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 dxtrace(command) command
80 #define dxtrace(command)
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
111 struct fake_dirent dot
;
113 struct fake_dirent dotdot
;
117 __le32 reserved_zero
;
119 u8 info_length
; /* 8 */
124 struct dx_entry entries
[0];
129 struct fake_dirent fake
;
130 struct dx_entry entries
[0];
136 struct buffer_head
*bh
;
137 struct dx_entry
*entries
;
148 static inline unsigned dx_get_block (struct dx_entry
*entry
);
149 static void dx_set_block (struct dx_entry
*entry
, unsigned value
);
150 static inline unsigned dx_get_hash (struct dx_entry
*entry
);
151 static void dx_set_hash (struct dx_entry
*entry
, unsigned value
);
152 static unsigned dx_get_count (struct dx_entry
*entries
);
153 static unsigned dx_get_limit (struct dx_entry
*entries
);
154 static void dx_set_count (struct dx_entry
*entries
, unsigned value
);
155 static void dx_set_limit (struct dx_entry
*entries
, unsigned value
);
156 static unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
);
157 static unsigned dx_node_limit (struct inode
*dir
);
158 static struct dx_frame
*dx_probe(struct qstr
*entry
,
160 struct dx_hash_info
*hinfo
,
161 struct dx_frame
*frame
,
163 static void dx_release (struct dx_frame
*frames
);
164 static int dx_make_map (struct ext3_dir_entry_2
*de
, int size
,
165 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
166 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
167 static struct ext3_dir_entry_2
*dx_move_dirents (char *from
, char *to
,
168 struct dx_map_entry
*offsets
, int count
);
169 static struct ext3_dir_entry_2
* dx_pack_dirents (char *base
, int size
);
170 static void dx_insert_block (struct dx_frame
*frame
, u32 hash
, u32 block
);
171 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
172 struct dx_frame
*frame
,
173 struct dx_frame
*frames
,
175 static struct buffer_head
* ext3_dx_find_entry(struct inode
*dir
,
176 struct qstr
*entry
, struct ext3_dir_entry_2
**res_dir
,
178 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
179 struct inode
*inode
);
182 * p is at least 6 bytes before the end of page
184 static inline struct ext3_dir_entry_2
*
185 ext3_next_entry(struct ext3_dir_entry_2
*p
)
187 return (struct ext3_dir_entry_2
*)((char *)p
+
188 ext3_rec_len_from_disk(p
->rec_len
));
192 * Future: use high four bits of block for coalesce-on-delete flags
193 * Mask them off for now.
196 static inline unsigned dx_get_block (struct dx_entry
*entry
)
198 return le32_to_cpu(entry
->block
) & 0x00ffffff;
201 static inline void dx_set_block (struct dx_entry
*entry
, unsigned value
)
203 entry
->block
= cpu_to_le32(value
);
206 static inline unsigned dx_get_hash (struct dx_entry
*entry
)
208 return le32_to_cpu(entry
->hash
);
211 static inline void dx_set_hash (struct dx_entry
*entry
, unsigned value
)
213 entry
->hash
= cpu_to_le32(value
);
216 static inline unsigned dx_get_count (struct dx_entry
*entries
)
218 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
221 static inline unsigned dx_get_limit (struct dx_entry
*entries
)
223 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
226 static inline void dx_set_count (struct dx_entry
*entries
, unsigned value
)
228 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
231 static inline void dx_set_limit (struct dx_entry
*entries
, unsigned value
)
233 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
236 static inline unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
)
238 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(1) -
239 EXT3_DIR_REC_LEN(2) - infosize
;
240 return entry_space
/ sizeof(struct dx_entry
);
243 static inline unsigned dx_node_limit (struct inode
*dir
)
245 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(0);
246 return entry_space
/ sizeof(struct dx_entry
);
253 static void dx_show_index (char * label
, struct dx_entry
*entries
)
255 int i
, n
= dx_get_count (entries
);
256 printk("%s index ", label
);
257 for (i
= 0; i
< n
; i
++)
259 printk("%x->%u ", i
? dx_get_hash(entries
+ i
): 0, dx_get_block(entries
+ i
));
271 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext3_dir_entry_2
*de
,
272 int size
, int show_names
)
274 unsigned names
= 0, space
= 0;
275 char *base
= (char *) de
;
276 struct dx_hash_info h
= *hinfo
;
279 while ((char *) de
< base
+ size
)
285 int len
= de
->name_len
;
286 char *name
= de
->name
;
287 while (len
--) printk("%c", *name
++);
288 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
289 printk(":%x.%u ", h
.hash
,
290 ((char *) de
- base
));
292 space
+= EXT3_DIR_REC_LEN(de
->name_len
);
295 de
= ext3_next_entry(de
);
297 printk("(%i)\n", names
);
298 return (struct stats
) { names
, space
, 1 };
301 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
302 struct dx_entry
*entries
, int levels
)
304 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
305 unsigned count
= dx_get_count (entries
), names
= 0, space
= 0, i
;
307 struct buffer_head
*bh
;
309 printk("%i indexed blocks...\n", count
);
310 for (i
= 0; i
< count
; i
++, entries
++)
312 u32 block
= dx_get_block(entries
), hash
= i
? dx_get_hash(entries
): 0;
313 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
315 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
316 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0,&err
))) continue;
318 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
319 dx_show_leaf(hinfo
, (struct ext3_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
320 names
+= stats
.names
;
321 space
+= stats
.space
;
322 bcount
+= stats
.bcount
;
326 printk("%snames %u, fullness %u (%u%%)\n", levels
?"":" ",
327 names
, space
/bcount
,(space
/bcount
)*100/blocksize
);
328 return (struct stats
) { names
, space
, bcount
};
330 #endif /* DX_DEBUG */
333 * Probe for a directory leaf block to search.
335 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
336 * error in the directory index, and the caller should fall back to
337 * searching the directory normally. The callers of dx_probe **MUST**
338 * check for this error code, and make sure it never gets reflected
341 static struct dx_frame
*
342 dx_probe(struct qstr
*entry
, struct inode
*dir
,
343 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
345 unsigned count
, indirect
;
346 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
347 struct dx_root
*root
;
348 struct buffer_head
*bh
;
349 struct dx_frame
*frame
= frame_in
;
353 if (!(bh
= ext3_bread (NULL
,dir
, 0, 0, err
)))
355 root
= (struct dx_root
*) bh
->b_data
;
356 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
357 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
358 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
359 ext3_warning(dir
->i_sb
, __func__
,
360 "Unrecognised inode hash code %d",
361 root
->info
.hash_version
);
363 *err
= ERR_BAD_DX_DIR
;
366 hinfo
->hash_version
= root
->info
.hash_version
;
367 hinfo
->seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
369 ext3fs_dirhash(entry
->name
, entry
->len
, hinfo
);
372 if (root
->info
.unused_flags
& 1) {
373 ext3_warning(dir
->i_sb
, __func__
,
374 "Unimplemented inode hash flags: %#06x",
375 root
->info
.unused_flags
);
377 *err
= ERR_BAD_DX_DIR
;
381 if ((indirect
= root
->info
.indirect_levels
) > 1) {
382 ext3_warning(dir
->i_sb
, __func__
,
383 "Unimplemented inode hash depth: %#06x",
384 root
->info
.indirect_levels
);
386 *err
= ERR_BAD_DX_DIR
;
390 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
391 root
->info
.info_length
);
393 if (dx_get_limit(entries
) != dx_root_limit(dir
,
394 root
->info
.info_length
)) {
395 ext3_warning(dir
->i_sb
, __func__
,
396 "dx entry: limit != root limit");
398 *err
= ERR_BAD_DX_DIR
;
402 dxtrace (printk("Look up %x", hash
));
405 count
= dx_get_count(entries
);
406 if (!count
|| count
> dx_get_limit(entries
)) {
407 ext3_warning(dir
->i_sb
, __func__
,
408 "dx entry: no count or count > limit");
410 *err
= ERR_BAD_DX_DIR
;
415 q
= entries
+ count
- 1;
419 dxtrace(printk("."));
420 if (dx_get_hash(m
) > hash
)
426 if (0) // linear search cross check
428 unsigned n
= count
- 1;
432 dxtrace(printk(","));
433 if (dx_get_hash(++at
) > hash
)
439 assert (at
== p
- 1);
443 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
445 frame
->entries
= entries
;
447 if (!indirect
--) return frame
;
448 if (!(bh
= ext3_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
450 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
451 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
452 ext3_warning(dir
->i_sb
, __func__
,
453 "dx entry: limit != node limit");
455 *err
= ERR_BAD_DX_DIR
;
462 while (frame
>= frame_in
) {
467 if (*err
== ERR_BAD_DX_DIR
)
468 ext3_warning(dir
->i_sb
, __func__
,
469 "Corrupt dir inode %ld, running e2fsck is "
470 "recommended.", dir
->i_ino
);
474 static void dx_release (struct dx_frame
*frames
)
476 if (frames
[0].bh
== NULL
)
479 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
480 brelse(frames
[1].bh
);
481 brelse(frames
[0].bh
);
485 * This function increments the frame pointer to search the next leaf
486 * block, and reads in the necessary intervening nodes if the search
487 * should be necessary. Whether or not the search is necessary is
488 * controlled by the hash parameter. If the hash value is even, then
489 * the search is only continued if the next block starts with that
490 * hash value. This is used if we are searching for a specific file.
492 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
494 * This function returns 1 if the caller should continue to search,
495 * or 0 if it should not. If there is an error reading one of the
496 * index blocks, it will a negative error code.
498 * If start_hash is non-null, it will be filled in with the starting
499 * hash of the next page.
501 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
502 struct dx_frame
*frame
,
503 struct dx_frame
*frames
,
507 struct buffer_head
*bh
;
508 int err
, num_frames
= 0;
513 * Find the next leaf page by incrementing the frame pointer.
514 * If we run out of entries in the interior node, loop around and
515 * increment pointer in the parent node. When we break out of
516 * this loop, num_frames indicates the number of interior
517 * nodes need to be read.
520 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
529 * If the hash is 1, then continue only if the next page has a
530 * continuation hash of any value. This is used for readdir
531 * handling. Otherwise, check to see if the hash matches the
532 * desired contiuation hash. If it doesn't, return since
533 * there's no point to read in the successive index pages.
535 bhash
= dx_get_hash(p
->at
);
538 if ((hash
& 1) == 0) {
539 if ((bhash
& ~1) != hash
)
543 * If the hash is HASH_NB_ALWAYS, we always go to the next
544 * block so no check is necessary
546 while (num_frames
--) {
547 if (!(bh
= ext3_bread(NULL
, dir
, dx_get_block(p
->at
),
549 return err
; /* Failure */
553 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
560 * This function fills a red-black tree with information from a
561 * directory block. It returns the number directory entries loaded
562 * into the tree. If there is an error it is returned in err.
564 static int htree_dirblock_to_tree(struct file
*dir_file
,
565 struct inode
*dir
, int block
,
566 struct dx_hash_info
*hinfo
,
567 __u32 start_hash
, __u32 start_minor_hash
)
569 struct buffer_head
*bh
;
570 struct ext3_dir_entry_2
*de
, *top
;
573 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block
));
574 if (!(bh
= ext3_bread (NULL
, dir
, block
, 0, &err
)))
577 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
578 top
= (struct ext3_dir_entry_2
*) ((char *) de
+
579 dir
->i_sb
->s_blocksize
-
580 EXT3_DIR_REC_LEN(0));
581 for (; de
< top
; de
= ext3_next_entry(de
)) {
582 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
583 (block
<<EXT3_BLOCK_SIZE_BITS(dir
->i_sb
))
584 +((char *)de
- bh
->b_data
))) {
585 /* On error, skip the f_pos to the next block. */
586 dir_file
->f_pos
= (dir_file
->f_pos
|
587 (dir
->i_sb
->s_blocksize
- 1)) + 1;
591 ext3fs_dirhash(de
->name
, de
->name_len
, hinfo
);
592 if ((hinfo
->hash
< start_hash
) ||
593 ((hinfo
->hash
== start_hash
) &&
594 (hinfo
->minor_hash
< start_minor_hash
)))
598 if ((err
= ext3_htree_store_dirent(dir_file
,
599 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
611 * This function fills a red-black tree with information from a
612 * directory. We start scanning the directory in hash order, starting
613 * at start_hash and start_minor_hash.
615 * This function returns the number of entries inserted into the tree,
616 * or a negative error code.
618 int ext3_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
619 __u32 start_minor_hash
, __u32
*next_hash
)
621 struct dx_hash_info hinfo
;
622 struct ext3_dir_entry_2
*de
;
623 struct dx_frame frames
[2], *frame
;
630 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash
,
632 dir
= dir_file
->f_path
.dentry
->d_inode
;
633 if (!(EXT3_I(dir
)->i_flags
& EXT3_INDEX_FL
)) {
634 hinfo
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
635 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
636 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
637 start_hash
, start_minor_hash
);
641 hinfo
.hash
= start_hash
;
642 hinfo
.minor_hash
= 0;
643 frame
= dx_probe(NULL
, dir_file
->f_path
.dentry
->d_inode
, &hinfo
, frames
, &err
);
647 /* Add '.' and '..' from the htree header */
648 if (!start_hash
&& !start_minor_hash
) {
649 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
650 if ((err
= ext3_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
654 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
655 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
656 de
= ext3_next_entry(de
);
657 if ((err
= ext3_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
663 block
= dx_get_block(frame
->at
);
664 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
665 start_hash
, start_minor_hash
);
672 ret
= ext3_htree_next_block(dir
, HASH_NB_ALWAYS
,
673 frame
, frames
, &hashval
);
674 *next_hash
= hashval
;
680 * Stop if: (a) there are no more entries, or
681 * (b) we have inserted at least one entry and the
682 * next hash value is not a continuation
685 (count
&& ((hashval
& 1) == 0)))
689 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
699 * Directory block splitting, compacting
703 * Create map of hash values, offsets, and sizes, stored at end of block.
704 * Returns number of entries mapped.
706 static int dx_make_map (struct ext3_dir_entry_2
*de
, int size
,
707 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
710 char *base
= (char *) de
;
711 struct dx_hash_info h
= *hinfo
;
713 while ((char *) de
< base
+ size
)
715 if (de
->name_len
&& de
->inode
) {
716 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
718 map_tail
->hash
= h
.hash
;
719 map_tail
->offs
= (u16
) ((char *) de
- base
);
720 map_tail
->size
= le16_to_cpu(de
->rec_len
);
724 /* XXX: do we need to check rec_len == 0 case? -Chris */
725 de
= ext3_next_entry(de
);
730 /* Sort map by hash value */
731 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
733 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
735 /* Combsort until bubble sort doesn't suck */
739 if (count
- 9 < 2) /* 9, 10 -> 11 */
741 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
742 if (p
->hash
< q
->hash
)
745 /* Garden variety bubble sort */
751 if (q
[1].hash
>= q
[0].hash
)
759 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, u32 block
)
761 struct dx_entry
*entries
= frame
->entries
;
762 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
763 int count
= dx_get_count(entries
);
765 assert(count
< dx_get_limit(entries
));
766 assert(old
< entries
+ count
);
767 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
768 dx_set_hash(new, hash
);
769 dx_set_block(new, block
);
770 dx_set_count(entries
, count
+ 1);
773 static void ext3_update_dx_flag(struct inode
*inode
)
775 if (!EXT3_HAS_COMPAT_FEATURE(inode
->i_sb
,
776 EXT3_FEATURE_COMPAT_DIR_INDEX
))
777 EXT3_I(inode
)->i_flags
&= ~EXT3_INDEX_FL
;
781 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
783 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
784 * `de != NULL' is guaranteed by caller.
786 static inline int ext3_match (int len
, const char * const name
,
787 struct ext3_dir_entry_2
* de
)
789 if (len
!= de
->name_len
)
793 return !memcmp(name
, de
->name
, len
);
797 * Returns 0 if not found, -1 on failure, and 1 on success
799 static inline int search_dirblock(struct buffer_head
* bh
,
802 unsigned long offset
,
803 struct ext3_dir_entry_2
** res_dir
)
805 struct ext3_dir_entry_2
* de
;
808 const char *name
= child
->name
;
809 int namelen
= child
->len
;
811 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
812 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
813 while ((char *) de
< dlimit
) {
814 /* this code is executed quadratically often */
815 /* do minimal checking `by hand' */
817 if ((char *) de
+ namelen
<= dlimit
&&
818 ext3_match (namelen
, name
, de
)) {
819 /* found a match - just to be sure, do a full check */
820 if (!ext3_check_dir_entry("ext3_find_entry",
821 dir
, de
, bh
, offset
))
826 /* prevent looping on a bad block */
827 de_len
= ext3_rec_len_from_disk(de
->rec_len
);
831 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ de_len
);
840 * finds an entry in the specified directory with the wanted name. It
841 * returns the cache buffer in which the entry was found, and the entry
842 * itself (as a parameter - res_dir). It does NOT read the inode of the
843 * entry - you'll have to do that yourself if you want to.
845 * The returned buffer_head has ->b_count elevated. The caller is expected
846 * to brelse() it when appropriate.
848 static struct buffer_head
*ext3_find_entry(struct inode
*dir
,
850 struct ext3_dir_entry_2
**res_dir
)
852 struct super_block
* sb
;
853 struct buffer_head
* bh_use
[NAMEI_RA_SIZE
];
854 struct buffer_head
* bh
, *ret
= NULL
;
855 unsigned long start
, block
, b
;
856 int ra_max
= 0; /* Number of bh's in the readahead
858 int ra_ptr
= 0; /* Current index into readahead
866 namelen
= entry
->len
;
867 if (namelen
> EXT3_NAME_LEN
)
870 bh
= ext3_dx_find_entry(dir
, entry
, res_dir
, &err
);
872 * On success, or if the error was file not found,
873 * return. Otherwise, fall back to doing a search the
876 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
878 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
880 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
881 start
= EXT3_I(dir
)->i_dir_start_lookup
;
882 if (start
>= nblocks
)
888 * We deal with the read-ahead logic here.
890 if (ra_ptr
>= ra_max
) {
891 /* Refill the readahead buffer */
894 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
896 * Terminate if we reach the end of the
897 * directory and must wrap, or if our
898 * search has finished at this block.
900 if (b
>= nblocks
|| (num
&& block
== start
)) {
901 bh_use
[ra_max
] = NULL
;
905 bh
= ext3_getblk(NULL
, dir
, b
++, 0, &err
);
908 ll_rw_block(READ_META
, 1, &bh
);
911 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
914 if (!buffer_uptodate(bh
)) {
915 /* read error, skip block & hope for the best */
916 ext3_error(sb
, __func__
, "reading directory #%lu "
917 "offset %lu", dir
->i_ino
, block
);
921 i
= search_dirblock(bh
, dir
, entry
,
922 block
<< EXT3_BLOCK_SIZE_BITS(sb
), res_dir
);
924 EXT3_I(dir
)->i_dir_start_lookup
= block
;
926 goto cleanup_and_exit
;
930 goto cleanup_and_exit
;
933 if (++block
>= nblocks
)
935 } while (block
!= start
);
938 * If the directory has grown while we were searching, then
939 * search the last part of the directory before giving up.
942 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
943 if (block
< nblocks
) {
949 /* Clean up the read-ahead blocks */
950 for (; ra_ptr
< ra_max
; ra_ptr
++)
951 brelse (bh_use
[ra_ptr
]);
955 static struct buffer_head
* ext3_dx_find_entry(struct inode
*dir
,
956 struct qstr
*entry
, struct ext3_dir_entry_2
**res_dir
,
959 struct super_block
* sb
;
960 struct dx_hash_info hinfo
;
962 struct dx_frame frames
[2], *frame
;
963 struct ext3_dir_entry_2
*de
, *top
;
964 struct buffer_head
*bh
;
967 int namelen
= entry
->len
;
968 const u8
*name
= entry
->name
;
971 /* NFS may look up ".." - look at dx_root directory block */
972 if (namelen
> 2 || name
[0] != '.'|| (namelen
== 2 && name
[1] != '.')) {
973 if (!(frame
= dx_probe(entry
, dir
, &hinfo
, frames
, err
)))
977 frame
->bh
= NULL
; /* for dx_release() */
978 frame
->at
= (struct dx_entry
*)frames
; /* hack for zero entry*/
979 dx_set_block(frame
->at
, 0); /* dx_root block is 0 */
983 block
= dx_get_block(frame
->at
);
984 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0, err
)))
986 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
987 top
= (struct ext3_dir_entry_2
*) ((char *) de
+ sb
->s_blocksize
-
988 EXT3_DIR_REC_LEN(0));
989 for (; de
< top
; de
= ext3_next_entry(de
)) {
990 int off
= (block
<< EXT3_BLOCK_SIZE_BITS(sb
))
991 + ((char *) de
- bh
->b_data
);
993 if (!ext3_check_dir_entry(__func__
, dir
, de
, bh
, off
)) {
995 *err
= ERR_BAD_DX_DIR
;
999 if (ext3_match(namelen
, name
, de
)) {
1006 /* Check to see if we should continue to search */
1007 retval
= ext3_htree_next_block(dir
, hash
, frame
,
1010 ext3_warning(sb
, __func__
,
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(dir
, &dentry
->d_name
, &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
);
1042 return ERR_PTR(-EIO
);
1044 inode
= ext3_iget(dir
->i_sb
, ino
);
1046 return ERR_CAST(inode
);
1048 return d_splice_alias(inode
, dentry
);
1052 struct dentry
*ext3_get_parent(struct dentry
*child
)
1055 struct qstr dotdot
= {.name
= "..", .len
= 2};
1056 struct ext3_dir_entry_2
* de
;
1057 struct buffer_head
*bh
;
1059 bh
= ext3_find_entry(child
->d_inode
, &dotdot
, &de
);
1061 return ERR_PTR(-ENOENT
);
1062 ino
= le32_to_cpu(de
->inode
);
1065 if (!ext3_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1066 ext3_error(child
->d_inode
->i_sb
, "ext3_get_parent",
1067 "bad inode number: %lu", ino
);
1068 return ERR_PTR(-EIO
);
1071 return d_obtain_alias(ext3_iget(child
->d_inode
->i_sb
, ino
));
1075 static unsigned char ext3_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1076 [S_IFREG
>> S_SHIFT
] = EXT3_FT_REG_FILE
,
1077 [S_IFDIR
>> S_SHIFT
] = EXT3_FT_DIR
,
1078 [S_IFCHR
>> S_SHIFT
] = EXT3_FT_CHRDEV
,
1079 [S_IFBLK
>> S_SHIFT
] = EXT3_FT_BLKDEV
,
1080 [S_IFIFO
>> S_SHIFT
] = EXT3_FT_FIFO
,
1081 [S_IFSOCK
>> S_SHIFT
] = EXT3_FT_SOCK
,
1082 [S_IFLNK
>> S_SHIFT
] = EXT3_FT_SYMLINK
,
1085 static inline void ext3_set_de_type(struct super_block
*sb
,
1086 struct ext3_dir_entry_2
*de
,
1088 if (EXT3_HAS_INCOMPAT_FEATURE(sb
, EXT3_FEATURE_INCOMPAT_FILETYPE
))
1089 de
->file_type
= ext3_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1093 * Move count entries from end of map between two memory locations.
1094 * Returns pointer to last entry moved.
1096 static struct ext3_dir_entry_2
*
1097 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1099 unsigned rec_len
= 0;
1102 struct ext3_dir_entry_2
*de
= (struct ext3_dir_entry_2
*) (from
+ map
->offs
);
1103 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1104 memcpy (to
, de
, rec_len
);
1105 ((struct ext3_dir_entry_2
*) to
)->rec_len
=
1106 ext3_rec_len_to_disk(rec_len
);
1111 return (struct ext3_dir_entry_2
*) (to
- rec_len
);
1115 * Compact each dir entry in the range to the minimal rec_len.
1116 * Returns pointer to last entry in range.
1118 static struct ext3_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1120 struct ext3_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext3_dir_entry_2
*) base
;
1121 unsigned rec_len
= 0;
1124 while ((char*)de
< base
+ size
) {
1125 next
= ext3_next_entry(de
);
1126 if (de
->inode
&& de
->name_len
) {
1127 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1129 memmove(to
, de
, rec_len
);
1130 to
->rec_len
= ext3_rec_len_to_disk(rec_len
);
1132 to
= (struct ext3_dir_entry_2
*) (((char *) to
) + rec_len
);
1140 * Split a full leaf block to make room for a new dir entry.
1141 * Allocate a new block, and move entries so that they are approx. equally full.
1142 * Returns pointer to de in block into which the new entry will be inserted.
1144 static struct ext3_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1145 struct buffer_head
**bh
,struct dx_frame
*frame
,
1146 struct dx_hash_info
*hinfo
, int *error
)
1148 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1149 unsigned count
, continued
;
1150 struct buffer_head
*bh2
;
1153 struct dx_map_entry
*map
;
1154 char *data1
= (*bh
)->b_data
, *data2
;
1155 unsigned split
, move
, size
, i
;
1156 struct ext3_dir_entry_2
*de
= NULL
, *de2
;
1159 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1166 BUFFER_TRACE(*bh
, "get_write_access");
1167 err
= ext3_journal_get_write_access(handle
, *bh
);
1171 BUFFER_TRACE(frame
->bh
, "get_write_access");
1172 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1176 data2
= bh2
->b_data
;
1178 /* create map in the end of data2 block */
1179 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1180 count
= dx_make_map ((struct ext3_dir_entry_2
*) data1
,
1181 blocksize
, hinfo
, map
);
1183 dx_sort_map (map
, count
);
1184 /* Split the existing block in the middle, size-wise */
1187 for (i
= count
-1; i
>= 0; i
--) {
1188 /* is more than half of this entry in 2nd half of the block? */
1189 if (size
+ map
[i
].size
/2 > blocksize
/2)
1191 size
+= map
[i
].size
;
1194 /* map index at which we will split */
1195 split
= count
- move
;
1196 hash2
= map
[split
].hash
;
1197 continued
= hash2
== map
[split
- 1].hash
;
1198 dxtrace(printk("Split block %i at %x, %i/%i\n",
1199 dx_get_block(frame
->at
), hash2
, split
, count
-split
));
1201 /* Fancy dance to stay within two buffers */
1202 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1203 de
= dx_pack_dirents(data1
,blocksize
);
1204 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1205 de2
->rec_len
= ext3_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1206 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data1
, blocksize
, 1));
1207 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data2
, blocksize
, 1));
1209 /* Which block gets the new entry? */
1210 if (hinfo
->hash
>= hash2
)
1215 dx_insert_block (frame
, hash2
+ continued
, newblock
);
1216 err
= ext3_journal_dirty_metadata (handle
, bh2
);
1219 err
= ext3_journal_dirty_metadata (handle
, frame
->bh
);
1223 dxtrace(dx_show_index ("frame", frame
->entries
));
1230 ext3_std_error(dir
->i_sb
, err
);
1238 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1239 * it points to a directory entry which is guaranteed to be large
1240 * enough for new directory entry. If de is NULL, then
1241 * add_dirent_to_buf will attempt search the directory block for
1242 * space. It will return -ENOSPC if no space is available, and -EIO
1243 * and -EEXIST if directory entry already exists.
1245 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1246 * all other cases bh is released.
1248 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1249 struct inode
*inode
, struct ext3_dir_entry_2
*de
,
1250 struct buffer_head
* bh
)
1252 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1253 const char *name
= dentry
->d_name
.name
;
1254 int namelen
= dentry
->d_name
.len
;
1255 unsigned long offset
= 0;
1256 unsigned short reclen
;
1257 int nlen
, rlen
, err
;
1260 reclen
= EXT3_DIR_REC_LEN(namelen
);
1262 de
= (struct ext3_dir_entry_2
*)bh
->b_data
;
1263 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1264 while ((char *) de
<= top
) {
1265 if (!ext3_check_dir_entry("ext3_add_entry", dir
, de
,
1270 if (ext3_match (namelen
, name
, de
)) {
1274 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1275 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1276 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1278 de
= (struct ext3_dir_entry_2
*)((char *)de
+ rlen
);
1281 if ((char *) de
> top
)
1284 BUFFER_TRACE(bh
, "get_write_access");
1285 err
= ext3_journal_get_write_access(handle
, bh
);
1287 ext3_std_error(dir
->i_sb
, err
);
1292 /* By now the buffer is marked for journaling */
1293 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1294 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1296 struct ext3_dir_entry_2
*de1
= (struct ext3_dir_entry_2
*)((char *)de
+ nlen
);
1297 de1
->rec_len
= ext3_rec_len_to_disk(rlen
- nlen
);
1298 de
->rec_len
= ext3_rec_len_to_disk(nlen
);
1301 de
->file_type
= EXT3_FT_UNKNOWN
;
1303 de
->inode
= cpu_to_le32(inode
->i_ino
);
1304 ext3_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1307 de
->name_len
= namelen
;
1308 memcpy (de
->name
, name
, namelen
);
1310 * XXX shouldn't update any times until successful
1311 * completion of syscall, but too many callers depend
1314 * XXX similarly, too many callers depend on
1315 * ext3_new_inode() setting the times, but error
1316 * recovery deletes the inode, so the worst that can
1317 * happen is that the times are slightly out of date
1318 * and/or different from the directory change time.
1320 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME_SEC
;
1321 ext3_update_dx_flag(dir
);
1323 ext3_mark_inode_dirty(handle
, dir
);
1324 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1325 err
= ext3_journal_dirty_metadata(handle
, bh
);
1327 ext3_std_error(dir
->i_sb
, err
);
1333 * This converts a one block unindexed directory to a 3 block indexed
1334 * directory, and adds the dentry to the indexed directory.
1336 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1337 struct inode
*inode
, struct buffer_head
*bh
)
1339 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1340 const char *name
= dentry
->d_name
.name
;
1341 int namelen
= dentry
->d_name
.len
;
1342 struct buffer_head
*bh2
;
1343 struct dx_root
*root
;
1344 struct dx_frame frames
[2], *frame
;
1345 struct dx_entry
*entries
;
1346 struct ext3_dir_entry_2
*de
, *de2
;
1351 struct dx_hash_info hinfo
;
1353 struct fake_dirent
*fde
;
1355 blocksize
= dir
->i_sb
->s_blocksize
;
1356 dxtrace(printk("Creating index\n"));
1357 retval
= ext3_journal_get_write_access(handle
, bh
);
1359 ext3_std_error(dir
->i_sb
, retval
);
1363 root
= (struct dx_root
*) bh
->b_data
;
1365 bh2
= ext3_append (handle
, dir
, &block
, &retval
);
1370 EXT3_I(dir
)->i_flags
|= EXT3_INDEX_FL
;
1371 data1
= bh2
->b_data
;
1373 /* The 0th block becomes the root, move the dirents out */
1374 fde
= &root
->dotdot
;
1375 de
= (struct ext3_dir_entry_2
*)((char *)fde
+
1376 ext3_rec_len_from_disk(fde
->rec_len
));
1377 len
= ((char *) root
) + blocksize
- (char *) de
;
1378 memcpy (data1
, de
, len
);
1379 de
= (struct ext3_dir_entry_2
*) data1
;
1381 while ((char *)(de2
= ext3_next_entry(de
)) < top
)
1383 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1384 /* Initialize the root; the dot dirents already exist */
1385 de
= (struct ext3_dir_entry_2
*) (&root
->dotdot
);
1386 de
->rec_len
= ext3_rec_len_to_disk(blocksize
- EXT3_DIR_REC_LEN(2));
1387 memset (&root
->info
, 0, sizeof(root
->info
));
1388 root
->info
.info_length
= sizeof(root
->info
);
1389 root
->info
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
1390 entries
= root
->entries
;
1391 dx_set_block (entries
, 1);
1392 dx_set_count (entries
, 1);
1393 dx_set_limit (entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1395 /* Initialize as for dx_probe */
1396 hinfo
.hash_version
= root
->info
.hash_version
;
1397 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
1398 ext3fs_dirhash(name
, namelen
, &hinfo
);
1400 frame
->entries
= entries
;
1401 frame
->at
= entries
;
1404 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1405 dx_release (frames
);
1409 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1415 * adds a file entry to the specified directory, using the same
1416 * semantics as ext3_find_entry(). It returns NULL if it failed.
1418 * NOTE!! The inode part of 'de' is left at 0 - which means you
1419 * may not sleep between calling this and putting something into
1420 * the entry, as someone else might have used it while you slept.
1422 static int ext3_add_entry (handle_t
*handle
, struct dentry
*dentry
,
1423 struct inode
*inode
)
1425 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1426 unsigned long offset
;
1427 struct buffer_head
* bh
;
1428 struct ext3_dir_entry_2
*de
;
1429 struct super_block
* sb
;
1436 blocksize
= sb
->s_blocksize
;
1437 if (!dentry
->d_name
.len
)
1440 retval
= ext3_dx_add_entry(handle
, dentry
, inode
);
1441 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1443 EXT3_I(dir
)->i_flags
&= ~EXT3_INDEX_FL
;
1445 ext3_mark_inode_dirty(handle
, dir
);
1447 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1448 for (block
= 0, offset
= 0; block
< blocks
; block
++) {
1449 bh
= ext3_bread(handle
, dir
, block
, 0, &retval
);
1452 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1453 if (retval
!= -ENOSPC
)
1456 if (blocks
== 1 && !dx_fallback
&&
1457 EXT3_HAS_COMPAT_FEATURE(sb
, EXT3_FEATURE_COMPAT_DIR_INDEX
))
1458 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1461 bh
= ext3_append(handle
, dir
, &block
, &retval
);
1464 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1466 de
->rec_len
= ext3_rec_len_to_disk(blocksize
);
1467 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1471 * Returns 0 for success, or a negative error value
1473 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1474 struct inode
*inode
)
1476 struct dx_frame frames
[2], *frame
;
1477 struct dx_entry
*entries
, *at
;
1478 struct dx_hash_info hinfo
;
1479 struct buffer_head
* bh
;
1480 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1481 struct super_block
* sb
= dir
->i_sb
;
1482 struct ext3_dir_entry_2
*de
;
1485 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1488 entries
= frame
->entries
;
1491 if (!(bh
= ext3_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1494 BUFFER_TRACE(bh
, "get_write_access");
1495 err
= ext3_journal_get_write_access(handle
, bh
);
1499 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1500 if (err
!= -ENOSPC
) {
1505 /* Block full, should compress but for now just split */
1506 dxtrace(printk("using %u of %u node entries\n",
1507 dx_get_count(entries
), dx_get_limit(entries
)));
1508 /* Need to split index? */
1509 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1511 unsigned icount
= dx_get_count(entries
);
1512 int levels
= frame
- frames
;
1513 struct dx_entry
*entries2
;
1514 struct dx_node
*node2
;
1515 struct buffer_head
*bh2
;
1517 if (levels
&& (dx_get_count(frames
->entries
) ==
1518 dx_get_limit(frames
->entries
))) {
1519 ext3_warning(sb
, __func__
,
1520 "Directory index full!");
1524 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1527 node2
= (struct dx_node
*)(bh2
->b_data
);
1528 entries2
= node2
->entries
;
1529 node2
->fake
.rec_len
= ext3_rec_len_to_disk(sb
->s_blocksize
);
1530 node2
->fake
.inode
= 0;
1531 BUFFER_TRACE(frame
->bh
, "get_write_access");
1532 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1536 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1537 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1538 dxtrace(printk("Split index %i/%i\n", icount1
, icount2
));
1540 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1541 err
= ext3_journal_get_write_access(handle
,
1546 memcpy ((char *) entries2
, (char *) (entries
+ icount1
),
1547 icount2
* sizeof(struct dx_entry
));
1548 dx_set_count (entries
, icount1
);
1549 dx_set_count (entries2
, icount2
);
1550 dx_set_limit (entries2
, dx_node_limit(dir
));
1552 /* Which index block gets the new entry? */
1553 if (at
- entries
>= icount1
) {
1554 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1555 frame
->entries
= entries
= entries2
;
1556 swap(frame
->bh
, bh2
);
1558 dx_insert_block (frames
+ 0, hash2
, newblock
);
1559 dxtrace(dx_show_index ("node", frames
[1].entries
));
1560 dxtrace(dx_show_index ("node",
1561 ((struct dx_node
*) bh2
->b_data
)->entries
));
1562 err
= ext3_journal_dirty_metadata(handle
, bh2
);
1567 dxtrace(printk("Creating second level index...\n"));
1568 memcpy((char *) entries2
, (char *) entries
,
1569 icount
* sizeof(struct dx_entry
));
1570 dx_set_limit(entries2
, dx_node_limit(dir
));
1573 dx_set_count(entries
, 1);
1574 dx_set_block(entries
+ 0, newblock
);
1575 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1577 /* Add new access path frame */
1579 frame
->at
= at
= at
- entries
+ entries2
;
1580 frame
->entries
= entries
= entries2
;
1582 err
= ext3_journal_get_write_access(handle
,
1587 ext3_journal_dirty_metadata(handle
, frames
[0].bh
);
1589 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1592 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1597 ext3_std_error(dir
->i_sb
, err
);
1606 * ext3_delete_entry deletes a directory entry by merging it with the
1609 static int ext3_delete_entry (handle_t
*handle
,
1611 struct ext3_dir_entry_2
* de_del
,
1612 struct buffer_head
* bh
)
1614 struct ext3_dir_entry_2
* de
, * pde
;
1619 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1620 while (i
< bh
->b_size
) {
1621 if (!ext3_check_dir_entry("ext3_delete_entry", dir
, de
, bh
, i
))
1624 BUFFER_TRACE(bh
, "get_write_access");
1625 ext3_journal_get_write_access(handle
, bh
);
1627 pde
->rec_len
= ext3_rec_len_to_disk(
1628 ext3_rec_len_from_disk(pde
->rec_len
) +
1629 ext3_rec_len_from_disk(de
->rec_len
));
1633 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1634 ext3_journal_dirty_metadata(handle
, bh
);
1637 i
+= ext3_rec_len_from_disk(de
->rec_len
);
1639 de
= ext3_next_entry(de
);
1644 static int ext3_add_nondir(handle_t
*handle
,
1645 struct dentry
*dentry
, struct inode
*inode
)
1647 int err
= ext3_add_entry(handle
, dentry
, inode
);
1649 ext3_mark_inode_dirty(handle
, inode
);
1650 d_instantiate(dentry
, inode
);
1651 unlock_new_inode(inode
);
1655 unlock_new_inode(inode
);
1661 * By the time this is called, we already have created
1662 * the directory cache entry for the new file, but it
1663 * is so far negative - it has no inode.
1665 * If the create succeeds, we fill in the inode information
1666 * with d_instantiate().
1668 static int ext3_create (struct inode
* dir
, struct dentry
* dentry
, int mode
,
1669 struct nameidata
*nd
)
1672 struct inode
* inode
;
1673 int err
, retries
= 0;
1676 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1677 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1678 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1680 return PTR_ERR(handle
);
1682 if (IS_DIRSYNC(dir
))
1685 inode
= ext3_new_inode (handle
, dir
, mode
);
1686 err
= PTR_ERR(inode
);
1687 if (!IS_ERR(inode
)) {
1688 inode
->i_op
= &ext3_file_inode_operations
;
1689 inode
->i_fop
= &ext3_file_operations
;
1690 ext3_set_aops(inode
);
1691 err
= ext3_add_nondir(handle
, dentry
, inode
);
1693 ext3_journal_stop(handle
);
1694 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1699 static int ext3_mknod (struct inode
* dir
, struct dentry
*dentry
,
1700 int mode
, dev_t rdev
)
1703 struct inode
*inode
;
1704 int err
, retries
= 0;
1706 if (!new_valid_dev(rdev
))
1710 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1711 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1712 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1714 return PTR_ERR(handle
);
1716 if (IS_DIRSYNC(dir
))
1719 inode
= ext3_new_inode (handle
, dir
, mode
);
1720 err
= PTR_ERR(inode
);
1721 if (!IS_ERR(inode
)) {
1722 init_special_inode(inode
, inode
->i_mode
, rdev
);
1723 #ifdef CONFIG_EXT3_FS_XATTR
1724 inode
->i_op
= &ext3_special_inode_operations
;
1726 err
= ext3_add_nondir(handle
, dentry
, inode
);
1728 ext3_journal_stop(handle
);
1729 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1734 static int ext3_mkdir(struct inode
* dir
, struct dentry
* dentry
, int mode
)
1737 struct inode
* inode
;
1738 struct buffer_head
* dir_block
;
1739 struct ext3_dir_entry_2
* de
;
1740 int err
, retries
= 0;
1742 if (dir
->i_nlink
>= EXT3_LINK_MAX
)
1746 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1747 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1748 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1750 return PTR_ERR(handle
);
1752 if (IS_DIRSYNC(dir
))
1755 inode
= ext3_new_inode (handle
, dir
, S_IFDIR
| mode
);
1756 err
= PTR_ERR(inode
);
1760 inode
->i_op
= &ext3_dir_inode_operations
;
1761 inode
->i_fop
= &ext3_dir_operations
;
1762 inode
->i_size
= EXT3_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1763 dir_block
= ext3_bread (handle
, inode
, 0, 1, &err
);
1765 drop_nlink(inode
); /* is this nlink == 0? */
1766 unlock_new_inode(inode
);
1767 ext3_mark_inode_dirty(handle
, inode
);
1771 BUFFER_TRACE(dir_block
, "get_write_access");
1772 ext3_journal_get_write_access(handle
, dir_block
);
1773 de
= (struct ext3_dir_entry_2
*) dir_block
->b_data
;
1774 de
->inode
= cpu_to_le32(inode
->i_ino
);
1776 de
->rec_len
= ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de
->name_len
));
1777 strcpy (de
->name
, ".");
1778 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1779 de
= ext3_next_entry(de
);
1780 de
->inode
= cpu_to_le32(dir
->i_ino
);
1781 de
->rec_len
= ext3_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1782 EXT3_DIR_REC_LEN(1));
1784 strcpy (de
->name
, "..");
1785 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1787 BUFFER_TRACE(dir_block
, "call ext3_journal_dirty_metadata");
1788 ext3_journal_dirty_metadata(handle
, dir_block
);
1790 ext3_mark_inode_dirty(handle
, inode
);
1791 err
= ext3_add_entry (handle
, dentry
, inode
);
1794 unlock_new_inode(inode
);
1795 ext3_mark_inode_dirty(handle
, inode
);
1800 ext3_update_dx_flag(dir
);
1801 ext3_mark_inode_dirty(handle
, dir
);
1802 d_instantiate(dentry
, inode
);
1803 unlock_new_inode(inode
);
1805 ext3_journal_stop(handle
);
1806 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1812 * routine to check that the specified directory is empty (for rmdir)
1814 static int empty_dir (struct inode
* inode
)
1816 unsigned long offset
;
1817 struct buffer_head
* bh
;
1818 struct ext3_dir_entry_2
* de
, * de1
;
1819 struct super_block
* sb
;
1823 if (inode
->i_size
< EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1824 !(bh
= ext3_bread (NULL
, inode
, 0, 0, &err
))) {
1826 ext3_error(inode
->i_sb
, __func__
,
1827 "error %d reading directory #%lu offset 0",
1830 ext3_warning(inode
->i_sb
, __func__
,
1831 "bad directory (dir #%lu) - no data block",
1835 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1836 de1
= ext3_next_entry(de
);
1837 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1838 !le32_to_cpu(de1
->inode
) ||
1839 strcmp (".", de
->name
) ||
1840 strcmp ("..", de1
->name
)) {
1841 ext3_warning (inode
->i_sb
, "empty_dir",
1842 "bad directory (dir #%lu) - no `.' or `..'",
1847 offset
= ext3_rec_len_from_disk(de
->rec_len
) +
1848 ext3_rec_len_from_disk(de1
->rec_len
);
1849 de
= ext3_next_entry(de1
);
1850 while (offset
< inode
->i_size
) {
1852 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1855 bh
= ext3_bread (NULL
, inode
,
1856 offset
>> EXT3_BLOCK_SIZE_BITS(sb
), 0, &err
);
1859 ext3_error(sb
, __func__
,
1860 "error %d reading directory"
1862 err
, inode
->i_ino
, offset
);
1863 offset
+= sb
->s_blocksize
;
1866 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1868 if (!ext3_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1869 de
= (struct ext3_dir_entry_2
*)(bh
->b_data
+
1871 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1874 if (le32_to_cpu(de
->inode
)) {
1878 offset
+= ext3_rec_len_from_disk(de
->rec_len
);
1879 de
= ext3_next_entry(de
);
1885 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1886 * such inodes, starting at the superblock, in case we crash before the
1887 * file is closed/deleted, or in case the inode truncate spans multiple
1888 * transactions and the last transaction is not recovered after a crash.
1890 * At filesystem recovery time, we walk this list deleting unlinked
1891 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1893 int ext3_orphan_add(handle_t
*handle
, struct inode
*inode
)
1895 struct super_block
*sb
= inode
->i_sb
;
1896 struct ext3_iloc iloc
;
1900 if (!list_empty(&EXT3_I(inode
)->i_orphan
))
1903 /* Orphan handling is only valid for files with data blocks
1904 * being truncated, or files being unlinked. */
1906 /* @@@ FIXME: Observation from aviro:
1907 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1908 * here (on lock_super()), so race with ext3_link() which might bump
1909 * ->i_nlink. For, say it, character device. Not a regular file,
1910 * not a directory, not a symlink and ->i_nlink > 0.
1912 J_ASSERT ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1913 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1915 BUFFER_TRACE(EXT3_SB(sb
)->s_sbh
, "get_write_access");
1916 err
= ext3_journal_get_write_access(handle
, EXT3_SB(sb
)->s_sbh
);
1920 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1924 /* Insert this inode at the head of the on-disk orphan list... */
1925 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT3_SB(sb
)->s_es
->s_last_orphan
);
1926 EXT3_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1927 err
= ext3_journal_dirty_metadata(handle
, EXT3_SB(sb
)->s_sbh
);
1928 rc
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
1932 /* Only add to the head of the in-memory list if all the
1933 * previous operations succeeded. If the orphan_add is going to
1934 * fail (possibly taking the journal offline), we can't risk
1935 * leaving the inode on the orphan list: stray orphan-list
1936 * entries can cause panics at unmount time.
1938 * This is safe: on error we're going to ignore the orphan list
1939 * anyway on the next recovery. */
1941 list_add(&EXT3_I(inode
)->i_orphan
, &EXT3_SB(sb
)->s_orphan
);
1943 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1944 jbd_debug(4, "orphan inode %lu will point to %d\n",
1945 inode
->i_ino
, NEXT_ORPHAN(inode
));
1948 ext3_std_error(inode
->i_sb
, err
);
1953 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1954 * of such inodes stored on disk, because it is finally being cleaned up.
1956 int ext3_orphan_del(handle_t
*handle
, struct inode
*inode
)
1958 struct list_head
*prev
;
1959 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1960 struct ext3_sb_info
*sbi
;
1961 unsigned long ino_next
;
1962 struct ext3_iloc iloc
;
1965 lock_super(inode
->i_sb
);
1966 if (list_empty(&ei
->i_orphan
)) {
1967 unlock_super(inode
->i_sb
);
1971 ino_next
= NEXT_ORPHAN(inode
);
1972 prev
= ei
->i_orphan
.prev
;
1973 sbi
= EXT3_SB(inode
->i_sb
);
1975 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
1977 list_del_init(&ei
->i_orphan
);
1979 /* If we're on an error path, we may not have a valid
1980 * transaction handle with which to update the orphan list on
1981 * disk, but we still need to remove the inode from the linked
1982 * list in memory. */
1986 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1990 if (prev
== &sbi
->s_orphan
) {
1991 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
1992 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1993 err
= ext3_journal_get_write_access(handle
, sbi
->s_sbh
);
1996 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
1997 err
= ext3_journal_dirty_metadata(handle
, sbi
->s_sbh
);
1999 struct ext3_iloc iloc2
;
2000 struct inode
*i_prev
=
2001 &list_entry(prev
, struct ext3_inode_info
, i_orphan
)->vfs_inode
;
2003 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2004 i_prev
->i_ino
, ino_next
);
2005 err
= ext3_reserve_inode_write(handle
, i_prev
, &iloc2
);
2008 NEXT_ORPHAN(i_prev
) = ino_next
;
2009 err
= ext3_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2013 NEXT_ORPHAN(inode
) = 0;
2014 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
2017 ext3_std_error(inode
->i_sb
, err
);
2019 unlock_super(inode
->i_sb
);
2027 static int ext3_rmdir (struct inode
* dir
, struct dentry
*dentry
)
2030 struct inode
* inode
;
2031 struct buffer_head
* bh
;
2032 struct ext3_dir_entry_2
* de
;
2035 /* Initialize quotas before so that eventual writes go in
2036 * separate transaction */
2037 DQUOT_INIT(dentry
->d_inode
);
2038 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2040 return PTR_ERR(handle
);
2043 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2047 if (IS_DIRSYNC(dir
))
2050 inode
= dentry
->d_inode
;
2053 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2056 retval
= -ENOTEMPTY
;
2057 if (!empty_dir (inode
))
2060 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2063 if (inode
->i_nlink
!= 2)
2064 ext3_warning (inode
->i_sb
, "ext3_rmdir",
2065 "empty directory has nlink!=2 (%d)",
2069 /* There's no need to set i_disksize: the fact that i_nlink is
2070 * zero will ensure that the right thing happens during any
2073 ext3_orphan_add(handle
, inode
);
2074 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2075 ext3_mark_inode_dirty(handle
, inode
);
2077 ext3_update_dx_flag(dir
);
2078 ext3_mark_inode_dirty(handle
, dir
);
2081 ext3_journal_stop(handle
);
2086 static int ext3_unlink(struct inode
* dir
, struct dentry
*dentry
)
2089 struct inode
* inode
;
2090 struct buffer_head
* bh
;
2091 struct ext3_dir_entry_2
* de
;
2094 /* Initialize quotas before so that eventual writes go
2095 * in separate transaction */
2096 DQUOT_INIT(dentry
->d_inode
);
2097 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2099 return PTR_ERR(handle
);
2101 if (IS_DIRSYNC(dir
))
2105 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2109 inode
= dentry
->d_inode
;
2112 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2115 if (!inode
->i_nlink
) {
2116 ext3_warning (inode
->i_sb
, "ext3_unlink",
2117 "Deleting nonexistent file (%lu), %d",
2118 inode
->i_ino
, inode
->i_nlink
);
2121 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2124 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2125 ext3_update_dx_flag(dir
);
2126 ext3_mark_inode_dirty(handle
, dir
);
2128 if (!inode
->i_nlink
)
2129 ext3_orphan_add(handle
, inode
);
2130 inode
->i_ctime
= dir
->i_ctime
;
2131 ext3_mark_inode_dirty(handle
, inode
);
2135 ext3_journal_stop(handle
);
2140 static int ext3_symlink (struct inode
* dir
,
2141 struct dentry
*dentry
, const char * symname
)
2144 struct inode
* inode
;
2145 int l
, err
, retries
= 0;
2147 l
= strlen(symname
)+1;
2148 if (l
> dir
->i_sb
->s_blocksize
)
2149 return -ENAMETOOLONG
;
2152 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2153 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2154 2*EXT3_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2156 return PTR_ERR(handle
);
2158 if (IS_DIRSYNC(dir
))
2161 inode
= ext3_new_inode (handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2162 err
= PTR_ERR(inode
);
2166 if (l
> sizeof (EXT3_I(inode
)->i_data
)) {
2167 inode
->i_op
= &ext3_symlink_inode_operations
;
2168 ext3_set_aops(inode
);
2170 * page_symlink() calls into ext3_prepare/commit_write.
2171 * We have a transaction open. All is sweetness. It also sets
2172 * i_size in generic_commit_write().
2174 err
= __page_symlink(inode
, symname
, l
, 1);
2177 unlock_new_inode(inode
);
2178 ext3_mark_inode_dirty(handle
, inode
);
2183 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2184 memcpy((char*)&EXT3_I(inode
)->i_data
,symname
,l
);
2185 inode
->i_size
= l
-1;
2187 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
2188 err
= ext3_add_nondir(handle
, dentry
, inode
);
2190 ext3_journal_stop(handle
);
2191 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2196 static int ext3_link (struct dentry
* old_dentry
,
2197 struct inode
* dir
, struct dentry
*dentry
)
2200 struct inode
*inode
= old_dentry
->d_inode
;
2201 int err
, retries
= 0;
2203 if (inode
->i_nlink
>= EXT3_LINK_MAX
)
2206 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2207 * otherwise has the potential to corrupt the orphan inode list.
2209 if (inode
->i_nlink
== 0)
2213 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2214 EXT3_INDEX_EXTRA_TRANS_BLOCKS
);
2216 return PTR_ERR(handle
);
2218 if (IS_DIRSYNC(dir
))
2221 inode
->i_ctime
= CURRENT_TIME_SEC
;
2223 atomic_inc(&inode
->i_count
);
2225 err
= ext3_add_entry(handle
, dentry
, inode
);
2227 ext3_mark_inode_dirty(handle
, inode
);
2228 d_instantiate(dentry
, inode
);
2233 ext3_journal_stop(handle
);
2234 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2239 #define PARENT_INO(buffer) \
2240 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2243 * Anybody can rename anything with this: the permission checks are left to the
2244 * higher-level routines.
2246 static int ext3_rename (struct inode
* old_dir
, struct dentry
*old_dentry
,
2247 struct inode
* new_dir
,struct dentry
*new_dentry
)
2250 struct inode
* old_inode
, * new_inode
;
2251 struct buffer_head
* old_bh
, * new_bh
, * dir_bh
;
2252 struct ext3_dir_entry_2
* old_de
, * new_de
;
2255 old_bh
= new_bh
= dir_bh
= NULL
;
2257 /* Initialize quotas before so that eventual writes go
2258 * in separate transaction */
2259 if (new_dentry
->d_inode
)
2260 DQUOT_INIT(new_dentry
->d_inode
);
2261 handle
= ext3_journal_start(old_dir
, 2 *
2262 EXT3_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2263 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2265 return PTR_ERR(handle
);
2267 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2270 old_bh
= ext3_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2272 * Check for inode number is _not_ due to possible IO errors.
2273 * We might rmdir the source, keep it as pwd of some process
2274 * and merrily kill the link to whatever was created under the
2275 * same name. Goodbye sticky bit ;-<
2277 old_inode
= old_dentry
->d_inode
;
2279 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2282 new_inode
= new_dentry
->d_inode
;
2283 new_bh
= ext3_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
2290 if (S_ISDIR(old_inode
->i_mode
)) {
2292 retval
= -ENOTEMPTY
;
2293 if (!empty_dir (new_inode
))
2297 dir_bh
= ext3_bread (handle
, old_inode
, 0, 0, &retval
);
2300 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2303 if (!new_inode
&& new_dir
!=old_dir
&&
2304 new_dir
->i_nlink
>= EXT3_LINK_MAX
)
2308 retval
= ext3_add_entry (handle
, new_dentry
, old_inode
);
2312 BUFFER_TRACE(new_bh
, "get write access");
2313 ext3_journal_get_write_access(handle
, new_bh
);
2314 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2315 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2316 EXT3_FEATURE_INCOMPAT_FILETYPE
))
2317 new_de
->file_type
= old_de
->file_type
;
2318 new_dir
->i_version
++;
2319 new_dir
->i_ctime
= new_dir
->i_mtime
= CURRENT_TIME_SEC
;
2320 ext3_mark_inode_dirty(handle
, new_dir
);
2321 BUFFER_TRACE(new_bh
, "call ext3_journal_dirty_metadata");
2322 ext3_journal_dirty_metadata(handle
, new_bh
);
2328 * Like most other Unix systems, set the ctime for inodes on a
2331 old_inode
->i_ctime
= CURRENT_TIME_SEC
;
2332 ext3_mark_inode_dirty(handle
, old_inode
);
2337 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2338 old_de
->name_len
!= old_dentry
->d_name
.len
||
2339 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2340 (retval
= ext3_delete_entry(handle
, old_dir
,
2341 old_de
, old_bh
)) == -ENOENT
) {
2342 /* old_de could have moved from under us during htree split, so
2343 * make sure that we are deleting the right entry. We might
2344 * also be pointing to a stale entry in the unused part of
2345 * old_bh so just checking inum and the name isn't enough. */
2346 struct buffer_head
*old_bh2
;
2347 struct ext3_dir_entry_2
*old_de2
;
2349 old_bh2
= ext3_find_entry(old_dir
, &old_dentry
->d_name
,
2352 retval
= ext3_delete_entry(handle
, old_dir
,
2358 ext3_warning(old_dir
->i_sb
, "ext3_rename",
2359 "Deleting old file (%lu), %d, error=%d",
2360 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2364 drop_nlink(new_inode
);
2365 new_inode
->i_ctime
= CURRENT_TIME_SEC
;
2367 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME_SEC
;
2368 ext3_update_dx_flag(old_dir
);
2370 BUFFER_TRACE(dir_bh
, "get_write_access");
2371 ext3_journal_get_write_access(handle
, dir_bh
);
2372 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2373 BUFFER_TRACE(dir_bh
, "call ext3_journal_dirty_metadata");
2374 ext3_journal_dirty_metadata(handle
, dir_bh
);
2375 drop_nlink(old_dir
);
2377 drop_nlink(new_inode
);
2380 ext3_update_dx_flag(new_dir
);
2381 ext3_mark_inode_dirty(handle
, new_dir
);
2384 ext3_mark_inode_dirty(handle
, old_dir
);
2386 ext3_mark_inode_dirty(handle
, new_inode
);
2387 if (!new_inode
->i_nlink
)
2388 ext3_orphan_add(handle
, new_inode
);
2396 ext3_journal_stop(handle
);
2401 * directories can handle most operations...
2403 const struct inode_operations ext3_dir_inode_operations
= {
2404 .create
= ext3_create
,
2405 .lookup
= ext3_lookup
,
2407 .unlink
= ext3_unlink
,
2408 .symlink
= ext3_symlink
,
2409 .mkdir
= ext3_mkdir
,
2410 .rmdir
= ext3_rmdir
,
2411 .mknod
= ext3_mknod
,
2412 .rename
= ext3_rename
,
2413 .setattr
= ext3_setattr
,
2414 #ifdef CONFIG_EXT3_FS_XATTR
2415 .setxattr
= generic_setxattr
,
2416 .getxattr
= generic_getxattr
,
2417 .listxattr
= ext3_listxattr
,
2418 .removexattr
= generic_removexattr
,
2420 .permission
= ext3_permission
,
2423 const struct inode_operations ext3_special_inode_operations
= {
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
,