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
27 #include <linux/quotaops.h>
34 * define how far ahead to read directories while searching them.
36 #define NAMEI_RA_CHUNKS 2
37 #define NAMEI_RA_BLOCKS 4
38 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
40 static struct buffer_head
*ext3_append(handle_t
*handle
,
44 struct buffer_head
*bh
;
46 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
48 if ((bh
= ext3_dir_bread(handle
, inode
, *block
, 1, err
))) {
49 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
50 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
51 *err
= ext3_journal_get_write_access(handle
, bh
);
61 #define assert(test) J_ASSERT(test)
65 #define dxtrace(command) command
67 #define dxtrace(command)
91 * dx_root_info is laid out so that if it should somehow get overlaid by a
92 * dirent the two low bits of the hash version will be zero. Therefore, the
93 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
98 struct fake_dirent dot
;
100 struct fake_dirent dotdot
;
104 __le32 reserved_zero
;
106 u8 info_length
; /* 8 */
111 struct dx_entry entries
[0];
116 struct fake_dirent fake
;
117 struct dx_entry entries
[0];
123 struct buffer_head
*bh
;
124 struct dx_entry
*entries
;
135 static inline unsigned dx_get_block (struct dx_entry
*entry
);
136 static void dx_set_block (struct dx_entry
*entry
, unsigned value
);
137 static inline unsigned dx_get_hash (struct dx_entry
*entry
);
138 static void dx_set_hash (struct dx_entry
*entry
, unsigned value
);
139 static unsigned dx_get_count (struct dx_entry
*entries
);
140 static unsigned dx_get_limit (struct dx_entry
*entries
);
141 static void dx_set_count (struct dx_entry
*entries
, unsigned value
);
142 static void dx_set_limit (struct dx_entry
*entries
, unsigned value
);
143 static unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
);
144 static unsigned dx_node_limit (struct inode
*dir
);
145 static struct dx_frame
*dx_probe(struct qstr
*entry
,
147 struct dx_hash_info
*hinfo
,
148 struct dx_frame
*frame
,
150 static void dx_release (struct dx_frame
*frames
);
151 static int dx_make_map(struct ext3_dir_entry_2
*de
, unsigned blocksize
,
152 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
153 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
154 static struct ext3_dir_entry_2
*dx_move_dirents (char *from
, char *to
,
155 struct dx_map_entry
*offsets
, int count
);
156 static struct ext3_dir_entry_2
*dx_pack_dirents(char *base
, unsigned blocksize
);
157 static void dx_insert_block (struct dx_frame
*frame
, u32 hash
, u32 block
);
158 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
159 struct dx_frame
*frame
,
160 struct dx_frame
*frames
,
162 static struct buffer_head
* ext3_dx_find_entry(struct inode
*dir
,
163 struct qstr
*entry
, struct ext3_dir_entry_2
**res_dir
,
165 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
166 struct inode
*inode
);
169 * p is at least 6 bytes before the end of page
171 static inline struct ext3_dir_entry_2
*
172 ext3_next_entry(struct ext3_dir_entry_2
*p
)
174 return (struct ext3_dir_entry_2
*)((char *)p
+
175 ext3_rec_len_from_disk(p
->rec_len
));
179 * Future: use high four bits of block for coalesce-on-delete flags
180 * Mask them off for now.
183 static inline unsigned dx_get_block (struct dx_entry
*entry
)
185 return le32_to_cpu(entry
->block
) & 0x00ffffff;
188 static inline void dx_set_block (struct dx_entry
*entry
, unsigned value
)
190 entry
->block
= cpu_to_le32(value
);
193 static inline unsigned dx_get_hash (struct dx_entry
*entry
)
195 return le32_to_cpu(entry
->hash
);
198 static inline void dx_set_hash (struct dx_entry
*entry
, unsigned value
)
200 entry
->hash
= cpu_to_le32(value
);
203 static inline unsigned dx_get_count (struct dx_entry
*entries
)
205 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
208 static inline unsigned dx_get_limit (struct dx_entry
*entries
)
210 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
213 static inline void dx_set_count (struct dx_entry
*entries
, unsigned value
)
215 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
218 static inline void dx_set_limit (struct dx_entry
*entries
, unsigned value
)
220 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
223 static inline unsigned dx_root_limit (struct inode
*dir
, unsigned infosize
)
225 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(1) -
226 EXT3_DIR_REC_LEN(2) - infosize
;
227 return entry_space
/ sizeof(struct dx_entry
);
230 static inline unsigned dx_node_limit (struct inode
*dir
)
232 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT3_DIR_REC_LEN(0);
233 return entry_space
/ sizeof(struct dx_entry
);
240 static void dx_show_index (char * label
, struct dx_entry
*entries
)
242 int i
, n
= dx_get_count (entries
);
243 printk("%s index ", label
);
244 for (i
= 0; i
< n
; i
++)
246 printk("%x->%u ", i
? dx_get_hash(entries
+ i
): 0, dx_get_block(entries
+ i
));
258 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext3_dir_entry_2
*de
,
259 int size
, int show_names
)
261 unsigned names
= 0, space
= 0;
262 char *base
= (char *) de
;
263 struct dx_hash_info h
= *hinfo
;
266 while ((char *) de
< base
+ size
)
272 int len
= de
->name_len
;
273 char *name
= de
->name
;
274 while (len
--) printk("%c", *name
++);
275 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
276 printk(":%x.%u ", h
.hash
,
277 (unsigned) ((char *) de
- base
));
279 space
+= EXT3_DIR_REC_LEN(de
->name_len
);
282 de
= ext3_next_entry(de
);
284 printk("(%i)\n", names
);
285 return (struct stats
) { names
, space
, 1 };
288 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
289 struct dx_entry
*entries
, int levels
)
291 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
292 unsigned count
= dx_get_count (entries
), names
= 0, space
= 0, i
;
294 struct buffer_head
*bh
;
296 printk("%i indexed blocks...\n", count
);
297 for (i
= 0; i
< count
; i
++, entries
++)
299 u32 block
= dx_get_block(entries
), hash
= i
? dx_get_hash(entries
): 0;
300 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
302 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
303 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0,&err
))) continue;
305 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
306 dx_show_leaf(hinfo
, (struct ext3_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
307 names
+= stats
.names
;
308 space
+= stats
.space
;
309 bcount
+= stats
.bcount
;
313 printk("%snames %u, fullness %u (%u%%)\n", levels
?"":" ",
314 names
, space
/bcount
,(space
/bcount
)*100/blocksize
);
315 return (struct stats
) { names
, space
, bcount
};
317 #endif /* DX_DEBUG */
320 * Probe for a directory leaf block to search.
322 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
323 * error in the directory index, and the caller should fall back to
324 * searching the directory normally. The callers of dx_probe **MUST**
325 * check for this error code, and make sure it never gets reflected
328 static struct dx_frame
*
329 dx_probe(struct qstr
*entry
, struct inode
*dir
,
330 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
332 unsigned count
, indirect
;
333 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
334 struct dx_root
*root
;
335 struct buffer_head
*bh
;
336 struct dx_frame
*frame
= frame_in
;
340 if (!(bh
= ext3_dir_bread(NULL
, dir
, 0, 0, err
))) {
341 *err
= ERR_BAD_DX_DIR
;
344 root
= (struct dx_root
*) bh
->b_data
;
345 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
346 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
347 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
348 ext3_warning(dir
->i_sb
, __func__
,
349 "Unrecognised inode hash code %d",
350 root
->info
.hash_version
);
352 *err
= ERR_BAD_DX_DIR
;
355 hinfo
->hash_version
= root
->info
.hash_version
;
356 if (hinfo
->hash_version
<= DX_HASH_TEA
)
357 hinfo
->hash_version
+= EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
358 hinfo
->seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
360 ext3fs_dirhash(entry
->name
, entry
->len
, hinfo
);
363 if (root
->info
.unused_flags
& 1) {
364 ext3_warning(dir
->i_sb
, __func__
,
365 "Unimplemented inode hash flags: %#06x",
366 root
->info
.unused_flags
);
368 *err
= ERR_BAD_DX_DIR
;
372 if ((indirect
= root
->info
.indirect_levels
) > 1) {
373 ext3_warning(dir
->i_sb
, __func__
,
374 "Unimplemented inode hash depth: %#06x",
375 root
->info
.indirect_levels
);
377 *err
= ERR_BAD_DX_DIR
;
381 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
382 root
->info
.info_length
);
384 if (dx_get_limit(entries
) != dx_root_limit(dir
,
385 root
->info
.info_length
)) {
386 ext3_warning(dir
->i_sb
, __func__
,
387 "dx entry: limit != root limit");
389 *err
= ERR_BAD_DX_DIR
;
393 dxtrace (printk("Look up %x", hash
));
396 count
= dx_get_count(entries
);
397 if (!count
|| count
> dx_get_limit(entries
)) {
398 ext3_warning(dir
->i_sb
, __func__
,
399 "dx entry: no count or count > limit");
401 *err
= ERR_BAD_DX_DIR
;
406 q
= entries
+ count
- 1;
410 dxtrace(printk("."));
411 if (dx_get_hash(m
) > hash
)
417 if (0) // linear search cross check
419 unsigned n
= count
- 1;
423 dxtrace(printk(","));
424 if (dx_get_hash(++at
) > hash
)
430 assert (at
== p
- 1);
434 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
436 frame
->entries
= entries
;
438 if (!indirect
--) return frame
;
439 if (!(bh
= ext3_dir_bread(NULL
, dir
, dx_get_block(at
), 0, err
))) {
440 *err
= ERR_BAD_DX_DIR
;
443 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
444 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
445 ext3_warning(dir
->i_sb
, __func__
,
446 "dx entry: limit != node limit");
448 *err
= ERR_BAD_DX_DIR
;
455 while (frame
>= frame_in
) {
460 if (*err
== ERR_BAD_DX_DIR
)
461 ext3_warning(dir
->i_sb
, __func__
,
462 "Corrupt dir inode %ld, running e2fsck is "
463 "recommended.", dir
->i_ino
);
467 static void dx_release (struct dx_frame
*frames
)
469 if (frames
[0].bh
== NULL
)
472 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
473 brelse(frames
[1].bh
);
474 brelse(frames
[0].bh
);
478 * This function increments the frame pointer to search the next leaf
479 * block, and reads in the necessary intervening nodes if the search
480 * should be necessary. Whether or not the search is necessary is
481 * controlled by the hash parameter. If the hash value is even, then
482 * the search is only continued if the next block starts with that
483 * hash value. This is used if we are searching for a specific file.
485 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
487 * This function returns 1 if the caller should continue to search,
488 * or 0 if it should not. If there is an error reading one of the
489 * index blocks, it will a negative error code.
491 * If start_hash is non-null, it will be filled in with the starting
492 * hash of the next page.
494 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
495 struct dx_frame
*frame
,
496 struct dx_frame
*frames
,
500 struct buffer_head
*bh
;
501 int err
, num_frames
= 0;
506 * Find the next leaf page by incrementing the frame pointer.
507 * If we run out of entries in the interior node, loop around and
508 * increment pointer in the parent node. When we break out of
509 * this loop, num_frames indicates the number of interior
510 * nodes need to be read.
513 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
522 * If the hash is 1, then continue only if the next page has a
523 * continuation hash of any value. This is used for readdir
524 * handling. Otherwise, check to see if the hash matches the
525 * desired contiuation hash. If it doesn't, return since
526 * there's no point to read in the successive index pages.
528 bhash
= dx_get_hash(p
->at
);
531 if ((hash
& 1) == 0) {
532 if ((bhash
& ~1) != hash
)
536 * If the hash is HASH_NB_ALWAYS, we always go to the next
537 * block so no check is necessary
539 while (num_frames
--) {
540 if (!(bh
= ext3_dir_bread(NULL
, dir
, dx_get_block(p
->at
),
542 return err
; /* Failure */
546 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
553 * This function fills a red-black tree with information from a
554 * directory block. It returns the number directory entries loaded
555 * into the tree. If there is an error it is returned in err.
557 static int htree_dirblock_to_tree(struct file
*dir_file
,
558 struct inode
*dir
, int block
,
559 struct dx_hash_info
*hinfo
,
560 __u32 start_hash
, __u32 start_minor_hash
)
562 struct buffer_head
*bh
;
563 struct ext3_dir_entry_2
*de
, *top
;
564 int err
= 0, count
= 0;
566 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block
));
568 if (!(bh
= ext3_dir_bread(NULL
, dir
, block
, 0, &err
)))
571 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
572 top
= (struct ext3_dir_entry_2
*) ((char *) de
+
573 dir
->i_sb
->s_blocksize
-
574 EXT3_DIR_REC_LEN(0));
575 for (; de
< top
; de
= ext3_next_entry(de
)) {
576 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
577 (block
<<EXT3_BLOCK_SIZE_BITS(dir
->i_sb
))
578 +((char *)de
- bh
->b_data
))) {
579 /* On error, skip the f_pos to the next block. */
580 dir_file
->f_pos
= (dir_file
->f_pos
|
581 (dir
->i_sb
->s_blocksize
- 1)) + 1;
585 ext3fs_dirhash(de
->name
, de
->name_len
, hinfo
);
586 if ((hinfo
->hash
< start_hash
) ||
587 ((hinfo
->hash
== start_hash
) &&
588 (hinfo
->minor_hash
< start_minor_hash
)))
592 if ((err
= ext3_htree_store_dirent(dir_file
,
593 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
605 * This function fills a red-black tree with information from a
606 * directory. We start scanning the directory in hash order, starting
607 * at start_hash and start_minor_hash.
609 * This function returns the number of entries inserted into the tree,
610 * or a negative error code.
612 int ext3_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
613 __u32 start_minor_hash
, __u32
*next_hash
)
615 struct dx_hash_info hinfo
;
616 struct ext3_dir_entry_2
*de
;
617 struct dx_frame frames
[2], *frame
;
624 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash
,
626 dir
= file_inode(dir_file
);
627 if (!(EXT3_I(dir
)->i_flags
& EXT3_INDEX_FL
)) {
628 hinfo
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
629 if (hinfo
.hash_version
<= DX_HASH_TEA
)
630 hinfo
.hash_version
+=
631 EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
632 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
633 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
634 start_hash
, start_minor_hash
);
638 hinfo
.hash
= start_hash
;
639 hinfo
.minor_hash
= 0;
640 frame
= dx_probe(NULL
, file_inode(dir_file
), &hinfo
, frames
, &err
);
644 /* Add '.' and '..' from the htree header */
645 if (!start_hash
&& !start_minor_hash
) {
646 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
647 if ((err
= ext3_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
651 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
652 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
653 de
= ext3_next_entry(de
);
654 if ((err
= ext3_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
660 block
= dx_get_block(frame
->at
);
661 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
662 start_hash
, start_minor_hash
);
669 ret
= ext3_htree_next_block(dir
, HASH_NB_ALWAYS
,
670 frame
, frames
, &hashval
);
671 *next_hash
= hashval
;
677 * Stop if: (a) there are no more entries, or
678 * (b) we have inserted at least one entry and the
679 * next hash value is not a continuation
682 (count
&& ((hashval
& 1) == 0)))
686 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
696 * Directory block splitting, compacting
700 * Create map of hash values, offsets, and sizes, stored at end of block.
701 * Returns number of entries mapped.
703 static int dx_make_map(struct ext3_dir_entry_2
*de
, unsigned blocksize
,
704 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
707 char *base
= (char *) de
;
708 struct dx_hash_info h
= *hinfo
;
710 while ((char *) de
< base
+ blocksize
)
712 if (de
->name_len
&& de
->inode
) {
713 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
715 map_tail
->hash
= h
.hash
;
716 map_tail
->offs
= (u16
) ((char *) de
- base
);
717 map_tail
->size
= le16_to_cpu(de
->rec_len
);
721 /* XXX: do we need to check rec_len == 0 case? -Chris */
722 de
= ext3_next_entry(de
);
727 /* Sort map by hash value */
728 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
730 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
732 /* Combsort until bubble sort doesn't suck */
736 if (count
- 9 < 2) /* 9, 10 -> 11 */
738 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
739 if (p
->hash
< q
->hash
)
742 /* Garden variety bubble sort */
748 if (q
[1].hash
>= q
[0].hash
)
756 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, u32 block
)
758 struct dx_entry
*entries
= frame
->entries
;
759 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
760 int count
= dx_get_count(entries
);
762 assert(count
< dx_get_limit(entries
));
763 assert(old
< entries
+ count
);
764 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
765 dx_set_hash(new, hash
);
766 dx_set_block(new, block
);
767 dx_set_count(entries
, count
+ 1);
770 static void ext3_update_dx_flag(struct inode
*inode
)
772 if (!EXT3_HAS_COMPAT_FEATURE(inode
->i_sb
,
773 EXT3_FEATURE_COMPAT_DIR_INDEX
))
774 EXT3_I(inode
)->i_flags
&= ~EXT3_INDEX_FL
;
778 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
780 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
781 * `de != NULL' is guaranteed by caller.
783 static inline int ext3_match (int len
, const char * const name
,
784 struct ext3_dir_entry_2
* de
)
786 if (len
!= de
->name_len
)
790 return !memcmp(name
, de
->name
, len
);
794 * Returns 0 if not found, -1 on failure, and 1 on success
796 static inline int search_dirblock(struct buffer_head
* bh
,
799 unsigned long offset
,
800 struct ext3_dir_entry_2
** res_dir
)
802 struct ext3_dir_entry_2
* de
;
805 const char *name
= child
->name
;
806 int namelen
= child
->len
;
808 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
809 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
810 while ((char *) de
< dlimit
) {
811 /* this code is executed quadratically often */
812 /* do minimal checking `by hand' */
814 if ((char *) de
+ namelen
<= dlimit
&&
815 ext3_match (namelen
, name
, de
)) {
816 /* found a match - just to be sure, do a full check */
817 if (!ext3_check_dir_entry("ext3_find_entry",
818 dir
, de
, bh
, offset
))
823 /* prevent looping on a bad block */
824 de_len
= ext3_rec_len_from_disk(de
->rec_len
);
828 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ de_len
);
837 * finds an entry in the specified directory with the wanted name. It
838 * returns the cache buffer in which the entry was found, and the entry
839 * itself (as a parameter - res_dir). It does NOT read the inode of the
840 * entry - you'll have to do that yourself if you want to.
842 * The returned buffer_head has ->b_count elevated. The caller is expected
843 * to brelse() it when appropriate.
845 static struct buffer_head
*ext3_find_entry(struct inode
*dir
,
847 struct ext3_dir_entry_2
**res_dir
)
849 struct super_block
* sb
;
850 struct buffer_head
* bh_use
[NAMEI_RA_SIZE
];
851 struct buffer_head
* bh
, *ret
= NULL
;
852 unsigned long start
, block
, b
;
853 const u8
*name
= entry
->name
;
854 int ra_max
= 0; /* Number of bh's in the readahead
856 int ra_ptr
= 0; /* Current index into readahead
864 namelen
= entry
->len
;
865 if (namelen
> EXT3_NAME_LEN
)
867 if ((namelen
<= 2) && (name
[0] == '.') &&
868 (name
[1] == '.' || name
[1] == 0)) {
870 * "." or ".." will only be in the first block
871 * NFS may look up ".."; "." should be handled by the VFS
878 bh
= ext3_dx_find_entry(dir
, entry
, res_dir
, &err
);
880 * On success, or if the error was file not found,
881 * return. Otherwise, fall back to doing a search the
884 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
886 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
888 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
889 start
= EXT3_I(dir
)->i_dir_start_lookup
;
890 if (start
>= nblocks
)
896 * We deal with the read-ahead logic here.
898 if (ra_ptr
>= ra_max
) {
899 /* Refill the readahead buffer */
902 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
904 * Terminate if we reach the end of the
905 * directory and must wrap, or if our
906 * search has finished at this block.
908 if (b
>= nblocks
|| (num
&& block
== start
)) {
909 bh_use
[ra_max
] = NULL
;
913 bh
= ext3_getblk(NULL
, dir
, b
++, 0, &err
);
915 if (bh
&& !bh_uptodate_or_lock(bh
)) {
917 bh
->b_end_io
= end_buffer_read_sync
;
918 submit_bh(READ
| REQ_META
| REQ_PRIO
,
923 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
926 if (!buffer_uptodate(bh
)) {
927 /* read error, skip block & hope for the best */
928 ext3_error(sb
, __func__
, "reading directory #%lu "
929 "offset %lu", dir
->i_ino
, block
);
933 i
= search_dirblock(bh
, dir
, entry
,
934 block
<< EXT3_BLOCK_SIZE_BITS(sb
), res_dir
);
936 EXT3_I(dir
)->i_dir_start_lookup
= block
;
938 goto cleanup_and_exit
;
942 goto cleanup_and_exit
;
945 if (++block
>= nblocks
)
947 } while (block
!= start
);
950 * If the directory has grown while we were searching, then
951 * search the last part of the directory before giving up.
954 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
955 if (block
< nblocks
) {
961 /* Clean up the read-ahead blocks */
962 for (; ra_ptr
< ra_max
; ra_ptr
++)
963 brelse (bh_use
[ra_ptr
]);
967 static struct buffer_head
* ext3_dx_find_entry(struct inode
*dir
,
968 struct qstr
*entry
, struct ext3_dir_entry_2
**res_dir
,
971 struct super_block
*sb
= dir
->i_sb
;
972 struct dx_hash_info hinfo
;
973 struct dx_frame frames
[2], *frame
;
974 struct buffer_head
*bh
;
978 if (!(frame
= dx_probe(entry
, dir
, &hinfo
, frames
, err
)))
981 block
= dx_get_block(frame
->at
);
982 if (!(bh
= ext3_dir_bread (NULL
, dir
, block
, 0, err
)))
985 retval
= search_dirblock(bh
, dir
, entry
,
986 block
<< EXT3_BLOCK_SIZE_BITS(sb
),
994 *err
= ERR_BAD_DX_DIR
;
998 /* Check to see if we should continue to search */
999 retval
= ext3_htree_next_block(dir
, hinfo
.hash
, frame
,
1002 ext3_warning(sb
, __func__
,
1003 "error reading index page in directory #%lu",
1008 } while (retval
== 1);
1012 dxtrace(printk("%s not found\n", entry
->name
));
1013 dx_release (frames
);
1017 static struct dentry
*ext3_lookup(struct inode
* dir
, struct dentry
*dentry
, unsigned int flags
)
1019 struct inode
* inode
;
1020 struct ext3_dir_entry_2
* de
;
1021 struct buffer_head
* bh
;
1023 if (dentry
->d_name
.len
> EXT3_NAME_LEN
)
1024 return ERR_PTR(-ENAMETOOLONG
);
1026 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
1029 unsigned long ino
= le32_to_cpu(de
->inode
);
1031 if (!ext3_valid_inum(dir
->i_sb
, ino
)) {
1032 ext3_error(dir
->i_sb
, "ext3_lookup",
1033 "bad inode number: %lu", ino
);
1034 return ERR_PTR(-EIO
);
1036 inode
= ext3_iget(dir
->i_sb
, ino
);
1037 if (inode
== ERR_PTR(-ESTALE
)) {
1038 ext3_error(dir
->i_sb
, __func__
,
1039 "deleted inode referenced: %lu",
1041 return ERR_PTR(-EIO
);
1044 return d_splice_alias(inode
, dentry
);
1048 struct dentry
*ext3_get_parent(struct dentry
*child
)
1051 struct qstr dotdot
= QSTR_INIT("..", 2);
1052 struct ext3_dir_entry_2
* de
;
1053 struct buffer_head
*bh
;
1055 bh
= ext3_find_entry(child
->d_inode
, &dotdot
, &de
);
1057 return ERR_PTR(-ENOENT
);
1058 ino
= le32_to_cpu(de
->inode
);
1061 if (!ext3_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1062 ext3_error(child
->d_inode
->i_sb
, "ext3_get_parent",
1063 "bad inode number: %lu", ino
);
1064 return ERR_PTR(-EIO
);
1067 return d_obtain_alias(ext3_iget(child
->d_inode
->i_sb
, ino
));
1071 static unsigned char ext3_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1072 [S_IFREG
>> S_SHIFT
] = EXT3_FT_REG_FILE
,
1073 [S_IFDIR
>> S_SHIFT
] = EXT3_FT_DIR
,
1074 [S_IFCHR
>> S_SHIFT
] = EXT3_FT_CHRDEV
,
1075 [S_IFBLK
>> S_SHIFT
] = EXT3_FT_BLKDEV
,
1076 [S_IFIFO
>> S_SHIFT
] = EXT3_FT_FIFO
,
1077 [S_IFSOCK
>> S_SHIFT
] = EXT3_FT_SOCK
,
1078 [S_IFLNK
>> S_SHIFT
] = EXT3_FT_SYMLINK
,
1081 static inline void ext3_set_de_type(struct super_block
*sb
,
1082 struct ext3_dir_entry_2
*de
,
1084 if (EXT3_HAS_INCOMPAT_FEATURE(sb
, EXT3_FEATURE_INCOMPAT_FILETYPE
))
1085 de
->file_type
= ext3_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1089 * Move count entries from end of map between two memory locations.
1090 * Returns pointer to last entry moved.
1092 static struct ext3_dir_entry_2
*
1093 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1095 unsigned rec_len
= 0;
1098 struct ext3_dir_entry_2
*de
= (struct ext3_dir_entry_2
*) (from
+ map
->offs
);
1099 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1100 memcpy (to
, de
, rec_len
);
1101 ((struct ext3_dir_entry_2
*) to
)->rec_len
=
1102 ext3_rec_len_to_disk(rec_len
);
1107 return (struct ext3_dir_entry_2
*) (to
- rec_len
);
1111 * Compact each dir entry in the range to the minimal rec_len.
1112 * Returns pointer to last entry in range.
1114 static struct ext3_dir_entry_2
*dx_pack_dirents(char *base
, unsigned blocksize
)
1116 struct ext3_dir_entry_2
*next
, *to
, *prev
;
1117 struct ext3_dir_entry_2
*de
= (struct ext3_dir_entry_2
*)base
;
1118 unsigned rec_len
= 0;
1121 while ((char *)de
< base
+ blocksize
) {
1122 next
= ext3_next_entry(de
);
1123 if (de
->inode
&& de
->name_len
) {
1124 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1126 memmove(to
, de
, rec_len
);
1127 to
->rec_len
= ext3_rec_len_to_disk(rec_len
);
1129 to
= (struct ext3_dir_entry_2
*) (((char *) to
) + rec_len
);
1137 * Split a full leaf block to make room for a new dir entry.
1138 * Allocate a new block, and move entries so that they are approx. equally full.
1139 * Returns pointer to de in block into which the new entry will be inserted.
1141 static struct ext3_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1142 struct buffer_head
**bh
,struct dx_frame
*frame
,
1143 struct dx_hash_info
*hinfo
, int *error
)
1145 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1146 unsigned count
, continued
;
1147 struct buffer_head
*bh2
;
1150 struct dx_map_entry
*map
;
1151 char *data1
= (*bh
)->b_data
, *data2
;
1152 unsigned split
, move
, size
;
1153 struct ext3_dir_entry_2
*de
= NULL
, *de2
;
1156 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1163 BUFFER_TRACE(*bh
, "get_write_access");
1164 err
= ext3_journal_get_write_access(handle
, *bh
);
1168 BUFFER_TRACE(frame
->bh
, "get_write_access");
1169 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1173 data2
= bh2
->b_data
;
1175 /* create map in the end of data2 block */
1176 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1177 count
= dx_make_map ((struct ext3_dir_entry_2
*) data1
,
1178 blocksize
, hinfo
, map
);
1180 dx_sort_map (map
, count
);
1181 /* Split the existing block in the middle, size-wise */
1184 for (i
= count
-1; i
>= 0; i
--) {
1185 /* is more than half of this entry in 2nd half of the block? */
1186 if (size
+ map
[i
].size
/2 > blocksize
/2)
1188 size
+= map
[i
].size
;
1191 /* map index at which we will split */
1192 split
= count
- move
;
1193 hash2
= map
[split
].hash
;
1194 continued
= hash2
== map
[split
- 1].hash
;
1195 dxtrace(printk("Split block %i at %x, %i/%i\n",
1196 dx_get_block(frame
->at
), hash2
, split
, count
-split
));
1198 /* Fancy dance to stay within two buffers */
1199 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1200 de
= dx_pack_dirents(data1
,blocksize
);
1201 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1202 de2
->rec_len
= ext3_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1203 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data1
, blocksize
, 1));
1204 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data2
, blocksize
, 1));
1206 /* Which block gets the new entry? */
1207 if (hinfo
->hash
>= hash2
)
1212 dx_insert_block (frame
, hash2
+ continued
, newblock
);
1213 err
= ext3_journal_dirty_metadata (handle
, bh2
);
1216 err
= ext3_journal_dirty_metadata (handle
, frame
->bh
);
1220 dxtrace(dx_show_index ("frame", frame
->entries
));
1227 ext3_std_error(dir
->i_sb
, err
);
1235 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1236 * it points to a directory entry which is guaranteed to be large
1237 * enough for new directory entry. If de is NULL, then
1238 * add_dirent_to_buf will attempt search the directory block for
1239 * space. It will return -ENOSPC if no space is available, and -EIO
1240 * and -EEXIST if directory entry already exists.
1242 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1243 * all other cases bh is released.
1245 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1246 struct inode
*inode
, struct ext3_dir_entry_2
*de
,
1247 struct buffer_head
* bh
)
1249 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1250 const char *name
= dentry
->d_name
.name
;
1251 int namelen
= dentry
->d_name
.len
;
1252 unsigned long offset
= 0;
1253 unsigned short reclen
;
1254 int nlen
, rlen
, err
;
1257 reclen
= EXT3_DIR_REC_LEN(namelen
);
1259 de
= (struct ext3_dir_entry_2
*)bh
->b_data
;
1260 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1261 while ((char *) de
<= top
) {
1262 if (!ext3_check_dir_entry("ext3_add_entry", dir
, de
,
1267 if (ext3_match (namelen
, name
, de
)) {
1271 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1272 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1273 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1275 de
= (struct ext3_dir_entry_2
*)((char *)de
+ rlen
);
1278 if ((char *) de
> top
)
1281 BUFFER_TRACE(bh
, "get_write_access");
1282 err
= ext3_journal_get_write_access(handle
, bh
);
1284 ext3_std_error(dir
->i_sb
, err
);
1289 /* By now the buffer is marked for journaling */
1290 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1291 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1293 struct ext3_dir_entry_2
*de1
= (struct ext3_dir_entry_2
*)((char *)de
+ nlen
);
1294 de1
->rec_len
= ext3_rec_len_to_disk(rlen
- nlen
);
1295 de
->rec_len
= ext3_rec_len_to_disk(nlen
);
1298 de
->file_type
= EXT3_FT_UNKNOWN
;
1300 de
->inode
= cpu_to_le32(inode
->i_ino
);
1301 ext3_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1304 de
->name_len
= namelen
;
1305 memcpy (de
->name
, name
, namelen
);
1307 * XXX shouldn't update any times until successful
1308 * completion of syscall, but too many callers depend
1311 * XXX similarly, too many callers depend on
1312 * ext3_new_inode() setting the times, but error
1313 * recovery deletes the inode, so the worst that can
1314 * happen is that the times are slightly out of date
1315 * and/or different from the directory change time.
1317 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME_SEC
;
1318 ext3_update_dx_flag(dir
);
1320 ext3_mark_inode_dirty(handle
, dir
);
1321 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1322 err
= ext3_journal_dirty_metadata(handle
, bh
);
1324 ext3_std_error(dir
->i_sb
, err
);
1330 * This converts a one block unindexed directory to a 3 block indexed
1331 * directory, and adds the dentry to the indexed directory.
1333 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1334 struct inode
*inode
, struct buffer_head
*bh
)
1336 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1337 const char *name
= dentry
->d_name
.name
;
1338 int namelen
= dentry
->d_name
.len
;
1339 struct buffer_head
*bh2
;
1340 struct dx_root
*root
;
1341 struct dx_frame frames
[2], *frame
;
1342 struct dx_entry
*entries
;
1343 struct ext3_dir_entry_2
*de
, *de2
;
1348 struct dx_hash_info hinfo
;
1350 struct fake_dirent
*fde
;
1352 blocksize
= dir
->i_sb
->s_blocksize
;
1353 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1354 retval
= ext3_journal_get_write_access(handle
, bh
);
1356 ext3_std_error(dir
->i_sb
, retval
);
1360 root
= (struct dx_root
*) bh
->b_data
;
1362 /* The 0th block becomes the root, move the dirents out */
1363 fde
= &root
->dotdot
;
1364 de
= (struct ext3_dir_entry_2
*)((char *)fde
+
1365 ext3_rec_len_from_disk(fde
->rec_len
));
1366 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1367 ext3_error(dir
->i_sb
, __func__
,
1368 "invalid rec_len for '..' in inode %lu",
1373 len
= ((char *) root
) + blocksize
- (char *) de
;
1375 bh2
= ext3_append (handle
, dir
, &block
, &retval
);
1380 EXT3_I(dir
)->i_flags
|= EXT3_INDEX_FL
;
1381 data1
= bh2
->b_data
;
1383 memcpy (data1
, de
, len
);
1384 de
= (struct ext3_dir_entry_2
*) data1
;
1386 while ((char *)(de2
= ext3_next_entry(de
)) < top
)
1388 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1389 /* Initialize the root; the dot dirents already exist */
1390 de
= (struct ext3_dir_entry_2
*) (&root
->dotdot
);
1391 de
->rec_len
= ext3_rec_len_to_disk(blocksize
- EXT3_DIR_REC_LEN(2));
1392 memset (&root
->info
, 0, sizeof(root
->info
));
1393 root
->info
.info_length
= sizeof(root
->info
);
1394 root
->info
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
1395 entries
= root
->entries
;
1396 dx_set_block (entries
, 1);
1397 dx_set_count (entries
, 1);
1398 dx_set_limit (entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1400 /* Initialize as for dx_probe */
1401 hinfo
.hash_version
= root
->info
.hash_version
;
1402 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1403 hinfo
.hash_version
+= EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
1404 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
1405 ext3fs_dirhash(name
, namelen
, &hinfo
);
1407 frame
->entries
= entries
;
1408 frame
->at
= entries
;
1412 * Mark buffers dirty here so that if do_split() fails we write a
1413 * consistent set of buffers to disk.
1415 ext3_journal_dirty_metadata(handle
, frame
->bh
);
1416 ext3_journal_dirty_metadata(handle
, bh
);
1417 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1419 ext3_mark_inode_dirty(handle
, dir
);
1425 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1431 * adds a file entry to the specified directory, using the same
1432 * semantics as ext3_find_entry(). It returns NULL if it failed.
1434 * NOTE!! The inode part of 'de' is left at 0 - which means you
1435 * may not sleep between calling this and putting something into
1436 * the entry, as someone else might have used it while you slept.
1438 static int ext3_add_entry (handle_t
*handle
, struct dentry
*dentry
,
1439 struct inode
*inode
)
1441 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1442 struct buffer_head
* bh
;
1443 struct ext3_dir_entry_2
*de
;
1444 struct super_block
* sb
;
1451 blocksize
= sb
->s_blocksize
;
1452 if (!dentry
->d_name
.len
)
1455 retval
= ext3_dx_add_entry(handle
, dentry
, inode
);
1456 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1458 EXT3_I(dir
)->i_flags
&= ~EXT3_INDEX_FL
;
1460 ext3_mark_inode_dirty(handle
, dir
);
1462 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1463 for (block
= 0; block
< blocks
; block
++) {
1464 if (!(bh
= ext3_dir_bread(handle
, dir
, block
, 0, &retval
)))
1467 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1468 if (retval
!= -ENOSPC
)
1471 if (blocks
== 1 && !dx_fallback
&&
1472 EXT3_HAS_COMPAT_FEATURE(sb
, EXT3_FEATURE_COMPAT_DIR_INDEX
))
1473 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1476 bh
= ext3_append(handle
, dir
, &block
, &retval
);
1479 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1481 de
->rec_len
= ext3_rec_len_to_disk(blocksize
);
1482 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1486 * Returns 0 for success, or a negative error value
1488 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1489 struct inode
*inode
)
1491 struct dx_frame frames
[2], *frame
;
1492 struct dx_entry
*entries
, *at
;
1493 struct dx_hash_info hinfo
;
1494 struct buffer_head
* bh
;
1495 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1496 struct super_block
* sb
= dir
->i_sb
;
1497 struct ext3_dir_entry_2
*de
;
1500 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1503 entries
= frame
->entries
;
1506 if (!(bh
= ext3_dir_bread(handle
, dir
, dx_get_block(frame
->at
), 0, &err
)))
1509 BUFFER_TRACE(bh
, "get_write_access");
1510 err
= ext3_journal_get_write_access(handle
, bh
);
1514 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1515 if (err
!= -ENOSPC
) {
1520 /* Block full, should compress but for now just split */
1521 dxtrace(printk("using %u of %u node entries\n",
1522 dx_get_count(entries
), dx_get_limit(entries
)));
1523 /* Need to split index? */
1524 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1526 unsigned icount
= dx_get_count(entries
);
1527 int levels
= frame
- frames
;
1528 struct dx_entry
*entries2
;
1529 struct dx_node
*node2
;
1530 struct buffer_head
*bh2
;
1532 if (levels
&& (dx_get_count(frames
->entries
) ==
1533 dx_get_limit(frames
->entries
))) {
1534 ext3_warning(sb
, __func__
,
1535 "Directory index full!");
1539 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1542 node2
= (struct dx_node
*)(bh2
->b_data
);
1543 entries2
= node2
->entries
;
1544 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
1545 node2
->fake
.rec_len
= ext3_rec_len_to_disk(sb
->s_blocksize
);
1546 BUFFER_TRACE(frame
->bh
, "get_write_access");
1547 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1551 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1552 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1553 dxtrace(printk("Split index %i/%i\n", icount1
, icount2
));
1555 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1556 err
= ext3_journal_get_write_access(handle
,
1561 memcpy ((char *) entries2
, (char *) (entries
+ icount1
),
1562 icount2
* sizeof(struct dx_entry
));
1563 dx_set_count (entries
, icount1
);
1564 dx_set_count (entries2
, icount2
);
1565 dx_set_limit (entries2
, dx_node_limit(dir
));
1567 /* Which index block gets the new entry? */
1568 if (at
- entries
>= icount1
) {
1569 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1570 frame
->entries
= entries
= entries2
;
1571 swap(frame
->bh
, bh2
);
1573 dx_insert_block (frames
+ 0, hash2
, newblock
);
1574 dxtrace(dx_show_index ("node", frames
[1].entries
));
1575 dxtrace(dx_show_index ("node",
1576 ((struct dx_node
*) bh2
->b_data
)->entries
));
1577 err
= ext3_journal_dirty_metadata(handle
, bh2
);
1582 dxtrace(printk("Creating second level index...\n"));
1583 memcpy((char *) entries2
, (char *) entries
,
1584 icount
* sizeof(struct dx_entry
));
1585 dx_set_limit(entries2
, dx_node_limit(dir
));
1588 dx_set_count(entries
, 1);
1589 dx_set_block(entries
+ 0, newblock
);
1590 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1592 /* Add new access path frame */
1594 frame
->at
= at
= at
- entries
+ entries2
;
1595 frame
->entries
= entries
= entries2
;
1597 err
= ext3_journal_get_write_access(handle
,
1602 err
= ext3_journal_dirty_metadata(handle
, frames
[0].bh
);
1606 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1609 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1614 ext3_std_error(dir
->i_sb
, err
);
1623 * ext3_delete_entry deletes a directory entry by merging it with the
1626 static int ext3_delete_entry (handle_t
*handle
,
1628 struct ext3_dir_entry_2
* de_del
,
1629 struct buffer_head
* bh
)
1631 struct ext3_dir_entry_2
* de
, * pde
;
1636 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1637 while (i
< bh
->b_size
) {
1638 if (!ext3_check_dir_entry("ext3_delete_entry", dir
, de
, bh
, i
))
1643 BUFFER_TRACE(bh
, "get_write_access");
1644 err
= ext3_journal_get_write_access(handle
, bh
);
1649 pde
->rec_len
= ext3_rec_len_to_disk(
1650 ext3_rec_len_from_disk(pde
->rec_len
) +
1651 ext3_rec_len_from_disk(de
->rec_len
));
1655 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1656 err
= ext3_journal_dirty_metadata(handle
, bh
);
1659 ext3_std_error(dir
->i_sb
, err
);
1664 i
+= ext3_rec_len_from_disk(de
->rec_len
);
1666 de
= ext3_next_entry(de
);
1671 static int ext3_add_nondir(handle_t
*handle
,
1672 struct dentry
*dentry
, struct inode
*inode
)
1674 int err
= ext3_add_entry(handle
, dentry
, inode
);
1676 ext3_mark_inode_dirty(handle
, inode
);
1677 unlock_new_inode(inode
);
1678 d_instantiate(dentry
, inode
);
1682 unlock_new_inode(inode
);
1688 * By the time this is called, we already have created
1689 * the directory cache entry for the new file, but it
1690 * is so far negative - it has no inode.
1692 * If the create succeeds, we fill in the inode information
1693 * with d_instantiate().
1695 static int ext3_create (struct inode
* dir
, struct dentry
* dentry
, umode_t mode
,
1699 struct inode
* inode
;
1700 int err
, retries
= 0;
1702 dquot_initialize(dir
);
1705 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1706 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1707 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1709 return PTR_ERR(handle
);
1711 if (IS_DIRSYNC(dir
))
1714 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, mode
);
1715 err
= PTR_ERR(inode
);
1716 if (!IS_ERR(inode
)) {
1717 inode
->i_op
= &ext3_file_inode_operations
;
1718 inode
->i_fop
= &ext3_file_operations
;
1719 ext3_set_aops(inode
);
1720 err
= ext3_add_nondir(handle
, dentry
, inode
);
1722 ext3_journal_stop(handle
);
1723 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1728 static int ext3_mknod (struct inode
* dir
, struct dentry
*dentry
,
1729 umode_t mode
, dev_t rdev
)
1732 struct inode
*inode
;
1733 int err
, retries
= 0;
1735 if (!new_valid_dev(rdev
))
1738 dquot_initialize(dir
);
1741 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1742 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1743 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1745 return PTR_ERR(handle
);
1747 if (IS_DIRSYNC(dir
))
1750 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, mode
);
1751 err
= PTR_ERR(inode
);
1752 if (!IS_ERR(inode
)) {
1753 init_special_inode(inode
, inode
->i_mode
, rdev
);
1754 #ifdef CONFIG_EXT3_FS_XATTR
1755 inode
->i_op
= &ext3_special_inode_operations
;
1757 err
= ext3_add_nondir(handle
, dentry
, inode
);
1759 ext3_journal_stop(handle
);
1760 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1765 static int ext3_mkdir(struct inode
* dir
, struct dentry
* dentry
, umode_t mode
)
1768 struct inode
* inode
;
1769 struct buffer_head
* dir_block
= NULL
;
1770 struct ext3_dir_entry_2
* de
;
1771 int err
, retries
= 0;
1773 if (dir
->i_nlink
>= EXT3_LINK_MAX
)
1776 dquot_initialize(dir
);
1779 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1780 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1781 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1783 return PTR_ERR(handle
);
1785 if (IS_DIRSYNC(dir
))
1788 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, S_IFDIR
| mode
);
1789 err
= PTR_ERR(inode
);
1793 inode
->i_op
= &ext3_dir_inode_operations
;
1794 inode
->i_fop
= &ext3_dir_operations
;
1795 inode
->i_size
= EXT3_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1796 if (!(dir_block
= ext3_dir_bread(handle
, inode
, 0, 1, &err
)))
1797 goto out_clear_inode
;
1799 BUFFER_TRACE(dir_block
, "get_write_access");
1800 err
= ext3_journal_get_write_access(handle
, dir_block
);
1802 goto out_clear_inode
;
1804 de
= (struct ext3_dir_entry_2
*) dir_block
->b_data
;
1805 de
->inode
= cpu_to_le32(inode
->i_ino
);
1807 de
->rec_len
= ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de
->name_len
));
1808 strcpy (de
->name
, ".");
1809 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1810 de
= ext3_next_entry(de
);
1811 de
->inode
= cpu_to_le32(dir
->i_ino
);
1812 de
->rec_len
= ext3_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1813 EXT3_DIR_REC_LEN(1));
1815 strcpy (de
->name
, "..");
1816 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1817 set_nlink(inode
, 2);
1818 BUFFER_TRACE(dir_block
, "call ext3_journal_dirty_metadata");
1819 err
= ext3_journal_dirty_metadata(handle
, dir_block
);
1821 goto out_clear_inode
;
1823 err
= ext3_mark_inode_dirty(handle
, inode
);
1825 err
= ext3_add_entry (handle
, dentry
, inode
);
1830 unlock_new_inode(inode
);
1831 ext3_mark_inode_dirty(handle
, inode
);
1836 ext3_update_dx_flag(dir
);
1837 err
= ext3_mark_inode_dirty(handle
, dir
);
1839 goto out_clear_inode
;
1841 unlock_new_inode(inode
);
1842 d_instantiate(dentry
, inode
);
1845 ext3_journal_stop(handle
);
1846 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1852 * routine to check that the specified directory is empty (for rmdir)
1854 static int empty_dir (struct inode
* inode
)
1856 unsigned long offset
;
1857 struct buffer_head
* bh
;
1858 struct ext3_dir_entry_2
* de
, * de1
;
1859 struct super_block
* sb
;
1863 if (inode
->i_size
< EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1864 !(bh
= ext3_dir_bread(NULL
, inode
, 0, 0, &err
))) {
1866 ext3_error(inode
->i_sb
, __func__
,
1867 "error %d reading directory #%lu offset 0",
1870 ext3_warning(inode
->i_sb
, __func__
,
1871 "bad directory (dir #%lu) - no data block",
1875 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1876 de1
= ext3_next_entry(de
);
1877 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1878 !le32_to_cpu(de1
->inode
) ||
1879 strcmp (".", de
->name
) ||
1880 strcmp ("..", de1
->name
)) {
1881 ext3_warning (inode
->i_sb
, "empty_dir",
1882 "bad directory (dir #%lu) - no `.' or `..'",
1887 offset
= ext3_rec_len_from_disk(de
->rec_len
) +
1888 ext3_rec_len_from_disk(de1
->rec_len
);
1889 de
= ext3_next_entry(de1
);
1890 while (offset
< inode
->i_size
) {
1892 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1895 if (!(bh
= ext3_dir_bread (NULL
, inode
,
1896 offset
>> EXT3_BLOCK_SIZE_BITS(sb
), 0, &err
))) {
1898 ext3_error(sb
, __func__
,
1899 "error %d reading directory"
1901 err
, inode
->i_ino
, offset
);
1902 offset
+= sb
->s_blocksize
;
1905 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1907 if (!ext3_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1908 de
= (struct ext3_dir_entry_2
*)(bh
->b_data
+
1910 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1913 if (le32_to_cpu(de
->inode
)) {
1917 offset
+= ext3_rec_len_from_disk(de
->rec_len
);
1918 de
= ext3_next_entry(de
);
1924 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1925 * such inodes, starting at the superblock, in case we crash before the
1926 * file is closed/deleted, or in case the inode truncate spans multiple
1927 * transactions and the last transaction is not recovered after a crash.
1929 * At filesystem recovery time, we walk this list deleting unlinked
1930 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1932 int ext3_orphan_add(handle_t
*handle
, struct inode
*inode
)
1934 struct super_block
*sb
= inode
->i_sb
;
1935 struct ext3_iloc iloc
;
1938 mutex_lock(&EXT3_SB(sb
)->s_orphan_lock
);
1939 if (!list_empty(&EXT3_I(inode
)->i_orphan
))
1942 /* Orphan handling is only valid for files with data blocks
1943 * being truncated, or files being unlinked. */
1945 /* @@@ FIXME: Observation from aviro:
1946 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1947 * here (on s_orphan_lock), so race with ext3_link() which might bump
1948 * ->i_nlink. For, say it, character device. Not a regular file,
1949 * not a directory, not a symlink and ->i_nlink > 0.
1951 * tytso, 4/25/2009: I'm not sure how that could happen;
1952 * shouldn't the fs core protect us from these sort of
1953 * unlink()/link() races?
1955 J_ASSERT ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1956 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1958 BUFFER_TRACE(EXT3_SB(sb
)->s_sbh
, "get_write_access");
1959 err
= ext3_journal_get_write_access(handle
, EXT3_SB(sb
)->s_sbh
);
1963 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1967 /* Insert this inode at the head of the on-disk orphan list... */
1968 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT3_SB(sb
)->s_es
->s_last_orphan
);
1969 EXT3_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1970 err
= ext3_journal_dirty_metadata(handle
, EXT3_SB(sb
)->s_sbh
);
1971 rc
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
1975 /* Only add to the head of the in-memory list if all the
1976 * previous operations succeeded. If the orphan_add is going to
1977 * fail (possibly taking the journal offline), we can't risk
1978 * leaving the inode on the orphan list: stray orphan-list
1979 * entries can cause panics at unmount time.
1981 * This is safe: on error we're going to ignore the orphan list
1982 * anyway on the next recovery. */
1984 list_add(&EXT3_I(inode
)->i_orphan
, &EXT3_SB(sb
)->s_orphan
);
1986 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1987 jbd_debug(4, "orphan inode %lu will point to %d\n",
1988 inode
->i_ino
, NEXT_ORPHAN(inode
));
1990 mutex_unlock(&EXT3_SB(sb
)->s_orphan_lock
);
1991 ext3_std_error(inode
->i_sb
, err
);
1996 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1997 * of such inodes stored on disk, because it is finally being cleaned up.
1999 int ext3_orphan_del(handle_t
*handle
, struct inode
*inode
)
2001 struct list_head
*prev
;
2002 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2003 struct ext3_sb_info
*sbi
;
2004 unsigned long ino_next
;
2005 struct ext3_iloc iloc
;
2008 mutex_lock(&EXT3_SB(inode
->i_sb
)->s_orphan_lock
);
2009 if (list_empty(&ei
->i_orphan
))
2012 ino_next
= NEXT_ORPHAN(inode
);
2013 prev
= ei
->i_orphan
.prev
;
2014 sbi
= EXT3_SB(inode
->i_sb
);
2016 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2018 list_del_init(&ei
->i_orphan
);
2020 /* If we're on an error path, we may not have a valid
2021 * transaction handle with which to update the orphan list on
2022 * disk, but we still need to remove the inode from the linked
2023 * list in memory. */
2027 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
2031 if (prev
== &sbi
->s_orphan
) {
2032 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
2033 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2034 err
= ext3_journal_get_write_access(handle
, sbi
->s_sbh
);
2037 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2038 err
= ext3_journal_dirty_metadata(handle
, sbi
->s_sbh
);
2040 struct ext3_iloc iloc2
;
2041 struct inode
*i_prev
=
2042 &list_entry(prev
, struct ext3_inode_info
, i_orphan
)->vfs_inode
;
2044 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2045 i_prev
->i_ino
, ino_next
);
2046 err
= ext3_reserve_inode_write(handle
, i_prev
, &iloc2
);
2049 NEXT_ORPHAN(i_prev
) = ino_next
;
2050 err
= ext3_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2054 NEXT_ORPHAN(inode
) = 0;
2055 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
2058 ext3_std_error(inode
->i_sb
, err
);
2060 mutex_unlock(&EXT3_SB(inode
->i_sb
)->s_orphan_lock
);
2068 static int ext3_rmdir (struct inode
* dir
, struct dentry
*dentry
)
2071 struct inode
* inode
;
2072 struct buffer_head
* bh
;
2073 struct ext3_dir_entry_2
* de
;
2076 /* Initialize quotas before so that eventual writes go in
2077 * separate transaction */
2078 dquot_initialize(dir
);
2079 dquot_initialize(dentry
->d_inode
);
2081 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2083 return PTR_ERR(handle
);
2086 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2090 if (IS_DIRSYNC(dir
))
2093 inode
= dentry
->d_inode
;
2096 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2099 retval
= -ENOTEMPTY
;
2100 if (!empty_dir (inode
))
2103 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2106 if (inode
->i_nlink
!= 2)
2107 ext3_warning (inode
->i_sb
, "ext3_rmdir",
2108 "empty directory has nlink!=2 (%d)",
2112 /* There's no need to set i_disksize: the fact that i_nlink is
2113 * zero will ensure that the right thing happens during any
2116 ext3_orphan_add(handle
, inode
);
2117 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2118 ext3_mark_inode_dirty(handle
, inode
);
2120 ext3_update_dx_flag(dir
);
2121 ext3_mark_inode_dirty(handle
, dir
);
2124 ext3_journal_stop(handle
);
2129 static int ext3_unlink(struct inode
* dir
, struct dentry
*dentry
)
2132 struct inode
* inode
;
2133 struct buffer_head
* bh
;
2134 struct ext3_dir_entry_2
* de
;
2137 trace_ext3_unlink_enter(dir
, dentry
);
2138 /* Initialize quotas before so that eventual writes go
2139 * in separate transaction */
2140 dquot_initialize(dir
);
2141 dquot_initialize(dentry
->d_inode
);
2143 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2145 return PTR_ERR(handle
);
2147 if (IS_DIRSYNC(dir
))
2151 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2155 inode
= dentry
->d_inode
;
2158 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2161 if (!inode
->i_nlink
) {
2162 ext3_warning (inode
->i_sb
, "ext3_unlink",
2163 "Deleting nonexistent file (%lu), %d",
2164 inode
->i_ino
, inode
->i_nlink
);
2165 set_nlink(inode
, 1);
2167 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2170 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2171 ext3_update_dx_flag(dir
);
2172 ext3_mark_inode_dirty(handle
, dir
);
2174 if (!inode
->i_nlink
)
2175 ext3_orphan_add(handle
, inode
);
2176 inode
->i_ctime
= dir
->i_ctime
;
2177 ext3_mark_inode_dirty(handle
, inode
);
2181 ext3_journal_stop(handle
);
2183 trace_ext3_unlink_exit(dentry
, retval
);
2187 static int ext3_symlink (struct inode
* dir
,
2188 struct dentry
*dentry
, const char * symname
)
2191 struct inode
* inode
;
2192 int l
, err
, retries
= 0;
2195 l
= strlen(symname
)+1;
2196 if (l
> dir
->i_sb
->s_blocksize
)
2197 return -ENAMETOOLONG
;
2199 dquot_initialize(dir
);
2201 if (l
> EXT3_N_BLOCKS
* 4) {
2203 * For non-fast symlinks, we just allocate inode and put it on
2204 * orphan list in the first transaction => we need bitmap,
2205 * group descriptor, sb, inode block, quota blocks, and
2206 * possibly selinux xattr blocks.
2208 credits
= 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2209 EXT3_XATTR_TRANS_BLOCKS
;
2212 * Fast symlink. We have to add entry to directory
2213 * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2214 * allocate new inode (bitmap, group descriptor, inode block,
2215 * quota blocks, sb is already counted in previous macros).
2217 credits
= EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2218 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
2219 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
);
2222 handle
= ext3_journal_start(dir
, credits
);
2224 return PTR_ERR(handle
);
2226 if (IS_DIRSYNC(dir
))
2229 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, S_IFLNK
|S_IRWXUGO
);
2230 err
= PTR_ERR(inode
);
2234 if (l
> EXT3_N_BLOCKS
* 4) {
2235 inode
->i_op
= &ext3_symlink_inode_operations
;
2236 ext3_set_aops(inode
);
2238 * We cannot call page_symlink() with transaction started
2239 * because it calls into ext3_write_begin() which acquires page
2240 * lock which ranks below transaction start (and it can also
2241 * wait for journal commit if we are running out of space). So
2242 * we have to stop transaction now and restart it when symlink
2243 * contents is written.
2245 * To keep fs consistent in case of crash, we have to put inode
2246 * to orphan list in the mean time.
2249 err
= ext3_orphan_add(handle
, inode
);
2250 ext3_journal_stop(handle
);
2252 goto err_drop_inode
;
2253 err
= __page_symlink(inode
, symname
, l
, 1);
2255 goto err_drop_inode
;
2257 * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2258 * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2260 handle
= ext3_journal_start(dir
,
2261 EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2262 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
2263 if (IS_ERR(handle
)) {
2264 err
= PTR_ERR(handle
);
2265 goto err_drop_inode
;
2267 set_nlink(inode
, 1);
2268 err
= ext3_orphan_del(handle
, inode
);
2270 ext3_journal_stop(handle
);
2272 goto err_drop_inode
;
2275 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2276 memcpy((char*)&EXT3_I(inode
)->i_data
,symname
,l
);
2277 inode
->i_size
= l
-1;
2279 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
2280 err
= ext3_add_nondir(handle
, dentry
, inode
);
2282 ext3_journal_stop(handle
);
2283 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2287 unlock_new_inode(inode
);
2292 static int ext3_link (struct dentry
* old_dentry
,
2293 struct inode
* dir
, struct dentry
*dentry
)
2296 struct inode
*inode
= old_dentry
->d_inode
;
2297 int err
, retries
= 0;
2299 if (inode
->i_nlink
>= EXT3_LINK_MAX
)
2302 dquot_initialize(dir
);
2305 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2306 EXT3_INDEX_EXTRA_TRANS_BLOCKS
);
2308 return PTR_ERR(handle
);
2310 if (IS_DIRSYNC(dir
))
2313 inode
->i_ctime
= CURRENT_TIME_SEC
;
2317 err
= ext3_add_entry(handle
, dentry
, inode
);
2319 ext3_mark_inode_dirty(handle
, inode
);
2320 d_instantiate(dentry
, inode
);
2325 ext3_journal_stop(handle
);
2326 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2331 #define PARENT_INO(buffer) \
2332 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2335 * Anybody can rename anything with this: the permission checks are left to the
2336 * higher-level routines.
2338 static int ext3_rename (struct inode
* old_dir
, struct dentry
*old_dentry
,
2339 struct inode
* new_dir
,struct dentry
*new_dentry
)
2342 struct inode
* old_inode
, * new_inode
;
2343 struct buffer_head
* old_bh
, * new_bh
, * dir_bh
;
2344 struct ext3_dir_entry_2
* old_de
, * new_de
;
2345 int retval
, flush_file
= 0;
2347 dquot_initialize(old_dir
);
2348 dquot_initialize(new_dir
);
2350 old_bh
= new_bh
= dir_bh
= NULL
;
2352 /* Initialize quotas before so that eventual writes go
2353 * in separate transaction */
2354 if (new_dentry
->d_inode
)
2355 dquot_initialize(new_dentry
->d_inode
);
2356 handle
= ext3_journal_start(old_dir
, 2 *
2357 EXT3_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2358 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2360 return PTR_ERR(handle
);
2362 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2365 old_bh
= ext3_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2367 * Check for inode number is _not_ due to possible IO errors.
2368 * We might rmdir the source, keep it as pwd of some process
2369 * and merrily kill the link to whatever was created under the
2370 * same name. Goodbye sticky bit ;-<
2372 old_inode
= old_dentry
->d_inode
;
2374 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2377 new_inode
= new_dentry
->d_inode
;
2378 new_bh
= ext3_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
2385 if (S_ISDIR(old_inode
->i_mode
)) {
2387 retval
= -ENOTEMPTY
;
2388 if (!empty_dir (new_inode
))
2392 dir_bh
= ext3_dir_bread(handle
, old_inode
, 0, 0, &retval
);
2395 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2398 if (!new_inode
&& new_dir
!=old_dir
&&
2399 new_dir
->i_nlink
>= EXT3_LINK_MAX
)
2403 retval
= ext3_add_entry (handle
, new_dentry
, old_inode
);
2407 BUFFER_TRACE(new_bh
, "get write access");
2408 retval
= ext3_journal_get_write_access(handle
, new_bh
);
2411 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2412 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2413 EXT3_FEATURE_INCOMPAT_FILETYPE
))
2414 new_de
->file_type
= old_de
->file_type
;
2415 new_dir
->i_version
++;
2416 new_dir
->i_ctime
= new_dir
->i_mtime
= CURRENT_TIME_SEC
;
2417 ext3_mark_inode_dirty(handle
, new_dir
);
2418 BUFFER_TRACE(new_bh
, "call ext3_journal_dirty_metadata");
2419 retval
= ext3_journal_dirty_metadata(handle
, new_bh
);
2427 * Like most other Unix systems, set the ctime for inodes on a
2430 old_inode
->i_ctime
= CURRENT_TIME_SEC
;
2431 ext3_mark_inode_dirty(handle
, old_inode
);
2436 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2437 old_de
->name_len
!= old_dentry
->d_name
.len
||
2438 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2439 (retval
= ext3_delete_entry(handle
, old_dir
,
2440 old_de
, old_bh
)) == -ENOENT
) {
2441 /* old_de could have moved from under us during htree split, so
2442 * make sure that we are deleting the right entry. We might
2443 * also be pointing to a stale entry in the unused part of
2444 * old_bh so just checking inum and the name isn't enough. */
2445 struct buffer_head
*old_bh2
;
2446 struct ext3_dir_entry_2
*old_de2
;
2448 old_bh2
= ext3_find_entry(old_dir
, &old_dentry
->d_name
,
2451 retval
= ext3_delete_entry(handle
, old_dir
,
2457 ext3_warning(old_dir
->i_sb
, "ext3_rename",
2458 "Deleting old file (%lu), %d, error=%d",
2459 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2463 drop_nlink(new_inode
);
2464 new_inode
->i_ctime
= CURRENT_TIME_SEC
;
2466 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME_SEC
;
2467 ext3_update_dx_flag(old_dir
);
2469 BUFFER_TRACE(dir_bh
, "get_write_access");
2470 retval
= ext3_journal_get_write_access(handle
, dir_bh
);
2473 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2474 BUFFER_TRACE(dir_bh
, "call ext3_journal_dirty_metadata");
2475 retval
= ext3_journal_dirty_metadata(handle
, dir_bh
);
2478 ext3_std_error(new_dir
->i_sb
, retval
);
2481 drop_nlink(old_dir
);
2483 drop_nlink(new_inode
);
2486 ext3_update_dx_flag(new_dir
);
2487 ext3_mark_inode_dirty(handle
, new_dir
);
2490 ext3_mark_inode_dirty(handle
, old_dir
);
2492 ext3_mark_inode_dirty(handle
, new_inode
);
2493 if (!new_inode
->i_nlink
)
2494 ext3_orphan_add(handle
, new_inode
);
2495 if (ext3_should_writeback_data(new_inode
))
2504 ext3_journal_stop(handle
);
2505 if (retval
== 0 && flush_file
)
2506 filemap_flush(old_inode
->i_mapping
);
2511 * directories can handle most operations...
2513 const struct inode_operations ext3_dir_inode_operations
= {
2514 .create
= ext3_create
,
2515 .lookup
= ext3_lookup
,
2517 .unlink
= ext3_unlink
,
2518 .symlink
= ext3_symlink
,
2519 .mkdir
= ext3_mkdir
,
2520 .rmdir
= ext3_rmdir
,
2521 .mknod
= ext3_mknod
,
2522 .rename
= ext3_rename
,
2523 .setattr
= ext3_setattr
,
2524 #ifdef CONFIG_EXT3_FS_XATTR
2525 .setxattr
= generic_setxattr
,
2526 .getxattr
= generic_getxattr
,
2527 .listxattr
= ext3_listxattr
,
2528 .removexattr
= generic_removexattr
,
2530 .get_acl
= ext3_get_acl
,
2533 const struct inode_operations ext3_special_inode_operations
= {
2534 .setattr
= ext3_setattr
,
2535 #ifdef CONFIG_EXT3_FS_XATTR
2536 .setxattr
= generic_setxattr
,
2537 .getxattr
= generic_getxattr
,
2538 .listxattr
= ext3_listxattr
,
2539 .removexattr
= generic_removexattr
,
2541 .get_acl
= ext3_get_acl
,