2 * linux/fs/ext4/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/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
38 #include "ext4_jbd2.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
*ext4_append(handle_t
*handle
,
54 ext4_lblk_t
*block
, int *err
)
56 struct buffer_head
*bh
;
58 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
60 bh
= ext4_bread(handle
, inode
, *block
, 1, err
);
62 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
63 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
64 *err
= ext4_journal_get_write_access(handle
, bh
);
74 #define assert(test) J_ASSERT(test)
78 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
82 #define dxtrace(command) command
84 #define dxtrace(command)
108 * dx_root_info is laid out so that if it should somehow get overlaid by a
109 * dirent the two low bits of the hash version will be zero. Therefore, the
110 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
115 struct fake_dirent dot
;
117 struct fake_dirent dotdot
;
121 __le32 reserved_zero
;
123 u8 info_length
; /* 8 */
128 struct dx_entry entries
[0];
133 struct fake_dirent fake
;
134 struct dx_entry entries
[0];
140 struct buffer_head
*bh
;
141 struct dx_entry
*entries
;
152 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
153 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
154 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
155 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
156 static unsigned dx_get_count(struct dx_entry
*entries
);
157 static unsigned dx_get_limit(struct dx_entry
*entries
);
158 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
159 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
160 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
161 static unsigned dx_node_limit(struct inode
*dir
);
162 static struct dx_frame
*dx_probe(const struct qstr
*d_name
,
164 struct dx_hash_info
*hinfo
,
165 struct dx_frame
*frame
,
167 static void dx_release(struct dx_frame
*frames
);
168 static int dx_make_map(struct ext4_dir_entry_2
*de
, int size
,
169 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
170 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
171 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
172 struct dx_map_entry
*offsets
, int count
);
173 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, int size
);
174 static void dx_insert_block(struct dx_frame
*frame
,
175 u32 hash
, ext4_lblk_t block
);
176 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
177 struct dx_frame
*frame
,
178 struct dx_frame
*frames
,
180 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
181 const struct qstr
*d_name
,
182 struct ext4_dir_entry_2
**res_dir
,
184 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
185 struct inode
*inode
);
188 * p is at least 6 bytes before the end of page
190 static inline struct ext4_dir_entry_2
*
191 ext4_next_entry(struct ext4_dir_entry_2
*p
)
193 return (struct ext4_dir_entry_2
*)((char *)p
+
194 ext4_rec_len_from_disk(p
->rec_len
));
198 * Future: use high four bits of block for coalesce-on-delete flags
199 * Mask them off for now.
202 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
204 return le32_to_cpu(entry
->block
) & 0x00ffffff;
207 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
209 entry
->block
= cpu_to_le32(value
);
212 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
214 return le32_to_cpu(entry
->hash
);
217 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
219 entry
->hash
= cpu_to_le32(value
);
222 static inline unsigned dx_get_count(struct dx_entry
*entries
)
224 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
227 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
229 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
232 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
234 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
237 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
239 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
242 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
244 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
245 EXT4_DIR_REC_LEN(2) - infosize
;
246 return entry_space
/ sizeof(struct dx_entry
);
249 static inline unsigned dx_node_limit(struct inode
*dir
)
251 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
252 return entry_space
/ sizeof(struct dx_entry
);
259 static void dx_show_index(char * label
, struct dx_entry
*entries
)
261 int i
, n
= dx_get_count (entries
);
262 printk(KERN_DEBUG
"%s index ", label
);
263 for (i
= 0; i
< n
; i
++) {
264 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
265 0, (unsigned long)dx_get_block(entries
+ i
));
277 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
278 int size
, int show_names
)
280 unsigned names
= 0, space
= 0;
281 char *base
= (char *) de
;
282 struct dx_hash_info h
= *hinfo
;
285 while ((char *) de
< base
+ size
)
291 int len
= de
->name_len
;
292 char *name
= de
->name
;
293 while (len
--) printk("%c", *name
++);
294 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
295 printk(":%x.%u ", h
.hash
,
296 ((char *) de
- base
));
298 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
301 de
= ext4_next_entry(de
);
303 printk("(%i)\n", names
);
304 return (struct stats
) { names
, space
, 1 };
307 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
308 struct dx_entry
*entries
, int levels
)
310 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
311 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
313 struct buffer_head
*bh
;
315 printk("%i indexed blocks...\n", count
);
316 for (i
= 0; i
< count
; i
++, entries
++)
318 ext4_lblk_t block
= dx_get_block(entries
);
319 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
320 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
322 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
323 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0,&err
))) continue;
325 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
326 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
327 names
+= stats
.names
;
328 space
+= stats
.space
;
329 bcount
+= stats
.bcount
;
333 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
334 levels
? "" : " ", names
, space
/bcount
,
335 (space
/bcount
)*100/blocksize
);
336 return (struct stats
) { names
, space
, bcount
};
338 #endif /* DX_DEBUG */
341 * Probe for a directory leaf block to search.
343 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
344 * error in the directory index, and the caller should fall back to
345 * searching the directory normally. The callers of dx_probe **MUST**
346 * check for this error code, and make sure it never gets reflected
349 static struct dx_frame
*
350 dx_probe(const struct qstr
*d_name
, struct inode
*dir
,
351 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
353 unsigned count
, indirect
;
354 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
355 struct dx_root
*root
;
356 struct buffer_head
*bh
;
357 struct dx_frame
*frame
= frame_in
;
361 if (!(bh
= ext4_bread (NULL
,dir
, 0, 0, err
)))
363 root
= (struct dx_root
*) bh
->b_data
;
364 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
365 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
366 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
367 ext4_warning(dir
->i_sb
, __func__
,
368 "Unrecognised inode hash code %d",
369 root
->info
.hash_version
);
371 *err
= ERR_BAD_DX_DIR
;
374 hinfo
->hash_version
= root
->info
.hash_version
;
375 if (hinfo
->hash_version
<= DX_HASH_TEA
)
376 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
377 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
379 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
382 if (root
->info
.unused_flags
& 1) {
383 ext4_warning(dir
->i_sb
, __func__
,
384 "Unimplemented inode hash flags: %#06x",
385 root
->info
.unused_flags
);
387 *err
= ERR_BAD_DX_DIR
;
391 if ((indirect
= root
->info
.indirect_levels
) > 1) {
392 ext4_warning(dir
->i_sb
, __func__
,
393 "Unimplemented inode hash depth: %#06x",
394 root
->info
.indirect_levels
);
396 *err
= ERR_BAD_DX_DIR
;
400 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
401 root
->info
.info_length
);
403 if (dx_get_limit(entries
) != dx_root_limit(dir
,
404 root
->info
.info_length
)) {
405 ext4_warning(dir
->i_sb
, __func__
,
406 "dx entry: limit != root limit");
408 *err
= ERR_BAD_DX_DIR
;
412 dxtrace(printk("Look up %x", hash
));
415 count
= dx_get_count(entries
);
416 if (!count
|| count
> dx_get_limit(entries
)) {
417 ext4_warning(dir
->i_sb
, __func__
,
418 "dx entry: no count or count > limit");
420 *err
= ERR_BAD_DX_DIR
;
425 q
= entries
+ count
- 1;
429 dxtrace(printk("."));
430 if (dx_get_hash(m
) > hash
)
436 if (0) // linear search cross check
438 unsigned n
= count
- 1;
442 dxtrace(printk(","));
443 if (dx_get_hash(++at
) > hash
)
449 assert (at
== p
- 1);
453 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
455 frame
->entries
= entries
;
457 if (!indirect
--) return frame
;
458 if (!(bh
= ext4_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
460 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
461 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
462 ext4_warning(dir
->i_sb
, __func__
,
463 "dx entry: limit != node limit");
465 *err
= ERR_BAD_DX_DIR
;
472 while (frame
>= frame_in
) {
477 if (*err
== ERR_BAD_DX_DIR
)
478 ext4_warning(dir
->i_sb
, __func__
,
479 "Corrupt dir inode %ld, running e2fsck is "
480 "recommended.", dir
->i_ino
);
484 static void dx_release (struct dx_frame
*frames
)
486 if (frames
[0].bh
== NULL
)
489 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
490 brelse(frames
[1].bh
);
491 brelse(frames
[0].bh
);
495 * This function increments the frame pointer to search the next leaf
496 * block, and reads in the necessary intervening nodes if the search
497 * should be necessary. Whether or not the search is necessary is
498 * controlled by the hash parameter. If the hash value is even, then
499 * the search is only continued if the next block starts with that
500 * hash value. This is used if we are searching for a specific file.
502 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
504 * This function returns 1 if the caller should continue to search,
505 * or 0 if it should not. If there is an error reading one of the
506 * index blocks, it will a negative error code.
508 * If start_hash is non-null, it will be filled in with the starting
509 * hash of the next page.
511 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
512 struct dx_frame
*frame
,
513 struct dx_frame
*frames
,
517 struct buffer_head
*bh
;
518 int err
, num_frames
= 0;
523 * Find the next leaf page by incrementing the frame pointer.
524 * If we run out of entries in the interior node, loop around and
525 * increment pointer in the parent node. When we break out of
526 * this loop, num_frames indicates the number of interior
527 * nodes need to be read.
530 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
539 * If the hash is 1, then continue only if the next page has a
540 * continuation hash of any value. This is used for readdir
541 * handling. Otherwise, check to see if the hash matches the
542 * desired contiuation hash. If it doesn't, return since
543 * there's no point to read in the successive index pages.
545 bhash
= dx_get_hash(p
->at
);
548 if ((hash
& 1) == 0) {
549 if ((bhash
& ~1) != hash
)
553 * If the hash is HASH_NB_ALWAYS, we always go to the next
554 * block so no check is necessary
556 while (num_frames
--) {
557 if (!(bh
= ext4_bread(NULL
, dir
, dx_get_block(p
->at
),
559 return err
; /* Failure */
563 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
570 * This function fills a red-black tree with information from a
571 * directory block. It returns the number directory entries loaded
572 * into the tree. If there is an error it is returned in err.
574 static int htree_dirblock_to_tree(struct file
*dir_file
,
575 struct inode
*dir
, ext4_lblk_t block
,
576 struct dx_hash_info
*hinfo
,
577 __u32 start_hash
, __u32 start_minor_hash
)
579 struct buffer_head
*bh
;
580 struct ext4_dir_entry_2
*de
, *top
;
583 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
584 (unsigned long)block
));
585 if (!(bh
= ext4_bread (NULL
, dir
, block
, 0, &err
)))
588 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
589 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
590 dir
->i_sb
->s_blocksize
-
591 EXT4_DIR_REC_LEN(0));
592 for (; de
< top
; de
= ext4_next_entry(de
)) {
593 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
594 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
595 +((char *)de
- bh
->b_data
))) {
596 /* On error, skip the f_pos to the next block. */
597 dir_file
->f_pos
= (dir_file
->f_pos
|
598 (dir
->i_sb
->s_blocksize
- 1)) + 1;
602 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
603 if ((hinfo
->hash
< start_hash
) ||
604 ((hinfo
->hash
== start_hash
) &&
605 (hinfo
->minor_hash
< start_minor_hash
)))
609 if ((err
= ext4_htree_store_dirent(dir_file
,
610 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
622 * This function fills a red-black tree with information from a
623 * directory. We start scanning the directory in hash order, starting
624 * at start_hash and start_minor_hash.
626 * This function returns the number of entries inserted into the tree,
627 * or a negative error code.
629 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
630 __u32 start_minor_hash
, __u32
*next_hash
)
632 struct dx_hash_info hinfo
;
633 struct ext4_dir_entry_2
*de
;
634 struct dx_frame frames
[2], *frame
;
641 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
642 start_hash
, start_minor_hash
));
643 dir
= dir_file
->f_path
.dentry
->d_inode
;
644 if (!(EXT4_I(dir
)->i_flags
& EXT4_INDEX_FL
)) {
645 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
646 if (hinfo
.hash_version
<= DX_HASH_TEA
)
647 hinfo
.hash_version
+=
648 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
649 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
650 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
651 start_hash
, start_minor_hash
);
655 hinfo
.hash
= start_hash
;
656 hinfo
.minor_hash
= 0;
657 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
, &err
);
661 /* Add '.' and '..' from the htree header */
662 if (!start_hash
&& !start_minor_hash
) {
663 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
664 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
668 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
669 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
670 de
= ext4_next_entry(de
);
671 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
677 block
= dx_get_block(frame
->at
);
678 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
679 start_hash
, start_minor_hash
);
686 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
687 frame
, frames
, &hashval
);
688 *next_hash
= hashval
;
694 * Stop if: (a) there are no more entries, or
695 * (b) we have inserted at least one entry and the
696 * next hash value is not a continuation
699 (count
&& ((hashval
& 1) == 0)))
703 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
704 "next hash: %x\n", count
, *next_hash
));
713 * Directory block splitting, compacting
717 * Create map of hash values, offsets, and sizes, stored at end of block.
718 * Returns number of entries mapped.
720 static int dx_make_map (struct ext4_dir_entry_2
*de
, int size
,
721 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
724 char *base
= (char *) de
;
725 struct dx_hash_info h
= *hinfo
;
727 while ((char *) de
< base
+ size
)
729 if (de
->name_len
&& de
->inode
) {
730 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
732 map_tail
->hash
= h
.hash
;
733 map_tail
->offs
= (u16
) ((char *) de
- base
);
734 map_tail
->size
= le16_to_cpu(de
->rec_len
);
738 /* XXX: do we need to check rec_len == 0 case? -Chris */
739 de
= ext4_next_entry(de
);
744 /* Sort map by hash value */
745 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
747 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
749 /* Combsort until bubble sort doesn't suck */
752 if (count
- 9 < 2) /* 9, 10 -> 11 */
754 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
755 if (p
->hash
< q
->hash
)
758 /* Garden variety bubble sort */
763 if (q
[1].hash
>= q
[0].hash
)
771 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
773 struct dx_entry
*entries
= frame
->entries
;
774 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
775 int count
= dx_get_count(entries
);
777 assert(count
< dx_get_limit(entries
));
778 assert(old
< entries
+ count
);
779 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
780 dx_set_hash(new, hash
);
781 dx_set_block(new, block
);
782 dx_set_count(entries
, count
+ 1);
785 static void ext4_update_dx_flag(struct inode
*inode
)
787 if (!EXT4_HAS_COMPAT_FEATURE(inode
->i_sb
,
788 EXT4_FEATURE_COMPAT_DIR_INDEX
))
789 EXT4_I(inode
)->i_flags
&= ~EXT4_INDEX_FL
;
793 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
795 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
796 * `de != NULL' is guaranteed by caller.
798 static inline int ext4_match (int len
, const char * const name
,
799 struct ext4_dir_entry_2
* de
)
801 if (len
!= de
->name_len
)
805 return !memcmp(name
, de
->name
, len
);
809 * Returns 0 if not found, -1 on failure, and 1 on success
811 static inline int search_dirblock(struct buffer_head
*bh
,
813 const struct qstr
*d_name
,
814 unsigned long offset
,
815 struct ext4_dir_entry_2
** res_dir
)
817 struct ext4_dir_entry_2
* de
;
820 const char *name
= d_name
->name
;
821 int namelen
= d_name
->len
;
823 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
824 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
825 while ((char *) de
< dlimit
) {
826 /* this code is executed quadratically often */
827 /* do minimal checking `by hand' */
829 if ((char *) de
+ namelen
<= dlimit
&&
830 ext4_match (namelen
, name
, de
)) {
831 /* found a match - just to be sure, do a full check */
832 if (!ext4_check_dir_entry("ext4_find_entry",
833 dir
, de
, bh
, offset
))
838 /* prevent looping on a bad block */
839 de_len
= ext4_rec_len_from_disk(de
->rec_len
);
843 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
852 * finds an entry in the specified directory with the wanted name. It
853 * returns the cache buffer in which the entry was found, and the entry
854 * itself (as a parameter - res_dir). It does NOT read the inode of the
855 * entry - you'll have to do that yourself if you want to.
857 * The returned buffer_head has ->b_count elevated. The caller is expected
858 * to brelse() it when appropriate.
860 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
861 const struct qstr
*d_name
,
862 struct ext4_dir_entry_2
** res_dir
)
864 struct super_block
*sb
;
865 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
866 struct buffer_head
*bh
, *ret
= NULL
;
867 ext4_lblk_t start
, block
, b
;
868 int ra_max
= 0; /* Number of bh's in the readahead
870 int ra_ptr
= 0; /* Current index into readahead
879 namelen
= d_name
->len
;
880 if (namelen
> EXT4_NAME_LEN
)
883 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
, &err
);
885 * On success, or if the error was file not found,
886 * return. Otherwise, fall back to doing a search the
889 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
891 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
894 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
895 start
= EXT4_I(dir
)->i_dir_start_lookup
;
896 if (start
>= nblocks
)
902 * We deal with the read-ahead logic here.
904 if (ra_ptr
>= ra_max
) {
905 /* Refill the readahead buffer */
908 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
910 * Terminate if we reach the end of the
911 * directory and must wrap, or if our
912 * search has finished at this block.
914 if (b
>= nblocks
|| (num
&& block
== start
)) {
915 bh_use
[ra_max
] = NULL
;
919 bh
= ext4_getblk(NULL
, dir
, b
++, 0, &err
);
922 ll_rw_block(READ_META
, 1, &bh
);
925 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
928 if (!buffer_uptodate(bh
)) {
929 /* read error, skip block & hope for the best */
930 ext4_error(sb
, __func__
, "reading directory #%lu "
931 "offset %lu", dir
->i_ino
,
932 (unsigned long)block
);
936 i
= search_dirblock(bh
, dir
, d_name
,
937 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
939 EXT4_I(dir
)->i_dir_start_lookup
= block
;
941 goto cleanup_and_exit
;
945 goto cleanup_and_exit
;
948 if (++block
>= nblocks
)
950 } while (block
!= start
);
953 * If the directory has grown while we were searching, then
954 * search the last part of the directory before giving up.
957 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
958 if (block
< nblocks
) {
964 /* Clean up the read-ahead blocks */
965 for (; ra_ptr
< ra_max
; ra_ptr
++)
966 brelse(bh_use
[ra_ptr
]);
970 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
971 struct ext4_dir_entry_2
**res_dir
, int *err
)
973 struct super_block
* sb
;
974 struct dx_hash_info hinfo
;
976 struct dx_frame frames
[2], *frame
;
977 struct ext4_dir_entry_2
*de
, *top
;
978 struct buffer_head
*bh
;
981 int namelen
= d_name
->len
;
982 const u8
*name
= d_name
->name
;
985 /* NFS may look up ".." - look at dx_root directory block */
986 if (namelen
> 2 || name
[0] != '.'||(name
[1] != '.' && name
[1] != '\0')){
987 if (!(frame
= dx_probe(d_name
, dir
, &hinfo
, frames
, err
)))
991 frame
->bh
= NULL
; /* for dx_release() */
992 frame
->at
= (struct dx_entry
*)frames
; /* hack for zero entry*/
993 dx_set_block(frame
->at
, 0); /* dx_root block is 0 */
997 block
= dx_get_block(frame
->at
);
998 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0, err
)))
1000 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1001 top
= (struct ext4_dir_entry_2
*) ((char *) de
+ sb
->s_blocksize
-
1002 EXT4_DIR_REC_LEN(0));
1003 for (; de
< top
; de
= ext4_next_entry(de
)) {
1004 int off
= (block
<< EXT4_BLOCK_SIZE_BITS(sb
))
1005 + ((char *) de
- bh
->b_data
);
1007 if (!ext4_check_dir_entry(__func__
, dir
, de
, bh
, off
)) {
1009 *err
= ERR_BAD_DX_DIR
;
1013 if (ext4_match(namelen
, name
, de
)) {
1020 /* Check to see if we should continue to search */
1021 retval
= ext4_htree_next_block(dir
, hash
, frame
,
1024 ext4_warning(sb
, __func__
,
1025 "error reading index page in directory #%lu",
1030 } while (retval
== 1);
1034 dxtrace(printk(KERN_DEBUG
"%s not found\n", name
));
1035 dx_release (frames
);
1039 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1041 struct inode
*inode
;
1042 struct ext4_dir_entry_2
*de
;
1043 struct buffer_head
*bh
;
1045 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1046 return ERR_PTR(-ENAMETOOLONG
);
1048 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
1051 unsigned long ino
= le32_to_cpu(de
->inode
);
1053 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1054 ext4_error(dir
->i_sb
, "ext4_lookup",
1055 "bad inode number: %lu", ino
);
1056 return ERR_PTR(-EIO
);
1058 inode
= ext4_iget(dir
->i_sb
, ino
);
1060 return ERR_CAST(inode
);
1062 return d_splice_alias(inode
, dentry
);
1066 struct dentry
*ext4_get_parent(struct dentry
*child
)
1069 struct inode
*inode
;
1070 static const struct qstr dotdot
= {
1074 struct ext4_dir_entry_2
* de
;
1075 struct buffer_head
*bh
;
1077 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
);
1080 return ERR_PTR(-ENOENT
);
1081 ino
= le32_to_cpu(de
->inode
);
1084 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1085 ext4_error(child
->d_inode
->i_sb
, "ext4_get_parent",
1086 "bad inode number: %lu", ino
);
1087 return ERR_PTR(-EIO
);
1090 return d_obtain_alias(ext4_iget(child
->d_inode
->i_sb
, ino
));
1094 static unsigned char ext4_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1095 [S_IFREG
>> S_SHIFT
] = EXT4_FT_REG_FILE
,
1096 [S_IFDIR
>> S_SHIFT
] = EXT4_FT_DIR
,
1097 [S_IFCHR
>> S_SHIFT
] = EXT4_FT_CHRDEV
,
1098 [S_IFBLK
>> S_SHIFT
] = EXT4_FT_BLKDEV
,
1099 [S_IFIFO
>> S_SHIFT
] = EXT4_FT_FIFO
,
1100 [S_IFSOCK
>> S_SHIFT
] = EXT4_FT_SOCK
,
1101 [S_IFLNK
>> S_SHIFT
] = EXT4_FT_SYMLINK
,
1104 static inline void ext4_set_de_type(struct super_block
*sb
,
1105 struct ext4_dir_entry_2
*de
,
1107 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FILETYPE
))
1108 de
->file_type
= ext4_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1112 * Move count entries from end of map between two memory locations.
1113 * Returns pointer to last entry moved.
1115 static struct ext4_dir_entry_2
*
1116 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1118 unsigned rec_len
= 0;
1121 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*) (from
+ map
->offs
);
1122 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1123 memcpy (to
, de
, rec_len
);
1124 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1125 ext4_rec_len_to_disk(rec_len
);
1130 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1134 * Compact each dir entry in the range to the minimal rec_len.
1135 * Returns pointer to last entry in range.
1137 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1139 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1140 unsigned rec_len
= 0;
1143 while ((char*)de
< base
+ size
) {
1144 next
= ext4_next_entry(de
);
1145 if (de
->inode
&& de
->name_len
) {
1146 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1148 memmove(to
, de
, rec_len
);
1149 to
->rec_len
= ext4_rec_len_to_disk(rec_len
);
1151 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1159 * Split a full leaf block to make room for a new dir entry.
1160 * Allocate a new block, and move entries so that they are approx. equally full.
1161 * Returns pointer to de in block into which the new entry will be inserted.
1163 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1164 struct buffer_head
**bh
,struct dx_frame
*frame
,
1165 struct dx_hash_info
*hinfo
, int *error
)
1167 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1168 unsigned count
, continued
;
1169 struct buffer_head
*bh2
;
1170 ext4_lblk_t newblock
;
1172 struct dx_map_entry
*map
;
1173 char *data1
= (*bh
)->b_data
, *data2
;
1174 unsigned split
, move
, size
, i
;
1175 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1178 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1185 BUFFER_TRACE(*bh
, "get_write_access");
1186 err
= ext4_journal_get_write_access(handle
, *bh
);
1190 BUFFER_TRACE(frame
->bh
, "get_write_access");
1191 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1195 data2
= bh2
->b_data
;
1197 /* create map in the end of data2 block */
1198 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1199 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1200 blocksize
, hinfo
, map
);
1202 dx_sort_map(map
, count
);
1203 /* Split the existing block in the middle, size-wise */
1206 for (i
= count
-1; i
>= 0; i
--) {
1207 /* is more than half of this entry in 2nd half of the block? */
1208 if (size
+ map
[i
].size
/2 > blocksize
/2)
1210 size
+= map
[i
].size
;
1213 /* map index at which we will split */
1214 split
= count
- move
;
1215 hash2
= map
[split
].hash
;
1216 continued
= hash2
== map
[split
- 1].hash
;
1217 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1218 (unsigned long)dx_get_block(frame
->at
),
1219 hash2
, split
, count
-split
));
1221 /* Fancy dance to stay within two buffers */
1222 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1223 de
= dx_pack_dirents(data1
, blocksize
);
1224 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1225 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1226 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1227 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1229 /* Which block gets the new entry? */
1230 if (hinfo
->hash
>= hash2
)
1235 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1236 err
= ext4_journal_dirty_metadata(handle
, bh2
);
1239 err
= ext4_journal_dirty_metadata(handle
, frame
->bh
);
1243 dxtrace(dx_show_index("frame", frame
->entries
));
1250 ext4_std_error(dir
->i_sb
, err
);
1257 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1258 * it points to a directory entry which is guaranteed to be large
1259 * enough for new directory entry. If de is NULL, then
1260 * add_dirent_to_buf will attempt search the directory block for
1261 * space. It will return -ENOSPC if no space is available, and -EIO
1262 * and -EEXIST if directory entry already exists.
1264 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1265 * all other cases bh is released.
1267 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1268 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1269 struct buffer_head
*bh
)
1271 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1272 const char *name
= dentry
->d_name
.name
;
1273 int namelen
= dentry
->d_name
.len
;
1274 unsigned long offset
= 0;
1275 unsigned short reclen
;
1276 int nlen
, rlen
, err
;
1279 reclen
= EXT4_DIR_REC_LEN(namelen
);
1281 de
= (struct ext4_dir_entry_2
*)bh
->b_data
;
1282 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1283 while ((char *) de
<= top
) {
1284 if (!ext4_check_dir_entry("ext4_add_entry", dir
, de
,
1289 if (ext4_match(namelen
, name
, de
)) {
1293 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1294 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1295 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1297 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1300 if ((char *) de
> top
)
1303 BUFFER_TRACE(bh
, "get_write_access");
1304 err
= ext4_journal_get_write_access(handle
, bh
);
1306 ext4_std_error(dir
->i_sb
, err
);
1311 /* By now the buffer is marked for journaling */
1312 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1313 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1315 struct ext4_dir_entry_2
*de1
= (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1316 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
);
1317 de
->rec_len
= ext4_rec_len_to_disk(nlen
);
1320 de
->file_type
= EXT4_FT_UNKNOWN
;
1322 de
->inode
= cpu_to_le32(inode
->i_ino
);
1323 ext4_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1326 de
->name_len
= namelen
;
1327 memcpy(de
->name
, name
, namelen
);
1329 * XXX shouldn't update any times until successful
1330 * completion of syscall, but too many callers depend
1333 * XXX similarly, too many callers depend on
1334 * ext4_new_inode() setting the times, but error
1335 * recovery deletes the inode, so the worst that can
1336 * happen is that the times are slightly out of date
1337 * and/or different from the directory change time.
1339 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1340 ext4_update_dx_flag(dir
);
1342 ext4_mark_inode_dirty(handle
, dir
);
1343 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1344 err
= ext4_journal_dirty_metadata(handle
, bh
);
1346 ext4_std_error(dir
->i_sb
, err
);
1352 * This converts a one block unindexed directory to a 3 block indexed
1353 * directory, and adds the dentry to the indexed directory.
1355 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1356 struct inode
*inode
, struct buffer_head
*bh
)
1358 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1359 const char *name
= dentry
->d_name
.name
;
1360 int namelen
= dentry
->d_name
.len
;
1361 struct buffer_head
*bh2
;
1362 struct dx_root
*root
;
1363 struct dx_frame frames
[2], *frame
;
1364 struct dx_entry
*entries
;
1365 struct ext4_dir_entry_2
*de
, *de2
;
1370 struct dx_hash_info hinfo
;
1372 struct fake_dirent
*fde
;
1374 blocksize
= dir
->i_sb
->s_blocksize
;
1375 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1376 retval
= ext4_journal_get_write_access(handle
, bh
);
1378 ext4_std_error(dir
->i_sb
, retval
);
1382 root
= (struct dx_root
*) bh
->b_data
;
1384 /* The 0th block becomes the root, move the dirents out */
1385 fde
= &root
->dotdot
;
1386 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1387 ext4_rec_len_from_disk(fde
->rec_len
));
1388 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1389 ext4_error(dir
->i_sb
, __func__
,
1390 "invalid rec_len for '..' in inode %lu",
1395 len
= ((char *) root
) + blocksize
- (char *) de
;
1397 /* Allocate new block for the 0th block's dirents */
1398 bh2
= ext4_append(handle
, dir
, &block
, &retval
);
1403 EXT4_I(dir
)->i_flags
|= EXT4_INDEX_FL
;
1404 data1
= bh2
->b_data
;
1406 memcpy (data1
, de
, len
);
1407 de
= (struct ext4_dir_entry_2
*) data1
;
1409 while ((char *)(de2
= ext4_next_entry(de
)) < top
)
1411 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1412 /* Initialize the root; the dot dirents already exist */
1413 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1414 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2));
1415 memset (&root
->info
, 0, sizeof(root
->info
));
1416 root
->info
.info_length
= sizeof(root
->info
);
1417 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1418 entries
= root
->entries
;
1419 dx_set_block(entries
, 1);
1420 dx_set_count(entries
, 1);
1421 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1423 /* Initialize as for dx_probe */
1424 hinfo
.hash_version
= root
->info
.hash_version
;
1425 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1426 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1427 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1428 ext4fs_dirhash(name
, namelen
, &hinfo
);
1430 frame
->entries
= entries
;
1431 frame
->at
= entries
;
1434 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1435 dx_release (frames
);
1439 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1445 * adds a file entry to the specified directory, using the same
1446 * semantics as ext4_find_entry(). It returns NULL if it failed.
1448 * NOTE!! The inode part of 'de' is left at 0 - which means you
1449 * may not sleep between calling this and putting something into
1450 * the entry, as someone else might have used it while you slept.
1452 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1453 struct inode
*inode
)
1455 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1456 unsigned long offset
;
1457 struct buffer_head
*bh
;
1458 struct ext4_dir_entry_2
*de
;
1459 struct super_block
*sb
;
1463 ext4_lblk_t block
, blocks
;
1466 blocksize
= sb
->s_blocksize
;
1467 if (!dentry
->d_name
.len
)
1470 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1471 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1473 EXT4_I(dir
)->i_flags
&= ~EXT4_INDEX_FL
;
1475 ext4_mark_inode_dirty(handle
, dir
);
1477 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1478 for (block
= 0, offset
= 0; block
< blocks
; block
++) {
1479 bh
= ext4_bread(handle
, dir
, block
, 0, &retval
);
1482 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1483 if (retval
!= -ENOSPC
)
1486 if (blocks
== 1 && !dx_fallback
&&
1487 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1488 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1491 bh
= ext4_append(handle
, dir
, &block
, &retval
);
1494 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1496 de
->rec_len
= ext4_rec_len_to_disk(blocksize
);
1497 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1501 * Returns 0 for success, or a negative error value
1503 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1504 struct inode
*inode
)
1506 struct dx_frame frames
[2], *frame
;
1507 struct dx_entry
*entries
, *at
;
1508 struct dx_hash_info hinfo
;
1509 struct buffer_head
*bh
;
1510 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1511 struct super_block
*sb
= dir
->i_sb
;
1512 struct ext4_dir_entry_2
*de
;
1515 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1518 entries
= frame
->entries
;
1521 if (!(bh
= ext4_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1524 BUFFER_TRACE(bh
, "get_write_access");
1525 err
= ext4_journal_get_write_access(handle
, bh
);
1529 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1530 if (err
!= -ENOSPC
) {
1535 /* Block full, should compress but for now just split */
1536 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
1537 dx_get_count(entries
), dx_get_limit(entries
)));
1538 /* Need to split index? */
1539 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1540 ext4_lblk_t newblock
;
1541 unsigned icount
= dx_get_count(entries
);
1542 int levels
= frame
- frames
;
1543 struct dx_entry
*entries2
;
1544 struct dx_node
*node2
;
1545 struct buffer_head
*bh2
;
1547 if (levels
&& (dx_get_count(frames
->entries
) ==
1548 dx_get_limit(frames
->entries
))) {
1549 ext4_warning(sb
, __func__
,
1550 "Directory index full!");
1554 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1557 node2
= (struct dx_node
*)(bh2
->b_data
);
1558 entries2
= node2
->entries
;
1559 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
);
1560 node2
->fake
.inode
= 0;
1561 BUFFER_TRACE(frame
->bh
, "get_write_access");
1562 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1566 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1567 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1568 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
1571 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1572 err
= ext4_journal_get_write_access(handle
,
1577 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
1578 icount2
* sizeof(struct dx_entry
));
1579 dx_set_count(entries
, icount1
);
1580 dx_set_count(entries2
, icount2
);
1581 dx_set_limit(entries2
, dx_node_limit(dir
));
1583 /* Which index block gets the new entry? */
1584 if (at
- entries
>= icount1
) {
1585 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1586 frame
->entries
= entries
= entries2
;
1587 swap(frame
->bh
, bh2
);
1589 dx_insert_block(frames
+ 0, hash2
, newblock
);
1590 dxtrace(dx_show_index("node", frames
[1].entries
));
1591 dxtrace(dx_show_index("node",
1592 ((struct dx_node
*) bh2
->b_data
)->entries
));
1593 err
= ext4_journal_dirty_metadata(handle
, bh2
);
1598 dxtrace(printk(KERN_DEBUG
1599 "Creating second level index...\n"));
1600 memcpy((char *) entries2
, (char *) entries
,
1601 icount
* sizeof(struct dx_entry
));
1602 dx_set_limit(entries2
, dx_node_limit(dir
));
1605 dx_set_count(entries
, 1);
1606 dx_set_block(entries
+ 0, newblock
);
1607 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1609 /* Add new access path frame */
1611 frame
->at
= at
= at
- entries
+ entries2
;
1612 frame
->entries
= entries
= entries2
;
1614 err
= ext4_journal_get_write_access(handle
,
1619 ext4_journal_dirty_metadata(handle
, frames
[0].bh
);
1621 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1624 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1629 ext4_std_error(dir
->i_sb
, err
);
1638 * ext4_delete_entry deletes a directory entry by merging it with the
1641 static int ext4_delete_entry(handle_t
*handle
,
1643 struct ext4_dir_entry_2
*de_del
,
1644 struct buffer_head
*bh
)
1646 struct ext4_dir_entry_2
*de
, *pde
;
1651 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1652 while (i
< bh
->b_size
) {
1653 if (!ext4_check_dir_entry("ext4_delete_entry", dir
, de
, bh
, i
))
1656 BUFFER_TRACE(bh
, "get_write_access");
1657 ext4_journal_get_write_access(handle
, bh
);
1659 pde
->rec_len
= ext4_rec_len_to_disk(
1660 ext4_rec_len_from_disk(pde
->rec_len
) +
1661 ext4_rec_len_from_disk(de
->rec_len
));
1665 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1666 ext4_journal_dirty_metadata(handle
, bh
);
1669 i
+= ext4_rec_len_from_disk(de
->rec_len
);
1671 de
= ext4_next_entry(de
);
1677 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1678 * since this indicates that nlinks count was previously 1.
1680 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
1683 if (is_dx(inode
) && inode
->i_nlink
> 1) {
1684 /* limit is 16-bit i_links_count */
1685 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
1687 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
1688 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
1694 * If a directory had nlink == 1, then we should let it be 1. This indicates
1695 * directory has >EXT4_LINK_MAX subdirs.
1697 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
1700 if (S_ISDIR(inode
->i_mode
) && inode
->i_nlink
== 0)
1705 static int ext4_add_nondir(handle_t
*handle
,
1706 struct dentry
*dentry
, struct inode
*inode
)
1708 int err
= ext4_add_entry(handle
, dentry
, inode
);
1710 ext4_mark_inode_dirty(handle
, inode
);
1711 d_instantiate(dentry
, inode
);
1720 * By the time this is called, we already have created
1721 * the directory cache entry for the new file, but it
1722 * is so far negative - it has no inode.
1724 * If the create succeeds, we fill in the inode information
1725 * with d_instantiate().
1727 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1728 struct nameidata
*nd
)
1731 struct inode
*inode
;
1732 int err
, retries
= 0;
1735 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1736 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1737 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1739 return PTR_ERR(handle
);
1741 if (IS_DIRSYNC(dir
))
1744 inode
= ext4_new_inode (handle
, dir
, mode
);
1745 err
= PTR_ERR(inode
);
1746 if (!IS_ERR(inode
)) {
1747 inode
->i_op
= &ext4_file_inode_operations
;
1748 inode
->i_fop
= &ext4_file_operations
;
1749 ext4_set_aops(inode
);
1750 err
= ext4_add_nondir(handle
, dentry
, inode
);
1752 ext4_journal_stop(handle
);
1753 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1758 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
1759 int mode
, dev_t rdev
)
1762 struct inode
*inode
;
1763 int err
, retries
= 0;
1765 if (!new_valid_dev(rdev
))
1769 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1770 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1771 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1773 return PTR_ERR(handle
);
1775 if (IS_DIRSYNC(dir
))
1778 inode
= ext4_new_inode(handle
, dir
, mode
);
1779 err
= PTR_ERR(inode
);
1780 if (!IS_ERR(inode
)) {
1781 init_special_inode(inode
, inode
->i_mode
, rdev
);
1782 #ifdef CONFIG_EXT4_FS_XATTR
1783 inode
->i_op
= &ext4_special_inode_operations
;
1785 err
= ext4_add_nondir(handle
, dentry
, inode
);
1787 ext4_journal_stop(handle
);
1788 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1793 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1796 struct inode
*inode
;
1797 struct buffer_head
*dir_block
;
1798 struct ext4_dir_entry_2
*de
;
1799 int err
, retries
= 0;
1801 if (EXT4_DIR_LINK_MAX(dir
))
1805 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1806 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1807 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1809 return PTR_ERR(handle
);
1811 if (IS_DIRSYNC(dir
))
1814 inode
= ext4_new_inode(handle
, dir
, S_IFDIR
| mode
);
1815 err
= PTR_ERR(inode
);
1819 inode
->i_op
= &ext4_dir_inode_operations
;
1820 inode
->i_fop
= &ext4_dir_operations
;
1821 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1822 dir_block
= ext4_bread(handle
, inode
, 0, 1, &err
);
1824 goto out_clear_inode
;
1825 BUFFER_TRACE(dir_block
, "get_write_access");
1826 ext4_journal_get_write_access(handle
, dir_block
);
1827 de
= (struct ext4_dir_entry_2
*) dir_block
->b_data
;
1828 de
->inode
= cpu_to_le32(inode
->i_ino
);
1830 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
));
1831 strcpy(de
->name
, ".");
1832 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1833 de
= ext4_next_entry(de
);
1834 de
->inode
= cpu_to_le32(dir
->i_ino
);
1835 de
->rec_len
= ext4_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1836 EXT4_DIR_REC_LEN(1));
1838 strcpy(de
->name
, "..");
1839 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1841 BUFFER_TRACE(dir_block
, "call ext4_journal_dirty_metadata");
1842 ext4_journal_dirty_metadata(handle
, dir_block
);
1844 ext4_mark_inode_dirty(handle
, inode
);
1845 err
= ext4_add_entry(handle
, dentry
, inode
);
1849 ext4_mark_inode_dirty(handle
, inode
);
1853 ext4_inc_count(handle
, dir
);
1854 ext4_update_dx_flag(dir
);
1855 ext4_mark_inode_dirty(handle
, dir
);
1856 d_instantiate(dentry
, inode
);
1858 ext4_journal_stop(handle
);
1859 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1865 * routine to check that the specified directory is empty (for rmdir)
1867 static int empty_dir(struct inode
*inode
)
1869 unsigned long offset
;
1870 struct buffer_head
*bh
;
1871 struct ext4_dir_entry_2
*de
, *de1
;
1872 struct super_block
*sb
;
1876 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1877 !(bh
= ext4_bread(NULL
, inode
, 0, 0, &err
))) {
1879 ext4_error(inode
->i_sb
, __func__
,
1880 "error %d reading directory #%lu offset 0",
1883 ext4_warning(inode
->i_sb
, __func__
,
1884 "bad directory (dir #%lu) - no data block",
1888 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1889 de1
= ext4_next_entry(de
);
1890 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1891 !le32_to_cpu(de1
->inode
) ||
1892 strcmp(".", de
->name
) ||
1893 strcmp("..", de1
->name
)) {
1894 ext4_warning(inode
->i_sb
, "empty_dir",
1895 "bad directory (dir #%lu) - no `.' or `..'",
1900 offset
= ext4_rec_len_from_disk(de
->rec_len
) +
1901 ext4_rec_len_from_disk(de1
->rec_len
);
1902 de
= ext4_next_entry(de1
);
1903 while (offset
< inode
->i_size
) {
1905 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1908 bh
= ext4_bread(NULL
, inode
,
1909 offset
>> EXT4_BLOCK_SIZE_BITS(sb
), 0, &err
);
1912 ext4_error(sb
, __func__
,
1913 "error %d reading directory"
1915 err
, inode
->i_ino
, offset
);
1916 offset
+= sb
->s_blocksize
;
1919 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1921 if (!ext4_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1922 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
1924 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1927 if (le32_to_cpu(de
->inode
)) {
1931 offset
+= ext4_rec_len_from_disk(de
->rec_len
);
1932 de
= ext4_next_entry(de
);
1938 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1939 * such inodes, starting at the superblock, in case we crash before the
1940 * file is closed/deleted, or in case the inode truncate spans multiple
1941 * transactions and the last transaction is not recovered after a crash.
1943 * At filesystem recovery time, we walk this list deleting unlinked
1944 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1946 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
1948 struct super_block
*sb
= inode
->i_sb
;
1949 struct ext4_iloc iloc
;
1953 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
1956 /* Orphan handling is only valid for files with data blocks
1957 * being truncated, or files being unlinked. */
1959 /* @@@ FIXME: Observation from aviro:
1960 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1961 * here (on lock_super()), so race with ext4_link() which might bump
1962 * ->i_nlink. For, say it, character device. Not a regular file,
1963 * not a directory, not a symlink and ->i_nlink > 0.
1965 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1966 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1968 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1969 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1973 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
1977 /* Insert this inode at the head of the on-disk orphan list... */
1978 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT4_SB(sb
)->s_es
->s_last_orphan
);
1979 EXT4_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1980 err
= ext4_journal_dirty_metadata(handle
, EXT4_SB(sb
)->s_sbh
);
1981 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1985 /* Only add to the head of the in-memory list if all the
1986 * previous operations succeeded. If the orphan_add is going to
1987 * fail (possibly taking the journal offline), we can't risk
1988 * leaving the inode on the orphan list: stray orphan-list
1989 * entries can cause panics at unmount time.
1991 * This is safe: on error we're going to ignore the orphan list
1992 * anyway on the next recovery. */
1994 list_add(&EXT4_I(inode
)->i_orphan
, &EXT4_SB(sb
)->s_orphan
);
1996 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1997 jbd_debug(4, "orphan inode %lu will point to %d\n",
1998 inode
->i_ino
, NEXT_ORPHAN(inode
));
2001 ext4_std_error(inode
->i_sb
, err
);
2006 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2007 * of such inodes stored on disk, because it is finally being cleaned up.
2009 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2011 struct list_head
*prev
;
2012 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2013 struct ext4_sb_info
*sbi
;
2014 unsigned long ino_next
;
2015 struct ext4_iloc iloc
;
2018 lock_super(inode
->i_sb
);
2019 if (list_empty(&ei
->i_orphan
)) {
2020 unlock_super(inode
->i_sb
);
2024 ino_next
= NEXT_ORPHAN(inode
);
2025 prev
= ei
->i_orphan
.prev
;
2026 sbi
= EXT4_SB(inode
->i_sb
);
2028 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2030 list_del_init(&ei
->i_orphan
);
2032 /* If we're on an error path, we may not have a valid
2033 * transaction handle with which to update the orphan list on
2034 * disk, but we still need to remove the inode from the linked
2035 * list in memory. */
2039 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2043 if (prev
== &sbi
->s_orphan
) {
2044 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
2045 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2046 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2049 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2050 err
= ext4_journal_dirty_metadata(handle
, sbi
->s_sbh
);
2052 struct ext4_iloc iloc2
;
2053 struct inode
*i_prev
=
2054 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2056 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2057 i_prev
->i_ino
, ino_next
);
2058 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2061 NEXT_ORPHAN(i_prev
) = ino_next
;
2062 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2066 NEXT_ORPHAN(inode
) = 0;
2067 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2070 ext4_std_error(inode
->i_sb
, err
);
2072 unlock_super(inode
->i_sb
);
2080 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2083 struct inode
*inode
;
2084 struct buffer_head
*bh
;
2085 struct ext4_dir_entry_2
*de
;
2088 /* Initialize quotas before so that eventual writes go in
2089 * separate transaction */
2090 DQUOT_INIT(dentry
->d_inode
);
2091 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2093 return PTR_ERR(handle
);
2096 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2100 if (IS_DIRSYNC(dir
))
2103 inode
= dentry
->d_inode
;
2106 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2109 retval
= -ENOTEMPTY
;
2110 if (!empty_dir(inode
))
2113 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2116 if (!EXT4_DIR_LINK_EMPTY(inode
))
2117 ext4_warning(inode
->i_sb
, "ext4_rmdir",
2118 "empty directory has too many links (%d)",
2122 /* There's no need to set i_disksize: the fact that i_nlink is
2123 * zero will ensure that the right thing happens during any
2126 ext4_orphan_add(handle
, inode
);
2127 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2128 ext4_mark_inode_dirty(handle
, inode
);
2129 ext4_dec_count(handle
, dir
);
2130 ext4_update_dx_flag(dir
);
2131 ext4_mark_inode_dirty(handle
, dir
);
2134 ext4_journal_stop(handle
);
2139 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2142 struct inode
*inode
;
2143 struct buffer_head
*bh
;
2144 struct ext4_dir_entry_2
*de
;
2147 /* Initialize quotas before so that eventual writes go
2148 * in separate transaction */
2149 DQUOT_INIT(dentry
->d_inode
);
2150 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2152 return PTR_ERR(handle
);
2154 if (IS_DIRSYNC(dir
))
2158 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2162 inode
= dentry
->d_inode
;
2165 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2168 if (!inode
->i_nlink
) {
2169 ext4_warning(inode
->i_sb
, "ext4_unlink",
2170 "Deleting nonexistent file (%lu), %d",
2171 inode
->i_ino
, inode
->i_nlink
);
2174 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2177 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2178 ext4_update_dx_flag(dir
);
2179 ext4_mark_inode_dirty(handle
, dir
);
2181 if (!inode
->i_nlink
)
2182 ext4_orphan_add(handle
, inode
);
2183 inode
->i_ctime
= ext4_current_time(inode
);
2184 ext4_mark_inode_dirty(handle
, inode
);
2188 ext4_journal_stop(handle
);
2193 static int ext4_symlink(struct inode
*dir
,
2194 struct dentry
*dentry
, const char *symname
)
2197 struct inode
*inode
;
2198 int l
, err
, retries
= 0;
2200 l
= strlen(symname
)+1;
2201 if (l
> dir
->i_sb
->s_blocksize
)
2202 return -ENAMETOOLONG
;
2205 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2206 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2207 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2209 return PTR_ERR(handle
);
2211 if (IS_DIRSYNC(dir
))
2214 inode
= ext4_new_inode(handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2215 err
= PTR_ERR(inode
);
2219 if (l
> sizeof(EXT4_I(inode
)->i_data
)) {
2220 inode
->i_op
= &ext4_symlink_inode_operations
;
2221 ext4_set_aops(inode
);
2223 * page_symlink() calls into ext4_prepare/commit_write.
2224 * We have a transaction open. All is sweetness. It also sets
2225 * i_size in generic_commit_write().
2227 err
= __page_symlink(inode
, symname
, l
, 1);
2230 ext4_mark_inode_dirty(handle
, inode
);
2235 /* clear the extent format for fast symlink */
2236 EXT4_I(inode
)->i_flags
&= ~EXT4_EXTENTS_FL
;
2237 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2238 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2239 inode
->i_size
= l
-1;
2241 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2242 err
= ext4_add_nondir(handle
, dentry
, inode
);
2244 ext4_journal_stop(handle
);
2245 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2250 static int ext4_link(struct dentry
*old_dentry
,
2251 struct inode
*dir
, struct dentry
*dentry
)
2254 struct inode
*inode
= old_dentry
->d_inode
;
2255 int err
, retries
= 0;
2257 if (EXT4_DIR_LINK_MAX(inode
))
2261 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2262 * otherwise has the potential to corrupt the orphan inode list.
2264 if (inode
->i_nlink
== 0)
2268 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2269 EXT4_INDEX_EXTRA_TRANS_BLOCKS
);
2271 return PTR_ERR(handle
);
2273 if (IS_DIRSYNC(dir
))
2276 inode
->i_ctime
= ext4_current_time(inode
);
2277 ext4_inc_count(handle
, inode
);
2278 atomic_inc(&inode
->i_count
);
2280 err
= ext4_add_nondir(handle
, dentry
, inode
);
2281 ext4_journal_stop(handle
);
2282 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2287 #define PARENT_INO(buffer) \
2288 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2291 * Anybody can rename anything with this: the permission checks are left to the
2292 * higher-level routines.
2294 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2295 struct inode
*new_dir
, struct dentry
*new_dentry
)
2298 struct inode
*old_inode
, *new_inode
;
2299 struct buffer_head
*old_bh
, *new_bh
, *dir_bh
;
2300 struct ext4_dir_entry_2
*old_de
, *new_de
;
2303 old_bh
= new_bh
= dir_bh
= NULL
;
2305 /* Initialize quotas before so that eventual writes go
2306 * in separate transaction */
2307 if (new_dentry
->d_inode
)
2308 DQUOT_INIT(new_dentry
->d_inode
);
2309 handle
= ext4_journal_start(old_dir
, 2 *
2310 EXT4_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2311 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2313 return PTR_ERR(handle
);
2315 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2318 old_bh
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2320 * Check for inode number is _not_ due to possible IO errors.
2321 * We might rmdir the source, keep it as pwd of some process
2322 * and merrily kill the link to whatever was created under the
2323 * same name. Goodbye sticky bit ;-<
2325 old_inode
= old_dentry
->d_inode
;
2327 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2330 new_inode
= new_dentry
->d_inode
;
2331 new_bh
= ext4_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
2338 if (S_ISDIR(old_inode
->i_mode
)) {
2340 retval
= -ENOTEMPTY
;
2341 if (!empty_dir(new_inode
))
2345 dir_bh
= ext4_bread(handle
, old_inode
, 0, 0, &retval
);
2348 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2351 if (!new_inode
&& new_dir
!= old_dir
&&
2352 new_dir
->i_nlink
>= EXT4_LINK_MAX
)
2356 retval
= ext4_add_entry(handle
, new_dentry
, old_inode
);
2360 BUFFER_TRACE(new_bh
, "get write access");
2361 ext4_journal_get_write_access(handle
, new_bh
);
2362 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2363 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2364 EXT4_FEATURE_INCOMPAT_FILETYPE
))
2365 new_de
->file_type
= old_de
->file_type
;
2366 new_dir
->i_version
++;
2367 new_dir
->i_ctime
= new_dir
->i_mtime
=
2368 ext4_current_time(new_dir
);
2369 ext4_mark_inode_dirty(handle
, new_dir
);
2370 BUFFER_TRACE(new_bh
, "call ext4_journal_dirty_metadata");
2371 ext4_journal_dirty_metadata(handle
, new_bh
);
2377 * Like most other Unix systems, set the ctime for inodes on a
2380 old_inode
->i_ctime
= ext4_current_time(old_inode
);
2381 ext4_mark_inode_dirty(handle
, old_inode
);
2386 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2387 old_de
->name_len
!= old_dentry
->d_name
.len
||
2388 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2389 (retval
= ext4_delete_entry(handle
, old_dir
,
2390 old_de
, old_bh
)) == -ENOENT
) {
2391 /* old_de could have moved from under us during htree split, so
2392 * make sure that we are deleting the right entry. We might
2393 * also be pointing to a stale entry in the unused part of
2394 * old_bh so just checking inum and the name isn't enough. */
2395 struct buffer_head
*old_bh2
;
2396 struct ext4_dir_entry_2
*old_de2
;
2398 old_bh2
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de2
);
2400 retval
= ext4_delete_entry(handle
, old_dir
,
2406 ext4_warning(old_dir
->i_sb
, "ext4_rename",
2407 "Deleting old file (%lu), %d, error=%d",
2408 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2412 ext4_dec_count(handle
, new_inode
);
2413 new_inode
->i_ctime
= ext4_current_time(new_inode
);
2415 old_dir
->i_ctime
= old_dir
->i_mtime
= ext4_current_time(old_dir
);
2416 ext4_update_dx_flag(old_dir
);
2418 BUFFER_TRACE(dir_bh
, "get_write_access");
2419 ext4_journal_get_write_access(handle
, dir_bh
);
2420 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2421 BUFFER_TRACE(dir_bh
, "call ext4_journal_dirty_metadata");
2422 ext4_journal_dirty_metadata(handle
, dir_bh
);
2423 ext4_dec_count(handle
, old_dir
);
2425 /* checked empty_dir above, can't have another parent,
2426 * ext4_dec_count() won't work for many-linked dirs */
2427 new_inode
->i_nlink
= 0;
2429 ext4_inc_count(handle
, new_dir
);
2430 ext4_update_dx_flag(new_dir
);
2431 ext4_mark_inode_dirty(handle
, new_dir
);
2434 ext4_mark_inode_dirty(handle
, old_dir
);
2436 ext4_mark_inode_dirty(handle
, new_inode
);
2437 if (!new_inode
->i_nlink
)
2438 ext4_orphan_add(handle
, new_inode
);
2446 ext4_journal_stop(handle
);
2451 * directories can handle most operations...
2453 const struct inode_operations ext4_dir_inode_operations
= {
2454 .create
= ext4_create
,
2455 .lookup
= ext4_lookup
,
2457 .unlink
= ext4_unlink
,
2458 .symlink
= ext4_symlink
,
2459 .mkdir
= ext4_mkdir
,
2460 .rmdir
= ext4_rmdir
,
2461 .mknod
= ext4_mknod
,
2462 .rename
= ext4_rename
,
2463 .setattr
= ext4_setattr
,
2464 #ifdef CONFIG_EXT4_FS_XATTR
2465 .setxattr
= generic_setxattr
,
2466 .getxattr
= generic_getxattr
,
2467 .listxattr
= ext4_listxattr
,
2468 .removexattr
= generic_removexattr
,
2470 .permission
= ext4_permission
,
2473 const struct inode_operations ext4_special_inode_operations
= {
2474 .setattr
= ext4_setattr
,
2475 #ifdef CONFIG_EXT4_FS_XATTR
2476 .setxattr
= generic_setxattr
,
2477 .getxattr
= generic_getxattr
,
2478 .listxattr
= ext4_listxattr
,
2479 .removexattr
= generic_removexattr
,
2481 .permission
= ext4_permission
,