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/ext4_fs.h>
32 #include <linux/ext4_jbd2.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/bio.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)
50 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
52 static struct buffer_head
*ext4_append(handle_t
*handle
,
54 ext4_lblk_t
*block
, int *err
)
56 struct buffer_head
*bh
;
58 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
60 if ((bh
= ext4_bread(handle
, inode
, *block
, 1, err
))) {
61 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
62 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
63 ext4_journal_get_write_access(handle
,bh
);
69 #define assert(test) J_ASSERT(test)
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
77 #define dxtrace(command) command
79 #define dxtrace(command)
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
110 struct fake_dirent dot
;
112 struct fake_dirent dotdot
;
116 __le32 reserved_zero
;
118 u8 info_length
; /* 8 */
123 struct dx_entry entries
[0];
128 struct fake_dirent fake
;
129 struct dx_entry entries
[0];
135 struct buffer_head
*bh
;
136 struct dx_entry
*entries
;
147 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
148 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
149 static inline unsigned dx_get_hash (struct dx_entry
*entry
);
150 static void dx_set_hash (struct dx_entry
*entry
, unsigned value
);
151 static unsigned dx_get_count (struct dx_entry
*entries
);
152 static unsigned dx_get_limit (struct dx_entry
*entries
);
153 static void dx_set_count (struct dx_entry
*entries
, unsigned value
);
154 static void dx_set_limit (struct dx_entry
*entries
, unsigned value
);
155 static unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
);
156 static unsigned dx_node_limit (struct inode
*dir
);
157 static struct dx_frame
*dx_probe(struct dentry
*dentry
,
159 struct dx_hash_info
*hinfo
,
160 struct dx_frame
*frame
,
162 static void dx_release (struct dx_frame
*frames
);
163 static int dx_make_map (struct ext4_dir_entry_2
*de
, int size
,
164 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
165 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
166 static struct ext4_dir_entry_2
*dx_move_dirents (char *from
, char *to
,
167 struct dx_map_entry
*offsets
, int count
);
168 static struct ext4_dir_entry_2
* dx_pack_dirents (char *base
, int size
);
169 static void dx_insert_block(struct dx_frame
*frame
,
170 u32 hash
, ext4_lblk_t block
);
171 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
172 struct dx_frame
*frame
,
173 struct dx_frame
*frames
,
175 static struct buffer_head
* ext4_dx_find_entry(struct dentry
*dentry
,
176 struct ext4_dir_entry_2
**res_dir
, int *err
);
177 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
178 struct inode
*inode
);
181 * Future: use high four bits of block for coalesce-on-delete flags
182 * Mask them off for now.
185 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
187 return le32_to_cpu(entry
->block
) & 0x00ffffff;
190 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
192 entry
->block
= cpu_to_le32(value
);
195 static inline unsigned dx_get_hash (struct dx_entry
*entry
)
197 return le32_to_cpu(entry
->hash
);
200 static inline void dx_set_hash (struct dx_entry
*entry
, unsigned value
)
202 entry
->hash
= cpu_to_le32(value
);
205 static inline unsigned dx_get_count (struct dx_entry
*entries
)
207 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
210 static inline unsigned dx_get_limit (struct dx_entry
*entries
)
212 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
215 static inline void dx_set_count (struct dx_entry
*entries
, unsigned value
)
217 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
220 static inline void dx_set_limit (struct dx_entry
*entries
, unsigned value
)
222 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
225 static inline unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
)
227 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
228 EXT4_DIR_REC_LEN(2) - infosize
;
229 return 0? 20: entry_space
/ sizeof(struct dx_entry
);
232 static inline unsigned dx_node_limit (struct inode
*dir
)
234 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
235 return 0? 22: entry_space
/ sizeof(struct dx_entry
);
242 static void dx_show_index (char * label
, struct dx_entry
*entries
)
244 int i
, n
= dx_get_count (entries
);
245 printk("%s index ", label
);
246 for (i
= 0; i
< n
; i
++) {
247 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
248 0, (unsigned long)dx_get_block(entries
+ i
));
260 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
261 int size
, int show_names
)
263 unsigned names
= 0, space
= 0;
264 char *base
= (char *) de
;
265 struct dx_hash_info h
= *hinfo
;
268 while ((char *) de
< base
+ size
)
274 int len
= de
->name_len
;
275 char *name
= de
->name
;
276 while (len
--) printk("%c", *name
++);
277 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
278 printk(":%x.%u ", h
.hash
,
279 ((char *) de
- base
));
281 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
284 de
= ext4_next_entry(de
);
286 printk("(%i)\n", names
);
287 return (struct stats
) { names
, space
, 1 };
290 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
291 struct dx_entry
*entries
, int levels
)
293 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
294 unsigned count
= dx_get_count (entries
), names
= 0, space
= 0, i
;
296 struct buffer_head
*bh
;
298 printk("%i indexed blocks...\n", count
);
299 for (i
= 0; i
< count
; i
++, entries
++)
301 ext4_lblk_t block
= dx_get_block(entries
);
302 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
303 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
305 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
306 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0,&err
))) continue;
308 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
309 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
310 names
+= stats
.names
;
311 space
+= stats
.space
;
312 bcount
+= stats
.bcount
;
316 printk("%snames %u, fullness %u (%u%%)\n", levels
?"":" ",
317 names
, space
/bcount
,(space
/bcount
)*100/blocksize
);
318 return (struct stats
) { names
, space
, bcount
};
320 #endif /* DX_DEBUG */
323 * Probe for a directory leaf block to search.
325 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
326 * error in the directory index, and the caller should fall back to
327 * searching the directory normally. The callers of dx_probe **MUST**
328 * check for this error code, and make sure it never gets reflected
331 static struct dx_frame
*
332 dx_probe(struct dentry
*dentry
, struct inode
*dir
,
333 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
335 unsigned count
, indirect
;
336 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
337 struct dx_root
*root
;
338 struct buffer_head
*bh
;
339 struct dx_frame
*frame
= frame_in
;
344 dir
= dentry
->d_parent
->d_inode
;
345 if (!(bh
= ext4_bread (NULL
,dir
, 0, 0, err
)))
347 root
= (struct dx_root
*) bh
->b_data
;
348 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
349 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
350 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
351 ext4_warning(dir
->i_sb
, __FUNCTION__
,
352 "Unrecognised inode hash code %d",
353 root
->info
.hash_version
);
355 *err
= ERR_BAD_DX_DIR
;
358 hinfo
->hash_version
= root
->info
.hash_version
;
359 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
361 ext4fs_dirhash(dentry
->d_name
.name
, dentry
->d_name
.len
, hinfo
);
364 if (root
->info
.unused_flags
& 1) {
365 ext4_warning(dir
->i_sb
, __FUNCTION__
,
366 "Unimplemented inode hash flags: %#06x",
367 root
->info
.unused_flags
);
369 *err
= ERR_BAD_DX_DIR
;
373 if ((indirect
= root
->info
.indirect_levels
) > 1) {
374 ext4_warning(dir
->i_sb
, __FUNCTION__
,
375 "Unimplemented inode hash depth: %#06x",
376 root
->info
.indirect_levels
);
378 *err
= ERR_BAD_DX_DIR
;
382 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
383 root
->info
.info_length
);
385 if (dx_get_limit(entries
) != dx_root_limit(dir
,
386 root
->info
.info_length
)) {
387 ext4_warning(dir
->i_sb
, __FUNCTION__
,
388 "dx entry: limit != root limit");
390 *err
= ERR_BAD_DX_DIR
;
394 dxtrace (printk("Look up %x", hash
));
397 count
= dx_get_count(entries
);
398 if (!count
|| count
> dx_get_limit(entries
)) {
399 ext4_warning(dir
->i_sb
, __FUNCTION__
,
400 "dx entry: no count or count > limit");
402 *err
= ERR_BAD_DX_DIR
;
407 q
= entries
+ count
- 1;
411 dxtrace(printk("."));
412 if (dx_get_hash(m
) > hash
)
418 if (0) // linear search cross check
420 unsigned n
= count
- 1;
424 dxtrace(printk(","));
425 if (dx_get_hash(++at
) > hash
)
431 assert (at
== p
- 1);
435 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
437 frame
->entries
= entries
;
439 if (!indirect
--) return frame
;
440 if (!(bh
= ext4_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
442 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
443 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
444 ext4_warning(dir
->i_sb
, __FUNCTION__
,
445 "dx entry: limit != node limit");
447 *err
= ERR_BAD_DX_DIR
;
454 while (frame
>= frame_in
) {
459 if (*err
== ERR_BAD_DX_DIR
)
460 ext4_warning(dir
->i_sb
, __FUNCTION__
,
461 "Corrupt dir inode %ld, running e2fsck is "
462 "recommended.", dir
->i_ino
);
466 static void dx_release (struct dx_frame
*frames
)
468 if (frames
[0].bh
== NULL
)
471 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
472 brelse(frames
[1].bh
);
473 brelse(frames
[0].bh
);
477 * This function increments the frame pointer to search the next leaf
478 * block, and reads in the necessary intervening nodes if the search
479 * should be necessary. Whether or not the search is necessary is
480 * controlled by the hash parameter. If the hash value is even, then
481 * the search is only continued if the next block starts with that
482 * hash value. This is used if we are searching for a specific file.
484 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
486 * This function returns 1 if the caller should continue to search,
487 * or 0 if it should not. If there is an error reading one of the
488 * index blocks, it will a negative error code.
490 * If start_hash is non-null, it will be filled in with the starting
491 * hash of the next page.
493 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
494 struct dx_frame
*frame
,
495 struct dx_frame
*frames
,
499 struct buffer_head
*bh
;
500 int err
, num_frames
= 0;
505 * Find the next leaf page by incrementing the frame pointer.
506 * If we run out of entries in the interior node, loop around and
507 * increment pointer in the parent node. When we break out of
508 * this loop, num_frames indicates the number of interior
509 * nodes need to be read.
512 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
521 * If the hash is 1, then continue only if the next page has a
522 * continuation hash of any value. This is used for readdir
523 * handling. Otherwise, check to see if the hash matches the
524 * desired contiuation hash. If it doesn't, return since
525 * there's no point to read in the successive index pages.
527 bhash
= dx_get_hash(p
->at
);
530 if ((hash
& 1) == 0) {
531 if ((bhash
& ~1) != hash
)
535 * If the hash is HASH_NB_ALWAYS, we always go to the next
536 * block so no check is necessary
538 while (num_frames
--) {
539 if (!(bh
= ext4_bread(NULL
, dir
, dx_get_block(p
->at
),
541 return err
; /* Failure */
545 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
552 * p is at least 6 bytes before the end of page
554 static inline struct ext4_dir_entry_2
*ext4_next_entry(struct ext4_dir_entry_2
*p
)
556 return (struct ext4_dir_entry_2
*)((char *)p
+
557 ext4_rec_len_from_disk(p
->rec_len
));
561 * This function fills a red-black tree with information from a
562 * directory block. It returns the number directory entries loaded
563 * into the tree. If there is an error it is returned in err.
565 static int htree_dirblock_to_tree(struct file
*dir_file
,
566 struct inode
*dir
, ext4_lblk_t block
,
567 struct dx_hash_info
*hinfo
,
568 __u32 start_hash
, __u32 start_minor_hash
)
570 struct buffer_head
*bh
;
571 struct ext4_dir_entry_2
*de
, *top
;
574 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
575 (unsigned long)block
));
576 if (!(bh
= ext4_bread (NULL
, dir
, block
, 0, &err
)))
579 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
580 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
581 dir
->i_sb
->s_blocksize
-
582 EXT4_DIR_REC_LEN(0));
583 for (; de
< top
; de
= ext4_next_entry(de
)) {
584 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
585 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
586 +((char *)de
- bh
->b_data
))) {
587 /* On error, skip the f_pos to the next block. */
588 dir_file
->f_pos
= (dir_file
->f_pos
|
589 (dir
->i_sb
->s_blocksize
- 1)) + 1;
593 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
594 if ((hinfo
->hash
< start_hash
) ||
595 ((hinfo
->hash
== start_hash
) &&
596 (hinfo
->minor_hash
< start_minor_hash
)))
600 if ((err
= ext4_htree_store_dirent(dir_file
,
601 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
613 * This function fills a red-black tree with information from a
614 * directory. We start scanning the directory in hash order, starting
615 * at start_hash and start_minor_hash.
617 * This function returns the number of entries inserted into the tree,
618 * or a negative error code.
620 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
621 __u32 start_minor_hash
, __u32
*next_hash
)
623 struct dx_hash_info hinfo
;
624 struct ext4_dir_entry_2
*de
;
625 struct dx_frame frames
[2], *frame
;
632 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash
,
634 dir
= dir_file
->f_path
.dentry
->d_inode
;
635 if (!(EXT4_I(dir
)->i_flags
& EXT4_INDEX_FL
)) {
636 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
637 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
638 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
639 start_hash
, start_minor_hash
);
643 hinfo
.hash
= start_hash
;
644 hinfo
.minor_hash
= 0;
645 frame
= dx_probe(NULL
, dir_file
->f_path
.dentry
->d_inode
, &hinfo
, frames
, &err
);
649 /* Add '.' and '..' from the htree header */
650 if (!start_hash
&& !start_minor_hash
) {
651 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
652 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
656 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
657 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
658 de
= ext4_next_entry(de
);
659 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
665 block
= dx_get_block(frame
->at
);
666 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
667 start_hash
, start_minor_hash
);
674 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
675 frame
, frames
, &hashval
);
676 *next_hash
= hashval
;
682 * Stop if: (a) there are no more entries, or
683 * (b) we have inserted at least one entry and the
684 * next hash value is not a continuation
687 (count
&& ((hashval
& 1) == 0)))
691 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
701 * Directory block splitting, compacting
705 * Create map of hash values, offsets, and sizes, stored at end of block.
706 * Returns number of entries mapped.
708 static int dx_make_map (struct ext4_dir_entry_2
*de
, int size
,
709 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
712 char *base
= (char *) de
;
713 struct dx_hash_info h
= *hinfo
;
715 while ((char *) de
< base
+ size
)
717 if (de
->name_len
&& de
->inode
) {
718 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
720 map_tail
->hash
= h
.hash
;
721 map_tail
->offs
= (u16
) ((char *) de
- base
);
722 map_tail
->size
= le16_to_cpu(de
->rec_len
);
726 /* XXX: do we need to check rec_len == 0 case? -Chris */
727 de
= ext4_next_entry(de
);
732 /* Sort map by hash value */
733 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
735 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
737 /* Combsort until bubble sort doesn't suck */
740 if (count
- 9 < 2) /* 9, 10 -> 11 */
742 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
743 if (p
->hash
< q
->hash
)
746 /* Garden variety bubble sort */
751 if (q
[1].hash
>= q
[0].hash
)
759 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
761 struct dx_entry
*entries
= frame
->entries
;
762 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
763 int count
= dx_get_count(entries
);
765 assert(count
< dx_get_limit(entries
));
766 assert(old
< entries
+ count
);
767 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
768 dx_set_hash(new, hash
);
769 dx_set_block(new, block
);
770 dx_set_count(entries
, count
+ 1);
773 static void ext4_update_dx_flag(struct inode
*inode
)
775 if (!EXT4_HAS_COMPAT_FEATURE(inode
->i_sb
,
776 EXT4_FEATURE_COMPAT_DIR_INDEX
))
777 EXT4_I(inode
)->i_flags
&= ~EXT4_INDEX_FL
;
781 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
783 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
784 * `de != NULL' is guaranteed by caller.
786 static inline int ext4_match (int len
, const char * const name
,
787 struct ext4_dir_entry_2
* de
)
789 if (len
!= de
->name_len
)
793 return !memcmp(name
, de
->name
, len
);
797 * Returns 0 if not found, -1 on failure, and 1 on success
799 static inline int search_dirblock(struct buffer_head
* bh
,
801 struct dentry
*dentry
,
802 unsigned long offset
,
803 struct ext4_dir_entry_2
** res_dir
)
805 struct ext4_dir_entry_2
* de
;
808 const char *name
= dentry
->d_name
.name
;
809 int namelen
= dentry
->d_name
.len
;
811 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
812 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
813 while ((char *) de
< dlimit
) {
814 /* this code is executed quadratically often */
815 /* do minimal checking `by hand' */
817 if ((char *) de
+ namelen
<= dlimit
&&
818 ext4_match (namelen
, name
, de
)) {
819 /* found a match - just to be sure, do a full check */
820 if (!ext4_check_dir_entry("ext4_find_entry",
821 dir
, de
, bh
, offset
))
826 /* prevent looping on a bad block */
827 de_len
= ext4_rec_len_from_disk(de
->rec_len
);
831 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
840 * finds an entry in the specified directory with the wanted name. It
841 * returns the cache buffer in which the entry was found, and the entry
842 * itself (as a parameter - res_dir). It does NOT read the inode of the
843 * entry - you'll have to do that yourself if you want to.
845 * The returned buffer_head has ->b_count elevated. The caller is expected
846 * to brelse() it when appropriate.
848 static struct buffer_head
* ext4_find_entry (struct dentry
*dentry
,
849 struct ext4_dir_entry_2
** res_dir
)
851 struct super_block
* sb
;
852 struct buffer_head
* bh_use
[NAMEI_RA_SIZE
];
853 struct buffer_head
* bh
, *ret
= NULL
;
854 ext4_lblk_t start
, block
, b
;
855 int ra_max
= 0; /* Number of bh's in the readahead
857 int ra_ptr
= 0; /* Current index into readahead
862 struct inode
*dir
= dentry
->d_parent
->d_inode
;
869 blocksize
= sb
->s_blocksize
;
870 namelen
= dentry
->d_name
.len
;
871 name
= dentry
->d_name
.name
;
872 if (namelen
> EXT4_NAME_LEN
)
875 bh
= ext4_dx_find_entry(dentry
, res_dir
, &err
);
877 * On success, or if the error was file not found,
878 * return. Otherwise, fall back to doing a search the
881 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
883 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
885 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
886 start
= EXT4_I(dir
)->i_dir_start_lookup
;
887 if (start
>= nblocks
)
893 * We deal with the read-ahead logic here.
895 if (ra_ptr
>= ra_max
) {
896 /* Refill the readahead buffer */
899 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
901 * Terminate if we reach the end of the
902 * directory and must wrap, or if our
903 * search has finished at this block.
905 if (b
>= nblocks
|| (num
&& block
== start
)) {
906 bh_use
[ra_max
] = NULL
;
910 bh
= ext4_getblk(NULL
, dir
, b
++, 0, &err
);
913 ll_rw_block(READ_META
, 1, &bh
);
916 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
919 if (!buffer_uptodate(bh
)) {
920 /* read error, skip block & hope for the best */
921 ext4_error(sb
, __FUNCTION__
, "reading directory #%lu "
922 "offset %lu", dir
->i_ino
,
923 (unsigned long)block
);
927 i
= search_dirblock(bh
, dir
, dentry
,
928 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
930 EXT4_I(dir
)->i_dir_start_lookup
= block
;
932 goto cleanup_and_exit
;
936 goto cleanup_and_exit
;
939 if (++block
>= nblocks
)
941 } while (block
!= start
);
944 * If the directory has grown while we were searching, then
945 * search the last part of the directory before giving up.
948 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
949 if (block
< nblocks
) {
955 /* Clean up the read-ahead blocks */
956 for (; ra_ptr
< ra_max
; ra_ptr
++)
957 brelse (bh_use
[ra_ptr
]);
961 static struct buffer_head
* ext4_dx_find_entry(struct dentry
*dentry
,
962 struct ext4_dir_entry_2
**res_dir
, int *err
)
964 struct super_block
* sb
;
965 struct dx_hash_info hinfo
;
967 struct dx_frame frames
[2], *frame
;
968 struct ext4_dir_entry_2
*de
, *top
;
969 struct buffer_head
*bh
;
972 int namelen
= dentry
->d_name
.len
;
973 const u8
*name
= dentry
->d_name
.name
;
974 struct inode
*dir
= dentry
->d_parent
->d_inode
;
977 /* NFS may look up ".." - look at dx_root directory block */
978 if (namelen
> 2 || name
[0] != '.'||(name
[1] != '.' && name
[1] != '\0')){
979 if (!(frame
= dx_probe(dentry
, NULL
, &hinfo
, frames
, err
)))
983 frame
->bh
= NULL
; /* for dx_release() */
984 frame
->at
= (struct dx_entry
*)frames
; /* hack for zero entry*/
985 dx_set_block(frame
->at
, 0); /* dx_root block is 0 */
989 block
= dx_get_block(frame
->at
);
990 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0, err
)))
992 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
993 top
= (struct ext4_dir_entry_2
*) ((char *) de
+ sb
->s_blocksize
-
994 EXT4_DIR_REC_LEN(0));
995 for (; de
< top
; de
= ext4_next_entry(de
))
996 if (ext4_match (namelen
, name
, de
)) {
997 if (!ext4_check_dir_entry("ext4_find_entry",
999 (block
<<EXT4_BLOCK_SIZE_BITS(sb
))
1000 +((char *)de
- bh
->b_data
))) {
1002 *err
= ERR_BAD_DX_DIR
;
1006 dx_release (frames
);
1010 /* Check to see if we should continue to search */
1011 retval
= ext4_htree_next_block(dir
, hash
, frame
,
1014 ext4_warning(sb
, __FUNCTION__
,
1015 "error reading index page in directory #%lu",
1020 } while (retval
== 1);
1024 dxtrace(printk("%s not found\n", name
));
1025 dx_release (frames
);
1029 static struct dentry
*ext4_lookup(struct inode
* dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1031 struct inode
* inode
;
1032 struct ext4_dir_entry_2
* de
;
1033 struct buffer_head
* bh
;
1035 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1036 return ERR_PTR(-ENAMETOOLONG
);
1038 bh
= ext4_find_entry(dentry
, &de
);
1041 unsigned long ino
= le32_to_cpu(de
->inode
);
1043 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1044 ext4_error(dir
->i_sb
, "ext4_lookup",
1045 "bad inode number: %lu", ino
);
1048 inode
= iget(dir
->i_sb
, ino
);
1051 return ERR_PTR(-EACCES
);
1053 if (is_bad_inode(inode
)) {
1055 return ERR_PTR(-ENOENT
);
1058 return d_splice_alias(inode
, dentry
);
1062 struct dentry
*ext4_get_parent(struct dentry
*child
)
1065 struct dentry
*parent
;
1066 struct inode
*inode
;
1067 struct dentry dotdot
;
1068 struct ext4_dir_entry_2
* de
;
1069 struct buffer_head
*bh
;
1071 dotdot
.d_name
.name
= "..";
1072 dotdot
.d_name
.len
= 2;
1073 dotdot
.d_parent
= child
; /* confusing, isn't it! */
1075 bh
= ext4_find_entry(&dotdot
, &de
);
1078 return ERR_PTR(-ENOENT
);
1079 ino
= le32_to_cpu(de
->inode
);
1082 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1083 ext4_error(child
->d_inode
->i_sb
, "ext4_get_parent",
1084 "bad inode number: %lu", ino
);
1087 inode
= iget(child
->d_inode
->i_sb
, ino
);
1090 return ERR_PTR(-EACCES
);
1092 if (is_bad_inode(inode
)) {
1094 return ERR_PTR(-ENOENT
);
1097 parent
= d_alloc_anon(inode
);
1100 parent
= ERR_PTR(-ENOMEM
);
1106 static unsigned char ext4_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1107 [S_IFREG
>> S_SHIFT
] = EXT4_FT_REG_FILE
,
1108 [S_IFDIR
>> S_SHIFT
] = EXT4_FT_DIR
,
1109 [S_IFCHR
>> S_SHIFT
] = EXT4_FT_CHRDEV
,
1110 [S_IFBLK
>> S_SHIFT
] = EXT4_FT_BLKDEV
,
1111 [S_IFIFO
>> S_SHIFT
] = EXT4_FT_FIFO
,
1112 [S_IFSOCK
>> S_SHIFT
] = EXT4_FT_SOCK
,
1113 [S_IFLNK
>> S_SHIFT
] = EXT4_FT_SYMLINK
,
1116 static inline void ext4_set_de_type(struct super_block
*sb
,
1117 struct ext4_dir_entry_2
*de
,
1119 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FILETYPE
))
1120 de
->file_type
= ext4_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1124 * Move count entries from end of map between two memory locations.
1125 * Returns pointer to last entry moved.
1127 static struct ext4_dir_entry_2
*
1128 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1130 unsigned rec_len
= 0;
1133 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*) (from
+ map
->offs
);
1134 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1135 memcpy (to
, de
, rec_len
);
1136 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1137 ext4_rec_len_to_disk(rec_len
);
1142 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1146 * Compact each dir entry in the range to the minimal rec_len.
1147 * Returns pointer to last entry in range.
1149 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1151 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1152 unsigned rec_len
= 0;
1155 while ((char*)de
< base
+ size
) {
1156 next
= ext4_next_entry(de
);
1157 if (de
->inode
&& de
->name_len
) {
1158 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1160 memmove(to
, de
, rec_len
);
1161 to
->rec_len
= ext4_rec_len_to_disk(rec_len
);
1163 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1171 * Split a full leaf block to make room for a new dir entry.
1172 * Allocate a new block, and move entries so that they are approx. equally full.
1173 * Returns pointer to de in block into which the new entry will be inserted.
1175 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1176 struct buffer_head
**bh
,struct dx_frame
*frame
,
1177 struct dx_hash_info
*hinfo
, int *error
)
1179 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1180 unsigned count
, continued
;
1181 struct buffer_head
*bh2
;
1182 ext4_lblk_t newblock
;
1184 struct dx_map_entry
*map
;
1185 char *data1
= (*bh
)->b_data
, *data2
;
1186 unsigned split
, move
, size
, i
;
1187 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1190 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1197 BUFFER_TRACE(*bh
, "get_write_access");
1198 err
= ext4_journal_get_write_access(handle
, *bh
);
1202 BUFFER_TRACE(frame
->bh
, "get_write_access");
1203 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1207 data2
= bh2
->b_data
;
1209 /* create map in the end of data2 block */
1210 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1211 count
= dx_make_map ((struct ext4_dir_entry_2
*) data1
,
1212 blocksize
, hinfo
, map
);
1214 dx_sort_map (map
, count
);
1215 /* Split the existing block in the middle, size-wise */
1218 for (i
= count
-1; i
>= 0; i
--) {
1219 /* is more than half of this entry in 2nd half of the block? */
1220 if (size
+ map
[i
].size
/2 > blocksize
/2)
1222 size
+= map
[i
].size
;
1225 /* map index at which we will split */
1226 split
= count
- move
;
1227 hash2
= map
[split
].hash
;
1228 continued
= hash2
== map
[split
- 1].hash
;
1229 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1230 (unsigned long)dx_get_block(frame
->at
),
1231 hash2
, split
, count
-split
));
1233 /* Fancy dance to stay within two buffers */
1234 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1235 de
= dx_pack_dirents(data1
,blocksize
);
1236 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1237 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1238 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1239 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1241 /* Which block gets the new entry? */
1242 if (hinfo
->hash
>= hash2
)
1247 dx_insert_block (frame
, hash2
+ continued
, newblock
);
1248 err
= ext4_journal_dirty_metadata (handle
, bh2
);
1251 err
= ext4_journal_dirty_metadata (handle
, frame
->bh
);
1255 dxtrace(dx_show_index ("frame", frame
->entries
));
1262 ext4_std_error(dir
->i_sb
, err
);
1269 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1270 * it points to a directory entry which is guaranteed to be large
1271 * enough for new directory entry. If de is NULL, then
1272 * add_dirent_to_buf will attempt search the directory block for
1273 * space. It will return -ENOSPC if no space is available, and -EIO
1274 * and -EEXIST if directory entry already exists.
1276 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1277 * all other cases bh is released.
1279 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1280 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1281 struct buffer_head
* bh
)
1283 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1284 const char *name
= dentry
->d_name
.name
;
1285 int namelen
= dentry
->d_name
.len
;
1286 unsigned long offset
= 0;
1287 unsigned short reclen
;
1288 int nlen
, rlen
, err
;
1291 reclen
= EXT4_DIR_REC_LEN(namelen
);
1293 de
= (struct ext4_dir_entry_2
*)bh
->b_data
;
1294 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1295 while ((char *) de
<= top
) {
1296 if (!ext4_check_dir_entry("ext4_add_entry", dir
, de
,
1301 if (ext4_match (namelen
, name
, de
)) {
1305 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1306 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1307 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1309 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1312 if ((char *) de
> top
)
1315 BUFFER_TRACE(bh
, "get_write_access");
1316 err
= ext4_journal_get_write_access(handle
, bh
);
1318 ext4_std_error(dir
->i_sb
, err
);
1323 /* By now the buffer is marked for journaling */
1324 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1325 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1327 struct ext4_dir_entry_2
*de1
= (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1328 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
);
1329 de
->rec_len
= ext4_rec_len_to_disk(nlen
);
1332 de
->file_type
= EXT4_FT_UNKNOWN
;
1334 de
->inode
= cpu_to_le32(inode
->i_ino
);
1335 ext4_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1338 de
->name_len
= namelen
;
1339 memcpy (de
->name
, name
, namelen
);
1341 * XXX shouldn't update any times until successful
1342 * completion of syscall, but too many callers depend
1345 * XXX similarly, too many callers depend on
1346 * ext4_new_inode() setting the times, but error
1347 * recovery deletes the inode, so the worst that can
1348 * happen is that the times are slightly out of date
1349 * and/or different from the directory change time.
1351 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1352 ext4_update_dx_flag(dir
);
1354 ext4_mark_inode_dirty(handle
, dir
);
1355 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1356 err
= ext4_journal_dirty_metadata(handle
, bh
);
1358 ext4_std_error(dir
->i_sb
, err
);
1364 * This converts a one block unindexed directory to a 3 block indexed
1365 * directory, and adds the dentry to the indexed directory.
1367 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1368 struct inode
*inode
, struct buffer_head
*bh
)
1370 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1371 const char *name
= dentry
->d_name
.name
;
1372 int namelen
= dentry
->d_name
.len
;
1373 struct buffer_head
*bh2
;
1374 struct dx_root
*root
;
1375 struct dx_frame frames
[2], *frame
;
1376 struct dx_entry
*entries
;
1377 struct ext4_dir_entry_2
*de
, *de2
;
1382 struct dx_hash_info hinfo
;
1384 struct fake_dirent
*fde
;
1386 blocksize
= dir
->i_sb
->s_blocksize
;
1387 dxtrace(printk("Creating index\n"));
1388 retval
= ext4_journal_get_write_access(handle
, bh
);
1390 ext4_std_error(dir
->i_sb
, retval
);
1394 root
= (struct dx_root
*) bh
->b_data
;
1396 bh2
= ext4_append (handle
, dir
, &block
, &retval
);
1401 EXT4_I(dir
)->i_flags
|= EXT4_INDEX_FL
;
1402 data1
= bh2
->b_data
;
1404 /* The 0th block becomes the root, move the dirents out */
1405 fde
= &root
->dotdot
;
1406 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1407 ext4_rec_len_from_disk(fde
->rec_len
));
1408 len
= ((char *) root
) + blocksize
- (char *) de
;
1409 memcpy (data1
, de
, len
);
1410 de
= (struct ext4_dir_entry_2
*) data1
;
1412 while ((char *)(de2
= ext4_next_entry(de
)) < top
)
1414 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1415 /* Initialize the root; the dot dirents already exist */
1416 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1417 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2));
1418 memset (&root
->info
, 0, sizeof(root
->info
));
1419 root
->info
.info_length
= sizeof(root
->info
);
1420 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1421 entries
= root
->entries
;
1422 dx_set_block (entries
, 1);
1423 dx_set_count (entries
, 1);
1424 dx_set_limit (entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1426 /* Initialize as for dx_probe */
1427 hinfo
.hash_version
= root
->info
.hash_version
;
1428 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1429 ext4fs_dirhash(name
, namelen
, &hinfo
);
1431 frame
->entries
= entries
;
1432 frame
->at
= entries
;
1435 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1436 dx_release (frames
);
1440 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1446 * adds a file entry to the specified directory, using the same
1447 * semantics as ext4_find_entry(). It returns NULL if it failed.
1449 * NOTE!! The inode part of 'de' is left at 0 - which means you
1450 * may not sleep between calling this and putting something into
1451 * the entry, as someone else might have used it while you slept.
1453 static int ext4_add_entry (handle_t
*handle
, struct dentry
*dentry
,
1454 struct inode
*inode
)
1456 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1457 unsigned long offset
;
1458 struct buffer_head
* bh
;
1459 struct ext4_dir_entry_2
*de
;
1460 struct super_block
* sb
;
1464 ext4_lblk_t block
, blocks
;
1467 blocksize
= sb
->s_blocksize
;
1468 if (!dentry
->d_name
.len
)
1471 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1472 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1474 EXT4_I(dir
)->i_flags
&= ~EXT4_INDEX_FL
;
1476 ext4_mark_inode_dirty(handle
, dir
);
1478 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1479 for (block
= 0, offset
= 0; block
< blocks
; block
++) {
1480 bh
= ext4_bread(handle
, dir
, block
, 0, &retval
);
1483 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1484 if (retval
!= -ENOSPC
)
1487 if (blocks
== 1 && !dx_fallback
&&
1488 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1489 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1492 bh
= ext4_append(handle
, dir
, &block
, &retval
);
1495 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1497 de
->rec_len
= ext4_rec_len_to_disk(blocksize
);
1498 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1502 * Returns 0 for success, or a negative error value
1504 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1505 struct inode
*inode
)
1507 struct dx_frame frames
[2], *frame
;
1508 struct dx_entry
*entries
, *at
;
1509 struct dx_hash_info hinfo
;
1510 struct buffer_head
* bh
;
1511 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1512 struct super_block
* sb
= dir
->i_sb
;
1513 struct ext4_dir_entry_2
*de
;
1516 frame
= dx_probe(dentry
, NULL
, &hinfo
, frames
, &err
);
1519 entries
= frame
->entries
;
1522 if (!(bh
= ext4_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1525 BUFFER_TRACE(bh
, "get_write_access");
1526 err
= ext4_journal_get_write_access(handle
, bh
);
1530 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1531 if (err
!= -ENOSPC
) {
1536 /* Block full, should compress but for now just split */
1537 dxtrace(printk("using %u of %u node entries\n",
1538 dx_get_count(entries
), dx_get_limit(entries
)));
1539 /* Need to split index? */
1540 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1541 ext4_lblk_t newblock
;
1542 unsigned icount
= dx_get_count(entries
);
1543 int levels
= frame
- frames
;
1544 struct dx_entry
*entries2
;
1545 struct dx_node
*node2
;
1546 struct buffer_head
*bh2
;
1548 if (levels
&& (dx_get_count(frames
->entries
) ==
1549 dx_get_limit(frames
->entries
))) {
1550 ext4_warning(sb
, __FUNCTION__
,
1551 "Directory index full!");
1555 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1558 node2
= (struct dx_node
*)(bh2
->b_data
);
1559 entries2
= node2
->entries
;
1560 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
);
1561 node2
->fake
.inode
= 0;
1562 BUFFER_TRACE(frame
->bh
, "get_write_access");
1563 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1567 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1568 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1569 dxtrace(printk("Split index %i/%i\n", icount1
, icount2
));
1571 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1572 err
= ext4_journal_get_write_access(handle
,
1577 memcpy ((char *) entries2
, (char *) (entries
+ icount1
),
1578 icount2
* sizeof(struct dx_entry
));
1579 dx_set_count (entries
, icount1
);
1580 dx_set_count (entries2
, icount2
);
1581 dx_set_limit (entries2
, dx_node_limit(dir
));
1583 /* Which index block gets the new entry? */
1584 if (at
- entries
>= icount1
) {
1585 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1586 frame
->entries
= entries
= entries2
;
1587 swap(frame
->bh
, bh2
);
1589 dx_insert_block (frames
+ 0, hash2
, newblock
);
1590 dxtrace(dx_show_index ("node", frames
[1].entries
));
1591 dxtrace(dx_show_index ("node",
1592 ((struct dx_node
*) bh2
->b_data
)->entries
));
1593 err
= ext4_journal_dirty_metadata(handle
, bh2
);
1598 dxtrace(printk("Creating second level index...\n"));
1599 memcpy((char *) entries2
, (char *) entries
,
1600 icount
* sizeof(struct dx_entry
));
1601 dx_set_limit(entries2
, dx_node_limit(dir
));
1604 dx_set_count(entries
, 1);
1605 dx_set_block(entries
+ 0, newblock
);
1606 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1608 /* Add new access path frame */
1610 frame
->at
= at
= at
- entries
+ entries2
;
1611 frame
->entries
= entries
= entries2
;
1613 err
= ext4_journal_get_write_access(handle
,
1618 ext4_journal_dirty_metadata(handle
, frames
[0].bh
);
1620 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1623 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1628 ext4_std_error(dir
->i_sb
, err
);
1637 * ext4_delete_entry deletes a directory entry by merging it with the
1640 static int ext4_delete_entry (handle_t
*handle
,
1642 struct ext4_dir_entry_2
* de_del
,
1643 struct buffer_head
* bh
)
1645 struct ext4_dir_entry_2
* de
, * pde
;
1650 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1651 while (i
< bh
->b_size
) {
1652 if (!ext4_check_dir_entry("ext4_delete_entry", dir
, de
, bh
, i
))
1655 BUFFER_TRACE(bh
, "get_write_access");
1656 ext4_journal_get_write_access(handle
, bh
);
1658 pde
->rec_len
= ext4_rec_len_to_disk(
1659 ext4_rec_len_from_disk(pde
->rec_len
) +
1660 ext4_rec_len_from_disk(de
->rec_len
));
1664 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1665 ext4_journal_dirty_metadata(handle
, bh
);
1668 i
+= ext4_rec_len_from_disk(de
->rec_len
);
1670 de
= ext4_next_entry(de
);
1676 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1677 * since this indicates that nlinks count was previously 1.
1679 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
1682 if (is_dx(inode
) && inode
->i_nlink
> 1) {
1683 /* limit is 16-bit i_links_count */
1684 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
1686 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
1687 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
1693 * If a directory had nlink == 1, then we should let it be 1. This indicates
1694 * directory has >EXT4_LINK_MAX subdirs.
1696 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
1699 if (S_ISDIR(inode
->i_mode
) && inode
->i_nlink
== 0)
1704 static int ext4_add_nondir(handle_t
*handle
,
1705 struct dentry
*dentry
, struct inode
*inode
)
1707 int err
= ext4_add_entry(handle
, dentry
, inode
);
1709 ext4_mark_inode_dirty(handle
, inode
);
1710 d_instantiate(dentry
, inode
);
1719 * By the time this is called, we already have created
1720 * the directory cache entry for the new file, but it
1721 * is so far negative - it has no inode.
1723 * If the create succeeds, we fill in the inode information
1724 * with d_instantiate().
1726 static int ext4_create (struct inode
* dir
, struct dentry
* dentry
, int mode
,
1727 struct nameidata
*nd
)
1730 struct inode
* inode
;
1731 int err
, retries
= 0;
1734 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1735 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1736 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1738 return PTR_ERR(handle
);
1740 if (IS_DIRSYNC(dir
))
1743 inode
= ext4_new_inode (handle
, dir
, mode
);
1744 err
= PTR_ERR(inode
);
1745 if (!IS_ERR(inode
)) {
1746 inode
->i_op
= &ext4_file_inode_operations
;
1747 inode
->i_fop
= &ext4_file_operations
;
1748 ext4_set_aops(inode
);
1749 err
= ext4_add_nondir(handle
, dentry
, inode
);
1751 ext4_journal_stop(handle
);
1752 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1757 static int ext4_mknod (struct inode
* dir
, struct dentry
*dentry
,
1758 int mode
, dev_t rdev
)
1761 struct inode
*inode
;
1762 int err
, retries
= 0;
1764 if (!new_valid_dev(rdev
))
1768 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1769 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1770 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1772 return PTR_ERR(handle
);
1774 if (IS_DIRSYNC(dir
))
1777 inode
= ext4_new_inode (handle
, dir
, mode
);
1778 err
= PTR_ERR(inode
);
1779 if (!IS_ERR(inode
)) {
1780 init_special_inode(inode
, inode
->i_mode
, rdev
);
1781 #ifdef CONFIG_EXT4DEV_FS_XATTR
1782 inode
->i_op
= &ext4_special_inode_operations
;
1784 err
= ext4_add_nondir(handle
, dentry
, inode
);
1786 ext4_journal_stop(handle
);
1787 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1792 static int ext4_mkdir(struct inode
* dir
, struct dentry
* dentry
, int mode
)
1795 struct inode
* inode
;
1796 struct buffer_head
* dir_block
;
1797 struct ext4_dir_entry_2
* de
;
1798 int err
, retries
= 0;
1800 if (EXT4_DIR_LINK_MAX(dir
))
1804 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1805 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1806 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1808 return PTR_ERR(handle
);
1810 if (IS_DIRSYNC(dir
))
1813 inode
= ext4_new_inode (handle
, dir
, S_IFDIR
| mode
);
1814 err
= PTR_ERR(inode
);
1818 inode
->i_op
= &ext4_dir_inode_operations
;
1819 inode
->i_fop
= &ext4_dir_operations
;
1820 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1821 dir_block
= ext4_bread (handle
, inode
, 0, 1, &err
);
1823 ext4_dec_count(handle
, inode
); /* is this nlink == 0? */
1824 ext4_mark_inode_dirty(handle
, inode
);
1828 BUFFER_TRACE(dir_block
, "get_write_access");
1829 ext4_journal_get_write_access(handle
, dir_block
);
1830 de
= (struct ext4_dir_entry_2
*) dir_block
->b_data
;
1831 de
->inode
= cpu_to_le32(inode
->i_ino
);
1833 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
));
1834 strcpy (de
->name
, ".");
1835 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1836 de
= ext4_next_entry(de
);
1837 de
->inode
= cpu_to_le32(dir
->i_ino
);
1838 de
->rec_len
= ext4_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1839 EXT4_DIR_REC_LEN(1));
1841 strcpy (de
->name
, "..");
1842 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1844 BUFFER_TRACE(dir_block
, "call ext4_journal_dirty_metadata");
1845 ext4_journal_dirty_metadata(handle
, dir_block
);
1847 ext4_mark_inode_dirty(handle
, inode
);
1848 err
= ext4_add_entry (handle
, dentry
, inode
);
1851 ext4_mark_inode_dirty(handle
, inode
);
1855 ext4_inc_count(handle
, dir
);
1856 ext4_update_dx_flag(dir
);
1857 ext4_mark_inode_dirty(handle
, dir
);
1858 d_instantiate(dentry
, inode
);
1860 ext4_journal_stop(handle
);
1861 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1867 * routine to check that the specified directory is empty (for rmdir)
1869 static int empty_dir (struct inode
* inode
)
1871 unsigned long offset
;
1872 struct buffer_head
* bh
;
1873 struct ext4_dir_entry_2
* de
, * de1
;
1874 struct super_block
* sb
;
1878 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1879 !(bh
= ext4_bread (NULL
, inode
, 0, 0, &err
))) {
1881 ext4_error(inode
->i_sb
, __FUNCTION__
,
1882 "error %d reading directory #%lu offset 0",
1885 ext4_warning(inode
->i_sb
, __FUNCTION__
,
1886 "bad directory (dir #%lu) - no data block",
1890 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1891 de1
= ext4_next_entry(de
);
1892 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1893 !le32_to_cpu(de1
->inode
) ||
1894 strcmp (".", de
->name
) ||
1895 strcmp ("..", de1
->name
)) {
1896 ext4_warning (inode
->i_sb
, "empty_dir",
1897 "bad directory (dir #%lu) - no `.' or `..'",
1902 offset
= ext4_rec_len_from_disk(de
->rec_len
) +
1903 ext4_rec_len_from_disk(de1
->rec_len
);
1904 de
= ext4_next_entry(de1
);
1905 while (offset
< inode
->i_size
) {
1907 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1910 bh
= ext4_bread (NULL
, inode
,
1911 offset
>> EXT4_BLOCK_SIZE_BITS(sb
), 0, &err
);
1914 ext4_error(sb
, __FUNCTION__
,
1915 "error %d reading directory"
1917 err
, inode
->i_ino
, offset
);
1918 offset
+= sb
->s_blocksize
;
1921 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1923 if (!ext4_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1924 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
1926 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1929 if (le32_to_cpu(de
->inode
)) {
1933 offset
+= ext4_rec_len_from_disk(de
->rec_len
);
1934 de
= ext4_next_entry(de
);
1940 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1941 * such inodes, starting at the superblock, in case we crash before the
1942 * file is closed/deleted, or in case the inode truncate spans multiple
1943 * transactions and the last transaction is not recovered after a crash.
1945 * At filesystem recovery time, we walk this list deleting unlinked
1946 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1948 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
1950 struct super_block
*sb
= inode
->i_sb
;
1951 struct ext4_iloc iloc
;
1955 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
1958 /* Orphan handling is only valid for files with data blocks
1959 * being truncated, or files being unlinked. */
1961 /* @@@ FIXME: Observation from aviro:
1962 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1963 * here (on lock_super()), so race with ext4_link() which might bump
1964 * ->i_nlink. For, say it, character device. Not a regular file,
1965 * not a directory, not a symlink and ->i_nlink > 0.
1967 J_ASSERT ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1968 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1970 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1971 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1975 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
1979 /* Insert this inode at the head of the on-disk orphan list... */
1980 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT4_SB(sb
)->s_es
->s_last_orphan
);
1981 EXT4_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1982 err
= ext4_journal_dirty_metadata(handle
, EXT4_SB(sb
)->s_sbh
);
1983 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1987 /* Only add to the head of the in-memory list if all the
1988 * previous operations succeeded. If the orphan_add is going to
1989 * fail (possibly taking the journal offline), we can't risk
1990 * leaving the inode on the orphan list: stray orphan-list
1991 * entries can cause panics at unmount time.
1993 * This is safe: on error we're going to ignore the orphan list
1994 * anyway on the next recovery. */
1996 list_add(&EXT4_I(inode
)->i_orphan
, &EXT4_SB(sb
)->s_orphan
);
1998 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1999 jbd_debug(4, "orphan inode %lu will point to %d\n",
2000 inode
->i_ino
, NEXT_ORPHAN(inode
));
2003 ext4_std_error(inode
->i_sb
, err
);
2008 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2009 * of such inodes stored on disk, because it is finally being cleaned up.
2011 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2013 struct list_head
*prev
;
2014 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2015 struct ext4_sb_info
*sbi
;
2016 unsigned long ino_next
;
2017 struct ext4_iloc iloc
;
2020 lock_super(inode
->i_sb
);
2021 if (list_empty(&ei
->i_orphan
)) {
2022 unlock_super(inode
->i_sb
);
2026 ino_next
= NEXT_ORPHAN(inode
);
2027 prev
= ei
->i_orphan
.prev
;
2028 sbi
= EXT4_SB(inode
->i_sb
);
2030 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2032 list_del_init(&ei
->i_orphan
);
2034 /* If we're on an error path, we may not have a valid
2035 * transaction handle with which to update the orphan list on
2036 * disk, but we still need to remove the inode from the linked
2037 * list in memory. */
2041 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2045 if (prev
== &sbi
->s_orphan
) {
2046 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
2047 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2048 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2051 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2052 err
= ext4_journal_dirty_metadata(handle
, sbi
->s_sbh
);
2054 struct ext4_iloc iloc2
;
2055 struct inode
*i_prev
=
2056 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2058 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2059 i_prev
->i_ino
, ino_next
);
2060 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2063 NEXT_ORPHAN(i_prev
) = ino_next
;
2064 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2068 NEXT_ORPHAN(inode
) = 0;
2069 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2072 ext4_std_error(inode
->i_sb
, err
);
2074 unlock_super(inode
->i_sb
);
2082 static int ext4_rmdir (struct inode
* dir
, struct dentry
*dentry
)
2085 struct inode
* inode
;
2086 struct buffer_head
* bh
;
2087 struct ext4_dir_entry_2
* de
;
2090 /* Initialize quotas before so that eventual writes go in
2091 * separate transaction */
2092 DQUOT_INIT(dentry
->d_inode
);
2093 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2095 return PTR_ERR(handle
);
2098 bh
= ext4_find_entry (dentry
, &de
);
2102 if (IS_DIRSYNC(dir
))
2105 inode
= dentry
->d_inode
;
2108 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2111 retval
= -ENOTEMPTY
;
2112 if (!empty_dir (inode
))
2115 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2118 if (!EXT4_DIR_LINK_EMPTY(inode
))
2119 ext4_warning (inode
->i_sb
, "ext4_rmdir",
2120 "empty directory has too many links (%d)",
2124 /* There's no need to set i_disksize: the fact that i_nlink is
2125 * zero will ensure that the right thing happens during any
2128 ext4_orphan_add(handle
, inode
);
2129 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2130 ext4_mark_inode_dirty(handle
, inode
);
2131 ext4_dec_count(handle
, dir
);
2132 ext4_update_dx_flag(dir
);
2133 ext4_mark_inode_dirty(handle
, dir
);
2136 ext4_journal_stop(handle
);
2141 static int ext4_unlink(struct inode
* dir
, struct dentry
*dentry
)
2144 struct inode
* inode
;
2145 struct buffer_head
* bh
;
2146 struct ext4_dir_entry_2
* de
;
2149 /* Initialize quotas before so that eventual writes go
2150 * in separate transaction */
2151 DQUOT_INIT(dentry
->d_inode
);
2152 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2154 return PTR_ERR(handle
);
2156 if (IS_DIRSYNC(dir
))
2160 bh
= ext4_find_entry (dentry
, &de
);
2164 inode
= dentry
->d_inode
;
2167 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2170 if (!inode
->i_nlink
) {
2171 ext4_warning (inode
->i_sb
, "ext4_unlink",
2172 "Deleting nonexistent file (%lu), %d",
2173 inode
->i_ino
, inode
->i_nlink
);
2176 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2179 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2180 ext4_update_dx_flag(dir
);
2181 ext4_mark_inode_dirty(handle
, dir
);
2182 ext4_dec_count(handle
, inode
);
2183 if (!inode
->i_nlink
)
2184 ext4_orphan_add(handle
, inode
);
2185 inode
->i_ctime
= ext4_current_time(inode
);
2186 ext4_mark_inode_dirty(handle
, inode
);
2190 ext4_journal_stop(handle
);
2195 static int ext4_symlink (struct inode
* dir
,
2196 struct dentry
*dentry
, const char * symname
)
2199 struct inode
* inode
;
2200 int l
, err
, retries
= 0;
2202 l
= strlen(symname
)+1;
2203 if (l
> dir
->i_sb
->s_blocksize
)
2204 return -ENAMETOOLONG
;
2207 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2208 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2209 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2211 return PTR_ERR(handle
);
2213 if (IS_DIRSYNC(dir
))
2216 inode
= ext4_new_inode (handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2217 err
= PTR_ERR(inode
);
2221 if (l
> sizeof (EXT4_I(inode
)->i_data
)) {
2222 inode
->i_op
= &ext4_symlink_inode_operations
;
2223 ext4_set_aops(inode
);
2225 * page_symlink() calls into ext4_prepare/commit_write.
2226 * We have a transaction open. All is sweetness. It also sets
2227 * i_size in generic_commit_write().
2229 err
= __page_symlink(inode
, symname
, l
,
2230 mapping_gfp_mask(inode
->i_mapping
) & ~__GFP_FS
);
2232 ext4_dec_count(handle
, inode
);
2233 ext4_mark_inode_dirty(handle
, inode
);
2238 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2239 memcpy((char*)&EXT4_I(inode
)->i_data
,symname
,l
);
2240 inode
->i_size
= l
-1;
2242 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2243 err
= ext4_add_nondir(handle
, dentry
, inode
);
2245 ext4_journal_stop(handle
);
2246 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2251 static int ext4_link (struct dentry
* old_dentry
,
2252 struct inode
* dir
, struct dentry
*dentry
)
2255 struct inode
*inode
= old_dentry
->d_inode
;
2256 int err
, retries
= 0;
2258 if (EXT4_DIR_LINK_MAX(inode
))
2262 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2263 * otherwise has the potential to corrupt the orphan inode list.
2265 if (inode
->i_nlink
== 0)
2269 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2270 EXT4_INDEX_EXTRA_TRANS_BLOCKS
);
2272 return PTR_ERR(handle
);
2274 if (IS_DIRSYNC(dir
))
2277 inode
->i_ctime
= ext4_current_time(inode
);
2278 ext4_inc_count(handle
, inode
);
2279 atomic_inc(&inode
->i_count
);
2281 err
= ext4_add_nondir(handle
, dentry
, inode
);
2282 ext4_journal_stop(handle
);
2283 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2288 #define PARENT_INO(buffer) \
2289 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2292 * Anybody can rename anything with this: the permission checks are left to the
2293 * higher-level routines.
2295 static int ext4_rename (struct inode
* old_dir
, struct dentry
*old_dentry
,
2296 struct inode
* new_dir
,struct dentry
*new_dentry
)
2299 struct inode
* old_inode
, * new_inode
;
2300 struct buffer_head
* old_bh
, * new_bh
, * dir_bh
;
2301 struct ext4_dir_entry_2
* old_de
, * new_de
;
2304 old_bh
= new_bh
= dir_bh
= NULL
;
2306 /* Initialize quotas before so that eventual writes go
2307 * in separate transaction */
2308 if (new_dentry
->d_inode
)
2309 DQUOT_INIT(new_dentry
->d_inode
);
2310 handle
= ext4_journal_start(old_dir
, 2 *
2311 EXT4_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2312 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2314 return PTR_ERR(handle
);
2316 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2319 old_bh
= ext4_find_entry (old_dentry
, &old_de
);
2321 * Check for inode number is _not_ due to possible IO errors.
2322 * We might rmdir the source, keep it as pwd of some process
2323 * and merrily kill the link to whatever was created under the
2324 * same name. Goodbye sticky bit ;-<
2326 old_inode
= old_dentry
->d_inode
;
2328 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2331 new_inode
= new_dentry
->d_inode
;
2332 new_bh
= ext4_find_entry (new_dentry
, &new_de
);
2339 if (S_ISDIR(old_inode
->i_mode
)) {
2341 retval
= -ENOTEMPTY
;
2342 if (!empty_dir (new_inode
))
2346 dir_bh
= ext4_bread (handle
, old_inode
, 0, 0, &retval
);
2349 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2352 if (!new_inode
&& new_dir
!=old_dir
&&
2353 new_dir
->i_nlink
>= EXT4_LINK_MAX
)
2357 retval
= ext4_add_entry (handle
, new_dentry
, old_inode
);
2361 BUFFER_TRACE(new_bh
, "get write access");
2362 ext4_journal_get_write_access(handle
, new_bh
);
2363 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2364 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2365 EXT4_FEATURE_INCOMPAT_FILETYPE
))
2366 new_de
->file_type
= old_de
->file_type
;
2367 new_dir
->i_version
++;
2368 BUFFER_TRACE(new_bh
, "call ext4_journal_dirty_metadata");
2369 ext4_journal_dirty_metadata(handle
, new_bh
);
2375 * Like most other Unix systems, set the ctime for inodes on a
2378 old_inode
->i_ctime
= ext4_current_time(old_inode
);
2379 ext4_mark_inode_dirty(handle
, old_inode
);
2384 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2385 old_de
->name_len
!= old_dentry
->d_name
.len
||
2386 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2387 (retval
= ext4_delete_entry(handle
, old_dir
,
2388 old_de
, old_bh
)) == -ENOENT
) {
2389 /* old_de could have moved from under us during htree split, so
2390 * make sure that we are deleting the right entry. We might
2391 * also be pointing to a stale entry in the unused part of
2392 * old_bh so just checking inum and the name isn't enough. */
2393 struct buffer_head
*old_bh2
;
2394 struct ext4_dir_entry_2
*old_de2
;
2396 old_bh2
= ext4_find_entry(old_dentry
, &old_de2
);
2398 retval
= ext4_delete_entry(handle
, old_dir
,
2404 ext4_warning(old_dir
->i_sb
, "ext4_rename",
2405 "Deleting old file (%lu), %d, error=%d",
2406 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2410 ext4_dec_count(handle
, new_inode
);
2411 new_inode
->i_ctime
= ext4_current_time(new_inode
);
2413 old_dir
->i_ctime
= old_dir
->i_mtime
= ext4_current_time(old_dir
);
2414 ext4_update_dx_flag(old_dir
);
2416 BUFFER_TRACE(dir_bh
, "get_write_access");
2417 ext4_journal_get_write_access(handle
, dir_bh
);
2418 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2419 BUFFER_TRACE(dir_bh
, "call ext4_journal_dirty_metadata");
2420 ext4_journal_dirty_metadata(handle
, dir_bh
);
2421 ext4_dec_count(handle
, old_dir
);
2423 /* checked empty_dir above, can't have another parent,
2424 * ext3_dec_count() won't work for many-linked dirs */
2425 new_inode
->i_nlink
= 0;
2427 ext4_inc_count(handle
, new_dir
);
2428 ext4_update_dx_flag(new_dir
);
2429 ext4_mark_inode_dirty(handle
, new_dir
);
2432 ext4_mark_inode_dirty(handle
, old_dir
);
2434 ext4_mark_inode_dirty(handle
, new_inode
);
2435 if (!new_inode
->i_nlink
)
2436 ext4_orphan_add(handle
, new_inode
);
2444 ext4_journal_stop(handle
);
2449 * directories can handle most operations...
2451 const struct inode_operations ext4_dir_inode_operations
= {
2452 .create
= ext4_create
,
2453 .lookup
= ext4_lookup
,
2455 .unlink
= ext4_unlink
,
2456 .symlink
= ext4_symlink
,
2457 .mkdir
= ext4_mkdir
,
2458 .rmdir
= ext4_rmdir
,
2459 .mknod
= ext4_mknod
,
2460 .rename
= ext4_rename
,
2461 .setattr
= ext4_setattr
,
2462 #ifdef CONFIG_EXT4DEV_FS_XATTR
2463 .setxattr
= generic_setxattr
,
2464 .getxattr
= generic_getxattr
,
2465 .listxattr
= ext4_listxattr
,
2466 .removexattr
= generic_removexattr
,
2468 .permission
= ext4_permission
,
2471 const struct inode_operations ext4_special_inode_operations
= {
2472 .setattr
= ext4_setattr
,
2473 #ifdef CONFIG_EXT4DEV_FS_XATTR
2474 .setxattr
= generic_setxattr
,
2475 .getxattr
= generic_getxattr
,
2476 .listxattr
= ext4_listxattr
,
2477 .removexattr
= generic_removexattr
,
2479 .permission
= ext4_permission
,