2 * linux/fs/ext3/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/jbd.h>
30 #include <linux/time.h>
31 #include <linux/ext3_fs.h>
32 #include <linux/ext3_jbd.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/smp_lock.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
*ext3_append(handle_t
*handle
,
56 struct buffer_head
*bh
;
58 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
60 if ((bh
= ext3_bread(handle
, inode
, *block
, 1, err
))) {
61 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
62 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
63 ext3_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
;
146 #ifdef CONFIG_EXT3_INDEX
147 static inline unsigned dx_get_block (struct dx_entry
*entry
);
148 static void dx_set_block (struct dx_entry
*entry
, unsigned 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 ext3_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 ext3_dir_entry_2
*dx_move_dirents (char *from
, char *to
,
167 struct dx_map_entry
*offsets
, int count
);
168 static struct ext3_dir_entry_2
* dx_pack_dirents (char *base
, int size
);
169 static void dx_insert_block (struct dx_frame
*frame
, u32 hash
, u32 block
);
170 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
171 struct dx_frame
*frame
,
172 struct dx_frame
*frames
,
174 static struct buffer_head
* ext3_dx_find_entry(struct dentry
*dentry
,
175 struct ext3_dir_entry_2
**res_dir
, int *err
);
176 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
177 struct inode
*inode
);
180 * Future: use high four bits of block for coalesce-on-delete flags
181 * Mask them off for now.
184 static inline unsigned dx_get_block (struct dx_entry
*entry
)
186 return le32_to_cpu(entry
->block
) & 0x00ffffff;
189 static inline void dx_set_block (struct dx_entry
*entry
, unsigned value
)
191 entry
->block
= cpu_to_le32(value
);
194 static inline unsigned dx_get_hash (struct dx_entry
*entry
)
196 return le32_to_cpu(entry
->hash
);
199 static inline void dx_set_hash (struct dx_entry
*entry
, unsigned value
)
201 entry
->hash
= cpu_to_le32(value
);
204 static inline unsigned dx_get_count (struct dx_entry
*entries
)
206 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
209 static inline unsigned dx_get_limit (struct dx_entry
*entries
)
211 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
214 static inline void dx_set_count (struct dx_entry
*entries
, unsigned value
)
216 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
219 static inline void dx_set_limit (struct dx_entry
*entries
, unsigned value
)
221 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
224 static inline unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
)
226 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(1) -
227 EXT3_DIR_REC_LEN(2) - infosize
;
228 return 0? 20: entry_space
/ sizeof(struct dx_entry
);
231 static inline unsigned dx_node_limit (struct inode
*dir
)
233 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(0);
234 return 0? 22: entry_space
/ sizeof(struct dx_entry
);
241 static void dx_show_index (char * label
, struct dx_entry
*entries
)
243 int i
, n
= dx_get_count (entries
);
244 printk("%s index ", label
);
245 for (i
= 0; i
< n
; i
++)
247 printk("%x->%u ", i
? dx_get_hash(entries
+ i
): 0, dx_get_block(entries
+ i
));
259 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext3_dir_entry_2
*de
,
260 int size
, int show_names
)
262 unsigned names
= 0, space
= 0;
263 char *base
= (char *) de
;
264 struct dx_hash_info h
= *hinfo
;
267 while ((char *) de
< base
+ size
)
273 int len
= de
->name_len
;
274 char *name
= de
->name
;
275 while (len
--) printk("%c", *name
++);
276 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
277 printk(":%x.%u ", h
.hash
,
278 ((char *) de
- base
));
280 space
+= EXT3_DIR_REC_LEN(de
->name_len
);
283 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ le16_to_cpu(de
->rec_len
));
285 printk("(%i)\n", names
);
286 return (struct stats
) { names
, space
, 1 };
289 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
290 struct dx_entry
*entries
, int levels
)
292 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
293 unsigned count
= dx_get_count (entries
), names
= 0, space
= 0, i
;
295 struct buffer_head
*bh
;
297 printk("%i indexed blocks...\n", count
);
298 for (i
= 0; i
< count
; i
++, entries
++)
300 u32 block
= dx_get_block(entries
), hash
= i
? dx_get_hash(entries
): 0;
301 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
303 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
304 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0,&err
))) continue;
306 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
307 dx_show_leaf(hinfo
, (struct ext3_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
308 names
+= stats
.names
;
309 space
+= stats
.space
;
310 bcount
+= stats
.bcount
;
314 printk("%snames %u, fullness %u (%u%%)\n", levels
?"":" ",
315 names
, space
/bcount
,(space
/bcount
)*100/blocksize
);
316 return (struct stats
) { names
, space
, bcount
};
318 #endif /* DX_DEBUG */
321 * Probe for a directory leaf block to search.
323 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324 * error in the directory index, and the caller should fall back to
325 * searching the directory normally. The callers of dx_probe **MUST**
326 * check for this error code, and make sure it never gets reflected
329 static struct dx_frame
*
330 dx_probe(struct dentry
*dentry
, struct inode
*dir
,
331 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
333 unsigned count
, indirect
;
334 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
335 struct dx_root
*root
;
336 struct buffer_head
*bh
;
337 struct dx_frame
*frame
= frame_in
;
342 dir
= dentry
->d_parent
->d_inode
;
343 if (!(bh
= ext3_bread (NULL
,dir
, 0, 0, err
)))
345 root
= (struct dx_root
*) bh
->b_data
;
346 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
347 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
348 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
349 ext3_warning(dir
->i_sb
, __FUNCTION__
,
350 "Unrecognised inode hash code %d",
351 root
->info
.hash_version
);
353 *err
= ERR_BAD_DX_DIR
;
356 hinfo
->hash_version
= root
->info
.hash_version
;
357 hinfo
->seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
359 ext3fs_dirhash(dentry
->d_name
.name
, dentry
->d_name
.len
, hinfo
);
362 if (root
->info
.unused_flags
& 1) {
363 ext3_warning(dir
->i_sb
, __FUNCTION__
,
364 "Unimplemented inode hash flags: %#06x",
365 root
->info
.unused_flags
);
367 *err
= ERR_BAD_DX_DIR
;
371 if ((indirect
= root
->info
.indirect_levels
) > 1) {
372 ext3_warning(dir
->i_sb
, __FUNCTION__
,
373 "Unimplemented inode hash depth: %#06x",
374 root
->info
.indirect_levels
);
376 *err
= ERR_BAD_DX_DIR
;
380 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
381 root
->info
.info_length
);
382 assert(dx_get_limit(entries
) == dx_root_limit(dir
,
383 root
->info
.info_length
));
384 dxtrace (printk("Look up %x", hash
));
387 count
= dx_get_count(entries
);
388 assert (count
&& count
<= dx_get_limit(entries
));
390 q
= entries
+ count
- 1;
394 dxtrace(printk("."));
395 if (dx_get_hash(m
) > hash
)
401 if (0) // linear search cross check
403 unsigned n
= count
- 1;
407 dxtrace(printk(","));
408 if (dx_get_hash(++at
) > hash
)
414 assert (at
== p
- 1);
418 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
420 frame
->entries
= entries
;
422 if (!indirect
--) return frame
;
423 if (!(bh
= ext3_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
425 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
426 assert (dx_get_limit(entries
) == dx_node_limit (dir
));
430 while (frame
>= frame_in
) {
438 static void dx_release (struct dx_frame
*frames
)
440 if (frames
[0].bh
== NULL
)
443 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
444 brelse(frames
[1].bh
);
445 brelse(frames
[0].bh
);
449 * This function increments the frame pointer to search the next leaf
450 * block, and reads in the necessary intervening nodes if the search
451 * should be necessary. Whether or not the search is necessary is
452 * controlled by the hash parameter. If the hash value is even, then
453 * the search is only continued if the next block starts with that
454 * hash value. This is used if we are searching for a specific file.
456 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
458 * This function returns 1 if the caller should continue to search,
459 * or 0 if it should not. If there is an error reading one of the
460 * index blocks, it will a negative error code.
462 * If start_hash is non-null, it will be filled in with the starting
463 * hash of the next page.
465 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
466 struct dx_frame
*frame
,
467 struct dx_frame
*frames
,
471 struct buffer_head
*bh
;
472 int err
, num_frames
= 0;
477 * Find the next leaf page by incrementing the frame pointer.
478 * If we run out of entries in the interior node, loop around and
479 * increment pointer in the parent node. When we break out of
480 * this loop, num_frames indicates the number of interior
481 * nodes need to be read.
484 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
493 * If the hash is 1, then continue only if the next page has a
494 * continuation hash of any value. This is used for readdir
495 * handling. Otherwise, check to see if the hash matches the
496 * desired contiuation hash. If it doesn't, return since
497 * there's no point to read in the successive index pages.
499 bhash
= dx_get_hash(p
->at
);
502 if ((hash
& 1) == 0) {
503 if ((bhash
& ~1) != hash
)
507 * If the hash is HASH_NB_ALWAYS, we always go to the next
508 * block so no check is necessary
510 while (num_frames
--) {
511 if (!(bh
= ext3_bread(NULL
, dir
, dx_get_block(p
->at
),
513 return err
; /* Failure */
517 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
524 * p is at least 6 bytes before the end of page
526 static inline struct ext3_dir_entry_2
*ext3_next_entry(struct ext3_dir_entry_2
*p
)
528 return (struct ext3_dir_entry_2
*)((char*)p
+ le16_to_cpu(p
->rec_len
));
532 * This function fills a red-black tree with information from a
533 * directory block. It returns the number directory entries loaded
534 * into the tree. If there is an error it is returned in err.
536 static int htree_dirblock_to_tree(struct file
*dir_file
,
537 struct inode
*dir
, int block
,
538 struct dx_hash_info
*hinfo
,
539 __u32 start_hash
, __u32 start_minor_hash
)
541 struct buffer_head
*bh
;
542 struct ext3_dir_entry_2
*de
, *top
;
545 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block
));
546 if (!(bh
= ext3_bread (NULL
, dir
, block
, 0, &err
)))
549 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
550 top
= (struct ext3_dir_entry_2
*) ((char *) de
+
551 dir
->i_sb
->s_blocksize
-
552 EXT3_DIR_REC_LEN(0));
553 for (; de
< top
; de
= ext3_next_entry(de
)) {
554 ext3fs_dirhash(de
->name
, de
->name_len
, hinfo
);
555 if ((hinfo
->hash
< start_hash
) ||
556 ((hinfo
->hash
== start_hash
) &&
557 (hinfo
->minor_hash
< start_minor_hash
)))
561 if ((err
= ext3_htree_store_dirent(dir_file
,
562 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
574 * This function fills a red-black tree with information from a
575 * directory. We start scanning the directory in hash order, starting
576 * at start_hash and start_minor_hash.
578 * This function returns the number of entries inserted into the tree,
579 * or a negative error code.
581 int ext3_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
582 __u32 start_minor_hash
, __u32
*next_hash
)
584 struct dx_hash_info hinfo
;
585 struct ext3_dir_entry_2
*de
;
586 struct dx_frame frames
[2], *frame
;
593 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash
,
595 dir
= dir_file
->f_dentry
->d_inode
;
596 if (!(EXT3_I(dir
)->i_flags
& EXT3_INDEX_FL
)) {
597 hinfo
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
598 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
599 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
600 start_hash
, start_minor_hash
);
604 hinfo
.hash
= start_hash
;
605 hinfo
.minor_hash
= 0;
606 frame
= dx_probe(NULL
, dir_file
->f_dentry
->d_inode
, &hinfo
, frames
, &err
);
610 /* Add '.' and '..' from the htree header */
611 if (!start_hash
&& !start_minor_hash
) {
612 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
613 if ((err
= ext3_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
617 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
618 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
619 de
= ext3_next_entry(de
);
620 if ((err
= ext3_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
626 block
= dx_get_block(frame
->at
);
627 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
628 start_hash
, start_minor_hash
);
635 ret
= ext3_htree_next_block(dir
, HASH_NB_ALWAYS
,
636 frame
, frames
, &hashval
);
637 *next_hash
= hashval
;
643 * Stop if: (a) there are no more entries, or
644 * (b) we have inserted at least one entry and the
645 * next hash value is not a continuation
648 (count
&& ((hashval
& 1) == 0)))
652 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
662 * Directory block splitting, compacting
665 static int dx_make_map (struct ext3_dir_entry_2
*de
, int size
,
666 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
669 char *base
= (char *) de
;
670 struct dx_hash_info h
= *hinfo
;
672 while ((char *) de
< base
+ size
)
674 if (de
->name_len
&& de
->inode
) {
675 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
677 map_tail
->hash
= h
.hash
;
678 map_tail
->offs
= (u32
) ((char *) de
- base
);
682 /* XXX: do we need to check rec_len == 0 case? -Chris */
683 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ le16_to_cpu(de
->rec_len
));
688 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
690 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
692 /* Combsort until bubble sort doesn't suck */
696 if (count
- 9 < 2) /* 9, 10 -> 11 */
698 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
699 if (p
->hash
< q
->hash
)
702 /* Garden variety bubble sort */
708 if (q
[1].hash
>= q
[0].hash
)
716 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, u32 block
)
718 struct dx_entry
*entries
= frame
->entries
;
719 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
720 int count
= dx_get_count(entries
);
722 assert(count
< dx_get_limit(entries
));
723 assert(old
< entries
+ count
);
724 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
725 dx_set_hash(new, hash
);
726 dx_set_block(new, block
);
727 dx_set_count(entries
, count
+ 1);
732 static void ext3_update_dx_flag(struct inode
*inode
)
734 if (!EXT3_HAS_COMPAT_FEATURE(inode
->i_sb
,
735 EXT3_FEATURE_COMPAT_DIR_INDEX
))
736 EXT3_I(inode
)->i_flags
&= ~EXT3_INDEX_FL
;
740 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
742 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
743 * `de != NULL' is guaranteed by caller.
745 static inline int ext3_match (int len
, const char * const name
,
746 struct ext3_dir_entry_2
* de
)
748 if (len
!= de
->name_len
)
752 return !memcmp(name
, de
->name
, len
);
756 * Returns 0 if not found, -1 on failure, and 1 on success
758 static inline int search_dirblock(struct buffer_head
* bh
,
760 struct dentry
*dentry
,
761 unsigned long offset
,
762 struct ext3_dir_entry_2
** res_dir
)
764 struct ext3_dir_entry_2
* de
;
767 const char *name
= dentry
->d_name
.name
;
768 int namelen
= dentry
->d_name
.len
;
770 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
771 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
772 while ((char *) de
< dlimit
) {
773 /* this code is executed quadratically often */
774 /* do minimal checking `by hand' */
776 if ((char *) de
+ namelen
<= dlimit
&&
777 ext3_match (namelen
, name
, de
)) {
778 /* found a match - just to be sure, do a full check */
779 if (!ext3_check_dir_entry("ext3_find_entry",
780 dir
, de
, bh
, offset
))
785 /* prevent looping on a bad block */
786 de_len
= le16_to_cpu(de
->rec_len
);
790 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ de_len
);
799 * finds an entry in the specified directory with the wanted name. It
800 * returns the cache buffer in which the entry was found, and the entry
801 * itself (as a parameter - res_dir). It does NOT read the inode of the
802 * entry - you'll have to do that yourself if you want to.
804 * The returned buffer_head has ->b_count elevated. The caller is expected
805 * to brelse() it when appropriate.
807 static struct buffer_head
* ext3_find_entry (struct dentry
*dentry
,
808 struct ext3_dir_entry_2
** res_dir
)
810 struct super_block
* sb
;
811 struct buffer_head
* bh_use
[NAMEI_RA_SIZE
];
812 struct buffer_head
* bh
, *ret
= NULL
;
813 unsigned long start
, block
, b
;
814 int ra_max
= 0; /* Number of bh's in the readahead
816 int ra_ptr
= 0; /* Current index into readahead
820 struct inode
*dir
= dentry
->d_parent
->d_inode
;
827 blocksize
= sb
->s_blocksize
;
828 namelen
= dentry
->d_name
.len
;
829 name
= dentry
->d_name
.name
;
830 if (namelen
> EXT3_NAME_LEN
)
832 #ifdef CONFIG_EXT3_INDEX
834 bh
= ext3_dx_find_entry(dentry
, res_dir
, &err
);
836 * On success, or if the error was file not found,
837 * return. Otherwise, fall back to doing a search the
840 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
842 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
845 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
846 start
= EXT3_I(dir
)->i_dir_start_lookup
;
847 if (start
>= nblocks
)
853 * We deal with the read-ahead logic here.
855 if (ra_ptr
>= ra_max
) {
856 /* Refill the readahead buffer */
859 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
861 * Terminate if we reach the end of the
862 * directory and must wrap, or if our
863 * search has finished at this block.
865 if (b
>= nblocks
|| (num
&& block
== start
)) {
866 bh_use
[ra_max
] = NULL
;
870 bh
= ext3_getblk(NULL
, dir
, b
++, 0, &err
);
873 ll_rw_block(READ
, 1, &bh
);
876 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
879 if (!buffer_uptodate(bh
)) {
880 /* read error, skip block & hope for the best */
881 ext3_error(sb
, __FUNCTION__
, "reading directory #%lu "
882 "offset %lu", dir
->i_ino
, block
);
886 i
= search_dirblock(bh
, dir
, dentry
,
887 block
<< EXT3_BLOCK_SIZE_BITS(sb
), res_dir
);
889 EXT3_I(dir
)->i_dir_start_lookup
= block
;
891 goto cleanup_and_exit
;
895 goto cleanup_and_exit
;
898 if (++block
>= nblocks
)
900 } while (block
!= start
);
903 * If the directory has grown while we were searching, then
904 * search the last part of the directory before giving up.
907 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
908 if (block
< nblocks
) {
914 /* Clean up the read-ahead blocks */
915 for (; ra_ptr
< ra_max
; ra_ptr
++)
916 brelse (bh_use
[ra_ptr
]);
920 #ifdef CONFIG_EXT3_INDEX
921 static struct buffer_head
* ext3_dx_find_entry(struct dentry
*dentry
,
922 struct ext3_dir_entry_2
**res_dir
, int *err
)
924 struct super_block
* sb
;
925 struct dx_hash_info hinfo
;
927 struct dx_frame frames
[2], *frame
;
928 struct ext3_dir_entry_2
*de
, *top
;
929 struct buffer_head
*bh
;
932 int namelen
= dentry
->d_name
.len
;
933 const u8
*name
= dentry
->d_name
.name
;
934 struct inode
*dir
= dentry
->d_parent
->d_inode
;
937 /* NFS may look up ".." - look at dx_root directory block */
938 if (namelen
> 2 || name
[0] != '.'||(name
[1] != '.' && name
[1] != '\0')){
939 if (!(frame
= dx_probe(dentry
, NULL
, &hinfo
, frames
, err
)))
943 frame
->bh
= NULL
; /* for dx_release() */
944 frame
->at
= (struct dx_entry
*)frames
; /* hack for zero entry*/
945 dx_set_block(frame
->at
, 0); /* dx_root block is 0 */
949 block
= dx_get_block(frame
->at
);
950 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0, err
)))
952 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
953 top
= (struct ext3_dir_entry_2
*) ((char *) de
+ sb
->s_blocksize
-
954 EXT3_DIR_REC_LEN(0));
955 for (; de
< top
; de
= ext3_next_entry(de
))
956 if (ext3_match (namelen
, name
, de
)) {
957 if (!ext3_check_dir_entry("ext3_find_entry",
959 (block
<<EXT3_BLOCK_SIZE_BITS(sb
))
960 +((char *)de
- bh
->b_data
))) {
969 /* Check to see if we should continue to search */
970 retval
= ext3_htree_next_block(dir
, hash
, frame
,
973 ext3_warning(sb
, __FUNCTION__
,
974 "error reading index page in directory #%lu",
979 } while (retval
== 1);
983 dxtrace(printk("%s not found\n", name
));
989 static struct dentry
*ext3_lookup(struct inode
* dir
, struct dentry
*dentry
, struct nameidata
*nd
)
991 struct inode
* inode
;
992 struct ext3_dir_entry_2
* de
;
993 struct buffer_head
* bh
;
995 if (dentry
->d_name
.len
> EXT3_NAME_LEN
)
996 return ERR_PTR(-ENAMETOOLONG
);
998 bh
= ext3_find_entry(dentry
, &de
);
1001 unsigned long ino
= le32_to_cpu(de
->inode
);
1003 inode
= iget(dir
->i_sb
, ino
);
1006 return ERR_PTR(-EACCES
);
1009 return d_splice_alias(inode
, dentry
);
1010 d_add(dentry
, inode
);
1015 struct dentry
*ext3_get_parent(struct dentry
*child
)
1018 struct dentry
*parent
;
1019 struct inode
*inode
;
1020 struct dentry dotdot
;
1021 struct ext3_dir_entry_2
* de
;
1022 struct buffer_head
*bh
;
1024 dotdot
.d_name
.name
= "..";
1025 dotdot
.d_name
.len
= 2;
1026 dotdot
.d_parent
= child
; /* confusing, isn't it! */
1028 bh
= ext3_find_entry(&dotdot
, &de
);
1031 return ERR_PTR(-ENOENT
);
1032 ino
= le32_to_cpu(de
->inode
);
1034 inode
= iget(child
->d_inode
->i_sb
, ino
);
1037 return ERR_PTR(-EACCES
);
1039 parent
= d_alloc_anon(inode
);
1042 parent
= ERR_PTR(-ENOMEM
);
1048 static unsigned char ext3_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1049 [S_IFREG
>> S_SHIFT
] = EXT3_FT_REG_FILE
,
1050 [S_IFDIR
>> S_SHIFT
] = EXT3_FT_DIR
,
1051 [S_IFCHR
>> S_SHIFT
] = EXT3_FT_CHRDEV
,
1052 [S_IFBLK
>> S_SHIFT
] = EXT3_FT_BLKDEV
,
1053 [S_IFIFO
>> S_SHIFT
] = EXT3_FT_FIFO
,
1054 [S_IFSOCK
>> S_SHIFT
] = EXT3_FT_SOCK
,
1055 [S_IFLNK
>> S_SHIFT
] = EXT3_FT_SYMLINK
,
1058 static inline void ext3_set_de_type(struct super_block
*sb
,
1059 struct ext3_dir_entry_2
*de
,
1061 if (EXT3_HAS_INCOMPAT_FEATURE(sb
, EXT3_FEATURE_INCOMPAT_FILETYPE
))
1062 de
->file_type
= ext3_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1065 #ifdef CONFIG_EXT3_INDEX
1066 static struct ext3_dir_entry_2
*
1067 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1069 unsigned rec_len
= 0;
1072 struct ext3_dir_entry_2
*de
= (struct ext3_dir_entry_2
*) (from
+ map
->offs
);
1073 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1074 memcpy (to
, de
, rec_len
);
1075 ((struct ext3_dir_entry_2
*) to
)->rec_len
=
1076 cpu_to_le16(rec_len
);
1081 return (struct ext3_dir_entry_2
*) (to
- rec_len
);
1084 static struct ext3_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1086 struct ext3_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext3_dir_entry_2
*) base
;
1087 unsigned rec_len
= 0;
1090 while ((char*)de
< base
+ size
) {
1091 next
= (struct ext3_dir_entry_2
*) ((char *) de
+
1092 le16_to_cpu(de
->rec_len
));
1093 if (de
->inode
&& de
->name_len
) {
1094 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1096 memmove(to
, de
, rec_len
);
1097 to
->rec_len
= cpu_to_le16(rec_len
);
1099 to
= (struct ext3_dir_entry_2
*) (((char *) to
) + rec_len
);
1106 static struct ext3_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1107 struct buffer_head
**bh
,struct dx_frame
*frame
,
1108 struct dx_hash_info
*hinfo
, int *error
)
1110 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1111 unsigned count
, continued
;
1112 struct buffer_head
*bh2
;
1115 struct dx_map_entry
*map
;
1116 char *data1
= (*bh
)->b_data
, *data2
;
1118 struct ext3_dir_entry_2
*de
= NULL
, *de2
;
1121 bh2
= ext3_append (handle
, dir
, &newblock
, error
);
1128 BUFFER_TRACE(*bh
, "get_write_access");
1129 err
= ext3_journal_get_write_access(handle
, *bh
);
1135 ext3_std_error(dir
->i_sb
, err
);
1138 BUFFER_TRACE(frame
->bh
, "get_write_access");
1139 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1143 data2
= bh2
->b_data
;
1145 /* create map in the end of data2 block */
1146 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1147 count
= dx_make_map ((struct ext3_dir_entry_2
*) data1
,
1148 blocksize
, hinfo
, map
);
1150 split
= count
/2; // need to adjust to actual middle
1151 dx_sort_map (map
, count
);
1152 hash2
= map
[split
].hash
;
1153 continued
= hash2
== map
[split
- 1].hash
;
1154 dxtrace(printk("Split block %i at %x, %i/%i\n",
1155 dx_get_block(frame
->at
), hash2
, split
, count
-split
));
1157 /* Fancy dance to stay within two buffers */
1158 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1159 de
= dx_pack_dirents(data1
,blocksize
);
1160 de
->rec_len
= cpu_to_le16(data1
+ blocksize
- (char *) de
);
1161 de2
->rec_len
= cpu_to_le16(data2
+ blocksize
- (char *) de2
);
1162 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data1
, blocksize
, 1));
1163 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data2
, blocksize
, 1));
1165 /* Which block gets the new entry? */
1166 if (hinfo
->hash
>= hash2
)
1171 dx_insert_block (frame
, hash2
+ continued
, newblock
);
1172 err
= ext3_journal_dirty_metadata (handle
, bh2
);
1175 err
= ext3_journal_dirty_metadata (handle
, frame
->bh
);
1179 dxtrace(dx_show_index ("frame", frame
->entries
));
1187 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1188 * it points to a directory entry which is guaranteed to be large
1189 * enough for new directory entry. If de is NULL, then
1190 * add_dirent_to_buf will attempt search the directory block for
1191 * space. It will return -ENOSPC if no space is available, and -EIO
1192 * and -EEXIST if directory entry already exists.
1194 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1195 * all other cases bh is released.
1197 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1198 struct inode
*inode
, struct ext3_dir_entry_2
*de
,
1199 struct buffer_head
* bh
)
1201 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1202 const char *name
= dentry
->d_name
.name
;
1203 int namelen
= dentry
->d_name
.len
;
1204 unsigned long offset
= 0;
1205 unsigned short reclen
;
1206 int nlen
, rlen
, err
;
1209 reclen
= EXT3_DIR_REC_LEN(namelen
);
1211 de
= (struct ext3_dir_entry_2
*)bh
->b_data
;
1212 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1213 while ((char *) de
<= top
) {
1214 if (!ext3_check_dir_entry("ext3_add_entry", dir
, de
,
1219 if (ext3_match (namelen
, name
, de
)) {
1223 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1224 rlen
= le16_to_cpu(de
->rec_len
);
1225 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1227 de
= (struct ext3_dir_entry_2
*)((char *)de
+ rlen
);
1230 if ((char *) de
> top
)
1233 BUFFER_TRACE(bh
, "get_write_access");
1234 err
= ext3_journal_get_write_access(handle
, bh
);
1236 ext3_std_error(dir
->i_sb
, err
);
1241 /* By now the buffer is marked for journaling */
1242 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1243 rlen
= le16_to_cpu(de
->rec_len
);
1245 struct ext3_dir_entry_2
*de1
= (struct ext3_dir_entry_2
*)((char *)de
+ nlen
);
1246 de1
->rec_len
= cpu_to_le16(rlen
- nlen
);
1247 de
->rec_len
= cpu_to_le16(nlen
);
1250 de
->file_type
= EXT3_FT_UNKNOWN
;
1252 de
->inode
= cpu_to_le32(inode
->i_ino
);
1253 ext3_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1256 de
->name_len
= namelen
;
1257 memcpy (de
->name
, name
, namelen
);
1259 * XXX shouldn't update any times until successful
1260 * completion of syscall, but too many callers depend
1263 * XXX similarly, too many callers depend on
1264 * ext3_new_inode() setting the times, but error
1265 * recovery deletes the inode, so the worst that can
1266 * happen is that the times are slightly out of date
1267 * and/or different from the directory change time.
1269 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME_SEC
;
1270 ext3_update_dx_flag(dir
);
1272 ext3_mark_inode_dirty(handle
, dir
);
1273 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1274 err
= ext3_journal_dirty_metadata(handle
, bh
);
1276 ext3_std_error(dir
->i_sb
, err
);
1281 #ifdef CONFIG_EXT3_INDEX
1283 * This converts a one block unindexed directory to a 3 block indexed
1284 * directory, and adds the dentry to the indexed directory.
1286 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1287 struct inode
*inode
, struct buffer_head
*bh
)
1289 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1290 const char *name
= dentry
->d_name
.name
;
1291 int namelen
= dentry
->d_name
.len
;
1292 struct buffer_head
*bh2
;
1293 struct dx_root
*root
;
1294 struct dx_frame frames
[2], *frame
;
1295 struct dx_entry
*entries
;
1296 struct ext3_dir_entry_2
*de
, *de2
;
1301 struct dx_hash_info hinfo
;
1303 struct fake_dirent
*fde
;
1305 blocksize
= dir
->i_sb
->s_blocksize
;
1306 dxtrace(printk("Creating index\n"));
1307 retval
= ext3_journal_get_write_access(handle
, bh
);
1309 ext3_std_error(dir
->i_sb
, retval
);
1313 root
= (struct dx_root
*) bh
->b_data
;
1315 bh2
= ext3_append (handle
, dir
, &block
, &retval
);
1320 EXT3_I(dir
)->i_flags
|= EXT3_INDEX_FL
;
1321 data1
= bh2
->b_data
;
1323 /* The 0th block becomes the root, move the dirents out */
1324 fde
= &root
->dotdot
;
1325 de
= (struct ext3_dir_entry_2
*)((char *)fde
+ le16_to_cpu(fde
->rec_len
));
1326 len
= ((char *) root
) + blocksize
- (char *) de
;
1327 memcpy (data1
, de
, len
);
1328 de
= (struct ext3_dir_entry_2
*) data1
;
1330 while ((char *)(de2
=(void*)de
+le16_to_cpu(de
->rec_len
)) < top
)
1332 de
->rec_len
= cpu_to_le16(data1
+ blocksize
- (char *) de
);
1333 /* Initialize the root; the dot dirents already exist */
1334 de
= (struct ext3_dir_entry_2
*) (&root
->dotdot
);
1335 de
->rec_len
= cpu_to_le16(blocksize
- EXT3_DIR_REC_LEN(2));
1336 memset (&root
->info
, 0, sizeof(root
->info
));
1337 root
->info
.info_length
= sizeof(root
->info
);
1338 root
->info
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
1339 entries
= root
->entries
;
1340 dx_set_block (entries
, 1);
1341 dx_set_count (entries
, 1);
1342 dx_set_limit (entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1344 /* Initialize as for dx_probe */
1345 hinfo
.hash_version
= root
->info
.hash_version
;
1346 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
1347 ext3fs_dirhash(name
, namelen
, &hinfo
);
1349 frame
->entries
= entries
;
1350 frame
->at
= entries
;
1353 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1354 dx_release (frames
);
1358 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1365 * adds a file entry to the specified directory, using the same
1366 * semantics as ext3_find_entry(). It returns NULL if it failed.
1368 * NOTE!! The inode part of 'de' is left at 0 - which means you
1369 * may not sleep between calling this and putting something into
1370 * the entry, as someone else might have used it while you slept.
1372 static int ext3_add_entry (handle_t
*handle
, struct dentry
*dentry
,
1373 struct inode
*inode
)
1375 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1376 unsigned long offset
;
1377 struct buffer_head
* bh
;
1378 struct ext3_dir_entry_2
*de
;
1379 struct super_block
* sb
;
1381 #ifdef CONFIG_EXT3_INDEX
1385 unsigned nlen
, rlen
;
1389 blocksize
= sb
->s_blocksize
;
1390 if (!dentry
->d_name
.len
)
1392 #ifdef CONFIG_EXT3_INDEX
1394 retval
= ext3_dx_add_entry(handle
, dentry
, inode
);
1395 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1397 EXT3_I(dir
)->i_flags
&= ~EXT3_INDEX_FL
;
1399 ext3_mark_inode_dirty(handle
, dir
);
1402 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1403 for (block
= 0, offset
= 0; block
< blocks
; block
++) {
1404 bh
= ext3_bread(handle
, dir
, block
, 0, &retval
);
1407 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1408 if (retval
!= -ENOSPC
)
1411 #ifdef CONFIG_EXT3_INDEX
1412 if (blocks
== 1 && !dx_fallback
&&
1413 EXT3_HAS_COMPAT_FEATURE(sb
, EXT3_FEATURE_COMPAT_DIR_INDEX
))
1414 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1418 bh
= ext3_append(handle
, dir
, &block
, &retval
);
1421 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1423 de
->rec_len
= cpu_to_le16(rlen
= blocksize
);
1425 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1428 #ifdef CONFIG_EXT3_INDEX
1430 * Returns 0 for success, or a negative error value
1432 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1433 struct inode
*inode
)
1435 struct dx_frame frames
[2], *frame
;
1436 struct dx_entry
*entries
, *at
;
1437 struct dx_hash_info hinfo
;
1438 struct buffer_head
* bh
;
1439 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1440 struct super_block
* sb
= dir
->i_sb
;
1441 struct ext3_dir_entry_2
*de
;
1444 frame
= dx_probe(dentry
, NULL
, &hinfo
, frames
, &err
);
1447 entries
= frame
->entries
;
1450 if (!(bh
= ext3_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1453 BUFFER_TRACE(bh
, "get_write_access");
1454 err
= ext3_journal_get_write_access(handle
, bh
);
1458 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1459 if (err
!= -ENOSPC
) {
1464 /* Block full, should compress but for now just split */
1465 dxtrace(printk("using %u of %u node entries\n",
1466 dx_get_count(entries
), dx_get_limit(entries
)));
1467 /* Need to split index? */
1468 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1470 unsigned icount
= dx_get_count(entries
);
1471 int levels
= frame
- frames
;
1472 struct dx_entry
*entries2
;
1473 struct dx_node
*node2
;
1474 struct buffer_head
*bh2
;
1476 if (levels
&& (dx_get_count(frames
->entries
) ==
1477 dx_get_limit(frames
->entries
))) {
1478 ext3_warning(sb
, __FUNCTION__
,
1479 "Directory index full!\n");
1483 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1486 node2
= (struct dx_node
*)(bh2
->b_data
);
1487 entries2
= node2
->entries
;
1488 node2
->fake
.rec_len
= cpu_to_le16(sb
->s_blocksize
);
1489 node2
->fake
.inode
= 0;
1490 BUFFER_TRACE(frame
->bh
, "get_write_access");
1491 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1495 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1496 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1497 dxtrace(printk("Split index %i/%i\n", icount1
, icount2
));
1499 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1500 err
= ext3_journal_get_write_access(handle
,
1505 memcpy ((char *) entries2
, (char *) (entries
+ icount1
),
1506 icount2
* sizeof(struct dx_entry
));
1507 dx_set_count (entries
, icount1
);
1508 dx_set_count (entries2
, icount2
);
1509 dx_set_limit (entries2
, dx_node_limit(dir
));
1511 /* Which index block gets the new entry? */
1512 if (at
- entries
>= icount1
) {
1513 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1514 frame
->entries
= entries
= entries2
;
1515 swap(frame
->bh
, bh2
);
1517 dx_insert_block (frames
+ 0, hash2
, newblock
);
1518 dxtrace(dx_show_index ("node", frames
[1].entries
));
1519 dxtrace(dx_show_index ("node",
1520 ((struct dx_node
*) bh2
->b_data
)->entries
));
1521 err
= ext3_journal_dirty_metadata(handle
, bh2
);
1526 dxtrace(printk("Creating second level index...\n"));
1527 memcpy((char *) entries2
, (char *) entries
,
1528 icount
* sizeof(struct dx_entry
));
1529 dx_set_limit(entries2
, dx_node_limit(dir
));
1532 dx_set_count(entries
, 1);
1533 dx_set_block(entries
+ 0, newblock
);
1534 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1536 /* Add new access path frame */
1538 frame
->at
= at
= at
- entries
+ entries2
;
1539 frame
->entries
= entries
= entries2
;
1541 err
= ext3_journal_get_write_access(handle
,
1546 ext3_journal_dirty_metadata(handle
, frames
[0].bh
);
1548 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1551 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1556 ext3_std_error(dir
->i_sb
, err
);
1566 * ext3_delete_entry deletes a directory entry by merging it with the
1569 static int ext3_delete_entry (handle_t
*handle
,
1571 struct ext3_dir_entry_2
* de_del
,
1572 struct buffer_head
* bh
)
1574 struct ext3_dir_entry_2
* de
, * pde
;
1579 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1580 while (i
< bh
->b_size
) {
1581 if (!ext3_check_dir_entry("ext3_delete_entry", dir
, de
, bh
, i
))
1584 BUFFER_TRACE(bh
, "get_write_access");
1585 ext3_journal_get_write_access(handle
, bh
);
1588 cpu_to_le16(le16_to_cpu(pde
->rec_len
) +
1589 le16_to_cpu(de
->rec_len
));
1593 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1594 ext3_journal_dirty_metadata(handle
, bh
);
1597 i
+= le16_to_cpu(de
->rec_len
);
1599 de
= (struct ext3_dir_entry_2
*)
1600 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1606 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1607 * do not perform it in these functions. We perform it at the call site,
1610 static inline void ext3_inc_count(handle_t
*handle
, struct inode
*inode
)
1615 static inline void ext3_dec_count(handle_t
*handle
, struct inode
*inode
)
1620 static int ext3_add_nondir(handle_t
*handle
,
1621 struct dentry
*dentry
, struct inode
*inode
)
1623 int err
= ext3_add_entry(handle
, dentry
, inode
);
1625 ext3_mark_inode_dirty(handle
, inode
);
1626 d_instantiate(dentry
, inode
);
1629 ext3_dec_count(handle
, inode
);
1635 * By the time this is called, we already have created
1636 * the directory cache entry for the new file, but it
1637 * is so far negative - it has no inode.
1639 * If the create succeeds, we fill in the inode information
1640 * with d_instantiate().
1642 static int ext3_create (struct inode
* dir
, struct dentry
* dentry
, int mode
,
1643 struct nameidata
*nd
)
1646 struct inode
* inode
;
1647 int err
, retries
= 0;
1650 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1651 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1652 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1654 return PTR_ERR(handle
);
1656 if (IS_DIRSYNC(dir
))
1659 inode
= ext3_new_inode (handle
, dir
, mode
);
1660 err
= PTR_ERR(inode
);
1661 if (!IS_ERR(inode
)) {
1662 inode
->i_op
= &ext3_file_inode_operations
;
1663 inode
->i_fop
= &ext3_file_operations
;
1664 ext3_set_aops(inode
);
1665 err
= ext3_add_nondir(handle
, dentry
, inode
);
1667 ext3_journal_stop(handle
);
1668 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1673 static int ext3_mknod (struct inode
* dir
, struct dentry
*dentry
,
1674 int mode
, dev_t rdev
)
1677 struct inode
*inode
;
1678 int err
, retries
= 0;
1680 if (!new_valid_dev(rdev
))
1684 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1685 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1686 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1688 return PTR_ERR(handle
);
1690 if (IS_DIRSYNC(dir
))
1693 inode
= ext3_new_inode (handle
, dir
, mode
);
1694 err
= PTR_ERR(inode
);
1695 if (!IS_ERR(inode
)) {
1696 init_special_inode(inode
, inode
->i_mode
, rdev
);
1697 #ifdef CONFIG_EXT3_FS_XATTR
1698 inode
->i_op
= &ext3_special_inode_operations
;
1700 err
= ext3_add_nondir(handle
, dentry
, inode
);
1702 ext3_journal_stop(handle
);
1703 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1708 static int ext3_mkdir(struct inode
* dir
, struct dentry
* dentry
, int mode
)
1711 struct inode
* inode
;
1712 struct buffer_head
* dir_block
;
1713 struct ext3_dir_entry_2
* de
;
1714 int err
, retries
= 0;
1716 if (dir
->i_nlink
>= EXT3_LINK_MAX
)
1720 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1721 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1722 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1724 return PTR_ERR(handle
);
1726 if (IS_DIRSYNC(dir
))
1729 inode
= ext3_new_inode (handle
, dir
, S_IFDIR
| mode
);
1730 err
= PTR_ERR(inode
);
1734 inode
->i_op
= &ext3_dir_inode_operations
;
1735 inode
->i_fop
= &ext3_dir_operations
;
1736 inode
->i_size
= EXT3_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1737 dir_block
= ext3_bread (handle
, inode
, 0, 1, &err
);
1739 inode
->i_nlink
--; /* is this nlink == 0? */
1740 ext3_mark_inode_dirty(handle
, inode
);
1744 BUFFER_TRACE(dir_block
, "get_write_access");
1745 ext3_journal_get_write_access(handle
, dir_block
);
1746 de
= (struct ext3_dir_entry_2
*) dir_block
->b_data
;
1747 de
->inode
= cpu_to_le32(inode
->i_ino
);
1749 de
->rec_len
= cpu_to_le16(EXT3_DIR_REC_LEN(de
->name_len
));
1750 strcpy (de
->name
, ".");
1751 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1752 de
= (struct ext3_dir_entry_2
*)
1753 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1754 de
->inode
= cpu_to_le32(dir
->i_ino
);
1755 de
->rec_len
= cpu_to_le16(inode
->i_sb
->s_blocksize
-EXT3_DIR_REC_LEN(1));
1757 strcpy (de
->name
, "..");
1758 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1760 BUFFER_TRACE(dir_block
, "call ext3_journal_dirty_metadata");
1761 ext3_journal_dirty_metadata(handle
, dir_block
);
1763 ext3_mark_inode_dirty(handle
, inode
);
1764 err
= ext3_add_entry (handle
, dentry
, inode
);
1767 ext3_mark_inode_dirty(handle
, inode
);
1772 ext3_update_dx_flag(dir
);
1773 ext3_mark_inode_dirty(handle
, dir
);
1774 d_instantiate(dentry
, inode
);
1776 ext3_journal_stop(handle
);
1777 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1783 * routine to check that the specified directory is empty (for rmdir)
1785 static int empty_dir (struct inode
* inode
)
1787 unsigned long offset
;
1788 struct buffer_head
* bh
;
1789 struct ext3_dir_entry_2
* de
, * de1
;
1790 struct super_block
* sb
;
1794 if (inode
->i_size
< EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1795 !(bh
= ext3_bread (NULL
, inode
, 0, 0, &err
))) {
1797 ext3_error(inode
->i_sb
, __FUNCTION__
,
1798 "error %d reading directory #%lu offset 0",
1801 ext3_warning(inode
->i_sb
, __FUNCTION__
,
1802 "bad directory (dir #%lu) - no data block",
1806 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1807 de1
= (struct ext3_dir_entry_2
*)
1808 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1809 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1810 !le32_to_cpu(de1
->inode
) ||
1811 strcmp (".", de
->name
) ||
1812 strcmp ("..", de1
->name
)) {
1813 ext3_warning (inode
->i_sb
, "empty_dir",
1814 "bad directory (dir #%lu) - no `.' or `..'",
1819 offset
= le16_to_cpu(de
->rec_len
) + le16_to_cpu(de1
->rec_len
);
1820 de
= (struct ext3_dir_entry_2
*)
1821 ((char *) de1
+ le16_to_cpu(de1
->rec_len
));
1822 while (offset
< inode
->i_size
) {
1824 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1827 bh
= ext3_bread (NULL
, inode
,
1828 offset
>> EXT3_BLOCK_SIZE_BITS(sb
), 0, &err
);
1831 ext3_error(sb
, __FUNCTION__
,
1832 "error %d reading directory"
1834 err
, inode
->i_ino
, offset
);
1835 offset
+= sb
->s_blocksize
;
1838 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1840 if (!ext3_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1841 de
= (struct ext3_dir_entry_2
*)(bh
->b_data
+
1843 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1846 if (le32_to_cpu(de
->inode
)) {
1850 offset
+= le16_to_cpu(de
->rec_len
);
1851 de
= (struct ext3_dir_entry_2
*)
1852 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1858 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1859 * such inodes, starting at the superblock, in case we crash before the
1860 * file is closed/deleted, or in case the inode truncate spans multiple
1861 * transactions and the last transaction is not recovered after a crash.
1863 * At filesystem recovery time, we walk this list deleting unlinked
1864 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1866 int ext3_orphan_add(handle_t
*handle
, struct inode
*inode
)
1868 struct super_block
*sb
= inode
->i_sb
;
1869 struct ext3_iloc iloc
;
1873 if (!list_empty(&EXT3_I(inode
)->i_orphan
))
1876 /* Orphan handling is only valid for files with data blocks
1877 * being truncated, or files being unlinked. */
1879 /* @@@ FIXME: Observation from aviro:
1880 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1881 * here (on lock_super()), so race with ext3_link() which might bump
1882 * ->i_nlink. For, say it, character device. Not a regular file,
1883 * not a directory, not a symlink and ->i_nlink > 0.
1885 J_ASSERT ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1886 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1888 BUFFER_TRACE(EXT3_SB(sb
)->s_sbh
, "get_write_access");
1889 err
= ext3_journal_get_write_access(handle
, EXT3_SB(sb
)->s_sbh
);
1893 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1897 /* Insert this inode at the head of the on-disk orphan list... */
1898 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT3_SB(sb
)->s_es
->s_last_orphan
);
1899 EXT3_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1900 err
= ext3_journal_dirty_metadata(handle
, EXT3_SB(sb
)->s_sbh
);
1901 rc
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
1905 /* Only add to the head of the in-memory list if all the
1906 * previous operations succeeded. If the orphan_add is going to
1907 * fail (possibly taking the journal offline), we can't risk
1908 * leaving the inode on the orphan list: stray orphan-list
1909 * entries can cause panics at unmount time.
1911 * This is safe: on error we're going to ignore the orphan list
1912 * anyway on the next recovery. */
1914 list_add(&EXT3_I(inode
)->i_orphan
, &EXT3_SB(sb
)->s_orphan
);
1916 jbd_debug(4, "superblock will point to %ld\n", inode
->i_ino
);
1917 jbd_debug(4, "orphan inode %ld will point to %d\n",
1918 inode
->i_ino
, NEXT_ORPHAN(inode
));
1921 ext3_std_error(inode
->i_sb
, err
);
1926 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1927 * of such inodes stored on disk, because it is finally being cleaned up.
1929 int ext3_orphan_del(handle_t
*handle
, struct inode
*inode
)
1931 struct list_head
*prev
;
1932 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1933 struct ext3_sb_info
*sbi
;
1934 unsigned long ino_next
;
1935 struct ext3_iloc iloc
;
1938 lock_super(inode
->i_sb
);
1939 if (list_empty(&ei
->i_orphan
)) {
1940 unlock_super(inode
->i_sb
);
1944 ino_next
= NEXT_ORPHAN(inode
);
1945 prev
= ei
->i_orphan
.prev
;
1946 sbi
= EXT3_SB(inode
->i_sb
);
1948 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
1950 list_del_init(&ei
->i_orphan
);
1952 /* If we're on an error path, we may not have a valid
1953 * transaction handle with which to update the orphan list on
1954 * disk, but we still need to remove the inode from the linked
1955 * list in memory. */
1959 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1963 if (prev
== &sbi
->s_orphan
) {
1964 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
1965 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1966 err
= ext3_journal_get_write_access(handle
, sbi
->s_sbh
);
1969 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
1970 err
= ext3_journal_dirty_metadata(handle
, sbi
->s_sbh
);
1972 struct ext3_iloc iloc2
;
1973 struct inode
*i_prev
=
1974 &list_entry(prev
, struct ext3_inode_info
, i_orphan
)->vfs_inode
;
1976 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1977 i_prev
->i_ino
, ino_next
);
1978 err
= ext3_reserve_inode_write(handle
, i_prev
, &iloc2
);
1981 NEXT_ORPHAN(i_prev
) = ino_next
;
1982 err
= ext3_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
1986 NEXT_ORPHAN(inode
) = 0;
1987 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
1990 ext3_std_error(inode
->i_sb
, err
);
1992 unlock_super(inode
->i_sb
);
2000 static int ext3_rmdir (struct inode
* dir
, struct dentry
*dentry
)
2003 struct inode
* inode
;
2004 struct buffer_head
* bh
;
2005 struct ext3_dir_entry_2
* de
;
2008 /* Initialize quotas before so that eventual writes go in
2009 * separate transaction */
2010 DQUOT_INIT(dentry
->d_inode
);
2011 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2013 return PTR_ERR(handle
);
2016 bh
= ext3_find_entry (dentry
, &de
);
2020 if (IS_DIRSYNC(dir
))
2023 inode
= dentry
->d_inode
;
2026 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2029 retval
= -ENOTEMPTY
;
2030 if (!empty_dir (inode
))
2033 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2036 if (inode
->i_nlink
!= 2)
2037 ext3_warning (inode
->i_sb
, "ext3_rmdir",
2038 "empty directory has nlink!=2 (%d)",
2042 /* There's no need to set i_disksize: the fact that i_nlink is
2043 * zero will ensure that the right thing happens during any
2046 ext3_orphan_add(handle
, inode
);
2047 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2048 ext3_mark_inode_dirty(handle
, inode
);
2050 ext3_update_dx_flag(dir
);
2051 ext3_mark_inode_dirty(handle
, dir
);
2054 ext3_journal_stop(handle
);
2059 static int ext3_unlink(struct inode
* dir
, struct dentry
*dentry
)
2062 struct inode
* inode
;
2063 struct buffer_head
* bh
;
2064 struct ext3_dir_entry_2
* de
;
2067 /* Initialize quotas before so that eventual writes go
2068 * in separate transaction */
2069 DQUOT_INIT(dentry
->d_inode
);
2070 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2072 return PTR_ERR(handle
);
2074 if (IS_DIRSYNC(dir
))
2078 bh
= ext3_find_entry (dentry
, &de
);
2082 inode
= dentry
->d_inode
;
2085 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2088 if (!inode
->i_nlink
) {
2089 ext3_warning (inode
->i_sb
, "ext3_unlink",
2090 "Deleting nonexistent file (%lu), %d",
2091 inode
->i_ino
, inode
->i_nlink
);
2094 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2097 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2098 ext3_update_dx_flag(dir
);
2099 ext3_mark_inode_dirty(handle
, dir
);
2101 if (!inode
->i_nlink
)
2102 ext3_orphan_add(handle
, inode
);
2103 inode
->i_ctime
= dir
->i_ctime
;
2104 ext3_mark_inode_dirty(handle
, inode
);
2108 ext3_journal_stop(handle
);
2113 static int ext3_symlink (struct inode
* dir
,
2114 struct dentry
*dentry
, const char * symname
)
2117 struct inode
* inode
;
2118 int l
, err
, retries
= 0;
2120 l
= strlen(symname
)+1;
2121 if (l
> dir
->i_sb
->s_blocksize
)
2122 return -ENAMETOOLONG
;
2125 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2126 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2127 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2129 return PTR_ERR(handle
);
2131 if (IS_DIRSYNC(dir
))
2134 inode
= ext3_new_inode (handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2135 err
= PTR_ERR(inode
);
2139 if (l
> sizeof (EXT3_I(inode
)->i_data
)) {
2140 inode
->i_op
= &ext3_symlink_inode_operations
;
2141 ext3_set_aops(inode
);
2143 * page_symlink() calls into ext3_prepare/commit_write.
2144 * We have a transaction open. All is sweetness. It also sets
2145 * i_size in generic_commit_write().
2147 err
= page_symlink(inode
, symname
, l
);
2149 ext3_dec_count(handle
, inode
);
2150 ext3_mark_inode_dirty(handle
, inode
);
2155 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2156 memcpy((char*)&EXT3_I(inode
)->i_data
,symname
,l
);
2157 inode
->i_size
= l
-1;
2159 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
2160 err
= ext3_add_nondir(handle
, dentry
, inode
);
2162 ext3_journal_stop(handle
);
2163 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2168 static int ext3_link (struct dentry
* old_dentry
,
2169 struct inode
* dir
, struct dentry
*dentry
)
2172 struct inode
*inode
= old_dentry
->d_inode
;
2173 int err
, retries
= 0;
2175 if (inode
->i_nlink
>= EXT3_LINK_MAX
)
2179 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2180 EXT3_INDEX_EXTRA_TRANS_BLOCKS
);
2182 return PTR_ERR(handle
);
2184 if (IS_DIRSYNC(dir
))
2187 inode
->i_ctime
= CURRENT_TIME_SEC
;
2188 ext3_inc_count(handle
, inode
);
2189 atomic_inc(&inode
->i_count
);
2191 err
= ext3_add_nondir(handle
, dentry
, inode
);
2192 ext3_journal_stop(handle
);
2193 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2198 #define PARENT_INO(buffer) \
2199 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2200 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2203 * Anybody can rename anything with this: the permission checks are left to the
2204 * higher-level routines.
2206 static int ext3_rename (struct inode
* old_dir
, struct dentry
*old_dentry
,
2207 struct inode
* new_dir
,struct dentry
*new_dentry
)
2210 struct inode
* old_inode
, * new_inode
;
2211 struct buffer_head
* old_bh
, * new_bh
, * dir_bh
;
2212 struct ext3_dir_entry_2
* old_de
, * new_de
;
2215 old_bh
= new_bh
= dir_bh
= NULL
;
2217 /* Initialize quotas before so that eventual writes go
2218 * in separate transaction */
2219 if (new_dentry
->d_inode
)
2220 DQUOT_INIT(new_dentry
->d_inode
);
2221 handle
= ext3_journal_start(old_dir
, 2 *
2222 EXT3_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2223 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2225 return PTR_ERR(handle
);
2227 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2230 old_bh
= ext3_find_entry (old_dentry
, &old_de
);
2232 * Check for inode number is _not_ due to possible IO errors.
2233 * We might rmdir the source, keep it as pwd of some process
2234 * and merrily kill the link to whatever was created under the
2235 * same name. Goodbye sticky bit ;-<
2237 old_inode
= old_dentry
->d_inode
;
2239 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2242 new_inode
= new_dentry
->d_inode
;
2243 new_bh
= ext3_find_entry (new_dentry
, &new_de
);
2250 if (S_ISDIR(old_inode
->i_mode
)) {
2252 retval
= -ENOTEMPTY
;
2253 if (!empty_dir (new_inode
))
2257 dir_bh
= ext3_bread (handle
, old_inode
, 0, 0, &retval
);
2260 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2263 if (!new_inode
&& new_dir
!=old_dir
&&
2264 new_dir
->i_nlink
>= EXT3_LINK_MAX
)
2268 retval
= ext3_add_entry (handle
, new_dentry
, old_inode
);
2272 BUFFER_TRACE(new_bh
, "get write access");
2273 ext3_journal_get_write_access(handle
, new_bh
);
2274 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2275 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2276 EXT3_FEATURE_INCOMPAT_FILETYPE
))
2277 new_de
->file_type
= old_de
->file_type
;
2278 new_dir
->i_version
++;
2279 BUFFER_TRACE(new_bh
, "call ext3_journal_dirty_metadata");
2280 ext3_journal_dirty_metadata(handle
, new_bh
);
2286 * Like most other Unix systems, set the ctime for inodes on a
2289 old_inode
->i_ctime
= CURRENT_TIME_SEC
;
2290 ext3_mark_inode_dirty(handle
, old_inode
);
2295 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2296 old_de
->name_len
!= old_dentry
->d_name
.len
||
2297 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2298 (retval
= ext3_delete_entry(handle
, old_dir
,
2299 old_de
, old_bh
)) == -ENOENT
) {
2300 /* old_de could have moved from under us during htree split, so
2301 * make sure that we are deleting the right entry. We might
2302 * also be pointing to a stale entry in the unused part of
2303 * old_bh so just checking inum and the name isn't enough. */
2304 struct buffer_head
*old_bh2
;
2305 struct ext3_dir_entry_2
*old_de2
;
2307 old_bh2
= ext3_find_entry(old_dentry
, &old_de2
);
2309 retval
= ext3_delete_entry(handle
, old_dir
,
2315 ext3_warning(old_dir
->i_sb
, "ext3_rename",
2316 "Deleting old file (%lu), %d, error=%d",
2317 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2321 new_inode
->i_nlink
--;
2322 new_inode
->i_ctime
= CURRENT_TIME_SEC
;
2324 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME_SEC
;
2325 ext3_update_dx_flag(old_dir
);
2327 BUFFER_TRACE(dir_bh
, "get_write_access");
2328 ext3_journal_get_write_access(handle
, dir_bh
);
2329 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2330 BUFFER_TRACE(dir_bh
, "call ext3_journal_dirty_metadata");
2331 ext3_journal_dirty_metadata(handle
, dir_bh
);
2334 new_inode
->i_nlink
--;
2337 ext3_update_dx_flag(new_dir
);
2338 ext3_mark_inode_dirty(handle
, new_dir
);
2341 ext3_mark_inode_dirty(handle
, old_dir
);
2343 ext3_mark_inode_dirty(handle
, new_inode
);
2344 if (!new_inode
->i_nlink
)
2345 ext3_orphan_add(handle
, new_inode
);
2353 ext3_journal_stop(handle
);
2358 * directories can handle most operations...
2360 struct inode_operations ext3_dir_inode_operations
= {
2361 .create
= ext3_create
,
2362 .lookup
= ext3_lookup
,
2364 .unlink
= ext3_unlink
,
2365 .symlink
= ext3_symlink
,
2366 .mkdir
= ext3_mkdir
,
2367 .rmdir
= ext3_rmdir
,
2368 .mknod
= ext3_mknod
,
2369 .rename
= ext3_rename
,
2370 .setattr
= ext3_setattr
,
2371 #ifdef CONFIG_EXT3_FS_XATTR
2372 .setxattr
= generic_setxattr
,
2373 .getxattr
= generic_getxattr
,
2374 .listxattr
= ext3_listxattr
,
2375 .removexattr
= generic_removexattr
,
2377 .permission
= ext3_permission
,
2380 struct inode_operations ext3_special_inode_operations
= {
2381 .setattr
= ext3_setattr
,
2382 #ifdef CONFIG_EXT3_FS_XATTR
2383 .setxattr
= generic_setxattr
,
2384 .getxattr
= generic_getxattr
,
2385 .listxattr
= ext3_listxattr
,
2386 .removexattr
= generic_removexattr
,
2388 .permission
= ext3_permission
,