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
, unsigned blocksize
,
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
, unsigned blocksize
);
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 if (hinfo
->hash_version
<= DX_HASH_TEA
)
368 hinfo
->hash_version
+= EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
369 hinfo
->seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
371 ext3fs_dirhash(entry
->name
, entry
->len
, hinfo
);
374 if (root
->info
.unused_flags
& 1) {
375 ext3_warning(dir
->i_sb
, __func__
,
376 "Unimplemented inode hash flags: %#06x",
377 root
->info
.unused_flags
);
379 *err
= ERR_BAD_DX_DIR
;
383 if ((indirect
= root
->info
.indirect_levels
) > 1) {
384 ext3_warning(dir
->i_sb
, __func__
,
385 "Unimplemented inode hash depth: %#06x",
386 root
->info
.indirect_levels
);
388 *err
= ERR_BAD_DX_DIR
;
392 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
393 root
->info
.info_length
);
395 if (dx_get_limit(entries
) != dx_root_limit(dir
,
396 root
->info
.info_length
)) {
397 ext3_warning(dir
->i_sb
, __func__
,
398 "dx entry: limit != root limit");
400 *err
= ERR_BAD_DX_DIR
;
404 dxtrace (printk("Look up %x", hash
));
407 count
= dx_get_count(entries
);
408 if (!count
|| count
> dx_get_limit(entries
)) {
409 ext3_warning(dir
->i_sb
, __func__
,
410 "dx entry: no count or count > limit");
412 *err
= ERR_BAD_DX_DIR
;
417 q
= entries
+ count
- 1;
421 dxtrace(printk("."));
422 if (dx_get_hash(m
) > hash
)
428 if (0) // linear search cross check
430 unsigned n
= count
- 1;
434 dxtrace(printk(","));
435 if (dx_get_hash(++at
) > hash
)
441 assert (at
== p
- 1);
445 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
447 frame
->entries
= entries
;
449 if (!indirect
--) return frame
;
450 if (!(bh
= ext3_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
452 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
453 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
454 ext3_warning(dir
->i_sb
, __func__
,
455 "dx entry: limit != node limit");
457 *err
= ERR_BAD_DX_DIR
;
464 while (frame
>= frame_in
) {
469 if (*err
== ERR_BAD_DX_DIR
)
470 ext3_warning(dir
->i_sb
, __func__
,
471 "Corrupt dir inode %ld, running e2fsck is "
472 "recommended.", dir
->i_ino
);
476 static void dx_release (struct dx_frame
*frames
)
478 if (frames
[0].bh
== NULL
)
481 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
482 brelse(frames
[1].bh
);
483 brelse(frames
[0].bh
);
487 * This function increments the frame pointer to search the next leaf
488 * block, and reads in the necessary intervening nodes if the search
489 * should be necessary. Whether or not the search is necessary is
490 * controlled by the hash parameter. If the hash value is even, then
491 * the search is only continued if the next block starts with that
492 * hash value. This is used if we are searching for a specific file.
494 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
496 * This function returns 1 if the caller should continue to search,
497 * or 0 if it should not. If there is an error reading one of the
498 * index blocks, it will a negative error code.
500 * If start_hash is non-null, it will be filled in with the starting
501 * hash of the next page.
503 static int ext3_htree_next_block(struct inode
*dir
, __u32 hash
,
504 struct dx_frame
*frame
,
505 struct dx_frame
*frames
,
509 struct buffer_head
*bh
;
510 int err
, num_frames
= 0;
515 * Find the next leaf page by incrementing the frame pointer.
516 * If we run out of entries in the interior node, loop around and
517 * increment pointer in the parent node. When we break out of
518 * this loop, num_frames indicates the number of interior
519 * nodes need to be read.
522 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
531 * If the hash is 1, then continue only if the next page has a
532 * continuation hash of any value. This is used for readdir
533 * handling. Otherwise, check to see if the hash matches the
534 * desired contiuation hash. If it doesn't, return since
535 * there's no point to read in the successive index pages.
537 bhash
= dx_get_hash(p
->at
);
540 if ((hash
& 1) == 0) {
541 if ((bhash
& ~1) != hash
)
545 * If the hash is HASH_NB_ALWAYS, we always go to the next
546 * block so no check is necessary
548 while (num_frames
--) {
549 if (!(bh
= ext3_bread(NULL
, dir
, dx_get_block(p
->at
),
551 return err
; /* Failure */
555 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
562 * This function fills a red-black tree with information from a
563 * directory block. It returns the number directory entries loaded
564 * into the tree. If there is an error it is returned in err.
566 static int htree_dirblock_to_tree(struct file
*dir_file
,
567 struct inode
*dir
, int block
,
568 struct dx_hash_info
*hinfo
,
569 __u32 start_hash
, __u32 start_minor_hash
)
571 struct buffer_head
*bh
;
572 struct ext3_dir_entry_2
*de
, *top
;
575 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block
));
576 if (!(bh
= ext3_bread (NULL
, dir
, block
, 0, &err
)))
579 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
580 top
= (struct ext3_dir_entry_2
*) ((char *) de
+
581 dir
->i_sb
->s_blocksize
-
582 EXT3_DIR_REC_LEN(0));
583 for (; de
< top
; de
= ext3_next_entry(de
)) {
584 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
585 (block
<<EXT3_BLOCK_SIZE_BITS(dir
->i_sb
))
586 +((char *)de
- bh
->b_data
))) {
587 /* On error, skip the f_pos to the next block. */
588 dir_file
->f_pos
= (dir_file
->f_pos
|
589 (dir
->i_sb
->s_blocksize
- 1)) + 1;
593 ext3fs_dirhash(de
->name
, de
->name_len
, hinfo
);
594 if ((hinfo
->hash
< start_hash
) ||
595 ((hinfo
->hash
== start_hash
) &&
596 (hinfo
->minor_hash
< start_minor_hash
)))
600 if ((err
= ext3_htree_store_dirent(dir_file
,
601 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
613 * This function fills a red-black tree with information from a
614 * directory. We start scanning the directory in hash order, starting
615 * at start_hash and start_minor_hash.
617 * This function returns the number of entries inserted into the tree,
618 * or a negative error code.
620 int ext3_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
621 __u32 start_minor_hash
, __u32
*next_hash
)
623 struct dx_hash_info hinfo
;
624 struct ext3_dir_entry_2
*de
;
625 struct dx_frame frames
[2], *frame
;
632 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash
,
634 dir
= dir_file
->f_path
.dentry
->d_inode
;
635 if (!(EXT3_I(dir
)->i_flags
& EXT3_INDEX_FL
)) {
636 hinfo
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
637 if (hinfo
.hash_version
<= DX_HASH_TEA
)
638 hinfo
.hash_version
+=
639 EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
640 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
641 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
642 start_hash
, start_minor_hash
);
646 hinfo
.hash
= start_hash
;
647 hinfo
.minor_hash
= 0;
648 frame
= dx_probe(NULL
, dir_file
->f_path
.dentry
->d_inode
, &hinfo
, frames
, &err
);
652 /* Add '.' and '..' from the htree header */
653 if (!start_hash
&& !start_minor_hash
) {
654 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
655 if ((err
= ext3_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
659 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
660 de
= (struct ext3_dir_entry_2
*) frames
[0].bh
->b_data
;
661 de
= ext3_next_entry(de
);
662 if ((err
= ext3_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
668 block
= dx_get_block(frame
->at
);
669 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
670 start_hash
, start_minor_hash
);
677 ret
= ext3_htree_next_block(dir
, HASH_NB_ALWAYS
,
678 frame
, frames
, &hashval
);
679 *next_hash
= hashval
;
685 * Stop if: (a) there are no more entries, or
686 * (b) we have inserted at least one entry and the
687 * next hash value is not a continuation
690 (count
&& ((hashval
& 1) == 0)))
694 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
704 * Directory block splitting, compacting
708 * Create map of hash values, offsets, and sizes, stored at end of block.
709 * Returns number of entries mapped.
711 static int dx_make_map(struct ext3_dir_entry_2
*de
, unsigned blocksize
,
712 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
715 char *base
= (char *) de
;
716 struct dx_hash_info h
= *hinfo
;
718 while ((char *) de
< base
+ blocksize
)
720 if (de
->name_len
&& de
->inode
) {
721 ext3fs_dirhash(de
->name
, de
->name_len
, &h
);
723 map_tail
->hash
= h
.hash
;
724 map_tail
->offs
= (u16
) ((char *) de
- base
);
725 map_tail
->size
= le16_to_cpu(de
->rec_len
);
729 /* XXX: do we need to check rec_len == 0 case? -Chris */
730 de
= ext3_next_entry(de
);
735 /* Sort map by hash value */
736 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
738 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
740 /* Combsort until bubble sort doesn't suck */
744 if (count
- 9 < 2) /* 9, 10 -> 11 */
746 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
747 if (p
->hash
< q
->hash
)
750 /* Garden variety bubble sort */
756 if (q
[1].hash
>= q
[0].hash
)
764 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, u32 block
)
766 struct dx_entry
*entries
= frame
->entries
;
767 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
768 int count
= dx_get_count(entries
);
770 assert(count
< dx_get_limit(entries
));
771 assert(old
< entries
+ count
);
772 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
773 dx_set_hash(new, hash
);
774 dx_set_block(new, block
);
775 dx_set_count(entries
, count
+ 1);
778 static void ext3_update_dx_flag(struct inode
*inode
)
780 if (!EXT3_HAS_COMPAT_FEATURE(inode
->i_sb
,
781 EXT3_FEATURE_COMPAT_DIR_INDEX
))
782 EXT3_I(inode
)->i_flags
&= ~EXT3_INDEX_FL
;
786 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
788 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
789 * `de != NULL' is guaranteed by caller.
791 static inline int ext3_match (int len
, const char * const name
,
792 struct ext3_dir_entry_2
* de
)
794 if (len
!= de
->name_len
)
798 return !memcmp(name
, de
->name
, len
);
802 * Returns 0 if not found, -1 on failure, and 1 on success
804 static inline int search_dirblock(struct buffer_head
* bh
,
807 unsigned long offset
,
808 struct ext3_dir_entry_2
** res_dir
)
810 struct ext3_dir_entry_2
* de
;
813 const char *name
= child
->name
;
814 int namelen
= child
->len
;
816 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
817 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
818 while ((char *) de
< dlimit
) {
819 /* this code is executed quadratically often */
820 /* do minimal checking `by hand' */
822 if ((char *) de
+ namelen
<= dlimit
&&
823 ext3_match (namelen
, name
, de
)) {
824 /* found a match - just to be sure, do a full check */
825 if (!ext3_check_dir_entry("ext3_find_entry",
826 dir
, de
, bh
, offset
))
831 /* prevent looping on a bad block */
832 de_len
= ext3_rec_len_from_disk(de
->rec_len
);
836 de
= (struct ext3_dir_entry_2
*) ((char *) de
+ de_len
);
845 * finds an entry in the specified directory with the wanted name. It
846 * returns the cache buffer in which the entry was found, and the entry
847 * itself (as a parameter - res_dir). It does NOT read the inode of the
848 * entry - you'll have to do that yourself if you want to.
850 * The returned buffer_head has ->b_count elevated. The caller is expected
851 * to brelse() it when appropriate.
853 static struct buffer_head
*ext3_find_entry(struct inode
*dir
,
855 struct ext3_dir_entry_2
**res_dir
)
857 struct super_block
* sb
;
858 struct buffer_head
* bh_use
[NAMEI_RA_SIZE
];
859 struct buffer_head
* bh
, *ret
= NULL
;
860 unsigned long start
, block
, b
;
861 const u8
*name
= entry
->name
;
862 int ra_max
= 0; /* Number of bh's in the readahead
864 int ra_ptr
= 0; /* Current index into readahead
872 namelen
= entry
->len
;
873 if (namelen
> EXT3_NAME_LEN
)
875 if ((namelen
<= 2) && (name
[0] == '.') &&
876 (name
[1] == '.' || name
[1] == 0)) {
878 * "." or ".." will only be in the first block
879 * NFS may look up ".."; "." should be handled by the VFS
886 bh
= ext3_dx_find_entry(dir
, entry
, res_dir
, &err
);
888 * On success, or if the error was file not found,
889 * return. Otherwise, fall back to doing a search the
892 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
894 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
896 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
897 start
= EXT3_I(dir
)->i_dir_start_lookup
;
898 if (start
>= nblocks
)
904 * We deal with the read-ahead logic here.
906 if (ra_ptr
>= ra_max
) {
907 /* Refill the readahead buffer */
910 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
912 * Terminate if we reach the end of the
913 * directory and must wrap, or if our
914 * search has finished at this block.
916 if (b
>= nblocks
|| (num
&& block
== start
)) {
917 bh_use
[ra_max
] = NULL
;
921 bh
= ext3_getblk(NULL
, dir
, b
++, 0, &err
);
924 ll_rw_block(READ_META
, 1, &bh
);
927 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
930 if (!buffer_uptodate(bh
)) {
931 /* read error, skip block & hope for the best */
932 ext3_error(sb
, __func__
, "reading directory #%lu "
933 "offset %lu", dir
->i_ino
, block
);
937 i
= search_dirblock(bh
, dir
, entry
,
938 block
<< EXT3_BLOCK_SIZE_BITS(sb
), res_dir
);
940 EXT3_I(dir
)->i_dir_start_lookup
= block
;
942 goto cleanup_and_exit
;
946 goto cleanup_and_exit
;
949 if (++block
>= nblocks
)
951 } while (block
!= start
);
954 * If the directory has grown while we were searching, then
955 * search the last part of the directory before giving up.
958 nblocks
= dir
->i_size
>> EXT3_BLOCK_SIZE_BITS(sb
);
959 if (block
< nblocks
) {
965 /* Clean up the read-ahead blocks */
966 for (; ra_ptr
< ra_max
; ra_ptr
++)
967 brelse (bh_use
[ra_ptr
]);
971 static struct buffer_head
* ext3_dx_find_entry(struct inode
*dir
,
972 struct qstr
*entry
, struct ext3_dir_entry_2
**res_dir
,
975 struct super_block
*sb
= dir
->i_sb
;
976 struct dx_hash_info hinfo
;
977 struct dx_frame frames
[2], *frame
;
978 struct buffer_head
*bh
;
982 if (!(frame
= dx_probe(entry
, dir
, &hinfo
, frames
, err
)))
985 block
= dx_get_block(frame
->at
);
986 if (!(bh
= ext3_bread (NULL
,dir
, block
, 0, err
)))
989 retval
= search_dirblock(bh
, dir
, entry
,
990 block
<< EXT3_BLOCK_SIZE_BITS(sb
),
998 *err
= ERR_BAD_DX_DIR
;
1002 /* Check to see if we should continue to search */
1003 retval
= ext3_htree_next_block(dir
, hinfo
.hash
, frame
,
1006 ext3_warning(sb
, __func__
,
1007 "error reading index page in directory #%lu",
1012 } while (retval
== 1);
1016 dxtrace(printk("%s not found\n", name
));
1017 dx_release (frames
);
1021 static struct dentry
*ext3_lookup(struct inode
* dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1023 struct inode
* inode
;
1024 struct ext3_dir_entry_2
* de
;
1025 struct buffer_head
* bh
;
1027 if (dentry
->d_name
.len
> EXT3_NAME_LEN
)
1028 return ERR_PTR(-ENAMETOOLONG
);
1030 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
1033 unsigned long ino
= le32_to_cpu(de
->inode
);
1035 if (!ext3_valid_inum(dir
->i_sb
, ino
)) {
1036 ext3_error(dir
->i_sb
, "ext3_lookup",
1037 "bad inode number: %lu", ino
);
1038 return ERR_PTR(-EIO
);
1040 inode
= ext3_iget(dir
->i_sb
, ino
);
1041 if (inode
== ERR_PTR(-ESTALE
)) {
1042 ext3_error(dir
->i_sb
, __func__
,
1043 "deleted inode referenced: %lu",
1045 return ERR_PTR(-EIO
);
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
, unsigned blocksize
)
1120 struct ext3_dir_entry_2
*next
, *to
, *prev
;
1121 struct ext3_dir_entry_2
*de
= (struct ext3_dir_entry_2
*)base
;
1122 unsigned rec_len
= 0;
1125 while ((char *)de
< base
+ blocksize
) {
1126 next
= ext3_next_entry(de
);
1127 if (de
->inode
&& de
->name_len
) {
1128 rec_len
= EXT3_DIR_REC_LEN(de
->name_len
);
1130 memmove(to
, de
, rec_len
);
1131 to
->rec_len
= ext3_rec_len_to_disk(rec_len
);
1133 to
= (struct ext3_dir_entry_2
*) (((char *) to
) + rec_len
);
1141 * Split a full leaf block to make room for a new dir entry.
1142 * Allocate a new block, and move entries so that they are approx. equally full.
1143 * Returns pointer to de in block into which the new entry will be inserted.
1145 static struct ext3_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1146 struct buffer_head
**bh
,struct dx_frame
*frame
,
1147 struct dx_hash_info
*hinfo
, int *error
)
1149 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1150 unsigned count
, continued
;
1151 struct buffer_head
*bh2
;
1154 struct dx_map_entry
*map
;
1155 char *data1
= (*bh
)->b_data
, *data2
;
1156 unsigned split
, move
, size
;
1157 struct ext3_dir_entry_2
*de
= NULL
, *de2
;
1160 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1167 BUFFER_TRACE(*bh
, "get_write_access");
1168 err
= ext3_journal_get_write_access(handle
, *bh
);
1172 BUFFER_TRACE(frame
->bh
, "get_write_access");
1173 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1177 data2
= bh2
->b_data
;
1179 /* create map in the end of data2 block */
1180 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1181 count
= dx_make_map ((struct ext3_dir_entry_2
*) data1
,
1182 blocksize
, hinfo
, map
);
1184 dx_sort_map (map
, count
);
1185 /* Split the existing block in the middle, size-wise */
1188 for (i
= count
-1; i
>= 0; i
--) {
1189 /* is more than half of this entry in 2nd half of the block? */
1190 if (size
+ map
[i
].size
/2 > blocksize
/2)
1192 size
+= map
[i
].size
;
1195 /* map index at which we will split */
1196 split
= count
- move
;
1197 hash2
= map
[split
].hash
;
1198 continued
= hash2
== map
[split
- 1].hash
;
1199 dxtrace(printk("Split block %i at %x, %i/%i\n",
1200 dx_get_block(frame
->at
), hash2
, split
, count
-split
));
1202 /* Fancy dance to stay within two buffers */
1203 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1204 de
= dx_pack_dirents(data1
,blocksize
);
1205 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1206 de2
->rec_len
= ext3_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1207 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data1
, blocksize
, 1));
1208 dxtrace(dx_show_leaf (hinfo
, (struct ext3_dir_entry_2
*) data2
, blocksize
, 1));
1210 /* Which block gets the new entry? */
1211 if (hinfo
->hash
>= hash2
)
1216 dx_insert_block (frame
, hash2
+ continued
, newblock
);
1217 err
= ext3_journal_dirty_metadata (handle
, bh2
);
1220 err
= ext3_journal_dirty_metadata (handle
, frame
->bh
);
1224 dxtrace(dx_show_index ("frame", frame
->entries
));
1231 ext3_std_error(dir
->i_sb
, err
);
1239 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1240 * it points to a directory entry which is guaranteed to be large
1241 * enough for new directory entry. If de is NULL, then
1242 * add_dirent_to_buf will attempt search the directory block for
1243 * space. It will return -ENOSPC if no space is available, and -EIO
1244 * and -EEXIST if directory entry already exists.
1246 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1247 * all other cases bh is released.
1249 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1250 struct inode
*inode
, struct ext3_dir_entry_2
*de
,
1251 struct buffer_head
* bh
)
1253 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1254 const char *name
= dentry
->d_name
.name
;
1255 int namelen
= dentry
->d_name
.len
;
1256 unsigned long offset
= 0;
1257 unsigned short reclen
;
1258 int nlen
, rlen
, err
;
1261 reclen
= EXT3_DIR_REC_LEN(namelen
);
1263 de
= (struct ext3_dir_entry_2
*)bh
->b_data
;
1264 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1265 while ((char *) de
<= top
) {
1266 if (!ext3_check_dir_entry("ext3_add_entry", dir
, de
,
1271 if (ext3_match (namelen
, name
, de
)) {
1275 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1276 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1277 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1279 de
= (struct ext3_dir_entry_2
*)((char *)de
+ rlen
);
1282 if ((char *) de
> top
)
1285 BUFFER_TRACE(bh
, "get_write_access");
1286 err
= ext3_journal_get_write_access(handle
, bh
);
1288 ext3_std_error(dir
->i_sb
, err
);
1293 /* By now the buffer is marked for journaling */
1294 nlen
= EXT3_DIR_REC_LEN(de
->name_len
);
1295 rlen
= ext3_rec_len_from_disk(de
->rec_len
);
1297 struct ext3_dir_entry_2
*de1
= (struct ext3_dir_entry_2
*)((char *)de
+ nlen
);
1298 de1
->rec_len
= ext3_rec_len_to_disk(rlen
- nlen
);
1299 de
->rec_len
= ext3_rec_len_to_disk(nlen
);
1302 de
->file_type
= EXT3_FT_UNKNOWN
;
1304 de
->inode
= cpu_to_le32(inode
->i_ino
);
1305 ext3_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1308 de
->name_len
= namelen
;
1309 memcpy (de
->name
, name
, namelen
);
1311 * XXX shouldn't update any times until successful
1312 * completion of syscall, but too many callers depend
1315 * XXX similarly, too many callers depend on
1316 * ext3_new_inode() setting the times, but error
1317 * recovery deletes the inode, so the worst that can
1318 * happen is that the times are slightly out of date
1319 * and/or different from the directory change time.
1321 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME_SEC
;
1322 ext3_update_dx_flag(dir
);
1324 ext3_mark_inode_dirty(handle
, dir
);
1325 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1326 err
= ext3_journal_dirty_metadata(handle
, bh
);
1328 ext3_std_error(dir
->i_sb
, err
);
1334 * This converts a one block unindexed directory to a 3 block indexed
1335 * directory, and adds the dentry to the indexed directory.
1337 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1338 struct inode
*inode
, struct buffer_head
*bh
)
1340 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1341 const char *name
= dentry
->d_name
.name
;
1342 int namelen
= dentry
->d_name
.len
;
1343 struct buffer_head
*bh2
;
1344 struct dx_root
*root
;
1345 struct dx_frame frames
[2], *frame
;
1346 struct dx_entry
*entries
;
1347 struct ext3_dir_entry_2
*de
, *de2
;
1352 struct dx_hash_info hinfo
;
1354 struct fake_dirent
*fde
;
1356 blocksize
= dir
->i_sb
->s_blocksize
;
1357 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1358 retval
= ext3_journal_get_write_access(handle
, bh
);
1360 ext3_std_error(dir
->i_sb
, retval
);
1364 root
= (struct dx_root
*) bh
->b_data
;
1366 /* The 0th block becomes the root, move the dirents out */
1367 fde
= &root
->dotdot
;
1368 de
= (struct ext3_dir_entry_2
*)((char *)fde
+
1369 ext3_rec_len_from_disk(fde
->rec_len
));
1370 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1371 ext3_error(dir
->i_sb
, __func__
,
1372 "invalid rec_len for '..' in inode %lu",
1377 len
= ((char *) root
) + blocksize
- (char *) de
;
1379 bh2
= ext3_append (handle
, dir
, &block
, &retval
);
1384 EXT3_I(dir
)->i_flags
|= EXT3_INDEX_FL
;
1385 data1
= bh2
->b_data
;
1387 memcpy (data1
, de
, len
);
1388 de
= (struct ext3_dir_entry_2
*) data1
;
1390 while ((char *)(de2
= ext3_next_entry(de
)) < top
)
1392 de
->rec_len
= ext3_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1393 /* Initialize the root; the dot dirents already exist */
1394 de
= (struct ext3_dir_entry_2
*) (&root
->dotdot
);
1395 de
->rec_len
= ext3_rec_len_to_disk(blocksize
- EXT3_DIR_REC_LEN(2));
1396 memset (&root
->info
, 0, sizeof(root
->info
));
1397 root
->info
.info_length
= sizeof(root
->info
);
1398 root
->info
.hash_version
= EXT3_SB(dir
->i_sb
)->s_def_hash_version
;
1399 entries
= root
->entries
;
1400 dx_set_block (entries
, 1);
1401 dx_set_count (entries
, 1);
1402 dx_set_limit (entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1404 /* Initialize as for dx_probe */
1405 hinfo
.hash_version
= root
->info
.hash_version
;
1406 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1407 hinfo
.hash_version
+= EXT3_SB(dir
->i_sb
)->s_hash_unsigned
;
1408 hinfo
.seed
= EXT3_SB(dir
->i_sb
)->s_hash_seed
;
1409 ext3fs_dirhash(name
, namelen
, &hinfo
);
1411 frame
->entries
= entries
;
1412 frame
->at
= entries
;
1416 * Mark buffers dirty here so that if do_split() fails we write a
1417 * consistent set of buffers to disk.
1419 ext3_journal_dirty_metadata(handle
, frame
->bh
);
1420 ext3_journal_dirty_metadata(handle
, bh
);
1421 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1423 ext3_mark_inode_dirty(handle
, dir
);
1429 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1435 * adds a file entry to the specified directory, using the same
1436 * semantics as ext3_find_entry(). It returns NULL if it failed.
1438 * NOTE!! The inode part of 'de' is left at 0 - which means you
1439 * may not sleep between calling this and putting something into
1440 * the entry, as someone else might have used it while you slept.
1442 static int ext3_add_entry (handle_t
*handle
, struct dentry
*dentry
,
1443 struct inode
*inode
)
1445 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1446 struct buffer_head
* bh
;
1447 struct ext3_dir_entry_2
*de
;
1448 struct super_block
* sb
;
1455 blocksize
= sb
->s_blocksize
;
1456 if (!dentry
->d_name
.len
)
1459 retval
= ext3_dx_add_entry(handle
, dentry
, inode
);
1460 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1462 EXT3_I(dir
)->i_flags
&= ~EXT3_INDEX_FL
;
1464 ext3_mark_inode_dirty(handle
, dir
);
1466 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1467 for (block
= 0; block
< blocks
; block
++) {
1468 bh
= ext3_bread(handle
, dir
, block
, 0, &retval
);
1471 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1472 if (retval
!= -ENOSPC
)
1475 if (blocks
== 1 && !dx_fallback
&&
1476 EXT3_HAS_COMPAT_FEATURE(sb
, EXT3_FEATURE_COMPAT_DIR_INDEX
))
1477 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1480 bh
= ext3_append(handle
, dir
, &block
, &retval
);
1483 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1485 de
->rec_len
= ext3_rec_len_to_disk(blocksize
);
1486 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1490 * Returns 0 for success, or a negative error value
1492 static int ext3_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1493 struct inode
*inode
)
1495 struct dx_frame frames
[2], *frame
;
1496 struct dx_entry
*entries
, *at
;
1497 struct dx_hash_info hinfo
;
1498 struct buffer_head
* bh
;
1499 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1500 struct super_block
* sb
= dir
->i_sb
;
1501 struct ext3_dir_entry_2
*de
;
1504 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1507 entries
= frame
->entries
;
1510 if (!(bh
= ext3_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1513 BUFFER_TRACE(bh
, "get_write_access");
1514 err
= ext3_journal_get_write_access(handle
, bh
);
1518 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1519 if (err
!= -ENOSPC
) {
1524 /* Block full, should compress but for now just split */
1525 dxtrace(printk("using %u of %u node entries\n",
1526 dx_get_count(entries
), dx_get_limit(entries
)));
1527 /* Need to split index? */
1528 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1530 unsigned icount
= dx_get_count(entries
);
1531 int levels
= frame
- frames
;
1532 struct dx_entry
*entries2
;
1533 struct dx_node
*node2
;
1534 struct buffer_head
*bh2
;
1536 if (levels
&& (dx_get_count(frames
->entries
) ==
1537 dx_get_limit(frames
->entries
))) {
1538 ext3_warning(sb
, __func__
,
1539 "Directory index full!");
1543 bh2
= ext3_append (handle
, dir
, &newblock
, &err
);
1546 node2
= (struct dx_node
*)(bh2
->b_data
);
1547 entries2
= node2
->entries
;
1548 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
1549 node2
->fake
.rec_len
= ext3_rec_len_to_disk(sb
->s_blocksize
);
1550 BUFFER_TRACE(frame
->bh
, "get_write_access");
1551 err
= ext3_journal_get_write_access(handle
, frame
->bh
);
1555 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1556 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1557 dxtrace(printk("Split index %i/%i\n", icount1
, icount2
));
1559 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1560 err
= ext3_journal_get_write_access(handle
,
1565 memcpy ((char *) entries2
, (char *) (entries
+ icount1
),
1566 icount2
* sizeof(struct dx_entry
));
1567 dx_set_count (entries
, icount1
);
1568 dx_set_count (entries2
, icount2
);
1569 dx_set_limit (entries2
, dx_node_limit(dir
));
1571 /* Which index block gets the new entry? */
1572 if (at
- entries
>= icount1
) {
1573 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1574 frame
->entries
= entries
= entries2
;
1575 swap(frame
->bh
, bh2
);
1577 dx_insert_block (frames
+ 0, hash2
, newblock
);
1578 dxtrace(dx_show_index ("node", frames
[1].entries
));
1579 dxtrace(dx_show_index ("node",
1580 ((struct dx_node
*) bh2
->b_data
)->entries
));
1581 err
= ext3_journal_dirty_metadata(handle
, bh2
);
1586 dxtrace(printk("Creating second level index...\n"));
1587 memcpy((char *) entries2
, (char *) entries
,
1588 icount
* sizeof(struct dx_entry
));
1589 dx_set_limit(entries2
, dx_node_limit(dir
));
1592 dx_set_count(entries
, 1);
1593 dx_set_block(entries
+ 0, newblock
);
1594 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1596 /* Add new access path frame */
1598 frame
->at
= at
= at
- entries
+ entries2
;
1599 frame
->entries
= entries
= entries2
;
1601 err
= ext3_journal_get_write_access(handle
,
1606 err
= ext3_journal_dirty_metadata(handle
, frames
[0].bh
);
1610 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1613 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1618 ext3_std_error(dir
->i_sb
, err
);
1627 * ext3_delete_entry deletes a directory entry by merging it with the
1630 static int ext3_delete_entry (handle_t
*handle
,
1632 struct ext3_dir_entry_2
* de_del
,
1633 struct buffer_head
* bh
)
1635 struct ext3_dir_entry_2
* de
, * pde
;
1640 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1641 while (i
< bh
->b_size
) {
1642 if (!ext3_check_dir_entry("ext3_delete_entry", dir
, de
, bh
, i
))
1647 BUFFER_TRACE(bh
, "get_write_access");
1648 err
= ext3_journal_get_write_access(handle
, bh
);
1653 pde
->rec_len
= ext3_rec_len_to_disk(
1654 ext3_rec_len_from_disk(pde
->rec_len
) +
1655 ext3_rec_len_from_disk(de
->rec_len
));
1659 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1660 err
= ext3_journal_dirty_metadata(handle
, bh
);
1663 ext3_std_error(dir
->i_sb
, err
);
1668 i
+= ext3_rec_len_from_disk(de
->rec_len
);
1670 de
= ext3_next_entry(de
);
1675 static int ext3_add_nondir(handle_t
*handle
,
1676 struct dentry
*dentry
, struct inode
*inode
)
1678 int err
= ext3_add_entry(handle
, dentry
, inode
);
1680 ext3_mark_inode_dirty(handle
, inode
);
1681 d_instantiate(dentry
, inode
);
1682 unlock_new_inode(inode
);
1686 unlock_new_inode(inode
);
1692 * By the time this is called, we already have created
1693 * the directory cache entry for the new file, but it
1694 * is so far negative - it has no inode.
1696 * If the create succeeds, we fill in the inode information
1697 * with d_instantiate().
1699 static int ext3_create (struct inode
* dir
, struct dentry
* dentry
, int mode
,
1700 struct nameidata
*nd
)
1703 struct inode
* inode
;
1704 int err
, retries
= 0;
1706 dquot_initialize(dir
);
1709 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1710 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1711 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1713 return PTR_ERR(handle
);
1715 if (IS_DIRSYNC(dir
))
1718 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, mode
);
1719 err
= PTR_ERR(inode
);
1720 if (!IS_ERR(inode
)) {
1721 inode
->i_op
= &ext3_file_inode_operations
;
1722 inode
->i_fop
= &ext3_file_operations
;
1723 ext3_set_aops(inode
);
1724 err
= ext3_add_nondir(handle
, dentry
, inode
);
1726 ext3_journal_stop(handle
);
1727 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1732 static int ext3_mknod (struct inode
* dir
, struct dentry
*dentry
,
1733 int mode
, dev_t rdev
)
1736 struct inode
*inode
;
1737 int err
, retries
= 0;
1739 if (!new_valid_dev(rdev
))
1742 dquot_initialize(dir
);
1745 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1746 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1747 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1749 return PTR_ERR(handle
);
1751 if (IS_DIRSYNC(dir
))
1754 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, mode
);
1755 err
= PTR_ERR(inode
);
1756 if (!IS_ERR(inode
)) {
1757 init_special_inode(inode
, inode
->i_mode
, rdev
);
1758 #ifdef CONFIG_EXT3_FS_XATTR
1759 inode
->i_op
= &ext3_special_inode_operations
;
1761 err
= ext3_add_nondir(handle
, dentry
, inode
);
1763 ext3_journal_stop(handle
);
1764 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1769 static int ext3_mkdir(struct inode
* dir
, struct dentry
* dentry
, int mode
)
1772 struct inode
* inode
;
1773 struct buffer_head
* dir_block
= NULL
;
1774 struct ext3_dir_entry_2
* de
;
1775 int err
, retries
= 0;
1777 if (dir
->i_nlink
>= EXT3_LINK_MAX
)
1780 dquot_initialize(dir
);
1783 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1784 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1785 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1787 return PTR_ERR(handle
);
1789 if (IS_DIRSYNC(dir
))
1792 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, S_IFDIR
| mode
);
1793 err
= PTR_ERR(inode
);
1797 inode
->i_op
= &ext3_dir_inode_operations
;
1798 inode
->i_fop
= &ext3_dir_operations
;
1799 inode
->i_size
= EXT3_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1800 dir_block
= ext3_bread (handle
, inode
, 0, 1, &err
);
1802 goto out_clear_inode
;
1804 BUFFER_TRACE(dir_block
, "get_write_access");
1805 err
= ext3_journal_get_write_access(handle
, dir_block
);
1807 goto out_clear_inode
;
1809 de
= (struct ext3_dir_entry_2
*) dir_block
->b_data
;
1810 de
->inode
= cpu_to_le32(inode
->i_ino
);
1812 de
->rec_len
= ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de
->name_len
));
1813 strcpy (de
->name
, ".");
1814 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1815 de
= ext3_next_entry(de
);
1816 de
->inode
= cpu_to_le32(dir
->i_ino
);
1817 de
->rec_len
= ext3_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1818 EXT3_DIR_REC_LEN(1));
1820 strcpy (de
->name
, "..");
1821 ext3_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1823 BUFFER_TRACE(dir_block
, "call ext3_journal_dirty_metadata");
1824 err
= ext3_journal_dirty_metadata(handle
, dir_block
);
1826 goto out_clear_inode
;
1828 err
= ext3_mark_inode_dirty(handle
, inode
);
1830 err
= ext3_add_entry (handle
, dentry
, inode
);
1835 unlock_new_inode(inode
);
1836 ext3_mark_inode_dirty(handle
, inode
);
1841 ext3_update_dx_flag(dir
);
1842 err
= ext3_mark_inode_dirty(handle
, dir
);
1844 goto out_clear_inode
;
1846 d_instantiate(dentry
, inode
);
1847 unlock_new_inode(inode
);
1850 ext3_journal_stop(handle
);
1851 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
1857 * routine to check that the specified directory is empty (for rmdir)
1859 static int empty_dir (struct inode
* inode
)
1861 unsigned long offset
;
1862 struct buffer_head
* bh
;
1863 struct ext3_dir_entry_2
* de
, * de1
;
1864 struct super_block
* sb
;
1868 if (inode
->i_size
< EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1869 !(bh
= ext3_bread (NULL
, inode
, 0, 0, &err
))) {
1871 ext3_error(inode
->i_sb
, __func__
,
1872 "error %d reading directory #%lu offset 0",
1875 ext3_warning(inode
->i_sb
, __func__
,
1876 "bad directory (dir #%lu) - no data block",
1880 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1881 de1
= ext3_next_entry(de
);
1882 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1883 !le32_to_cpu(de1
->inode
) ||
1884 strcmp (".", de
->name
) ||
1885 strcmp ("..", de1
->name
)) {
1886 ext3_warning (inode
->i_sb
, "empty_dir",
1887 "bad directory (dir #%lu) - no `.' or `..'",
1892 offset
= ext3_rec_len_from_disk(de
->rec_len
) +
1893 ext3_rec_len_from_disk(de1
->rec_len
);
1894 de
= ext3_next_entry(de1
);
1895 while (offset
< inode
->i_size
) {
1897 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1900 bh
= ext3_bread (NULL
, inode
,
1901 offset
>> EXT3_BLOCK_SIZE_BITS(sb
), 0, &err
);
1904 ext3_error(sb
, __func__
,
1905 "error %d reading directory"
1907 err
, inode
->i_ino
, offset
);
1908 offset
+= sb
->s_blocksize
;
1911 de
= (struct ext3_dir_entry_2
*) bh
->b_data
;
1913 if (!ext3_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1914 de
= (struct ext3_dir_entry_2
*)(bh
->b_data
+
1916 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1919 if (le32_to_cpu(de
->inode
)) {
1923 offset
+= ext3_rec_len_from_disk(de
->rec_len
);
1924 de
= ext3_next_entry(de
);
1930 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1931 * such inodes, starting at the superblock, in case we crash before the
1932 * file is closed/deleted, or in case the inode truncate spans multiple
1933 * transactions and the last transaction is not recovered after a crash.
1935 * At filesystem recovery time, we walk this list deleting unlinked
1936 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1938 int ext3_orphan_add(handle_t
*handle
, struct inode
*inode
)
1940 struct super_block
*sb
= inode
->i_sb
;
1941 struct ext3_iloc iloc
;
1944 mutex_lock(&EXT3_SB(sb
)->s_orphan_lock
);
1945 if (!list_empty(&EXT3_I(inode
)->i_orphan
))
1948 /* Orphan handling is only valid for files with data blocks
1949 * being truncated, or files being unlinked. */
1951 /* @@@ FIXME: Observation from aviro:
1952 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1953 * here (on s_orphan_lock), so race with ext3_link() which might bump
1954 * ->i_nlink. For, say it, character device. Not a regular file,
1955 * not a directory, not a symlink and ->i_nlink > 0.
1957 * tytso, 4/25/2009: I'm not sure how that could happen;
1958 * shouldn't the fs core protect us from these sort of
1959 * unlink()/link() races?
1961 J_ASSERT ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1962 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1964 BUFFER_TRACE(EXT3_SB(sb
)->s_sbh
, "get_write_access");
1965 err
= ext3_journal_get_write_access(handle
, EXT3_SB(sb
)->s_sbh
);
1969 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
1973 /* Insert this inode at the head of the on-disk orphan list... */
1974 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT3_SB(sb
)->s_es
->s_last_orphan
);
1975 EXT3_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1976 err
= ext3_journal_dirty_metadata(handle
, EXT3_SB(sb
)->s_sbh
);
1977 rc
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
1981 /* Only add to the head of the in-memory list if all the
1982 * previous operations succeeded. If the orphan_add is going to
1983 * fail (possibly taking the journal offline), we can't risk
1984 * leaving the inode on the orphan list: stray orphan-list
1985 * entries can cause panics at unmount time.
1987 * This is safe: on error we're going to ignore the orphan list
1988 * anyway on the next recovery. */
1990 list_add(&EXT3_I(inode
)->i_orphan
, &EXT3_SB(sb
)->s_orphan
);
1992 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1993 jbd_debug(4, "orphan inode %lu will point to %d\n",
1994 inode
->i_ino
, NEXT_ORPHAN(inode
));
1996 mutex_unlock(&EXT3_SB(sb
)->s_orphan_lock
);
1997 ext3_std_error(inode
->i_sb
, err
);
2002 * ext3_orphan_del() removes an unlinked or truncated inode from the list
2003 * of such inodes stored on disk, because it is finally being cleaned up.
2005 int ext3_orphan_del(handle_t
*handle
, struct inode
*inode
)
2007 struct list_head
*prev
;
2008 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2009 struct ext3_sb_info
*sbi
;
2010 unsigned long ino_next
;
2011 struct ext3_iloc iloc
;
2014 mutex_lock(&EXT3_SB(inode
->i_sb
)->s_orphan_lock
);
2015 if (list_empty(&ei
->i_orphan
))
2018 ino_next
= NEXT_ORPHAN(inode
);
2019 prev
= ei
->i_orphan
.prev
;
2020 sbi
= EXT3_SB(inode
->i_sb
);
2022 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2024 list_del_init(&ei
->i_orphan
);
2026 /* If we're on an error path, we may not have a valid
2027 * transaction handle with which to update the orphan list on
2028 * disk, but we still need to remove the inode from the linked
2029 * list in memory. */
2033 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
2037 if (prev
== &sbi
->s_orphan
) {
2038 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
2039 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2040 err
= ext3_journal_get_write_access(handle
, sbi
->s_sbh
);
2043 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2044 err
= ext3_journal_dirty_metadata(handle
, sbi
->s_sbh
);
2046 struct ext3_iloc iloc2
;
2047 struct inode
*i_prev
=
2048 &list_entry(prev
, struct ext3_inode_info
, i_orphan
)->vfs_inode
;
2050 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2051 i_prev
->i_ino
, ino_next
);
2052 err
= ext3_reserve_inode_write(handle
, i_prev
, &iloc2
);
2055 NEXT_ORPHAN(i_prev
) = ino_next
;
2056 err
= ext3_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2060 NEXT_ORPHAN(inode
) = 0;
2061 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
2064 ext3_std_error(inode
->i_sb
, err
);
2066 mutex_unlock(&EXT3_SB(inode
->i_sb
)->s_orphan_lock
);
2074 static int ext3_rmdir (struct inode
* dir
, struct dentry
*dentry
)
2077 struct inode
* inode
;
2078 struct buffer_head
* bh
;
2079 struct ext3_dir_entry_2
* de
;
2082 /* Initialize quotas before so that eventual writes go in
2083 * separate transaction */
2084 dquot_initialize(dir
);
2085 dquot_initialize(dentry
->d_inode
);
2087 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2089 return PTR_ERR(handle
);
2092 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2096 if (IS_DIRSYNC(dir
))
2099 inode
= dentry
->d_inode
;
2102 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2105 retval
= -ENOTEMPTY
;
2106 if (!empty_dir (inode
))
2109 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2112 if (inode
->i_nlink
!= 2)
2113 ext3_warning (inode
->i_sb
, "ext3_rmdir",
2114 "empty directory has nlink!=2 (%d)",
2118 /* There's no need to set i_disksize: the fact that i_nlink is
2119 * zero will ensure that the right thing happens during any
2122 ext3_orphan_add(handle
, inode
);
2123 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2124 ext3_mark_inode_dirty(handle
, inode
);
2126 ext3_update_dx_flag(dir
);
2127 ext3_mark_inode_dirty(handle
, dir
);
2130 ext3_journal_stop(handle
);
2135 static int ext3_unlink(struct inode
* dir
, struct dentry
*dentry
)
2138 struct inode
* inode
;
2139 struct buffer_head
* bh
;
2140 struct ext3_dir_entry_2
* de
;
2143 /* Initialize quotas before so that eventual writes go
2144 * in separate transaction */
2145 dquot_initialize(dir
);
2146 dquot_initialize(dentry
->d_inode
);
2148 handle
= ext3_journal_start(dir
, EXT3_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2150 return PTR_ERR(handle
);
2152 if (IS_DIRSYNC(dir
))
2156 bh
= ext3_find_entry(dir
, &dentry
->d_name
, &de
);
2160 inode
= dentry
->d_inode
;
2163 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2166 if (!inode
->i_nlink
) {
2167 ext3_warning (inode
->i_sb
, "ext3_unlink",
2168 "Deleting nonexistent file (%lu), %d",
2169 inode
->i_ino
, inode
->i_nlink
);
2172 retval
= ext3_delete_entry(handle
, dir
, de
, bh
);
2175 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME_SEC
;
2176 ext3_update_dx_flag(dir
);
2177 ext3_mark_inode_dirty(handle
, dir
);
2179 if (!inode
->i_nlink
)
2180 ext3_orphan_add(handle
, inode
);
2181 inode
->i_ctime
= dir
->i_ctime
;
2182 ext3_mark_inode_dirty(handle
, inode
);
2186 ext3_journal_stop(handle
);
2191 static int ext3_symlink (struct inode
* dir
,
2192 struct dentry
*dentry
, const char * symname
)
2195 struct inode
* inode
;
2196 int l
, err
, retries
= 0;
2199 l
= strlen(symname
)+1;
2200 if (l
> dir
->i_sb
->s_blocksize
)
2201 return -ENAMETOOLONG
;
2203 dquot_initialize(dir
);
2205 if (l
> EXT3_N_BLOCKS
* 4) {
2207 * For non-fast symlinks, we just allocate inode and put it on
2208 * orphan list in the first transaction => we need bitmap,
2209 * group descriptor, sb, inode block, quota blocks.
2211 credits
= 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
);
2214 * Fast symlink. We have to add entry to directory
2215 * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2216 * allocate new inode (bitmap, group descriptor, inode block,
2217 * quota blocks, sb is already counted in previous macros).
2219 credits
= EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2220 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
2221 EXT3_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
);
2224 handle
= ext3_journal_start(dir
, credits
);
2226 return PTR_ERR(handle
);
2228 if (IS_DIRSYNC(dir
))
2231 inode
= ext3_new_inode (handle
, dir
, &dentry
->d_name
, S_IFLNK
|S_IRWXUGO
);
2232 err
= PTR_ERR(inode
);
2236 if (l
> EXT3_N_BLOCKS
* 4) {
2237 inode
->i_op
= &ext3_symlink_inode_operations
;
2238 ext3_set_aops(inode
);
2240 * We cannot call page_symlink() with transaction started
2241 * because it calls into ext3_write_begin() which acquires page
2242 * lock which ranks below transaction start (and it can also
2243 * wait for journal commit if we are running out of space). So
2244 * we have to stop transaction now and restart it when symlink
2245 * contents is written.
2247 * To keep fs consistent in case of crash, we have to put inode
2248 * to orphan list in the mean time.
2251 err
= ext3_orphan_add(handle
, inode
);
2252 ext3_journal_stop(handle
);
2254 goto err_drop_inode
;
2255 err
= __page_symlink(inode
, symname
, l
, 1);
2257 goto err_drop_inode
;
2259 * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2260 * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2262 handle
= ext3_journal_start(dir
,
2263 EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2264 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
2265 if (IS_ERR(handle
)) {
2266 err
= PTR_ERR(handle
);
2267 goto err_drop_inode
;
2270 err
= ext3_orphan_del(handle
, inode
);
2272 ext3_journal_stop(handle
);
2274 goto err_drop_inode
;
2277 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2278 memcpy((char*)&EXT3_I(inode
)->i_data
,symname
,l
);
2279 inode
->i_size
= l
-1;
2281 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
2282 err
= ext3_add_nondir(handle
, dentry
, inode
);
2284 ext3_journal_stop(handle
);
2285 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2289 unlock_new_inode(inode
);
2294 static int ext3_link (struct dentry
* old_dentry
,
2295 struct inode
* dir
, struct dentry
*dentry
)
2298 struct inode
*inode
= old_dentry
->d_inode
;
2299 int err
, retries
= 0;
2301 if (inode
->i_nlink
>= EXT3_LINK_MAX
)
2304 dquot_initialize(dir
);
2307 handle
= ext3_journal_start(dir
, EXT3_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2308 EXT3_INDEX_EXTRA_TRANS_BLOCKS
);
2310 return PTR_ERR(handle
);
2312 if (IS_DIRSYNC(dir
))
2315 inode
->i_ctime
= CURRENT_TIME_SEC
;
2319 err
= ext3_add_entry(handle
, dentry
, inode
);
2321 ext3_mark_inode_dirty(handle
, inode
);
2322 d_instantiate(dentry
, inode
);
2327 ext3_journal_stop(handle
);
2328 if (err
== -ENOSPC
&& ext3_should_retry_alloc(dir
->i_sb
, &retries
))
2333 #define PARENT_INO(buffer) \
2334 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2337 * Anybody can rename anything with this: the permission checks are left to the
2338 * higher-level routines.
2340 static int ext3_rename (struct inode
* old_dir
, struct dentry
*old_dentry
,
2341 struct inode
* new_dir
,struct dentry
*new_dentry
)
2344 struct inode
* old_inode
, * new_inode
;
2345 struct buffer_head
* old_bh
, * new_bh
, * dir_bh
;
2346 struct ext3_dir_entry_2
* old_de
, * new_de
;
2347 int retval
, flush_file
= 0;
2349 dquot_initialize(old_dir
);
2350 dquot_initialize(new_dir
);
2352 old_bh
= new_bh
= dir_bh
= NULL
;
2354 /* Initialize quotas before so that eventual writes go
2355 * in separate transaction */
2356 if (new_dentry
->d_inode
)
2357 dquot_initialize(new_dentry
->d_inode
);
2358 handle
= ext3_journal_start(old_dir
, 2 *
2359 EXT3_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2360 EXT3_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2362 return PTR_ERR(handle
);
2364 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2367 old_bh
= ext3_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2369 * Check for inode number is _not_ due to possible IO errors.
2370 * We might rmdir the source, keep it as pwd of some process
2371 * and merrily kill the link to whatever was created under the
2372 * same name. Goodbye sticky bit ;-<
2374 old_inode
= old_dentry
->d_inode
;
2376 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2379 new_inode
= new_dentry
->d_inode
;
2380 new_bh
= ext3_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
2387 if (S_ISDIR(old_inode
->i_mode
)) {
2389 retval
= -ENOTEMPTY
;
2390 if (!empty_dir (new_inode
))
2394 dir_bh
= ext3_bread (handle
, old_inode
, 0, 0, &retval
);
2397 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2400 if (!new_inode
&& new_dir
!=old_dir
&&
2401 new_dir
->i_nlink
>= EXT3_LINK_MAX
)
2405 retval
= ext3_add_entry (handle
, new_dentry
, old_inode
);
2409 BUFFER_TRACE(new_bh
, "get write access");
2410 retval
= ext3_journal_get_write_access(handle
, new_bh
);
2413 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2414 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2415 EXT3_FEATURE_INCOMPAT_FILETYPE
))
2416 new_de
->file_type
= old_de
->file_type
;
2417 new_dir
->i_version
++;
2418 new_dir
->i_ctime
= new_dir
->i_mtime
= CURRENT_TIME_SEC
;
2419 ext3_mark_inode_dirty(handle
, new_dir
);
2420 BUFFER_TRACE(new_bh
, "call ext3_journal_dirty_metadata");
2421 retval
= ext3_journal_dirty_metadata(handle
, new_bh
);
2429 * Like most other Unix systems, set the ctime for inodes on a
2432 old_inode
->i_ctime
= CURRENT_TIME_SEC
;
2433 ext3_mark_inode_dirty(handle
, old_inode
);
2438 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2439 old_de
->name_len
!= old_dentry
->d_name
.len
||
2440 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2441 (retval
= ext3_delete_entry(handle
, old_dir
,
2442 old_de
, old_bh
)) == -ENOENT
) {
2443 /* old_de could have moved from under us during htree split, so
2444 * make sure that we are deleting the right entry. We might
2445 * also be pointing to a stale entry in the unused part of
2446 * old_bh so just checking inum and the name isn't enough. */
2447 struct buffer_head
*old_bh2
;
2448 struct ext3_dir_entry_2
*old_de2
;
2450 old_bh2
= ext3_find_entry(old_dir
, &old_dentry
->d_name
,
2453 retval
= ext3_delete_entry(handle
, old_dir
,
2459 ext3_warning(old_dir
->i_sb
, "ext3_rename",
2460 "Deleting old file (%lu), %d, error=%d",
2461 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2465 drop_nlink(new_inode
);
2466 new_inode
->i_ctime
= CURRENT_TIME_SEC
;
2468 old_dir
->i_ctime
= old_dir
->i_mtime
= CURRENT_TIME_SEC
;
2469 ext3_update_dx_flag(old_dir
);
2471 BUFFER_TRACE(dir_bh
, "get_write_access");
2472 retval
= ext3_journal_get_write_access(handle
, dir_bh
);
2475 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2476 BUFFER_TRACE(dir_bh
, "call ext3_journal_dirty_metadata");
2477 retval
= ext3_journal_dirty_metadata(handle
, dir_bh
);
2480 ext3_std_error(new_dir
->i_sb
, retval
);
2483 drop_nlink(old_dir
);
2485 drop_nlink(new_inode
);
2488 ext3_update_dx_flag(new_dir
);
2489 ext3_mark_inode_dirty(handle
, new_dir
);
2492 ext3_mark_inode_dirty(handle
, old_dir
);
2494 ext3_mark_inode_dirty(handle
, new_inode
);
2495 if (!new_inode
->i_nlink
)
2496 ext3_orphan_add(handle
, new_inode
);
2497 if (ext3_should_writeback_data(new_inode
))
2506 ext3_journal_stop(handle
);
2507 if (retval
== 0 && flush_file
)
2508 filemap_flush(old_inode
->i_mapping
);
2513 * directories can handle most operations...
2515 const struct inode_operations ext3_dir_inode_operations
= {
2516 .create
= ext3_create
,
2517 .lookup
= ext3_lookup
,
2519 .unlink
= ext3_unlink
,
2520 .symlink
= ext3_symlink
,
2521 .mkdir
= ext3_mkdir
,
2522 .rmdir
= ext3_rmdir
,
2523 .mknod
= ext3_mknod
,
2524 .rename
= ext3_rename
,
2525 .setattr
= ext3_setattr
,
2526 #ifdef CONFIG_EXT3_FS_XATTR
2527 .setxattr
= generic_setxattr
,
2528 .getxattr
= generic_getxattr
,
2529 .listxattr
= ext3_listxattr
,
2530 .removexattr
= generic_removexattr
,
2532 .get_acl
= ext3_get_acl
,
2535 const struct inode_operations ext3_special_inode_operations
= {
2536 .setattr
= ext3_setattr
,
2537 #ifdef CONFIG_EXT3_FS_XATTR
2538 .setxattr
= generic_setxattr
,
2539 .getxattr
= generic_getxattr
,
2540 .listxattr
= ext3_listxattr
,
2541 .removexattr
= generic_removexattr
,
2543 .get_acl
= ext3_get_acl
,