_mark_inode_dirty(handle
, inode
);
2424 ext4_inc_count(handle
, dir
);
2425 ext4_update_dx_flag(dir
);
2426 err
= ext4_mark_inode_dirty(handle
, dir
);
2428 goto out_clear_inode
;
2429 unlock_new_inode(inode
);
2430 d_instantiate(dentry
, inode
);
2431 if (IS_DIRSYNC(dir
))
2432 ext4_handle_sync(handle
);
2436 ext4_journal_stop(handle
);
2437 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2443 * routine to check that the specified directory is empty (for rmdir)
2445 static int empty_dir(struct inode
*inode
)
2447 unsigned int offset
;
2448 struct buffer_head
*bh
;
2449 struct ext4_dir_entry_2
*de
, *de1
;
2450 struct super_block
*sb
;
2453 if (ext4_has_inline_data(inode
)) {
2454 int has_inline_data
= 1;
2456 err
= empty_inline_dir(inode
, &has_inline_data
);
2457 if (has_inline_data
)
2462 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2463 EXT4_ERROR_INODE(inode
, "invalid size");
2466 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2470 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2471 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2472 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2473 !le32_to_cpu(de1
->inode
) ||
2474 strcmp(".", de
->name
) ||
2475 strcmp("..", de1
->name
)) {
2476 ext4_warning(inode
->i_sb
,
2477 "bad directory (dir #%lu) - no `.' or `..'",
2482 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2483 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2484 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2485 while (offset
< inode
->i_size
) {
2486 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2487 unsigned int lblock
;
2490 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2491 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2494 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2496 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2497 bh
->b_data
, bh
->b_size
, offset
)) {
2498 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2500 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2503 if (le32_to_cpu(de
->inode
)) {
2507 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2508 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2515 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2516 * such inodes, starting at the superblock, in case we crash before the
2517 * file is closed/deleted, or in case the inode truncate spans multiple
2518 * transactions and the last transaction is not recovered after a crash.
2520 * At filesystem recovery time, we walk this list deleting unlinked
2521 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2523 * Orphan list manipulation functions must be called under i_mutex unless
2524 * we are just creating the inode or deleting it.
2526 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2528 struct super_block
*sb
= inode
->i_sb
;
2529 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2530 struct ext4_iloc iloc
;
2534 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2537 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2538 !mutex_is_locked(&inode
->i_mutex
));
2540 * Exit early if inode already is on orphan list. This is a big speedup
2541 * since we don't have to contend on the global s_orphan_lock.
2543 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2547 * Orphan handling is only valid for files with data blocks
2548 * being truncated, or files being unlinked. Note that we either
2549 * hold i_mutex, or the inode can not be referenced from outside,
2550 * so i_nlink should not be bumped due to race
2552 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2553 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2555 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2556 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2560 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2564 mutex_lock(&sbi
->s_orphan_lock
);
2566 * Due to previous errors inode may be already a part of on-disk
2567 * orphan list. If so skip on-disk list modification.
2569 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2570 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2571 /* Insert this inode at the head of the on-disk orphan list */
2572 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2573 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2576 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2577 mutex_unlock(&sbi
->s_orphan_lock
);
2580 err
= ext4_handle_dirty_super(handle
, sb
);
2581 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2586 * We have to remove inode from in-memory list if
2587 * addition to on disk orphan list failed. Stray orphan
2588 * list entries can cause panics at unmount time.
2590 mutex_lock(&sbi
->s_orphan_lock
);
2591 list_del(&EXT4_I(inode
)->i_orphan
);
2592 mutex_unlock(&sbi
->s_orphan_lock
);
2595 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2596 jbd_debug(4, "orphan inode %lu will point to %d\n",
2597 inode
->i_ino
, NEXT_ORPHAN(inode
));
2599 ext4_std_error(sb
, err
);
2604 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2605 * of such inodes stored on disk, because it is finally being cleaned up.
2607 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2609 struct list_head
*prev
;
2610 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2611 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2613 struct ext4_iloc iloc
;
2616 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2619 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2620 !mutex_is_locked(&inode
->i_mutex
));
2621 /* Do this quick check before taking global s_orphan_lock. */
2622 if (list_empty(&ei
->i_orphan
))
2626 /* Grab inode buffer early before taking global s_orphan_lock */
2627 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2630 mutex_lock(&sbi
->s_orphan_lock
);
2631 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2633 prev
= ei
->i_orphan
.prev
;
2634 list_del_init(&ei
->i_orphan
);
2636 /* If we're on an error path, we may not have a valid
2637 * transaction handle with which to update the orphan list on
2638 * disk, but we still need to remove the inode from the linked
2639 * list in memory. */
2640 if (!handle
|| err
) {
2641 mutex_unlock(&sbi
->s_orphan_lock
);
2645 ino_next
= NEXT_ORPHAN(inode
);
2646 if (prev
== &sbi
->s_orphan
) {
2647 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2648 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2649 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2651 mutex_unlock(&sbi
->s_orphan_lock
);
2654 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2655 mutex_unlock(&sbi
->s_orphan_lock
);
2656 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2658 struct ext4_iloc iloc2
;
2659 struct inode
*i_prev
=
2660 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2662 jbd_debug(4, "orphan inode %lu will point to %u\n",
2663 i_prev
->i_ino
, ino_next
);
2664 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2666 mutex_unlock(&sbi
->s_orphan_lock
);
2669 NEXT_ORPHAN(i_prev
) = ino_next
;
2670 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2671 mutex_unlock(&sbi
->s_orphan_lock
);
2675 NEXT_ORPHAN(inode
) = 0;
2676 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2678 ext4_std_error(inode
->i_sb
, err
);
2686 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2689 struct inode
*inode
;
2690 struct buffer_head
*bh
;
2691 struct ext4_dir_entry_2
*de
;
2692 handle_t
*handle
= NULL
;
2694 /* Initialize quotas before so that eventual writes go in
2695 * separate transaction */
2696 dquot_initialize(dir
);
2697 dquot_initialize(dentry
->d_inode
);
2700 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2706 inode
= dentry
->d_inode
;
2709 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2712 retval
= -ENOTEMPTY
;
2713 if (!empty_dir(inode
))
2716 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2717 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2718 if (IS_ERR(handle
)) {
2719 retval
= PTR_ERR(handle
);
2724 if (IS_DIRSYNC(dir
))
2725 ext4_handle_sync(handle
);
2727 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2730 if (!EXT4_DIR_LINK_EMPTY(inode
))
2731 ext4_warning(inode
->i_sb
,
2732 "empty directory has too many links (%d)",
2736 /* There's no need to set i_disksize: the fact that i_nlink is
2737 * zero will ensure that the right thing happens during any
2740 ext4_orphan_add(handle
, inode
);
2741 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2742 ext4_mark_inode_dirty(handle
, inode
);
2743 ext4_dec_count(handle
, dir
);
2744 ext4_update_dx_flag(dir
);
2745 ext4_mark_inode_dirty(handle
, dir
);
2750 ext4_journal_stop(handle
);
2754 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2757 struct inode
*inode
;
2758 struct buffer_head
*bh
;
2759 struct ext4_dir_entry_2
*de
;
2760 handle_t
*handle
= NULL
;
2762 trace_ext4_unlink_enter(dir
, dentry
);
2763 /* Initialize quotas before so that eventual writes go
2764 * in separate transaction */
2765 dquot_initialize(dir
);
2766 dquot_initialize(dentry
->d_inode
);
2769 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2775 inode
= dentry
->d_inode
;
2778 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2781 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2782 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2783 if (IS_ERR(handle
)) {
2784 retval
= PTR_ERR(handle
);
2789 if (IS_DIRSYNC(dir
))
2790 ext4_handle_sync(handle
);
2792 if (!inode
->i_nlink
) {
2793 ext4_warning(inode
->i_sb
,
2794 "Deleting nonexistent file (%lu), %d",
2795 inode
->i_ino
, inode
->i_nlink
);
2796 set_nlink(inode
, 1);
2798 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2801 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2802 ext4_update_dx_flag(dir
);
2803 ext4_mark_inode_dirty(handle
, dir
);
2805 if (!inode
->i_nlink
)
2806 ext4_orphan_add(handle
, inode
);
2807 inode
->i_ctime
= ext4_current_time(inode
);
2808 ext4_mark_inode_dirty(handle
, inode
);
2814 ext4_journal_stop(handle
);
2815 trace_ext4_unlink_exit(dentry
, retval
);
2819 static int ext4_symlink(struct inode
*dir
,
2820 struct dentry
*dentry
, const char *symname
)
2823 struct inode
*inode
;
2824 int l
, err
, retries
= 0;
2827 l
= strlen(symname
)+1;
2828 if (l
> dir
->i_sb
->s_blocksize
)
2829 return -ENAMETOOLONG
;
2831 dquot_initialize(dir
);
2833 if (l
> EXT4_N_BLOCKS
* 4) {
2835 * For non-fast symlinks, we just allocate inode and put it on
2836 * orphan list in the first transaction => we need bitmap,
2837 * group descriptor, sb, inode block, quota blocks, and
2838 * possibly selinux xattr blocks.
2840 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2841 EXT4_XATTR_TRANS_BLOCKS
;
2844 * Fast symlink. We have to add entry to directory
2845 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2846 * allocate new inode (bitmap, group descriptor, inode block,
2847 * quota blocks, sb is already counted in previous macros).
2849 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2850 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
2853 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
2854 &dentry
->d_name
, 0, NULL
,
2855 EXT4_HT_DIR
, credits
);
2856 handle
= ext4_journal_current_handle();
2857 err
= PTR_ERR(inode
);
2861 if (l
> EXT4_N_BLOCKS
* 4) {
2862 inode
->i_op
= &ext4_symlink_inode_operations
;
2863 ext4_set_aops(inode
);
2865 * We cannot call page_symlink() with transaction started
2866 * because it calls into ext4_write_begin() which can wait
2867 * for transaction commit if we are running out of space
2868 * and thus we deadlock. So we have to stop transaction now
2869 * and restart it when symlink contents is written.
2871 * To keep fs consistent in case of crash, we have to put inode
2872 * to orphan list in the mean time.
2875 err
= ext4_orphan_add(handle
, inode
);
2876 ext4_journal_stop(handle
);
2878 goto err_drop_inode
;
2879 err
= __page_symlink(inode
, symname
, l
, 1);
2881 goto err_drop_inode
;
2883 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2884 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2886 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2887 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2888 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
2889 if (IS_ERR(handle
)) {
2890 err
= PTR_ERR(handle
);
2891 goto err_drop_inode
;
2893 set_nlink(inode
, 1);
2894 err
= ext4_orphan_del(handle
, inode
);
2896 ext4_journal_stop(handle
);
2898 goto err_drop_inode
;
2901 /* clear the extent format for fast symlink */
2902 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
2903 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2904 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2905 inode
->i_size
= l
-1;
2907 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2908 err
= ext4_add_nondir(handle
, dentry
, inode
);
2909 if (!err
&& IS_DIRSYNC(dir
))
2910 ext4_handle_sync(handle
);
2914 ext4_journal_stop(handle
);
2915 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2919 unlock_new_inode(inode
);
2924 static int ext4_link(struct dentry
*old_dentry
,
2925 struct inode
*dir
, struct dentry
*dentry
)
2928 struct inode
*inode
= old_dentry
->d_inode
;
2929 int err
, retries
= 0;
2931 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
2934 dquot_initialize(dir
);
2937 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2938 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2939 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
2941 return PTR_ERR(handle
);
2943 if (IS_DIRSYNC(dir
))
2944 ext4_handle_sync(handle
);
2946 inode
->i_ctime
= ext4_current_time(inode
);
2947 ext4_inc_count(handle
, inode
);
2950 err
= ext4_add_entry(handle
, dentry
, inode
);
2952 ext4_mark_inode_dirty(handle
, inode
);
2953 /* this can happen only for tmpfile being
2954 * linked the first time
2956 if (inode
->i_nlink
== 1)
2957 ext4_orphan_del(handle
, inode
);
2958 d_instantiate(dentry
, inode
);
2963 ext4_journal_stop(handle
);
2964 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2971 * Try to find buffer head where contains the parent block.
2972 * It should be the inode block if it is inlined or the 1st block
2973 * if it is a normal dir.
2975 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
2976 struct inode
*inode
,
2978 struct ext4_dir_entry_2
**parent_de
,
2981 struct buffer_head
*bh
;
2983 if (!ext4_has_inline_data(inode
)) {
2984 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2986 *retval
= PTR_ERR(bh
);
2989 *parent_de
= ext4_next_entry(
2990 (struct ext4_dir_entry_2
*)bh
->b_data
,
2991 inode
->i_sb
->s_blocksize
);
2996 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
2999 struct ext4_renament
{
3001 struct dentry
*dentry
;
3002 struct inode
*inode
;
3004 int dir_nlink_delta
;
3006 /* entry for "dentry" */
3007 struct buffer_head
*bh
;
3008 struct ext4_dir_entry_2
*de
;
3011 /* entry for ".." in inode if it's a directory */
3012 struct buffer_head
*dir_bh
;
3013 struct ext4_dir_entry_2
*parent_de
;
3017 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3021 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3022 &retval
, &ent
->parent_de
,
3026 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3028 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3029 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3032 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3037 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3038 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3039 if (!ent
->dir_inlined
) {
3040 if (is_dx(ent
->inode
)) {
3041 retval
= ext4_handle_dirty_dx_node(handle
,
3045 retval
= ext4_handle_dirty_dirent_node(handle
,
3050 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3053 ext4_std_error(ent
->dir
->i_sb
, retval
);
3059 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3060 unsigned ino
, unsigned file_type
)
3064 BUFFER_TRACE(ent
->bh
, "get write access");
3065 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3068 ent
->de
->inode
= cpu_to_le32(ino
);
3069 if (EXT4_HAS_INCOMPAT_FEATURE(ent
->dir
->i_sb
,
3070 EXT4_FEATURE_INCOMPAT_FILETYPE
))
3071 ent
->de
->file_type
= file_type
;
3072 ent
->dir
->i_version
++;
3073 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3074 ext4_current_time(ent
->dir
);
3075 ext4_mark_inode_dirty(handle
, ent
->dir
);
3076 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3077 if (!ent
->inlined
) {
3078 retval
= ext4_handle_dirty_dirent_node(handle
,
3080 if (unlikely(retval
)) {
3081 ext4_std_error(ent
->dir
->i_sb
, retval
);
3091 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3092 const struct qstr
*d_name
)
3094 int retval
= -ENOENT
;
3095 struct buffer_head
*bh
;
3096 struct ext4_dir_entry_2
*de
;
3098 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3102 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3108 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3113 * ent->de could have moved from under us during htree split, so make
3114 * sure that we are deleting the right entry. We might also be pointing
3115 * to a stale entry in the unused part of ent->bh so just checking inum
3116 * and the name isn't enough.
3118 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3119 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3120 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3121 ent
->de
->name_len
) ||
3123 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3124 &ent
->dentry
->d_name
);
3126 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3127 if (retval
== -ENOENT
) {
3128 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3129 &ent
->dentry
->d_name
);
3134 ext4_warning(ent
->dir
->i_sb
,
3135 "Deleting old file (%lu), %d, error=%d",
3136 ent
->dir
->i_ino
, ent
->dir
->i_nlink
, retval
);
3140 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3142 if (ent
->dir_nlink_delta
) {
3143 if (ent
->dir_nlink_delta
== -1)
3144 ext4_dec_count(handle
, ent
->dir
);
3146 ext4_inc_count(handle
, ent
->dir
);
3147 ext4_mark_inode_dirty(handle
, ent
->dir
);
3152 * Anybody can rename anything with this: the permission checks are left to the
3153 * higher-level routines.
3155 * n.b. old_{dentry,inode) refers to the source dentry/inode
3156 * while new_{dentry,inode) refers to the destination dentry/inode
3157 * This comes from rename(const char *oldpath, const char *newpath)
3159 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3160 struct inode
*new_dir
, struct dentry
*new_dentry
)
3162 handle_t
*handle
= NULL
;
3163 struct ext4_renament old
= {
3165 .dentry
= old_dentry
,
3166 .inode
= old_dentry
->d_inode
,
3168 struct ext4_renament
new = {
3170 .dentry
= new_dentry
,
3171 .inode
= new_dentry
->d_inode
,
3176 dquot_initialize(old
.dir
);
3177 dquot_initialize(new.dir
);
3179 /* Initialize quotas before so that eventual writes go
3180 * in separate transaction */
3182 dquot_initialize(new.inode
);
3184 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3186 return PTR_ERR(old
.bh
);
3188 * Check for inode number is _not_ due to possible IO errors.
3189 * We might rmdir the source, keep it as pwd of some process
3190 * and merrily kill the link to whatever was created under the
3191 * same name. Goodbye sticky bit ;-<
3194 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3197 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3198 &new.de
, &new.inlined
);
3199 if (IS_ERR(new.bh
)) {
3200 retval
= PTR_ERR(new.bh
);
3209 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3210 ext4_alloc_da_blocks(old
.inode
);
3212 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3213 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3214 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3216 return PTR_ERR(handle
);
3218 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3219 ext4_handle_sync(handle
);
3221 if (S_ISDIR(old
.inode
->i_mode
)) {
3223 retval
= -ENOTEMPTY
;
3224 if (!empty_dir(new.inode
))
3228 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3231 retval
= ext4_rename_dir_prepare(handle
, &old
);
3236 * If we're renaming a file within an inline_data dir and adding or
3237 * setting the new dirent causes a conversion from inline_data to
3238 * extents/blockmap, we need to force the dirent delete code to
3239 * re-read the directory, or else we end up trying to delete a dirent
3240 * from what is now the extent tree root (or a block map).
3242 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3243 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3245 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3249 retval
= ext4_setent(handle
, &new,
3250 old
.inode
->i_ino
, old
.de
->file_type
);
3255 force_reread
= !ext4_test_inode_flag(new.dir
,
3256 EXT4_INODE_INLINE_DATA
);
3259 * Like most other Unix systems, set the ctime for inodes on a
3262 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3263 ext4_mark_inode_dirty(handle
, old
.inode
);
3268 ext4_rename_delete(handle
, &old
, force_reread
);
3271 ext4_dec_count(handle
, new.inode
);
3272 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3274 old
.dir
->i_ctime
= old
.dir
->i_mtime
= ext4_current_time(old
.dir
);
3275 ext4_update_dx_flag(old
.dir
);
3277 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3281 ext4_dec_count(handle
, old
.dir
);
3283 /* checked empty_dir above, can't have another parent,
3284 * ext4_dec_count() won't work for many-linked dirs */
3285 clear_nlink(new.inode
);
3287 ext4_inc_count(handle
, new.dir
);
3288 ext4_update_dx_flag(new.dir
);
3289 ext4_mark_inode_dirty(handle
, new.dir
);
3292 ext4_mark_inode_dirty(handle
, old
.dir
);
3294 ext4_mark_inode_dirty(handle
, new.inode
);
3295 if (!new.inode
->i_nlink
)
3296 ext4_orphan_add(handle
, new.inode
);
3305 ext4_journal_stop(handle
);
3309 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3310 struct inode
*new_dir
, struct dentry
*new_dentry
)
3312 handle_t
*handle
= NULL
;
3313 struct ext4_renament old
= {
3315 .dentry
= old_dentry
,
3316 .inode
= old_dentry
->d_inode
,
3318 struct ext4_renament
new = {
3320 .dentry
= new_dentry
,
3321 .inode
= new_dentry
->d_inode
,
3326 dquot_initialize(old
.dir
);
3327 dquot_initialize(new.dir
);
3329 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3330 &old
.de
, &old
.inlined
);
3332 return PTR_ERR(old
.bh
);
3334 * Check for inode number is _not_ due to possible IO errors.
3335 * We might rmdir the source, keep it as pwd of some process
3336 * and merrily kill the link to whatever was created under the
3337 * same name. Goodbye sticky bit ;-<
3340 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3343 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3344 &new.de
, &new.inlined
);
3345 if (IS_ERR(new.bh
)) {
3346 retval
= PTR_ERR(new.bh
);
3350 /* RENAME_EXCHANGE case: old *and* new must both exist */
3351 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3354 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3355 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3356 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3358 return PTR_ERR(handle
);
3360 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3361 ext4_handle_sync(handle
);
3363 if (S_ISDIR(old
.inode
->i_mode
)) {
3365 retval
= ext4_rename_dir_prepare(handle
, &old
);
3369 if (S_ISDIR(new.inode
->i_mode
)) {
3371 retval
= ext4_rename_dir_prepare(handle
, &new);
3377 * Other than the special case of overwriting a directory, parents'
3378 * nlink only needs to be modified if this is a cross directory rename.
3380 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3381 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3382 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3384 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3385 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3389 new_file_type
= new.de
->file_type
;
3390 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3394 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3399 * Like most other Unix systems, set the ctime for inodes on a
3402 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3403 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3404 ext4_mark_inode_dirty(handle
, old
.inode
);
3405 ext4_mark_inode_dirty(handle
, new.inode
);
3408 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3413 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3417 ext4_update_dir_count(handle
, &old
);
3418 ext4_update_dir_count(handle
, &new);
3427 ext4_journal_stop(handle
);
3431 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3432 struct inode
*new_dir
, struct dentry
*new_dentry
,
3435 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
))
3438 if (flags
& RENAME_EXCHANGE
) {
3439 return ext4_cross_rename(old_dir
, old_dentry
,
3440 new_dir
, new_dentry
);
3443 * Existence checking was done by the VFS, otherwise "RENAME_NOREPLACE"
3444 * is equivalent to regular rename.
3446 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3450 * directories can handle most operations...
3452 const struct inode_operations ext4_dir_inode_operations
= {
3453 .create
= ext4_create
,
3454 .lookup
= ext4_lookup
,
3456 .unlink
= ext4_unlink
,
3457 .symlink
= ext4_symlink
,
3458 .mkdir
= ext4_mkdir
,
3459 .rmdir
= ext4_rmdir
,
3460 .mknod
= ext4_mknod
,
3461 .tmpfile
= ext4_tmpfile
,
3462 .rename2
= ext4_rename2
,
3463 .setattr
= ext4_setattr
,
3464 .setxattr
= generic_setxattr
,
3465 .getxattr
= generic_getxattr
,
3466 .listxattr
= ext4_listxattr
,
3467 .removexattr
= generic_removexattr
,
3468 .get_acl
= ext4_get_acl
,
3469 .set_acl
= ext4_set_acl
,
3470 .fiemap
= ext4_fiemap
,
3473 const struct inode_operations ext4_special_inode_operations
= {
3474 .setattr
= ext4_setattr
,
3475 .setxattr
= generic_setxattr
,
3476 .getxattr
= generic_getxattr
,
3477 .listxattr
= ext4_listxattr
,
3478 .removexattr
= generic_removexattr
,
3479 .get_acl
= ext4_get_acl
,
3480 .set_acl
= ext4_set_acl
,