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"
43 #include <trace/events/ext4.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
, unsigned blocksize
,
165 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
166 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
167 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
168 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
169 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
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
, unsigned long blocksize
)
189 return (struct ext4_dir_entry_2
*)((char *)p
+
190 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
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 (unsigned) ((char *) de
- base
));
294 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
297 de
= ext4_next_entry(de
, size
);
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
, "Unrecognised inode hash code %d",
364 root
->info
.hash_version
);
366 *err
= ERR_BAD_DX_DIR
;
369 hinfo
->hash_version
= root
->info
.hash_version
;
370 if (hinfo
->hash_version
<= DX_HASH_TEA
)
371 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
372 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
374 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
377 if (root
->info
.unused_flags
& 1) {
378 ext4_warning(dir
->i_sb
, "Unimplemented inode hash flags: %#06x",
379 root
->info
.unused_flags
);
381 *err
= ERR_BAD_DX_DIR
;
385 if ((indirect
= root
->info
.indirect_levels
) > 1) {
386 ext4_warning(dir
->i_sb
, "Unimplemented inode hash depth: %#06x",
387 root
->info
.indirect_levels
);
389 *err
= ERR_BAD_DX_DIR
;
393 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
394 root
->info
.info_length
);
396 if (dx_get_limit(entries
) != dx_root_limit(dir
,
397 root
->info
.info_length
)) {
398 ext4_warning(dir
->i_sb
, "dx entry: limit != root limit");
400 *err
= ERR_BAD_DX_DIR
;
404 dxtrace(printk("Look up %x", hash
));
407 count
= dx_get_count(entries
);
408 if (!count
|| count
> dx_get_limit(entries
)) {
409 ext4_warning(dir
->i_sb
,
410 "dx entry: no count or count > limit");
412 *err
= ERR_BAD_DX_DIR
;
417 q
= entries
+ count
- 1;
421 dxtrace(printk("."));
422 if (dx_get_hash(m
) > hash
)
428 if (0) // linear search cross check
430 unsigned n
= count
- 1;
434 dxtrace(printk(","));
435 if (dx_get_hash(++at
) > hash
)
441 assert (at
== p
- 1);
445 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
447 frame
->entries
= entries
;
449 if (!indirect
--) return frame
;
450 if (!(bh
= ext4_bread (NULL
,dir
, dx_get_block(at
), 0, err
)))
452 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
453 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
454 ext4_warning(dir
->i_sb
,
455 "dx entry: limit != node limit");
457 *err
= ERR_BAD_DX_DIR
;
464 while (frame
>= frame_in
) {
469 if (*err
== ERR_BAD_DX_DIR
)
470 ext4_warning(dir
->i_sb
,
471 "Corrupt dir inode %lu, running e2fsck is "
472 "recommended.", dir
->i_ino
);
476 static void dx_release (struct dx_frame
*frames
)
478 if (frames
[0].bh
== NULL
)
481 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
482 brelse(frames
[1].bh
);
483 brelse(frames
[0].bh
);
487 * This function increments the frame pointer to search the next leaf
488 * block, and reads in the necessary intervening nodes if the search
489 * should be necessary. Whether or not the search is necessary is
490 * controlled by the hash parameter. If the hash value is even, then
491 * the search is only continued if the next block starts with that
492 * hash value. This is used if we are searching for a specific file.
494 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
496 * This function returns 1 if the caller should continue to search,
497 * or 0 if it should not. If there is an error reading one of the
498 * index blocks, it will a negative error code.
500 * If start_hash is non-null, it will be filled in with the starting
501 * hash of the next page.
503 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
504 struct dx_frame
*frame
,
505 struct dx_frame
*frames
,
509 struct buffer_head
*bh
;
510 int err
, num_frames
= 0;
515 * Find the next leaf page by incrementing the frame pointer.
516 * If we run out of entries in the interior node, loop around and
517 * increment pointer in the parent node. When we break out of
518 * this loop, num_frames indicates the number of interior
519 * nodes need to be read.
522 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
531 * If the hash is 1, then continue only if the next page has a
532 * continuation hash of any value. This is used for readdir
533 * handling. Otherwise, check to see if the hash matches the
534 * desired contiuation hash. If it doesn't, return since
535 * there's no point to read in the successive index pages.
537 bhash
= dx_get_hash(p
->at
);
540 if ((hash
& 1) == 0) {
541 if ((bhash
& ~1) != hash
)
545 * If the hash is HASH_NB_ALWAYS, we always go to the next
546 * block so no check is necessary
548 while (num_frames
--) {
549 if (!(bh
= ext4_bread(NULL
, dir
, dx_get_block(p
->at
),
551 return err
; /* Failure */
555 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
562 * This function fills a red-black tree with information from a
563 * directory block. It returns the number directory entries loaded
564 * into the tree. If there is an error it is returned in err.
566 static int htree_dirblock_to_tree(struct file
*dir_file
,
567 struct inode
*dir
, ext4_lblk_t block
,
568 struct dx_hash_info
*hinfo
,
569 __u32 start_hash
, __u32 start_minor_hash
)
571 struct buffer_head
*bh
;
572 struct ext4_dir_entry_2
*de
, *top
;
575 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
576 (unsigned long)block
));
577 if (!(bh
= ext4_bread (NULL
, dir
, block
, 0, &err
)))
580 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
581 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
582 dir
->i_sb
->s_blocksize
-
583 EXT4_DIR_REC_LEN(0));
584 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
585 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
586 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
587 + ((char *)de
- bh
->b_data
))) {
588 /* On error, skip the f_pos to the next block. */
589 dir_file
->f_pos
= (dir_file
->f_pos
|
590 (dir
->i_sb
->s_blocksize
- 1)) + 1;
594 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
595 if ((hinfo
->hash
< start_hash
) ||
596 ((hinfo
->hash
== start_hash
) &&
597 (hinfo
->minor_hash
< start_minor_hash
)))
601 if ((err
= ext4_htree_store_dirent(dir_file
,
602 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
614 * This function fills a red-black tree with information from a
615 * directory. We start scanning the directory in hash order, starting
616 * at start_hash and start_minor_hash.
618 * This function returns the number of entries inserted into the tree,
619 * or a negative error code.
621 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
622 __u32 start_minor_hash
, __u32
*next_hash
)
624 struct dx_hash_info hinfo
;
625 struct ext4_dir_entry_2
*de
;
626 struct dx_frame frames
[2], *frame
;
633 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
634 start_hash
, start_minor_hash
));
635 dir
= dir_file
->f_path
.dentry
->d_inode
;
636 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
637 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
638 if (hinfo
.hash_version
<= DX_HASH_TEA
)
639 hinfo
.hash_version
+=
640 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
641 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
642 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
643 start_hash
, start_minor_hash
);
647 hinfo
.hash
= start_hash
;
648 hinfo
.minor_hash
= 0;
649 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
, &err
);
653 /* Add '.' and '..' from the htree header */
654 if (!start_hash
&& !start_minor_hash
) {
655 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
656 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
660 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
661 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
662 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
663 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
669 block
= dx_get_block(frame
->at
);
670 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
671 start_hash
, start_minor_hash
);
678 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
679 frame
, frames
, &hashval
);
680 *next_hash
= hashval
;
686 * Stop if: (a) there are no more entries, or
687 * (b) we have inserted at least one entry and the
688 * next hash value is not a continuation
691 (count
&& ((hashval
& 1) == 0)))
695 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
696 "next hash: %x\n", count
, *next_hash
));
705 * Directory block splitting, compacting
709 * Create map of hash values, offsets, and sizes, stored at end of block.
710 * Returns number of entries mapped.
712 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
713 struct dx_hash_info
*hinfo
,
714 struct dx_map_entry
*map_tail
)
717 char *base
= (char *) de
;
718 struct dx_hash_info h
= *hinfo
;
720 while ((char *) de
< base
+ blocksize
) {
721 if (de
->name_len
&& de
->inode
) {
722 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
724 map_tail
->hash
= h
.hash
;
725 map_tail
->offs
= ((char *) de
- base
)>>2;
726 map_tail
->size
= le16_to_cpu(de
->rec_len
);
730 /* XXX: do we need to check rec_len == 0 case? -Chris */
731 de
= ext4_next_entry(de
, blocksize
);
736 /* Sort map by hash value */
737 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
739 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
741 /* Combsort until bubble sort doesn't suck */
744 if (count
- 9 < 2) /* 9, 10 -> 11 */
746 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
747 if (p
->hash
< q
->hash
)
750 /* Garden variety bubble sort */
755 if (q
[1].hash
>= q
[0].hash
)
763 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
765 struct dx_entry
*entries
= frame
->entries
;
766 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
767 int count
= dx_get_count(entries
);
769 assert(count
< dx_get_limit(entries
));
770 assert(old
< entries
+ count
);
771 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
772 dx_set_hash(new, hash
);
773 dx_set_block(new, block
);
774 dx_set_count(entries
, count
+ 1);
777 static void ext4_update_dx_flag(struct inode
*inode
)
779 if (!EXT4_HAS_COMPAT_FEATURE(inode
->i_sb
,
780 EXT4_FEATURE_COMPAT_DIR_INDEX
))
781 ext4_clear_inode_flag(inode
, EXT4_INODE_INDEX
);
785 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
787 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
788 * `de != NULL' is guaranteed by caller.
790 static inline int ext4_match (int len
, const char * const name
,
791 struct ext4_dir_entry_2
* de
)
793 if (len
!= de
->name_len
)
797 return !memcmp(name
, de
->name
, len
);
801 * Returns 0 if not found, -1 on failure, and 1 on success
803 static inline int search_dirblock(struct buffer_head
*bh
,
805 const struct qstr
*d_name
,
807 struct ext4_dir_entry_2
** res_dir
)
809 struct ext4_dir_entry_2
* de
;
812 const char *name
= d_name
->name
;
813 int namelen
= d_name
->len
;
815 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
816 dlimit
= bh
->b_data
+ dir
->i_sb
->s_blocksize
;
817 while ((char *) de
< dlimit
) {
818 /* this code is executed quadratically often */
819 /* do minimal checking `by hand' */
821 if ((char *) de
+ namelen
<= dlimit
&&
822 ext4_match (namelen
, name
, de
)) {
823 /* found a match - just to be sure, do a full check */
824 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, offset
))
829 /* prevent looping on a bad block */
830 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
831 dir
->i_sb
->s_blocksize
);
835 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
844 * finds an entry in the specified directory with the wanted name. It
845 * returns the cache buffer in which the entry was found, and the entry
846 * itself (as a parameter - res_dir). It does NOT read the inode of the
847 * entry - you'll have to do that yourself if you want to.
849 * The returned buffer_head has ->b_count elevated. The caller is expected
850 * to brelse() it when appropriate.
852 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
853 const struct qstr
*d_name
,
854 struct ext4_dir_entry_2
** res_dir
)
856 struct super_block
*sb
;
857 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
858 struct buffer_head
*bh
, *ret
= NULL
;
859 ext4_lblk_t start
, block
, b
;
860 const u8
*name
= d_name
->name
;
861 int ra_max
= 0; /* Number of bh's in the readahead
863 int ra_ptr
= 0; /* Current index into readahead
872 namelen
= d_name
->len
;
873 if (namelen
> EXT4_NAME_LEN
)
875 if ((namelen
<= 2) && (name
[0] == '.') &&
876 (name
[1] == '.' || name
[1] == '\0')) {
878 * "." or ".." will only be in the first block
879 * NFS may look up ".."; "." should be handled by the VFS
886 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
, &err
);
888 * On success, or if the error was file not found,
889 * return. Otherwise, fall back to doing a search the
892 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
894 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
897 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
898 start
= EXT4_I(dir
)->i_dir_start_lookup
;
899 if (start
>= nblocks
)
905 * We deal with the read-ahead logic here.
907 if (ra_ptr
>= ra_max
) {
908 /* Refill the readahead buffer */
911 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
913 * Terminate if we reach the end of the
914 * directory and must wrap, or if our
915 * search has finished at this block.
917 if (b
>= nblocks
|| (num
&& block
== start
)) {
918 bh_use
[ra_max
] = NULL
;
922 bh
= ext4_getblk(NULL
, dir
, b
++, 0, &err
);
925 ll_rw_block(READ
| REQ_META
| REQ_PRIO
,
929 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
932 if (!buffer_uptodate(bh
)) {
933 /* read error, skip block & hope for the best */
934 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
935 (unsigned long) block
);
939 i
= search_dirblock(bh
, dir
, d_name
,
940 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
942 EXT4_I(dir
)->i_dir_start_lookup
= block
;
944 goto cleanup_and_exit
;
948 goto cleanup_and_exit
;
951 if (++block
>= nblocks
)
953 } while (block
!= start
);
956 * If the directory has grown while we were searching, then
957 * search the last part of the directory before giving up.
960 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
961 if (block
< nblocks
) {
967 /* Clean up the read-ahead blocks */
968 for (; ra_ptr
< ra_max
; ra_ptr
++)
969 brelse(bh_use
[ra_ptr
]);
973 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
974 struct ext4_dir_entry_2
**res_dir
, int *err
)
976 struct super_block
* sb
= dir
->i_sb
;
977 struct dx_hash_info hinfo
;
978 struct dx_frame frames
[2], *frame
;
979 struct buffer_head
*bh
;
983 if (!(frame
= dx_probe(d_name
, dir
, &hinfo
, frames
, err
)))
986 block
= dx_get_block(frame
->at
);
987 if (!(bh
= ext4_bread(NULL
, dir
, block
, 0, err
)))
990 retval
= search_dirblock(bh
, dir
, d_name
,
991 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
993 if (retval
== 1) { /* Success! */
999 *err
= ERR_BAD_DX_DIR
;
1003 /* Check to see if we should continue to search */
1004 retval
= ext4_htree_next_block(dir
, hinfo
.hash
, frame
,
1008 "error reading index page in directory #%lu",
1013 } while (retval
== 1);
1017 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1018 dx_release (frames
);
1022 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
1024 struct inode
*inode
;
1025 struct ext4_dir_entry_2
*de
;
1026 struct buffer_head
*bh
;
1028 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1029 return ERR_PTR(-ENAMETOOLONG
);
1031 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
1034 __u32 ino
= le32_to_cpu(de
->inode
);
1036 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1037 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1038 return ERR_PTR(-EIO
);
1040 inode
= ext4_iget(dir
->i_sb
, ino
);
1041 if (inode
== ERR_PTR(-ESTALE
)) {
1042 EXT4_ERROR_INODE(dir
,
1043 "deleted inode referenced: %u",
1045 return ERR_PTR(-EIO
);
1048 return d_splice_alias(inode
, dentry
);
1052 struct dentry
*ext4_get_parent(struct dentry
*child
)
1055 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1056 struct ext4_dir_entry_2
* de
;
1057 struct buffer_head
*bh
;
1059 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
);
1061 return ERR_PTR(-ENOENT
);
1062 ino
= le32_to_cpu(de
->inode
);
1065 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1066 EXT4_ERROR_INODE(child
->d_inode
,
1067 "bad parent inode number: %u", ino
);
1068 return ERR_PTR(-EIO
);
1071 return d_obtain_alias(ext4_iget(child
->d_inode
->i_sb
, ino
));
1075 static unsigned char ext4_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1076 [S_IFREG
>> S_SHIFT
] = EXT4_FT_REG_FILE
,
1077 [S_IFDIR
>> S_SHIFT
] = EXT4_FT_DIR
,
1078 [S_IFCHR
>> S_SHIFT
] = EXT4_FT_CHRDEV
,
1079 [S_IFBLK
>> S_SHIFT
] = EXT4_FT_BLKDEV
,
1080 [S_IFIFO
>> S_SHIFT
] = EXT4_FT_FIFO
,
1081 [S_IFSOCK
>> S_SHIFT
] = EXT4_FT_SOCK
,
1082 [S_IFLNK
>> S_SHIFT
] = EXT4_FT_SYMLINK
,
1085 static inline void ext4_set_de_type(struct super_block
*sb
,
1086 struct ext4_dir_entry_2
*de
,
1088 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FILETYPE
))
1089 de
->file_type
= ext4_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1093 * Move count entries from end of map between two memory locations.
1094 * Returns pointer to last entry moved.
1096 static struct ext4_dir_entry_2
*
1097 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1100 unsigned rec_len
= 0;
1103 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1104 (from
+ (map
->offs
<<2));
1105 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1106 memcpy (to
, de
, rec_len
);
1107 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1108 ext4_rec_len_to_disk(rec_len
, blocksize
);
1113 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1117 * Compact each dir entry in the range to the minimal rec_len.
1118 * Returns pointer to last entry in range.
1120 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1122 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1123 unsigned rec_len
= 0;
1126 while ((char*)de
< base
+ blocksize
) {
1127 next
= ext4_next_entry(de
, blocksize
);
1128 if (de
->inode
&& de
->name_len
) {
1129 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1131 memmove(to
, de
, rec_len
);
1132 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1134 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1142 * Split a full leaf block to make room for a new dir entry.
1143 * Allocate a new block, and move entries so that they are approx. equally full.
1144 * Returns pointer to de in block into which the new entry will be inserted.
1146 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1147 struct buffer_head
**bh
,struct dx_frame
*frame
,
1148 struct dx_hash_info
*hinfo
, int *error
)
1150 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1151 unsigned count
, continued
;
1152 struct buffer_head
*bh2
;
1153 ext4_lblk_t newblock
;
1155 struct dx_map_entry
*map
;
1156 char *data1
= (*bh
)->b_data
, *data2
;
1157 unsigned split
, move
, size
;
1158 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1161 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1168 BUFFER_TRACE(*bh
, "get_write_access");
1169 err
= ext4_journal_get_write_access(handle
, *bh
);
1173 BUFFER_TRACE(frame
->bh
, "get_write_access");
1174 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1178 data2
= bh2
->b_data
;
1180 /* create map in the end of data2 block */
1181 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1182 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1183 blocksize
, hinfo
, map
);
1185 dx_sort_map(map
, count
);
1186 /* Split the existing block in the middle, size-wise */
1189 for (i
= count
-1; i
>= 0; i
--) {
1190 /* is more than half of this entry in 2nd half of the block? */
1191 if (size
+ map
[i
].size
/2 > blocksize
/2)
1193 size
+= map
[i
].size
;
1196 /* map index at which we will split */
1197 split
= count
- move
;
1198 hash2
= map
[split
].hash
;
1199 continued
= hash2
== map
[split
- 1].hash
;
1200 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1201 (unsigned long)dx_get_block(frame
->at
),
1202 hash2
, split
, count
-split
));
1204 /* Fancy dance to stay within two buffers */
1205 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
, blocksize
);
1206 de
= dx_pack_dirents(data1
, blocksize
);
1207 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
,
1209 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ blocksize
- (char *) de2
,
1211 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1212 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1214 /* Which block gets the new entry? */
1215 if (hinfo
->hash
>= hash2
)
1220 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1221 err
= ext4_handle_dirty_metadata(handle
, dir
, bh2
);
1224 err
= ext4_handle_dirty_metadata(handle
, dir
, frame
->bh
);
1228 dxtrace(dx_show_index("frame", frame
->entries
));
1235 ext4_std_error(dir
->i_sb
, err
);
1242 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1243 * it points to a directory entry which is guaranteed to be large
1244 * enough for new directory entry. If de is NULL, then
1245 * add_dirent_to_buf will attempt search the directory block for
1246 * space. It will return -ENOSPC if no space is available, and -EIO
1247 * and -EEXIST if directory entry already exists.
1249 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1250 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1251 struct buffer_head
*bh
)
1253 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1254 const char *name
= dentry
->d_name
.name
;
1255 int namelen
= dentry
->d_name
.len
;
1256 unsigned int offset
= 0;
1257 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1258 unsigned short reclen
;
1259 int nlen
, rlen
, err
;
1262 reclen
= EXT4_DIR_REC_LEN(namelen
);
1264 de
= (struct ext4_dir_entry_2
*)bh
->b_data
;
1265 top
= bh
->b_data
+ blocksize
- reclen
;
1266 while ((char *) de
<= top
) {
1267 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, offset
))
1269 if (ext4_match(namelen
, name
, de
))
1271 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1272 rlen
= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
1273 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1275 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1278 if ((char *) de
> top
)
1281 BUFFER_TRACE(bh
, "get_write_access");
1282 err
= ext4_journal_get_write_access(handle
, bh
);
1284 ext4_std_error(dir
->i_sb
, err
);
1288 /* By now the buffer is marked for journaling */
1289 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1290 rlen
= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
1292 struct ext4_dir_entry_2
*de1
= (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1293 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, blocksize
);
1294 de
->rec_len
= ext4_rec_len_to_disk(nlen
, blocksize
);
1297 de
->file_type
= EXT4_FT_UNKNOWN
;
1299 de
->inode
= cpu_to_le32(inode
->i_ino
);
1300 ext4_set_de_type(dir
->i_sb
, de
, inode
->i_mode
);
1303 de
->name_len
= namelen
;
1304 memcpy(de
->name
, name
, namelen
);
1306 * XXX shouldn't update any times until successful
1307 * completion of syscall, but too many callers depend
1310 * XXX similarly, too many callers depend on
1311 * ext4_new_inode() setting the times, but error
1312 * recovery deletes the inode, so the worst that can
1313 * happen is that the times are slightly out of date
1314 * and/or different from the directory change time.
1316 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1317 ext4_update_dx_flag(dir
);
1319 ext4_mark_inode_dirty(handle
, dir
);
1320 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1321 err
= ext4_handle_dirty_metadata(handle
, dir
, bh
);
1323 ext4_std_error(dir
->i_sb
, err
);
1328 * This converts a one block unindexed directory to a 3 block indexed
1329 * directory, and adds the dentry to the indexed directory.
1331 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1332 struct inode
*inode
, struct buffer_head
*bh
)
1334 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1335 const char *name
= dentry
->d_name
.name
;
1336 int namelen
= dentry
->d_name
.len
;
1337 struct buffer_head
*bh2
;
1338 struct dx_root
*root
;
1339 struct dx_frame frames
[2], *frame
;
1340 struct dx_entry
*entries
;
1341 struct ext4_dir_entry_2
*de
, *de2
;
1346 struct dx_hash_info hinfo
;
1348 struct fake_dirent
*fde
;
1350 blocksize
= dir
->i_sb
->s_blocksize
;
1351 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1352 retval
= ext4_journal_get_write_access(handle
, bh
);
1354 ext4_std_error(dir
->i_sb
, retval
);
1358 root
= (struct dx_root
*) bh
->b_data
;
1360 /* The 0th block becomes the root, move the dirents out */
1361 fde
= &root
->dotdot
;
1362 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1363 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1364 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1365 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1369 len
= ((char *) root
) + blocksize
- (char *) de
;
1371 /* Allocate new block for the 0th block's dirents */
1372 bh2
= ext4_append(handle
, dir
, &block
, &retval
);
1377 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1378 data1
= bh2
->b_data
;
1380 memcpy (data1
, de
, len
);
1381 de
= (struct ext4_dir_entry_2
*) data1
;
1383 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1385 de
->rec_len
= ext4_rec_len_to_disk(data1
+ blocksize
- (char *) de
,
1387 /* Initialize the root; the dot dirents already exist */
1388 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1389 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1391 memset (&root
->info
, 0, sizeof(root
->info
));
1392 root
->info
.info_length
= sizeof(root
->info
);
1393 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1394 entries
= root
->entries
;
1395 dx_set_block(entries
, 1);
1396 dx_set_count(entries
, 1);
1397 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1399 /* Initialize as for dx_probe */
1400 hinfo
.hash_version
= root
->info
.hash_version
;
1401 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1402 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1403 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1404 ext4fs_dirhash(name
, namelen
, &hinfo
);
1406 frame
->entries
= entries
;
1407 frame
->at
= entries
;
1411 ext4_handle_dirty_metadata(handle
, dir
, frame
->bh
);
1412 ext4_handle_dirty_metadata(handle
, dir
, bh
);
1414 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1417 * Even if the block split failed, we have to properly write
1418 * out all the changes we did so far. Otherwise we can end up
1419 * with corrupted filesystem.
1421 ext4_mark_inode_dirty(handle
, dir
);
1427 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1435 * adds a file entry to the specified directory, using the same
1436 * semantics as ext4_find_entry(). It returns NULL if it failed.
1438 * NOTE!! The inode part of 'de' is left at 0 - which means you
1439 * may not sleep between calling this and putting something into
1440 * the entry, as someone else might have used it while you slept.
1442 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1443 struct inode
*inode
)
1445 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1446 struct buffer_head
*bh
;
1447 struct ext4_dir_entry_2
*de
;
1448 struct super_block
*sb
;
1452 ext4_lblk_t block
, blocks
;
1455 blocksize
= sb
->s_blocksize
;
1456 if (!dentry
->d_name
.len
)
1459 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1460 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1462 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
1464 ext4_mark_inode_dirty(handle
, dir
);
1466 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1467 for (block
= 0; block
< blocks
; block
++) {
1468 bh
= ext4_bread(handle
, dir
, block
, 0, &retval
);
1471 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1472 if (retval
!= -ENOSPC
) {
1477 if (blocks
== 1 && !dx_fallback
&&
1478 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1479 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1482 bh
= ext4_append(handle
, dir
, &block
, &retval
);
1485 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1487 de
->rec_len
= ext4_rec_len_to_disk(blocksize
, blocksize
);
1488 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1491 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
1496 * Returns 0 for success, or a negative error value
1498 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1499 struct inode
*inode
)
1501 struct dx_frame frames
[2], *frame
;
1502 struct dx_entry
*entries
, *at
;
1503 struct dx_hash_info hinfo
;
1504 struct buffer_head
*bh
;
1505 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1506 struct super_block
*sb
= dir
->i_sb
;
1507 struct ext4_dir_entry_2
*de
;
1510 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1513 entries
= frame
->entries
;
1516 if (!(bh
= ext4_bread(handle
,dir
, dx_get_block(frame
->at
), 0, &err
)))
1519 BUFFER_TRACE(bh
, "get_write_access");
1520 err
= ext4_journal_get_write_access(handle
, bh
);
1524 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1528 /* Block full, should compress but for now just split */
1529 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
1530 dx_get_count(entries
), dx_get_limit(entries
)));
1531 /* Need to split index? */
1532 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
1533 ext4_lblk_t newblock
;
1534 unsigned icount
= dx_get_count(entries
);
1535 int levels
= frame
- frames
;
1536 struct dx_entry
*entries2
;
1537 struct dx_node
*node2
;
1538 struct buffer_head
*bh2
;
1540 if (levels
&& (dx_get_count(frames
->entries
) ==
1541 dx_get_limit(frames
->entries
))) {
1542 ext4_warning(sb
, "Directory index full!");
1546 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1549 node2
= (struct dx_node
*)(bh2
->b_data
);
1550 entries2
= node2
->entries
;
1551 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
1552 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
1554 BUFFER_TRACE(frame
->bh
, "get_write_access");
1555 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1559 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
1560 unsigned hash2
= dx_get_hash(entries
+ icount1
);
1561 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
1564 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
1565 err
= ext4_journal_get_write_access(handle
,
1570 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
1571 icount2
* sizeof(struct dx_entry
));
1572 dx_set_count(entries
, icount1
);
1573 dx_set_count(entries2
, icount2
);
1574 dx_set_limit(entries2
, dx_node_limit(dir
));
1576 /* Which index block gets the new entry? */
1577 if (at
- entries
>= icount1
) {
1578 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
1579 frame
->entries
= entries
= entries2
;
1580 swap(frame
->bh
, bh2
);
1582 dx_insert_block(frames
+ 0, hash2
, newblock
);
1583 dxtrace(dx_show_index("node", frames
[1].entries
));
1584 dxtrace(dx_show_index("node",
1585 ((struct dx_node
*) bh2
->b_data
)->entries
));
1586 err
= ext4_handle_dirty_metadata(handle
, dir
, bh2
);
1591 dxtrace(printk(KERN_DEBUG
1592 "Creating second level index...\n"));
1593 memcpy((char *) entries2
, (char *) entries
,
1594 icount
* sizeof(struct dx_entry
));
1595 dx_set_limit(entries2
, dx_node_limit(dir
));
1598 dx_set_count(entries
, 1);
1599 dx_set_block(entries
+ 0, newblock
);
1600 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
1602 /* Add new access path frame */
1604 frame
->at
= at
= at
- entries
+ entries2
;
1605 frame
->entries
= entries
= entries2
;
1607 err
= ext4_journal_get_write_access(handle
,
1612 err
= ext4_handle_dirty_metadata(handle
, dir
, frames
[0].bh
);
1614 ext4_std_error(inode
->i_sb
, err
);
1618 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
1621 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
;
1643 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1648 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1649 while (i
< bh
->b_size
) {
1650 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, i
))
1653 BUFFER_TRACE(bh
, "get_write_access");
1654 err
= ext4_journal_get_write_access(handle
, bh
);
1655 if (unlikely(err
)) {
1656 ext4_std_error(dir
->i_sb
, err
);
1660 pde
->rec_len
= ext4_rec_len_to_disk(
1661 ext4_rec_len_from_disk(pde
->rec_len
,
1663 ext4_rec_len_from_disk(de
->rec_len
,
1669 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1670 err
= ext4_handle_dirty_metadata(handle
, dir
, bh
);
1671 if (unlikely(err
)) {
1672 ext4_std_error(dir
->i_sb
, err
);
1677 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
1679 de
= ext4_next_entry(de
, blocksize
);
1685 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1686 * since this indicates that nlinks count was previously 1.
1688 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
1691 if (is_dx(inode
) && inode
->i_nlink
> 1) {
1692 /* limit is 16-bit i_links_count */
1693 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
1694 set_nlink(inode
, 1);
1695 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
1696 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
1702 * If a directory had nlink == 1, then we should let it be 1. This indicates
1703 * directory has >EXT4_LINK_MAX subdirs.
1705 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
1707 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
1712 static int ext4_add_nondir(handle_t
*handle
,
1713 struct dentry
*dentry
, struct inode
*inode
)
1715 int err
= ext4_add_entry(handle
, dentry
, inode
);
1717 ext4_mark_inode_dirty(handle
, inode
);
1718 d_instantiate(dentry
, inode
);
1719 unlock_new_inode(inode
);
1723 unlock_new_inode(inode
);
1729 * By the time this is called, we already have created
1730 * the directory cache entry for the new file, but it
1731 * is so far negative - it has no inode.
1733 * If the create succeeds, we fill in the inode information
1734 * with d_instantiate().
1736 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
1737 struct nameidata
*nd
)
1740 struct inode
*inode
;
1741 int err
, retries
= 0;
1743 dquot_initialize(dir
);
1746 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1747 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1748 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1750 return PTR_ERR(handle
);
1752 if (IS_DIRSYNC(dir
))
1753 ext4_handle_sync(handle
);
1755 inode
= ext4_new_inode(handle
, dir
, mode
, &dentry
->d_name
, 0, NULL
);
1756 err
= PTR_ERR(inode
);
1757 if (!IS_ERR(inode
)) {
1758 inode
->i_op
= &ext4_file_inode_operations
;
1759 inode
->i_fop
= &ext4_file_operations
;
1760 ext4_set_aops(inode
);
1761 err
= ext4_add_nondir(handle
, dentry
, inode
);
1763 ext4_journal_stop(handle
);
1764 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1769 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
1770 umode_t mode
, dev_t rdev
)
1773 struct inode
*inode
;
1774 int err
, retries
= 0;
1776 if (!new_valid_dev(rdev
))
1779 dquot_initialize(dir
);
1782 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1783 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1784 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1786 return PTR_ERR(handle
);
1788 if (IS_DIRSYNC(dir
))
1789 ext4_handle_sync(handle
);
1791 inode
= ext4_new_inode(handle
, dir
, mode
, &dentry
->d_name
, 0, NULL
);
1792 err
= PTR_ERR(inode
);
1793 if (!IS_ERR(inode
)) {
1794 init_special_inode(inode
, inode
->i_mode
, rdev
);
1795 #ifdef CONFIG_EXT4_FS_XATTR
1796 inode
->i_op
= &ext4_special_inode_operations
;
1798 err
= ext4_add_nondir(handle
, dentry
, inode
);
1800 ext4_journal_stop(handle
);
1801 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1806 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1809 struct inode
*inode
;
1810 struct buffer_head
*dir_block
= NULL
;
1811 struct ext4_dir_entry_2
*de
;
1812 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1813 int err
, retries
= 0;
1815 if (EXT4_DIR_LINK_MAX(dir
))
1818 dquot_initialize(dir
);
1821 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
1822 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
1823 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
1825 return PTR_ERR(handle
);
1827 if (IS_DIRSYNC(dir
))
1828 ext4_handle_sync(handle
);
1830 inode
= ext4_new_inode(handle
, dir
, S_IFDIR
| mode
,
1831 &dentry
->d_name
, 0, NULL
);
1832 err
= PTR_ERR(inode
);
1836 inode
->i_op
= &ext4_dir_inode_operations
;
1837 inode
->i_fop
= &ext4_dir_operations
;
1838 inode
->i_size
= EXT4_I(inode
)->i_disksize
= inode
->i_sb
->s_blocksize
;
1839 dir_block
= ext4_bread(handle
, inode
, 0, 1, &err
);
1841 goto out_clear_inode
;
1842 BUFFER_TRACE(dir_block
, "get_write_access");
1843 err
= ext4_journal_get_write_access(handle
, dir_block
);
1845 goto out_clear_inode
;
1846 de
= (struct ext4_dir_entry_2
*) dir_block
->b_data
;
1847 de
->inode
= cpu_to_le32(inode
->i_ino
);
1849 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
1851 strcpy(de
->name
, ".");
1852 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1853 de
= ext4_next_entry(de
, blocksize
);
1854 de
->inode
= cpu_to_le32(dir
->i_ino
);
1855 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(1),
1858 strcpy(de
->name
, "..");
1859 ext4_set_de_type(dir
->i_sb
, de
, S_IFDIR
);
1860 set_nlink(inode
, 2);
1861 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
1862 err
= ext4_handle_dirty_metadata(handle
, inode
, dir_block
);
1864 goto out_clear_inode
;
1865 err
= ext4_mark_inode_dirty(handle
, inode
);
1867 err
= ext4_add_entry(handle
, dentry
, inode
);
1871 unlock_new_inode(inode
);
1872 ext4_mark_inode_dirty(handle
, inode
);
1876 ext4_inc_count(handle
, dir
);
1877 ext4_update_dx_flag(dir
);
1878 err
= ext4_mark_inode_dirty(handle
, dir
);
1880 goto out_clear_inode
;
1881 d_instantiate(dentry
, inode
);
1882 unlock_new_inode(inode
);
1885 ext4_journal_stop(handle
);
1886 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
1892 * routine to check that the specified directory is empty (for rmdir)
1894 static int empty_dir(struct inode
*inode
)
1896 unsigned int offset
;
1897 struct buffer_head
*bh
;
1898 struct ext4_dir_entry_2
*de
, *de1
;
1899 struct super_block
*sb
;
1903 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1904 !(bh
= ext4_bread(NULL
, inode
, 0, 0, &err
))) {
1906 EXT4_ERROR_INODE(inode
,
1907 "error %d reading directory lblock 0", err
);
1909 ext4_warning(inode
->i_sb
,
1910 "bad directory (dir #%lu) - no data block",
1914 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1915 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
1916 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
1917 !le32_to_cpu(de1
->inode
) ||
1918 strcmp(".", de
->name
) ||
1919 strcmp("..", de1
->name
)) {
1920 ext4_warning(inode
->i_sb
,
1921 "bad directory (dir #%lu) - no `.' or `..'",
1926 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
1927 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
1928 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
1929 while (offset
< inode
->i_size
) {
1931 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
1932 unsigned int lblock
;
1935 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
1936 bh
= ext4_bread(NULL
, inode
, lblock
, 0, &err
);
1939 EXT4_ERROR_INODE(inode
,
1940 "error %d reading directory "
1941 "lblock %u", err
, lblock
);
1942 offset
+= sb
->s_blocksize
;
1945 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1947 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
, offset
)) {
1948 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
1950 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
1953 if (le32_to_cpu(de
->inode
)) {
1957 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
1958 de
= ext4_next_entry(de
, sb
->s_blocksize
);
1964 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1965 * such inodes, starting at the superblock, in case we crash before the
1966 * file is closed/deleted, or in case the inode truncate spans multiple
1967 * transactions and the last transaction is not recovered after a crash.
1969 * At filesystem recovery time, we walk this list deleting unlinked
1970 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1972 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
1974 struct super_block
*sb
= inode
->i_sb
;
1975 struct ext4_iloc iloc
;
1978 if (!ext4_handle_valid(handle
))
1981 mutex_lock(&EXT4_SB(sb
)->s_orphan_lock
);
1982 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
1986 * Orphan handling is only valid for files with data blocks
1987 * being truncated, or files being unlinked. Note that we either
1988 * hold i_mutex, or the inode can not be referenced from outside,
1989 * so i_nlink should not be bumped due to race
1991 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1992 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
1994 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1995 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1999 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2003 * Due to previous errors inode may be already a part of on-disk
2004 * orphan list. If so skip on-disk list modification.
2006 if (NEXT_ORPHAN(inode
) && NEXT_ORPHAN(inode
) <=
2007 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
)))
2010 /* Insert this inode at the head of the on-disk orphan list... */
2011 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT4_SB(sb
)->s_es
->s_last_orphan
);
2012 EXT4_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2013 err
= ext4_handle_dirty_metadata(handle
, NULL
, EXT4_SB(sb
)->s_sbh
);
2014 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2018 /* Only add to the head of the in-memory list if all the
2019 * previous operations succeeded. If the orphan_add is going to
2020 * fail (possibly taking the journal offline), we can't risk
2021 * leaving the inode on the orphan list: stray orphan-list
2022 * entries can cause panics at unmount time.
2024 * This is safe: on error we're going to ignore the orphan list
2025 * anyway on the next recovery. */
2028 list_add(&EXT4_I(inode
)->i_orphan
, &EXT4_SB(sb
)->s_orphan
);
2030 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2031 jbd_debug(4, "orphan inode %lu will point to %d\n",
2032 inode
->i_ino
, NEXT_ORPHAN(inode
));
2034 mutex_unlock(&EXT4_SB(sb
)->s_orphan_lock
);
2035 ext4_std_error(inode
->i_sb
, err
);
2040 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2041 * of such inodes stored on disk, because it is finally being cleaned up.
2043 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2045 struct list_head
*prev
;
2046 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2047 struct ext4_sb_info
*sbi
;
2049 struct ext4_iloc iloc
;
2052 /* ext4_handle_valid() assumes a valid handle_t pointer */
2053 if (handle
&& !ext4_handle_valid(handle
))
2056 mutex_lock(&EXT4_SB(inode
->i_sb
)->s_orphan_lock
);
2057 if (list_empty(&ei
->i_orphan
))
2060 ino_next
= NEXT_ORPHAN(inode
);
2061 prev
= ei
->i_orphan
.prev
;
2062 sbi
= EXT4_SB(inode
->i_sb
);
2064 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2066 list_del_init(&ei
->i_orphan
);
2068 /* If we're on an error path, we may not have a valid
2069 * transaction handle with which to update the orphan list on
2070 * disk, but we still need to remove the inode from the linked
2071 * list in memory. */
2072 if (sbi
->s_journal
&& !handle
)
2075 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2079 if (prev
== &sbi
->s_orphan
) {
2080 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2081 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2082 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2085 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2086 err
= ext4_handle_dirty_metadata(handle
, NULL
, sbi
->s_sbh
);
2088 struct ext4_iloc iloc2
;
2089 struct inode
*i_prev
=
2090 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2092 jbd_debug(4, "orphan inode %lu will point to %u\n",
2093 i_prev
->i_ino
, ino_next
);
2094 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2097 NEXT_ORPHAN(i_prev
) = ino_next
;
2098 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2102 NEXT_ORPHAN(inode
) = 0;
2103 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2106 ext4_std_error(inode
->i_sb
, err
);
2108 mutex_unlock(&EXT4_SB(inode
->i_sb
)->s_orphan_lock
);
2116 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2119 struct inode
*inode
;
2120 struct buffer_head
*bh
;
2121 struct ext4_dir_entry_2
*de
;
2124 /* Initialize quotas before so that eventual writes go in
2125 * separate transaction */
2126 dquot_initialize(dir
);
2127 dquot_initialize(dentry
->d_inode
);
2129 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2131 return PTR_ERR(handle
);
2134 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2138 if (IS_DIRSYNC(dir
))
2139 ext4_handle_sync(handle
);
2141 inode
= dentry
->d_inode
;
2144 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2147 retval
= -ENOTEMPTY
;
2148 if (!empty_dir(inode
))
2151 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2154 if (!EXT4_DIR_LINK_EMPTY(inode
))
2155 ext4_warning(inode
->i_sb
,
2156 "empty directory has too many links (%d)",
2160 /* There's no need to set i_disksize: the fact that i_nlink is
2161 * zero will ensure that the right thing happens during any
2164 ext4_orphan_add(handle
, inode
);
2165 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2166 ext4_mark_inode_dirty(handle
, inode
);
2167 ext4_dec_count(handle
, dir
);
2168 ext4_update_dx_flag(dir
);
2169 ext4_mark_inode_dirty(handle
, dir
);
2172 ext4_journal_stop(handle
);
2177 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2180 struct inode
*inode
;
2181 struct buffer_head
*bh
;
2182 struct ext4_dir_entry_2
*de
;
2185 trace_ext4_unlink_enter(dir
, dentry
);
2186 /* Initialize quotas before so that eventual writes go
2187 * in separate transaction */
2188 dquot_initialize(dir
);
2189 dquot_initialize(dentry
->d_inode
);
2191 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2193 return PTR_ERR(handle
);
2195 if (IS_DIRSYNC(dir
))
2196 ext4_handle_sync(handle
);
2199 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2203 inode
= dentry
->d_inode
;
2206 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2209 if (!inode
->i_nlink
) {
2210 ext4_warning(inode
->i_sb
,
2211 "Deleting nonexistent file (%lu), %d",
2212 inode
->i_ino
, inode
->i_nlink
);
2213 set_nlink(inode
, 1);
2215 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2218 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2219 ext4_update_dx_flag(dir
);
2220 ext4_mark_inode_dirty(handle
, dir
);
2222 if (!inode
->i_nlink
)
2223 ext4_orphan_add(handle
, inode
);
2224 inode
->i_ctime
= ext4_current_time(inode
);
2225 ext4_mark_inode_dirty(handle
, inode
);
2229 ext4_journal_stop(handle
);
2231 trace_ext4_unlink_exit(dentry
, retval
);
2235 static int ext4_symlink(struct inode
*dir
,
2236 struct dentry
*dentry
, const char *symname
)
2239 struct inode
*inode
;
2240 int l
, err
, retries
= 0;
2243 l
= strlen(symname
)+1;
2244 if (l
> dir
->i_sb
->s_blocksize
)
2245 return -ENAMETOOLONG
;
2247 dquot_initialize(dir
);
2249 if (l
> EXT4_N_BLOCKS
* 4) {
2251 * For non-fast symlinks, we just allocate inode and put it on
2252 * orphan list in the first transaction => we need bitmap,
2253 * group descriptor, sb, inode block, quota blocks, and
2254 * possibly selinux xattr blocks.
2256 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2257 EXT4_XATTR_TRANS_BLOCKS
;
2260 * Fast symlink. We have to add entry to directory
2261 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2262 * allocate new inode (bitmap, group descriptor, inode block,
2263 * quota blocks, sb is already counted in previous macros).
2265 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2266 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
2267 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
);
2270 handle
= ext4_journal_start(dir
, credits
);
2272 return PTR_ERR(handle
);
2274 if (IS_DIRSYNC(dir
))
2275 ext4_handle_sync(handle
);
2277 inode
= ext4_new_inode(handle
, dir
, S_IFLNK
|S_IRWXUGO
,
2278 &dentry
->d_name
, 0, NULL
);
2279 err
= PTR_ERR(inode
);
2283 if (l
> EXT4_N_BLOCKS
* 4) {
2284 inode
->i_op
= &ext4_symlink_inode_operations
;
2285 ext4_set_aops(inode
);
2287 * We cannot call page_symlink() with transaction started
2288 * because it calls into ext4_write_begin() which can wait
2289 * for transaction commit if we are running out of space
2290 * and thus we deadlock. So we have to stop transaction now
2291 * and restart it when symlink contents is written.
2293 * To keep fs consistent in case of crash, we have to put inode
2294 * to orphan list in the mean time.
2297 err
= ext4_orphan_add(handle
, inode
);
2298 ext4_journal_stop(handle
);
2300 goto err_drop_inode
;
2301 err
= __page_symlink(inode
, symname
, l
, 1);
2303 goto err_drop_inode
;
2305 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2306 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2308 handle
= ext4_journal_start(dir
,
2309 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2310 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
2311 if (IS_ERR(handle
)) {
2312 err
= PTR_ERR(handle
);
2313 goto err_drop_inode
;
2315 set_nlink(inode
, 1);
2316 err
= ext4_orphan_del(handle
, inode
);
2318 ext4_journal_stop(handle
);
2320 goto err_drop_inode
;
2323 /* clear the extent format for fast symlink */
2324 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
2325 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2326 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2327 inode
->i_size
= l
-1;
2329 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2330 err
= ext4_add_nondir(handle
, dentry
, inode
);
2332 ext4_journal_stop(handle
);
2333 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2337 unlock_new_inode(inode
);
2342 static int ext4_link(struct dentry
*old_dentry
,
2343 struct inode
*dir
, struct dentry
*dentry
)
2346 struct inode
*inode
= old_dentry
->d_inode
;
2347 int err
, retries
= 0;
2349 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
2352 dquot_initialize(dir
);
2355 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2356 EXT4_INDEX_EXTRA_TRANS_BLOCKS
);
2358 return PTR_ERR(handle
);
2360 if (IS_DIRSYNC(dir
))
2361 ext4_handle_sync(handle
);
2363 inode
->i_ctime
= ext4_current_time(inode
);
2364 ext4_inc_count(handle
, inode
);
2367 err
= ext4_add_entry(handle
, dentry
, inode
);
2369 ext4_mark_inode_dirty(handle
, inode
);
2370 d_instantiate(dentry
, inode
);
2375 ext4_journal_stop(handle
);
2376 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2381 #define PARENT_INO(buffer, size) \
2382 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2385 * Anybody can rename anything with this: the permission checks are left to the
2386 * higher-level routines.
2388 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2389 struct inode
*new_dir
, struct dentry
*new_dentry
)
2392 struct inode
*old_inode
, *new_inode
;
2393 struct buffer_head
*old_bh
, *new_bh
, *dir_bh
;
2394 struct ext4_dir_entry_2
*old_de
, *new_de
;
2395 int retval
, force_da_alloc
= 0;
2397 dquot_initialize(old_dir
);
2398 dquot_initialize(new_dir
);
2400 old_bh
= new_bh
= dir_bh
= NULL
;
2402 /* Initialize quotas before so that eventual writes go
2403 * in separate transaction */
2404 if (new_dentry
->d_inode
)
2405 dquot_initialize(new_dentry
->d_inode
);
2406 handle
= ext4_journal_start(old_dir
, 2 *
2407 EXT4_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2408 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2410 return PTR_ERR(handle
);
2412 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2413 ext4_handle_sync(handle
);
2415 old_bh
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2417 * Check for inode number is _not_ due to possible IO errors.
2418 * We might rmdir the source, keep it as pwd of some process
2419 * and merrily kill the link to whatever was created under the
2420 * same name. Goodbye sticky bit ;-<
2422 old_inode
= old_dentry
->d_inode
;
2424 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
2427 new_inode
= new_dentry
->d_inode
;
2428 new_bh
= ext4_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
2435 if (S_ISDIR(old_inode
->i_mode
)) {
2437 retval
= -ENOTEMPTY
;
2438 if (!empty_dir(new_inode
))
2442 dir_bh
= ext4_bread(handle
, old_inode
, 0, 0, &retval
);
2445 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
,
2446 old_dir
->i_sb
->s_blocksize
)) != old_dir
->i_ino
)
2449 if (!new_inode
&& new_dir
!= old_dir
&&
2450 EXT4_DIR_LINK_MAX(new_dir
))
2452 BUFFER_TRACE(dir_bh
, "get_write_access");
2453 retval
= ext4_journal_get_write_access(handle
, dir_bh
);
2458 retval
= ext4_add_entry(handle
, new_dentry
, old_inode
);
2462 BUFFER_TRACE(new_bh
, "get write access");
2463 retval
= ext4_journal_get_write_access(handle
, new_bh
);
2466 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
2467 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
2468 EXT4_FEATURE_INCOMPAT_FILETYPE
))
2469 new_de
->file_type
= old_de
->file_type
;
2470 new_dir
->i_version
++;
2471 new_dir
->i_ctime
= new_dir
->i_mtime
=
2472 ext4_current_time(new_dir
);
2473 ext4_mark_inode_dirty(handle
, new_dir
);
2474 BUFFER_TRACE(new_bh
, "call ext4_handle_dirty_metadata");
2475 retval
= ext4_handle_dirty_metadata(handle
, new_dir
, new_bh
);
2476 if (unlikely(retval
)) {
2477 ext4_std_error(new_dir
->i_sb
, retval
);
2485 * Like most other Unix systems, set the ctime for inodes on a
2488 old_inode
->i_ctime
= ext4_current_time(old_inode
);
2489 ext4_mark_inode_dirty(handle
, old_inode
);
2494 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
2495 old_de
->name_len
!= old_dentry
->d_name
.len
||
2496 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
2497 (retval
= ext4_delete_entry(handle
, old_dir
,
2498 old_de
, old_bh
)) == -ENOENT
) {
2499 /* old_de could have moved from under us during htree split, so
2500 * make sure that we are deleting the right entry. We might
2501 * also be pointing to a stale entry in the unused part of
2502 * old_bh so just checking inum and the name isn't enough. */
2503 struct buffer_head
*old_bh2
;
2504 struct ext4_dir_entry_2
*old_de2
;
2506 old_bh2
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de2
);
2508 retval
= ext4_delete_entry(handle
, old_dir
,
2514 ext4_warning(old_dir
->i_sb
,
2515 "Deleting old file (%lu), %d, error=%d",
2516 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
2520 ext4_dec_count(handle
, new_inode
);
2521 new_inode
->i_ctime
= ext4_current_time(new_inode
);
2523 old_dir
->i_ctime
= old_dir
->i_mtime
= ext4_current_time(old_dir
);
2524 ext4_update_dx_flag(old_dir
);
2526 PARENT_INO(dir_bh
->b_data
, new_dir
->i_sb
->s_blocksize
) =
2527 cpu_to_le32(new_dir
->i_ino
);
2528 BUFFER_TRACE(dir_bh
, "call ext4_handle_dirty_metadata");
2529 retval
= ext4_handle_dirty_metadata(handle
, old_inode
, dir_bh
);
2531 ext4_std_error(old_dir
->i_sb
, retval
);
2534 ext4_dec_count(handle
, old_dir
);
2536 /* checked empty_dir above, can't have another parent,
2537 * ext4_dec_count() won't work for many-linked dirs */
2538 clear_nlink(new_inode
);
2540 ext4_inc_count(handle
, new_dir
);
2541 ext4_update_dx_flag(new_dir
);
2542 ext4_mark_inode_dirty(handle
, new_dir
);
2545 ext4_mark_inode_dirty(handle
, old_dir
);
2547 ext4_mark_inode_dirty(handle
, new_inode
);
2548 if (!new_inode
->i_nlink
)
2549 ext4_orphan_add(handle
, new_inode
);
2550 if (!test_opt(new_dir
->i_sb
, NO_AUTO_DA_ALLOC
))
2559 ext4_journal_stop(handle
);
2560 if (retval
== 0 && force_da_alloc
)
2561 ext4_alloc_da_blocks(old_inode
);
2566 * directories can handle most operations...
2568 const struct inode_operations ext4_dir_inode_operations
= {
2569 .create
= ext4_create
,
2570 .lookup
= ext4_lookup
,
2572 .unlink
= ext4_unlink
,
2573 .symlink
= ext4_symlink
,
2574 .mkdir
= ext4_mkdir
,
2575 .rmdir
= ext4_rmdir
,
2576 .mknod
= ext4_mknod
,
2577 .rename
= ext4_rename
,
2578 .setattr
= ext4_setattr
,
2579 #ifdef CONFIG_EXT4_FS_XATTR
2580 .setxattr
= generic_setxattr
,
2581 .getxattr
= generic_getxattr
,
2582 .listxattr
= ext4_listxattr
,
2583 .removexattr
= generic_removexattr
,
2585 .get_acl
= ext4_get_acl
,
2586 .fiemap
= ext4_fiemap
,
2589 const struct inode_operations ext4_special_inode_operations
= {
2590 .setattr
= ext4_setattr
,
2591 #ifdef CONFIG_EXT4_FS_XATTR
2592 .setxattr
= generic_setxattr
,
2593 .getxattr
= generic_getxattr
,
2594 .listxattr
= ext4_listxattr
,
2595 .removexattr
= generic_removexattr
,
2597 .get_acl
= ext4_get_acl
,