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
, 1);
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), __LINE__)
89 static struct buffer_head
*__ext4_read_dirblock(struct inode
*inode
,
94 struct buffer_head
*bh
;
95 struct ext4_dir_entry
*dirent
;
98 bh
= ext4_bread(NULL
, inode
, block
, 0);
100 __ext4_warning(inode
->i_sb
, __func__
, line
,
101 "error %ld reading directory block "
102 "(ino %lu, block %lu)", PTR_ERR(bh
), inode
->i_ino
,
103 (unsigned long) block
);
108 ext4_error_inode(inode
, __func__
, line
, block
, "Directory hole found");
109 return ERR_PTR(-EIO
);
111 dirent
= (struct ext4_dir_entry
*) bh
->b_data
;
112 /* Determine whether or not we have an index block */
116 else if (ext4_rec_len_from_disk(dirent
->rec_len
,
117 inode
->i_sb
->s_blocksize
) ==
118 inode
->i_sb
->s_blocksize
)
121 if (!is_dx_block
&& type
== INDEX
) {
122 ext4_error_inode(inode
, __func__
, line
, block
,
123 "directory leaf block found instead of index block");
124 return ERR_PTR(-EIO
);
126 if (!ext4_has_metadata_csum(inode
->i_sb
) ||
131 * An empty leaf block can get mistaken for a index block; for
132 * this reason, we can only check the index checksum when the
133 * caller is sure it should be an index block.
135 if (is_dx_block
&& type
== INDEX
) {
136 if (ext4_dx_csum_verify(inode
, dirent
))
137 set_buffer_verified(bh
);
139 ext4_error_inode(inode
, __func__
, line
, block
,
140 "Directory index failed checksum");
142 return ERR_PTR(-EIO
);
146 if (ext4_dirent_csum_verify(inode
, dirent
))
147 set_buffer_verified(bh
);
149 ext4_error_inode(inode
, __func__
, line
, block
,
150 "Directory block failed checksum");
152 return ERR_PTR(-EIO
);
159 #define assert(test) J_ASSERT(test)
163 #define dxtrace(command) command
165 #define dxtrace(command)
189 * dx_root_info is laid out so that if it should somehow get overlaid by a
190 * dirent the two low bits of the hash version will be zero. Therefore, the
191 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
196 struct fake_dirent dot
;
198 struct fake_dirent dotdot
;
202 __le32 reserved_zero
;
204 u8 info_length
; /* 8 */
209 struct dx_entry entries
[0];
214 struct fake_dirent fake
;
215 struct dx_entry entries
[0];
221 struct buffer_head
*bh
;
222 struct dx_entry
*entries
;
234 * This goes at the end of each htree block.
238 __le32 dt_checksum
; /* crc32c(uuid+inum+dirblock) */
241 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
242 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
243 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
244 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
245 static unsigned dx_get_count(struct dx_entry
*entries
);
246 static unsigned dx_get_limit(struct dx_entry
*entries
);
247 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
248 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
249 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
250 static unsigned dx_node_limit(struct inode
*dir
);
251 static struct dx_frame
*dx_probe(const struct qstr
*d_name
,
253 struct dx_hash_info
*hinfo
,
254 struct dx_frame
*frame
);
255 static void dx_release(struct dx_frame
*frames
);
256 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
257 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
258 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
259 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
260 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
261 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
262 static void dx_insert_block(struct dx_frame
*frame
,
263 u32 hash
, ext4_lblk_t block
);
264 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
265 struct dx_frame
*frame
,
266 struct dx_frame
*frames
,
268 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
269 const struct qstr
*d_name
,
270 struct ext4_dir_entry_2
**res_dir
);
271 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
272 struct inode
*inode
);
274 /* checksumming functions */
275 void initialize_dirent_tail(struct ext4_dir_entry_tail
*t
,
276 unsigned int blocksize
)
278 memset(t
, 0, sizeof(struct ext4_dir_entry_tail
));
279 t
->det_rec_len
= ext4_rec_len_to_disk(
280 sizeof(struct ext4_dir_entry_tail
), blocksize
);
281 t
->det_reserved_ft
= EXT4_FT_DIR_CSUM
;
284 /* Walk through a dirent block to find a checksum "dirent" at the tail */
285 static struct ext4_dir_entry_tail
*get_dirent_tail(struct inode
*inode
,
286 struct ext4_dir_entry
*de
)
288 struct ext4_dir_entry_tail
*t
;
291 struct ext4_dir_entry
*d
, *top
;
294 top
= (struct ext4_dir_entry
*)(((void *)de
) +
295 (EXT4_BLOCK_SIZE(inode
->i_sb
) -
296 sizeof(struct ext4_dir_entry_tail
)));
297 while (d
< top
&& d
->rec_len
)
298 d
= (struct ext4_dir_entry
*)(((void *)d
) +
299 le16_to_cpu(d
->rec_len
));
304 t
= (struct ext4_dir_entry_tail
*)d
;
306 t
= EXT4_DIRENT_TAIL(de
, EXT4_BLOCK_SIZE(inode
->i_sb
));
309 if (t
->det_reserved_zero1
||
310 le16_to_cpu(t
->det_rec_len
) != sizeof(struct ext4_dir_entry_tail
) ||
311 t
->det_reserved_zero2
||
312 t
->det_reserved_ft
!= EXT4_FT_DIR_CSUM
)
318 static __le32
ext4_dirent_csum(struct inode
*inode
,
319 struct ext4_dir_entry
*dirent
, int size
)
321 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
322 struct ext4_inode_info
*ei
= EXT4_I(inode
);
325 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
326 return cpu_to_le32(csum
);
329 static void warn_no_space_for_csum(struct inode
*inode
)
331 ext4_warning(inode
->i_sb
, "no space in directory inode %lu leaf for "
332 "checksum. Please run e2fsck -D.", inode
->i_ino
);
335 int ext4_dirent_csum_verify(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
337 struct ext4_dir_entry_tail
*t
;
339 if (!ext4_has_metadata_csum(inode
->i_sb
))
342 t
= get_dirent_tail(inode
, dirent
);
344 warn_no_space_for_csum(inode
);
348 if (t
->det_checksum
!= ext4_dirent_csum(inode
, dirent
,
349 (void *)t
- (void *)dirent
))
355 static void ext4_dirent_csum_set(struct inode
*inode
,
356 struct ext4_dir_entry
*dirent
)
358 struct ext4_dir_entry_tail
*t
;
360 if (!ext4_has_metadata_csum(inode
->i_sb
))
363 t
= get_dirent_tail(inode
, dirent
);
365 warn_no_space_for_csum(inode
);
369 t
->det_checksum
= ext4_dirent_csum(inode
, dirent
,
370 (void *)t
- (void *)dirent
);
373 int ext4_handle_dirty_dirent_node(handle_t
*handle
,
375 struct buffer_head
*bh
)
377 ext4_dirent_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
378 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
381 static struct dx_countlimit
*get_dx_countlimit(struct inode
*inode
,
382 struct ext4_dir_entry
*dirent
,
385 struct ext4_dir_entry
*dp
;
386 struct dx_root_info
*root
;
389 if (le16_to_cpu(dirent
->rec_len
) == EXT4_BLOCK_SIZE(inode
->i_sb
))
391 else if (le16_to_cpu(dirent
->rec_len
) == 12) {
392 dp
= (struct ext4_dir_entry
*)(((void *)dirent
) + 12);
393 if (le16_to_cpu(dp
->rec_len
) !=
394 EXT4_BLOCK_SIZE(inode
->i_sb
) - 12)
396 root
= (struct dx_root_info
*)(((void *)dp
+ 12));
397 if (root
->reserved_zero
||
398 root
->info_length
!= sizeof(struct dx_root_info
))
405 *offset
= count_offset
;
406 return (struct dx_countlimit
*)(((void *)dirent
) + count_offset
);
409 static __le32
ext4_dx_csum(struct inode
*inode
, struct ext4_dir_entry
*dirent
,
410 int count_offset
, int count
, struct dx_tail
*t
)
412 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
413 struct ext4_inode_info
*ei
= EXT4_I(inode
);
418 size
= count_offset
+ (count
* sizeof(struct dx_entry
));
419 save_csum
= t
->dt_checksum
;
421 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
422 csum
= ext4_chksum(sbi
, csum
, (__u8
*)t
, sizeof(struct dx_tail
));
423 t
->dt_checksum
= save_csum
;
425 return cpu_to_le32(csum
);
428 static int ext4_dx_csum_verify(struct inode
*inode
,
429 struct ext4_dir_entry
*dirent
)
431 struct dx_countlimit
*c
;
433 int count_offset
, limit
, count
;
435 if (!ext4_has_metadata_csum(inode
->i_sb
))
438 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
440 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
443 limit
= le16_to_cpu(c
->limit
);
444 count
= le16_to_cpu(c
->count
);
445 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
446 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
447 warn_no_space_for_csum(inode
);
450 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
452 if (t
->dt_checksum
!= ext4_dx_csum(inode
, dirent
, count_offset
,
458 static void ext4_dx_csum_set(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
460 struct dx_countlimit
*c
;
462 int count_offset
, limit
, count
;
464 if (!ext4_has_metadata_csum(inode
->i_sb
))
467 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
469 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
472 limit
= le16_to_cpu(c
->limit
);
473 count
= le16_to_cpu(c
->count
);
474 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
475 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
476 warn_no_space_for_csum(inode
);
479 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
481 t
->dt_checksum
= ext4_dx_csum(inode
, dirent
, count_offset
, count
, t
);
484 static inline int ext4_handle_dirty_dx_node(handle_t
*handle
,
486 struct buffer_head
*bh
)
488 ext4_dx_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
489 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
493 * p is at least 6 bytes before the end of page
495 static inline struct ext4_dir_entry_2
*
496 ext4_next_entry(struct ext4_dir_entry_2
*p
, unsigned long blocksize
)
498 return (struct ext4_dir_entry_2
*)((char *)p
+
499 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
503 * Future: use high four bits of block for coalesce-on-delete flags
504 * Mask them off for now.
507 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
509 return le32_to_cpu(entry
->block
) & 0x00ffffff;
512 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
514 entry
->block
= cpu_to_le32(value
);
517 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
519 return le32_to_cpu(entry
->hash
);
522 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
524 entry
->hash
= cpu_to_le32(value
);
527 static inline unsigned dx_get_count(struct dx_entry
*entries
)
529 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
532 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
534 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
537 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
539 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
542 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
544 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
547 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
549 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
550 EXT4_DIR_REC_LEN(2) - infosize
;
552 if (ext4_has_metadata_csum(dir
->i_sb
))
553 entry_space
-= sizeof(struct dx_tail
);
554 return entry_space
/ sizeof(struct dx_entry
);
557 static inline unsigned dx_node_limit(struct inode
*dir
)
559 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
561 if (ext4_has_metadata_csum(dir
->i_sb
))
562 entry_space
-= sizeof(struct dx_tail
);
563 return entry_space
/ sizeof(struct dx_entry
);
570 static void dx_show_index(char * label
, struct dx_entry
*entries
)
572 int i
, n
= dx_get_count (entries
);
573 printk(KERN_DEBUG
"%s index ", label
);
574 for (i
= 0; i
< n
; i
++) {
575 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
576 0, (unsigned long)dx_get_block(entries
+ i
));
588 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
589 int size
, int show_names
)
591 unsigned names
= 0, space
= 0;
592 char *base
= (char *) de
;
593 struct dx_hash_info h
= *hinfo
;
596 while ((char *) de
< base
+ size
)
602 int len
= de
->name_len
;
603 char *name
= de
->name
;
604 while (len
--) printk("%c", *name
++);
605 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
606 printk(":%x.%u ", h
.hash
,
607 (unsigned) ((char *) de
- base
));
609 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
612 de
= ext4_next_entry(de
, size
);
614 printk("(%i)\n", names
);
615 return (struct stats
) { names
, space
, 1 };
618 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
619 struct dx_entry
*entries
, int levels
)
621 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
622 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
624 struct buffer_head
*bh
;
626 printk("%i indexed blocks...\n", count
);
627 for (i
= 0; i
< count
; i
++, entries
++)
629 ext4_lblk_t block
= dx_get_block(entries
);
630 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
631 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
633 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
634 bh
= ext4_bread(NULL
,dir
, block
, 0);
635 if (!bh
|| IS_ERR(bh
))
638 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
639 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
640 names
+= stats
.names
;
641 space
+= stats
.space
;
642 bcount
+= stats
.bcount
;
646 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
647 levels
? "" : " ", names
, space
/bcount
,
648 (space
/bcount
)*100/blocksize
);
649 return (struct stats
) { names
, space
, bcount
};
651 #endif /* DX_DEBUG */
654 * Probe for a directory leaf block to search.
656 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
657 * error in the directory index, and the caller should fall back to
658 * searching the directory normally. The callers of dx_probe **MUST**
659 * check for this error code, and make sure it never gets reflected
662 static struct dx_frame
*
663 dx_probe(const struct qstr
*d_name
, struct inode
*dir
,
664 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
)
666 unsigned count
, indirect
;
667 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
668 struct dx_root
*root
;
669 struct dx_frame
*frame
= frame_in
;
670 struct dx_frame
*ret_err
= ERR_PTR(ERR_BAD_DX_DIR
);
673 frame
->bh
= ext4_read_dirblock(dir
, 0, INDEX
);
674 if (IS_ERR(frame
->bh
))
675 return (struct dx_frame
*) frame
->bh
;
677 root
= (struct dx_root
*) frame
->bh
->b_data
;
678 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
679 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
680 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
681 ext4_warning(dir
->i_sb
, "Unrecognised inode hash code %d",
682 root
->info
.hash_version
);
685 hinfo
->hash_version
= root
->info
.hash_version
;
686 if (hinfo
->hash_version
<= DX_HASH_TEA
)
687 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
688 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
690 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
693 if (root
->info
.unused_flags
& 1) {
694 ext4_warning(dir
->i_sb
, "Unimplemented inode hash flags: %#06x",
695 root
->info
.unused_flags
);
699 if ((indirect
= root
->info
.indirect_levels
) > 1) {
700 ext4_warning(dir
->i_sb
, "Unimplemented inode hash depth: %#06x",
701 root
->info
.indirect_levels
);
705 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
706 root
->info
.info_length
);
708 if (dx_get_limit(entries
) != dx_root_limit(dir
,
709 root
->info
.info_length
)) {
710 ext4_warning(dir
->i_sb
, "dx entry: limit != root limit");
714 dxtrace(printk("Look up %x", hash
));
716 count
= dx_get_count(entries
);
717 if (!count
|| count
> dx_get_limit(entries
)) {
718 ext4_warning(dir
->i_sb
,
719 "dx entry: no count or count > limit");
724 q
= entries
+ count
- 1;
727 dxtrace(printk("."));
728 if (dx_get_hash(m
) > hash
)
734 if (0) { // linear search cross check
735 unsigned n
= count
- 1;
739 dxtrace(printk(","));
740 if (dx_get_hash(++at
) > hash
)
746 assert (at
== p
- 1);
750 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
751 frame
->entries
= entries
;
756 frame
->bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
757 if (IS_ERR(frame
->bh
)) {
758 ret_err
= (struct dx_frame
*) frame
->bh
;
762 entries
= ((struct dx_node
*) frame
->bh
->b_data
)->entries
;
764 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
765 ext4_warning(dir
->i_sb
,
766 "dx entry: limit != node limit");
771 while (frame
>= frame_in
) {
775 if (ret_err
== ERR_PTR(ERR_BAD_DX_DIR
))
776 ext4_warning(dir
->i_sb
,
777 "Corrupt dir inode %lu, running e2fsck is "
778 "recommended.", dir
->i_ino
);
782 static void dx_release (struct dx_frame
*frames
)
784 if (frames
[0].bh
== NULL
)
787 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
788 brelse(frames
[1].bh
);
789 brelse(frames
[0].bh
);
793 * This function increments the frame pointer to search the next leaf
794 * block, and reads in the necessary intervening nodes if the search
795 * should be necessary. Whether or not the search is necessary is
796 * controlled by the hash parameter. If the hash value is even, then
797 * the search is only continued if the next block starts with that
798 * hash value. This is used if we are searching for a specific file.
800 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
802 * This function returns 1 if the caller should continue to search,
803 * or 0 if it should not. If there is an error reading one of the
804 * index blocks, it will a negative error code.
806 * If start_hash is non-null, it will be filled in with the starting
807 * hash of the next page.
809 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
810 struct dx_frame
*frame
,
811 struct dx_frame
*frames
,
815 struct buffer_head
*bh
;
821 * Find the next leaf page by incrementing the frame pointer.
822 * If we run out of entries in the interior node, loop around and
823 * increment pointer in the parent node. When we break out of
824 * this loop, num_frames indicates the number of interior
825 * nodes need to be read.
828 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
837 * If the hash is 1, then continue only if the next page has a
838 * continuation hash of any value. This is used for readdir
839 * handling. Otherwise, check to see if the hash matches the
840 * desired contiuation hash. If it doesn't, return since
841 * there's no point to read in the successive index pages.
843 bhash
= dx_get_hash(p
->at
);
846 if ((hash
& 1) == 0) {
847 if ((bhash
& ~1) != hash
)
851 * If the hash is HASH_NB_ALWAYS, we always go to the next
852 * block so no check is necessary
854 while (num_frames
--) {
855 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
861 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
868 * This function fills a red-black tree with information from a
869 * directory block. It returns the number directory entries loaded
870 * into the tree. If there is an error it is returned in err.
872 static int htree_dirblock_to_tree(struct file
*dir_file
,
873 struct inode
*dir
, ext4_lblk_t block
,
874 struct dx_hash_info
*hinfo
,
875 __u32 start_hash
, __u32 start_minor_hash
)
877 struct buffer_head
*bh
;
878 struct ext4_dir_entry_2
*de
, *top
;
879 int err
= 0, count
= 0;
881 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
882 (unsigned long)block
));
883 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
887 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
888 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
889 dir
->i_sb
->s_blocksize
-
890 EXT4_DIR_REC_LEN(0));
891 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
892 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
893 bh
->b_data
, bh
->b_size
,
894 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
895 + ((char *)de
- bh
->b_data
))) {
896 /* silently ignore the rest of the block */
899 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
900 if ((hinfo
->hash
< start_hash
) ||
901 ((hinfo
->hash
== start_hash
) &&
902 (hinfo
->minor_hash
< start_minor_hash
)))
906 if ((err
= ext4_htree_store_dirent(dir_file
,
907 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
919 * This function fills a red-black tree with information from a
920 * directory. We start scanning the directory in hash order, starting
921 * at start_hash and start_minor_hash.
923 * This function returns the number of entries inserted into the tree,
924 * or a negative error code.
926 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
927 __u32 start_minor_hash
, __u32
*next_hash
)
929 struct dx_hash_info hinfo
;
930 struct ext4_dir_entry_2
*de
;
931 struct dx_frame frames
[2], *frame
;
938 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
939 start_hash
, start_minor_hash
));
940 dir
= file_inode(dir_file
);
941 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
942 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
943 if (hinfo
.hash_version
<= DX_HASH_TEA
)
944 hinfo
.hash_version
+=
945 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
946 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
947 if (ext4_has_inline_data(dir
)) {
948 int has_inline_data
= 1;
949 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
953 if (has_inline_data
) {
958 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
959 start_hash
, start_minor_hash
);
963 hinfo
.hash
= start_hash
;
964 hinfo
.minor_hash
= 0;
965 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
);
967 return PTR_ERR(frame
);
969 /* Add '.' and '..' from the htree header */
970 if (!start_hash
&& !start_minor_hash
) {
971 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
972 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
976 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
977 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
978 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
979 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
985 block
= dx_get_block(frame
->at
);
986 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
987 start_hash
, start_minor_hash
);
994 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
995 frame
, frames
, &hashval
);
996 *next_hash
= hashval
;
1002 * Stop if: (a) there are no more entries, or
1003 * (b) we have inserted at least one entry and the
1004 * next hash value is not a continuation
1007 (count
&& ((hashval
& 1) == 0)))
1011 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1012 "next hash: %x\n", count
, *next_hash
));
1019 static inline int search_dirblock(struct buffer_head
*bh
,
1021 const struct qstr
*d_name
,
1022 unsigned int offset
,
1023 struct ext4_dir_entry_2
**res_dir
)
1025 return search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1026 d_name
, offset
, res_dir
);
1030 * Directory block splitting, compacting
1034 * Create map of hash values, offsets, and sizes, stored at end of block.
1035 * Returns number of entries mapped.
1037 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
1038 struct dx_hash_info
*hinfo
,
1039 struct dx_map_entry
*map_tail
)
1042 char *base
= (char *) de
;
1043 struct dx_hash_info h
= *hinfo
;
1045 while ((char *) de
< base
+ blocksize
) {
1046 if (de
->name_len
&& de
->inode
) {
1047 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1049 map_tail
->hash
= h
.hash
;
1050 map_tail
->offs
= ((char *) de
- base
)>>2;
1051 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1055 /* XXX: do we need to check rec_len == 0 case? -Chris */
1056 de
= ext4_next_entry(de
, blocksize
);
1061 /* Sort map by hash value */
1062 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1064 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1066 /* Combsort until bubble sort doesn't suck */
1068 count
= count
*10/13;
1069 if (count
- 9 < 2) /* 9, 10 -> 11 */
1071 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1072 if (p
->hash
< q
->hash
)
1075 /* Garden variety bubble sort */
1080 if (q
[1].hash
>= q
[0].hash
)
1088 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1090 struct dx_entry
*entries
= frame
->entries
;
1091 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1092 int count
= dx_get_count(entries
);
1094 assert(count
< dx_get_limit(entries
));
1095 assert(old
< entries
+ count
);
1096 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1097 dx_set_hash(new, hash
);
1098 dx_set_block(new, block
);
1099 dx_set_count(entries
, count
+ 1);
1103 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1105 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1106 * `de != NULL' is guaranteed by caller.
1108 static inline int ext4_match (int len
, const char * const name
,
1109 struct ext4_dir_entry_2
* de
)
1111 if (len
!= de
->name_len
)
1115 return !memcmp(name
, de
->name
, len
);
1119 * Returns 0 if not found, -1 on failure, and 1 on success
1121 int search_dir(struct buffer_head
*bh
,
1125 const struct qstr
*d_name
,
1126 unsigned int offset
,
1127 struct ext4_dir_entry_2
**res_dir
)
1129 struct ext4_dir_entry_2
* de
;
1132 const char *name
= d_name
->name
;
1133 int namelen
= d_name
->len
;
1135 de
= (struct ext4_dir_entry_2
*)search_buf
;
1136 dlimit
= search_buf
+ buf_size
;
1137 while ((char *) de
< dlimit
) {
1138 /* this code is executed quadratically often */
1139 /* do minimal checking `by hand' */
1141 if ((char *) de
+ namelen
<= dlimit
&&
1142 ext4_match (namelen
, name
, de
)) {
1143 /* found a match - just to be sure, do a full check */
1144 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, bh
->b_data
,
1145 bh
->b_size
, offset
))
1150 /* prevent looping on a bad block */
1151 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1152 dir
->i_sb
->s_blocksize
);
1156 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1161 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1162 struct ext4_dir_entry
*de
)
1164 struct super_block
*sb
= dir
->i_sb
;
1170 if (de
->inode
== 0 &&
1171 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1180 * finds an entry in the specified directory with the wanted name. It
1181 * returns the cache buffer in which the entry was found, and the entry
1182 * itself (as a parameter - res_dir). It does NOT read the inode of the
1183 * entry - you'll have to do that yourself if you want to.
1185 * The returned buffer_head has ->b_count elevated. The caller is expected
1186 * to brelse() it when appropriate.
1188 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1189 const struct qstr
*d_name
,
1190 struct ext4_dir_entry_2
**res_dir
,
1193 struct super_block
*sb
;
1194 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1195 struct buffer_head
*bh
, *ret
= NULL
;
1196 ext4_lblk_t start
, block
, b
;
1197 const u8
*name
= d_name
->name
;
1198 int ra_max
= 0; /* Number of bh's in the readahead
1200 int ra_ptr
= 0; /* Current index into readahead
1203 ext4_lblk_t nblocks
;
1208 namelen
= d_name
->len
;
1209 if (namelen
> EXT4_NAME_LEN
)
1212 if (ext4_has_inline_data(dir
)) {
1213 int has_inline_data
= 1;
1214 ret
= ext4_find_inline_entry(dir
, d_name
, res_dir
,
1216 if (has_inline_data
) {
1223 if ((namelen
<= 2) && (name
[0] == '.') &&
1224 (name
[1] == '.' || name
[1] == '\0')) {
1226 * "." or ".." will only be in the first block
1227 * NFS may look up ".."; "." should be handled by the VFS
1234 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
);
1236 * On success, or if the error was file not found,
1237 * return. Otherwise, fall back to doing a search the
1238 * old fashioned way.
1240 if (!IS_ERR(bh
) || PTR_ERR(bh
) != ERR_BAD_DX_DIR
)
1242 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1245 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1246 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1247 if (start
>= nblocks
)
1253 * We deal with the read-ahead logic here.
1255 if (ra_ptr
>= ra_max
) {
1256 /* Refill the readahead buffer */
1259 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1261 * Terminate if we reach the end of the
1262 * directory and must wrap, or if our
1263 * search has finished at this block.
1265 if (b
>= nblocks
|| (num
&& block
== start
)) {
1266 bh_use
[ra_max
] = NULL
;
1270 bh
= ext4_getblk(NULL
, dir
, b
++, 0);
1271 if (unlikely(IS_ERR(bh
))) {
1276 bh_use
[ra_max
] = bh
;
1278 ll_rw_block(READ
| REQ_META
| REQ_PRIO
,
1282 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1285 if (!buffer_uptodate(bh
)) {
1286 /* read error, skip block & hope for the best */
1287 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1288 (unsigned long) block
);
1292 if (!buffer_verified(bh
) &&
1293 !is_dx_internal_node(dir
, block
,
1294 (struct ext4_dir_entry
*)bh
->b_data
) &&
1295 !ext4_dirent_csum_verify(dir
,
1296 (struct ext4_dir_entry
*)bh
->b_data
)) {
1297 EXT4_ERROR_INODE(dir
, "checksumming directory "
1298 "block %lu", (unsigned long)block
);
1302 set_buffer_verified(bh
);
1303 i
= search_dirblock(bh
, dir
, d_name
,
1304 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1306 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1308 goto cleanup_and_exit
;
1312 goto cleanup_and_exit
;
1315 if (++block
>= nblocks
)
1317 } while (block
!= start
);
1320 * If the directory has grown while we were searching, then
1321 * search the last part of the directory before giving up.
1324 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1325 if (block
< nblocks
) {
1331 /* Clean up the read-ahead blocks */
1332 for (; ra_ptr
< ra_max
; ra_ptr
++)
1333 brelse(bh_use
[ra_ptr
]);
1337 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
1338 struct ext4_dir_entry_2
**res_dir
)
1340 struct super_block
* sb
= dir
->i_sb
;
1341 struct dx_hash_info hinfo
;
1342 struct dx_frame frames
[2], *frame
;
1343 struct buffer_head
*bh
;
1347 frame
= dx_probe(d_name
, dir
, &hinfo
, frames
);
1349 return (struct buffer_head
*) frame
;
1351 block
= dx_get_block(frame
->at
);
1352 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1356 retval
= search_dirblock(bh
, dir
, d_name
,
1357 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1363 bh
= ERR_PTR(ERR_BAD_DX_DIR
);
1367 /* Check to see if we should continue to search */
1368 retval
= ext4_htree_next_block(dir
, hinfo
.hash
, frame
,
1372 "error %d reading index page in directory #%lu",
1373 retval
, dir
->i_ino
);
1374 bh
= ERR_PTR(retval
);
1377 } while (retval
== 1);
1381 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1387 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1389 struct inode
*inode
;
1390 struct ext4_dir_entry_2
*de
;
1391 struct buffer_head
*bh
;
1393 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1394 return ERR_PTR(-ENAMETOOLONG
);
1396 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1398 return (struct dentry
*) bh
;
1401 __u32 ino
= le32_to_cpu(de
->inode
);
1403 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1404 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1405 return ERR_PTR(-EIO
);
1407 if (unlikely(ino
== dir
->i_ino
)) {
1408 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1410 return ERR_PTR(-EIO
);
1412 inode
= ext4_iget_normal(dir
->i_sb
, ino
);
1413 if (inode
== ERR_PTR(-ESTALE
)) {
1414 EXT4_ERROR_INODE(dir
,
1415 "deleted inode referenced: %u",
1417 return ERR_PTR(-EIO
);
1420 return d_splice_alias(inode
, dentry
);
1424 struct dentry
*ext4_get_parent(struct dentry
*child
)
1427 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1428 struct ext4_dir_entry_2
* de
;
1429 struct buffer_head
*bh
;
1431 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
, NULL
);
1433 return (struct dentry
*) bh
;
1435 return ERR_PTR(-ENOENT
);
1436 ino
= le32_to_cpu(de
->inode
);
1439 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1440 EXT4_ERROR_INODE(child
->d_inode
,
1441 "bad parent inode number: %u", ino
);
1442 return ERR_PTR(-EIO
);
1445 return d_obtain_alias(ext4_iget_normal(child
->d_inode
->i_sb
, ino
));
1449 * Move count entries from end of map between two memory locations.
1450 * Returns pointer to last entry moved.
1452 static struct ext4_dir_entry_2
*
1453 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1456 unsigned rec_len
= 0;
1459 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1460 (from
+ (map
->offs
<<2));
1461 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1462 memcpy (to
, de
, rec_len
);
1463 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1464 ext4_rec_len_to_disk(rec_len
, blocksize
);
1469 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1473 * Compact each dir entry in the range to the minimal rec_len.
1474 * Returns pointer to last entry in range.
1476 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1478 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1479 unsigned rec_len
= 0;
1482 while ((char*)de
< base
+ blocksize
) {
1483 next
= ext4_next_entry(de
, blocksize
);
1484 if (de
->inode
&& de
->name_len
) {
1485 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1487 memmove(to
, de
, rec_len
);
1488 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1490 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1498 * Split a full leaf block to make room for a new dir entry.
1499 * Allocate a new block, and move entries so that they are approx. equally full.
1500 * Returns pointer to de in block into which the new entry will be inserted.
1502 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1503 struct buffer_head
**bh
,struct dx_frame
*frame
,
1504 struct dx_hash_info
*hinfo
)
1506 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1507 unsigned count
, continued
;
1508 struct buffer_head
*bh2
;
1509 ext4_lblk_t newblock
;
1511 struct dx_map_entry
*map
;
1512 char *data1
= (*bh
)->b_data
, *data2
;
1513 unsigned split
, move
, size
;
1514 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1515 struct ext4_dir_entry_tail
*t
;
1519 if (ext4_has_metadata_csum(dir
->i_sb
))
1520 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1522 bh2
= ext4_append(handle
, dir
, &newblock
);
1526 return (struct ext4_dir_entry_2
*) bh2
;
1529 BUFFER_TRACE(*bh
, "get_write_access");
1530 err
= ext4_journal_get_write_access(handle
, *bh
);
1534 BUFFER_TRACE(frame
->bh
, "get_write_access");
1535 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1539 data2
= bh2
->b_data
;
1541 /* create map in the end of data2 block */
1542 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1543 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1544 blocksize
, hinfo
, map
);
1546 dx_sort_map(map
, count
);
1547 /* Split the existing block in the middle, size-wise */
1550 for (i
= count
-1; i
>= 0; i
--) {
1551 /* is more than half of this entry in 2nd half of the block? */
1552 if (size
+ map
[i
].size
/2 > blocksize
/2)
1554 size
+= map
[i
].size
;
1557 /* map index at which we will split */
1558 split
= count
- move
;
1559 hash2
= map
[split
].hash
;
1560 continued
= hash2
== map
[split
- 1].hash
;
1561 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1562 (unsigned long)dx_get_block(frame
->at
),
1563 hash2
, split
, count
-split
));
1565 /* Fancy dance to stay within two buffers */
1566 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
, blocksize
);
1567 de
= dx_pack_dirents(data1
, blocksize
);
1568 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1571 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1575 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1576 initialize_dirent_tail(t
, blocksize
);
1578 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1579 initialize_dirent_tail(t
, blocksize
);
1582 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1583 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1585 /* Which block gets the new entry? */
1586 if (hinfo
->hash
>= hash2
) {
1590 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1591 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1594 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1598 dxtrace(dx_show_index("frame", frame
->entries
));
1605 ext4_std_error(dir
->i_sb
, err
);
1606 return ERR_PTR(err
);
1609 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1610 struct buffer_head
*bh
,
1611 void *buf
, int buf_size
,
1612 const char *name
, int namelen
,
1613 struct ext4_dir_entry_2
**dest_de
)
1615 struct ext4_dir_entry_2
*de
;
1616 unsigned short reclen
= EXT4_DIR_REC_LEN(namelen
);
1618 unsigned int offset
= 0;
1621 de
= (struct ext4_dir_entry_2
*)buf
;
1622 top
= buf
+ buf_size
- reclen
;
1623 while ((char *) de
<= top
) {
1624 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1625 buf
, buf_size
, offset
))
1627 if (ext4_match(namelen
, name
, de
))
1629 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1630 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1631 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1633 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1636 if ((char *) de
> top
)
1643 void ext4_insert_dentry(struct inode
*inode
,
1644 struct ext4_dir_entry_2
*de
,
1646 const char *name
, int namelen
)
1651 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1652 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1654 struct ext4_dir_entry_2
*de1
=
1655 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1656 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1657 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1660 de
->file_type
= EXT4_FT_UNKNOWN
;
1661 de
->inode
= cpu_to_le32(inode
->i_ino
);
1662 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1663 de
->name_len
= namelen
;
1664 memcpy(de
->name
, name
, namelen
);
1667 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1668 * it points to a directory entry which is guaranteed to be large
1669 * enough for new directory entry. If de is NULL, then
1670 * add_dirent_to_buf will attempt search the directory block for
1671 * space. It will return -ENOSPC if no space is available, and -EIO
1672 * and -EEXIST if directory entry already exists.
1674 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1675 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1676 struct buffer_head
*bh
)
1678 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1679 const char *name
= dentry
->d_name
.name
;
1680 int namelen
= dentry
->d_name
.len
;
1681 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1685 if (ext4_has_metadata_csum(inode
->i_sb
))
1686 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1689 err
= ext4_find_dest_de(dir
, inode
,
1690 bh
, bh
->b_data
, blocksize
- csum_size
,
1691 name
, namelen
, &de
);
1695 BUFFER_TRACE(bh
, "get_write_access");
1696 err
= ext4_journal_get_write_access(handle
, bh
);
1698 ext4_std_error(dir
->i_sb
, err
);
1702 /* By now the buffer is marked for journaling */
1703 ext4_insert_dentry(inode
, de
, blocksize
, name
, namelen
);
1706 * XXX shouldn't update any times until successful
1707 * completion of syscall, but too many callers depend
1710 * XXX similarly, too many callers depend on
1711 * ext4_new_inode() setting the times, but error
1712 * recovery deletes the inode, so the worst that can
1713 * happen is that the times are slightly out of date
1714 * and/or different from the directory change time.
1716 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1717 ext4_update_dx_flag(dir
);
1719 ext4_mark_inode_dirty(handle
, dir
);
1720 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1721 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1723 ext4_std_error(dir
->i_sb
, err
);
1728 * This converts a one block unindexed directory to a 3 block indexed
1729 * directory, and adds the dentry to the indexed directory.
1731 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1732 struct inode
*inode
, struct buffer_head
*bh
)
1734 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1735 const char *name
= dentry
->d_name
.name
;
1736 int namelen
= dentry
->d_name
.len
;
1737 struct buffer_head
*bh2
;
1738 struct dx_root
*root
;
1739 struct dx_frame frames
[2], *frame
;
1740 struct dx_entry
*entries
;
1741 struct ext4_dir_entry_2
*de
, *de2
;
1742 struct ext4_dir_entry_tail
*t
;
1747 struct dx_hash_info hinfo
;
1749 struct fake_dirent
*fde
;
1752 if (ext4_has_metadata_csum(inode
->i_sb
))
1753 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1755 blocksize
= dir
->i_sb
->s_blocksize
;
1756 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1757 BUFFER_TRACE(bh
, "get_write_access");
1758 retval
= ext4_journal_get_write_access(handle
, bh
);
1760 ext4_std_error(dir
->i_sb
, retval
);
1764 root
= (struct dx_root
*) bh
->b_data
;
1766 /* The 0th block becomes the root, move the dirents out */
1767 fde
= &root
->dotdot
;
1768 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1769 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1770 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1771 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1775 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1777 /* Allocate new block for the 0th block's dirents */
1778 bh2
= ext4_append(handle
, dir
, &block
);
1781 return PTR_ERR(bh2
);
1783 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1784 data1
= bh2
->b_data
;
1786 memcpy (data1
, de
, len
);
1787 de
= (struct ext4_dir_entry_2
*) data1
;
1789 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1791 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1796 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1797 initialize_dirent_tail(t
, blocksize
);
1800 /* Initialize the root; the dot dirents already exist */
1801 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1802 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1804 memset (&root
->info
, 0, sizeof(root
->info
));
1805 root
->info
.info_length
= sizeof(root
->info
);
1806 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1807 entries
= root
->entries
;
1808 dx_set_block(entries
, 1);
1809 dx_set_count(entries
, 1);
1810 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1812 /* Initialize as for dx_probe */
1813 hinfo
.hash_version
= root
->info
.hash_version
;
1814 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1815 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1816 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1817 ext4fs_dirhash(name
, namelen
, &hinfo
);
1818 memset(frames
, 0, sizeof(frames
));
1820 frame
->entries
= entries
;
1821 frame
->at
= entries
;
1825 retval
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1828 retval
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1832 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
);
1834 retval
= PTR_ERR(de
);
1839 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1844 * Even if the block split failed, we have to properly write
1845 * out all the changes we did so far. Otherwise we can end up
1846 * with corrupted filesystem.
1848 ext4_mark_inode_dirty(handle
, dir
);
1856 * adds a file entry to the specified directory, using the same
1857 * semantics as ext4_find_entry(). It returns NULL if it failed.
1859 * NOTE!! The inode part of 'de' is left at 0 - which means you
1860 * may not sleep between calling this and putting something into
1861 * the entry, as someone else might have used it while you slept.
1863 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1864 struct inode
*inode
)
1866 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1867 struct buffer_head
*bh
;
1868 struct ext4_dir_entry_2
*de
;
1869 struct ext4_dir_entry_tail
*t
;
1870 struct super_block
*sb
;
1874 ext4_lblk_t block
, blocks
;
1877 if (ext4_has_metadata_csum(inode
->i_sb
))
1878 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1881 blocksize
= sb
->s_blocksize
;
1882 if (!dentry
->d_name
.len
)
1885 if (ext4_has_inline_data(dir
)) {
1886 retval
= ext4_try_add_inline_entry(handle
, dentry
, inode
);
1896 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1897 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1899 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
1901 ext4_mark_inode_dirty(handle
, dir
);
1903 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1904 for (block
= 0; block
< blocks
; block
++) {
1905 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1909 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1910 if (retval
!= -ENOSPC
) {
1915 if (blocks
== 1 && !dx_fallback
&&
1916 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1917 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1920 bh
= ext4_append(handle
, dir
, &block
);
1923 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1925 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
1928 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
1929 initialize_dirent_tail(t
, blocksize
);
1932 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1935 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
1940 * Returns 0 for success, or a negative error value
1942 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1943 struct inode
*inode
)
1945 struct dx_frame frames
[2], *frame
;
1946 struct dx_entry
*entries
, *at
;
1947 struct dx_hash_info hinfo
;
1948 struct buffer_head
*bh
;
1949 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1950 struct super_block
*sb
= dir
->i_sb
;
1951 struct ext4_dir_entry_2
*de
;
1954 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
);
1956 return PTR_ERR(frame
);
1957 entries
= frame
->entries
;
1959 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
1966 BUFFER_TRACE(bh
, "get_write_access");
1967 err
= ext4_journal_get_write_access(handle
, bh
);
1971 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1975 /* Block full, should compress but for now just split */
1976 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
1977 dx_get_count(entries
), dx_get_limit(entries
)));
1978 /* Need to split index? */
1979 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1980 ext4_lblk_t newblock
;
1981 unsigned icount
= dx_get_count(entries
);
1982 int levels
= frame
- frames
;
1983 struct dx_entry
*entries2
;
1984 struct dx_node
*node2
;
1985 struct buffer_head
*bh2
;
1987 if (levels
&& (dx_get_count(frames
->entries
) ==
1988 dx_get_limit(frames
->entries
))) {
1989 ext4_warning(sb
, "Directory index full!");
1993 bh2
= ext4_append(handle
, dir
, &newblock
);
1998 node2
= (struct dx_node
*)(bh2
->b_data
);
1999 entries2
= node2
->entries
;
2000 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2001 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2003 BUFFER_TRACE(frame
->bh
, "get_write_access");
2004 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2008 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2009 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2010 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2013 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2014 err
= ext4_journal_get_write_access(handle
,
2019 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2020 icount2
* sizeof(struct dx_entry
));
2021 dx_set_count(entries
, icount1
);
2022 dx_set_count(entries2
, icount2
);
2023 dx_set_limit(entries2
, dx_node_limit(dir
));
2025 /* Which index block gets the new entry? */
2026 if (at
- entries
>= icount1
) {
2027 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2028 frame
->entries
= entries
= entries2
;
2029 swap(frame
->bh
, bh2
);
2031 dx_insert_block(frames
+ 0, hash2
, newblock
);
2032 dxtrace(dx_show_index("node", frames
[1].entries
));
2033 dxtrace(dx_show_index("node",
2034 ((struct dx_node
*) bh2
->b_data
)->entries
));
2035 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2040 dxtrace(printk(KERN_DEBUG
2041 "Creating second level index...\n"));
2042 memcpy((char *) entries2
, (char *) entries
,
2043 icount
* sizeof(struct dx_entry
));
2044 dx_set_limit(entries2
, dx_node_limit(dir
));
2047 dx_set_count(entries
, 1);
2048 dx_set_block(entries
+ 0, newblock
);
2049 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2051 /* Add new access path frame */
2053 frame
->at
= at
= at
- entries
+ entries2
;
2054 frame
->entries
= entries
= entries2
;
2056 err
= ext4_journal_get_write_access(handle
,
2061 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2063 ext4_std_error(inode
->i_sb
, err
);
2067 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
);
2072 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
2076 ext4_std_error(dir
->i_sb
, err
);
2084 * ext4_generic_delete_entry deletes a directory entry by merging it
2085 * with the previous entry
2087 int ext4_generic_delete_entry(handle_t
*handle
,
2089 struct ext4_dir_entry_2
*de_del
,
2090 struct buffer_head
*bh
,
2095 struct ext4_dir_entry_2
*de
, *pde
;
2096 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2101 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2102 while (i
< buf_size
- csum_size
) {
2103 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2104 bh
->b_data
, bh
->b_size
, i
))
2108 pde
->rec_len
= ext4_rec_len_to_disk(
2109 ext4_rec_len_from_disk(pde
->rec_len
,
2111 ext4_rec_len_from_disk(de
->rec_len
,
2119 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2121 de
= ext4_next_entry(de
, blocksize
);
2126 static int ext4_delete_entry(handle_t
*handle
,
2128 struct ext4_dir_entry_2
*de_del
,
2129 struct buffer_head
*bh
)
2131 int err
, csum_size
= 0;
2133 if (ext4_has_inline_data(dir
)) {
2134 int has_inline_data
= 1;
2135 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2137 if (has_inline_data
)
2141 if (ext4_has_metadata_csum(dir
->i_sb
))
2142 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2144 BUFFER_TRACE(bh
, "get_write_access");
2145 err
= ext4_journal_get_write_access(handle
, bh
);
2149 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2151 dir
->i_sb
->s_blocksize
, csum_size
);
2155 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2156 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2163 ext4_std_error(dir
->i_sb
, err
);
2168 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2169 * since this indicates that nlinks count was previously 1.
2171 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2174 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2175 /* limit is 16-bit i_links_count */
2176 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2177 set_nlink(inode
, 1);
2178 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
2179 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
2185 * If a directory had nlink == 1, then we should let it be 1. This indicates
2186 * directory has >EXT4_LINK_MAX subdirs.
2188 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2190 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2195 static int ext4_add_nondir(handle_t
*handle
,
2196 struct dentry
*dentry
, struct inode
*inode
)
2198 int err
= ext4_add_entry(handle
, dentry
, inode
);
2200 ext4_mark_inode_dirty(handle
, inode
);
2201 unlock_new_inode(inode
);
2202 d_instantiate(dentry
, inode
);
2206 unlock_new_inode(inode
);
2212 * By the time this is called, we already have created
2213 * the directory cache entry for the new file, but it
2214 * is so far negative - it has no inode.
2216 * If the create succeeds, we fill in the inode information
2217 * with d_instantiate().
2219 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2223 struct inode
*inode
;
2224 int err
, credits
, retries
= 0;
2226 dquot_initialize(dir
);
2228 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2229 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2231 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2232 NULL
, EXT4_HT_DIR
, credits
);
2233 handle
= ext4_journal_current_handle();
2234 err
= PTR_ERR(inode
);
2235 if (!IS_ERR(inode
)) {
2236 inode
->i_op
= &ext4_file_inode_operations
;
2237 if (test_opt(inode
->i_sb
, DAX
))
2238 inode
->i_fop
= &ext4_dax_file_operations
;
2240 inode
->i_fop
= &ext4_file_operations
;
2241 ext4_set_aops(inode
);
2242 err
= ext4_add_nondir(handle
, dentry
, inode
);
2243 if (!err
&& IS_DIRSYNC(dir
))
2244 ext4_handle_sync(handle
);
2247 ext4_journal_stop(handle
);
2248 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2253 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2254 umode_t mode
, dev_t rdev
)
2257 struct inode
*inode
;
2258 int err
, credits
, retries
= 0;
2260 if (!new_valid_dev(rdev
))
2263 dquot_initialize(dir
);
2265 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2266 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2268 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2269 NULL
, EXT4_HT_DIR
, credits
);
2270 handle
= ext4_journal_current_handle();
2271 err
= PTR_ERR(inode
);
2272 if (!IS_ERR(inode
)) {
2273 init_special_inode(inode
, inode
->i_mode
, rdev
);
2274 inode
->i_op
= &ext4_special_inode_operations
;
2275 err
= ext4_add_nondir(handle
, dentry
, inode
);
2276 if (!err
&& IS_DIRSYNC(dir
))
2277 ext4_handle_sync(handle
);
2280 ext4_journal_stop(handle
);
2281 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2286 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2289 struct inode
*inode
;
2290 int err
, retries
= 0;
2292 dquot_initialize(dir
);
2295 inode
= ext4_new_inode_start_handle(dir
, mode
,
2298 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2299 4 + EXT4_XATTR_TRANS_BLOCKS
);
2300 handle
= ext4_journal_current_handle();
2301 err
= PTR_ERR(inode
);
2302 if (!IS_ERR(inode
)) {
2303 inode
->i_op
= &ext4_file_inode_operations
;
2304 if (test_opt(inode
->i_sb
, DAX
))
2305 inode
->i_fop
= &ext4_dax_file_operations
;
2307 inode
->i_fop
= &ext4_file_operations
;
2308 ext4_set_aops(inode
);
2309 d_tmpfile(dentry
, inode
);
2310 err
= ext4_orphan_add(handle
, inode
);
2312 goto err_unlock_inode
;
2313 mark_inode_dirty(inode
);
2314 unlock_new_inode(inode
);
2317 ext4_journal_stop(handle
);
2318 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2322 ext4_journal_stop(handle
);
2323 unlock_new_inode(inode
);
2327 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2328 struct ext4_dir_entry_2
*de
,
2329 int blocksize
, int csum_size
,
2330 unsigned int parent_ino
, int dotdot_real_len
)
2332 de
->inode
= cpu_to_le32(inode
->i_ino
);
2334 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2336 strcpy(de
->name
, ".");
2337 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2339 de
= ext4_next_entry(de
, blocksize
);
2340 de
->inode
= cpu_to_le32(parent_ino
);
2342 if (!dotdot_real_len
)
2343 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2344 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2347 de
->rec_len
= ext4_rec_len_to_disk(
2348 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2349 strcpy(de
->name
, "..");
2350 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2352 return ext4_next_entry(de
, blocksize
);
2355 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2356 struct inode
*inode
)
2358 struct buffer_head
*dir_block
= NULL
;
2359 struct ext4_dir_entry_2
*de
;
2360 struct ext4_dir_entry_tail
*t
;
2361 ext4_lblk_t block
= 0;
2362 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2366 if (ext4_has_metadata_csum(dir
->i_sb
))
2367 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2369 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2370 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2371 if (err
< 0 && err
!= -ENOSPC
)
2378 dir_block
= ext4_append(handle
, inode
, &block
);
2379 if (IS_ERR(dir_block
))
2380 return PTR_ERR(dir_block
);
2381 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2382 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2383 set_nlink(inode
, 2);
2385 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2386 initialize_dirent_tail(t
, blocksize
);
2389 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2390 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2393 set_buffer_verified(dir_block
);
2399 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2402 struct inode
*inode
;
2403 int err
, credits
, retries
= 0;
2405 if (EXT4_DIR_LINK_MAX(dir
))
2408 dquot_initialize(dir
);
2410 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2411 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2413 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2415 0, NULL
, EXT4_HT_DIR
, credits
);
2416 handle
= ext4_journal_current_handle();
2417 err
= PTR_ERR(inode
);
2421 inode
->i_op
= &ext4_dir_inode_operations
;
2422 inode
->i_fop
= &ext4_dir_operations
;
2423 err
= ext4_init_new_dir(handle
, dir
, inode
);
2425 goto out_clear_inode
;
2426 err
= ext4_mark_inode_dirty(handle
, inode
);
2428 err
= ext4_add_entry(handle
, dentry
, inode
);
2432 unlock_new_inode(inode
);
2433 ext4_mark_inode_dirty(handle
, inode
);
2437 ext4_inc_count(handle
, dir
);
2438 ext4_update_dx_flag(dir
);
2439 err
= ext4_mark_inode_dirty(handle
, dir
);
2441 goto out_clear_inode
;
2442 unlock_new_inode(inode
);
2443 d_instantiate(dentry
, inode
);
2444 if (IS_DIRSYNC(dir
))
2445 ext4_handle_sync(handle
);
2449 ext4_journal_stop(handle
);
2450 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2456 * routine to check that the specified directory is empty (for rmdir)
2458 static int empty_dir(struct inode
*inode
)
2460 unsigned int offset
;
2461 struct buffer_head
*bh
;
2462 struct ext4_dir_entry_2
*de
, *de1
;
2463 struct super_block
*sb
;
2466 if (ext4_has_inline_data(inode
)) {
2467 int has_inline_data
= 1;
2469 err
= empty_inline_dir(inode
, &has_inline_data
);
2470 if (has_inline_data
)
2475 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2476 EXT4_ERROR_INODE(inode
, "invalid size");
2479 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2483 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2484 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2485 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2486 !le32_to_cpu(de1
->inode
) ||
2487 strcmp(".", de
->name
) ||
2488 strcmp("..", de1
->name
)) {
2489 ext4_warning(inode
->i_sb
,
2490 "bad directory (dir #%lu) - no `.' or `..'",
2495 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2496 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2497 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2498 while (offset
< inode
->i_size
) {
2499 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2500 unsigned int lblock
;
2503 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2504 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2507 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2509 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2510 bh
->b_data
, bh
->b_size
, offset
)) {
2511 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2513 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2516 if (le32_to_cpu(de
->inode
)) {
2520 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2521 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2528 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2529 * such inodes, starting at the superblock, in case we crash before the
2530 * file is closed/deleted, or in case the inode truncate spans multiple
2531 * transactions and the last transaction is not recovered after a crash.
2533 * At filesystem recovery time, we walk this list deleting unlinked
2534 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2536 * Orphan list manipulation functions must be called under i_mutex unless
2537 * we are just creating the inode or deleting it.
2539 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2541 struct super_block
*sb
= inode
->i_sb
;
2542 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2543 struct ext4_iloc iloc
;
2547 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2550 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2551 !mutex_is_locked(&inode
->i_mutex
));
2553 * Exit early if inode already is on orphan list. This is a big speedup
2554 * since we don't have to contend on the global s_orphan_lock.
2556 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2560 * Orphan handling is only valid for files with data blocks
2561 * being truncated, or files being unlinked. Note that we either
2562 * hold i_mutex, or the inode can not be referenced from outside,
2563 * so i_nlink should not be bumped due to race
2565 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2566 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2568 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2569 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2573 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2577 mutex_lock(&sbi
->s_orphan_lock
);
2579 * Due to previous errors inode may be already a part of on-disk
2580 * orphan list. If so skip on-disk list modification.
2582 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2583 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2584 /* Insert this inode at the head of the on-disk orphan list */
2585 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2586 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2589 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2590 mutex_unlock(&sbi
->s_orphan_lock
);
2593 err
= ext4_handle_dirty_super(handle
, sb
);
2594 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2599 * We have to remove inode from in-memory list if
2600 * addition to on disk orphan list failed. Stray orphan
2601 * list entries can cause panics at unmount time.
2603 mutex_lock(&sbi
->s_orphan_lock
);
2604 list_del(&EXT4_I(inode
)->i_orphan
);
2605 mutex_unlock(&sbi
->s_orphan_lock
);
2608 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2609 jbd_debug(4, "orphan inode %lu will point to %d\n",
2610 inode
->i_ino
, NEXT_ORPHAN(inode
));
2612 ext4_std_error(sb
, err
);
2617 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2618 * of such inodes stored on disk, because it is finally being cleaned up.
2620 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2622 struct list_head
*prev
;
2623 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2624 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2626 struct ext4_iloc iloc
;
2629 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2632 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2633 !mutex_is_locked(&inode
->i_mutex
));
2634 /* Do this quick check before taking global s_orphan_lock. */
2635 if (list_empty(&ei
->i_orphan
))
2639 /* Grab inode buffer early before taking global s_orphan_lock */
2640 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2643 mutex_lock(&sbi
->s_orphan_lock
);
2644 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2646 prev
= ei
->i_orphan
.prev
;
2647 list_del_init(&ei
->i_orphan
);
2649 /* If we're on an error path, we may not have a valid
2650 * transaction handle with which to update the orphan list on
2651 * disk, but we still need to remove the inode from the linked
2652 * list in memory. */
2653 if (!handle
|| err
) {
2654 mutex_unlock(&sbi
->s_orphan_lock
);
2658 ino_next
= NEXT_ORPHAN(inode
);
2659 if (prev
== &sbi
->s_orphan
) {
2660 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2661 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2662 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2664 mutex_unlock(&sbi
->s_orphan_lock
);
2667 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2668 mutex_unlock(&sbi
->s_orphan_lock
);
2669 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2671 struct ext4_iloc iloc2
;
2672 struct inode
*i_prev
=
2673 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2675 jbd_debug(4, "orphan inode %lu will point to %u\n",
2676 i_prev
->i_ino
, ino_next
);
2677 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2679 mutex_unlock(&sbi
->s_orphan_lock
);
2682 NEXT_ORPHAN(i_prev
) = ino_next
;
2683 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2684 mutex_unlock(&sbi
->s_orphan_lock
);
2688 NEXT_ORPHAN(inode
) = 0;
2689 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2691 ext4_std_error(inode
->i_sb
, err
);
2699 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2702 struct inode
*inode
;
2703 struct buffer_head
*bh
;
2704 struct ext4_dir_entry_2
*de
;
2705 handle_t
*handle
= NULL
;
2707 /* Initialize quotas before so that eventual writes go in
2708 * separate transaction */
2709 dquot_initialize(dir
);
2710 dquot_initialize(dentry
->d_inode
);
2713 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2719 inode
= dentry
->d_inode
;
2722 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2725 retval
= -ENOTEMPTY
;
2726 if (!empty_dir(inode
))
2729 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2730 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2731 if (IS_ERR(handle
)) {
2732 retval
= PTR_ERR(handle
);
2737 if (IS_DIRSYNC(dir
))
2738 ext4_handle_sync(handle
);
2740 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2743 if (!EXT4_DIR_LINK_EMPTY(inode
))
2744 ext4_warning(inode
->i_sb
,
2745 "empty directory has too many links (%d)",
2749 /* There's no need to set i_disksize: the fact that i_nlink is
2750 * zero will ensure that the right thing happens during any
2753 ext4_orphan_add(handle
, inode
);
2754 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2755 ext4_mark_inode_dirty(handle
, inode
);
2756 ext4_dec_count(handle
, dir
);
2757 ext4_update_dx_flag(dir
);
2758 ext4_mark_inode_dirty(handle
, dir
);
2763 ext4_journal_stop(handle
);
2767 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2770 struct inode
*inode
;
2771 struct buffer_head
*bh
;
2772 struct ext4_dir_entry_2
*de
;
2773 handle_t
*handle
= NULL
;
2775 trace_ext4_unlink_enter(dir
, dentry
);
2776 /* Initialize quotas before so that eventual writes go
2777 * in separate transaction */
2778 dquot_initialize(dir
);
2779 dquot_initialize(dentry
->d_inode
);
2782 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2788 inode
= dentry
->d_inode
;
2791 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2794 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2795 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2796 if (IS_ERR(handle
)) {
2797 retval
= PTR_ERR(handle
);
2802 if (IS_DIRSYNC(dir
))
2803 ext4_handle_sync(handle
);
2805 if (!inode
->i_nlink
) {
2806 ext4_warning(inode
->i_sb
,
2807 "Deleting nonexistent file (%lu), %d",
2808 inode
->i_ino
, inode
->i_nlink
);
2809 set_nlink(inode
, 1);
2811 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2814 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2815 ext4_update_dx_flag(dir
);
2816 ext4_mark_inode_dirty(handle
, dir
);
2818 if (!inode
->i_nlink
)
2819 ext4_orphan_add(handle
, inode
);
2820 inode
->i_ctime
= ext4_current_time(inode
);
2821 ext4_mark_inode_dirty(handle
, inode
);
2826 ext4_journal_stop(handle
);
2827 trace_ext4_unlink_exit(dentry
, retval
);
2831 static int ext4_symlink(struct inode
*dir
,
2832 struct dentry
*dentry
, const char *symname
)
2835 struct inode
*inode
;
2836 int l
, err
, retries
= 0;
2839 l
= strlen(symname
)+1;
2840 if (l
> dir
->i_sb
->s_blocksize
)
2841 return -ENAMETOOLONG
;
2843 dquot_initialize(dir
);
2845 if (l
> EXT4_N_BLOCKS
* 4) {
2847 * For non-fast symlinks, we just allocate inode and put it on
2848 * orphan list in the first transaction => we need bitmap,
2849 * group descriptor, sb, inode block, quota blocks, and
2850 * possibly selinux xattr blocks.
2852 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2853 EXT4_XATTR_TRANS_BLOCKS
;
2856 * Fast symlink. We have to add entry to directory
2857 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2858 * allocate new inode (bitmap, group descriptor, inode block,
2859 * quota blocks, sb is already counted in previous macros).
2861 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2862 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
2865 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
2866 &dentry
->d_name
, 0, NULL
,
2867 EXT4_HT_DIR
, credits
);
2868 handle
= ext4_journal_current_handle();
2869 err
= PTR_ERR(inode
);
2873 if (l
> EXT4_N_BLOCKS
* 4) {
2874 inode
->i_op
= &ext4_symlink_inode_operations
;
2875 ext4_set_aops(inode
);
2877 * We cannot call page_symlink() with transaction started
2878 * because it calls into ext4_write_begin() which can wait
2879 * for transaction commit if we are running out of space
2880 * and thus we deadlock. So we have to stop transaction now
2881 * and restart it when symlink contents is written.
2883 * To keep fs consistent in case of crash, we have to put inode
2884 * to orphan list in the mean time.
2887 err
= ext4_orphan_add(handle
, inode
);
2888 ext4_journal_stop(handle
);
2890 goto err_drop_inode
;
2891 err
= __page_symlink(inode
, symname
, l
, 1);
2893 goto err_drop_inode
;
2895 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2896 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2898 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2899 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2900 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
2901 if (IS_ERR(handle
)) {
2902 err
= PTR_ERR(handle
);
2903 goto err_drop_inode
;
2905 set_nlink(inode
, 1);
2906 err
= ext4_orphan_del(handle
, inode
);
2908 ext4_journal_stop(handle
);
2910 goto err_drop_inode
;
2913 /* clear the extent format for fast symlink */
2914 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
2915 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2916 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2917 inode
->i_size
= l
-1;
2919 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2920 err
= ext4_add_nondir(handle
, dentry
, inode
);
2921 if (!err
&& IS_DIRSYNC(dir
))
2922 ext4_handle_sync(handle
);
2926 ext4_journal_stop(handle
);
2927 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2931 unlock_new_inode(inode
);
2936 static int ext4_link(struct dentry
*old_dentry
,
2937 struct inode
*dir
, struct dentry
*dentry
)
2940 struct inode
*inode
= old_dentry
->d_inode
;
2941 int err
, retries
= 0;
2943 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
2946 dquot_initialize(dir
);
2949 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2950 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2951 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
2953 return PTR_ERR(handle
);
2955 if (IS_DIRSYNC(dir
))
2956 ext4_handle_sync(handle
);
2958 inode
->i_ctime
= ext4_current_time(inode
);
2959 ext4_inc_count(handle
, inode
);
2962 err
= ext4_add_entry(handle
, dentry
, inode
);
2964 ext4_mark_inode_dirty(handle
, inode
);
2965 /* this can happen only for tmpfile being
2966 * linked the first time
2968 if (inode
->i_nlink
== 1)
2969 ext4_orphan_del(handle
, inode
);
2970 d_instantiate(dentry
, inode
);
2975 ext4_journal_stop(handle
);
2976 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2983 * Try to find buffer head where contains the parent block.
2984 * It should be the inode block if it is inlined or the 1st block
2985 * if it is a normal dir.
2987 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
2988 struct inode
*inode
,
2990 struct ext4_dir_entry_2
**parent_de
,
2993 struct buffer_head
*bh
;
2995 if (!ext4_has_inline_data(inode
)) {
2996 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2998 *retval
= PTR_ERR(bh
);
3001 *parent_de
= ext4_next_entry(
3002 (struct ext4_dir_entry_2
*)bh
->b_data
,
3003 inode
->i_sb
->s_blocksize
);
3008 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3011 struct ext4_renament
{
3013 struct dentry
*dentry
;
3014 struct inode
*inode
;
3016 int dir_nlink_delta
;
3018 /* entry for "dentry" */
3019 struct buffer_head
*bh
;
3020 struct ext4_dir_entry_2
*de
;
3023 /* entry for ".." in inode if it's a directory */
3024 struct buffer_head
*dir_bh
;
3025 struct ext4_dir_entry_2
*parent_de
;
3029 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3033 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3034 &retval
, &ent
->parent_de
,
3038 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3040 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3041 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3044 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3049 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3050 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3051 if (!ent
->dir_inlined
) {
3052 if (is_dx(ent
->inode
)) {
3053 retval
= ext4_handle_dirty_dx_node(handle
,
3057 retval
= ext4_handle_dirty_dirent_node(handle
,
3062 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3065 ext4_std_error(ent
->dir
->i_sb
, retval
);
3071 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3072 unsigned ino
, unsigned file_type
)
3076 BUFFER_TRACE(ent
->bh
, "get write access");
3077 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3080 ent
->de
->inode
= cpu_to_le32(ino
);
3081 if (EXT4_HAS_INCOMPAT_FEATURE(ent
->dir
->i_sb
,
3082 EXT4_FEATURE_INCOMPAT_FILETYPE
))
3083 ent
->de
->file_type
= file_type
;
3084 ent
->dir
->i_version
++;
3085 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3086 ext4_current_time(ent
->dir
);
3087 ext4_mark_inode_dirty(handle
, ent
->dir
);
3088 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3089 if (!ent
->inlined
) {
3090 retval
= ext4_handle_dirty_dirent_node(handle
,
3092 if (unlikely(retval
)) {
3093 ext4_std_error(ent
->dir
->i_sb
, retval
);
3103 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3104 const struct qstr
*d_name
)
3106 int retval
= -ENOENT
;
3107 struct buffer_head
*bh
;
3108 struct ext4_dir_entry_2
*de
;
3110 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3114 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3120 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3125 * ent->de could have moved from under us during htree split, so make
3126 * sure that we are deleting the right entry. We might also be pointing
3127 * to a stale entry in the unused part of ent->bh so just checking inum
3128 * and the name isn't enough.
3130 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3131 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3132 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3133 ent
->de
->name_len
) ||
3135 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3136 &ent
->dentry
->d_name
);
3138 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3139 if (retval
== -ENOENT
) {
3140 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3141 &ent
->dentry
->d_name
);
3146 ext4_warning(ent
->dir
->i_sb
,
3147 "Deleting old file (%lu), %d, error=%d",
3148 ent
->dir
->i_ino
, ent
->dir
->i_nlink
, retval
);
3152 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3154 if (ent
->dir_nlink_delta
) {
3155 if (ent
->dir_nlink_delta
== -1)
3156 ext4_dec_count(handle
, ent
->dir
);
3158 ext4_inc_count(handle
, ent
->dir
);
3159 ext4_mark_inode_dirty(handle
, ent
->dir
);
3163 static struct inode
*ext4_whiteout_for_rename(struct ext4_renament
*ent
,
3164 int credits
, handle_t
**h
)
3171 * for inode block, sb block, group summaries,
3174 credits
+= (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent
->dir
->i_sb
) +
3175 EXT4_XATTR_TRANS_BLOCKS
+ 4);
3177 wh
= ext4_new_inode_start_handle(ent
->dir
, S_IFCHR
| WHITEOUT_MODE
,
3178 &ent
->dentry
->d_name
, 0, NULL
,
3179 EXT4_HT_DIR
, credits
);
3181 handle
= ext4_journal_current_handle();
3184 ext4_journal_stop(handle
);
3185 if (PTR_ERR(wh
) == -ENOSPC
&&
3186 ext4_should_retry_alloc(ent
->dir
->i_sb
, &retries
))
3190 init_special_inode(wh
, wh
->i_mode
, WHITEOUT_DEV
);
3191 wh
->i_op
= &ext4_special_inode_operations
;
3197 * Anybody can rename anything with this: the permission checks are left to the
3198 * higher-level routines.
3200 * n.b. old_{dentry,inode) refers to the source dentry/inode
3201 * while new_{dentry,inode) refers to the destination dentry/inode
3202 * This comes from rename(const char *oldpath, const char *newpath)
3204 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3205 struct inode
*new_dir
, struct dentry
*new_dentry
,
3208 handle_t
*handle
= NULL
;
3209 struct ext4_renament old
= {
3211 .dentry
= old_dentry
,
3212 .inode
= old_dentry
->d_inode
,
3214 struct ext4_renament
new = {
3216 .dentry
= new_dentry
,
3217 .inode
= new_dentry
->d_inode
,
3221 struct inode
*whiteout
= NULL
;
3225 dquot_initialize(old
.dir
);
3226 dquot_initialize(new.dir
);
3228 /* Initialize quotas before so that eventual writes go
3229 * in separate transaction */
3231 dquot_initialize(new.inode
);
3233 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3235 return PTR_ERR(old
.bh
);
3237 * Check for inode number is _not_ due to possible IO errors.
3238 * We might rmdir the source, keep it as pwd of some process
3239 * and merrily kill the link to whatever was created under the
3240 * same name. Goodbye sticky bit ;-<
3243 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3246 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3247 &new.de
, &new.inlined
);
3248 if (IS_ERR(new.bh
)) {
3249 retval
= PTR_ERR(new.bh
);
3259 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3260 ext4_alloc_da_blocks(old
.inode
);
3262 credits
= (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3263 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
3264 if (!(flags
& RENAME_WHITEOUT
)) {
3265 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
, credits
);
3266 if (IS_ERR(handle
)) {
3267 retval
= PTR_ERR(handle
);
3272 whiteout
= ext4_whiteout_for_rename(&old
, credits
, &handle
);
3273 if (IS_ERR(whiteout
)) {
3274 retval
= PTR_ERR(whiteout
);
3280 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3281 ext4_handle_sync(handle
);
3283 if (S_ISDIR(old
.inode
->i_mode
)) {
3285 retval
= -ENOTEMPTY
;
3286 if (!empty_dir(new.inode
))
3290 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3293 retval
= ext4_rename_dir_prepare(handle
, &old
);
3298 * If we're renaming a file within an inline_data dir and adding or
3299 * setting the new dirent causes a conversion from inline_data to
3300 * extents/blockmap, we need to force the dirent delete code to
3301 * re-read the directory, or else we end up trying to delete a dirent
3302 * from what is now the extent tree root (or a block map).
3304 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3305 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3307 old_file_type
= old
.de
->file_type
;
3310 * Do this before adding a new entry, so the old entry is sure
3311 * to be still pointing to the valid old entry.
3313 retval
= ext4_setent(handle
, &old
, whiteout
->i_ino
,
3317 ext4_mark_inode_dirty(handle
, whiteout
);
3320 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3324 retval
= ext4_setent(handle
, &new,
3325 old
.inode
->i_ino
, old_file_type
);
3330 force_reread
= !ext4_test_inode_flag(new.dir
,
3331 EXT4_INODE_INLINE_DATA
);
3334 * Like most other Unix systems, set the ctime for inodes on a
3337 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3338 ext4_mark_inode_dirty(handle
, old
.inode
);
3344 ext4_rename_delete(handle
, &old
, force_reread
);
3348 ext4_dec_count(handle
, new.inode
);
3349 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3351 old
.dir
->i_ctime
= old
.dir
->i_mtime
= ext4_current_time(old
.dir
);
3352 ext4_update_dx_flag(old
.dir
);
3354 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3358 ext4_dec_count(handle
, old
.dir
);
3360 /* checked empty_dir above, can't have another parent,
3361 * ext4_dec_count() won't work for many-linked dirs */
3362 clear_nlink(new.inode
);
3364 ext4_inc_count(handle
, new.dir
);
3365 ext4_update_dx_flag(new.dir
);
3366 ext4_mark_inode_dirty(handle
, new.dir
);
3369 ext4_mark_inode_dirty(handle
, old
.dir
);
3371 ext4_mark_inode_dirty(handle
, new.inode
);
3372 if (!new.inode
->i_nlink
)
3373 ext4_orphan_add(handle
, new.inode
);
3383 drop_nlink(whiteout
);
3384 unlock_new_inode(whiteout
);
3388 ext4_journal_stop(handle
);
3392 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3393 struct inode
*new_dir
, struct dentry
*new_dentry
)
3395 handle_t
*handle
= NULL
;
3396 struct ext4_renament old
= {
3398 .dentry
= old_dentry
,
3399 .inode
= old_dentry
->d_inode
,
3401 struct ext4_renament
new = {
3403 .dentry
= new_dentry
,
3404 .inode
= new_dentry
->d_inode
,
3409 dquot_initialize(old
.dir
);
3410 dquot_initialize(new.dir
);
3412 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3413 &old
.de
, &old
.inlined
);
3415 return PTR_ERR(old
.bh
);
3417 * Check for inode number is _not_ due to possible IO errors.
3418 * We might rmdir the source, keep it as pwd of some process
3419 * and merrily kill the link to whatever was created under the
3420 * same name. Goodbye sticky bit ;-<
3423 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3426 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3427 &new.de
, &new.inlined
);
3428 if (IS_ERR(new.bh
)) {
3429 retval
= PTR_ERR(new.bh
);
3434 /* RENAME_EXCHANGE case: old *and* new must both exist */
3435 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3438 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3439 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3440 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3441 if (IS_ERR(handle
)) {
3442 retval
= PTR_ERR(handle
);
3447 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3448 ext4_handle_sync(handle
);
3450 if (S_ISDIR(old
.inode
->i_mode
)) {
3452 retval
= ext4_rename_dir_prepare(handle
, &old
);
3456 if (S_ISDIR(new.inode
->i_mode
)) {
3458 retval
= ext4_rename_dir_prepare(handle
, &new);
3464 * Other than the special case of overwriting a directory, parents'
3465 * nlink only needs to be modified if this is a cross directory rename.
3467 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3468 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3469 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3471 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3472 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3476 new_file_type
= new.de
->file_type
;
3477 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3481 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3486 * Like most other Unix systems, set the ctime for inodes on a
3489 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3490 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3491 ext4_mark_inode_dirty(handle
, old
.inode
);
3492 ext4_mark_inode_dirty(handle
, new.inode
);
3495 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3500 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3504 ext4_update_dir_count(handle
, &old
);
3505 ext4_update_dir_count(handle
, &new);
3514 ext4_journal_stop(handle
);
3518 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3519 struct inode
*new_dir
, struct dentry
*new_dentry
,
3522 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3525 if (flags
& RENAME_EXCHANGE
) {
3526 return ext4_cross_rename(old_dir
, old_dentry
,
3527 new_dir
, new_dentry
);
3530 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
3534 * directories can handle most operations...
3536 const struct inode_operations ext4_dir_inode_operations
= {
3537 .create
= ext4_create
,
3538 .lookup
= ext4_lookup
,
3540 .unlink
= ext4_unlink
,
3541 .symlink
= ext4_symlink
,
3542 .mkdir
= ext4_mkdir
,
3543 .rmdir
= ext4_rmdir
,
3544 .mknod
= ext4_mknod
,
3545 .tmpfile
= ext4_tmpfile
,
3546 .rename2
= ext4_rename2
,
3547 .setattr
= ext4_setattr
,
3548 .setxattr
= generic_setxattr
,
3549 .getxattr
= generic_getxattr
,
3550 .listxattr
= ext4_listxattr
,
3551 .removexattr
= generic_removexattr
,
3552 .get_acl
= ext4_get_acl
,
3553 .set_acl
= ext4_set_acl
,
3554 .fiemap
= ext4_fiemap
,
3557 const struct inode_operations ext4_special_inode_operations
= {
3558 .setattr
= ext4_setattr
,
3559 .setxattr
= generic_setxattr
,
3560 .getxattr
= generic_getxattr
,
3561 .listxattr
= ext4_listxattr
,
3562 .removexattr
= generic_removexattr
,
3563 .get_acl
= ext4_get_acl
,
3564 .set_acl
= ext4_set_acl
,