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/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
*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
;
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
);
383 if (dx_get_limit(entries
) != dx_root_limit(dir
,
384 root
->info
.info_length
)) {
385 ext3_warning(dir
->i_sb
, __FUNCTION__
,
386 "dx entry: limit != root limit");
388 *err
= ERR_BAD_DX_DIR
;
392 dxtrace (printk("Look up %x", hash
));
395 count
= dx_get_count(entries
);
396 if (!count
|| count
> dx_get_limit(entries
)) {
397 ext3_warning(dir
->i_sb
, __FUNCTION__
,
398 "dx entry: no count or count > limit");
400 *err
= ERR_BAD_DX_DIR
;
405 q
= entries
+ count
- 1;
409 dxtrace(printk("."));
410 if (dx_get_hash(m
) > hash
)
416 if (0) // linear search cross check
418 unsigned n
= count
- 1;
422 dxtrace(printk(","));
423 if (dx_get_hash(++at
) > hash
)
429 assert (at
== p
- 1);
433 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
435 frame
->entries
= entries
;
437 if (!indirect
--) return frame
;
438 if (!(bh
= ext3_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
440 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
441 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
442 ext3_warning(dir
->i_sb
, __FUNCTION__
,
443 "dx entry: limit != node limit");
445 *err
= ERR_BAD_DX_DIR
;
452 while (frame
>= frame_in
) {
457 if (*err
== ERR_BAD_DX_DIR
)
458 ext3_warning(dir
->i_sb
, __FUNCTION__
,
459 "Corrupt dir inode %ld, running e2fsck is "
460 "recommended.", dir
->i_ino
);
464 static void dx_release (struct dx_frame
*frames
)
466 if (frames
[0].bh
== NULL
)
469 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
470 brelse(frames
[1].bh
);
471 brelse(frames
[0].bh
);
475 * This function increments the frame pointer to search the next leaf
476 * block, and reads in the necessary intervening nodes if the search
477 * should be necessary. Whether or not the search is necessary is
478 * controlled by the hash parameter. If the hash value is even, then
479 * the search is only continued if the next block starts with that
480 * hash value. This is used if we are searching for a specific file.
482 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
484 * This function returns 1 if the caller should continue to search,
485 * or 0 if it should not. If there is an error reading one of the
486 * index blocks, it will a negative error code.
488 * If start_hash is non-null, it will be filled in with the starting
489 * hash of the next page.
491 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
492 struct dx_frame
*frame
,
493 struct dx_frame
*frames
,
497 struct buffer_head
*bh
;
498 int err
, num_frames
= 0;
503 * Find the next leaf page by incrementing the frame pointer.
504 * If we run out of entries in the interior node, loop around and
505 * increment pointer in the parent node. When we break out of
506 * this loop, num_frames indicates the number of interior
507 * nodes need to be read.
510 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
519 * If the hash is 1, then continue only if the next page has a
520 * continuation hash of any value. This is used for readdir
521 * handling. Otherwise, check to see if the hash matches the
522 * desired contiuation hash. If it doesn't, return since
523 * there's no point to read in the successive index pages.
525 bhash
= dx_get_hash(p
->at
);
528 if ((hash
& 1) == 0) {
529 if ((bhash
& ~1) != hash
)
533 * If the hash is HASH_NB_ALWAYS, we always go to the next
534 * block so no check is necessary
536 while (num_frames
--) {
537 if (!(bh
= ext3_bread(NULL
, dir
, dx_get_block(p
->at
),
539 return err
; /* Failure */
543 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
550 * p is at least 6 bytes before the end of page
552 static inline struct ext3_dir_entry_2
*ext3_next_entry(struct ext3_dir_entry_2
*p
)
554 return (struct ext3_dir_entry_2
*)((char*)p
+ le16_to_cpu(p
->rec_len
));
558 * This function fills a red-black tree with information from a
559 * directory block. It returns the number directory entries loaded
560 * into the tree. If there is an error it is returned in err.
562 static int htree_dirblock_to_tree(struct file
*dir_file
,
563 struct inode
*dir
, int block
,
564 struct dx_hash_info
*hinfo
,
565 __u32 start_hash
, __u32 start_minor_hash
)
567 struct buffer_head
*bh
;
568 struct ext3_dir_entry_2
*de
, *top
;
571 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block
));
572 if (!(bh
= ext3_bread (NULL
, dir
, block
, 0, &err
)))
575 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
576 top
= (struct ext3_dir_entry_2
*) ((char *) de
+
577 dir
->i_sb
->s_blocksize
-
578 EXT3_DIR_REC_LEN(0));
579 for (; de
< top
; de
= ext3_next_entry(de
)) {
580 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
581 (block
<<EXT3_BLOCK_SIZE_BITS(dir
->i_sb
))
582 +((char *)de
- bh
->b_data
))) {
583 /* On error, skip the f_pos to the next block. */
584 dir_file
->f_pos
= (dir_file
->f_pos
|
585 (dir
->i_sb
->s_blocksize
- 1)) + 1;
589 ext3fs_dirhash(de
->name
, de
->name_len
, hinfo
);
590 if ((hinfo
->hash
< start_hash
) ||
591 ((hinfo
->hash
== start_hash
) &&
592 (hinfo
->minor_hash
< start_minor_hash
)))
596 if ((err
= ext3_htree_store_dirent(dir_file
,
597 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
609 * This function fills a red-black tree with information from a
610 * directory. We start scanning the directory in hash order, starting
611 * at start_hash and start_minor_hash.
613 * This function returns the number of entries inserted into the tree,
614 * or a negative error code.
616 int ext3_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
617 __u32 start_minor_hash
, __u32
*next_hash
)
619 struct dx_hash_info hinfo
;
620 struct ext3_dir_entry_2
*de
;
621 struct dx_frame frames
[2], *frame
;
628 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash
,
630 dir
= dir_file
->f_path
.dentry
->d_inode
;
631 if (!(EXT3_I(dir
)->i_flags
& EXT3_INDEX_FL
)) {
632 hinfo
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
633 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
634 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
635 start_hash
, start_minor_hash
);
639 hinfo
.hash
= start_hash
;
640 hinfo
.minor_hash
= 0;
641 frame
= dx_probe(NULL
, dir_file
->f_path
.dentry
->d_inode
, &hinfo
, frames
, &err
);
645 /* Add '.' and '..' from the htree header */
646 if (!start_hash
&& !start_minor_hash
) {
647 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
648 if ((err
= ext3_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
652 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
653 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
654 de
= ext3_next_entry(de
);
655 if ((err
= ext3_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
661 block
= dx_get_block(frame
->at
);
662 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
663 start_hash
, start_minor_hash
);
670 ret
= ext3_htree_next_block(dir
, HASH_NB_ALWAYS
,
671 frame
, frames
, &hashval
);
672 *next_hash
= hashval
;
678 * Stop if: (a) there are no more entries, or
679 * (b) we have inserted at least one entry and the
680 * next hash value is not a continuation
683 (count
&& ((hashval
& 1) == 0)))
687 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
697 * Directory block splitting, compacting
701 * Create map of hash values, offsets, and sizes, stored at end of block.
702 * Returns number of entries mapped.
704 static int dx_make_map (struct ext3_dir_entry_2
*de
, int size
,
705 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
708 char *base
= (char *) de
;
709 struct dx_hash_info h
= *hinfo
;
711 while ((char *) de
< base
+ size
)
713 if (de
->name_len
&& de
->inode
) {
714 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
716 map_tail
->hash
= h
.hash
;
717 map_tail
->offs
= (u16
) ((char *) de
- base
);
718 map_tail
->size
= le16_to_cpu(de
->rec_len
);
722 /* XXX: do we need to check rec_len == 0 case? -Chris */
723 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ le16_to_cpu(de
->rec_len
));
728 /* Sort map by hash value */
729 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
731 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
733 /* Combsort until bubble sort doesn't suck */
737 if (count
- 9 < 2) /* 9, 10 -> 11 */
739 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
740 if (p
->hash
< q
->hash
)
743 /* Garden variety bubble sort */
749 if (q
[1].hash
>= q
[0].hash
)
757 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, u32 block
)
759 struct dx_entry
*entries
= frame
->entries
;
760 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
761 int count
= dx_get_count(entries
);
763 assert(count
< dx_get_limit(entries
));
764 assert(old
< entries
+ count
);
765 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
766 dx_set_hash(new, hash
);
767 dx_set_block(new, block
);
768 dx_set_count(entries
, count
+ 1);
771 static void ext3_update_dx_flag(struct inode
*inode
)
773 if (!EXT3_HAS_COMPAT_FEATURE(inode
->i_sb
,
774 EXT3_FEATURE_COMPAT_DIR_INDEX
))
775 EXT3_I(inode
)->i_flags
&= ~EXT3_INDEX_FL
;
779 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
781 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
782 * `de != NULL' is guaranteed by caller.
784 static inline int ext3_match (int len
, const char * const name
,
785 struct ext3_dir_entry_2
* de
)
787 if (len
!= de
->name_len
)
791 return !memcmp(name
, de
->name
, len
);
795 * Returns 0 if not found, -1 on failure, and 1 on success
797 static inline int search_dirblock(struct buffer_head
* bh
,
799 struct dentry
*dentry
,
800 unsigned long offset
,
801 struct ext3_dir_entry_2
** res_dir
)
803 struct ext3_dir_entry_2
* de
;
806 const char *name
= dentry
->d_name
.name
;
807 int namelen
= dentry
->d_name
.len
;
809 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
810 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
811 while ((char *) de
< dlimit
) {
812 /* this code is executed quadratically often */
813 /* do minimal checking `by hand' */
815 if ((char *) de
+ namelen
<= dlimit
&&
816 ext3_match (namelen
, name
, de
)) {
817 /* found a match - just to be sure, do a full check */
818 if (!ext3_check_dir_entry("ext3_find_entry",
819 dir
, de
, bh
, offset
))
824 /* prevent looping on a bad block */
825 de_len
= le16_to_cpu(de
->rec_len
);
829 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ de_len
);
838 * finds an entry in the specified directory with the wanted name. It
839 * returns the cache buffer in which the entry was found, and the entry
840 * itself (as a parameter - res_dir). It does NOT read the inode of the
841 * entry - you'll have to do that yourself if you want to.
843 * The returned buffer_head has ->b_count elevated. The caller is expected
844 * to brelse() it when appropriate.
846 static struct buffer_head
* ext3_find_entry (struct dentry
*dentry
,
847 struct ext3_dir_entry_2
** res_dir
)
849 struct super_block
* sb
;
850 struct buffer_head
* bh_use
[NAMEI_RA_SIZE
];
851 struct buffer_head
* bh
, *ret
= NULL
;
852 unsigned long start
, block
, b
;
853 int ra_max
= 0; /* Number of bh's in the readahead
855 int ra_ptr
= 0; /* Current index into readahead
859 struct inode
*dir
= dentry
->d_parent
->d_inode
;
866 blocksize
= sb
->s_blocksize
;
867 namelen
= dentry
->d_name
.len
;
868 name
= dentry
->d_name
.name
;
869 if (namelen
> EXT3_NAME_LEN
)
872 bh
= ext3_dx_find_entry(dentry
, res_dir
, &err
);
874 * On success, or if the error was file not found,
875 * return. Otherwise, fall back to doing a search the
878 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
880 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
882 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
883 start
= EXT3_I(dir
)->i_dir_start_lookup
;
884 if (start
>= nblocks
)
890 * We deal with the read-ahead logic here.
892 if (ra_ptr
>= ra_max
) {
893 /* Refill the readahead buffer */
896 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
898 * Terminate if we reach the end of the
899 * directory and must wrap, or if our
900 * search has finished at this block.
902 if (b
>= nblocks
|| (num
&& block
== start
)) {
903 bh_use
[ra_max
] = NULL
;
907 bh
= ext3_getblk(NULL
, dir
, b
++, 0, &err
);
910 ll_rw_block(READ_META
, 1, &bh
);
913 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
916 if (!buffer_uptodate(bh
)) {
917 /* read error, skip block & hope for the best */
918 ext3_error(sb
, __FUNCTION__
, "reading directory #%lu "
919 "offset %lu", dir
->i_ino
, block
);
923 i
= search_dirblock(bh
, dir
, dentry
,
924 block
<< EXT3_BLOCK_SIZE_BITS(sb
), res_dir
);
926 EXT3_I(dir
)->i_dir_start_lookup
= block
;
928 goto cleanup_and_exit
;
932 goto cleanup_and_exit
;
935 if (++block
>= nblocks
)
937 } while (block
!= start
);
940 * If the directory has grown while we were searching, then
941 * search the last part of the directory before giving up.
944 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
945 if (block
< nblocks
) {
951 /* Clean up the read-ahead blocks */
952 for (; ra_ptr
< ra_max
; ra_ptr
++)
953 brelse (bh_use
[ra_ptr
]);
957 static struct buffer_head
* ext3_dx_find_entry(struct dentry
*dentry
,
958 struct ext3_dir_entry_2
**res_dir
, int *err
)
960 struct super_block
* sb
;
961 struct dx_hash_info hinfo
;
963 struct dx_frame frames
[2], *frame
;
964 struct ext3_dir_entry_2
*de
, *top
;
965 struct buffer_head
*bh
;
968 int namelen
= dentry
->d_name
.len
;
969 const u8
*name
= dentry
->d_name
.name
;
970 struct inode
*dir
= dentry
->d_parent
->d_inode
;
973 /* NFS may look up ".." - look at dx_root directory block */
974 if (namelen
> 2 || name
[0] != '.'||(name
[1] != '.' && name
[1] != '\0')){
975 if (!(frame
= dx_probe(dentry
, NULL
, &hinfo
, frames
, err
)))
979 frame
->bh
= NULL
; /* for dx_release() */
980 frame
->at
= (struct dx_entry
*)frames
; /* hack for zero entry*/
981 dx_set_block(frame
->at
, 0); /* dx_root block is 0 */
985 block
= dx_get_block(frame
->at
);
986 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0, err
)))
988 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
989 top
= (struct ext3_dir_entry_2
*) ((char *) de
+ sb
->s_blocksize
-
990 EXT3_DIR_REC_LEN(0));
991 for (; de
< top
; de
= ext3_next_entry(de
))
992 if (ext3_match (namelen
, name
, de
)) {
993 if (!ext3_check_dir_entry("ext3_find_entry",
995 (block
<<EXT3_BLOCK_SIZE_BITS(sb
))
996 +((char *)de
- bh
->b_data
))) {
998 *err
= ERR_BAD_DX_DIR
;
1002 dx_release (frames
);
1006 /* Check to see if we should continue to search */
1007 retval
= ext3_htree_next_block(dir
, hash
, frame
,
1010 ext3_warning(sb
, __FUNCTION__
,
1011 "error reading index page in directory #%lu",
1016 } while (retval
== 1);
1020 dxtrace(printk("%s not found\n", name
));
1021 dx_release (frames
);
1025 static struct dentry
*ext3_lookup(struct inode
* dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1027 struct inode
* inode
;
1028 struct ext3_dir_entry_2
* de
;
1029 struct buffer_head
* bh
;
1031 if (dentry
->d_name
.len
> EXT3_NAME_LEN
)
1032 return ERR_PTR(-ENAMETOOLONG
);
1034 bh
= ext3_find_entry(dentry
, &de
);
1037 unsigned long ino
= le32_to_cpu(de
->inode
);
1039 if (!ext3_valid_inum(dir
->i_sb
, ino
)) {
1040 ext3_error(dir
->i_sb
, "ext3_lookup",
1041 "bad inode number: %lu", ino
);
1044 inode
= iget(dir
->i_sb
, ino
);
1047 return ERR_PTR(-EACCES
);
1049 if (is_bad_inode(inode
)) {
1051 return ERR_PTR(-ENOENT
);
1054 return d_splice_alias(inode
, dentry
);
1058 struct dentry
*ext3_get_parent(struct dentry
*child
)
1061 struct dentry
*parent
;
1062 struct inode
*inode
;
1063 struct dentry dotdot
;
1064 struct ext3_dir_entry_2
* de
;
1065 struct buffer_head
*bh
;
1067 dotdot
.d_name
.name
= "..";
1068 dotdot
.d_name
.len
= 2;
1069 dotdot
.d_parent
= child
; /* confusing, isn't it! */
1071 bh
= ext3_find_entry(&dotdot
, &de
);
1074 return ERR_PTR(-ENOENT
);
1075 ino
= le32_to_cpu(de
->inode
);
1078 if (!ext3_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1079 ext3_error(child
->d_inode
->i_sb
, "ext3_get_parent",
1080 "bad inode number: %lu", ino
);
1083 inode
= iget(child
->d_inode
->i_sb
, ino
);
1086 return ERR_PTR(-EACCES
);
1088 if (is_bad_inode(inode
)) {
1090 return ERR_PTR(-ENOENT
);
1093 parent
= d_alloc_anon(inode
);
1096 parent
= ERR_PTR(-ENOMEM
);
1102 static unsigned char ext3_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1103 [S_IFREG
>> S_SHIFT
] = EXT3_FT_REG_FILE
,
1104 [S_IFDIR
>> S_SHIFT
] = EXT3_FT_DIR
,
1105 [S_IFCHR
>> S_SHIFT
] = EXT3_FT_CHRDEV
,
1106 [S_IFBLK
>> S_SHIFT
] = EXT3_FT_BLKDEV
,
1107 [S_IFIFO
>> S_SHIFT
] = EXT3_FT_FIFO
,
1108 [S_IFSOCK
>> S_SHIFT
] = EXT3_FT_SOCK
,
1109 [S_IFLNK
>> S_SHIFT
] = EXT3_FT_SYMLINK
,
1112 static inline void ext3_set_de_type(struct super_block
*sb
,
1113 struct ext3_dir_entry_2
*de
,
1115 if (EXT3_HAS_INCOMPAT_FEATURE(sb
, EXT3_FEATURE_INCOMPAT_FILETYPE
))
1116 de
->file_type
= ext3_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1120 * Move count entries from end of map between two memory locations.
1121 * Returns pointer to last entry moved.
1123 static struct ext3_dir_entry_2
*
1124 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1126 unsigned rec_len
= 0;
1129 struct ext3_dir_entry_2
*de
= (struct ext3_dir_entry_2
*) (from
+ map
->offs
);
1130 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1131 memcpy (to
, de
, rec_len
);
1132 ((struct ext3_dir_entry_2
*) to
)->rec_len
=
1133 cpu_to_le16(rec_len
);
1138 return (struct ext3_dir_entry_2
*) (to
- rec_len
);
1142 * Compact each dir entry in the range to the minimal rec_len.
1143 * Returns pointer to last entry in range.
1145 static struct ext3_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1147 struct ext3_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext3_dir_entry_2
*) base
;
1148 unsigned rec_len
= 0;
1151 while ((char*)de
< base
+ size
) {
1152 next
= (struct ext3_dir_entry_2
*) ((char *) de
+
1153 le16_to_cpu(de
->rec_len
));
1154 if (de
->inode
&& de
->name_len
) {
1155 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1157 memmove(to
, de
, rec_len
);
1158 to
->rec_len
= cpu_to_le16(rec_len
);
1160 to
= (struct ext3_dir_entry_2
*) (((char *) to
) + rec_len
);
1168 * Split a full leaf block to make room for a new dir entry.
1169 * Allocate a new block, and move entries so that they are approx. equally full.
1170 * Returns pointer to de in block into which the new entry will be inserted.
1172 static struct ext3_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1173 struct buffer_head
**bh
,struct dx_frame
*frame
,
1174 struct dx_hash_info
*hinfo
, int *error
)
1176 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1177 unsigned count
, continued
;
1178 struct buffer_head
*bh2
;
1181 struct dx_map_entry
*map
;
1182 char *data1
= (*bh
)->b_data
, *data2
;
1183 unsigned split
, move
, size
, i
;
1184 struct ext3_dir_entry_2
*de
= NULL
, *de2
;
1187 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1194 BUFFER_TRACE(*bh
, "get_write_access");
1195 err
= ext3_journal_get_write_access(handle
, *bh
);
1199 BUFFER_TRACE(frame
->bh
, "get_write_access");
1200 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1204 data2
= bh2
->b_data
;
1206 /* create map in the end of data2 block */
1207 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1208 count
= dx_make_map ((struct ext3_dir_entry_2
*) data1
,
1209 blocksize
, hinfo
, map
);
1211 dx_sort_map (map
, count
);
1212 /* Split the existing block in the middle, size-wise */
1215 for (i
= count
-1; i
>= 0; i
--) {
1216 /* is more than half of this entry in 2nd half of the block? */
1217 if (size
+ map
[i
].size
/2 > blocksize
/2)
1219 size
+= map
[i
].size
;
1222 /* map index at which we will split */
1223 split
= count
- move
;
1224 hash2
= map
[split
].hash
;
1225 continued
= hash2
== map
[split
- 1].hash
;
1226 dxtrace(printk("Split block %i at %x, %i/%i\n",
1227 dx_get_block(frame
->at
), hash2
, split
, count
-split
));
1229 /* Fancy dance to stay within two buffers */
1230 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1231 de
= dx_pack_dirents(data1
,blocksize
);
1232 de
->rec_len
= cpu_to_le16(data1
+ blocksize
- (char *) de
);
1233 de2
->rec_len
= cpu_to_le16(data2
+ blocksize
- (char *) de2
);
1234 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data1
, blocksize
, 1));
1235 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data2
, blocksize
, 1));
1237 /* Which block gets the new entry? */
1238 if (hinfo
->hash
>= hash2
)
1243 dx_insert_block (frame
, hash2
+ continued
, newblock
);
1244 err
= ext3_journal_dirty_metadata (handle
, bh2
);
1247 err
= ext3_journal_dirty_metadata (handle
, frame
->bh
);
1251 dxtrace(dx_show_index ("frame", frame
->entries
));
1258 ext3_std_error(dir
->i_sb
, err
);
1266 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1267 * it points to a directory entry which is guaranteed to be large
1268 * enough for new directory entry. If de is NULL, then
1269 * add_dirent_to_buf will attempt search the directory block for
1270 * space. It will return -ENOSPC if no space is available, and -EIO
1271 * and -EEXIST if directory entry already exists.
1273 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1274 * all other cases bh is released.
1276 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1277 struct inode
*inode
, struct ext3_dir_entry_2
*de
,
1278 struct buffer_head
* bh
)
1280 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1281 const char *name
= dentry
->d_name
.name
;
1282 int namelen
= dentry
->d_name
.len
;
1283 unsigned long offset
= 0;
1284 unsigned short reclen
;
1285 int nlen
, rlen
, err
;
1288 reclen
= EXT3_DIR_REC_LEN(namelen
);
1290 de
= (struct ext3_dir_entry_2
*)bh
->b_data
;
1291 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1292 while ((char *) de
<= top
) {
1293 if (!ext3_check_dir_entry("ext3_add_entry", dir
, de
,
1298 if (ext3_match (namelen
, name
, de
)) {
1302 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1303 rlen
= le16_to_cpu(de
->rec_len
);
1304 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1306 de
= (struct ext3_dir_entry_2
*)((char *)de
+ rlen
);
1309 if ((char *) de
> top
)
1312 BUFFER_TRACE(bh
, "get_write_access");
1313 err
= ext3_journal_get_write_access(handle
, bh
);
1315 ext3_std_error(dir
->i_sb
, err
);
1320 /* By now the buffer is marked for journaling */
1321 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1322 rlen
= le16_to_cpu(de
->rec_len
);
1324 struct ext3_dir_entry_2
*de1
= (struct ext3_dir_entry_2
*)((char *)de
+ nlen
);
1325 de1
->rec_len
= cpu_to_le16(rlen
- nlen
);
1326 de
->rec_len
= cpu_to_le16(nlen
);
1329 de
->file_type
= EXT3_FT_UNKNOWN
;
1331 de
->inode
= cpu_to_le32(inode
->i_ino
);
1332 ext3_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1335 de
->name_len
= namelen
;
1336 memcpy (de
->name
, name
, namelen
);
1338 * XXX shouldn't update any times until successful
1339 * completion of syscall, but too many callers depend
1342 * XXX similarly, too many callers depend on
1343 * ext3_new_inode() setting the times, but error
1344 * recovery deletes the inode, so the worst that can
1345 * happen is that the times are slightly out of date
1346 * and/or different from the directory change time.
1348 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME_SEC
;
1349 ext3_update_dx_flag(dir
);
1351 ext3_mark_inode_dirty(handle
, dir
);
1352 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1353 err
= ext3_journal_dirty_metadata(handle
, bh
);
1355 ext3_std_error(dir
->i_sb
, err
);
1361 * This converts a one block unindexed directory to a 3 block indexed
1362 * directory, and adds the dentry to the indexed directory.
1364 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1365 struct inode
*inode
, struct buffer_head
*bh
)
1367 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1368 const char *name
= dentry
->d_name
.name
;
1369 int namelen
= dentry
->d_name
.len
;
1370 struct buffer_head
*bh2
;
1371 struct dx_root
*root
;
1372 struct dx_frame frames
[2], *frame
;
1373 struct dx_entry
*entries
;
1374 struct ext3_dir_entry_2
*de
, *de2
;
1379 struct dx_hash_info hinfo
;
1381 struct fake_dirent
*fde
;
1383 blocksize
= dir
->i_sb
->s_blocksize
;
1384 dxtrace(printk("Creating index\n"));
1385 retval
= ext3_journal_get_write_access(handle
, bh
);
1387 ext3_std_error(dir
->i_sb
, retval
);
1391 root
= (struct dx_root
*) bh
->b_data
;
1393 bh2
= ext3_append (handle
, dir
, &block
, &retval
);
1398 EXT3_I(dir
)->i_flags
|= EXT3_INDEX_FL
;
1399 data1
= bh2
->b_data
;
1401 /* The 0th block becomes the root, move the dirents out */
1402 fde
= &root
->dotdot
;
1403 de
= (struct ext3_dir_entry_2
*)((char *)fde
+ le16_to_cpu(fde
->rec_len
));
1404 len
= ((char *) root
) + blocksize
- (char *) de
;
1405 memcpy (data1
, de
, len
);
1406 de
= (struct ext3_dir_entry_2
*) data1
;
1408 while ((char *)(de2
=(void*)de
+le16_to_cpu(de
->rec_len
)) < top
)
1410 de
->rec_len
= cpu_to_le16(data1
+ blocksize
- (char *) de
);
1411 /* Initialize the root; the dot dirents already exist */
1412 de
= (struct ext3_dir_entry_2
*) (&root
->dotdot
);
1413 de
->rec_len
= cpu_to_le16(blocksize
- EXT3_DIR_REC_LEN(2));
1414 memset (&root
->info
, 0, sizeof(root
->info
));
1415 root
->info
.info_length
= sizeof(root
->info
);
1416 root
->info
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
1417 entries
= root
->entries
;
1418 dx_set_block (entries
, 1);
1419 dx_set_count (entries
, 1);
1420 dx_set_limit (entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1422 /* Initialize as for dx_probe */
1423 hinfo
.hash_version
= root
->info
.hash_version
;
1424 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
1425 ext3fs_dirhash(name
, namelen
, &hinfo
);
1427 frame
->entries
= entries
;
1428 frame
->at
= entries
;
1431 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1432 dx_release (frames
);
1436 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1442 * adds a file entry to the specified directory, using the same
1443 * semantics as ext3_find_entry(). It returns NULL if it failed.
1445 * NOTE!! The inode part of 'de' is left at 0 - which means you
1446 * may not sleep between calling this and putting something into
1447 * the entry, as someone else might have used it while you slept.
1449 static int ext3_add_entry (handle_t
*handle
, struct dentry
*dentry
,
1450 struct inode
*inode
)
1452 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1453 unsigned long offset
;
1454 struct buffer_head
* bh
;
1455 struct ext3_dir_entry_2
*de
;
1456 struct super_block
* sb
;
1463 blocksize
= sb
->s_blocksize
;
1464 if (!dentry
->d_name
.len
)
1467 retval
= ext3_dx_add_entry(handle
, dentry
, inode
);
1468 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1470 EXT3_I(dir
)->i_flags
&= ~EXT3_INDEX_FL
;
1472 ext3_mark_inode_dirty(handle
, dir
);
1474 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1475 for (block
= 0, offset
= 0; block
< blocks
; block
++) {
1476 bh
= ext3_bread(handle
, dir
, block
, 0, &retval
);
1479 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1480 if (retval
!= -ENOSPC
)
1483 if (blocks
== 1 && !dx_fallback
&&
1484 EXT3_HAS_COMPAT_FEATURE(sb
, EXT3_FEATURE_COMPAT_DIR_INDEX
))
1485 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1488 bh
= ext3_append(handle
, dir
, &block
, &retval
);
1491 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1493 de
->rec_len
= cpu_to_le16(blocksize
);
1494 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1498 * Returns 0 for success, or a negative error value
1500 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1501 struct inode
*inode
)
1503 struct dx_frame frames
[2], *frame
;
1504 struct dx_entry
*entries
, *at
;
1505 struct dx_hash_info hinfo
;
1506 struct buffer_head
* bh
;
1507 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1508 struct super_block
* sb
= dir
->i_sb
;
1509 struct ext3_dir_entry_2
*de
;
1512 frame
= dx_probe(dentry
, NULL
, &hinfo
, frames
, &err
);
1515 entries
= frame
->entries
;
1518 if (!(bh
= ext3_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1521 BUFFER_TRACE(bh
, "get_write_access");
1522 err
= ext3_journal_get_write_access(handle
, bh
);
1526 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1527 if (err
!= -ENOSPC
) {
1532 /* Block full, should compress but for now just split */
1533 dxtrace(printk("using %u of %u node entries\n",
1534 dx_get_count(entries
), dx_get_limit(entries
)));
1535 /* Need to split index? */
1536 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1538 unsigned icount
= dx_get_count(entries
);
1539 int levels
= frame
- frames
;
1540 struct dx_entry
*entries2
;
1541 struct dx_node
*node2
;
1542 struct buffer_head
*bh2
;
1544 if (levels
&& (dx_get_count(frames
->entries
) ==
1545 dx_get_limit(frames
->entries
))) {
1546 ext3_warning(sb
, __FUNCTION__
,
1547 "Directory index full!");
1551 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1554 node2
= (struct dx_node
*)(bh2
->b_data
);
1555 entries2
= node2
->entries
;
1556 node2
->fake
.rec_len
= cpu_to_le16(sb
->s_blocksize
);
1557 node2
->fake
.inode
= 0;
1558 BUFFER_TRACE(frame
->bh
, "get_write_access");
1559 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1563 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1564 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1565 dxtrace(printk("Split index %i/%i\n", icount1
, icount2
));
1567 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1568 err
= ext3_journal_get_write_access(handle
,
1573 memcpy ((char *) entries2
, (char *) (entries
+ icount1
),
1574 icount2
* sizeof(struct dx_entry
));
1575 dx_set_count (entries
, icount1
);
1576 dx_set_count (entries2
, icount2
);
1577 dx_set_limit (entries2
, dx_node_limit(dir
));
1579 /* Which index block gets the new entry? */
1580 if (at
- entries
>= icount1
) {
1581 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1582 frame
->entries
= entries
= entries2
;
1583 swap(frame
->bh
, bh2
);
1585 dx_insert_block (frames
+ 0, hash2
, newblock
);
1586 dxtrace(dx_show_index ("node", frames
[1].entries
));
1587 dxtrace(dx_show_index ("node",
1588 ((struct dx_node
*) bh2
->b_data
)->entries
));
1589 err
= ext3_journal_dirty_metadata(handle
, bh2
);
1594 dxtrace(printk("Creating second level index...\n"));
1595 memcpy((char *) entries2
, (char *) entries
,
1596 icount
* sizeof(struct dx_entry
));
1597 dx_set_limit(entries2
, dx_node_limit(dir
));
1600 dx_set_count(entries
, 1);
1601 dx_set_block(entries
+ 0, newblock
);
1602 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1604 /* Add new access path frame */
1606 frame
->at
= at
= at
- entries
+ entries2
;
1607 frame
->entries
= entries
= entries2
;
1609 err
= ext3_journal_get_write_access(handle
,
1614 ext3_journal_dirty_metadata(handle
, frames
[0].bh
);
1616 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1619 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1624 ext3_std_error(dir
->i_sb
, err
);
1633 * ext3_delete_entry deletes a directory entry by merging it with the
1636 static int ext3_delete_entry (handle_t
*handle
,
1638 struct ext3_dir_entry_2
* de_del
,
1639 struct buffer_head
* bh
)
1641 struct ext3_dir_entry_2
* de
, * pde
;
1646 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1647 while (i
< bh
->b_size
) {
1648 if (!ext3_check_dir_entry("ext3_delete_entry", dir
, de
, bh
, i
))
1651 BUFFER_TRACE(bh
, "get_write_access");
1652 ext3_journal_get_write_access(handle
, bh
);
1655 cpu_to_le16(le16_to_cpu(pde
->rec_len
) +
1656 le16_to_cpu(de
->rec_len
));
1660 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1661 ext3_journal_dirty_metadata(handle
, bh
);
1664 i
+= le16_to_cpu(de
->rec_len
);
1666 de
= (struct ext3_dir_entry_2
*)
1667 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1672 static int ext3_add_nondir(handle_t
*handle
,
1673 struct dentry
*dentry
, struct inode
*inode
)
1675 int err
= ext3_add_entry(handle
, dentry
, inode
);
1677 ext3_mark_inode_dirty(handle
, inode
);
1678 d_instantiate(dentry
, inode
);
1687 * By the time this is called, we already have created
1688 * the directory cache entry for the new file, but it
1689 * is so far negative - it has no inode.
1691 * If the create succeeds, we fill in the inode information
1692 * with d_instantiate().
1694 static int ext3_create (struct inode
* dir
, struct dentry
* dentry
, int mode
,
1695 struct nameidata
*nd
)
1698 struct inode
* inode
;
1699 int err
, retries
= 0;
1702 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1703 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1704 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1706 return PTR_ERR(handle
);
1708 if (IS_DIRSYNC(dir
))
1711 inode
= ext3_new_inode (handle
, dir
, mode
);
1712 err
= PTR_ERR(inode
);
1713 if (!IS_ERR(inode
)) {
1714 inode
->i_op
= &ext3_file_inode_operations
;
1715 inode
->i_fop
= &ext3_file_operations
;
1716 ext3_set_aops(inode
);
1717 err
= ext3_add_nondir(handle
, dentry
, inode
);
1719 ext3_journal_stop(handle
);
1720 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1725 static int ext3_mknod (struct inode
* dir
, struct dentry
*dentry
,
1726 int mode
, dev_t rdev
)
1729 struct inode
*inode
;
1730 int err
, retries
= 0;
1732 if (!new_valid_dev(rdev
))
1736 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1737 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1738 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1740 return PTR_ERR(handle
);
1742 if (IS_DIRSYNC(dir
))
1745 inode
= ext3_new_inode (handle
, dir
, mode
);
1746 err
= PTR_ERR(inode
);
1747 if (!IS_ERR(inode
)) {
1748 init_special_inode(inode
, inode
->i_mode
, rdev
);
1749 #ifdef CONFIG_EXT3_FS_XATTR
1750 inode
->i_op
= &ext3_special_inode_operations
;
1752 err
= ext3_add_nondir(handle
, dentry
, inode
);
1754 ext3_journal_stop(handle
);
1755 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1760 static int ext3_mkdir(struct inode
* dir
, struct dentry
* dentry
, int mode
)
1763 struct inode
* inode
;
1764 struct buffer_head
* dir_block
;
1765 struct ext3_dir_entry_2
* de
;
1766 int err
, retries
= 0;
1768 if (dir
->i_nlink
>= EXT3_LINK_MAX
)
1772 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1773 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1774 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1776 return PTR_ERR(handle
);
1778 if (IS_DIRSYNC(dir
))
1781 inode
= ext3_new_inode (handle
, dir
, S_IFDIR
| mode
);
1782 err
= PTR_ERR(inode
);
1786 inode
->i_op
= &ext3_dir_inode_operations
;
1787 inode
->i_fop
= &ext3_dir_operations
;
1788 inode
->i_size
= EXT3_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1789 dir_block
= ext3_bread (handle
, inode
, 0, 1, &err
);
1791 drop_nlink(inode
); /* is this nlink == 0? */
1792 ext3_mark_inode_dirty(handle
, inode
);
1796 BUFFER_TRACE(dir_block
, "get_write_access");
1797 ext3_journal_get_write_access(handle
, dir_block
);
1798 de
= (struct ext3_dir_entry_2
*) dir_block
->b_data
;
1799 de
->inode
= cpu_to_le32(inode
->i_ino
);
1801 de
->rec_len
= cpu_to_le16(EXT3_DIR_REC_LEN(de
->name_len
));
1802 strcpy (de
->name
, ".");
1803 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1804 de
= (struct ext3_dir_entry_2
*)
1805 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1806 de
->inode
= cpu_to_le32(dir
->i_ino
);
1807 de
->rec_len
= cpu_to_le16(inode
->i_sb
->s_blocksize
-EXT3_DIR_REC_LEN(1));
1809 strcpy (de
->name
, "..");
1810 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1812 BUFFER_TRACE(dir_block
, "call ext3_journal_dirty_metadata");
1813 ext3_journal_dirty_metadata(handle
, dir_block
);
1815 ext3_mark_inode_dirty(handle
, inode
);
1816 err
= ext3_add_entry (handle
, dentry
, inode
);
1819 ext3_mark_inode_dirty(handle
, inode
);
1824 ext3_update_dx_flag(dir
);
1825 ext3_mark_inode_dirty(handle
, dir
);
1826 d_instantiate(dentry
, inode
);
1828 ext3_journal_stop(handle
);
1829 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1835 * routine to check that the specified directory is empty (for rmdir)
1837 static int empty_dir (struct inode
* inode
)
1839 unsigned long offset
;
1840 struct buffer_head
* bh
;
1841 struct ext3_dir_entry_2
* de
, * de1
;
1842 struct super_block
* sb
;
1846 if (inode
->i_size
< EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1847 !(bh
= ext3_bread (NULL
, inode
, 0, 0, &err
))) {
1849 ext3_error(inode
->i_sb
, __FUNCTION__
,
1850 "error %d reading directory #%lu offset 0",
1853 ext3_warning(inode
->i_sb
, __FUNCTION__
,
1854 "bad directory (dir #%lu) - no data block",
1858 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1859 de1
= (struct ext3_dir_entry_2
*)
1860 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1861 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1862 !le32_to_cpu(de1
->inode
) ||
1863 strcmp (".", de
->name
) ||
1864 strcmp ("..", de1
->name
)) {
1865 ext3_warning (inode
->i_sb
, "empty_dir",
1866 "bad directory (dir #%lu) - no `.' or `..'",
1871 offset
= le16_to_cpu(de
->rec_len
) + le16_to_cpu(de1
->rec_len
);
1872 de
= (struct ext3_dir_entry_2
*)
1873 ((char *) de1
+ le16_to_cpu(de1
->rec_len
));
1874 while (offset
< inode
->i_size
) {
1876 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1879 bh
= ext3_bread (NULL
, inode
,
1880 offset
>> EXT3_BLOCK_SIZE_BITS(sb
), 0, &err
);
1883 ext3_error(sb
, __FUNCTION__
,
1884 "error %d reading directory"
1886 err
, inode
->i_ino
, offset
);
1887 offset
+= sb
->s_blocksize
;
1890 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1892 if (!ext3_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1893 de
= (struct ext3_dir_entry_2
*)(bh
->b_data
+
1895 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1898 if (le32_to_cpu(de
->inode
)) {
1902 offset
+= le16_to_cpu(de
->rec_len
);
1903 de
= (struct ext3_dir_entry_2
*)
1904 ((char *) de
+ le16_to_cpu(de
->rec_len
));
1910 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1911 * such inodes, starting at the superblock, in case we crash before the
1912 * file is closed/deleted, or in case the inode truncate spans multiple
1913 * transactions and the last transaction is not recovered after a crash.
1915 * At filesystem recovery time, we walk this list deleting unlinked
1916 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1918 int ext3_orphan_add(handle_t
*handle
, struct inode
*inode
)
1920 struct super_block
*sb
= inode
->i_sb
;
1921 struct ext3_iloc iloc
;
1925 if (!list_empty(&EXT3_I(inode
)->i_orphan
))
1928 /* Orphan handling is only valid for files with data blocks
1929 * being truncated, or files being unlinked. */
1931 /* @@@ FIXME: Observation from aviro:
1932 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1933 * here (on lock_super()), so race with ext3_link() which might bump
1934 * ->i_nlink. For, say it, character device. Not a regular file,
1935 * not a directory, not a symlink and ->i_nlink > 0.
1937 J_ASSERT ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1938 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1940 BUFFER_TRACE(EXT3_SB(sb
)->s_sbh
, "get_write_access");
1941 err
= ext3_journal_get_write_access(handle
, EXT3_SB(sb
)->s_sbh
);
1945 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1949 /* Insert this inode at the head of the on-disk orphan list... */
1950 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT3_SB(sb
)->s_es
->s_last_orphan
);
1951 EXT3_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1952 err
= ext3_journal_dirty_metadata(handle
, EXT3_SB(sb
)->s_sbh
);
1953 rc
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
1957 /* Only add to the head of the in-memory list if all the
1958 * previous operations succeeded. If the orphan_add is going to
1959 * fail (possibly taking the journal offline), we can't risk
1960 * leaving the inode on the orphan list: stray orphan-list
1961 * entries can cause panics at unmount time.
1963 * This is safe: on error we're going to ignore the orphan list
1964 * anyway on the next recovery. */
1966 list_add(&EXT3_I(inode
)->i_orphan
, &EXT3_SB(sb
)->s_orphan
);
1968 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1969 jbd_debug(4, "orphan inode %lu will point to %d\n",
1970 inode
->i_ino
, NEXT_ORPHAN(inode
));
1973 ext3_std_error(inode
->i_sb
, err
);
1978 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1979 * of such inodes stored on disk, because it is finally being cleaned up.
1981 int ext3_orphan_del(handle_t
*handle
, struct inode
*inode
)
1983 struct list_head
*prev
;
1984 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1985 struct ext3_sb_info
*sbi
;
1986 unsigned long ino_next
;
1987 struct ext3_iloc iloc
;
1990 lock_super(inode
->i_sb
);
1991 if (list_empty(&ei
->i_orphan
)) {
1992 unlock_super(inode
->i_sb
);
1996 ino_next
= NEXT_ORPHAN(inode
);
1997 prev
= ei
->i_orphan
.prev
;
1998 sbi
= EXT3_SB(inode
->i_sb
);
2000 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2002 list_del_init(&ei
->i_orphan
);
2004 /* If we're on an error path, we may not have a valid
2005 * transaction handle with which to update the orphan list on
2006 * disk, but we still need to remove the inode from the linked
2007 * list in memory. */
2011 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
2015 if (prev
== &sbi
->s_orphan
) {
2016 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
2017 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2018 err
= ext3_journal_get_write_access(handle
, sbi
->s_sbh
);
2021 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2022 err
= ext3_journal_dirty_metadata(handle
, sbi
->s_sbh
);
2024 struct ext3_iloc iloc2
;
2025 struct inode
*i_prev
=
2026 &list_entry(prev
, struct ext3_inode_info
, i_orphan
)->vfs_inode
;
2028 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2029 i_prev
->i_ino
, ino_next
);
2030 err
= ext3_reserve_inode_write(handle
, i_prev
, &iloc2
);
2033 NEXT_ORPHAN(i_prev
) = ino_next
;
2034 err
= ext3_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2038 NEXT_ORPHAN(inode
) = 0;
2039 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
2042 ext3_std_error(inode
->i_sb
, err
);
2044 unlock_super(inode
->i_sb
);
2052 static int ext3_rmdir (struct inode
* dir
, struct dentry
*dentry
)
2055 struct inode
* inode
;
2056 struct buffer_head
* bh
;
2057 struct ext3_dir_entry_2
* de
;
2060 /* Initialize quotas before so that eventual writes go in
2061 * separate transaction */
2062 DQUOT_INIT(dentry
->d_inode
);
2063 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2065 return PTR_ERR(handle
);
2068 bh
= ext3_find_entry (dentry
, &de
);
2072 if (IS_DIRSYNC(dir
))
2075 inode
= dentry
->d_inode
;
2078 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2081 retval
= -ENOTEMPTY
;
2082 if (!empty_dir (inode
))
2085 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2088 if (inode
->i_nlink
!= 2)
2089 ext3_warning (inode
->i_sb
, "ext3_rmdir",
2090 "empty directory has nlink!=2 (%d)",
2094 /* There's no need to set i_disksize: the fact that i_nlink is
2095 * zero will ensure that the right thing happens during any
2098 ext3_orphan_add(handle
, inode
);
2099 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2100 ext3_mark_inode_dirty(handle
, inode
);
2102 ext3_update_dx_flag(dir
);
2103 ext3_mark_inode_dirty(handle
, dir
);
2106 ext3_journal_stop(handle
);
2111 static int ext3_unlink(struct inode
* dir
, struct dentry
*dentry
)
2114 struct inode
* inode
;
2115 struct buffer_head
* bh
;
2116 struct ext3_dir_entry_2
* de
;
2119 /* Initialize quotas before so that eventual writes go
2120 * in separate transaction */
2121 DQUOT_INIT(dentry
->d_inode
);
2122 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2124 return PTR_ERR(handle
);
2126 if (IS_DIRSYNC(dir
))
2130 bh
= ext3_find_entry (dentry
, &de
);
2134 inode
= dentry
->d_inode
;
2137 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2140 if (!inode
->i_nlink
) {
2141 ext3_warning (inode
->i_sb
, "ext3_unlink",
2142 "Deleting nonexistent file (%lu), %d",
2143 inode
->i_ino
, inode
->i_nlink
);
2146 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2149 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2150 ext3_update_dx_flag(dir
);
2151 ext3_mark_inode_dirty(handle
, dir
);
2153 if (!inode
->i_nlink
)
2154 ext3_orphan_add(handle
, inode
);
2155 inode
->i_ctime
= dir
->i_ctime
;
2156 ext3_mark_inode_dirty(handle
, inode
);
2160 ext3_journal_stop(handle
);
2165 static int ext3_symlink (struct inode
* dir
,
2166 struct dentry
*dentry
, const char * symname
)
2169 struct inode
* inode
;
2170 int l
, err
, retries
= 0;
2172 l
= strlen(symname
)+1;
2173 if (l
> dir
->i_sb
->s_blocksize
)
2174 return -ENAMETOOLONG
;
2177 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2178 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2179 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2181 return PTR_ERR(handle
);
2183 if (IS_DIRSYNC(dir
))
2186 inode
= ext3_new_inode (handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2187 err
= PTR_ERR(inode
);
2191 if (l
> sizeof (EXT3_I(inode
)->i_data
)) {
2192 inode
->i_op
= &ext3_symlink_inode_operations
;
2193 ext3_set_aops(inode
);
2195 * page_symlink() calls into ext3_prepare/commit_write.
2196 * We have a transaction open. All is sweetness. It also sets
2197 * i_size in generic_commit_write().
2199 err
= __page_symlink(inode
, symname
, l
,
2200 mapping_gfp_mask(inode
->i_mapping
) & ~__GFP_FS
);
2203 ext3_mark_inode_dirty(handle
, inode
);
2208 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2209 memcpy((char*)&EXT3_I(inode
)->i_data
,symname
,l
);
2210 inode
->i_size
= l
-1;
2212 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
2213 err
= ext3_add_nondir(handle
, dentry
, inode
);
2215 ext3_journal_stop(handle
);
2216 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2221 static int ext3_link (struct dentry
* old_dentry
,
2222 struct inode
* dir
, struct dentry
*dentry
)
2225 struct inode
*inode
= old_dentry
->d_inode
;
2226 int err
, retries
= 0;
2228 if (inode
->i_nlink
>= EXT3_LINK_MAX
)
2231 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2232 * otherwise has the potential to corrupt the orphan inode list.
2234 if (inode
->i_nlink
== 0)
2238 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2239 EXT3_INDEX_EXTRA_TRANS_BLOCKS
);
2241 return PTR_ERR(handle
);
2243 if (IS_DIRSYNC(dir
))
2246 inode
->i_ctime
= CURRENT_TIME_SEC
;
2248 atomic_inc(&inode
->i_count
);
2250 err
= ext3_add_nondir(handle
, dentry
, inode
);
2251 ext3_journal_stop(handle
);
2252 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2257 #define PARENT_INO(buffer) \
2258 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2259 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2262 * Anybody can rename anything with this: the permission checks are left to the
2263 * higher-level routines.
2265 static int ext3_rename (struct inode
* old_dir
, struct dentry
*old_dentry
,
2266 struct inode
* new_dir
,struct dentry
*new_dentry
)
2269 struct inode
* old_inode
, * new_inode
;
2270 struct buffer_head
* old_bh
, * new_bh
, * dir_bh
;
2271 struct ext3_dir_entry_2
* old_de
, * new_de
;
2274 old_bh
= new_bh
= dir_bh
= NULL
;
2276 /* Initialize quotas before so that eventual writes go
2277 * in separate transaction */
2278 if (new_dentry
->d_inode
)
2279 DQUOT_INIT(new_dentry
->d_inode
);
2280 handle
= ext3_journal_start(old_dir
, 2 *
2281 EXT3_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2282 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2284 return PTR_ERR(handle
);
2286 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2289 old_bh
= ext3_find_entry (old_dentry
, &old_de
);
2291 * Check for inode number is _not_ due to possible IO errors.
2292 * We might rmdir the source, keep it as pwd of some process
2293 * and merrily kill the link to whatever was created under the
2294 * same name. Goodbye sticky bit ;-<
2296 old_inode
= old_dentry
->d_inode
;
2298 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2301 new_inode
= new_dentry
->d_inode
;
2302 new_bh
= ext3_find_entry (new_dentry
, &new_de
);
2309 if (S_ISDIR(old_inode
->i_mode
)) {
2311 retval
= -ENOTEMPTY
;
2312 if (!empty_dir (new_inode
))
2316 dir_bh
= ext3_bread (handle
, old_inode
, 0, 0, &retval
);
2319 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2322 if (!new_inode
&& new_dir
!=old_dir
&&
2323 new_dir
->i_nlink
>= EXT3_LINK_MAX
)
2327 retval
= ext3_add_entry (handle
, new_dentry
, old_inode
);
2331 BUFFER_TRACE(new_bh
, "get write access");
2332 ext3_journal_get_write_access(handle
, new_bh
);
2333 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2334 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2335 EXT3_FEATURE_INCOMPAT_FILETYPE
))
2336 new_de
->file_type
= old_de
->file_type
;
2337 new_dir
->i_version
++;
2338 BUFFER_TRACE(new_bh
, "call ext3_journal_dirty_metadata");
2339 ext3_journal_dirty_metadata(handle
, new_bh
);
2345 * Like most other Unix systems, set the ctime for inodes on a
2348 old_inode
->i_ctime
= CURRENT_TIME_SEC
;
2349 ext3_mark_inode_dirty(handle
, old_inode
);
2354 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2355 old_de
->name_len
!= old_dentry
->d_name
.len
||
2356 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2357 (retval
= ext3_delete_entry(handle
, old_dir
,
2358 old_de
, old_bh
)) == -ENOENT
) {
2359 /* old_de could have moved from under us during htree split, so
2360 * make sure that we are deleting the right entry. We might
2361 * also be pointing to a stale entry in the unused part of
2362 * old_bh so just checking inum and the name isn't enough. */
2363 struct buffer_head
*old_bh2
;
2364 struct ext3_dir_entry_2
*old_de2
;
2366 old_bh2
= ext3_find_entry(old_dentry
, &old_de2
);
2368 retval
= ext3_delete_entry(handle
, old_dir
,
2374 ext3_warning(old_dir
->i_sb
, "ext3_rename",
2375 "Deleting old file (%lu), %d, error=%d",
2376 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2380 drop_nlink(new_inode
);
2381 new_inode
->i_ctime
= CURRENT_TIME_SEC
;
2383 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME_SEC
;
2384 ext3_update_dx_flag(old_dir
);
2386 BUFFER_TRACE(dir_bh
, "get_write_access");
2387 ext3_journal_get_write_access(handle
, dir_bh
);
2388 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2389 BUFFER_TRACE(dir_bh
, "call ext3_journal_dirty_metadata");
2390 ext3_journal_dirty_metadata(handle
, dir_bh
);
2391 drop_nlink(old_dir
);
2393 drop_nlink(new_inode
);
2396 ext3_update_dx_flag(new_dir
);
2397 ext3_mark_inode_dirty(handle
, new_dir
);
2400 ext3_mark_inode_dirty(handle
, old_dir
);
2402 ext3_mark_inode_dirty(handle
, new_inode
);
2403 if (!new_inode
->i_nlink
)
2404 ext3_orphan_add(handle
, new_inode
);
2412 ext3_journal_stop(handle
);
2417 * directories can handle most operations...
2419 const struct inode_operations ext3_dir_inode_operations
= {
2420 .create
= ext3_create
,
2421 .lookup
= ext3_lookup
,
2423 .unlink
= ext3_unlink
,
2424 .symlink
= ext3_symlink
,
2425 .mkdir
= ext3_mkdir
,
2426 .rmdir
= ext3_rmdir
,
2427 .mknod
= ext3_mknod
,
2428 .rename
= ext3_rename
,
2429 .setattr
= ext3_setattr
,
2430 #ifdef CONFIG_EXT3_FS_XATTR
2431 .setxattr
= generic_setxattr
,
2432 .getxattr
= generic_getxattr
,
2433 .listxattr
= ext3_listxattr
,
2434 .removexattr
= generic_removexattr
,
2436 .permission
= ext3_permission
,
2439 const struct inode_operations ext3_special_inode_operations
= {
2440 .setattr
= ext3_setattr
,
2441 #ifdef CONFIG_EXT3_FS_XATTR
2442 .setxattr
= generic_setxattr
,
2443 .getxattr
= generic_getxattr
,
2444 .listxattr
= ext3_listxattr
,
2445 .removexattr
= generic_removexattr
,
2447 .permission
= ext3_permission
,