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 if (unlikely(EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
&&
59 ((inode
->i_size
>> 10) >=
60 EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
))) {
65 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
67 bh
= ext4_bread(handle
, inode
, *block
, 1, err
);
69 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
70 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
71 *err
= ext4_journal_get_write_access(handle
, bh
);
79 ext4_error(inode
->i_sb
,
80 "Directory hole detected on inode %lu\n",
87 #define assert(test) J_ASSERT(test)
91 #define dxtrace(command) command
93 #define dxtrace(command)
117 * dx_root_info is laid out so that if it should somehow get overlaid by a
118 * dirent the two low bits of the hash version will be zero. Therefore, the
119 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
124 struct fake_dirent dot
;
126 struct fake_dirent dotdot
;
130 __le32 reserved_zero
;
132 u8 info_length
; /* 8 */
137 struct dx_entry entries
[0];
142 struct fake_dirent fake
;
143 struct dx_entry entries
[0];
149 struct buffer_head
*bh
;
150 struct dx_entry
*entries
;
162 * This goes at the end of each htree block.
166 __le32 dt_checksum
; /* crc32c(uuid+inum+dirblock) */
169 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
170 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
171 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
172 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
173 static unsigned dx_get_count(struct dx_entry
*entries
);
174 static unsigned dx_get_limit(struct dx_entry
*entries
);
175 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
176 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
177 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
178 static unsigned dx_node_limit(struct inode
*dir
);
179 static struct dx_frame
*dx_probe(const struct qstr
*d_name
,
181 struct dx_hash_info
*hinfo
,
182 struct dx_frame
*frame
,
184 static void dx_release(struct dx_frame
*frames
);
185 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
186 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
187 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
188 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
189 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
190 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
191 static void dx_insert_block(struct dx_frame
*frame
,
192 u32 hash
, ext4_lblk_t block
);
193 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
194 struct dx_frame
*frame
,
195 struct dx_frame
*frames
,
197 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
198 const struct qstr
*d_name
,
199 struct ext4_dir_entry_2
**res_dir
,
201 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
202 struct inode
*inode
);
204 /* checksumming functions */
205 void initialize_dirent_tail(struct ext4_dir_entry_tail
*t
,
206 unsigned int blocksize
)
208 memset(t
, 0, sizeof(struct ext4_dir_entry_tail
));
209 t
->det_rec_len
= ext4_rec_len_to_disk(
210 sizeof(struct ext4_dir_entry_tail
), blocksize
);
211 t
->det_reserved_ft
= EXT4_FT_DIR_CSUM
;
214 /* Walk through a dirent block to find a checksum "dirent" at the tail */
215 static struct ext4_dir_entry_tail
*get_dirent_tail(struct inode
*inode
,
216 struct ext4_dir_entry
*de
)
218 struct ext4_dir_entry_tail
*t
;
221 struct ext4_dir_entry
*d
, *top
;
224 top
= (struct ext4_dir_entry
*)(((void *)de
) +
225 (EXT4_BLOCK_SIZE(inode
->i_sb
) -
226 sizeof(struct ext4_dir_entry_tail
)));
227 while (d
< top
&& d
->rec_len
)
228 d
= (struct ext4_dir_entry
*)(((void *)d
) +
229 le16_to_cpu(d
->rec_len
));
234 t
= (struct ext4_dir_entry_tail
*)d
;
236 t
= EXT4_DIRENT_TAIL(de
, EXT4_BLOCK_SIZE(inode
->i_sb
));
239 if (t
->det_reserved_zero1
||
240 le16_to_cpu(t
->det_rec_len
) != sizeof(struct ext4_dir_entry_tail
) ||
241 t
->det_reserved_zero2
||
242 t
->det_reserved_ft
!= EXT4_FT_DIR_CSUM
)
248 static __le32
ext4_dirent_csum(struct inode
*inode
,
249 struct ext4_dir_entry
*dirent
, int size
)
251 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
252 struct ext4_inode_info
*ei
= EXT4_I(inode
);
255 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
256 return cpu_to_le32(csum
);
259 static void warn_no_space_for_csum(struct inode
*inode
)
261 ext4_warning(inode
->i_sb
, "no space in directory inode %lu leaf for "
262 "checksum. Please run e2fsck -D.", inode
->i_ino
);
265 int ext4_dirent_csum_verify(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
267 struct ext4_dir_entry_tail
*t
;
269 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
270 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
273 t
= get_dirent_tail(inode
, dirent
);
275 warn_no_space_for_csum(inode
);
279 if (t
->det_checksum
!= ext4_dirent_csum(inode
, dirent
,
280 (void *)t
- (void *)dirent
))
286 static void ext4_dirent_csum_set(struct inode
*inode
,
287 struct ext4_dir_entry
*dirent
)
289 struct ext4_dir_entry_tail
*t
;
291 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
292 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
295 t
= get_dirent_tail(inode
, dirent
);
297 warn_no_space_for_csum(inode
);
301 t
->det_checksum
= ext4_dirent_csum(inode
, dirent
,
302 (void *)t
- (void *)dirent
);
305 int ext4_handle_dirty_dirent_node(handle_t
*handle
,
307 struct buffer_head
*bh
)
309 ext4_dirent_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
310 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
313 static struct dx_countlimit
*get_dx_countlimit(struct inode
*inode
,
314 struct ext4_dir_entry
*dirent
,
317 struct ext4_dir_entry
*dp
;
318 struct dx_root_info
*root
;
321 if (le16_to_cpu(dirent
->rec_len
) == EXT4_BLOCK_SIZE(inode
->i_sb
))
323 else if (le16_to_cpu(dirent
->rec_len
) == 12) {
324 dp
= (struct ext4_dir_entry
*)(((void *)dirent
) + 12);
325 if (le16_to_cpu(dp
->rec_len
) !=
326 EXT4_BLOCK_SIZE(inode
->i_sb
) - 12)
328 root
= (struct dx_root_info
*)(((void *)dp
+ 12));
329 if (root
->reserved_zero
||
330 root
->info_length
!= sizeof(struct dx_root_info
))
337 *offset
= count_offset
;
338 return (struct dx_countlimit
*)(((void *)dirent
) + count_offset
);
341 static __le32
ext4_dx_csum(struct inode
*inode
, struct ext4_dir_entry
*dirent
,
342 int count_offset
, int count
, struct dx_tail
*t
)
344 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
345 struct ext4_inode_info
*ei
= EXT4_I(inode
);
346 __u32 csum
, old_csum
;
349 size
= count_offset
+ (count
* sizeof(struct dx_entry
));
350 old_csum
= t
->dt_checksum
;
352 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
353 csum
= ext4_chksum(sbi
, csum
, (__u8
*)t
, sizeof(struct dx_tail
));
354 t
->dt_checksum
= old_csum
;
356 return cpu_to_le32(csum
);
359 static int ext4_dx_csum_verify(struct inode
*inode
,
360 struct ext4_dir_entry
*dirent
)
362 struct dx_countlimit
*c
;
364 int count_offset
, limit
, count
;
366 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
367 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
370 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
372 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
375 limit
= le16_to_cpu(c
->limit
);
376 count
= le16_to_cpu(c
->count
);
377 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
378 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
379 warn_no_space_for_csum(inode
);
382 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
384 if (t
->dt_checksum
!= ext4_dx_csum(inode
, dirent
, count_offset
,
390 static void ext4_dx_csum_set(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
392 struct dx_countlimit
*c
;
394 int count_offset
, limit
, count
;
396 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
397 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
400 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
402 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
405 limit
= le16_to_cpu(c
->limit
);
406 count
= le16_to_cpu(c
->count
);
407 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
408 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
409 warn_no_space_for_csum(inode
);
412 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
414 t
->dt_checksum
= ext4_dx_csum(inode
, dirent
, count_offset
, count
, t
);
417 static inline int ext4_handle_dirty_dx_node(handle_t
*handle
,
419 struct buffer_head
*bh
)
421 ext4_dx_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
422 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
426 * p is at least 6 bytes before the end of page
428 static inline struct ext4_dir_entry_2
*
429 ext4_next_entry(struct ext4_dir_entry_2
*p
, unsigned long blocksize
)
431 return (struct ext4_dir_entry_2
*)((char *)p
+
432 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
436 * Future: use high four bits of block for coalesce-on-delete flags
437 * Mask them off for now.
440 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
442 return le32_to_cpu(entry
->block
) & 0x00ffffff;
445 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
447 entry
->block
= cpu_to_le32(value
);
450 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
452 return le32_to_cpu(entry
->hash
);
455 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
457 entry
->hash
= cpu_to_le32(value
);
460 static inline unsigned dx_get_count(struct dx_entry
*entries
)
462 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
465 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
467 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
470 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
472 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
475 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
477 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
480 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
482 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
483 EXT4_DIR_REC_LEN(2) - infosize
;
485 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
486 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
487 entry_space
-= sizeof(struct dx_tail
);
488 return entry_space
/ sizeof(struct dx_entry
);
491 static inline unsigned dx_node_limit(struct inode
*dir
)
493 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
495 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
496 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
497 entry_space
-= sizeof(struct dx_tail
);
498 return entry_space
/ sizeof(struct dx_entry
);
505 static void dx_show_index(char * label
, struct dx_entry
*entries
)
507 int i
, n
= dx_get_count (entries
);
508 printk(KERN_DEBUG
"%s index ", label
);
509 for (i
= 0; i
< n
; i
++) {
510 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
511 0, (unsigned long)dx_get_block(entries
+ i
));
523 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
524 int size
, int show_names
)
526 unsigned names
= 0, space
= 0;
527 char *base
= (char *) de
;
528 struct dx_hash_info h
= *hinfo
;
531 while ((char *) de
< base
+ size
)
537 int len
= de
->name_len
;
538 char *name
= de
->name
;
539 while (len
--) printk("%c", *name
++);
540 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
541 printk(":%x.%u ", h
.hash
,
542 (unsigned) ((char *) de
- base
));
544 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
547 de
= ext4_next_entry(de
, size
);
549 printk("(%i)\n", names
);
550 return (struct stats
) { names
, space
, 1 };
553 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
554 struct dx_entry
*entries
, int levels
)
556 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
557 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
559 struct buffer_head
*bh
;
561 printk("%i indexed blocks...\n", count
);
562 for (i
= 0; i
< count
; i
++, entries
++)
564 ext4_lblk_t block
= dx_get_block(entries
);
565 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
566 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
568 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
569 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0,&err
))) continue;
571 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
572 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
573 names
+= stats
.names
;
574 space
+= stats
.space
;
575 bcount
+= stats
.bcount
;
579 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
580 levels
? "" : " ", names
, space
/bcount
,
581 (space
/bcount
)*100/blocksize
);
582 return (struct stats
) { names
, space
, bcount
};
584 #endif /* DX_DEBUG */
587 * Probe for a directory leaf block to search.
589 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
590 * error in the directory index, and the caller should fall back to
591 * searching the directory normally. The callers of dx_probe **MUST**
592 * check for this error code, and make sure it never gets reflected
595 static struct dx_frame
*
596 dx_probe(const struct qstr
*d_name
, struct inode
*dir
,
597 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
599 unsigned count
, indirect
;
600 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
601 struct dx_root
*root
;
602 struct buffer_head
*bh
;
603 struct dx_frame
*frame
= frame_in
;
607 if (!(bh
= ext4_bread(NULL
, dir
, 0, 0, err
))) {
609 *err
= ERR_BAD_DX_DIR
;
612 root
= (struct dx_root
*) bh
->b_data
;
613 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
614 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
615 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
616 ext4_warning(dir
->i_sb
, "Unrecognised inode hash code %d",
617 root
->info
.hash_version
);
619 *err
= ERR_BAD_DX_DIR
;
622 hinfo
->hash_version
= root
->info
.hash_version
;
623 if (hinfo
->hash_version
<= DX_HASH_TEA
)
624 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
625 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
627 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
630 if (root
->info
.unused_flags
& 1) {
631 ext4_warning(dir
->i_sb
, "Unimplemented inode hash flags: %#06x",
632 root
->info
.unused_flags
);
634 *err
= ERR_BAD_DX_DIR
;
638 if ((indirect
= root
->info
.indirect_levels
) > 1) {
639 ext4_warning(dir
->i_sb
, "Unimplemented inode hash depth: %#06x",
640 root
->info
.indirect_levels
);
642 *err
= ERR_BAD_DX_DIR
;
646 if (!buffer_verified(bh
) &&
647 !ext4_dx_csum_verify(dir
, (struct ext4_dir_entry
*)bh
->b_data
)) {
648 ext4_warning(dir
->i_sb
, "Root failed checksum");
650 *err
= ERR_BAD_DX_DIR
;
653 set_buffer_verified(bh
);
655 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
656 root
->info
.info_length
);
658 if (dx_get_limit(entries
) != dx_root_limit(dir
,
659 root
->info
.info_length
)) {
660 ext4_warning(dir
->i_sb
, "dx entry: limit != root limit");
662 *err
= ERR_BAD_DX_DIR
;
666 dxtrace(printk("Look up %x", hash
));
669 count
= dx_get_count(entries
);
670 if (!count
|| count
> dx_get_limit(entries
)) {
671 ext4_warning(dir
->i_sb
,
672 "dx entry: no count or count > limit");
674 *err
= ERR_BAD_DX_DIR
;
679 q
= entries
+ count
- 1;
683 dxtrace(printk("."));
684 if (dx_get_hash(m
) > hash
)
690 if (0) // linear search cross check
692 unsigned n
= count
- 1;
696 dxtrace(printk(","));
697 if (dx_get_hash(++at
) > hash
)
703 assert (at
== p
- 1);
707 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
709 frame
->entries
= entries
;
711 if (!indirect
--) return frame
;
712 if (!(bh
= ext4_bread(NULL
, dir
, dx_get_block(at
), 0, err
))) {
714 *err
= ERR_BAD_DX_DIR
;
717 at
= entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
719 if (!buffer_verified(bh
) &&
720 !ext4_dx_csum_verify(dir
,
721 (struct ext4_dir_entry
*)bh
->b_data
)) {
722 ext4_warning(dir
->i_sb
, "Node failed checksum");
724 *err
= ERR_BAD_DX_DIR
;
727 set_buffer_verified(bh
);
729 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
730 ext4_warning(dir
->i_sb
,
731 "dx entry: limit != node limit");
733 *err
= ERR_BAD_DX_DIR
;
740 while (frame
>= frame_in
) {
745 if (*err
== ERR_BAD_DX_DIR
)
746 ext4_warning(dir
->i_sb
,
747 "Corrupt dir inode %lu, running e2fsck is "
748 "recommended.", dir
->i_ino
);
752 static void dx_release (struct dx_frame
*frames
)
754 if (frames
[0].bh
== NULL
)
757 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
758 brelse(frames
[1].bh
);
759 brelse(frames
[0].bh
);
763 * This function increments the frame pointer to search the next leaf
764 * block, and reads in the necessary intervening nodes if the search
765 * should be necessary. Whether or not the search is necessary is
766 * controlled by the hash parameter. If the hash value is even, then
767 * the search is only continued if the next block starts with that
768 * hash value. This is used if we are searching for a specific file.
770 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
772 * This function returns 1 if the caller should continue to search,
773 * or 0 if it should not. If there is an error reading one of the
774 * index blocks, it will a negative error code.
776 * If start_hash is non-null, it will be filled in with the starting
777 * hash of the next page.
779 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
780 struct dx_frame
*frame
,
781 struct dx_frame
*frames
,
785 struct buffer_head
*bh
;
786 int err
, num_frames
= 0;
791 * Find the next leaf page by incrementing the frame pointer.
792 * If we run out of entries in the interior node, loop around and
793 * increment pointer in the parent node. When we break out of
794 * this loop, num_frames indicates the number of interior
795 * nodes need to be read.
798 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
807 * If the hash is 1, then continue only if the next page has a
808 * continuation hash of any value. This is used for readdir
809 * handling. Otherwise, check to see if the hash matches the
810 * desired contiuation hash. If it doesn't, return since
811 * there's no point to read in the successive index pages.
813 bhash
= dx_get_hash(p
->at
);
816 if ((hash
& 1) == 0) {
817 if ((bhash
& ~1) != hash
)
821 * If the hash is HASH_NB_ALWAYS, we always go to the next
822 * block so no check is necessary
824 while (num_frames
--) {
825 if (!(bh
= ext4_bread(NULL
, dir
, dx_get_block(p
->at
),
828 ext4_error(dir
->i_sb
,
829 "Directory hole detected on inode %lu\n",
833 return err
; /* Failure */
836 if (!buffer_verified(bh
) &&
837 !ext4_dx_csum_verify(dir
,
838 (struct ext4_dir_entry
*)bh
->b_data
)) {
839 ext4_warning(dir
->i_sb
, "Node failed checksum");
842 set_buffer_verified(bh
);
847 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
854 * This function fills a red-black tree with information from a
855 * directory block. It returns the number directory entries loaded
856 * into the tree. If there is an error it is returned in err.
858 static int htree_dirblock_to_tree(struct file
*dir_file
,
859 struct inode
*dir
, ext4_lblk_t block
,
860 struct dx_hash_info
*hinfo
,
861 __u32 start_hash
, __u32 start_minor_hash
)
863 struct buffer_head
*bh
;
864 struct ext4_dir_entry_2
*de
, *top
;
865 int err
= 0, count
= 0;
867 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
868 (unsigned long)block
));
869 if (!(bh
= ext4_bread(NULL
, dir
, block
, 0, &err
))) {
872 ext4_error(dir
->i_sb
,
873 "Directory hole detected on inode %lu\n",
879 if (!buffer_verified(bh
) &&
880 !ext4_dirent_csum_verify(dir
, (struct ext4_dir_entry
*)bh
->b_data
))
882 set_buffer_verified(bh
);
884 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
885 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
886 dir
->i_sb
->s_blocksize
-
887 EXT4_DIR_REC_LEN(0));
888 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
889 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
890 bh
->b_data
, bh
->b_size
,
891 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
892 + ((char *)de
- bh
->b_data
))) {
893 /* On error, skip the f_pos to the next block. */
894 dir_file
->f_pos
= (dir_file
->f_pos
|
895 (dir
->i_sb
->s_blocksize
- 1)) + 1;
899 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
900 if ((hinfo
->hash
< start_hash
) ||
901 ((hinfo
->hash
== start_hash
) &&
902 (hinfo
->minor_hash
< start_minor_hash
)))
906 if ((err
= ext4_htree_store_dirent(dir_file
,
907 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
919 * This function fills a red-black tree with information from a
920 * directory. We start scanning the directory in hash order, starting
921 * at start_hash and start_minor_hash.
923 * This function returns the number of entries inserted into the tree,
924 * or a negative error code.
926 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
927 __u32 start_minor_hash
, __u32
*next_hash
)
929 struct dx_hash_info hinfo
;
930 struct ext4_dir_entry_2
*de
;
931 struct dx_frame frames
[2], *frame
;
938 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
939 start_hash
, start_minor_hash
));
940 dir
= dir_file
->f_path
.dentry
->d_inode
;
941 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
942 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
943 if (hinfo
.hash_version
<= DX_HASH_TEA
)
944 hinfo
.hash_version
+=
945 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
946 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
947 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
948 start_hash
, start_minor_hash
);
952 hinfo
.hash
= start_hash
;
953 hinfo
.minor_hash
= 0;
954 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
, &err
);
958 /* Add '.' and '..' from the htree header */
959 if (!start_hash
&& !start_minor_hash
) {
960 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
961 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
965 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
966 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
967 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
968 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
974 block
= dx_get_block(frame
->at
);
975 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
976 start_hash
, start_minor_hash
);
983 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
984 frame
, frames
, &hashval
);
985 *next_hash
= hashval
;
991 * Stop if: (a) there are no more entries, or
992 * (b) we have inserted at least one entry and the
993 * next hash value is not a continuation
996 (count
&& ((hashval
& 1) == 0)))
1000 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1001 "next hash: %x\n", count
, *next_hash
));
1008 static inline int search_dirblock(struct buffer_head
*bh
,
1010 const struct qstr
*d_name
,
1011 unsigned int offset
,
1012 struct ext4_dir_entry_2
**res_dir
)
1014 return search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1015 d_name
, offset
, res_dir
);
1019 * Directory block splitting, compacting
1023 * Create map of hash values, offsets, and sizes, stored at end of block.
1024 * Returns number of entries mapped.
1026 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
1027 struct dx_hash_info
*hinfo
,
1028 struct dx_map_entry
*map_tail
)
1031 char *base
= (char *) de
;
1032 struct dx_hash_info h
= *hinfo
;
1034 while ((char *) de
< base
+ blocksize
) {
1035 if (de
->name_len
&& de
->inode
) {
1036 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1038 map_tail
->hash
= h
.hash
;
1039 map_tail
->offs
= ((char *) de
- base
)>>2;
1040 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1044 /* XXX: do we need to check rec_len == 0 case? -Chris */
1045 de
= ext4_next_entry(de
, blocksize
);
1050 /* Sort map by hash value */
1051 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1053 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1055 /* Combsort until bubble sort doesn't suck */
1057 count
= count
*10/13;
1058 if (count
- 9 < 2) /* 9, 10 -> 11 */
1060 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1061 if (p
->hash
< q
->hash
)
1064 /* Garden variety bubble sort */
1069 if (q
[1].hash
>= q
[0].hash
)
1077 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1079 struct dx_entry
*entries
= frame
->entries
;
1080 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1081 int count
= dx_get_count(entries
);
1083 assert(count
< dx_get_limit(entries
));
1084 assert(old
< entries
+ count
);
1085 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1086 dx_set_hash(new, hash
);
1087 dx_set_block(new, block
);
1088 dx_set_count(entries
, count
+ 1);
1092 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1094 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1095 * `de != NULL' is guaranteed by caller.
1097 static inline int ext4_match (int len
, const char * const name
,
1098 struct ext4_dir_entry_2
* de
)
1100 if (len
!= de
->name_len
)
1104 return !memcmp(name
, de
->name
, len
);
1108 * Returns 0 if not found, -1 on failure, and 1 on success
1110 int search_dir(struct buffer_head
*bh
,
1114 const struct qstr
*d_name
,
1115 unsigned int offset
,
1116 struct ext4_dir_entry_2
**res_dir
)
1118 struct ext4_dir_entry_2
* de
;
1121 const char *name
= d_name
->name
;
1122 int namelen
= d_name
->len
;
1124 de
= (struct ext4_dir_entry_2
*)search_buf
;
1125 dlimit
= search_buf
+ buf_size
;
1126 while ((char *) de
< dlimit
) {
1127 /* this code is executed quadratically often */
1128 /* do minimal checking `by hand' */
1130 if ((char *) de
+ namelen
<= dlimit
&&
1131 ext4_match (namelen
, name
, de
)) {
1132 /* found a match - just to be sure, do a full check */
1133 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, bh
->b_data
,
1134 bh
->b_size
, offset
))
1139 /* prevent looping on a bad block */
1140 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1141 dir
->i_sb
->s_blocksize
);
1145 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1150 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1151 struct ext4_dir_entry
*de
)
1153 struct super_block
*sb
= dir
->i_sb
;
1159 if (de
->inode
== 0 &&
1160 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1169 * finds an entry in the specified directory with the wanted name. It
1170 * returns the cache buffer in which the entry was found, and the entry
1171 * itself (as a parameter - res_dir). It does NOT read the inode of the
1172 * entry - you'll have to do that yourself if you want to.
1174 * The returned buffer_head has ->b_count elevated. The caller is expected
1175 * to brelse() it when appropriate.
1177 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1178 const struct qstr
*d_name
,
1179 struct ext4_dir_entry_2
** res_dir
)
1181 struct super_block
*sb
;
1182 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1183 struct buffer_head
*bh
, *ret
= NULL
;
1184 ext4_lblk_t start
, block
, b
;
1185 const u8
*name
= d_name
->name
;
1186 int ra_max
= 0; /* Number of bh's in the readahead
1188 int ra_ptr
= 0; /* Current index into readahead
1191 ext4_lblk_t nblocks
;
1197 namelen
= d_name
->len
;
1198 if (namelen
> EXT4_NAME_LEN
)
1201 if (ext4_has_inline_data(dir
)) {
1202 int has_inline_data
= 1;
1203 ret
= ext4_find_inline_entry(dir
, d_name
, res_dir
,
1205 if (has_inline_data
)
1209 if ((namelen
<= 2) && (name
[0] == '.') &&
1210 (name
[1] == '.' || name
[1] == '\0')) {
1212 * "." or ".." will only be in the first block
1213 * NFS may look up ".."; "." should be handled by the VFS
1220 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
, &err
);
1222 * On success, or if the error was file not found,
1223 * return. Otherwise, fall back to doing a search the
1224 * old fashioned way.
1226 if (bh
|| (err
!= ERR_BAD_DX_DIR
))
1228 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1231 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1232 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1233 if (start
>= nblocks
)
1239 * We deal with the read-ahead logic here.
1241 if (ra_ptr
>= ra_max
) {
1242 /* Refill the readahead buffer */
1245 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1247 * Terminate if we reach the end of the
1248 * directory and must wrap, or if our
1249 * search has finished at this block.
1251 if (b
>= nblocks
|| (num
&& block
== start
)) {
1252 bh_use
[ra_max
] = NULL
;
1256 bh
= ext4_getblk(NULL
, dir
, b
++, 0, &err
);
1257 bh_use
[ra_max
] = bh
;
1259 ll_rw_block(READ
| REQ_META
| REQ_PRIO
,
1263 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1266 if (!buffer_uptodate(bh
)) {
1267 /* read error, skip block & hope for the best */
1268 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1269 (unsigned long) block
);
1273 if (!buffer_verified(bh
) &&
1274 !is_dx_internal_node(dir
, block
,
1275 (struct ext4_dir_entry
*)bh
->b_data
) &&
1276 !ext4_dirent_csum_verify(dir
,
1277 (struct ext4_dir_entry
*)bh
->b_data
)) {
1278 EXT4_ERROR_INODE(dir
, "checksumming directory "
1279 "block %lu", (unsigned long)block
);
1283 set_buffer_verified(bh
);
1284 i
= search_dirblock(bh
, dir
, d_name
,
1285 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1287 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1289 goto cleanup_and_exit
;
1293 goto cleanup_and_exit
;
1296 if (++block
>= nblocks
)
1298 } while (block
!= start
);
1301 * If the directory has grown while we were searching, then
1302 * search the last part of the directory before giving up.
1305 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1306 if (block
< nblocks
) {
1312 /* Clean up the read-ahead blocks */
1313 for (; ra_ptr
< ra_max
; ra_ptr
++)
1314 brelse(bh_use
[ra_ptr
]);
1318 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
1319 struct ext4_dir_entry_2
**res_dir
, int *err
)
1321 struct super_block
* sb
= dir
->i_sb
;
1322 struct dx_hash_info hinfo
;
1323 struct dx_frame frames
[2], *frame
;
1324 struct buffer_head
*bh
;
1328 if (!(frame
= dx_probe(d_name
, dir
, &hinfo
, frames
, err
)))
1331 block
= dx_get_block(frame
->at
);
1332 if (!(bh
= ext4_bread(NULL
, dir
, block
, 0, err
))) {
1335 ext4_error(dir
->i_sb
,
1336 "Directory hole detected on inode %lu\n",
1342 if (!buffer_verified(bh
) &&
1343 !ext4_dirent_csum_verify(dir
,
1344 (struct ext4_dir_entry
*)bh
->b_data
)) {
1345 EXT4_ERROR_INODE(dir
, "checksumming directory "
1346 "block %lu", (unsigned long)block
);
1351 set_buffer_verified(bh
);
1352 retval
= search_dirblock(bh
, dir
, d_name
,
1353 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1355 if (retval
== 1) { /* Success! */
1361 *err
= ERR_BAD_DX_DIR
;
1365 /* Check to see if we should continue to search */
1366 retval
= ext4_htree_next_block(dir
, hinfo
.hash
, frame
,
1370 "error reading index page in directory #%lu",
1375 } while (retval
== 1);
1379 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1380 dx_release (frames
);
1384 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1386 struct inode
*inode
;
1387 struct ext4_dir_entry_2
*de
;
1388 struct buffer_head
*bh
;
1390 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1391 return ERR_PTR(-ENAMETOOLONG
);
1393 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
1396 __u32 ino
= le32_to_cpu(de
->inode
);
1398 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1399 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1400 return ERR_PTR(-EIO
);
1402 if (unlikely(ino
== dir
->i_ino
)) {
1403 EXT4_ERROR_INODE(dir
, "'%.*s' linked to parent dir",
1405 dentry
->d_name
.name
);
1406 return ERR_PTR(-EIO
);
1408 inode
= ext4_iget(dir
->i_sb
, ino
);
1409 if (inode
== ERR_PTR(-ESTALE
)) {
1410 EXT4_ERROR_INODE(dir
,
1411 "deleted inode referenced: %u",
1413 return ERR_PTR(-EIO
);
1416 return d_splice_alias(inode
, dentry
);
1420 struct dentry
*ext4_get_parent(struct dentry
*child
)
1423 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1424 struct ext4_dir_entry_2
* de
;
1425 struct buffer_head
*bh
;
1427 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
);
1429 return ERR_PTR(-ENOENT
);
1430 ino
= le32_to_cpu(de
->inode
);
1433 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1434 EXT4_ERROR_INODE(child
->d_inode
,
1435 "bad parent inode number: %u", ino
);
1436 return ERR_PTR(-EIO
);
1439 return d_obtain_alias(ext4_iget(child
->d_inode
->i_sb
, ino
));
1443 static unsigned char ext4_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
1444 [S_IFREG
>> S_SHIFT
] = EXT4_FT_REG_FILE
,
1445 [S_IFDIR
>> S_SHIFT
] = EXT4_FT_DIR
,
1446 [S_IFCHR
>> S_SHIFT
] = EXT4_FT_CHRDEV
,
1447 [S_IFBLK
>> S_SHIFT
] = EXT4_FT_BLKDEV
,
1448 [S_IFIFO
>> S_SHIFT
] = EXT4_FT_FIFO
,
1449 [S_IFSOCK
>> S_SHIFT
] = EXT4_FT_SOCK
,
1450 [S_IFLNK
>> S_SHIFT
] = EXT4_FT_SYMLINK
,
1453 static inline void ext4_set_de_type(struct super_block
*sb
,
1454 struct ext4_dir_entry_2
*de
,
1456 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FILETYPE
))
1457 de
->file_type
= ext4_type_by_mode
[(mode
& S_IFMT
)>>S_SHIFT
];
1461 * Move count entries from end of map between two memory locations.
1462 * Returns pointer to last entry moved.
1464 static struct ext4_dir_entry_2
*
1465 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1468 unsigned rec_len
= 0;
1471 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1472 (from
+ (map
->offs
<<2));
1473 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1474 memcpy (to
, de
, rec_len
);
1475 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1476 ext4_rec_len_to_disk(rec_len
, blocksize
);
1481 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1485 * Compact each dir entry in the range to the minimal rec_len.
1486 * Returns pointer to last entry in range.
1488 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1490 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1491 unsigned rec_len
= 0;
1494 while ((char*)de
< base
+ blocksize
) {
1495 next
= ext4_next_entry(de
, blocksize
);
1496 if (de
->inode
&& de
->name_len
) {
1497 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1499 memmove(to
, de
, rec_len
);
1500 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1502 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1510 * Split a full leaf block to make room for a new dir entry.
1511 * Allocate a new block, and move entries so that they are approx. equally full.
1512 * Returns pointer to de in block into which the new entry will be inserted.
1514 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1515 struct buffer_head
**bh
,struct dx_frame
*frame
,
1516 struct dx_hash_info
*hinfo
, int *error
)
1518 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1519 unsigned count
, continued
;
1520 struct buffer_head
*bh2
;
1521 ext4_lblk_t newblock
;
1523 struct dx_map_entry
*map
;
1524 char *data1
= (*bh
)->b_data
, *data2
;
1525 unsigned split
, move
, size
;
1526 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1527 struct ext4_dir_entry_tail
*t
;
1531 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
1532 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1533 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1535 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
1542 BUFFER_TRACE(*bh
, "get_write_access");
1543 err
= ext4_journal_get_write_access(handle
, *bh
);
1547 BUFFER_TRACE(frame
->bh
, "get_write_access");
1548 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1552 data2
= bh2
->b_data
;
1554 /* create map in the end of data2 block */
1555 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1556 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1557 blocksize
, hinfo
, map
);
1559 dx_sort_map(map
, count
);
1560 /* Split the existing block in the middle, size-wise */
1563 for (i
= count
-1; i
>= 0; i
--) {
1564 /* is more than half of this entry in 2nd half of the block? */
1565 if (size
+ map
[i
].size
/2 > blocksize
/2)
1567 size
+= map
[i
].size
;
1570 /* map index at which we will split */
1571 split
= count
- move
;
1572 hash2
= map
[split
].hash
;
1573 continued
= hash2
== map
[split
- 1].hash
;
1574 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1575 (unsigned long)dx_get_block(frame
->at
),
1576 hash2
, split
, count
-split
));
1578 /* Fancy dance to stay within two buffers */
1579 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
, blocksize
);
1580 de
= dx_pack_dirents(data1
, blocksize
);
1581 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1584 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1588 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1589 initialize_dirent_tail(t
, blocksize
);
1591 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1592 initialize_dirent_tail(t
, blocksize
);
1595 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1596 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1598 /* Which block gets the new entry? */
1599 if (hinfo
->hash
>= hash2
)
1604 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1605 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1608 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1612 dxtrace(dx_show_index("frame", frame
->entries
));
1619 ext4_std_error(dir
->i_sb
, err
);
1625 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1626 struct buffer_head
*bh
,
1627 void *buf
, int buf_size
,
1628 const char *name
, int namelen
,
1629 struct ext4_dir_entry_2
**dest_de
)
1631 struct ext4_dir_entry_2
*de
;
1632 unsigned short reclen
= EXT4_DIR_REC_LEN(namelen
);
1634 unsigned int offset
= 0;
1637 de
= (struct ext4_dir_entry_2
*)buf
;
1638 top
= buf
+ buf_size
- reclen
;
1639 while ((char *) de
<= top
) {
1640 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1641 buf
, buf_size
, offset
))
1643 if (ext4_match(namelen
, name
, de
))
1645 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1646 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1647 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1649 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1652 if ((char *) de
> top
)
1659 void ext4_insert_dentry(struct inode
*inode
,
1660 struct ext4_dir_entry_2
*de
,
1662 const char *name
, int namelen
)
1667 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1668 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1670 struct ext4_dir_entry_2
*de1
=
1671 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1672 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1673 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1676 de
->file_type
= EXT4_FT_UNKNOWN
;
1677 de
->inode
= cpu_to_le32(inode
->i_ino
);
1678 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1679 de
->name_len
= namelen
;
1680 memcpy(de
->name
, name
, namelen
);
1683 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1684 * it points to a directory entry which is guaranteed to be large
1685 * enough for new directory entry. If de is NULL, then
1686 * add_dirent_to_buf will attempt search the directory block for
1687 * space. It will return -ENOSPC if no space is available, and -EIO
1688 * and -EEXIST if directory entry already exists.
1690 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1691 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1692 struct buffer_head
*bh
)
1694 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1695 const char *name
= dentry
->d_name
.name
;
1696 int namelen
= dentry
->d_name
.len
;
1697 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1698 unsigned short reclen
;
1702 if (EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
1703 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1704 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1706 reclen
= EXT4_DIR_REC_LEN(namelen
);
1708 err
= ext4_find_dest_de(dir
, inode
,
1709 bh
, bh
->b_data
, blocksize
- csum_size
,
1710 name
, namelen
, &de
);
1714 BUFFER_TRACE(bh
, "get_write_access");
1715 err
= ext4_journal_get_write_access(handle
, bh
);
1717 ext4_std_error(dir
->i_sb
, err
);
1721 /* By now the buffer is marked for journaling */
1722 ext4_insert_dentry(inode
, de
, blocksize
, name
, namelen
);
1725 * XXX shouldn't update any times until successful
1726 * completion of syscall, but too many callers depend
1729 * XXX similarly, too many callers depend on
1730 * ext4_new_inode() setting the times, but error
1731 * recovery deletes the inode, so the worst that can
1732 * happen is that the times are slightly out of date
1733 * and/or different from the directory change time.
1735 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1736 ext4_update_dx_flag(dir
);
1738 ext4_mark_inode_dirty(handle
, dir
);
1739 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1740 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1742 ext4_std_error(dir
->i_sb
, err
);
1747 * This converts a one block unindexed directory to a 3 block indexed
1748 * directory, and adds the dentry to the indexed directory.
1750 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1751 struct inode
*inode
, struct buffer_head
*bh
)
1753 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1754 const char *name
= dentry
->d_name
.name
;
1755 int namelen
= dentry
->d_name
.len
;
1756 struct buffer_head
*bh2
;
1757 struct dx_root
*root
;
1758 struct dx_frame frames
[2], *frame
;
1759 struct dx_entry
*entries
;
1760 struct ext4_dir_entry_2
*de
, *de2
;
1761 struct ext4_dir_entry_tail
*t
;
1766 struct dx_hash_info hinfo
;
1768 struct fake_dirent
*fde
;
1771 if (EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
1772 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1773 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1775 blocksize
= dir
->i_sb
->s_blocksize
;
1776 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1777 retval
= ext4_journal_get_write_access(handle
, bh
);
1779 ext4_std_error(dir
->i_sb
, retval
);
1783 root
= (struct dx_root
*) bh
->b_data
;
1785 /* The 0th block becomes the root, move the dirents out */
1786 fde
= &root
->dotdot
;
1787 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1788 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1789 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1790 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1794 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1796 /* Allocate new block for the 0th block's dirents */
1797 bh2
= ext4_append(handle
, dir
, &block
, &retval
);
1802 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1803 data1
= bh2
->b_data
;
1805 memcpy (data1
, de
, len
);
1806 de
= (struct ext4_dir_entry_2
*) data1
;
1808 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1810 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1815 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1816 initialize_dirent_tail(t
, blocksize
);
1819 /* Initialize the root; the dot dirents already exist */
1820 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1821 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1823 memset (&root
->info
, 0, sizeof(root
->info
));
1824 root
->info
.info_length
= sizeof(root
->info
);
1825 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1826 entries
= root
->entries
;
1827 dx_set_block(entries
, 1);
1828 dx_set_count(entries
, 1);
1829 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1831 /* Initialize as for dx_probe */
1832 hinfo
.hash_version
= root
->info
.hash_version
;
1833 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1834 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1835 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1836 ext4fs_dirhash(name
, namelen
, &hinfo
);
1838 frame
->entries
= entries
;
1839 frame
->at
= entries
;
1843 ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1844 ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1846 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1849 * Even if the block split failed, we have to properly write
1850 * out all the changes we did so far. Otherwise we can end up
1851 * with corrupted filesystem.
1853 ext4_mark_inode_dirty(handle
, dir
);
1859 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1867 * adds a file entry to the specified directory, using the same
1868 * semantics as ext4_find_entry(). It returns NULL if it failed.
1870 * NOTE!! The inode part of 'de' is left at 0 - which means you
1871 * may not sleep between calling this and putting something into
1872 * the entry, as someone else might have used it while you slept.
1874 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1875 struct inode
*inode
)
1877 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1878 struct buffer_head
*bh
;
1879 struct ext4_dir_entry_2
*de
;
1880 struct ext4_dir_entry_tail
*t
;
1881 struct super_block
*sb
;
1885 ext4_lblk_t block
, blocks
;
1888 if (EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
1889 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1890 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1893 blocksize
= sb
->s_blocksize
;
1894 if (!dentry
->d_name
.len
)
1897 if (ext4_has_inline_data(dir
)) {
1898 retval
= ext4_try_add_inline_entry(handle
, dentry
, inode
);
1908 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1909 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1911 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
1913 ext4_mark_inode_dirty(handle
, dir
);
1915 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1916 for (block
= 0; block
< blocks
; block
++) {
1917 if (!(bh
= ext4_bread(handle
, dir
, block
, 0, &retval
))) {
1920 ext4_error(inode
->i_sb
,
1921 "Directory hole detected on inode %lu\n",
1926 if (!buffer_verified(bh
) &&
1927 !ext4_dirent_csum_verify(dir
,
1928 (struct ext4_dir_entry
*)bh
->b_data
))
1930 set_buffer_verified(bh
);
1931 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1932 if (retval
!= -ENOSPC
) {
1937 if (blocks
== 1 && !dx_fallback
&&
1938 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1939 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1942 bh
= ext4_append(handle
, dir
, &block
, &retval
);
1945 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1947 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
1950 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
1951 initialize_dirent_tail(t
, blocksize
);
1954 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1957 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
1962 * Returns 0 for success, or a negative error value
1964 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1965 struct inode
*inode
)
1967 struct dx_frame frames
[2], *frame
;
1968 struct dx_entry
*entries
, *at
;
1969 struct dx_hash_info hinfo
;
1970 struct buffer_head
*bh
;
1971 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1972 struct super_block
*sb
= dir
->i_sb
;
1973 struct ext4_dir_entry_2
*de
;
1976 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1979 entries
= frame
->entries
;
1982 if (!(bh
= ext4_bread(handle
, dir
, dx_get_block(frame
->at
), 0, &err
))) {
1985 ext4_error(dir
->i_sb
,
1986 "Directory hole detected on inode %lu\n",
1992 if (!buffer_verified(bh
) &&
1993 !ext4_dirent_csum_verify(dir
, (struct ext4_dir_entry
*)bh
->b_data
))
1995 set_buffer_verified(bh
);
1997 BUFFER_TRACE(bh
, "get_write_access");
1998 err
= ext4_journal_get_write_access(handle
, bh
);
2002 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
2006 /* Block full, should compress but for now just split */
2007 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
2008 dx_get_count(entries
), dx_get_limit(entries
)));
2009 /* Need to split index? */
2010 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
2011 ext4_lblk_t newblock
;
2012 unsigned icount
= dx_get_count(entries
);
2013 int levels
= frame
- frames
;
2014 struct dx_entry
*entries2
;
2015 struct dx_node
*node2
;
2016 struct buffer_head
*bh2
;
2018 if (levels
&& (dx_get_count(frames
->entries
) ==
2019 dx_get_limit(frames
->entries
))) {
2020 ext4_warning(sb
, "Directory index full!");
2024 bh2
= ext4_append (handle
, dir
, &newblock
, &err
);
2027 node2
= (struct dx_node
*)(bh2
->b_data
);
2028 entries2
= node2
->entries
;
2029 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2030 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2032 BUFFER_TRACE(frame
->bh
, "get_write_access");
2033 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2037 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2038 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2039 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2042 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2043 err
= ext4_journal_get_write_access(handle
,
2048 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2049 icount2
* sizeof(struct dx_entry
));
2050 dx_set_count(entries
, icount1
);
2051 dx_set_count(entries2
, icount2
);
2052 dx_set_limit(entries2
, dx_node_limit(dir
));
2054 /* Which index block gets the new entry? */
2055 if (at
- entries
>= icount1
) {
2056 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2057 frame
->entries
= entries
= entries2
;
2058 swap(frame
->bh
, bh2
);
2060 dx_insert_block(frames
+ 0, hash2
, newblock
);
2061 dxtrace(dx_show_index("node", frames
[1].entries
));
2062 dxtrace(dx_show_index("node",
2063 ((struct dx_node
*) bh2
->b_data
)->entries
));
2064 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2069 dxtrace(printk(KERN_DEBUG
2070 "Creating second level index...\n"));
2071 memcpy((char *) entries2
, (char *) entries
,
2072 icount
* sizeof(struct dx_entry
));
2073 dx_set_limit(entries2
, dx_node_limit(dir
));
2076 dx_set_count(entries
, 1);
2077 dx_set_block(entries
+ 0, newblock
);
2078 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2080 /* Add new access path frame */
2082 frame
->at
= at
= at
- entries
+ entries2
;
2083 frame
->entries
= entries
= entries2
;
2085 err
= ext4_journal_get_write_access(handle
,
2090 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2092 ext4_std_error(inode
->i_sb
, err
);
2096 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
2099 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
2103 ext4_std_error(dir
->i_sb
, err
);
2112 * ext4_generic_delete_entry deletes a directory entry by merging it
2113 * with the previous entry
2115 int ext4_generic_delete_entry(handle_t
*handle
,
2117 struct ext4_dir_entry_2
*de_del
,
2118 struct buffer_head
*bh
,
2123 struct ext4_dir_entry_2
*de
, *pde
;
2124 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2129 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2130 while (i
< buf_size
- csum_size
) {
2131 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2132 bh
->b_data
, bh
->b_size
, i
))
2136 pde
->rec_len
= ext4_rec_len_to_disk(
2137 ext4_rec_len_from_disk(pde
->rec_len
,
2139 ext4_rec_len_from_disk(de
->rec_len
,
2147 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2149 de
= ext4_next_entry(de
, blocksize
);
2154 static int ext4_delete_entry(handle_t
*handle
,
2156 struct ext4_dir_entry_2
*de_del
,
2157 struct buffer_head
*bh
)
2159 int err
, csum_size
= 0;
2161 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
2162 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
2163 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2165 BUFFER_TRACE(bh
, "get_write_access");
2166 err
= ext4_journal_get_write_access(handle
, bh
);
2170 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2172 dir
->i_sb
->s_blocksize
, csum_size
);
2176 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2177 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2184 ext4_std_error(dir
->i_sb
, err
);
2189 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2190 * since this indicates that nlinks count was previously 1.
2192 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2195 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2196 /* limit is 16-bit i_links_count */
2197 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2198 set_nlink(inode
, 1);
2199 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
2200 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
2206 * If a directory had nlink == 1, then we should let it be 1. This indicates
2207 * directory has >EXT4_LINK_MAX subdirs.
2209 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2211 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2216 static int ext4_add_nondir(handle_t
*handle
,
2217 struct dentry
*dentry
, struct inode
*inode
)
2219 int err
= ext4_add_entry(handle
, dentry
, inode
);
2221 ext4_mark_inode_dirty(handle
, inode
);
2222 unlock_new_inode(inode
);
2223 d_instantiate(dentry
, inode
);
2227 unlock_new_inode(inode
);
2233 * By the time this is called, we already have created
2234 * the directory cache entry for the new file, but it
2235 * is so far negative - it has no inode.
2237 * If the create succeeds, we fill in the inode information
2238 * with d_instantiate().
2240 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2244 struct inode
*inode
;
2245 int err
, retries
= 0;
2247 dquot_initialize(dir
);
2250 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2251 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
2252 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
2254 return PTR_ERR(handle
);
2256 if (IS_DIRSYNC(dir
))
2257 ext4_handle_sync(handle
);
2259 inode
= ext4_new_inode(handle
, dir
, mode
, &dentry
->d_name
, 0, NULL
);
2260 err
= PTR_ERR(inode
);
2261 if (!IS_ERR(inode
)) {
2262 inode
->i_op
= &ext4_file_inode_operations
;
2263 inode
->i_fop
= &ext4_file_operations
;
2264 ext4_set_aops(inode
);
2265 err
= ext4_add_nondir(handle
, dentry
, inode
);
2267 ext4_journal_stop(handle
);
2268 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2273 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2274 umode_t mode
, dev_t rdev
)
2277 struct inode
*inode
;
2278 int err
, retries
= 0;
2280 if (!new_valid_dev(rdev
))
2283 dquot_initialize(dir
);
2286 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2287 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
2288 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
2290 return PTR_ERR(handle
);
2292 if (IS_DIRSYNC(dir
))
2293 ext4_handle_sync(handle
);
2295 inode
= ext4_new_inode(handle
, dir
, mode
, &dentry
->d_name
, 0, NULL
);
2296 err
= PTR_ERR(inode
);
2297 if (!IS_ERR(inode
)) {
2298 init_special_inode(inode
, inode
->i_mode
, rdev
);
2299 inode
->i_op
= &ext4_special_inode_operations
;
2300 err
= ext4_add_nondir(handle
, dentry
, inode
);
2302 ext4_journal_stop(handle
);
2303 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2308 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2309 struct ext4_dir_entry_2
*de
,
2310 int blocksize
, int csum_size
,
2311 unsigned int parent_ino
, int dotdot_real_len
)
2313 de
->inode
= cpu_to_le32(inode
->i_ino
);
2315 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2317 strcpy(de
->name
, ".");
2318 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2320 de
= ext4_next_entry(de
, blocksize
);
2321 de
->inode
= cpu_to_le32(parent_ino
);
2323 if (!dotdot_real_len
)
2324 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2325 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2328 de
->rec_len
= ext4_rec_len_to_disk(
2329 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2330 strcpy(de
->name
, "..");
2331 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2333 return ext4_next_entry(de
, blocksize
);
2336 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2337 struct inode
*inode
)
2339 struct buffer_head
*dir_block
= NULL
;
2340 struct ext4_dir_entry_2
*de
;
2341 struct ext4_dir_entry_tail
*t
;
2342 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2346 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
2347 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
2348 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2350 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2351 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2352 if (err
< 0 && err
!= -ENOSPC
)
2358 inode
->i_size
= EXT4_I(inode
)->i_disksize
= blocksize
;
2359 dir_block
= ext4_bread(handle
, inode
, 0, 1, &err
);
2360 if (!(dir_block
= ext4_bread(handle
, inode
, 0, 1, &err
))) {
2363 ext4_error(inode
->i_sb
,
2364 "Directory hole detected on inode %lu\n",
2369 BUFFER_TRACE(dir_block
, "get_write_access");
2370 err
= ext4_journal_get_write_access(handle
, dir_block
);
2373 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2374 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2375 set_nlink(inode
, 2);
2377 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2378 initialize_dirent_tail(t
, blocksize
);
2381 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2382 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2385 set_buffer_verified(dir_block
);
2391 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2394 struct inode
*inode
;
2395 int err
, retries
= 0;
2397 if (EXT4_DIR_LINK_MAX(dir
))
2400 dquot_initialize(dir
);
2403 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2404 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
2405 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
));
2407 return PTR_ERR(handle
);
2409 if (IS_DIRSYNC(dir
))
2410 ext4_handle_sync(handle
);
2412 inode
= ext4_new_inode(handle
, dir
, S_IFDIR
| mode
,
2413 &dentry
->d_name
, 0, NULL
);
2414 err
= PTR_ERR(inode
);
2418 inode
->i_op
= &ext4_dir_inode_operations
;
2419 inode
->i_fop
= &ext4_dir_operations
;
2420 err
= ext4_init_new_dir(handle
, dir
, inode
);
2422 goto out_clear_inode
;
2423 err
= ext4_mark_inode_dirty(handle
, inode
);
2425 err
= ext4_add_entry(handle
, dentry
, inode
);
2429 unlock_new_inode(inode
);
2430 ext4_mark_inode_dirty(handle
, inode
);
2434 ext4_inc_count(handle
, dir
);
2435 ext4_update_dx_flag(dir
);
2436 err
= ext4_mark_inode_dirty(handle
, dir
);
2438 goto out_clear_inode
;
2439 unlock_new_inode(inode
);
2440 d_instantiate(dentry
, inode
);
2442 ext4_journal_stop(handle
);
2443 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2449 * routine to check that the specified directory is empty (for rmdir)
2451 static int empty_dir(struct inode
*inode
)
2453 unsigned int offset
;
2454 struct buffer_head
*bh
;
2455 struct ext4_dir_entry_2
*de
, *de1
;
2456 struct super_block
*sb
;
2460 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
2461 !(bh
= ext4_bread(NULL
, inode
, 0, 0, &err
))) {
2463 EXT4_ERROR_INODE(inode
,
2464 "error %d reading directory lblock 0", err
);
2466 ext4_warning(inode
->i_sb
,
2467 "bad directory (dir #%lu) - no data block",
2471 if (!buffer_verified(bh
) &&
2472 !ext4_dirent_csum_verify(inode
,
2473 (struct ext4_dir_entry
*)bh
->b_data
)) {
2474 EXT4_ERROR_INODE(inode
, "checksum error reading directory "
2478 set_buffer_verified(bh
);
2479 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2480 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2481 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2482 !le32_to_cpu(de1
->inode
) ||
2483 strcmp(".", de
->name
) ||
2484 strcmp("..", de1
->name
)) {
2485 ext4_warning(inode
->i_sb
,
2486 "bad directory (dir #%lu) - no `.' or `..'",
2491 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2492 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2493 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2494 while (offset
< inode
->i_size
) {
2496 (void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2497 unsigned int lblock
;
2500 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2501 bh
= ext4_bread(NULL
, inode
, lblock
, 0, &err
);
2504 EXT4_ERROR_INODE(inode
,
2505 "error %d reading directory "
2506 "lblock %u", err
, lblock
);
2508 ext4_warning(inode
->i_sb
,
2509 "bad directory (dir #%lu) - no data block",
2512 offset
+= sb
->s_blocksize
;
2515 if (!buffer_verified(bh
) &&
2516 !ext4_dirent_csum_verify(inode
,
2517 (struct ext4_dir_entry
*)bh
->b_data
)) {
2518 EXT4_ERROR_INODE(inode
, "checksum error "
2519 "reading directory lblock 0");
2522 set_buffer_verified(bh
);
2523 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2525 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2526 bh
->b_data
, bh
->b_size
, offset
)) {
2527 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2529 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2532 if (le32_to_cpu(de
->inode
)) {
2536 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2537 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2543 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2544 * such inodes, starting at the superblock, in case we crash before the
2545 * file is closed/deleted, or in case the inode truncate spans multiple
2546 * transactions and the last transaction is not recovered after a crash.
2548 * At filesystem recovery time, we walk this list deleting unlinked
2549 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2551 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2553 struct super_block
*sb
= inode
->i_sb
;
2554 struct ext4_iloc iloc
;
2557 if (!EXT4_SB(sb
)->s_journal
)
2560 mutex_lock(&EXT4_SB(sb
)->s_orphan_lock
);
2561 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2565 * Orphan handling is only valid for files with data blocks
2566 * being truncated, or files being unlinked. Note that we either
2567 * hold i_mutex, or the inode can not be referenced from outside,
2568 * so i_nlink should not be bumped due to race
2570 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2571 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2573 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
2574 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
2578 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2582 * Due to previous errors inode may be already a part of on-disk
2583 * orphan list. If so skip on-disk list modification.
2585 if (NEXT_ORPHAN(inode
) && NEXT_ORPHAN(inode
) <=
2586 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
)))
2589 /* Insert this inode at the head of the on-disk orphan list... */
2590 NEXT_ORPHAN(inode
) = le32_to_cpu(EXT4_SB(sb
)->s_es
->s_last_orphan
);
2591 EXT4_SB(sb
)->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2592 err
= ext4_handle_dirty_super(handle
, sb
);
2593 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2597 /* Only add to the head of the in-memory list if all the
2598 * previous operations succeeded. If the orphan_add is going to
2599 * fail (possibly taking the journal offline), we can't risk
2600 * leaving the inode on the orphan list: stray orphan-list
2601 * entries can cause panics at unmount time.
2603 * This is safe: on error we're going to ignore the orphan list
2604 * anyway on the next recovery. */
2607 list_add(&EXT4_I(inode
)->i_orphan
, &EXT4_SB(sb
)->s_orphan
);
2609 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2610 jbd_debug(4, "orphan inode %lu will point to %d\n",
2611 inode
->i_ino
, NEXT_ORPHAN(inode
));
2613 mutex_unlock(&EXT4_SB(sb
)->s_orphan_lock
);
2614 ext4_std_error(inode
->i_sb
, err
);
2619 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2620 * of such inodes stored on disk, because it is finally being cleaned up.
2622 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2624 struct list_head
*prev
;
2625 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2626 struct ext4_sb_info
*sbi
;
2628 struct ext4_iloc iloc
;
2631 if (!EXT4_SB(inode
->i_sb
)->s_journal
)
2634 mutex_lock(&EXT4_SB(inode
->i_sb
)->s_orphan_lock
);
2635 if (list_empty(&ei
->i_orphan
))
2638 ino_next
= NEXT_ORPHAN(inode
);
2639 prev
= ei
->i_orphan
.prev
;
2640 sbi
= EXT4_SB(inode
->i_sb
);
2642 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2644 list_del_init(&ei
->i_orphan
);
2646 /* If we're on an error path, we may not have a valid
2647 * transaction handle with which to update the orphan list on
2648 * disk, but we still need to remove the inode from the linked
2649 * list in memory. */
2653 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2657 if (prev
== &sbi
->s_orphan
) {
2658 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2659 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2660 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2663 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2664 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2666 struct ext4_iloc iloc2
;
2667 struct inode
*i_prev
=
2668 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2670 jbd_debug(4, "orphan inode %lu will point to %u\n",
2671 i_prev
->i_ino
, ino_next
);
2672 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2675 NEXT_ORPHAN(i_prev
) = ino_next
;
2676 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2680 NEXT_ORPHAN(inode
) = 0;
2681 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2684 ext4_std_error(inode
->i_sb
, err
);
2686 mutex_unlock(&EXT4_SB(inode
->i_sb
)->s_orphan_lock
);
2694 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2697 struct inode
*inode
;
2698 struct buffer_head
*bh
;
2699 struct ext4_dir_entry_2
*de
;
2702 /* Initialize quotas before so that eventual writes go in
2703 * separate transaction */
2704 dquot_initialize(dir
);
2705 dquot_initialize(dentry
->d_inode
);
2707 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2709 return PTR_ERR(handle
);
2712 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2716 if (IS_DIRSYNC(dir
))
2717 ext4_handle_sync(handle
);
2719 inode
= dentry
->d_inode
;
2722 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2725 retval
= -ENOTEMPTY
;
2726 if (!empty_dir(inode
))
2729 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2732 if (!EXT4_DIR_LINK_EMPTY(inode
))
2733 ext4_warning(inode
->i_sb
,
2734 "empty directory has too many links (%d)",
2738 /* There's no need to set i_disksize: the fact that i_nlink is
2739 * zero will ensure that the right thing happens during any
2742 ext4_orphan_add(handle
, inode
);
2743 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2744 ext4_mark_inode_dirty(handle
, inode
);
2745 ext4_dec_count(handle
, dir
);
2746 ext4_update_dx_flag(dir
);
2747 ext4_mark_inode_dirty(handle
, dir
);
2750 ext4_journal_stop(handle
);
2755 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2758 struct inode
*inode
;
2759 struct buffer_head
*bh
;
2760 struct ext4_dir_entry_2
*de
;
2763 trace_ext4_unlink_enter(dir
, dentry
);
2764 /* Initialize quotas before so that eventual writes go
2765 * in separate transaction */
2766 dquot_initialize(dir
);
2767 dquot_initialize(dentry
->d_inode
);
2769 handle
= ext4_journal_start(dir
, EXT4_DELETE_TRANS_BLOCKS(dir
->i_sb
));
2771 return PTR_ERR(handle
);
2773 if (IS_DIRSYNC(dir
))
2774 ext4_handle_sync(handle
);
2777 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
);
2781 inode
= dentry
->d_inode
;
2784 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2787 if (!inode
->i_nlink
) {
2788 ext4_warning(inode
->i_sb
,
2789 "Deleting nonexistent file (%lu), %d",
2790 inode
->i_ino
, inode
->i_nlink
);
2791 set_nlink(inode
, 1);
2793 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2796 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2797 ext4_update_dx_flag(dir
);
2798 ext4_mark_inode_dirty(handle
, dir
);
2800 if (!inode
->i_nlink
)
2801 ext4_orphan_add(handle
, inode
);
2802 inode
->i_ctime
= ext4_current_time(inode
);
2803 ext4_mark_inode_dirty(handle
, inode
);
2807 ext4_journal_stop(handle
);
2809 trace_ext4_unlink_exit(dentry
, retval
);
2813 static int ext4_symlink(struct inode
*dir
,
2814 struct dentry
*dentry
, const char *symname
)
2817 struct inode
*inode
;
2818 int l
, err
, retries
= 0;
2821 l
= strlen(symname
)+1;
2822 if (l
> dir
->i_sb
->s_blocksize
)
2823 return -ENAMETOOLONG
;
2825 dquot_initialize(dir
);
2827 if (l
> EXT4_N_BLOCKS
* 4) {
2829 * For non-fast symlinks, we just allocate inode and put it on
2830 * orphan list in the first transaction => we need bitmap,
2831 * group descriptor, sb, inode block, quota blocks, and
2832 * possibly selinux xattr blocks.
2834 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2835 EXT4_XATTR_TRANS_BLOCKS
;
2838 * Fast symlink. We have to add entry to directory
2839 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2840 * allocate new inode (bitmap, group descriptor, inode block,
2841 * quota blocks, sb is already counted in previous macros).
2843 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2844 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3 +
2845 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
);
2848 handle
= ext4_journal_start(dir
, credits
);
2850 return PTR_ERR(handle
);
2852 if (IS_DIRSYNC(dir
))
2853 ext4_handle_sync(handle
);
2855 inode
= ext4_new_inode(handle
, dir
, S_IFLNK
|S_IRWXUGO
,
2856 &dentry
->d_name
, 0, NULL
);
2857 err
= PTR_ERR(inode
);
2861 if (l
> EXT4_N_BLOCKS
* 4) {
2862 inode
->i_op
= &ext4_symlink_inode_operations
;
2863 ext4_set_aops(inode
);
2865 * We cannot call page_symlink() with transaction started
2866 * because it calls into ext4_write_begin() which can wait
2867 * for transaction commit if we are running out of space
2868 * and thus we deadlock. So we have to stop transaction now
2869 * and restart it when symlink contents is written.
2871 * To keep fs consistent in case of crash, we have to put inode
2872 * to orphan list in the mean time.
2875 err
= ext4_orphan_add(handle
, inode
);
2876 ext4_journal_stop(handle
);
2878 goto err_drop_inode
;
2879 err
= __page_symlink(inode
, symname
, l
, 1);
2881 goto err_drop_inode
;
2883 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2884 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2886 handle
= ext4_journal_start(dir
,
2887 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2888 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
2889 if (IS_ERR(handle
)) {
2890 err
= PTR_ERR(handle
);
2891 goto err_drop_inode
;
2893 set_nlink(inode
, 1);
2894 err
= ext4_orphan_del(handle
, inode
);
2896 ext4_journal_stop(handle
);
2898 goto err_drop_inode
;
2901 /* clear the extent format for fast symlink */
2902 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
2903 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2904 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2905 inode
->i_size
= l
-1;
2907 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2908 err
= ext4_add_nondir(handle
, dentry
, inode
);
2910 ext4_journal_stop(handle
);
2911 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2915 unlock_new_inode(inode
);
2920 static int ext4_link(struct dentry
*old_dentry
,
2921 struct inode
*dir
, struct dentry
*dentry
)
2924 struct inode
*inode
= old_dentry
->d_inode
;
2925 int err
, retries
= 0;
2927 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
2930 dquot_initialize(dir
);
2933 handle
= ext4_journal_start(dir
, EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2934 EXT4_INDEX_EXTRA_TRANS_BLOCKS
);
2936 return PTR_ERR(handle
);
2938 if (IS_DIRSYNC(dir
))
2939 ext4_handle_sync(handle
);
2941 inode
->i_ctime
= ext4_current_time(inode
);
2942 ext4_inc_count(handle
, inode
);
2945 err
= ext4_add_entry(handle
, dentry
, inode
);
2947 ext4_mark_inode_dirty(handle
, inode
);
2948 d_instantiate(dentry
, inode
);
2953 ext4_journal_stop(handle
);
2954 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2959 #define PARENT_INO(buffer, size) \
2960 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2963 * Anybody can rename anything with this: the permission checks are left to the
2964 * higher-level routines.
2966 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2967 struct inode
*new_dir
, struct dentry
*new_dentry
)
2970 struct inode
*old_inode
, *new_inode
;
2971 struct buffer_head
*old_bh
, *new_bh
, *dir_bh
;
2972 struct ext4_dir_entry_2
*old_de
, *new_de
;
2973 int retval
, force_da_alloc
= 0;
2975 dquot_initialize(old_dir
);
2976 dquot_initialize(new_dir
);
2978 old_bh
= new_bh
= dir_bh
= NULL
;
2980 /* Initialize quotas before so that eventual writes go
2981 * in separate transaction */
2982 if (new_dentry
->d_inode
)
2983 dquot_initialize(new_dentry
->d_inode
);
2984 handle
= ext4_journal_start(old_dir
, 2 *
2985 EXT4_DATA_TRANS_BLOCKS(old_dir
->i_sb
) +
2986 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
2988 return PTR_ERR(handle
);
2990 if (IS_DIRSYNC(old_dir
) || IS_DIRSYNC(new_dir
))
2991 ext4_handle_sync(handle
);
2993 old_bh
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de
);
2995 * Check for inode number is _not_ due to possible IO errors.
2996 * We might rmdir the source, keep it as pwd of some process
2997 * and merrily kill the link to whatever was created under the
2998 * same name. Goodbye sticky bit ;-<
3000 old_inode
= old_dentry
->d_inode
;
3002 if (!old_bh
|| le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
)
3005 new_inode
= new_dentry
->d_inode
;
3006 new_bh
= ext4_find_entry(new_dir
, &new_dentry
->d_name
, &new_de
);
3013 if (S_ISDIR(old_inode
->i_mode
)) {
3015 retval
= -ENOTEMPTY
;
3016 if (!empty_dir(new_inode
))
3020 if (!(dir_bh
= ext4_bread(handle
, old_inode
, 0, 0, &retval
))) {
3023 ext4_error(old_inode
->i_sb
,
3024 "Directory hole detected on inode %lu\n",
3029 if (!buffer_verified(dir_bh
) &&
3030 !ext4_dirent_csum_verify(old_inode
,
3031 (struct ext4_dir_entry
*)dir_bh
->b_data
))
3033 set_buffer_verified(dir_bh
);
3034 if (le32_to_cpu(PARENT_INO(dir_bh
->b_data
,
3035 old_dir
->i_sb
->s_blocksize
)) != old_dir
->i_ino
)
3038 if (!new_inode
&& new_dir
!= old_dir
&&
3039 EXT4_DIR_LINK_MAX(new_dir
))
3041 BUFFER_TRACE(dir_bh
, "get_write_access");
3042 retval
= ext4_journal_get_write_access(handle
, dir_bh
);
3047 retval
= ext4_add_entry(handle
, new_dentry
, old_inode
);
3051 BUFFER_TRACE(new_bh
, "get write access");
3052 retval
= ext4_journal_get_write_access(handle
, new_bh
);
3055 new_de
->inode
= cpu_to_le32(old_inode
->i_ino
);
3056 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir
->i_sb
,
3057 EXT4_FEATURE_INCOMPAT_FILETYPE
))
3058 new_de
->file_type
= old_de
->file_type
;
3059 new_dir
->i_version
++;
3060 new_dir
->i_ctime
= new_dir
->i_mtime
=
3061 ext4_current_time(new_dir
);
3062 ext4_mark_inode_dirty(handle
, new_dir
);
3063 BUFFER_TRACE(new_bh
, "call ext4_handle_dirty_metadata");
3064 retval
= ext4_handle_dirty_dirent_node(handle
, new_dir
, new_bh
);
3065 if (unlikely(retval
)) {
3066 ext4_std_error(new_dir
->i_sb
, retval
);
3074 * Like most other Unix systems, set the ctime for inodes on a
3077 old_inode
->i_ctime
= ext4_current_time(old_inode
);
3078 ext4_mark_inode_dirty(handle
, old_inode
);
3083 if (le32_to_cpu(old_de
->inode
) != old_inode
->i_ino
||
3084 old_de
->name_len
!= old_dentry
->d_name
.len
||
3085 strncmp(old_de
->name
, old_dentry
->d_name
.name
, old_de
->name_len
) ||
3086 (retval
= ext4_delete_entry(handle
, old_dir
,
3087 old_de
, old_bh
)) == -ENOENT
) {
3088 /* old_de could have moved from under us during htree split, so
3089 * make sure that we are deleting the right entry. We might
3090 * also be pointing to a stale entry in the unused part of
3091 * old_bh so just checking inum and the name isn't enough. */
3092 struct buffer_head
*old_bh2
;
3093 struct ext4_dir_entry_2
*old_de2
;
3095 old_bh2
= ext4_find_entry(old_dir
, &old_dentry
->d_name
, &old_de2
);
3097 retval
= ext4_delete_entry(handle
, old_dir
,
3103 ext4_warning(old_dir
->i_sb
,
3104 "Deleting old file (%lu), %d, error=%d",
3105 old_dir
->i_ino
, old_dir
->i_nlink
, retval
);
3109 ext4_dec_count(handle
, new_inode
);
3110 new_inode
->i_ctime
= ext4_current_time(new_inode
);
3112 old_dir
->i_ctime
= old_dir
->i_mtime
= ext4_current_time(old_dir
);
3113 ext4_update_dx_flag(old_dir
);
3115 PARENT_INO(dir_bh
->b_data
, new_dir
->i_sb
->s_blocksize
) =
3116 cpu_to_le32(new_dir
->i_ino
);
3117 BUFFER_TRACE(dir_bh
, "call ext4_handle_dirty_metadata");
3118 if (is_dx(old_inode
)) {
3119 retval
= ext4_handle_dirty_dx_node(handle
,
3123 retval
= ext4_handle_dirty_dirent_node(handle
,
3128 ext4_std_error(old_dir
->i_sb
, retval
);
3131 ext4_dec_count(handle
, old_dir
);
3133 /* checked empty_dir above, can't have another parent,
3134 * ext4_dec_count() won't work for many-linked dirs */
3135 clear_nlink(new_inode
);
3137 ext4_inc_count(handle
, new_dir
);
3138 ext4_update_dx_flag(new_dir
);
3139 ext4_mark_inode_dirty(handle
, new_dir
);
3142 ext4_mark_inode_dirty(handle
, old_dir
);
3144 ext4_mark_inode_dirty(handle
, new_inode
);
3145 if (!new_inode
->i_nlink
)
3146 ext4_orphan_add(handle
, new_inode
);
3147 if (!test_opt(new_dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3156 ext4_journal_stop(handle
);
3157 if (retval
== 0 && force_da_alloc
)
3158 ext4_alloc_da_blocks(old_inode
);
3163 * directories can handle most operations...
3165 const struct inode_operations ext4_dir_inode_operations
= {
3166 .create
= ext4_create
,
3167 .lookup
= ext4_lookup
,
3169 .unlink
= ext4_unlink
,
3170 .symlink
= ext4_symlink
,
3171 .mkdir
= ext4_mkdir
,
3172 .rmdir
= ext4_rmdir
,
3173 .mknod
= ext4_mknod
,
3174 .rename
= ext4_rename
,
3175 .setattr
= ext4_setattr
,
3176 #ifdef CONFIG_EXT4_FS_XATTR
3177 .setxattr
= generic_setxattr
,
3178 .getxattr
= generic_getxattr
,
3179 .listxattr
= ext4_listxattr
,
3180 .removexattr
= generic_removexattr
,
3182 .get_acl
= ext4_get_acl
,
3183 .fiemap
= ext4_fiemap
,
3186 const struct inode_operations ext4_special_inode_operations
= {
3187 .setattr
= ext4_setattr
,
3188 #ifdef CONFIG_EXT4_FS_XATTR
3189 .setxattr
= generic_setxattr
,
3190 .getxattr
= generic_getxattr
,
3191 .listxattr
= ext4_listxattr
,
3192 .removexattr
= generic_removexattr
,
3194 .get_acl
= ext4_get_acl
,