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 bh
= ext3_bread(handle
, inode
, *block
, 1, err
);
62 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
63 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
64 *err
= ext3_journal_get_write_access(handle
, bh
);
74 #define assert(test) J_ASSERT(test)
78 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
82 #define dxtrace(command) command
84 #define dxtrace(command)
108 * dx_root_info is laid out so that if it should somehow get overlaid by a
109 * dirent the two low bits of the hash version will be zero. Therefore, the
110 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
115 struct fake_dirent dot
;
117 struct fake_dirent dotdot
;
121 __le32 reserved_zero
;
123 u8 info_length
; /* 8 */
128 struct dx_entry entries
[0];
133 struct fake_dirent fake
;
134 struct dx_entry entries
[0];
140 struct buffer_head
*bh
;
141 struct dx_entry
*entries
;
152 static inline unsigned dx_get_block (struct dx_entry
*entry
);
153 static void dx_set_block (struct dx_entry
*entry
, unsigned value
);
154 static inline unsigned dx_get_hash (struct dx_entry
*entry
);
155 static void dx_set_hash (struct dx_entry
*entry
, unsigned value
);
156 static unsigned dx_get_count (struct dx_entry
*entries
);
157 static unsigned dx_get_limit (struct dx_entry
*entries
);
158 static void dx_set_count (struct dx_entry
*entries
, unsigned value
);
159 static void dx_set_limit (struct dx_entry
*entries
, unsigned value
);
160 static unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
);
161 static unsigned dx_node_limit (struct inode
*dir
);
162 static struct dx_frame
*dx_probe(struct qstr
*entry
,
164 struct dx_hash_info
*hinfo
,
165 struct dx_frame
*frame
,
167 static void dx_release (struct dx_frame
*frames
);
168 static int dx_make_map (struct ext3_dir_entry_2
*de
, int size
,
169 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
170 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
171 static struct ext3_dir_entry_2
*dx_move_dirents (char *from
, char *to
,
172 struct dx_map_entry
*offsets
, int count
);
173 static struct ext3_dir_entry_2
* dx_pack_dirents (char *base
, int size
);
174 static void dx_insert_block (struct dx_frame
*frame
, u32 hash
, u32 block
);
175 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
176 struct dx_frame
*frame
,
177 struct dx_frame
*frames
,
179 static struct buffer_head
* ext3_dx_find_entry(struct inode
*dir
,
180 struct qstr
*entry
, struct ext3_dir_entry_2
**res_dir
,
182 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
183 struct inode
*inode
);
186 * p is at least 6 bytes before the end of page
188 static inline struct ext3_dir_entry_2
*
189 ext3_next_entry(struct ext3_dir_entry_2
*p
)
191 return (struct ext3_dir_entry_2
*)((char *)p
+
192 ext3_rec_len_from_disk(p
->rec_len
));
196 * Future: use high four bits of block for coalesce-on-delete flags
197 * Mask them off for now.
200 static inline unsigned dx_get_block (struct dx_entry
*entry
)
202 return le32_to_cpu(entry
->block
) & 0x00ffffff;
205 static inline void dx_set_block (struct dx_entry
*entry
, unsigned value
)
207 entry
->block
= cpu_to_le32(value
);
210 static inline unsigned dx_get_hash (struct dx_entry
*entry
)
212 return le32_to_cpu(entry
->hash
);
215 static inline void dx_set_hash (struct dx_entry
*entry
, unsigned value
)
217 entry
->hash
= cpu_to_le32(value
);
220 static inline unsigned dx_get_count (struct dx_entry
*entries
)
222 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
225 static inline unsigned dx_get_limit (struct dx_entry
*entries
)
227 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
230 static inline void dx_set_count (struct dx_entry
*entries
, unsigned value
)
232 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
235 static inline void dx_set_limit (struct dx_entry
*entries
, unsigned value
)
237 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
240 static inline unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
)
242 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(1) -
243 EXT3_DIR_REC_LEN(2) - infosize
;
244 return entry_space
/ sizeof(struct dx_entry
);
247 static inline unsigned dx_node_limit (struct inode
*dir
)
249 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(0);
250 return entry_space
/ sizeof(struct dx_entry
);
257 static void dx_show_index (char * label
, struct dx_entry
*entries
)
259 int i
, n
= dx_get_count (entries
);
260 printk("%s index ", label
);
261 for (i
= 0; i
< n
; i
++)
263 printk("%x->%u ", i
? dx_get_hash(entries
+ i
): 0, dx_get_block(entries
+ i
));
275 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext3_dir_entry_2
*de
,
276 int size
, int show_names
)
278 unsigned names
= 0, space
= 0;
279 char *base
= (char *) de
;
280 struct dx_hash_info h
= *hinfo
;
283 while ((char *) de
< base
+ size
)
289 int len
= de
->name_len
;
290 char *name
= de
->name
;
291 while (len
--) printk("%c", *name
++);
292 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
293 printk(":%x.%u ", h
.hash
,
294 ((char *) de
- base
));
296 space
+= EXT3_DIR_REC_LEN(de
->name_len
);
299 de
= ext3_next_entry(de
);
301 printk("(%i)\n", names
);
302 return (struct stats
) { names
, space
, 1 };
305 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
306 struct dx_entry
*entries
, int levels
)
308 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
309 unsigned count
= dx_get_count (entries
), names
= 0, space
= 0, i
;
311 struct buffer_head
*bh
;
313 printk("%i indexed blocks...\n", count
);
314 for (i
= 0; i
< count
; i
++, entries
++)
316 u32 block
= dx_get_block(entries
), hash
= i
? dx_get_hash(entries
): 0;
317 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
319 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
320 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0,&err
))) continue;
322 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
323 dx_show_leaf(hinfo
, (struct ext3_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
324 names
+= stats
.names
;
325 space
+= stats
.space
;
326 bcount
+= stats
.bcount
;
330 printk("%snames %u, fullness %u (%u%%)\n", levels
?"":" ",
331 names
, space
/bcount
,(space
/bcount
)*100/blocksize
);
332 return (struct stats
) { names
, space
, bcount
};
334 #endif /* DX_DEBUG */
337 * Probe for a directory leaf block to search.
339 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
340 * error in the directory index, and the caller should fall back to
341 * searching the directory normally. The callers of dx_probe **MUST**
342 * check for this error code, and make sure it never gets reflected
345 static struct dx_frame
*
346 dx_probe(struct qstr
*entry
, struct inode
*dir
,
347 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
349 unsigned count
, indirect
;
350 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
351 struct dx_root
*root
;
352 struct buffer_head
*bh
;
353 struct dx_frame
*frame
= frame_in
;
357 if (!(bh
= ext3_bread (NULL
,dir
, 0, 0, err
)))
359 root
= (struct dx_root
*) bh
->b_data
;
360 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
361 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
362 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
363 ext3_warning(dir
->i_sb
, __func__
,
364 "Unrecognised inode hash code %d",
365 root
->info
.hash_version
);
367 *err
= ERR_BAD_DX_DIR
;
370 hinfo
->hash_version
= root
->info
.hash_version
;
371 if (hinfo
->hash_version
<= DX_HASH_TEA
)
372 hinfo
->hash_version
+= EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
373 hinfo
->seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
375 ext3fs_dirhash(entry
->name
, entry
->len
, hinfo
);
378 if (root
->info
.unused_flags
& 1) {
379 ext3_warning(dir
->i_sb
, __func__
,
380 "Unimplemented inode hash flags: %#06x",
381 root
->info
.unused_flags
);
383 *err
= ERR_BAD_DX_DIR
;
387 if ((indirect
= root
->info
.indirect_levels
) > 1) {
388 ext3_warning(dir
->i_sb
, __func__
,
389 "Unimplemented inode hash depth: %#06x",
390 root
->info
.indirect_levels
);
392 *err
= ERR_BAD_DX_DIR
;
396 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
397 root
->info
.info_length
);
399 if (dx_get_limit(entries
) != dx_root_limit(dir
,
400 root
->info
.info_length
)) {
401 ext3_warning(dir
->i_sb
, __func__
,
402 "dx entry: limit != root limit");
404 *err
= ERR_BAD_DX_DIR
;
408 dxtrace (printk("Look up %x", hash
));
411 count
= dx_get_count(entries
);
412 if (!count
|| count
> dx_get_limit(entries
)) {
413 ext3_warning(dir
->i_sb
, __func__
,
414 "dx entry: no count or count > limit");
416 *err
= ERR_BAD_DX_DIR
;
421 q
= entries
+ count
- 1;
425 dxtrace(printk("."));
426 if (dx_get_hash(m
) > hash
)
432 if (0) // linear search cross check
434 unsigned n
= count
- 1;
438 dxtrace(printk(","));
439 if (dx_get_hash(++at
) > hash
)
445 assert (at
== p
- 1);
449 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
451 frame
->entries
= entries
;
453 if (!indirect
--) return frame
;
454 if (!(bh
= ext3_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
456 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
457 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
458 ext3_warning(dir
->i_sb
, __func__
,
459 "dx entry: limit != node limit");
461 *err
= ERR_BAD_DX_DIR
;
468 while (frame
>= frame_in
) {
473 if (*err
== ERR_BAD_DX_DIR
)
474 ext3_warning(dir
->i_sb
, __func__
,
475 "Corrupt dir inode %ld, running e2fsck is "
476 "recommended.", dir
->i_ino
);
480 static void dx_release (struct dx_frame
*frames
)
482 if (frames
[0].bh
== NULL
)
485 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
486 brelse(frames
[1].bh
);
487 brelse(frames
[0].bh
);
491 * This function increments the frame pointer to search the next leaf
492 * block, and reads in the necessary intervening nodes if the search
493 * should be necessary. Whether or not the search is necessary is
494 * controlled by the hash parameter. If the hash value is even, then
495 * the search is only continued if the next block starts with that
496 * hash value. This is used if we are searching for a specific file.
498 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
500 * This function returns 1 if the caller should continue to search,
501 * or 0 if it should not. If there is an error reading one of the
502 * index blocks, it will a negative error code.
504 * If start_hash is non-null, it will be filled in with the starting
505 * hash of the next page.
507 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
508 struct dx_frame
*frame
,
509 struct dx_frame
*frames
,
513 struct buffer_head
*bh
;
514 int err
, num_frames
= 0;
519 * Find the next leaf page by incrementing the frame pointer.
520 * If we run out of entries in the interior node, loop around and
521 * increment pointer in the parent node. When we break out of
522 * this loop, num_frames indicates the number of interior
523 * nodes need to be read.
526 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
535 * If the hash is 1, then continue only if the next page has a
536 * continuation hash of any value. This is used for readdir
537 * handling. Otherwise, check to see if the hash matches the
538 * desired contiuation hash. If it doesn't, return since
539 * there's no point to read in the successive index pages.
541 bhash
= dx_get_hash(p
->at
);
544 if ((hash
& 1) == 0) {
545 if ((bhash
& ~1) != hash
)
549 * If the hash is HASH_NB_ALWAYS, we always go to the next
550 * block so no check is necessary
552 while (num_frames
--) {
553 if (!(bh
= ext3_bread(NULL
, dir
, dx_get_block(p
->at
),
555 return err
; /* Failure */
559 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
566 * This function fills a red-black tree with information from a
567 * directory block. It returns the number directory entries loaded
568 * into the tree. If there is an error it is returned in err.
570 static int htree_dirblock_to_tree(struct file
*dir_file
,
571 struct inode
*dir
, int block
,
572 struct dx_hash_info
*hinfo
,
573 __u32 start_hash
, __u32 start_minor_hash
)
575 struct buffer_head
*bh
;
576 struct ext3_dir_entry_2
*de
, *top
;
579 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block
));
580 if (!(bh
= ext3_bread (NULL
, dir
, block
, 0, &err
)))
583 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
584 top
= (struct ext3_dir_entry_2
*) ((char *) de
+
585 dir
->i_sb
->s_blocksize
-
586 EXT3_DIR_REC_LEN(0));
587 for (; de
< top
; de
= ext3_next_entry(de
)) {
588 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
589 (block
<<EXT3_BLOCK_SIZE_BITS(dir
->i_sb
))
590 +((char *)de
- bh
->b_data
))) {
591 /* On error, skip the f_pos to the next block. */
592 dir_file
->f_pos
= (dir_file
->f_pos
|
593 (dir
->i_sb
->s_blocksize
- 1)) + 1;
597 ext3fs_dirhash(de
->name
, de
->name_len
, hinfo
);
598 if ((hinfo
->hash
< start_hash
) ||
599 ((hinfo
->hash
== start_hash
) &&
600 (hinfo
->minor_hash
< start_minor_hash
)))
604 if ((err
= ext3_htree_store_dirent(dir_file
,
605 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
617 * This function fills a red-black tree with information from a
618 * directory. We start scanning the directory in hash order, starting
619 * at start_hash and start_minor_hash.
621 * This function returns the number of entries inserted into the tree,
622 * or a negative error code.
624 int ext3_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
625 __u32 start_minor_hash
, __u32
*next_hash
)
627 struct dx_hash_info hinfo
;
628 struct ext3_dir_entry_2
*de
;
629 struct dx_frame frames
[2], *frame
;
636 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash
,
638 dir
= dir_file
->f_path
.dentry
->d_inode
;
639 if (!(EXT3_I(dir
)->i_flags
& EXT3_INDEX_FL
)) {
640 hinfo
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
641 if (hinfo
.hash_version
<= DX_HASH_TEA
)
642 hinfo
.hash_version
+=
643 EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
644 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
645 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
646 start_hash
, start_minor_hash
);
650 hinfo
.hash
= start_hash
;
651 hinfo
.minor_hash
= 0;
652 frame
= dx_probe(NULL
, dir_file
->f_path
.dentry
->d_inode
, &hinfo
, frames
, &err
);
656 /* Add '.' and '..' from the htree header */
657 if (!start_hash
&& !start_minor_hash
) {
658 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
659 if ((err
= ext3_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
663 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
664 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
665 de
= ext3_next_entry(de
);
666 if ((err
= ext3_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
672 block
= dx_get_block(frame
->at
);
673 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
674 start_hash
, start_minor_hash
);
681 ret
= ext3_htree_next_block(dir
, HASH_NB_ALWAYS
,
682 frame
, frames
, &hashval
);
683 *next_hash
= hashval
;
689 * Stop if: (a) there are no more entries, or
690 * (b) we have inserted at least one entry and the
691 * next hash value is not a continuation
694 (count
&& ((hashval
& 1) == 0)))
698 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
708 * Directory block splitting, compacting
712 * Create map of hash values, offsets, and sizes, stored at end of block.
713 * Returns number of entries mapped.
715 static int dx_make_map (struct ext3_dir_entry_2
*de
, int size
,
716 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
719 char *base
= (char *) de
;
720 struct dx_hash_info h
= *hinfo
;
722 while ((char *) de
< base
+ size
)
724 if (de
->name_len
&& de
->inode
) {
725 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
727 map_tail
->hash
= h
.hash
;
728 map_tail
->offs
= (u16
) ((char *) de
- base
);
729 map_tail
->size
= le16_to_cpu(de
->rec_len
);
733 /* XXX: do we need to check rec_len == 0 case? -Chris */
734 de
= ext3_next_entry(de
);
739 /* Sort map by hash value */
740 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
742 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
744 /* Combsort until bubble sort doesn't suck */
748 if (count
- 9 < 2) /* 9, 10 -> 11 */
750 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
751 if (p
->hash
< q
->hash
)
754 /* Garden variety bubble sort */
760 if (q
[1].hash
>= q
[0].hash
)
768 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, u32 block
)
770 struct dx_entry
*entries
= frame
->entries
;
771 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
772 int count
= dx_get_count(entries
);
774 assert(count
< dx_get_limit(entries
));
775 assert(old
< entries
+ count
);
776 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
777 dx_set_hash(new, hash
);
778 dx_set_block(new, block
);
779 dx_set_count(entries
, count
+ 1);
782 static void ext3_update_dx_flag(struct inode
*inode
)
784 if (!EXT3_HAS_COMPAT_FEATURE(inode
->i_sb
,
785 EXT3_FEATURE_COMPAT_DIR_INDEX
))
786 EXT3_I(inode
)->i_flags
&= ~EXT3_INDEX_FL
;
790 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
792 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
793 * `de != NULL' is guaranteed by caller.
795 static inline int ext3_match (int len
, const char * const name
,
796 struct ext3_dir_entry_2
* de
)
798 if (len
!= de
->name_len
)
802 return !memcmp(name
, de
->name
, len
);
806 * Returns 0 if not found, -1 on failure, and 1 on success
808 static inline int search_dirblock(struct buffer_head
* bh
,
811 unsigned long offset
,
812 struct ext3_dir_entry_2
** res_dir
)
814 struct ext3_dir_entry_2
* de
;
817 const char *name
= child
->name
;
818 int namelen
= child
->len
;
820 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
821 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
822 while ((char *) de
< dlimit
) {
823 /* this code is executed quadratically often */
824 /* do minimal checking `by hand' */
826 if ((char *) de
+ namelen
<= dlimit
&&
827 ext3_match (namelen
, name
, de
)) {
828 /* found a match - just to be sure, do a full check */
829 if (!ext3_check_dir_entry("ext3_find_entry",
830 dir
, de
, bh
, offset
))
835 /* prevent looping on a bad block */
836 de_len
= ext3_rec_len_from_disk(de
->rec_len
);
840 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ de_len
);
849 * finds an entry in the specified directory with the wanted name. It
850 * returns the cache buffer in which the entry was found, and the entry
851 * itself (as a parameter - res_dir). It does NOT read the inode of the
852 * entry - you'll have to do that yourself if you want to.
854 * The returned buffer_head has ->b_count elevated. The caller is expected
855 * to brelse() it when appropriate.
857 static struct buffer_head
*ext3_find_entry(struct inode
*dir
,
859 struct ext3_dir_entry_2
**res_dir
)
861 struct super_block
* sb
;
862 struct buffer_head
* bh_use
[NAMEI_RA_SIZE
];
863 struct buffer_head
* bh
, *ret
= NULL
;
864 unsigned long start
, block
, b
;
865 int ra_max
= 0; /* Number of bh's in the readahead
867 int ra_ptr
= 0; /* Current index into readahead
875 namelen
= entry
->len
;
876 if (namelen
> EXT3_NAME_LEN
)
879 bh
= ext3_dx_find_entry(dir
, entry
, res_dir
, &err
);
881 * On success, or if the error was file not found,
882 * return. Otherwise, fall back to doing a search the
885 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
887 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
889 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
890 start
= EXT3_I(dir
)->i_dir_start_lookup
;
891 if (start
>= nblocks
)
897 * We deal with the read-ahead logic here.
899 if (ra_ptr
>= ra_max
) {
900 /* Refill the readahead buffer */
903 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
905 * Terminate if we reach the end of the
906 * directory and must wrap, or if our
907 * search has finished at this block.
909 if (b
>= nblocks
|| (num
&& block
== start
)) {
910 bh_use
[ra_max
] = NULL
;
914 bh
= ext3_getblk(NULL
, dir
, b
++, 0, &err
);
917 ll_rw_block(READ_META
, 1, &bh
);
920 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
923 if (!buffer_uptodate(bh
)) {
924 /* read error, skip block & hope for the best */
925 ext3_error(sb
, __func__
, "reading directory #%lu "
926 "offset %lu", dir
->i_ino
, block
);
930 i
= search_dirblock(bh
, dir
, entry
,
931 block
<< EXT3_BLOCK_SIZE_BITS(sb
), res_dir
);
933 EXT3_I(dir
)->i_dir_start_lookup
= block
;
935 goto cleanup_and_exit
;
939 goto cleanup_and_exit
;
942 if (++block
>= nblocks
)
944 } while (block
!= start
);
947 * If the directory has grown while we were searching, then
948 * search the last part of the directory before giving up.
951 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
952 if (block
< nblocks
) {
958 /* Clean up the read-ahead blocks */
959 for (; ra_ptr
< ra_max
; ra_ptr
++)
960 brelse (bh_use
[ra_ptr
]);
964 static struct buffer_head
* ext3_dx_find_entry(struct inode
*dir
,
965 struct qstr
*entry
, struct ext3_dir_entry_2
**res_dir
,
968 struct super_block
* sb
;
969 struct dx_hash_info hinfo
;
971 struct dx_frame frames
[2], *frame
;
972 struct ext3_dir_entry_2
*de
, *top
;
973 struct buffer_head
*bh
;
976 int namelen
= entry
->len
;
977 const u8
*name
= entry
->name
;
980 /* NFS may look up ".." - look at dx_root directory block */
981 if (namelen
> 2 || name
[0] != '.'|| (namelen
== 2 && name
[1] != '.')) {
982 if (!(frame
= dx_probe(entry
, dir
, &hinfo
, frames
, err
)))
986 frame
->bh
= NULL
; /* for dx_release() */
987 frame
->at
= (struct dx_entry
*)frames
; /* hack for zero entry*/
988 dx_set_block(frame
->at
, 0); /* dx_root block is 0 */
992 block
= dx_get_block(frame
->at
);
993 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0, err
)))
995 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
996 top
= (struct ext3_dir_entry_2
*) ((char *) de
+ sb
->s_blocksize
-
997 EXT3_DIR_REC_LEN(0));
998 for (; de
< top
; de
= ext3_next_entry(de
)) {
999 int off
= (block
<< EXT3_BLOCK_SIZE_BITS(sb
))
1000 + ((char *) de
- bh
->b_data
);
1002 if (!ext3_check_dir_entry(__func__
, dir
, de
, bh
, off
)) {
1004 *err
= ERR_BAD_DX_DIR
;
1008 if (ext3_match(namelen
, name
, de
)) {
1015 /* Check to see if we should continue to search */
1016 retval
= ext3_htree_next_block(dir
, hash
, frame
,
1019 ext3_warning(sb
, __func__
,
1020 "error reading index page in directory #%lu",
1025 } while (retval
== 1);
1029 dxtrace(printk("%s not found\n", name
));
1030 dx_release (frames
);
1034 static struct dentry
*ext3_lookup(struct inode
* dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1036 struct inode
* inode
;
1037 struct ext3_dir_entry_2
* de
;
1038 struct buffer_head
* bh
;
1040 if (dentry
->d_name
.len
> EXT3_NAME_LEN
)
1041 return ERR_PTR(-ENAMETOOLONG
);
1043 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
1046 unsigned long ino
= le32_to_cpu(de
->inode
);
1048 if (!ext3_valid_inum(dir
->i_sb
, ino
)) {
1049 ext3_error(dir
->i_sb
, "ext3_lookup",
1050 "bad inode number: %lu", ino
);
1051 return ERR_PTR(-EIO
);
1053 inode
= ext3_iget(dir
->i_sb
, ino
);
1055 return ERR_CAST(inode
);
1057 return d_splice_alias(inode
, dentry
);
1061 struct dentry
*ext3_get_parent(struct dentry
*child
)
1064 struct qstr dotdot
= {.name
= "..", .len
= 2};
1065 struct ext3_dir_entry_2
* de
;
1066 struct buffer_head
*bh
;
1068 bh
= ext3_find_entry(child
->d_inode
, &dotdot
, &de
);
1070 return ERR_PTR(-ENOENT
);
1071 ino
= le32_to_cpu(de
->inode
);
1074 if (!ext3_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1075 ext3_error(child
->d_inode
->i_sb
, "ext3_get_parent",
1076 "bad inode number: %lu", ino
);
1077 return ERR_PTR(-EIO
);
1080 return d_obtain_alias(ext3_iget(child
->d_inode
->i_sb
, ino
));
1084 static unsigned char ext3_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1085 [S_IFREG
>> S_SHIFT
] = EXT3_FT_REG_FILE
,
1086 [S_IFDIR
>> S_SHIFT
] = EXT3_FT_DIR
,
1087 [S_IFCHR
>> S_SHIFT
] = EXT3_FT_CHRDEV
,
1088 [S_IFBLK
>> S_SHIFT
] = EXT3_FT_BLKDEV
,
1089 [S_IFIFO
>> S_SHIFT
] = EXT3_FT_FIFO
,
1090 [S_IFSOCK
>> S_SHIFT
] = EXT3_FT_SOCK
,
1091 [S_IFLNK
>> S_SHIFT
] = EXT3_FT_SYMLINK
,
1094 static inline void ext3_set_de_type(struct super_block
*sb
,
1095 struct ext3_dir_entry_2
*de
,
1097 if (EXT3_HAS_INCOMPAT_FEATURE(sb
, EXT3_FEATURE_INCOMPAT_FILETYPE
))
1098 de
->file_type
= ext3_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1102 * Move count entries from end of map between two memory locations.
1103 * Returns pointer to last entry moved.
1105 static struct ext3_dir_entry_2
*
1106 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1108 unsigned rec_len
= 0;
1111 struct ext3_dir_entry_2
*de
= (struct ext3_dir_entry_2
*) (from
+ map
->offs
);
1112 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1113 memcpy (to
, de
, rec_len
);
1114 ((struct ext3_dir_entry_2
*) to
)->rec_len
=
1115 ext3_rec_len_to_disk(rec_len
);
1120 return (struct ext3_dir_entry_2
*) (to
- rec_len
);
1124 * Compact each dir entry in the range to the minimal rec_len.
1125 * Returns pointer to last entry in range.
1127 static struct ext3_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1129 struct ext3_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext3_dir_entry_2
*) base
;
1130 unsigned rec_len
= 0;
1133 while ((char*)de
< base
+ size
) {
1134 next
= ext3_next_entry(de
);
1135 if (de
->inode
&& de
->name_len
) {
1136 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1138 memmove(to
, de
, rec_len
);
1139 to
->rec_len
= ext3_rec_len_to_disk(rec_len
);
1141 to
= (struct ext3_dir_entry_2
*) (((char *) to
) + rec_len
);
1149 * Split a full leaf block to make room for a new dir entry.
1150 * Allocate a new block, and move entries so that they are approx. equally full.
1151 * Returns pointer to de in block into which the new entry will be inserted.
1153 static struct ext3_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1154 struct buffer_head
**bh
,struct dx_frame
*frame
,
1155 struct dx_hash_info
*hinfo
, int *error
)
1157 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1158 unsigned count
, continued
;
1159 struct buffer_head
*bh2
;
1162 struct dx_map_entry
*map
;
1163 char *data1
= (*bh
)->b_data
, *data2
;
1164 unsigned split
, move
, size
;
1165 struct ext3_dir_entry_2
*de
= NULL
, *de2
;
1168 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1175 BUFFER_TRACE(*bh
, "get_write_access");
1176 err
= ext3_journal_get_write_access(handle
, *bh
);
1180 BUFFER_TRACE(frame
->bh
, "get_write_access");
1181 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1185 data2
= bh2
->b_data
;
1187 /* create map in the end of data2 block */
1188 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1189 count
= dx_make_map ((struct ext3_dir_entry_2
*) data1
,
1190 blocksize
, hinfo
, map
);
1192 dx_sort_map (map
, count
);
1193 /* Split the existing block in the middle, size-wise */
1196 for (i
= count
-1; i
>= 0; i
--) {
1197 /* is more than half of this entry in 2nd half of the block? */
1198 if (size
+ map
[i
].size
/2 > blocksize
/2)
1200 size
+= map
[i
].size
;
1203 /* map index at which we will split */
1204 split
= count
- move
;
1205 hash2
= map
[split
].hash
;
1206 continued
= hash2
== map
[split
- 1].hash
;
1207 dxtrace(printk("Split block %i at %x, %i/%i\n",
1208 dx_get_block(frame
->at
), hash2
, split
, count
-split
));
1210 /* Fancy dance to stay within two buffers */
1211 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1212 de
= dx_pack_dirents(data1
,blocksize
);
1213 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1214 de2
->rec_len
= ext3_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1215 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data1
, blocksize
, 1));
1216 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data2
, blocksize
, 1));
1218 /* Which block gets the new entry? */
1219 if (hinfo
->hash
>= hash2
)
1224 dx_insert_block (frame
, hash2
+ continued
, newblock
);
1225 err
= ext3_journal_dirty_metadata (handle
, bh2
);
1228 err
= ext3_journal_dirty_metadata (handle
, frame
->bh
);
1232 dxtrace(dx_show_index ("frame", frame
->entries
));
1239 ext3_std_error(dir
->i_sb
, err
);
1247 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1248 * it points to a directory entry which is guaranteed to be large
1249 * enough for new directory entry. If de is NULL, then
1250 * add_dirent_to_buf will attempt search the directory block for
1251 * space. It will return -ENOSPC if no space is available, and -EIO
1252 * and -EEXIST if directory entry already exists.
1254 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1255 * all other cases bh is released.
1257 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1258 struct inode
*inode
, struct ext3_dir_entry_2
*de
,
1259 struct buffer_head
* bh
)
1261 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1262 const char *name
= dentry
->d_name
.name
;
1263 int namelen
= dentry
->d_name
.len
;
1264 unsigned long offset
= 0;
1265 unsigned short reclen
;
1266 int nlen
, rlen
, err
;
1269 reclen
= EXT3_DIR_REC_LEN(namelen
);
1271 de
= (struct ext3_dir_entry_2
*)bh
->b_data
;
1272 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1273 while ((char *) de
<= top
) {
1274 if (!ext3_check_dir_entry("ext3_add_entry", dir
, de
,
1279 if (ext3_match (namelen
, name
, de
)) {
1283 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1284 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1285 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1287 de
= (struct ext3_dir_entry_2
*)((char *)de
+ rlen
);
1290 if ((char *) de
> top
)
1293 BUFFER_TRACE(bh
, "get_write_access");
1294 err
= ext3_journal_get_write_access(handle
, bh
);
1296 ext3_std_error(dir
->i_sb
, err
);
1301 /* By now the buffer is marked for journaling */
1302 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1303 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1305 struct ext3_dir_entry_2
*de1
= (struct ext3_dir_entry_2
*)((char *)de
+ nlen
);
1306 de1
->rec_len
= ext3_rec_len_to_disk(rlen
- nlen
);
1307 de
->rec_len
= ext3_rec_len_to_disk(nlen
);
1310 de
->file_type
= EXT3_FT_UNKNOWN
;
1312 de
->inode
= cpu_to_le32(inode
->i_ino
);
1313 ext3_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1316 de
->name_len
= namelen
;
1317 memcpy (de
->name
, name
, namelen
);
1319 * XXX shouldn't update any times until successful
1320 * completion of syscall, but too many callers depend
1323 * XXX similarly, too many callers depend on
1324 * ext3_new_inode() setting the times, but error
1325 * recovery deletes the inode, so the worst that can
1326 * happen is that the times are slightly out of date
1327 * and/or different from the directory change time.
1329 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME_SEC
;
1330 ext3_update_dx_flag(dir
);
1332 ext3_mark_inode_dirty(handle
, dir
);
1333 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1334 err
= ext3_journal_dirty_metadata(handle
, bh
);
1336 ext3_std_error(dir
->i_sb
, err
);
1342 * This converts a one block unindexed directory to a 3 block indexed
1343 * directory, and adds the dentry to the indexed directory.
1345 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1346 struct inode
*inode
, struct buffer_head
*bh
)
1348 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1349 const char *name
= dentry
->d_name
.name
;
1350 int namelen
= dentry
->d_name
.len
;
1351 struct buffer_head
*bh2
;
1352 struct dx_root
*root
;
1353 struct dx_frame frames
[2], *frame
;
1354 struct dx_entry
*entries
;
1355 struct ext3_dir_entry_2
*de
, *de2
;
1360 struct dx_hash_info hinfo
;
1362 struct fake_dirent
*fde
;
1364 blocksize
= dir
->i_sb
->s_blocksize
;
1365 dxtrace(printk("Creating index\n"));
1366 retval
= ext3_journal_get_write_access(handle
, bh
);
1368 ext3_std_error(dir
->i_sb
, retval
);
1372 root
= (struct dx_root
*) bh
->b_data
;
1374 bh2
= ext3_append (handle
, dir
, &block
, &retval
);
1379 EXT3_I(dir
)->i_flags
|= EXT3_INDEX_FL
;
1380 data1
= bh2
->b_data
;
1382 /* The 0th block becomes the root, move the dirents out */
1383 fde
= &root
->dotdot
;
1384 de
= (struct ext3_dir_entry_2
*)((char *)fde
+
1385 ext3_rec_len_from_disk(fde
->rec_len
));
1386 len
= ((char *) root
) + blocksize
- (char *) de
;
1387 memcpy (data1
, de
, len
);
1388 de
= (struct ext3_dir_entry_2
*) data1
;
1390 while ((char *)(de2
= ext3_next_entry(de
)) < top
)
1392 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1393 /* Initialize the root; the dot dirents already exist */
1394 de
= (struct ext3_dir_entry_2
*) (&root
->dotdot
);
1395 de
->rec_len
= ext3_rec_len_to_disk(blocksize
- EXT3_DIR_REC_LEN(2));
1396 memset (&root
->info
, 0, sizeof(root
->info
));
1397 root
->info
.info_length
= sizeof(root
->info
);
1398 root
->info
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
1399 entries
= root
->entries
;
1400 dx_set_block (entries
, 1);
1401 dx_set_count (entries
, 1);
1402 dx_set_limit (entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1404 /* Initialize as for dx_probe */
1405 hinfo
.hash_version
= root
->info
.hash_version
;
1406 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1407 hinfo
.hash_version
+= EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
1408 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
1409 ext3fs_dirhash(name
, namelen
, &hinfo
);
1411 frame
->entries
= entries
;
1412 frame
->at
= entries
;
1415 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1416 dx_release (frames
);
1420 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1426 * adds a file entry to the specified directory, using the same
1427 * semantics as ext3_find_entry(). It returns NULL if it failed.
1429 * NOTE!! The inode part of 'de' is left at 0 - which means you
1430 * may not sleep between calling this and putting something into
1431 * the entry, as someone else might have used it while you slept.
1433 static int ext3_add_entry (handle_t
*handle
, struct dentry
*dentry
,
1434 struct inode
*inode
)
1436 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1437 unsigned long offset
;
1438 struct buffer_head
* bh
;
1439 struct ext3_dir_entry_2
*de
;
1440 struct super_block
* sb
;
1447 blocksize
= sb
->s_blocksize
;
1448 if (!dentry
->d_name
.len
)
1451 retval
= ext3_dx_add_entry(handle
, dentry
, inode
);
1452 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1454 EXT3_I(dir
)->i_flags
&= ~EXT3_INDEX_FL
;
1456 ext3_mark_inode_dirty(handle
, dir
);
1458 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1459 for (block
= 0, offset
= 0; block
< blocks
; block
++) {
1460 bh
= ext3_bread(handle
, dir
, block
, 0, &retval
);
1463 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1464 if (retval
!= -ENOSPC
)
1467 if (blocks
== 1 && !dx_fallback
&&
1468 EXT3_HAS_COMPAT_FEATURE(sb
, EXT3_FEATURE_COMPAT_DIR_INDEX
))
1469 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1472 bh
= ext3_append(handle
, dir
, &block
, &retval
);
1475 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1477 de
->rec_len
= ext3_rec_len_to_disk(blocksize
);
1478 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1482 * Returns 0 for success, or a negative error value
1484 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1485 struct inode
*inode
)
1487 struct dx_frame frames
[2], *frame
;
1488 struct dx_entry
*entries
, *at
;
1489 struct dx_hash_info hinfo
;
1490 struct buffer_head
* bh
;
1491 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1492 struct super_block
* sb
= dir
->i_sb
;
1493 struct ext3_dir_entry_2
*de
;
1496 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1499 entries
= frame
->entries
;
1502 if (!(bh
= ext3_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1505 BUFFER_TRACE(bh
, "get_write_access");
1506 err
= ext3_journal_get_write_access(handle
, bh
);
1510 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1511 if (err
!= -ENOSPC
) {
1516 /* Block full, should compress but for now just split */
1517 dxtrace(printk("using %u of %u node entries\n",
1518 dx_get_count(entries
), dx_get_limit(entries
)));
1519 /* Need to split index? */
1520 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1522 unsigned icount
= dx_get_count(entries
);
1523 int levels
= frame
- frames
;
1524 struct dx_entry
*entries2
;
1525 struct dx_node
*node2
;
1526 struct buffer_head
*bh2
;
1528 if (levels
&& (dx_get_count(frames
->entries
) ==
1529 dx_get_limit(frames
->entries
))) {
1530 ext3_warning(sb
, __func__
,
1531 "Directory index full!");
1535 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1538 node2
= (struct dx_node
*)(bh2
->b_data
);
1539 entries2
= node2
->entries
;
1540 node2
->fake
.rec_len
= ext3_rec_len_to_disk(sb
->s_blocksize
);
1541 node2
->fake
.inode
= 0;
1542 BUFFER_TRACE(frame
->bh
, "get_write_access");
1543 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1547 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1548 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1549 dxtrace(printk("Split index %i/%i\n", icount1
, icount2
));
1551 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1552 err
= ext3_journal_get_write_access(handle
,
1557 memcpy ((char *) entries2
, (char *) (entries
+ icount1
),
1558 icount2
* sizeof(struct dx_entry
));
1559 dx_set_count (entries
, icount1
);
1560 dx_set_count (entries2
, icount2
);
1561 dx_set_limit (entries2
, dx_node_limit(dir
));
1563 /* Which index block gets the new entry? */
1564 if (at
- entries
>= icount1
) {
1565 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1566 frame
->entries
= entries
= entries2
;
1567 swap(frame
->bh
, bh2
);
1569 dx_insert_block (frames
+ 0, hash2
, newblock
);
1570 dxtrace(dx_show_index ("node", frames
[1].entries
));
1571 dxtrace(dx_show_index ("node",
1572 ((struct dx_node
*) bh2
->b_data
)->entries
));
1573 err
= ext3_journal_dirty_metadata(handle
, bh2
);
1578 dxtrace(printk("Creating second level index...\n"));
1579 memcpy((char *) entries2
, (char *) entries
,
1580 icount
* sizeof(struct dx_entry
));
1581 dx_set_limit(entries2
, dx_node_limit(dir
));
1584 dx_set_count(entries
, 1);
1585 dx_set_block(entries
+ 0, newblock
);
1586 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1588 /* Add new access path frame */
1590 frame
->at
= at
= at
- entries
+ entries2
;
1591 frame
->entries
= entries
= entries2
;
1593 err
= ext3_journal_get_write_access(handle
,
1598 ext3_journal_dirty_metadata(handle
, frames
[0].bh
);
1600 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1603 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1608 ext3_std_error(dir
->i_sb
, err
);
1617 * ext3_delete_entry deletes a directory entry by merging it with the
1620 static int ext3_delete_entry (handle_t
*handle
,
1622 struct ext3_dir_entry_2
* de_del
,
1623 struct buffer_head
* bh
)
1625 struct ext3_dir_entry_2
* de
, * pde
;
1630 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1631 while (i
< bh
->b_size
) {
1632 if (!ext3_check_dir_entry("ext3_delete_entry", dir
, de
, bh
, i
))
1635 BUFFER_TRACE(bh
, "get_write_access");
1636 ext3_journal_get_write_access(handle
, bh
);
1638 pde
->rec_len
= ext3_rec_len_to_disk(
1639 ext3_rec_len_from_disk(pde
->rec_len
) +
1640 ext3_rec_len_from_disk(de
->rec_len
));
1644 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1645 ext3_journal_dirty_metadata(handle
, bh
);
1648 i
+= ext3_rec_len_from_disk(de
->rec_len
);
1650 de
= ext3_next_entry(de
);
1655 static int ext3_add_nondir(handle_t
*handle
,
1656 struct dentry
*dentry
, struct inode
*inode
)
1658 int err
= ext3_add_entry(handle
, dentry
, inode
);
1660 ext3_mark_inode_dirty(handle
, inode
);
1661 d_instantiate(dentry
, inode
);
1662 unlock_new_inode(inode
);
1666 unlock_new_inode(inode
);
1672 * By the time this is called, we already have created
1673 * the directory cache entry for the new file, but it
1674 * is so far negative - it has no inode.
1676 * If the create succeeds, we fill in the inode information
1677 * with d_instantiate().
1679 static int ext3_create (struct inode
* dir
, struct dentry
* dentry
, int mode
,
1680 struct nameidata
*nd
)
1683 struct inode
* inode
;
1684 int err
, retries
= 0;
1687 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1688 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1689 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1691 return PTR_ERR(handle
);
1693 if (IS_DIRSYNC(dir
))
1696 inode
= ext3_new_inode (handle
, dir
, mode
);
1697 err
= PTR_ERR(inode
);
1698 if (!IS_ERR(inode
)) {
1699 inode
->i_op
= &ext3_file_inode_operations
;
1700 inode
->i_fop
= &ext3_file_operations
;
1701 ext3_set_aops(inode
);
1702 err
= ext3_add_nondir(handle
, dentry
, inode
);
1704 ext3_journal_stop(handle
);
1705 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1710 static int ext3_mknod (struct inode
* dir
, struct dentry
*dentry
,
1711 int mode
, dev_t rdev
)
1714 struct inode
*inode
;
1715 int err
, retries
= 0;
1717 if (!new_valid_dev(rdev
))
1721 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1722 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1723 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1725 return PTR_ERR(handle
);
1727 if (IS_DIRSYNC(dir
))
1730 inode
= ext3_new_inode (handle
, dir
, mode
);
1731 err
= PTR_ERR(inode
);
1732 if (!IS_ERR(inode
)) {
1733 init_special_inode(inode
, inode
->i_mode
, rdev
);
1734 #ifdef CONFIG_EXT3_FS_XATTR
1735 inode
->i_op
= &ext3_special_inode_operations
;
1737 err
= ext3_add_nondir(handle
, dentry
, inode
);
1739 ext3_journal_stop(handle
);
1740 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1745 static int ext3_mkdir(struct inode
* dir
, struct dentry
* dentry
, int mode
)
1748 struct inode
* inode
;
1749 struct buffer_head
* dir_block
;
1750 struct ext3_dir_entry_2
* de
;
1751 int err
, retries
= 0;
1753 if (dir
->i_nlink
>= EXT3_LINK_MAX
)
1757 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1758 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1759 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1761 return PTR_ERR(handle
);
1763 if (IS_DIRSYNC(dir
))
1766 inode
= ext3_new_inode (handle
, dir
, S_IFDIR
| mode
);
1767 err
= PTR_ERR(inode
);
1771 inode
->i_op
= &ext3_dir_inode_operations
;
1772 inode
->i_fop
= &ext3_dir_operations
;
1773 inode
->i_size
= EXT3_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1774 dir_block
= ext3_bread (handle
, inode
, 0, 1, &err
);
1776 drop_nlink(inode
); /* is this nlink == 0? */
1777 unlock_new_inode(inode
);
1778 ext3_mark_inode_dirty(handle
, inode
);
1782 BUFFER_TRACE(dir_block
, "get_write_access");
1783 ext3_journal_get_write_access(handle
, dir_block
);
1784 de
= (struct ext3_dir_entry_2
*) dir_block
->b_data
;
1785 de
->inode
= cpu_to_le32(inode
->i_ino
);
1787 de
->rec_len
= ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de
->name_len
));
1788 strcpy (de
->name
, ".");
1789 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1790 de
= ext3_next_entry(de
);
1791 de
->inode
= cpu_to_le32(dir
->i_ino
);
1792 de
->rec_len
= ext3_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1793 EXT3_DIR_REC_LEN(1));
1795 strcpy (de
->name
, "..");
1796 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1798 BUFFER_TRACE(dir_block
, "call ext3_journal_dirty_metadata");
1799 ext3_journal_dirty_metadata(handle
, dir_block
);
1801 ext3_mark_inode_dirty(handle
, inode
);
1802 err
= ext3_add_entry (handle
, dentry
, inode
);
1805 unlock_new_inode(inode
);
1806 ext3_mark_inode_dirty(handle
, inode
);
1811 ext3_update_dx_flag(dir
);
1812 ext3_mark_inode_dirty(handle
, dir
);
1813 d_instantiate(dentry
, inode
);
1814 unlock_new_inode(inode
);
1816 ext3_journal_stop(handle
);
1817 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1823 * routine to check that the specified directory is empty (for rmdir)
1825 static int empty_dir (struct inode
* inode
)
1827 unsigned long offset
;
1828 struct buffer_head
* bh
;
1829 struct ext3_dir_entry_2
* de
, * de1
;
1830 struct super_block
* sb
;
1834 if (inode
->i_size
< EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1835 !(bh
= ext3_bread (NULL
, inode
, 0, 0, &err
))) {
1837 ext3_error(inode
->i_sb
, __func__
,
1838 "error %d reading directory #%lu offset 0",
1841 ext3_warning(inode
->i_sb
, __func__
,
1842 "bad directory (dir #%lu) - no data block",
1846 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1847 de1
= ext3_next_entry(de
);
1848 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1849 !le32_to_cpu(de1
->inode
) ||
1850 strcmp (".", de
->name
) ||
1851 strcmp ("..", de1
->name
)) {
1852 ext3_warning (inode
->i_sb
, "empty_dir",
1853 "bad directory (dir #%lu) - no `.' or `..'",
1858 offset
= ext3_rec_len_from_disk(de
->rec_len
) +
1859 ext3_rec_len_from_disk(de1
->rec_len
);
1860 de
= ext3_next_entry(de1
);
1861 while (offset
< inode
->i_size
) {
1863 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1866 bh
= ext3_bread (NULL
, inode
,
1867 offset
>> EXT3_BLOCK_SIZE_BITS(sb
), 0, &err
);
1870 ext3_error(sb
, __func__
,
1871 "error %d reading directory"
1873 err
, inode
->i_ino
, offset
);
1874 offset
+= sb
->s_blocksize
;
1877 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1879 if (!ext3_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1880 de
= (struct ext3_dir_entry_2
*)(bh
->b_data
+
1882 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1885 if (le32_to_cpu(de
->inode
)) {
1889 offset
+= ext3_rec_len_from_disk(de
->rec_len
);
1890 de
= ext3_next_entry(de
);
1896 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1897 * such inodes, starting at the superblock, in case we crash before the
1898 * file is closed/deleted, or in case the inode truncate spans multiple
1899 * transactions and the last transaction is not recovered after a crash.
1901 * At filesystem recovery time, we walk this list deleting unlinked
1902 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1904 int ext3_orphan_add(handle_t
*handle
, struct inode
*inode
)
1906 struct super_block
*sb
= inode
->i_sb
;
1907 struct ext3_iloc iloc
;
1911 if (!list_empty(&EXT3_I(inode
)->i_orphan
))
1914 /* Orphan handling is only valid for files with data blocks
1915 * being truncated, or files being unlinked. */
1917 /* @@@ FIXME: Observation from aviro:
1918 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1919 * here (on lock_super()), so race with ext3_link() which might bump
1920 * ->i_nlink. For, say it, character device. Not a regular file,
1921 * not a directory, not a symlink and ->i_nlink > 0.
1923 J_ASSERT ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1924 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1926 BUFFER_TRACE(EXT3_SB(sb
)->s_sbh
, "get_write_access");
1927 err
= ext3_journal_get_write_access(handle
, EXT3_SB(sb
)->s_sbh
);
1931 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1935 /* Insert this inode at the head of the on-disk orphan list... */
1936 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT3_SB(sb
)->s_es
->s_last_orphan
);
1937 EXT3_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1938 err
= ext3_journal_dirty_metadata(handle
, EXT3_SB(sb
)->s_sbh
);
1939 rc
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
1943 /* Only add to the head of the in-memory list if all the
1944 * previous operations succeeded. If the orphan_add is going to
1945 * fail (possibly taking the journal offline), we can't risk
1946 * leaving the inode on the orphan list: stray orphan-list
1947 * entries can cause panics at unmount time.
1949 * This is safe: on error we're going to ignore the orphan list
1950 * anyway on the next recovery. */
1952 list_add(&EXT3_I(inode
)->i_orphan
, &EXT3_SB(sb
)->s_orphan
);
1954 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1955 jbd_debug(4, "orphan inode %lu will point to %d\n",
1956 inode
->i_ino
, NEXT_ORPHAN(inode
));
1959 ext3_std_error(inode
->i_sb
, err
);
1964 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1965 * of such inodes stored on disk, because it is finally being cleaned up.
1967 int ext3_orphan_del(handle_t
*handle
, struct inode
*inode
)
1969 struct list_head
*prev
;
1970 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1971 struct ext3_sb_info
*sbi
;
1972 unsigned long ino_next
;
1973 struct ext3_iloc iloc
;
1976 lock_super(inode
->i_sb
);
1977 if (list_empty(&ei
->i_orphan
)) {
1978 unlock_super(inode
->i_sb
);
1982 ino_next
= NEXT_ORPHAN(inode
);
1983 prev
= ei
->i_orphan
.prev
;
1984 sbi
= EXT3_SB(inode
->i_sb
);
1986 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
1988 list_del_init(&ei
->i_orphan
);
1990 /* If we're on an error path, we may not have a valid
1991 * transaction handle with which to update the orphan list on
1992 * disk, but we still need to remove the inode from the linked
1993 * list in memory. */
1997 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
2001 if (prev
== &sbi
->s_orphan
) {
2002 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
2003 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2004 err
= ext3_journal_get_write_access(handle
, sbi
->s_sbh
);
2007 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2008 err
= ext3_journal_dirty_metadata(handle
, sbi
->s_sbh
);
2010 struct ext3_iloc iloc2
;
2011 struct inode
*i_prev
=
2012 &list_entry(prev
, struct ext3_inode_info
, i_orphan
)->vfs_inode
;
2014 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2015 i_prev
->i_ino
, ino_next
);
2016 err
= ext3_reserve_inode_write(handle
, i_prev
, &iloc2
);
2019 NEXT_ORPHAN(i_prev
) = ino_next
;
2020 err
= ext3_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2024 NEXT_ORPHAN(inode
) = 0;
2025 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
2028 ext3_std_error(inode
->i_sb
, err
);
2030 unlock_super(inode
->i_sb
);
2038 static int ext3_rmdir (struct inode
* dir
, struct dentry
*dentry
)
2041 struct inode
* inode
;
2042 struct buffer_head
* bh
;
2043 struct ext3_dir_entry_2
* de
;
2046 /* Initialize quotas before so that eventual writes go in
2047 * separate transaction */
2048 DQUOT_INIT(dentry
->d_inode
);
2049 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2051 return PTR_ERR(handle
);
2054 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2058 if (IS_DIRSYNC(dir
))
2061 inode
= dentry
->d_inode
;
2064 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2067 retval
= -ENOTEMPTY
;
2068 if (!empty_dir (inode
))
2071 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2074 if (inode
->i_nlink
!= 2)
2075 ext3_warning (inode
->i_sb
, "ext3_rmdir",
2076 "empty directory has nlink!=2 (%d)",
2080 /* There's no need to set i_disksize: the fact that i_nlink is
2081 * zero will ensure that the right thing happens during any
2084 ext3_orphan_add(handle
, inode
);
2085 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2086 ext3_mark_inode_dirty(handle
, inode
);
2088 ext3_update_dx_flag(dir
);
2089 ext3_mark_inode_dirty(handle
, dir
);
2092 ext3_journal_stop(handle
);
2097 static int ext3_unlink(struct inode
* dir
, struct dentry
*dentry
)
2100 struct inode
* inode
;
2101 struct buffer_head
* bh
;
2102 struct ext3_dir_entry_2
* de
;
2105 /* Initialize quotas before so that eventual writes go
2106 * in separate transaction */
2107 DQUOT_INIT(dentry
->d_inode
);
2108 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2110 return PTR_ERR(handle
);
2112 if (IS_DIRSYNC(dir
))
2116 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2120 inode
= dentry
->d_inode
;
2123 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2126 if (!inode
->i_nlink
) {
2127 ext3_warning (inode
->i_sb
, "ext3_unlink",
2128 "Deleting nonexistent file (%lu), %d",
2129 inode
->i_ino
, inode
->i_nlink
);
2132 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2135 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2136 ext3_update_dx_flag(dir
);
2137 ext3_mark_inode_dirty(handle
, dir
);
2139 if (!inode
->i_nlink
)
2140 ext3_orphan_add(handle
, inode
);
2141 inode
->i_ctime
= dir
->i_ctime
;
2142 ext3_mark_inode_dirty(handle
, inode
);
2146 ext3_journal_stop(handle
);
2151 static int ext3_symlink (struct inode
* dir
,
2152 struct dentry
*dentry
, const char * symname
)
2155 struct inode
* inode
;
2156 int l
, err
, retries
= 0;
2158 l
= strlen(symname
)+1;
2159 if (l
> dir
->i_sb
->s_blocksize
)
2160 return -ENAMETOOLONG
;
2163 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2164 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2165 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2167 return PTR_ERR(handle
);
2169 if (IS_DIRSYNC(dir
))
2172 inode
= ext3_new_inode (handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2173 err
= PTR_ERR(inode
);
2177 if (l
> sizeof (EXT3_I(inode
)->i_data
)) {
2178 inode
->i_op
= &ext3_symlink_inode_operations
;
2179 ext3_set_aops(inode
);
2181 * page_symlink() calls into ext3_prepare/commit_write.
2182 * We have a transaction open. All is sweetness. It also sets
2183 * i_size in generic_commit_write().
2185 err
= __page_symlink(inode
, symname
, l
, 1);
2188 unlock_new_inode(inode
);
2189 ext3_mark_inode_dirty(handle
, inode
);
2194 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2195 memcpy((char*)&EXT3_I(inode
)->i_data
,symname
,l
);
2196 inode
->i_size
= l
-1;
2198 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
2199 err
= ext3_add_nondir(handle
, dentry
, inode
);
2201 ext3_journal_stop(handle
);
2202 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2207 static int ext3_link (struct dentry
* old_dentry
,
2208 struct inode
* dir
, struct dentry
*dentry
)
2211 struct inode
*inode
= old_dentry
->d_inode
;
2212 int err
, retries
= 0;
2214 if (inode
->i_nlink
>= EXT3_LINK_MAX
)
2217 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2218 * otherwise has the potential to corrupt the orphan inode list.
2220 if (inode
->i_nlink
== 0)
2224 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2225 EXT3_INDEX_EXTRA_TRANS_BLOCKS
);
2227 return PTR_ERR(handle
);
2229 if (IS_DIRSYNC(dir
))
2232 inode
->i_ctime
= CURRENT_TIME_SEC
;
2234 atomic_inc(&inode
->i_count
);
2236 err
= ext3_add_entry(handle
, dentry
, inode
);
2238 ext3_mark_inode_dirty(handle
, inode
);
2239 d_instantiate(dentry
, inode
);
2244 ext3_journal_stop(handle
);
2245 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2250 #define PARENT_INO(buffer) \
2251 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2254 * Anybody can rename anything with this: the permission checks are left to the
2255 * higher-level routines.
2257 static int ext3_rename (struct inode
* old_dir
, struct dentry
*old_dentry
,
2258 struct inode
* new_dir
,struct dentry
*new_dentry
)
2261 struct inode
* old_inode
, * new_inode
;
2262 struct buffer_head
* old_bh
, * new_bh
, * dir_bh
;
2263 struct ext3_dir_entry_2
* old_de
, * new_de
;
2266 old_bh
= new_bh
= dir_bh
= NULL
;
2268 /* Initialize quotas before so that eventual writes go
2269 * in separate transaction */
2270 if (new_dentry
->d_inode
)
2271 DQUOT_INIT(new_dentry
->d_inode
);
2272 handle
= ext3_journal_start(old_dir
, 2 *
2273 EXT3_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2274 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2276 return PTR_ERR(handle
);
2278 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2281 old_bh
= ext3_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2283 * Check for inode number is _not_ due to possible IO errors.
2284 * We might rmdir the source, keep it as pwd of some process
2285 * and merrily kill the link to whatever was created under the
2286 * same name. Goodbye sticky bit ;-<
2288 old_inode
= old_dentry
->d_inode
;
2290 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2293 new_inode
= new_dentry
->d_inode
;
2294 new_bh
= ext3_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
2301 if (S_ISDIR(old_inode
->i_mode
)) {
2303 retval
= -ENOTEMPTY
;
2304 if (!empty_dir (new_inode
))
2308 dir_bh
= ext3_bread (handle
, old_inode
, 0, 0, &retval
);
2311 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2314 if (!new_inode
&& new_dir
!=old_dir
&&
2315 new_dir
->i_nlink
>= EXT3_LINK_MAX
)
2319 retval
= ext3_add_entry (handle
, new_dentry
, old_inode
);
2323 BUFFER_TRACE(new_bh
, "get write access");
2324 ext3_journal_get_write_access(handle
, new_bh
);
2325 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2326 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2327 EXT3_FEATURE_INCOMPAT_FILETYPE
))
2328 new_de
->file_type
= old_de
->file_type
;
2329 new_dir
->i_version
++;
2330 new_dir
->i_ctime
= new_dir
->i_mtime
= CURRENT_TIME_SEC
;
2331 ext3_mark_inode_dirty(handle
, new_dir
);
2332 BUFFER_TRACE(new_bh
, "call ext3_journal_dirty_metadata");
2333 ext3_journal_dirty_metadata(handle
, new_bh
);
2339 * Like most other Unix systems, set the ctime for inodes on a
2342 old_inode
->i_ctime
= CURRENT_TIME_SEC
;
2343 ext3_mark_inode_dirty(handle
, old_inode
);
2348 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2349 old_de
->name_len
!= old_dentry
->d_name
.len
||
2350 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2351 (retval
= ext3_delete_entry(handle
, old_dir
,
2352 old_de
, old_bh
)) == -ENOENT
) {
2353 /* old_de could have moved from under us during htree split, so
2354 * make sure that we are deleting the right entry. We might
2355 * also be pointing to a stale entry in the unused part of
2356 * old_bh so just checking inum and the name isn't enough. */
2357 struct buffer_head
*old_bh2
;
2358 struct ext3_dir_entry_2
*old_de2
;
2360 old_bh2
= ext3_find_entry(old_dir
, &old_dentry
->d_name
,
2363 retval
= ext3_delete_entry(handle
, old_dir
,
2369 ext3_warning(old_dir
->i_sb
, "ext3_rename",
2370 "Deleting old file (%lu), %d, error=%d",
2371 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2375 drop_nlink(new_inode
);
2376 new_inode
->i_ctime
= CURRENT_TIME_SEC
;
2378 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME_SEC
;
2379 ext3_update_dx_flag(old_dir
);
2381 BUFFER_TRACE(dir_bh
, "get_write_access");
2382 ext3_journal_get_write_access(handle
, dir_bh
);
2383 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2384 BUFFER_TRACE(dir_bh
, "call ext3_journal_dirty_metadata");
2385 ext3_journal_dirty_metadata(handle
, dir_bh
);
2386 drop_nlink(old_dir
);
2388 drop_nlink(new_inode
);
2391 ext3_update_dx_flag(new_dir
);
2392 ext3_mark_inode_dirty(handle
, new_dir
);
2395 ext3_mark_inode_dirty(handle
, old_dir
);
2397 ext3_mark_inode_dirty(handle
, new_inode
);
2398 if (!new_inode
->i_nlink
)
2399 ext3_orphan_add(handle
, new_inode
);
2407 ext3_journal_stop(handle
);
2412 * directories can handle most operations...
2414 const struct inode_operations ext3_dir_inode_operations
= {
2415 .create
= ext3_create
,
2416 .lookup
= ext3_lookup
,
2418 .unlink
= ext3_unlink
,
2419 .symlink
= ext3_symlink
,
2420 .mkdir
= ext3_mkdir
,
2421 .rmdir
= ext3_rmdir
,
2422 .mknod
= ext3_mknod
,
2423 .rename
= ext3_rename
,
2424 .setattr
= ext3_setattr
,
2425 #ifdef CONFIG_EXT3_FS_XATTR
2426 .setxattr
= generic_setxattr
,
2427 .getxattr
= generic_getxattr
,
2428 .listxattr
= ext3_listxattr
,
2429 .removexattr
= generic_removexattr
,
2431 .permission
= ext3_permission
,
2434 const struct inode_operations ext3_special_inode_operations
= {
2435 .setattr
= ext3_setattr
,
2436 #ifdef CONFIG_EXT3_FS_XATTR
2437 .setxattr
= generic_setxattr
,
2438 .getxattr
= generic_getxattr
,
2439 .listxattr
= ext3_listxattr
,
2440 .removexattr
= generic_removexattr
,
2442 .permission
= ext3_permission
,