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 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 ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
149 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t 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(const struct qstr
*d_name
,
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 ext4_dir_entry_2
*de
, int size
,
165 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
166 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
167 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
168 struct dx_map_entry
*offsets
, int count
);
169 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, int size
);
170 static void dx_insert_block(struct dx_frame
*frame
,
171 u32 hash
, ext4_lblk_t block
);
172 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
173 struct dx_frame
*frame
,
174 struct dx_frame
*frames
,
176 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
177 const struct qstr
*d_name
,
178 struct ext4_dir_entry_2
**res_dir
,
180 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
181 struct inode
*inode
);
184 * p is at least 6 bytes before the end of page
186 static inline struct ext4_dir_entry_2
*
187 ext4_next_entry(struct ext4_dir_entry_2
*p
)
189 return (struct ext4_dir_entry_2
*)((char *)p
+
190 ext4_rec_len_from_disk(p
->rec_len
));
194 * Future: use high four bits of block for coalesce-on-delete flags
195 * Mask them off for now.
198 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
200 return le32_to_cpu(entry
->block
) & 0x00ffffff;
203 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
205 entry
->block
= cpu_to_le32(value
);
208 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
210 return le32_to_cpu(entry
->hash
);
213 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
215 entry
->hash
= cpu_to_le32(value
);
218 static inline unsigned dx_get_count(struct dx_entry
*entries
)
220 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
223 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
225 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
228 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
230 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
233 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
235 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
238 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
240 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
241 EXT4_DIR_REC_LEN(2) - infosize
;
242 return entry_space
/ sizeof(struct dx_entry
);
245 static inline unsigned dx_node_limit(struct inode
*dir
)
247 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
248 return entry_space
/ sizeof(struct dx_entry
);
255 static void dx_show_index(char * label
, struct dx_entry
*entries
)
257 int i
, n
= dx_get_count (entries
);
258 printk(KERN_DEBUG
"%s index ", label
);
259 for (i
= 0; i
< n
; i
++) {
260 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
261 0, (unsigned long)dx_get_block(entries
+ i
));
273 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
274 int size
, int show_names
)
276 unsigned names
= 0, space
= 0;
277 char *base
= (char *) de
;
278 struct dx_hash_info h
= *hinfo
;
281 while ((char *) de
< base
+ size
)
287 int len
= de
->name_len
;
288 char *name
= de
->name
;
289 while (len
--) printk("%c", *name
++);
290 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
291 printk(":%x.%u ", h
.hash
,
292 ((char *) de
- base
));
294 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
297 de
= ext4_next_entry(de
);
299 printk("(%i)\n", names
);
300 return (struct stats
) { names
, space
, 1 };
303 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
304 struct dx_entry
*entries
, int levels
)
306 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
307 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
309 struct buffer_head
*bh
;
311 printk("%i indexed blocks...\n", count
);
312 for (i
= 0; i
< count
; i
++, entries
++)
314 ext4_lblk_t block
= dx_get_block(entries
);
315 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
316 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
318 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
319 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0,&err
))) continue;
321 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
322 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
323 names
+= stats
.names
;
324 space
+= stats
.space
;
325 bcount
+= stats
.bcount
;
329 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
330 levels
? "" : " ", names
, space
/bcount
,
331 (space
/bcount
)*100/blocksize
);
332 return (struct stats
) { names
, space
, bcount
};
334 #endif /* DX_DEBUG */
337 * Probe for a directory leaf block to search.
339 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
340 * error in the directory index, and the caller should fall back to
341 * searching the directory normally. The callers of dx_probe **MUST**
342 * check for this error code, and make sure it never gets reflected
345 static struct dx_frame
*
346 dx_probe(const struct qstr
*d_name
, struct inode
*dir
,
347 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
349 unsigned count
, indirect
;
350 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
351 struct dx_root
*root
;
352 struct buffer_head
*bh
;
353 struct dx_frame
*frame
= frame_in
;
357 if (!(bh
= ext4_bread (NULL
,dir
, 0, 0, err
)))
359 root
= (struct dx_root
*) bh
->b_data
;
360 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
361 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
362 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
363 ext4_warning(dir
->i_sb
, __func__
,
364 "Unrecognised inode hash code %d",
365 root
->info
.hash_version
);
367 *err
= ERR_BAD_DX_DIR
;
370 hinfo
->hash_version
= root
->info
.hash_version
;
371 if (hinfo
->hash_version
<= DX_HASH_TEA
)
372 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
373 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
375 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
378 if (root
->info
.unused_flags
& 1) {
379 ext4_warning(dir
->i_sb
, __func__
,
380 "Unimplemented inode hash flags: %#06x",
381 root
->info
.unused_flags
);
383 *err
= ERR_BAD_DX_DIR
;
387 if ((indirect
= root
->info
.indirect_levels
) > 1) {
388 ext4_warning(dir
->i_sb
, __func__
,
389 "Unimplemented inode hash depth: %#06x",
390 root
->info
.indirect_levels
);
392 *err
= ERR_BAD_DX_DIR
;
396 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
397 root
->info
.info_length
);
399 if (dx_get_limit(entries
) != dx_root_limit(dir
,
400 root
->info
.info_length
)) {
401 ext4_warning(dir
->i_sb
, __func__
,
402 "dx entry: limit != root limit");
404 *err
= ERR_BAD_DX_DIR
;
408 dxtrace(printk("Look up %x", hash
));
411 count
= dx_get_count(entries
);
412 if (!count
|| count
> dx_get_limit(entries
)) {
413 ext4_warning(dir
->i_sb
, __func__
,
414 "dx entry: no count or count > limit");
416 *err
= ERR_BAD_DX_DIR
;
421 q
= entries
+ count
- 1;
425 dxtrace(printk("."));
426 if (dx_get_hash(m
) > hash
)
432 if (0) // linear search cross check
434 unsigned n
= count
- 1;
438 dxtrace(printk(","));
439 if (dx_get_hash(++at
) > hash
)
445 assert (at
== p
- 1);
449 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
451 frame
->entries
= entries
;
453 if (!indirect
--) return frame
;
454 if (!(bh
= ext4_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
456 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
457 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
458 ext4_warning(dir
->i_sb
, __func__
,
459 "dx entry: limit != node limit");
461 *err
= ERR_BAD_DX_DIR
;
468 while (frame
>= frame_in
) {
473 if (*err
== ERR_BAD_DX_DIR
)
474 ext4_warning(dir
->i_sb
, __func__
,
475 "Corrupt dir inode %ld, running e2fsck is "
476 "recommended.", dir
->i_ino
);
480 static void dx_release (struct dx_frame
*frames
)
482 if (frames
[0].bh
== NULL
)
485 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
486 brelse(frames
[1].bh
);
487 brelse(frames
[0].bh
);
491 * This function increments the frame pointer to search the next leaf
492 * block, and reads in the necessary intervening nodes if the search
493 * should be necessary. Whether or not the search is necessary is
494 * controlled by the hash parameter. If the hash value is even, then
495 * the search is only continued if the next block starts with that
496 * hash value. This is used if we are searching for a specific file.
498 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
500 * This function returns 1 if the caller should continue to search,
501 * or 0 if it should not. If there is an error reading one of the
502 * index blocks, it will a negative error code.
504 * If start_hash is non-null, it will be filled in with the starting
505 * hash of the next page.
507 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
508 struct dx_frame
*frame
,
509 struct dx_frame
*frames
,
513 struct buffer_head
*bh
;
514 int err
, num_frames
= 0;
519 * Find the next leaf page by incrementing the frame pointer.
520 * If we run out of entries in the interior node, loop around and
521 * increment pointer in the parent node. When we break out of
522 * this loop, num_frames indicates the number of interior
523 * nodes need to be read.
526 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
535 * If the hash is 1, then continue only if the next page has a
536 * continuation hash of any value. This is used for readdir
537 * handling. Otherwise, check to see if the hash matches the
538 * desired contiuation hash. If it doesn't, return since
539 * there's no point to read in the successive index pages.
541 bhash
= dx_get_hash(p
->at
);
544 if ((hash
& 1) == 0) {
545 if ((bhash
& ~1) != hash
)
549 * If the hash is HASH_NB_ALWAYS, we always go to the next
550 * block so no check is necessary
552 while (num_frames
--) {
553 if (!(bh
= ext4_bread(NULL
, dir
, dx_get_block(p
->at
),
555 return err
; /* Failure */
559 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
566 * This function fills a red-black tree with information from a
567 * directory block. It returns the number directory entries loaded
568 * into the tree. If there is an error it is returned in err.
570 static int htree_dirblock_to_tree(struct file
*dir_file
,
571 struct inode
*dir
, ext4_lblk_t block
,
572 struct dx_hash_info
*hinfo
,
573 __u32 start_hash
, __u32 start_minor_hash
)
575 struct buffer_head
*bh
;
576 struct ext4_dir_entry_2
*de
, *top
;
579 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
580 (unsigned long)block
));
581 if (!(bh
= ext4_bread (NULL
, dir
, block
, 0, &err
)))
584 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
585 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
586 dir
->i_sb
->s_blocksize
-
587 EXT4_DIR_REC_LEN(0));
588 for (; de
< top
; de
= ext4_next_entry(de
)) {
589 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
590 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
591 +((char *)de
- bh
->b_data
))) {
592 /* On error, skip the f_pos to the next block. */
593 dir_file
->f_pos
= (dir_file
->f_pos
|
594 (dir
->i_sb
->s_blocksize
- 1)) + 1;
598 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
599 if ((hinfo
->hash
< start_hash
) ||
600 ((hinfo
->hash
== start_hash
) &&
601 (hinfo
->minor_hash
< start_minor_hash
)))
605 if ((err
= ext4_htree_store_dirent(dir_file
,
606 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
618 * This function fills a red-black tree with information from a
619 * directory. We start scanning the directory in hash order, starting
620 * at start_hash and start_minor_hash.
622 * This function returns the number of entries inserted into the tree,
623 * or a negative error code.
625 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
626 __u32 start_minor_hash
, __u32
*next_hash
)
628 struct dx_hash_info hinfo
;
629 struct ext4_dir_entry_2
*de
;
630 struct dx_frame frames
[2], *frame
;
637 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
638 start_hash
, start_minor_hash
));
639 dir
= dir_file
->f_path
.dentry
->d_inode
;
640 if (!(EXT4_I(dir
)->i_flags
& EXT4_INDEX_FL
)) {
641 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
642 if (hinfo
.hash_version
<= DX_HASH_TEA
)
643 hinfo
.hash_version
+=
644 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
645 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
646 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
647 start_hash
, start_minor_hash
);
651 hinfo
.hash
= start_hash
;
652 hinfo
.minor_hash
= 0;
653 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
, &err
);
657 /* Add '.' and '..' from the htree header */
658 if (!start_hash
&& !start_minor_hash
) {
659 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
660 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
664 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
665 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
666 de
= ext4_next_entry(de
);
667 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
673 block
= dx_get_block(frame
->at
);
674 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
675 start_hash
, start_minor_hash
);
682 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
683 frame
, frames
, &hashval
);
684 *next_hash
= hashval
;
690 * Stop if: (a) there are no more entries, or
691 * (b) we have inserted at least one entry and the
692 * next hash value is not a continuation
695 (count
&& ((hashval
& 1) == 0)))
699 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
700 "next hash: %x\n", count
, *next_hash
));
709 * Directory block splitting, compacting
713 * Create map of hash values, offsets, and sizes, stored at end of block.
714 * Returns number of entries mapped.
716 static int dx_make_map (struct ext4_dir_entry_2
*de
, int size
,
717 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
720 char *base
= (char *) de
;
721 struct dx_hash_info h
= *hinfo
;
723 while ((char *) de
< base
+ size
)
725 if (de
->name_len
&& de
->inode
) {
726 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
728 map_tail
->hash
= h
.hash
;
729 map_tail
->offs
= (u16
) ((char *) de
- base
);
730 map_tail
->size
= le16_to_cpu(de
->rec_len
);
734 /* XXX: do we need to check rec_len == 0 case? -Chris */
735 de
= ext4_next_entry(de
);
740 /* Sort map by hash value */
741 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
743 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
745 /* Combsort until bubble sort doesn't suck */
748 if (count
- 9 < 2) /* 9, 10 -> 11 */
750 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
751 if (p
->hash
< q
->hash
)
754 /* Garden variety bubble sort */
759 if (q
[1].hash
>= q
[0].hash
)
767 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
769 struct dx_entry
*entries
= frame
->entries
;
770 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
771 int count
= dx_get_count(entries
);
773 assert(count
< dx_get_limit(entries
));
774 assert(old
< entries
+ count
);
775 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
776 dx_set_hash(new, hash
);
777 dx_set_block(new, block
);
778 dx_set_count(entries
, count
+ 1);
781 static void ext4_update_dx_flag(struct inode
*inode
)
783 if (!EXT4_HAS_COMPAT_FEATURE(inode
->i_sb
,
784 EXT4_FEATURE_COMPAT_DIR_INDEX
))
785 EXT4_I(inode
)->i_flags
&= ~EXT4_INDEX_FL
;
789 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
791 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
792 * `de != NULL' is guaranteed by caller.
794 static inline int ext4_match (int len
, const char * const name
,
795 struct ext4_dir_entry_2
* de
)
797 if (len
!= de
->name_len
)
801 return !memcmp(name
, de
->name
, len
);
805 * Returns 0 if not found, -1 on failure, and 1 on success
807 static inline int search_dirblock(struct buffer_head
*bh
,
809 const struct qstr
*d_name
,
811 struct ext4_dir_entry_2
** res_dir
)
813 struct ext4_dir_entry_2
* de
;
816 const char *name
= d_name
->name
;
817 int namelen
= d_name
->len
;
819 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
820 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
821 while ((char *) de
< dlimit
) {
822 /* this code is executed quadratically often */
823 /* do minimal checking `by hand' */
825 if ((char *) de
+ namelen
<= dlimit
&&
826 ext4_match (namelen
, name
, de
)) {
827 /* found a match - just to be sure, do a full check */
828 if (!ext4_check_dir_entry("ext4_find_entry",
829 dir
, de
, bh
, offset
))
834 /* prevent looping on a bad block */
835 de_len
= ext4_rec_len_from_disk(de
->rec_len
);
839 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
848 * finds an entry in the specified directory with the wanted name. It
849 * returns the cache buffer in which the entry was found, and the entry
850 * itself (as a parameter - res_dir). It does NOT read the inode of the
851 * entry - you'll have to do that yourself if you want to.
853 * The returned buffer_head has ->b_count elevated. The caller is expected
854 * to brelse() it when appropriate.
856 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
857 const struct qstr
*d_name
,
858 struct ext4_dir_entry_2
** res_dir
)
860 struct super_block
*sb
;
861 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
862 struct buffer_head
*bh
, *ret
= NULL
;
863 ext4_lblk_t start
, block
, b
;
864 int ra_max
= 0; /* Number of bh's in the readahead
866 int ra_ptr
= 0; /* Current index into readahead
875 namelen
= d_name
->len
;
876 if (namelen
> EXT4_NAME_LEN
)
879 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
, &err
);
881 * On success, or if the error was file not found,
882 * return. Otherwise, fall back to doing a search the
885 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
887 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
890 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
891 start
= EXT4_I(dir
)->i_dir_start_lookup
;
892 if (start
>= nblocks
)
898 * We deal with the read-ahead logic here.
900 if (ra_ptr
>= ra_max
) {
901 /* Refill the readahead buffer */
904 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
906 * Terminate if we reach the end of the
907 * directory and must wrap, or if our
908 * search has finished at this block.
910 if (b
>= nblocks
|| (num
&& block
== start
)) {
911 bh_use
[ra_max
] = NULL
;
915 bh
= ext4_getblk(NULL
, dir
, b
++, 0, &err
);
918 ll_rw_block(READ_META
, 1, &bh
);
921 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
924 if (!buffer_uptodate(bh
)) {
925 /* read error, skip block & hope for the best */
926 ext4_error(sb
, __func__
, "reading directory #%lu "
927 "offset %lu", dir
->i_ino
,
928 (unsigned long)block
);
932 i
= search_dirblock(bh
, dir
, d_name
,
933 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
935 EXT4_I(dir
)->i_dir_start_lookup
= block
;
937 goto cleanup_and_exit
;
941 goto cleanup_and_exit
;
944 if (++block
>= nblocks
)
946 } while (block
!= start
);
949 * If the directory has grown while we were searching, then
950 * search the last part of the directory before giving up.
953 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
954 if (block
< nblocks
) {
960 /* Clean up the read-ahead blocks */
961 for (; ra_ptr
< ra_max
; ra_ptr
++)
962 brelse(bh_use
[ra_ptr
]);
966 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
967 struct ext4_dir_entry_2
**res_dir
, int *err
)
969 struct super_block
* sb
;
970 struct dx_hash_info hinfo
;
972 struct dx_frame frames
[2], *frame
;
973 struct ext4_dir_entry_2
*de
, *top
;
974 struct buffer_head
*bh
;
977 int namelen
= d_name
->len
;
978 const u8
*name
= d_name
->name
;
981 /* NFS may look up ".." - look at dx_root directory block */
982 if (namelen
> 2 || name
[0] != '.'||(name
[1] != '.' && name
[1] != '\0')){
983 if (!(frame
= dx_probe(d_name
, dir
, &hinfo
, frames
, err
)))
987 frame
->bh
= NULL
; /* for dx_release() */
988 frame
->at
= (struct dx_entry
*)frames
; /* hack for zero entry*/
989 dx_set_block(frame
->at
, 0); /* dx_root block is 0 */
993 block
= dx_get_block(frame
->at
);
994 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0, err
)))
996 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
997 top
= (struct ext4_dir_entry_2
*) ((char *) de
+ sb
->s_blocksize
-
998 EXT4_DIR_REC_LEN(0));
999 for (; de
< top
; de
= ext4_next_entry(de
)) {
1000 int off
= (block
<< EXT4_BLOCK_SIZE_BITS(sb
))
1001 + ((char *) de
- bh
->b_data
);
1003 if (!ext4_check_dir_entry(__func__
, dir
, de
, bh
, off
)) {
1005 *err
= ERR_BAD_DX_DIR
;
1009 if (ext4_match(namelen
, name
, de
)) {
1016 /* Check to see if we should continue to search */
1017 retval
= ext4_htree_next_block(dir
, hash
, frame
,
1020 ext4_warning(sb
, __func__
,
1021 "error reading index page in directory #%lu",
1026 } while (retval
== 1);
1030 dxtrace(printk(KERN_DEBUG
"%s not found\n", name
));
1031 dx_release (frames
);
1035 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1037 struct inode
*inode
;
1038 struct ext4_dir_entry_2
*de
;
1039 struct buffer_head
*bh
;
1041 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1042 return ERR_PTR(-ENAMETOOLONG
);
1044 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
1047 __u32 ino
= le32_to_cpu(de
->inode
);
1049 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1050 ext4_error(dir
->i_sb
, "ext4_lookup",
1051 "bad inode number: %u", ino
);
1052 return ERR_PTR(-EIO
);
1054 inode
= ext4_iget(dir
->i_sb
, ino
);
1055 if (unlikely(IS_ERR(inode
))) {
1056 if (PTR_ERR(inode
) == -ESTALE
) {
1057 ext4_error(dir
->i_sb
, __func__
,
1058 "deleted inode referenced: %u",
1060 return ERR_PTR(-EIO
);
1062 return ERR_CAST(inode
);
1066 return d_splice_alias(inode
, dentry
);
1070 struct dentry
*ext4_get_parent(struct dentry
*child
)
1073 struct inode
*inode
;
1074 static const struct qstr dotdot
= {
1078 struct ext4_dir_entry_2
* de
;
1079 struct buffer_head
*bh
;
1081 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
);
1084 return ERR_PTR(-ENOENT
);
1085 ino
= le32_to_cpu(de
->inode
);
1088 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1089 ext4_error(child
->d_inode
->i_sb
, "ext4_get_parent",
1090 "bad inode number: %u", ino
);
1091 return ERR_PTR(-EIO
);
1094 return d_obtain_alias(ext4_iget(child
->d_inode
->i_sb
, ino
));
1098 static unsigned char ext4_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1099 [S_IFREG
>> S_SHIFT
] = EXT4_FT_REG_FILE
,
1100 [S_IFDIR
>> S_SHIFT
] = EXT4_FT_DIR
,
1101 [S_IFCHR
>> S_SHIFT
] = EXT4_FT_CHRDEV
,
1102 [S_IFBLK
>> S_SHIFT
] = EXT4_FT_BLKDEV
,
1103 [S_IFIFO
>> S_SHIFT
] = EXT4_FT_FIFO
,
1104 [S_IFSOCK
>> S_SHIFT
] = EXT4_FT_SOCK
,
1105 [S_IFLNK
>> S_SHIFT
] = EXT4_FT_SYMLINK
,
1108 static inline void ext4_set_de_type(struct super_block
*sb
,
1109 struct ext4_dir_entry_2
*de
,
1111 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FILETYPE
))
1112 de
->file_type
= ext4_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1116 * Move count entries from end of map between two memory locations.
1117 * Returns pointer to last entry moved.
1119 static struct ext4_dir_entry_2
*
1120 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1122 unsigned rec_len
= 0;
1125 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*) (from
+ map
->offs
);
1126 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1127 memcpy (to
, de
, rec_len
);
1128 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1129 ext4_rec_len_to_disk(rec_len
);
1134 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1138 * Compact each dir entry in the range to the minimal rec_len.
1139 * Returns pointer to last entry in range.
1141 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1143 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1144 unsigned rec_len
= 0;
1147 while ((char*)de
< base
+ size
) {
1148 next
= ext4_next_entry(de
);
1149 if (de
->inode
&& de
->name_len
) {
1150 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1152 memmove(to
, de
, rec_len
);
1153 to
->rec_len
= ext4_rec_len_to_disk(rec_len
);
1155 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1163 * Split a full leaf block to make room for a new dir entry.
1164 * Allocate a new block, and move entries so that they are approx. equally full.
1165 * Returns pointer to de in block into which the new entry will be inserted.
1167 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1168 struct buffer_head
**bh
,struct dx_frame
*frame
,
1169 struct dx_hash_info
*hinfo
, int *error
)
1171 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1172 unsigned count
, continued
;
1173 struct buffer_head
*bh2
;
1174 ext4_lblk_t newblock
;
1176 struct dx_map_entry
*map
;
1177 char *data1
= (*bh
)->b_data
, *data2
;
1178 unsigned split
, move
, size
;
1179 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1182 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1189 BUFFER_TRACE(*bh
, "get_write_access");
1190 err
= ext4_journal_get_write_access(handle
, *bh
);
1194 BUFFER_TRACE(frame
->bh
, "get_write_access");
1195 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1199 data2
= bh2
->b_data
;
1201 /* create map in the end of data2 block */
1202 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1203 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1204 blocksize
, hinfo
, map
);
1206 dx_sort_map(map
, count
);
1207 /* Split the existing block in the middle, size-wise */
1210 for (i
= count
-1; i
>= 0; i
--) {
1211 /* is more than half of this entry in 2nd half of the block? */
1212 if (size
+ map
[i
].size
/2 > blocksize
/2)
1214 size
+= map
[i
].size
;
1217 /* map index at which we will split */
1218 split
= count
- move
;
1219 hash2
= map
[split
].hash
;
1220 continued
= hash2
== map
[split
- 1].hash
;
1221 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1222 (unsigned long)dx_get_block(frame
->at
),
1223 hash2
, split
, count
-split
));
1225 /* Fancy dance to stay within two buffers */
1226 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1227 de
= dx_pack_dirents(data1
, blocksize
);
1228 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1229 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1230 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1231 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1233 /* Which block gets the new entry? */
1234 if (hinfo
->hash
>= hash2
)
1239 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1240 err
= ext4_handle_dirty_metadata(handle
, dir
, bh2
);
1243 err
= ext4_handle_dirty_metadata(handle
, dir
, frame
->bh
);
1247 dxtrace(dx_show_index("frame", frame
->entries
));
1254 ext4_std_error(dir
->i_sb
, err
);
1261 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1262 * it points to a directory entry which is guaranteed to be large
1263 * enough for new directory entry. If de is NULL, then
1264 * add_dirent_to_buf will attempt search the directory block for
1265 * space. It will return -ENOSPC if no space is available, and -EIO
1266 * and -EEXIST if directory entry already exists.
1268 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1269 * all other cases bh is released.
1271 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1272 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1273 struct buffer_head
*bh
)
1275 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1276 const char *name
= dentry
->d_name
.name
;
1277 int namelen
= dentry
->d_name
.len
;
1278 unsigned int offset
= 0;
1279 unsigned short reclen
;
1280 int nlen
, rlen
, err
;
1283 reclen
= EXT4_DIR_REC_LEN(namelen
);
1285 de
= (struct ext4_dir_entry_2
*)bh
->b_data
;
1286 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1287 while ((char *) de
<= top
) {
1288 if (!ext4_check_dir_entry("ext4_add_entry", dir
, de
,
1293 if (ext4_match(namelen
, name
, de
)) {
1297 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1298 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1299 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1301 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1304 if ((char *) de
> top
)
1307 BUFFER_TRACE(bh
, "get_write_access");
1308 err
= ext4_journal_get_write_access(handle
, bh
);
1310 ext4_std_error(dir
->i_sb
, err
);
1315 /* By now the buffer is marked for journaling */
1316 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1317 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1319 struct ext4_dir_entry_2
*de1
= (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1320 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
);
1321 de
->rec_len
= ext4_rec_len_to_disk(nlen
);
1324 de
->file_type
= EXT4_FT_UNKNOWN
;
1326 de
->inode
= cpu_to_le32(inode
->i_ino
);
1327 ext4_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1330 de
->name_len
= namelen
;
1331 memcpy(de
->name
, name
, namelen
);
1333 * XXX shouldn't update any times until successful
1334 * completion of syscall, but too many callers depend
1337 * XXX similarly, too many callers depend on
1338 * ext4_new_inode() setting the times, but error
1339 * recovery deletes the inode, so the worst that can
1340 * happen is that the times are slightly out of date
1341 * and/or different from the directory change time.
1343 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1344 ext4_update_dx_flag(dir
);
1346 ext4_mark_inode_dirty(handle
, dir
);
1347 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1348 err
= ext4_handle_dirty_metadata(handle
, dir
, bh
);
1350 ext4_std_error(dir
->i_sb
, err
);
1356 * This converts a one block unindexed directory to a 3 block indexed
1357 * directory, and adds the dentry to the indexed directory.
1359 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1360 struct inode
*inode
, struct buffer_head
*bh
)
1362 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1363 const char *name
= dentry
->d_name
.name
;
1364 int namelen
= dentry
->d_name
.len
;
1365 struct buffer_head
*bh2
;
1366 struct dx_root
*root
;
1367 struct dx_frame frames
[2], *frame
;
1368 struct dx_entry
*entries
;
1369 struct ext4_dir_entry_2
*de
, *de2
;
1374 struct dx_hash_info hinfo
;
1376 struct fake_dirent
*fde
;
1378 blocksize
= dir
->i_sb
->s_blocksize
;
1379 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1380 retval
= ext4_journal_get_write_access(handle
, bh
);
1382 ext4_std_error(dir
->i_sb
, retval
);
1386 root
= (struct dx_root
*) bh
->b_data
;
1388 /* The 0th block becomes the root, move the dirents out */
1389 fde
= &root
->dotdot
;
1390 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1391 ext4_rec_len_from_disk(fde
->rec_len
));
1392 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1393 ext4_error(dir
->i_sb
, __func__
,
1394 "invalid rec_len for '..' in inode %lu",
1399 len
= ((char *) root
) + blocksize
- (char *) de
;
1401 /* Allocate new block for the 0th block's dirents */
1402 bh2
= ext4_append(handle
, dir
, &block
, &retval
);
1407 EXT4_I(dir
)->i_flags
|= EXT4_INDEX_FL
;
1408 data1
= bh2
->b_data
;
1410 memcpy (data1
, de
, len
);
1411 de
= (struct ext4_dir_entry_2
*) data1
;
1413 while ((char *)(de2
= ext4_next_entry(de
)) < top
)
1415 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1416 /* Initialize the root; the dot dirents already exist */
1417 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1418 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2));
1419 memset (&root
->info
, 0, sizeof(root
->info
));
1420 root
->info
.info_length
= sizeof(root
->info
);
1421 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1422 entries
= root
->entries
;
1423 dx_set_block(entries
, 1);
1424 dx_set_count(entries
, 1);
1425 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1427 /* Initialize as for dx_probe */
1428 hinfo
.hash_version
= root
->info
.hash_version
;
1429 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1430 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1431 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1432 ext4fs_dirhash(name
, namelen
, &hinfo
);
1434 frame
->entries
= entries
;
1435 frame
->at
= entries
;
1438 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1439 dx_release (frames
);
1443 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1449 * adds a file entry to the specified directory, using the same
1450 * semantics as ext4_find_entry(). It returns NULL if it failed.
1452 * NOTE!! The inode part of 'de' is left at 0 - which means you
1453 * may not sleep between calling this and putting something into
1454 * the entry, as someone else might have used it while you slept.
1456 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1457 struct inode
*inode
)
1459 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1460 struct buffer_head
*bh
;
1461 struct ext4_dir_entry_2
*de
;
1462 struct super_block
*sb
;
1466 ext4_lblk_t block
, blocks
;
1469 blocksize
= sb
->s_blocksize
;
1470 if (!dentry
->d_name
.len
)
1473 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1474 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1476 EXT4_I(dir
)->i_flags
&= ~EXT4_INDEX_FL
;
1478 ext4_mark_inode_dirty(handle
, dir
);
1480 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1481 for (block
= 0; block
< blocks
; block
++) {
1482 bh
= ext4_bread(handle
, dir
, block
, 0, &retval
);
1485 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1486 if (retval
!= -ENOSPC
)
1489 if (blocks
== 1 && !dx_fallback
&&
1490 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1491 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1494 bh
= ext4_append(handle
, dir
, &block
, &retval
);
1497 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1499 de
->rec_len
= ext4_rec_len_to_disk(blocksize
);
1500 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1504 * Returns 0 for success, or a negative error value
1506 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1507 struct inode
*inode
)
1509 struct dx_frame frames
[2], *frame
;
1510 struct dx_entry
*entries
, *at
;
1511 struct dx_hash_info hinfo
;
1512 struct buffer_head
*bh
;
1513 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1514 struct super_block
*sb
= dir
->i_sb
;
1515 struct ext4_dir_entry_2
*de
;
1518 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1521 entries
= frame
->entries
;
1524 if (!(bh
= ext4_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1527 BUFFER_TRACE(bh
, "get_write_access");
1528 err
= ext4_journal_get_write_access(handle
, bh
);
1532 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1533 if (err
!= -ENOSPC
) {
1538 /* Block full, should compress but for now just split */
1539 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
1540 dx_get_count(entries
), dx_get_limit(entries
)));
1541 /* Need to split index? */
1542 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1543 ext4_lblk_t newblock
;
1544 unsigned icount
= dx_get_count(entries
);
1545 int levels
= frame
- frames
;
1546 struct dx_entry
*entries2
;
1547 struct dx_node
*node2
;
1548 struct buffer_head
*bh2
;
1550 if (levels
&& (dx_get_count(frames
->entries
) ==
1551 dx_get_limit(frames
->entries
))) {
1552 ext4_warning(sb
, __func__
,
1553 "Directory index full!");
1557 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1560 node2
= (struct dx_node
*)(bh2
->b_data
);
1561 entries2
= node2
->entries
;
1562 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
);
1563 node2
->fake
.inode
= 0;
1564 BUFFER_TRACE(frame
->bh
, "get_write_access");
1565 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1569 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1570 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1571 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
1574 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1575 err
= ext4_journal_get_write_access(handle
,
1580 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
1581 icount2
* sizeof(struct dx_entry
));
1582 dx_set_count(entries
, icount1
);
1583 dx_set_count(entries2
, icount2
);
1584 dx_set_limit(entries2
, dx_node_limit(dir
));
1586 /* Which index block gets the new entry? */
1587 if (at
- entries
>= icount1
) {
1588 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1589 frame
->entries
= entries
= entries2
;
1590 swap(frame
->bh
, bh2
);
1592 dx_insert_block(frames
+ 0, hash2
, newblock
);
1593 dxtrace(dx_show_index("node", frames
[1].entries
));
1594 dxtrace(dx_show_index("node",
1595 ((struct dx_node
*) bh2
->b_data
)->entries
));
1596 err
= ext4_handle_dirty_metadata(handle
, inode
, bh2
);
1601 dxtrace(printk(KERN_DEBUG
1602 "Creating second level index...\n"));
1603 memcpy((char *) entries2
, (char *) entries
,
1604 icount
* sizeof(struct dx_entry
));
1605 dx_set_limit(entries2
, dx_node_limit(dir
));
1608 dx_set_count(entries
, 1);
1609 dx_set_block(entries
+ 0, newblock
);
1610 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1612 /* Add new access path frame */
1614 frame
->at
= at
= at
- entries
+ entries2
;
1615 frame
->entries
= entries
= entries2
;
1617 err
= ext4_journal_get_write_access(handle
,
1622 ext4_handle_dirty_metadata(handle
, inode
, frames
[0].bh
);
1624 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1627 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1632 ext4_std_error(dir
->i_sb
, err
);
1641 * ext4_delete_entry deletes a directory entry by merging it with the
1644 static int ext4_delete_entry(handle_t
*handle
,
1646 struct ext4_dir_entry_2
*de_del
,
1647 struct buffer_head
*bh
)
1649 struct ext4_dir_entry_2
*de
, *pde
;
1654 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1655 while (i
< bh
->b_size
) {
1656 if (!ext4_check_dir_entry("ext4_delete_entry", dir
, de
, bh
, i
))
1659 BUFFER_TRACE(bh
, "get_write_access");
1660 ext4_journal_get_write_access(handle
, bh
);
1662 pde
->rec_len
= ext4_rec_len_to_disk(
1663 ext4_rec_len_from_disk(pde
->rec_len
) +
1664 ext4_rec_len_from_disk(de
->rec_len
));
1668 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1669 ext4_handle_dirty_metadata(handle
, dir
, bh
);
1672 i
+= ext4_rec_len_from_disk(de
->rec_len
);
1674 de
= ext4_next_entry(de
);
1680 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1681 * since this indicates that nlinks count was previously 1.
1683 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
1686 if (is_dx(inode
) && inode
->i_nlink
> 1) {
1687 /* limit is 16-bit i_links_count */
1688 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
1690 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
1691 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
1697 * If a directory had nlink == 1, then we should let it be 1. This indicates
1698 * directory has >EXT4_LINK_MAX subdirs.
1700 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
1703 if (S_ISDIR(inode
->i_mode
) && inode
->i_nlink
== 0)
1708 static int ext4_add_nondir(handle_t
*handle
,
1709 struct dentry
*dentry
, struct inode
*inode
)
1711 int err
= ext4_add_entry(handle
, dentry
, inode
);
1713 ext4_mark_inode_dirty(handle
, inode
);
1714 d_instantiate(dentry
, inode
);
1715 unlock_new_inode(inode
);
1719 unlock_new_inode(inode
);
1725 * By the time this is called, we already have created
1726 * the directory cache entry for the new file, but it
1727 * is so far negative - it has no inode.
1729 * If the create succeeds, we fill in the inode information
1730 * with d_instantiate().
1732 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1733 struct nameidata
*nd
)
1736 struct inode
*inode
;
1737 int err
, retries
= 0;
1740 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1741 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1742 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1744 return PTR_ERR(handle
);
1746 if (IS_DIRSYNC(dir
))
1747 ext4_handle_sync(handle
);
1749 inode
= ext4_new_inode (handle
, dir
, mode
);
1750 err
= PTR_ERR(inode
);
1751 if (!IS_ERR(inode
)) {
1752 inode
->i_op
= &ext4_file_inode_operations
;
1753 inode
->i_fop
= &ext4_file_operations
;
1754 ext4_set_aops(inode
);
1755 err
= ext4_add_nondir(handle
, dentry
, inode
);
1757 ext4_journal_stop(handle
);
1758 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1763 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
1764 int mode
, dev_t rdev
)
1767 struct inode
*inode
;
1768 int err
, retries
= 0;
1770 if (!new_valid_dev(rdev
))
1774 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1775 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1776 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1778 return PTR_ERR(handle
);
1780 if (IS_DIRSYNC(dir
))
1781 ext4_handle_sync(handle
);
1783 inode
= ext4_new_inode(handle
, dir
, mode
);
1784 err
= PTR_ERR(inode
);
1785 if (!IS_ERR(inode
)) {
1786 init_special_inode(inode
, inode
->i_mode
, rdev
);
1787 #ifdef CONFIG_EXT4_FS_XATTR
1788 inode
->i_op
= &ext4_special_inode_operations
;
1790 err
= ext4_add_nondir(handle
, dentry
, inode
);
1792 ext4_journal_stop(handle
);
1793 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1798 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1801 struct inode
*inode
;
1802 struct buffer_head
*dir_block
;
1803 struct ext4_dir_entry_2
*de
;
1804 int err
, retries
= 0;
1806 if (EXT4_DIR_LINK_MAX(dir
))
1810 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1811 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1812 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1814 return PTR_ERR(handle
);
1816 if (IS_DIRSYNC(dir
))
1817 ext4_handle_sync(handle
);
1819 inode
= ext4_new_inode(handle
, dir
, S_IFDIR
| mode
);
1820 err
= PTR_ERR(inode
);
1824 inode
->i_op
= &ext4_dir_inode_operations
;
1825 inode
->i_fop
= &ext4_dir_operations
;
1826 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1827 dir_block
= ext4_bread(handle
, inode
, 0, 1, &err
);
1829 goto out_clear_inode
;
1830 BUFFER_TRACE(dir_block
, "get_write_access");
1831 ext4_journal_get_write_access(handle
, dir_block
);
1832 de
= (struct ext4_dir_entry_2
*) dir_block
->b_data
;
1833 de
->inode
= cpu_to_le32(inode
->i_ino
);
1835 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
));
1836 strcpy(de
->name
, ".");
1837 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1838 de
= ext4_next_entry(de
);
1839 de
->inode
= cpu_to_le32(dir
->i_ino
);
1840 de
->rec_len
= ext4_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1841 EXT4_DIR_REC_LEN(1));
1843 strcpy(de
->name
, "..");
1844 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1846 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
1847 ext4_handle_dirty_metadata(handle
, dir
, dir_block
);
1849 ext4_mark_inode_dirty(handle
, inode
);
1850 err
= ext4_add_entry(handle
, dentry
, inode
);
1854 unlock_new_inode(inode
);
1855 ext4_mark_inode_dirty(handle
, inode
);
1859 ext4_inc_count(handle
, dir
);
1860 ext4_update_dx_flag(dir
);
1861 ext4_mark_inode_dirty(handle
, dir
);
1862 d_instantiate(dentry
, inode
);
1863 unlock_new_inode(inode
);
1865 ext4_journal_stop(handle
);
1866 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1872 * routine to check that the specified directory is empty (for rmdir)
1874 static int empty_dir(struct inode
*inode
)
1876 unsigned int offset
;
1877 struct buffer_head
*bh
;
1878 struct ext4_dir_entry_2
*de
, *de1
;
1879 struct super_block
*sb
;
1883 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1884 !(bh
= ext4_bread(NULL
, inode
, 0, 0, &err
))) {
1886 ext4_error(inode
->i_sb
, __func__
,
1887 "error %d reading directory #%lu offset 0",
1890 ext4_warning(inode
->i_sb
, __func__
,
1891 "bad directory (dir #%lu) - no data block",
1895 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1896 de1
= ext4_next_entry(de
);
1897 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1898 !le32_to_cpu(de1
->inode
) ||
1899 strcmp(".", de
->name
) ||
1900 strcmp("..", de1
->name
)) {
1901 ext4_warning(inode
->i_sb
, "empty_dir",
1902 "bad directory (dir #%lu) - no `.' or `..'",
1907 offset
= ext4_rec_len_from_disk(de
->rec_len
) +
1908 ext4_rec_len_from_disk(de1
->rec_len
);
1909 de
= ext4_next_entry(de1
);
1910 while (offset
< inode
->i_size
) {
1912 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1915 bh
= ext4_bread(NULL
, inode
,
1916 offset
>> EXT4_BLOCK_SIZE_BITS(sb
), 0, &err
);
1919 ext4_error(sb
, __func__
,
1920 "error %d reading directory"
1922 err
, inode
->i_ino
, offset
);
1923 offset
+= sb
->s_blocksize
;
1926 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1928 if (!ext4_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1929 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
1931 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1934 if (le32_to_cpu(de
->inode
)) {
1938 offset
+= ext4_rec_len_from_disk(de
->rec_len
);
1939 de
= ext4_next_entry(de
);
1945 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1946 * such inodes, starting at the superblock, in case we crash before the
1947 * file is closed/deleted, or in case the inode truncate spans multiple
1948 * transactions and the last transaction is not recovered after a crash.
1950 * At filesystem recovery time, we walk this list deleting unlinked
1951 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1953 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
1955 struct super_block
*sb
= inode
->i_sb
;
1956 struct ext4_iloc iloc
;
1959 if (!ext4_handle_valid(handle
))
1963 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
1966 /* Orphan handling is only valid for files with data blocks
1967 * being truncated, or files being unlinked. */
1969 /* @@@ FIXME: Observation from aviro:
1970 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1971 * here (on lock_super()), so race with ext4_link() which might bump
1972 * ->i_nlink. For, say it, character device. Not a regular file,
1973 * not a directory, not a symlink and ->i_nlink > 0.
1975 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1976 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1978 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1979 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1983 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
1987 /* Insert this inode at the head of the on-disk orphan list... */
1988 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT4_SB(sb
)->s_es
->s_last_orphan
);
1989 EXT4_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1990 err
= ext4_handle_dirty_metadata(handle
, inode
, EXT4_SB(sb
)->s_sbh
);
1991 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1995 /* Only add to the head of the in-memory list if all the
1996 * previous operations succeeded. If the orphan_add is going to
1997 * fail (possibly taking the journal offline), we can't risk
1998 * leaving the inode on the orphan list: stray orphan-list
1999 * entries can cause panics at unmount time.
2001 * This is safe: on error we're going to ignore the orphan list
2002 * anyway on the next recovery. */
2004 list_add(&EXT4_I(inode
)->i_orphan
, &EXT4_SB(sb
)->s_orphan
);
2006 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2007 jbd_debug(4, "orphan inode %lu will point to %d\n",
2008 inode
->i_ino
, NEXT_ORPHAN(inode
));
2011 ext4_std_error(inode
->i_sb
, err
);
2016 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2017 * of such inodes stored on disk, because it is finally being cleaned up.
2019 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2021 struct list_head
*prev
;
2022 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2023 struct ext4_sb_info
*sbi
;
2025 struct ext4_iloc iloc
;
2028 if (!ext4_handle_valid(handle
))
2031 lock_super(inode
->i_sb
);
2032 if (list_empty(&ei
->i_orphan
)) {
2033 unlock_super(inode
->i_sb
);
2037 ino_next
= NEXT_ORPHAN(inode
);
2038 prev
= ei
->i_orphan
.prev
;
2039 sbi
= EXT4_SB(inode
->i_sb
);
2041 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2043 list_del_init(&ei
->i_orphan
);
2045 /* If we're on an error path, we may not have a valid
2046 * transaction handle with which to update the orphan list on
2047 * disk, but we still need to remove the inode from the linked
2048 * list in memory. */
2049 if (sbi
->s_journal
&& !handle
)
2052 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2056 if (prev
== &sbi
->s_orphan
) {
2057 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2058 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2059 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2062 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2063 err
= ext4_handle_dirty_metadata(handle
, inode
, sbi
->s_sbh
);
2065 struct ext4_iloc iloc2
;
2066 struct inode
*i_prev
=
2067 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2069 jbd_debug(4, "orphan inode %lu will point to %u\n",
2070 i_prev
->i_ino
, ino_next
);
2071 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2074 NEXT_ORPHAN(i_prev
) = ino_next
;
2075 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2079 NEXT_ORPHAN(inode
) = 0;
2080 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2083 ext4_std_error(inode
->i_sb
, err
);
2085 unlock_super(inode
->i_sb
);
2093 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2096 struct inode
*inode
;
2097 struct buffer_head
*bh
;
2098 struct ext4_dir_entry_2
*de
;
2101 /* Initialize quotas before so that eventual writes go in
2102 * separate transaction */
2103 DQUOT_INIT(dentry
->d_inode
);
2104 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2106 return PTR_ERR(handle
);
2109 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2113 if (IS_DIRSYNC(dir
))
2114 ext4_handle_sync(handle
);
2116 inode
= dentry
->d_inode
;
2119 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2122 retval
= -ENOTEMPTY
;
2123 if (!empty_dir(inode
))
2126 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2129 if (!EXT4_DIR_LINK_EMPTY(inode
))
2130 ext4_warning(inode
->i_sb
, "ext4_rmdir",
2131 "empty directory has too many links (%d)",
2135 /* There's no need to set i_disksize: the fact that i_nlink is
2136 * zero will ensure that the right thing happens during any
2139 ext4_orphan_add(handle
, inode
);
2140 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2141 ext4_mark_inode_dirty(handle
, inode
);
2142 ext4_dec_count(handle
, dir
);
2143 ext4_update_dx_flag(dir
);
2144 ext4_mark_inode_dirty(handle
, dir
);
2147 ext4_journal_stop(handle
);
2152 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2155 struct inode
*inode
;
2156 struct buffer_head
*bh
;
2157 struct ext4_dir_entry_2
*de
;
2160 /* Initialize quotas before so that eventual writes go
2161 * in separate transaction */
2162 DQUOT_INIT(dentry
->d_inode
);
2163 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2165 return PTR_ERR(handle
);
2167 if (IS_DIRSYNC(dir
))
2168 ext4_handle_sync(handle
);
2171 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2175 inode
= dentry
->d_inode
;
2178 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2181 if (!inode
->i_nlink
) {
2182 ext4_warning(inode
->i_sb
, "ext4_unlink",
2183 "Deleting nonexistent file (%lu), %d",
2184 inode
->i_ino
, inode
->i_nlink
);
2187 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2190 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2191 ext4_update_dx_flag(dir
);
2192 ext4_mark_inode_dirty(handle
, dir
);
2194 if (!inode
->i_nlink
)
2195 ext4_orphan_add(handle
, inode
);
2196 inode
->i_ctime
= ext4_current_time(inode
);
2197 ext4_mark_inode_dirty(handle
, inode
);
2201 ext4_journal_stop(handle
);
2206 static int ext4_symlink(struct inode
*dir
,
2207 struct dentry
*dentry
, const char *symname
)
2210 struct inode
*inode
;
2211 int l
, err
, retries
= 0;
2213 l
= strlen(symname
)+1;
2214 if (l
> dir
->i_sb
->s_blocksize
)
2215 return -ENAMETOOLONG
;
2218 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2219 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2220 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2222 return PTR_ERR(handle
);
2224 if (IS_DIRSYNC(dir
))
2225 ext4_handle_sync(handle
);
2227 inode
= ext4_new_inode(handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2228 err
= PTR_ERR(inode
);
2232 if (l
> sizeof(EXT4_I(inode
)->i_data
)) {
2233 inode
->i_op
= &ext4_symlink_inode_operations
;
2234 ext4_set_aops(inode
);
2236 * page_symlink() calls into ext4_prepare/commit_write.
2237 * We have a transaction open. All is sweetness. It also sets
2238 * i_size in generic_commit_write().
2240 err
= __page_symlink(inode
, symname
, l
, 1);
2243 unlock_new_inode(inode
);
2244 ext4_mark_inode_dirty(handle
, inode
);
2249 /* clear the extent format for fast symlink */
2250 EXT4_I(inode
)->i_flags
&= ~EXT4_EXTENTS_FL
;
2251 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2252 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2253 inode
->i_size
= l
-1;
2255 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2256 err
= ext4_add_nondir(handle
, dentry
, inode
);
2258 ext4_journal_stop(handle
);
2259 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2264 static int ext4_link(struct dentry
*old_dentry
,
2265 struct inode
*dir
, struct dentry
*dentry
)
2268 struct inode
*inode
= old_dentry
->d_inode
;
2269 int err
, retries
= 0;
2271 if (EXT4_DIR_LINK_MAX(inode
))
2275 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2276 * otherwise has the potential to corrupt the orphan inode list.
2278 if (inode
->i_nlink
== 0)
2282 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2283 EXT4_INDEX_EXTRA_TRANS_BLOCKS
);
2285 return PTR_ERR(handle
);
2287 if (IS_DIRSYNC(dir
))
2288 ext4_handle_sync(handle
);
2290 inode
->i_ctime
= ext4_current_time(inode
);
2291 ext4_inc_count(handle
, inode
);
2292 atomic_inc(&inode
->i_count
);
2294 err
= ext4_add_entry(handle
, dentry
, inode
);
2296 ext4_mark_inode_dirty(handle
, inode
);
2297 d_instantiate(dentry
, inode
);
2302 ext4_journal_stop(handle
);
2303 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2308 #define PARENT_INO(buffer) \
2309 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2312 * Anybody can rename anything with this: the permission checks are left to the
2313 * higher-level routines.
2315 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2316 struct inode
*new_dir
, struct dentry
*new_dentry
)
2319 struct inode
*old_inode
, *new_inode
;
2320 struct buffer_head
*old_bh
, *new_bh
, *dir_bh
;
2321 struct ext4_dir_entry_2
*old_de
, *new_de
;
2322 int retval
, force_da_alloc
= 0;
2324 old_bh
= new_bh
= dir_bh
= NULL
;
2326 /* Initialize quotas before so that eventual writes go
2327 * in separate transaction */
2328 if (new_dentry
->d_inode
)
2329 DQUOT_INIT(new_dentry
->d_inode
);
2330 handle
= ext4_journal_start(old_dir
, 2 *
2331 EXT4_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2332 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2334 return PTR_ERR(handle
);
2336 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2337 ext4_handle_sync(handle
);
2339 old_bh
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2341 * Check for inode number is _not_ due to possible IO errors.
2342 * We might rmdir the source, keep it as pwd of some process
2343 * and merrily kill the link to whatever was created under the
2344 * same name. Goodbye sticky bit ;-<
2346 old_inode
= old_dentry
->d_inode
;
2348 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2351 new_inode
= new_dentry
->d_inode
;
2352 new_bh
= ext4_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
2359 if (S_ISDIR(old_inode
->i_mode
)) {
2361 retval
= -ENOTEMPTY
;
2362 if (!empty_dir(new_inode
))
2366 dir_bh
= ext4_bread(handle
, old_inode
, 0, 0, &retval
);
2369 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2372 if (!new_inode
&& new_dir
!= old_dir
&&
2373 new_dir
->i_nlink
>= EXT4_LINK_MAX
)
2377 retval
= ext4_add_entry(handle
, new_dentry
, old_inode
);
2381 BUFFER_TRACE(new_bh
, "get write access");
2382 ext4_journal_get_write_access(handle
, new_bh
);
2383 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2384 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2385 EXT4_FEATURE_INCOMPAT_FILETYPE
))
2386 new_de
->file_type
= old_de
->file_type
;
2387 new_dir
->i_version
++;
2388 new_dir
->i_ctime
= new_dir
->i_mtime
=
2389 ext4_current_time(new_dir
);
2390 ext4_mark_inode_dirty(handle
, new_dir
);
2391 BUFFER_TRACE(new_bh
, "call ext4_handle_dirty_metadata");
2392 ext4_handle_dirty_metadata(handle
, new_dir
, new_bh
);
2398 * Like most other Unix systems, set the ctime for inodes on a
2401 old_inode
->i_ctime
= ext4_current_time(old_inode
);
2402 ext4_mark_inode_dirty(handle
, old_inode
);
2407 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2408 old_de
->name_len
!= old_dentry
->d_name
.len
||
2409 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2410 (retval
= ext4_delete_entry(handle
, old_dir
,
2411 old_de
, old_bh
)) == -ENOENT
) {
2412 /* old_de could have moved from under us during htree split, so
2413 * make sure that we are deleting the right entry. We might
2414 * also be pointing to a stale entry in the unused part of
2415 * old_bh so just checking inum and the name isn't enough. */
2416 struct buffer_head
*old_bh2
;
2417 struct ext4_dir_entry_2
*old_de2
;
2419 old_bh2
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de2
);
2421 retval
= ext4_delete_entry(handle
, old_dir
,
2427 ext4_warning(old_dir
->i_sb
, "ext4_rename",
2428 "Deleting old file (%lu), %d, error=%d",
2429 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2433 ext4_dec_count(handle
, new_inode
);
2434 new_inode
->i_ctime
= ext4_current_time(new_inode
);
2436 old_dir
->i_ctime
= old_dir
->i_mtime
= ext4_current_time(old_dir
);
2437 ext4_update_dx_flag(old_dir
);
2439 BUFFER_TRACE(dir_bh
, "get_write_access");
2440 ext4_journal_get_write_access(handle
, dir_bh
);
2441 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2442 BUFFER_TRACE(dir_bh
, "call ext4_handle_dirty_metadata");
2443 ext4_handle_dirty_metadata(handle
, old_dir
, dir_bh
);
2444 ext4_dec_count(handle
, old_dir
);
2446 /* checked empty_dir above, can't have another parent,
2447 * ext4_dec_count() won't work for many-linked dirs */
2448 new_inode
->i_nlink
= 0;
2450 ext4_inc_count(handle
, new_dir
);
2451 ext4_update_dx_flag(new_dir
);
2452 ext4_mark_inode_dirty(handle
, new_dir
);
2455 ext4_mark_inode_dirty(handle
, old_dir
);
2457 ext4_mark_inode_dirty(handle
, new_inode
);
2458 if (!new_inode
->i_nlink
)
2459 ext4_orphan_add(handle
, new_inode
);
2468 ext4_journal_stop(handle
);
2469 if (retval
== 0 && force_da_alloc
)
2470 ext4_alloc_da_blocks(old_inode
);
2475 * directories can handle most operations...
2477 const struct inode_operations ext4_dir_inode_operations
= {
2478 .create
= ext4_create
,
2479 .lookup
= ext4_lookup
,
2481 .unlink
= ext4_unlink
,
2482 .symlink
= ext4_symlink
,
2483 .mkdir
= ext4_mkdir
,
2484 .rmdir
= ext4_rmdir
,
2485 .mknod
= ext4_mknod
,
2486 .rename
= ext4_rename
,
2487 .setattr
= ext4_setattr
,
2488 #ifdef CONFIG_EXT4_FS_XATTR
2489 .setxattr
= generic_setxattr
,
2490 .getxattr
= generic_getxattr
,
2491 .listxattr
= ext4_listxattr
,
2492 .removexattr
= generic_removexattr
,
2494 .permission
= ext4_permission
,
2497 const struct inode_operations ext4_special_inode_operations
= {
2498 .setattr
= ext4_setattr
,
2499 #ifdef CONFIG_EXT4_FS_XATTR
2500 .setxattr
= generic_setxattr
,
2501 .getxattr
= generic_getxattr
,
2502 .listxattr
= ext4_listxattr
,
2503 .removexattr
= generic_removexattr
,
2505 .permission
= ext4_permission
,