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(struct dentry
*dentry
,
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 dentry
*dentry
,
181 struct ext4_dir_entry_2
**res_dir
, int *err
);
182 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
183 struct inode
*inode
);
186 * p is at least 6 bytes before the end of page
188 static inline struct ext4_dir_entry_2
*
189 ext4_next_entry(struct ext4_dir_entry_2
*p
)
191 return (struct ext4_dir_entry_2
*)((char *)p
+
192 ext4_rec_len_from_disk(p
->rec_len
));
196 * Future: use high four bits of block for coalesce-on-delete flags
197 * Mask them off for now.
200 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
202 return le32_to_cpu(entry
->block
) & 0x00ffffff;
205 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
207 entry
->block
= cpu_to_le32(value
);
210 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
212 return le32_to_cpu(entry
->hash
);
215 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
217 entry
->hash
= cpu_to_le32(value
);
220 static inline unsigned dx_get_count(struct dx_entry
*entries
)
222 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
225 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
227 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
230 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
232 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
235 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
237 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
240 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
242 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
243 EXT4_DIR_REC_LEN(2) - infosize
;
244 return entry_space
/ sizeof(struct dx_entry
);
247 static inline unsigned dx_node_limit(struct inode
*dir
)
249 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
250 return entry_space
/ sizeof(struct dx_entry
);
257 static void dx_show_index(char * label
, struct dx_entry
*entries
)
259 int i
, n
= dx_get_count (entries
);
260 printk(KERN_DEBUG
"%s index ", label
);
261 for (i
= 0; i
< n
; i
++) {
262 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
263 0, (unsigned long)dx_get_block(entries
+ i
));
275 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
276 int size
, int show_names
)
278 unsigned names
= 0, space
= 0;
279 char *base
= (char *) de
;
280 struct dx_hash_info h
= *hinfo
;
283 while ((char *) de
< base
+ size
)
289 int len
= de
->name_len
;
290 char *name
= de
->name
;
291 while (len
--) printk("%c", *name
++);
292 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
293 printk(":%x.%u ", h
.hash
,
294 ((char *) de
- base
));
296 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
299 de
= ext4_next_entry(de
);
301 printk("(%i)\n", names
);
302 return (struct stats
) { names
, space
, 1 };
305 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
306 struct dx_entry
*entries
, int levels
)
308 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
309 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
311 struct buffer_head
*bh
;
313 printk("%i indexed blocks...\n", count
);
314 for (i
= 0; i
< count
; i
++, entries
++)
316 ext4_lblk_t block
= dx_get_block(entries
);
317 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
318 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
320 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
321 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0,&err
))) continue;
323 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
324 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
325 names
+= stats
.names
;
326 space
+= stats
.space
;
327 bcount
+= stats
.bcount
;
331 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
332 levels
? "" : " ", names
, space
/bcount
,
333 (space
/bcount
)*100/blocksize
);
334 return (struct stats
) { names
, space
, bcount
};
336 #endif /* DX_DEBUG */
339 * Probe for a directory leaf block to search.
341 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
342 * error in the directory index, and the caller should fall back to
343 * searching the directory normally. The callers of dx_probe **MUST**
344 * check for this error code, and make sure it never gets reflected
347 static struct dx_frame
*
348 dx_probe(struct dentry
*dentry
, struct inode
*dir
,
349 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
351 unsigned count
, indirect
;
352 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
353 struct dx_root
*root
;
354 struct buffer_head
*bh
;
355 struct dx_frame
*frame
= frame_in
;
360 dir
= dentry
->d_parent
->d_inode
;
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 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
377 ext4fs_dirhash(dentry
->d_name
.name
, dentry
->d_name
.len
, hinfo
);
380 if (root
->info
.unused_flags
& 1) {
381 ext4_warning(dir
->i_sb
, __func__
,
382 "Unimplemented inode hash flags: %#06x",
383 root
->info
.unused_flags
);
385 *err
= ERR_BAD_DX_DIR
;
389 if ((indirect
= root
->info
.indirect_levels
) > 1) {
390 ext4_warning(dir
->i_sb
, __func__
,
391 "Unimplemented inode hash depth: %#06x",
392 root
->info
.indirect_levels
);
394 *err
= ERR_BAD_DX_DIR
;
398 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
399 root
->info
.info_length
);
401 if (dx_get_limit(entries
) != dx_root_limit(dir
,
402 root
->info
.info_length
)) {
403 ext4_warning(dir
->i_sb
, __func__
,
404 "dx entry: limit != root limit");
406 *err
= ERR_BAD_DX_DIR
;
410 dxtrace(printk("Look up %x", hash
));
413 count
= dx_get_count(entries
);
414 if (!count
|| count
> dx_get_limit(entries
)) {
415 ext4_warning(dir
->i_sb
, __func__
,
416 "dx entry: no count or count > limit");
418 *err
= ERR_BAD_DX_DIR
;
423 q
= entries
+ count
- 1;
427 dxtrace(printk("."));
428 if (dx_get_hash(m
) > hash
)
434 if (0) // linear search cross check
436 unsigned n
= count
- 1;
440 dxtrace(printk(","));
441 if (dx_get_hash(++at
) > hash
)
447 assert (at
== p
- 1);
451 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
453 frame
->entries
= entries
;
455 if (!indirect
--) return frame
;
456 if (!(bh
= ext4_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
458 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
459 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
460 ext4_warning(dir
->i_sb
, __func__
,
461 "dx entry: limit != node limit");
463 *err
= ERR_BAD_DX_DIR
;
470 while (frame
>= frame_in
) {
475 if (*err
== ERR_BAD_DX_DIR
)
476 ext4_warning(dir
->i_sb
, __func__
,
477 "Corrupt dir inode %ld, running e2fsck is "
478 "recommended.", dir
->i_ino
);
482 static void dx_release (struct dx_frame
*frames
)
484 if (frames
[0].bh
== NULL
)
487 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
488 brelse(frames
[1].bh
);
489 brelse(frames
[0].bh
);
493 * This function increments the frame pointer to search the next leaf
494 * block, and reads in the necessary intervening nodes if the search
495 * should be necessary. Whether or not the search is necessary is
496 * controlled by the hash parameter. If the hash value is even, then
497 * the search is only continued if the next block starts with that
498 * hash value. This is used if we are searching for a specific file.
500 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
502 * This function returns 1 if the caller should continue to search,
503 * or 0 if it should not. If there is an error reading one of the
504 * index blocks, it will a negative error code.
506 * If start_hash is non-null, it will be filled in with the starting
507 * hash of the next page.
509 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
510 struct dx_frame
*frame
,
511 struct dx_frame
*frames
,
515 struct buffer_head
*bh
;
516 int err
, num_frames
= 0;
521 * Find the next leaf page by incrementing the frame pointer.
522 * If we run out of entries in the interior node, loop around and
523 * increment pointer in the parent node. When we break out of
524 * this loop, num_frames indicates the number of interior
525 * nodes need to be read.
528 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
537 * If the hash is 1, then continue only if the next page has a
538 * continuation hash of any value. This is used for readdir
539 * handling. Otherwise, check to see if the hash matches the
540 * desired contiuation hash. If it doesn't, return since
541 * there's no point to read in the successive index pages.
543 bhash
= dx_get_hash(p
->at
);
546 if ((hash
& 1) == 0) {
547 if ((bhash
& ~1) != hash
)
551 * If the hash is HASH_NB_ALWAYS, we always go to the next
552 * block so no check is necessary
554 while (num_frames
--) {
555 if (!(bh
= ext4_bread(NULL
, dir
, dx_get_block(p
->at
),
557 return err
; /* Failure */
561 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
568 * This function fills a red-black tree with information from a
569 * directory block. It returns the number directory entries loaded
570 * into the tree. If there is an error it is returned in err.
572 static int htree_dirblock_to_tree(struct file
*dir_file
,
573 struct inode
*dir
, ext4_lblk_t block
,
574 struct dx_hash_info
*hinfo
,
575 __u32 start_hash
, __u32 start_minor_hash
)
577 struct buffer_head
*bh
;
578 struct ext4_dir_entry_2
*de
, *top
;
581 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
582 (unsigned long)block
));
583 if (!(bh
= ext4_bread (NULL
, dir
, block
, 0, &err
)))
586 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
587 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
588 dir
->i_sb
->s_blocksize
-
589 EXT4_DIR_REC_LEN(0));
590 for (; de
< top
; de
= ext4_next_entry(de
)) {
591 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir
, de
, bh
,
592 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
593 +((char *)de
- bh
->b_data
))) {
594 /* On error, skip the f_pos to the next block. */
595 dir_file
->f_pos
= (dir_file
->f_pos
|
596 (dir
->i_sb
->s_blocksize
- 1)) + 1;
600 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
601 if ((hinfo
->hash
< start_hash
) ||
602 ((hinfo
->hash
== start_hash
) &&
603 (hinfo
->minor_hash
< start_minor_hash
)))
607 if ((err
= ext4_htree_store_dirent(dir_file
,
608 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
620 * This function fills a red-black tree with information from a
621 * directory. We start scanning the directory in hash order, starting
622 * at start_hash and start_minor_hash.
624 * This function returns the number of entries inserted into the tree,
625 * or a negative error code.
627 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
628 __u32 start_minor_hash
, __u32
*next_hash
)
630 struct dx_hash_info hinfo
;
631 struct ext4_dir_entry_2
*de
;
632 struct dx_frame frames
[2], *frame
;
639 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
640 start_hash
, start_minor_hash
));
641 dir
= dir_file
->f_path
.dentry
->d_inode
;
642 if (!(EXT4_I(dir
)->i_flags
& EXT4_INDEX_FL
)) {
643 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
644 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
645 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
646 start_hash
, start_minor_hash
);
650 hinfo
.hash
= start_hash
;
651 hinfo
.minor_hash
= 0;
652 frame
= dx_probe(NULL
, dir_file
->f_path
.dentry
->d_inode
, &hinfo
, frames
, &err
);
656 /* Add '.' and '..' from the htree header */
657 if (!start_hash
&& !start_minor_hash
) {
658 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
659 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
663 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
664 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
665 de
= ext4_next_entry(de
);
666 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
672 block
= dx_get_block(frame
->at
);
673 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
674 start_hash
, start_minor_hash
);
681 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
682 frame
, frames
, &hashval
);
683 *next_hash
= hashval
;
689 * Stop if: (a) there are no more entries, or
690 * (b) we have inserted at least one entry and the
691 * next hash value is not a continuation
694 (count
&& ((hashval
& 1) == 0)))
698 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
699 "next hash: %x\n", count
, *next_hash
));
708 * Directory block splitting, compacting
712 * Create map of hash values, offsets, and sizes, stored at end of block.
713 * Returns number of entries mapped.
715 static int dx_make_map (struct ext4_dir_entry_2
*de
, int size
,
716 struct dx_hash_info
*hinfo
, struct dx_map_entry
*map_tail
)
719 char *base
= (char *) de
;
720 struct dx_hash_info h
= *hinfo
;
722 while ((char *) de
< base
+ size
)
724 if (de
->name_len
&& de
->inode
) {
725 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
727 map_tail
->hash
= h
.hash
;
728 map_tail
->offs
= (u16
) ((char *) de
- base
);
729 map_tail
->size
= le16_to_cpu(de
->rec_len
);
733 /* XXX: do we need to check rec_len == 0 case? -Chris */
734 de
= ext4_next_entry(de
);
739 /* Sort map by hash value */
740 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
742 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
744 /* Combsort until bubble sort doesn't suck */
747 if (count
- 9 < 2) /* 9, 10 -> 11 */
749 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
750 if (p
->hash
< q
->hash
)
753 /* Garden variety bubble sort */
758 if (q
[1].hash
>= q
[0].hash
)
766 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
768 struct dx_entry
*entries
= frame
->entries
;
769 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
770 int count
= dx_get_count(entries
);
772 assert(count
< dx_get_limit(entries
));
773 assert(old
< entries
+ count
);
774 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
775 dx_set_hash(new, hash
);
776 dx_set_block(new, block
);
777 dx_set_count(entries
, count
+ 1);
780 static void ext4_update_dx_flag(struct inode
*inode
)
782 if (!EXT4_HAS_COMPAT_FEATURE(inode
->i_sb
,
783 EXT4_FEATURE_COMPAT_DIR_INDEX
))
784 EXT4_I(inode
)->i_flags
&= ~EXT4_INDEX_FL
;
788 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
790 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
791 * `de != NULL' is guaranteed by caller.
793 static inline int ext4_match (int len
, const char * const name
,
794 struct ext4_dir_entry_2
* de
)
796 if (len
!= de
->name_len
)
800 return !memcmp(name
, de
->name
, len
);
804 * Returns 0 if not found, -1 on failure, and 1 on success
806 static inline int search_dirblock(struct buffer_head
*bh
,
808 struct dentry
*dentry
,
809 unsigned long offset
,
810 struct ext4_dir_entry_2
** res_dir
)
812 struct ext4_dir_entry_2
* de
;
815 const char *name
= dentry
->d_name
.name
;
816 int namelen
= dentry
->d_name
.len
;
818 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
819 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
820 while ((char *) de
< dlimit
) {
821 /* this code is executed quadratically often */
822 /* do minimal checking `by hand' */
824 if ((char *) de
+ namelen
<= dlimit
&&
825 ext4_match (namelen
, name
, de
)) {
826 /* found a match - just to be sure, do a full check */
827 if (!ext4_check_dir_entry("ext4_find_entry",
828 dir
, de
, bh
, offset
))
833 /* prevent looping on a bad block */
834 de_len
= ext4_rec_len_from_disk(de
->rec_len
);
838 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
847 * finds an entry in the specified directory with the wanted name. It
848 * returns the cache buffer in which the entry was found, and the entry
849 * itself (as a parameter - res_dir). It does NOT read the inode of the
850 * entry - you'll have to do that yourself if you want to.
852 * The returned buffer_head has ->b_count elevated. The caller is expected
853 * to brelse() it when appropriate.
855 static struct buffer_head
* ext4_find_entry (struct dentry
*dentry
,
856 struct ext4_dir_entry_2
** res_dir
)
858 struct super_block
*sb
;
859 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
860 struct buffer_head
*bh
, *ret
= NULL
;
861 ext4_lblk_t start
, block
, b
;
862 int ra_max
= 0; /* Number of bh's in the readahead
864 int ra_ptr
= 0; /* Current index into readahead
869 struct inode
*dir
= dentry
->d_parent
->d_inode
;
874 namelen
= dentry
->d_name
.len
;
875 if (namelen
> EXT4_NAME_LEN
)
878 bh
= ext4_dx_find_entry(dentry
, res_dir
, &err
);
880 * On success, or if the error was file not found,
881 * return. Otherwise, fall back to doing a search the
884 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
886 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
889 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
890 start
= EXT4_I(dir
)->i_dir_start_lookup
;
891 if (start
>= nblocks
)
897 * We deal with the read-ahead logic here.
899 if (ra_ptr
>= ra_max
) {
900 /* Refill the readahead buffer */
903 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
905 * Terminate if we reach the end of the
906 * directory and must wrap, or if our
907 * search has finished at this block.
909 if (b
>= nblocks
|| (num
&& block
== start
)) {
910 bh_use
[ra_max
] = NULL
;
914 bh
= ext4_getblk(NULL
, dir
, b
++, 0, &err
);
917 ll_rw_block(READ_META
, 1, &bh
);
920 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
923 if (!buffer_uptodate(bh
)) {
924 /* read error, skip block & hope for the best */
925 ext4_error(sb
, __func__
, "reading directory #%lu "
926 "offset %lu", dir
->i_ino
,
927 (unsigned long)block
);
931 i
= search_dirblock(bh
, dir
, dentry
,
932 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
934 EXT4_I(dir
)->i_dir_start_lookup
= block
;
936 goto cleanup_and_exit
;
940 goto cleanup_and_exit
;
943 if (++block
>= nblocks
)
945 } while (block
!= start
);
948 * If the directory has grown while we were searching, then
949 * search the last part of the directory before giving up.
952 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
953 if (block
< nblocks
) {
959 /* Clean up the read-ahead blocks */
960 for (; ra_ptr
< ra_max
; ra_ptr
++)
961 brelse(bh_use
[ra_ptr
]);
965 static struct buffer_head
* ext4_dx_find_entry(struct dentry
*dentry
,
966 struct ext4_dir_entry_2
**res_dir
, int *err
)
968 struct super_block
* sb
;
969 struct dx_hash_info hinfo
;
971 struct dx_frame frames
[2], *frame
;
972 struct ext4_dir_entry_2
*de
, *top
;
973 struct buffer_head
*bh
;
976 int namelen
= dentry
->d_name
.len
;
977 const u8
*name
= dentry
->d_name
.name
;
978 struct inode
*dir
= dentry
->d_parent
->d_inode
;
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(dentry
, NULL
, &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(dentry
, &de
);
1047 unsigned long 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: %lu", ino
);
1052 return ERR_PTR(-EIO
);
1054 inode
= ext4_iget(dir
->i_sb
, ino
);
1056 return ERR_CAST(inode
);
1058 return d_splice_alias(inode
, dentry
);
1062 struct dentry
*ext4_get_parent(struct dentry
*child
)
1065 struct dentry
*parent
;
1066 struct inode
*inode
;
1067 struct dentry dotdot
;
1068 struct ext4_dir_entry_2
* de
;
1069 struct buffer_head
*bh
;
1071 dotdot
.d_name
.name
= "..";
1072 dotdot
.d_name
.len
= 2;
1073 dotdot
.d_parent
= child
; /* confusing, isn't it! */
1075 bh
= ext4_find_entry(&dotdot
, &de
);
1078 return ERR_PTR(-ENOENT
);
1079 ino
= le32_to_cpu(de
->inode
);
1082 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1083 ext4_error(child
->d_inode
->i_sb
, "ext4_get_parent",
1084 "bad inode number: %lu", ino
);
1085 return ERR_PTR(-EIO
);
1088 inode
= ext4_iget(child
->d_inode
->i_sb
, ino
);
1090 return ERR_CAST(inode
);
1092 parent
= d_alloc_anon(inode
);
1095 parent
= ERR_PTR(-ENOMEM
);
1101 static unsigned char ext4_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1102 [S_IFREG
>> S_SHIFT
] = EXT4_FT_REG_FILE
,
1103 [S_IFDIR
>> S_SHIFT
] = EXT4_FT_DIR
,
1104 [S_IFCHR
>> S_SHIFT
] = EXT4_FT_CHRDEV
,
1105 [S_IFBLK
>> S_SHIFT
] = EXT4_FT_BLKDEV
,
1106 [S_IFIFO
>> S_SHIFT
] = EXT4_FT_FIFO
,
1107 [S_IFSOCK
>> S_SHIFT
] = EXT4_FT_SOCK
,
1108 [S_IFLNK
>> S_SHIFT
] = EXT4_FT_SYMLINK
,
1111 static inline void ext4_set_de_type(struct super_block
*sb
,
1112 struct ext4_dir_entry_2
*de
,
1114 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FILETYPE
))
1115 de
->file_type
= ext4_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1119 * Move count entries from end of map between two memory locations.
1120 * Returns pointer to last entry moved.
1122 static struct ext4_dir_entry_2
*
1123 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
)
1125 unsigned rec_len
= 0;
1128 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*) (from
+ map
->offs
);
1129 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1130 memcpy (to
, de
, rec_len
);
1131 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1132 ext4_rec_len_to_disk(rec_len
);
1137 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1141 * Compact each dir entry in the range to the minimal rec_len.
1142 * Returns pointer to last entry in range.
1144 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, int size
)
1146 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1147 unsigned rec_len
= 0;
1150 while ((char*)de
< base
+ size
) {
1151 next
= ext4_next_entry(de
);
1152 if (de
->inode
&& de
->name_len
) {
1153 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1155 memmove(to
, de
, rec_len
);
1156 to
->rec_len
= ext4_rec_len_to_disk(rec_len
);
1158 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1166 * Split a full leaf block to make room for a new dir entry.
1167 * Allocate a new block, and move entries so that they are approx. equally full.
1168 * Returns pointer to de in block into which the new entry will be inserted.
1170 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1171 struct buffer_head
**bh
,struct dx_frame
*frame
,
1172 struct dx_hash_info
*hinfo
, int *error
)
1174 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1175 unsigned count
, continued
;
1176 struct buffer_head
*bh2
;
1177 ext4_lblk_t newblock
;
1179 struct dx_map_entry
*map
;
1180 char *data1
= (*bh
)->b_data
, *data2
;
1181 unsigned split
, move
, size
, i
;
1182 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1185 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1192 BUFFER_TRACE(*bh
, "get_write_access");
1193 err
= ext4_journal_get_write_access(handle
, *bh
);
1197 BUFFER_TRACE(frame
->bh
, "get_write_access");
1198 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1202 data2
= bh2
->b_data
;
1204 /* create map in the end of data2 block */
1205 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1206 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1207 blocksize
, hinfo
, map
);
1209 dx_sort_map(map
, count
);
1210 /* Split the existing block in the middle, size-wise */
1213 for (i
= count
-1; i
>= 0; i
--) {
1214 /* is more than half of this entry in 2nd half of the block? */
1215 if (size
+ map
[i
].size
/2 > blocksize
/2)
1217 size
+= map
[i
].size
;
1220 /* map index at which we will split */
1221 split
= count
- move
;
1222 hash2
= map
[split
].hash
;
1223 continued
= hash2
== map
[split
- 1].hash
;
1224 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1225 (unsigned long)dx_get_block(frame
->at
),
1226 hash2
, split
, count
-split
));
1228 /* Fancy dance to stay within two buffers */
1229 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
);
1230 de
= dx_pack_dirents(data1
, blocksize
);
1231 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1232 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ blocksize
- (char *) de2
);
1233 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1234 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1236 /* Which block gets the new entry? */
1237 if (hinfo
->hash
>= hash2
)
1242 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1243 err
= ext4_journal_dirty_metadata(handle
, bh2
);
1246 err
= ext4_journal_dirty_metadata(handle
, frame
->bh
);
1250 dxtrace(dx_show_index("frame", frame
->entries
));
1257 ext4_std_error(dir
->i_sb
, err
);
1264 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1265 * it points to a directory entry which is guaranteed to be large
1266 * enough for new directory entry. If de is NULL, then
1267 * add_dirent_to_buf will attempt search the directory block for
1268 * space. It will return -ENOSPC if no space is available, and -EIO
1269 * and -EEXIST if directory entry already exists.
1271 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1272 * all other cases bh is released.
1274 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1275 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1276 struct buffer_head
*bh
)
1278 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1279 const char *name
= dentry
->d_name
.name
;
1280 int namelen
= dentry
->d_name
.len
;
1281 unsigned long offset
= 0;
1282 unsigned short reclen
;
1283 int nlen
, rlen
, err
;
1286 reclen
= EXT4_DIR_REC_LEN(namelen
);
1288 de
= (struct ext4_dir_entry_2
*)bh
->b_data
;
1289 top
= bh
->b_data
+ dir
->i_sb
->s_blocksize
- reclen
;
1290 while ((char *) de
<= top
) {
1291 if (!ext4_check_dir_entry("ext4_add_entry", dir
, de
,
1296 if (ext4_match(namelen
, name
, de
)) {
1300 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1301 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1302 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1304 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1307 if ((char *) de
> top
)
1310 BUFFER_TRACE(bh
, "get_write_access");
1311 err
= ext4_journal_get_write_access(handle
, bh
);
1313 ext4_std_error(dir
->i_sb
, err
);
1318 /* By now the buffer is marked for journaling */
1319 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1320 rlen
= ext4_rec_len_from_disk(de
->rec_len
);
1322 struct ext4_dir_entry_2
*de1
= (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1323 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
);
1324 de
->rec_len
= ext4_rec_len_to_disk(nlen
);
1327 de
->file_type
= EXT4_FT_UNKNOWN
;
1329 de
->inode
= cpu_to_le32(inode
->i_ino
);
1330 ext4_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1333 de
->name_len
= namelen
;
1334 memcpy(de
->name
, name
, namelen
);
1336 * XXX shouldn't update any times until successful
1337 * completion of syscall, but too many callers depend
1340 * XXX similarly, too many callers depend on
1341 * ext4_new_inode() setting the times, but error
1342 * recovery deletes the inode, so the worst that can
1343 * happen is that the times are slightly out of date
1344 * and/or different from the directory change time.
1346 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1347 ext4_update_dx_flag(dir
);
1349 ext4_mark_inode_dirty(handle
, dir
);
1350 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1351 err
= ext4_journal_dirty_metadata(handle
, bh
);
1353 ext4_std_error(dir
->i_sb
, err
);
1359 * This converts a one block unindexed directory to a 3 block indexed
1360 * directory, and adds the dentry to the indexed directory.
1362 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1363 struct inode
*inode
, struct buffer_head
*bh
)
1365 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1366 const char *name
= dentry
->d_name
.name
;
1367 int namelen
= dentry
->d_name
.len
;
1368 struct buffer_head
*bh2
;
1369 struct dx_root
*root
;
1370 struct dx_frame frames
[2], *frame
;
1371 struct dx_entry
*entries
;
1372 struct ext4_dir_entry_2
*de
, *de2
;
1377 struct dx_hash_info hinfo
;
1379 struct fake_dirent
*fde
;
1381 blocksize
= dir
->i_sb
->s_blocksize
;
1382 dxtrace(printk(KERN_DEBUG
"Creating index\n"));
1383 retval
= ext4_journal_get_write_access(handle
, bh
);
1385 ext4_std_error(dir
->i_sb
, retval
);
1389 root
= (struct dx_root
*) bh
->b_data
;
1391 bh2
= ext4_append(handle
, dir
, &block
, &retval
);
1396 EXT4_I(dir
)->i_flags
|= EXT4_INDEX_FL
;
1397 data1
= bh2
->b_data
;
1399 /* The 0th block becomes the root, move the dirents out */
1400 fde
= &root
->dotdot
;
1401 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1402 ext4_rec_len_from_disk(fde
->rec_len
));
1403 len
= ((char *) root
) + blocksize
- (char *) de
;
1404 memcpy (data1
, de
, len
);
1405 de
= (struct ext4_dir_entry_2
*) data1
;
1407 while ((char *)(de2
= ext4_next_entry(de
)) < top
)
1409 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
);
1410 /* Initialize the root; the dot dirents already exist */
1411 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1412 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2));
1413 memset (&root
->info
, 0, sizeof(root
->info
));
1414 root
->info
.info_length
= sizeof(root
->info
);
1415 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1416 entries
= root
->entries
;
1417 dx_set_block(entries
, 1);
1418 dx_set_count(entries
, 1);
1419 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1421 /* Initialize as for dx_probe */
1422 hinfo
.hash_version
= root
->info
.hash_version
;
1423 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1424 ext4fs_dirhash(name
, namelen
, &hinfo
);
1426 frame
->entries
= entries
;
1427 frame
->at
= entries
;
1430 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1431 dx_release (frames
);
1435 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1441 * adds a file entry to the specified directory, using the same
1442 * semantics as ext4_find_entry(). It returns NULL if it failed.
1444 * NOTE!! The inode part of 'de' is left at 0 - which means you
1445 * may not sleep between calling this and putting something into
1446 * the entry, as someone else might have used it while you slept.
1448 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1449 struct inode
*inode
)
1451 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1452 unsigned long offset
;
1453 struct buffer_head
*bh
;
1454 struct ext4_dir_entry_2
*de
;
1455 struct super_block
*sb
;
1459 ext4_lblk_t block
, blocks
;
1462 blocksize
= sb
->s_blocksize
;
1463 if (!dentry
->d_name
.len
)
1466 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1467 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1469 EXT4_I(dir
)->i_flags
&= ~EXT4_INDEX_FL
;
1471 ext4_mark_inode_dirty(handle
, dir
);
1473 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1474 for (block
= 0, offset
= 0; block
< blocks
; block
++) {
1475 bh
= ext4_bread(handle
, dir
, block
, 0, &retval
);
1478 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1479 if (retval
!= -ENOSPC
)
1482 if (blocks
== 1 && !dx_fallback
&&
1483 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1484 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1487 bh
= ext4_append(handle
, dir
, &block
, &retval
);
1490 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1492 de
->rec_len
= ext4_rec_len_to_disk(blocksize
);
1493 return add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1497 * Returns 0 for success, or a negative error value
1499 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1500 struct inode
*inode
)
1502 struct dx_frame frames
[2], *frame
;
1503 struct dx_entry
*entries
, *at
;
1504 struct dx_hash_info hinfo
;
1505 struct buffer_head
*bh
;
1506 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1507 struct super_block
*sb
= dir
->i_sb
;
1508 struct ext4_dir_entry_2
*de
;
1511 frame
= dx_probe(dentry
, NULL
, &hinfo
, frames
, &err
);
1514 entries
= frame
->entries
;
1517 if (!(bh
= ext4_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1520 BUFFER_TRACE(bh
, "get_write_access");
1521 err
= ext4_journal_get_write_access(handle
, bh
);
1525 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1526 if (err
!= -ENOSPC
) {
1531 /* Block full, should compress but for now just split */
1532 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
1533 dx_get_count(entries
), dx_get_limit(entries
)));
1534 /* Need to split index? */
1535 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1536 ext4_lblk_t newblock
;
1537 unsigned icount
= dx_get_count(entries
);
1538 int levels
= frame
- frames
;
1539 struct dx_entry
*entries2
;
1540 struct dx_node
*node2
;
1541 struct buffer_head
*bh2
;
1543 if (levels
&& (dx_get_count(frames
->entries
) ==
1544 dx_get_limit(frames
->entries
))) {
1545 ext4_warning(sb
, __func__
,
1546 "Directory index full!");
1550 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1553 node2
= (struct dx_node
*)(bh2
->b_data
);
1554 entries2
= node2
->entries
;
1555 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
);
1556 node2
->fake
.inode
= 0;
1557 BUFFER_TRACE(frame
->bh
, "get_write_access");
1558 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1562 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1563 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1564 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
1567 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1568 err
= ext4_journal_get_write_access(handle
,
1573 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
1574 icount2
* sizeof(struct dx_entry
));
1575 dx_set_count(entries
, icount1
);
1576 dx_set_count(entries2
, icount2
);
1577 dx_set_limit(entries2
, dx_node_limit(dir
));
1579 /* Which index block gets the new entry? */
1580 if (at
- entries
>= icount1
) {
1581 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1582 frame
->entries
= entries
= entries2
;
1583 swap(frame
->bh
, bh2
);
1585 dx_insert_block(frames
+ 0, hash2
, newblock
);
1586 dxtrace(dx_show_index("node", frames
[1].entries
));
1587 dxtrace(dx_show_index("node",
1588 ((struct dx_node
*) bh2
->b_data
)->entries
));
1589 err
= ext4_journal_dirty_metadata(handle
, bh2
);
1594 dxtrace(printk(KERN_DEBUG
1595 "Creating second level index...\n"));
1596 memcpy((char *) entries2
, (char *) entries
,
1597 icount
* sizeof(struct dx_entry
));
1598 dx_set_limit(entries2
, dx_node_limit(dir
));
1601 dx_set_count(entries
, 1);
1602 dx_set_block(entries
+ 0, newblock
);
1603 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1605 /* Add new access path frame */
1607 frame
->at
= at
= at
- entries
+ entries2
;
1608 frame
->entries
= entries
= entries2
;
1610 err
= ext4_journal_get_write_access(handle
,
1615 ext4_journal_dirty_metadata(handle
, frames
[0].bh
);
1617 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1620 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1625 ext4_std_error(dir
->i_sb
, err
);
1634 * ext4_delete_entry deletes a directory entry by merging it with the
1637 static int ext4_delete_entry(handle_t
*handle
,
1639 struct ext4_dir_entry_2
*de_del
,
1640 struct buffer_head
*bh
)
1642 struct ext4_dir_entry_2
*de
, *pde
;
1647 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1648 while (i
< bh
->b_size
) {
1649 if (!ext4_check_dir_entry("ext4_delete_entry", dir
, de
, bh
, i
))
1652 BUFFER_TRACE(bh
, "get_write_access");
1653 ext4_journal_get_write_access(handle
, bh
);
1655 pde
->rec_len
= ext4_rec_len_to_disk(
1656 ext4_rec_len_from_disk(pde
->rec_len
) +
1657 ext4_rec_len_from_disk(de
->rec_len
));
1661 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1662 ext4_journal_dirty_metadata(handle
, bh
);
1665 i
+= ext4_rec_len_from_disk(de
->rec_len
);
1667 de
= ext4_next_entry(de
);
1673 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1674 * since this indicates that nlinks count was previously 1.
1676 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
1679 if (is_dx(inode
) && inode
->i_nlink
> 1) {
1680 /* limit is 16-bit i_links_count */
1681 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
1683 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
1684 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
1690 * If a directory had nlink == 1, then we should let it be 1. This indicates
1691 * directory has >EXT4_LINK_MAX subdirs.
1693 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
1696 if (S_ISDIR(inode
->i_mode
) && inode
->i_nlink
== 0)
1701 static int ext4_add_nondir(handle_t
*handle
,
1702 struct dentry
*dentry
, struct inode
*inode
)
1704 int err
= ext4_add_entry(handle
, dentry
, inode
);
1706 ext4_mark_inode_dirty(handle
, inode
);
1707 d_instantiate(dentry
, inode
);
1716 * By the time this is called, we already have created
1717 * the directory cache entry for the new file, but it
1718 * is so far negative - it has no inode.
1720 * If the create succeeds, we fill in the inode information
1721 * with d_instantiate().
1723 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1724 struct nameidata
*nd
)
1727 struct inode
*inode
;
1728 int err
, retries
= 0;
1731 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1732 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1733 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1735 return PTR_ERR(handle
);
1737 if (IS_DIRSYNC(dir
))
1740 inode
= ext4_new_inode (handle
, dir
, mode
);
1741 err
= PTR_ERR(inode
);
1742 if (!IS_ERR(inode
)) {
1743 inode
->i_op
= &ext4_file_inode_operations
;
1744 inode
->i_fop
= &ext4_file_operations
;
1745 ext4_set_aops(inode
);
1746 err
= ext4_add_nondir(handle
, dentry
, inode
);
1748 ext4_journal_stop(handle
);
1749 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1754 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
1755 int mode
, dev_t rdev
)
1758 struct inode
*inode
;
1759 int err
, retries
= 0;
1761 if (!new_valid_dev(rdev
))
1765 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1766 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1767 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1769 return PTR_ERR(handle
);
1771 if (IS_DIRSYNC(dir
))
1774 inode
= ext4_new_inode(handle
, dir
, mode
);
1775 err
= PTR_ERR(inode
);
1776 if (!IS_ERR(inode
)) {
1777 init_special_inode(inode
, inode
->i_mode
, rdev
);
1778 #ifdef CONFIG_EXT4DEV_FS_XATTR
1779 inode
->i_op
= &ext4_special_inode_operations
;
1781 err
= ext4_add_nondir(handle
, dentry
, inode
);
1783 ext4_journal_stop(handle
);
1784 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1789 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1792 struct inode
*inode
;
1793 struct buffer_head
*dir_block
;
1794 struct ext4_dir_entry_2
*de
;
1795 int err
, retries
= 0;
1797 if (EXT4_DIR_LINK_MAX(dir
))
1801 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1802 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1803 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
1805 return PTR_ERR(handle
);
1807 if (IS_DIRSYNC(dir
))
1810 inode
= ext4_new_inode(handle
, dir
, S_IFDIR
| mode
);
1811 err
= PTR_ERR(inode
);
1815 inode
->i_op
= &ext4_dir_inode_operations
;
1816 inode
->i_fop
= &ext4_dir_operations
;
1817 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1818 dir_block
= ext4_bread(handle
, inode
, 0, 1, &err
);
1820 goto out_clear_inode
;
1821 BUFFER_TRACE(dir_block
, "get_write_access");
1822 ext4_journal_get_write_access(handle
, dir_block
);
1823 de
= (struct ext4_dir_entry_2
*) dir_block
->b_data
;
1824 de
->inode
= cpu_to_le32(inode
->i_ino
);
1826 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
));
1827 strcpy(de
->name
, ".");
1828 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1829 de
= ext4_next_entry(de
);
1830 de
->inode
= cpu_to_le32(dir
->i_ino
);
1831 de
->rec_len
= ext4_rec_len_to_disk(inode
->i_sb
->s_blocksize
-
1832 EXT4_DIR_REC_LEN(1));
1834 strcpy(de
->name
, "..");
1835 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1837 BUFFER_TRACE(dir_block
, "call ext4_journal_dirty_metadata");
1838 ext4_journal_dirty_metadata(handle
, dir_block
);
1840 ext4_mark_inode_dirty(handle
, inode
);
1841 err
= ext4_add_entry(handle
, dentry
, inode
);
1845 ext4_mark_inode_dirty(handle
, inode
);
1849 ext4_inc_count(handle
, dir
);
1850 ext4_update_dx_flag(dir
);
1851 ext4_mark_inode_dirty(handle
, dir
);
1852 d_instantiate(dentry
, inode
);
1854 ext4_journal_stop(handle
);
1855 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1861 * routine to check that the specified directory is empty (for rmdir)
1863 static int empty_dir(struct inode
*inode
)
1865 unsigned long offset
;
1866 struct buffer_head
*bh
;
1867 struct ext4_dir_entry_2
*de
, *de1
;
1868 struct super_block
*sb
;
1872 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1873 !(bh
= ext4_bread(NULL
, inode
, 0, 0, &err
))) {
1875 ext4_error(inode
->i_sb
, __func__
,
1876 "error %d reading directory #%lu offset 0",
1879 ext4_warning(inode
->i_sb
, __func__
,
1880 "bad directory (dir #%lu) - no data block",
1884 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1885 de1
= ext4_next_entry(de
);
1886 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1887 !le32_to_cpu(de1
->inode
) ||
1888 strcmp(".", de
->name
) ||
1889 strcmp("..", de1
->name
)) {
1890 ext4_warning(inode
->i_sb
, "empty_dir",
1891 "bad directory (dir #%lu) - no `.' or `..'",
1896 offset
= ext4_rec_len_from_disk(de
->rec_len
) +
1897 ext4_rec_len_from_disk(de1
->rec_len
);
1898 de
= ext4_next_entry(de1
);
1899 while (offset
< inode
->i_size
) {
1901 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1904 bh
= ext4_bread(NULL
, inode
,
1905 offset
>> EXT4_BLOCK_SIZE_BITS(sb
), 0, &err
);
1908 ext4_error(sb
, __func__
,
1909 "error %d reading directory"
1911 err
, inode
->i_ino
, offset
);
1912 offset
+= sb
->s_blocksize
;
1915 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1917 if (!ext4_check_dir_entry("empty_dir", inode
, de
, bh
, offset
)) {
1918 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
1920 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1923 if (le32_to_cpu(de
->inode
)) {
1927 offset
+= ext4_rec_len_from_disk(de
->rec_len
);
1928 de
= ext4_next_entry(de
);
1934 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1935 * such inodes, starting at the superblock, in case we crash before the
1936 * file is closed/deleted, or in case the inode truncate spans multiple
1937 * transactions and the last transaction is not recovered after a crash.
1939 * At filesystem recovery time, we walk this list deleting unlinked
1940 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1942 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
1944 struct super_block
*sb
= inode
->i_sb
;
1945 struct ext4_iloc iloc
;
1949 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
1952 /* Orphan handling is only valid for files with data blocks
1953 * being truncated, or files being unlinked. */
1955 /* @@@ FIXME: Observation from aviro:
1956 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1957 * here (on lock_super()), so race with ext4_link() which might bump
1958 * ->i_nlink. For, say it, character device. Not a regular file,
1959 * not a directory, not a symlink and ->i_nlink > 0.
1961 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1962 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1964 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1965 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1969 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
1973 /* Insert this inode at the head of the on-disk orphan list... */
1974 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT4_SB(sb
)->s_es
->s_last_orphan
);
1975 EXT4_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
1976 err
= ext4_journal_dirty_metadata(handle
, EXT4_SB(sb
)->s_sbh
);
1977 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1981 /* Only add to the head of the in-memory list if all the
1982 * previous operations succeeded. If the orphan_add is going to
1983 * fail (possibly taking the journal offline), we can't risk
1984 * leaving the inode on the orphan list: stray orphan-list
1985 * entries can cause panics at unmount time.
1987 * This is safe: on error we're going to ignore the orphan list
1988 * anyway on the next recovery. */
1990 list_add(&EXT4_I(inode
)->i_orphan
, &EXT4_SB(sb
)->s_orphan
);
1992 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
1993 jbd_debug(4, "orphan inode %lu will point to %d\n",
1994 inode
->i_ino
, NEXT_ORPHAN(inode
));
1997 ext4_std_error(inode
->i_sb
, err
);
2002 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2003 * of such inodes stored on disk, because it is finally being cleaned up.
2005 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2007 struct list_head
*prev
;
2008 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2009 struct ext4_sb_info
*sbi
;
2010 unsigned long ino_next
;
2011 struct ext4_iloc iloc
;
2014 lock_super(inode
->i_sb
);
2015 if (list_empty(&ei
->i_orphan
)) {
2016 unlock_super(inode
->i_sb
);
2020 ino_next
= NEXT_ORPHAN(inode
);
2021 prev
= ei
->i_orphan
.prev
;
2022 sbi
= EXT4_SB(inode
->i_sb
);
2024 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2026 list_del_init(&ei
->i_orphan
);
2028 /* If we're on an error path, we may not have a valid
2029 * transaction handle with which to update the orphan list on
2030 * disk, but we still need to remove the inode from the linked
2031 * list in memory. */
2035 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2039 if (prev
== &sbi
->s_orphan
) {
2040 jbd_debug(4, "superblock will point to %lu\n", ino_next
);
2041 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2042 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2045 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2046 err
= ext4_journal_dirty_metadata(handle
, sbi
->s_sbh
);
2048 struct ext4_iloc iloc2
;
2049 struct inode
*i_prev
=
2050 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2052 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2053 i_prev
->i_ino
, ino_next
);
2054 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2057 NEXT_ORPHAN(i_prev
) = ino_next
;
2058 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2062 NEXT_ORPHAN(inode
) = 0;
2063 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2066 ext4_std_error(inode
->i_sb
, err
);
2068 unlock_super(inode
->i_sb
);
2076 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2079 struct inode
*inode
;
2080 struct buffer_head
*bh
;
2081 struct ext4_dir_entry_2
*de
;
2084 /* Initialize quotas before so that eventual writes go in
2085 * separate transaction */
2086 DQUOT_INIT(dentry
->d_inode
);
2087 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2089 return PTR_ERR(handle
);
2092 bh
= ext4_find_entry(dentry
, &de
);
2096 if (IS_DIRSYNC(dir
))
2099 inode
= dentry
->d_inode
;
2102 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2105 retval
= -ENOTEMPTY
;
2106 if (!empty_dir(inode
))
2109 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2112 if (!EXT4_DIR_LINK_EMPTY(inode
))
2113 ext4_warning(inode
->i_sb
, "ext4_rmdir",
2114 "empty directory has too many links (%d)",
2118 /* There's no need to set i_disksize: the fact that i_nlink is
2119 * zero will ensure that the right thing happens during any
2122 ext4_orphan_add(handle
, inode
);
2123 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2124 ext4_mark_inode_dirty(handle
, inode
);
2125 ext4_dec_count(handle
, dir
);
2126 ext4_update_dx_flag(dir
);
2127 ext4_mark_inode_dirty(handle
, dir
);
2130 ext4_journal_stop(handle
);
2135 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2138 struct inode
*inode
;
2139 struct buffer_head
*bh
;
2140 struct ext4_dir_entry_2
*de
;
2143 /* Initialize quotas before so that eventual writes go
2144 * in separate transaction */
2145 DQUOT_INIT(dentry
->d_inode
);
2146 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2148 return PTR_ERR(handle
);
2150 if (IS_DIRSYNC(dir
))
2154 bh
= ext4_find_entry(dentry
, &de
);
2158 inode
= dentry
->d_inode
;
2161 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2164 if (!inode
->i_nlink
) {
2165 ext4_warning(inode
->i_sb
, "ext4_unlink",
2166 "Deleting nonexistent file (%lu), %d",
2167 inode
->i_ino
, inode
->i_nlink
);
2170 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2173 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2174 ext4_update_dx_flag(dir
);
2175 ext4_mark_inode_dirty(handle
, dir
);
2177 if (!inode
->i_nlink
)
2178 ext4_orphan_add(handle
, inode
);
2179 inode
->i_ctime
= ext4_current_time(inode
);
2180 ext4_mark_inode_dirty(handle
, inode
);
2184 ext4_journal_stop(handle
);
2189 static int ext4_symlink(struct inode
*dir
,
2190 struct dentry
*dentry
, const char *symname
)
2193 struct inode
*inode
;
2194 int l
, err
, retries
= 0;
2196 l
= strlen(symname
)+1;
2197 if (l
> dir
->i_sb
->s_blocksize
)
2198 return -ENAMETOOLONG
;
2201 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2202 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 5 +
2203 2*EXT4_QUOTA_INIT_BLOCKS(dir
->i_sb
));
2205 return PTR_ERR(handle
);
2207 if (IS_DIRSYNC(dir
))
2210 inode
= ext4_new_inode(handle
, dir
, S_IFLNK
|S_IRWXUGO
);
2211 err
= PTR_ERR(inode
);
2215 if (l
> sizeof(EXT4_I(inode
)->i_data
)) {
2216 inode
->i_op
= &ext4_symlink_inode_operations
;
2217 ext4_set_aops(inode
);
2219 * page_symlink() calls into ext4_prepare/commit_write.
2220 * We have a transaction open. All is sweetness. It also sets
2221 * i_size in generic_commit_write().
2223 err
= __page_symlink(inode
, symname
, l
,
2224 mapping_gfp_mask(inode
->i_mapping
) & ~__GFP_FS
);
2227 ext4_mark_inode_dirty(handle
, inode
);
2232 /* clear the extent format for fast symlink */
2233 EXT4_I(inode
)->i_flags
&= ~EXT4_EXTENTS_FL
;
2234 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2235 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2236 inode
->i_size
= l
-1;
2238 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2239 err
= ext4_add_nondir(handle
, dentry
, inode
);
2241 ext4_journal_stop(handle
);
2242 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2247 static int ext4_link(struct dentry
*old_dentry
,
2248 struct inode
*dir
, struct dentry
*dentry
)
2251 struct inode
*inode
= old_dentry
->d_inode
;
2252 int err
, retries
= 0;
2254 if (EXT4_DIR_LINK_MAX(inode
))
2258 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2259 * otherwise has the potential to corrupt the orphan inode list.
2261 if (inode
->i_nlink
== 0)
2265 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2266 EXT4_INDEX_EXTRA_TRANS_BLOCKS
);
2268 return PTR_ERR(handle
);
2270 if (IS_DIRSYNC(dir
))
2273 inode
->i_ctime
= ext4_current_time(inode
);
2274 ext4_inc_count(handle
, inode
);
2275 atomic_inc(&inode
->i_count
);
2277 err
= ext4_add_nondir(handle
, dentry
, inode
);
2278 ext4_journal_stop(handle
);
2279 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2284 #define PARENT_INO(buffer) \
2285 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2288 * Anybody can rename anything with this: the permission checks are left to the
2289 * higher-level routines.
2291 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2292 struct inode
*new_dir
, struct dentry
*new_dentry
)
2295 struct inode
*old_inode
, *new_inode
;
2296 struct buffer_head
*old_bh
, *new_bh
, *dir_bh
;
2297 struct ext4_dir_entry_2
*old_de
, *new_de
;
2300 old_bh
= new_bh
= dir_bh
= NULL
;
2302 /* Initialize quotas before so that eventual writes go
2303 * in separate transaction */
2304 if (new_dentry
->d_inode
)
2305 DQUOT_INIT(new_dentry
->d_inode
);
2306 handle
= ext4_journal_start(old_dir
, 2 *
2307 EXT4_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2308 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2310 return PTR_ERR(handle
);
2312 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2315 old_bh
= ext4_find_entry(old_dentry
, &old_de
);
2317 * Check for inode number is _not_ due to possible IO errors.
2318 * We might rmdir the source, keep it as pwd of some process
2319 * and merrily kill the link to whatever was created under the
2320 * same name. Goodbye sticky bit ;-<
2322 old_inode
= old_dentry
->d_inode
;
2324 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2327 new_inode
= new_dentry
->d_inode
;
2328 new_bh
= ext4_find_entry(new_dentry
, &new_de
);
2335 if (S_ISDIR(old_inode
->i_mode
)) {
2337 retval
= -ENOTEMPTY
;
2338 if (!empty_dir(new_inode
))
2342 dir_bh
= ext4_bread(handle
, old_inode
, 0, 0, &retval
);
2345 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
)) != old_dir
->i_ino
)
2348 if (!new_inode
&& new_dir
!= old_dir
&&
2349 new_dir
->i_nlink
>= EXT4_LINK_MAX
)
2353 retval
= ext4_add_entry(handle
, new_dentry
, old_inode
);
2357 BUFFER_TRACE(new_bh
, "get write access");
2358 ext4_journal_get_write_access(handle
, new_bh
);
2359 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2360 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2361 EXT4_FEATURE_INCOMPAT_FILETYPE
))
2362 new_de
->file_type
= old_de
->file_type
;
2363 new_dir
->i_version
++;
2364 new_dir
->i_ctime
= new_dir
->i_mtime
=
2365 ext4_current_time(new_dir
);
2366 ext4_mark_inode_dirty(handle
, new_dir
);
2367 BUFFER_TRACE(new_bh
, "call ext4_journal_dirty_metadata");
2368 ext4_journal_dirty_metadata(handle
, new_bh
);
2374 * Like most other Unix systems, set the ctime for inodes on a
2377 old_inode
->i_ctime
= ext4_current_time(old_inode
);
2378 ext4_mark_inode_dirty(handle
, old_inode
);
2383 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2384 old_de
->name_len
!= old_dentry
->d_name
.len
||
2385 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2386 (retval
= ext4_delete_entry(handle
, old_dir
,
2387 old_de
, old_bh
)) == -ENOENT
) {
2388 /* old_de could have moved from under us during htree split, so
2389 * make sure that we are deleting the right entry. We might
2390 * also be pointing to a stale entry in the unused part of
2391 * old_bh so just checking inum and the name isn't enough. */
2392 struct buffer_head
*old_bh2
;
2393 struct ext4_dir_entry_2
*old_de2
;
2395 old_bh2
= ext4_find_entry(old_dentry
, &old_de2
);
2397 retval
= ext4_delete_entry(handle
, old_dir
,
2403 ext4_warning(old_dir
->i_sb
, "ext4_rename",
2404 "Deleting old file (%lu), %d, error=%d",
2405 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2409 ext4_dec_count(handle
, new_inode
);
2410 new_inode
->i_ctime
= ext4_current_time(new_inode
);
2412 old_dir
->i_ctime
= old_dir
->i_mtime
= ext4_current_time(old_dir
);
2413 ext4_update_dx_flag(old_dir
);
2415 BUFFER_TRACE(dir_bh
, "get_write_access");
2416 ext4_journal_get_write_access(handle
, dir_bh
);
2417 PARENT_INO(dir_bh
->b_data
) = cpu_to_le32(new_dir
->i_ino
);
2418 BUFFER_TRACE(dir_bh
, "call ext4_journal_dirty_metadata");
2419 ext4_journal_dirty_metadata(handle
, dir_bh
);
2420 ext4_dec_count(handle
, old_dir
);
2422 /* checked empty_dir above, can't have another parent,
2423 * ext4_dec_count() won't work for many-linked dirs */
2424 new_inode
->i_nlink
= 0;
2426 ext4_inc_count(handle
, new_dir
);
2427 ext4_update_dx_flag(new_dir
);
2428 ext4_mark_inode_dirty(handle
, new_dir
);
2431 ext4_mark_inode_dirty(handle
, old_dir
);
2433 ext4_mark_inode_dirty(handle
, new_inode
);
2434 if (!new_inode
->i_nlink
)
2435 ext4_orphan_add(handle
, new_inode
);
2443 ext4_journal_stop(handle
);
2448 * directories can handle most operations...
2450 const struct inode_operations ext4_dir_inode_operations
= {
2451 .create
= ext4_create
,
2452 .lookup
= ext4_lookup
,
2454 .unlink
= ext4_unlink
,
2455 .symlink
= ext4_symlink
,
2456 .mkdir
= ext4_mkdir
,
2457 .rmdir
= ext4_rmdir
,
2458 .mknod
= ext4_mknod
,
2459 .rename
= ext4_rename
,
2460 .setattr
= ext4_setattr
,
2461 #ifdef CONFIG_EXT4DEV_FS_XATTR
2462 .setxattr
= generic_setxattr
,
2463 .getxattr
= generic_getxattr
,
2464 .listxattr
= ext4_listxattr
,
2465 .removexattr
= generic_removexattr
,
2467 .permission
= ext4_permission
,
2470 const struct inode_operations ext4_special_inode_operations
= {
2471 .setattr
= ext4_setattr
,
2472 #ifdef CONFIG_EXT4DEV_FS_XATTR
2473 .setxattr
= generic_setxattr
,
2474 .getxattr
= generic_getxattr
,
2475 .listxattr
= ext4_listxattr
,
2476 .removexattr
= generic_removexattr
,
2478 .permission
= ext4_permission
,