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)
51 static struct buffer_head
*ext4_append(handle_t
*handle
,
55 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
)))
61 return ERR_PTR(-ENOSPC
);
63 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
65 bh
= ext4_bread(handle
, inode
, *block
, 1, &err
);
68 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
69 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
70 BUFFER_TRACE(bh
, "get_write_access");
71 err
= ext4_journal_get_write_access(handle
, bh
);
74 ext4_std_error(inode
->i_sb
, err
);
80 static int ext4_dx_csum_verify(struct inode
*inode
,
81 struct ext4_dir_entry
*dirent
);
87 #define ext4_read_dirblock(inode, block, type) \
88 __ext4_read_dirblock((inode), (block), (type), __LINE__)
90 static struct buffer_head
*__ext4_read_dirblock(struct inode
*inode
,
95 struct buffer_head
*bh
;
96 struct ext4_dir_entry
*dirent
;
97 int err
= 0, is_dx_block
= 0;
99 bh
= ext4_bread(NULL
, inode
, block
, 0, &err
);
102 ext4_error_inode(inode
, __func__
, line
, block
,
103 "Directory hole found");
104 return ERR_PTR(-EIO
);
106 __ext4_warning(inode
->i_sb
, __func__
, line
,
107 "error reading directory block "
108 "(ino %lu, block %lu)", inode
->i_ino
,
109 (unsigned long) block
);
112 dirent
= (struct ext4_dir_entry
*) bh
->b_data
;
113 /* Determine whether or not we have an index block */
117 else if (ext4_rec_len_from_disk(dirent
->rec_len
,
118 inode
->i_sb
->s_blocksize
) ==
119 inode
->i_sb
->s_blocksize
)
122 if (!is_dx_block
&& type
== INDEX
) {
123 ext4_error_inode(inode
, __func__
, line
, block
,
124 "directory leaf block found instead of index block");
125 return ERR_PTR(-EIO
);
127 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
128 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
) ||
133 * An empty leaf block can get mistaken for a index block; for
134 * this reason, we can only check the index checksum when the
135 * caller is sure it should be an index block.
137 if (is_dx_block
&& type
== INDEX
) {
138 if (ext4_dx_csum_verify(inode
, dirent
))
139 set_buffer_verified(bh
);
141 ext4_error_inode(inode
, __func__
, line
, block
,
142 "Directory index failed checksum");
144 return ERR_PTR(-EIO
);
148 if (ext4_dirent_csum_verify(inode
, dirent
))
149 set_buffer_verified(bh
);
151 ext4_error_inode(inode
, __func__
, line
, block
,
152 "Directory block failed checksum");
154 return ERR_PTR(-EIO
);
161 #define assert(test) J_ASSERT(test)
165 #define dxtrace(command) command
167 #define dxtrace(command)
191 * dx_root_info is laid out so that if it should somehow get overlaid by a
192 * dirent the two low bits of the hash version will be zero. Therefore, the
193 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
198 struct fake_dirent dot
;
200 struct fake_dirent dotdot
;
204 __le32 reserved_zero
;
206 u8 info_length
; /* 8 */
211 struct dx_entry entries
[0];
216 struct fake_dirent fake
;
217 struct dx_entry entries
[0];
223 struct buffer_head
*bh
;
224 struct dx_entry
*entries
;
236 * This goes at the end of each htree block.
240 __le32 dt_checksum
; /* crc32c(uuid+inum+dirblock) */
243 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
244 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
245 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
246 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
247 static unsigned dx_get_count(struct dx_entry
*entries
);
248 static unsigned dx_get_limit(struct dx_entry
*entries
);
249 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
250 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
251 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
252 static unsigned dx_node_limit(struct inode
*dir
);
253 static struct dx_frame
*dx_probe(const struct qstr
*d_name
,
255 struct dx_hash_info
*hinfo
,
256 struct dx_frame
*frame
,
258 static void dx_release(struct dx_frame
*frames
);
259 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
260 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
261 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
262 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
263 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
264 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
265 static void dx_insert_block(struct dx_frame
*frame
,
266 u32 hash
, ext4_lblk_t block
);
267 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
268 struct dx_frame
*frame
,
269 struct dx_frame
*frames
,
271 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
272 const struct qstr
*d_name
,
273 struct ext4_dir_entry_2
**res_dir
,
275 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
276 struct inode
*inode
);
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail
*t
,
280 unsigned int blocksize
)
282 memset(t
, 0, sizeof(struct ext4_dir_entry_tail
));
283 t
->det_rec_len
= ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail
), blocksize
);
285 t
->det_reserved_ft
= EXT4_FT_DIR_CSUM
;
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail
*get_dirent_tail(struct inode
*inode
,
290 struct ext4_dir_entry
*de
)
292 struct ext4_dir_entry_tail
*t
;
295 struct ext4_dir_entry
*d
, *top
;
298 top
= (struct ext4_dir_entry
*)(((void *)de
) +
299 (EXT4_BLOCK_SIZE(inode
->i_sb
) -
300 sizeof(struct ext4_dir_entry_tail
)));
301 while (d
< top
&& d
->rec_len
)
302 d
= (struct ext4_dir_entry
*)(((void *)d
) +
303 le16_to_cpu(d
->rec_len
));
308 t
= (struct ext4_dir_entry_tail
*)d
;
310 t
= EXT4_DIRENT_TAIL(de
, EXT4_BLOCK_SIZE(inode
->i_sb
));
313 if (t
->det_reserved_zero1
||
314 le16_to_cpu(t
->det_rec_len
) != sizeof(struct ext4_dir_entry_tail
) ||
315 t
->det_reserved_zero2
||
316 t
->det_reserved_ft
!= EXT4_FT_DIR_CSUM
)
322 static __le32
ext4_dirent_csum(struct inode
*inode
,
323 struct ext4_dir_entry
*dirent
, int size
)
325 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
326 struct ext4_inode_info
*ei
= EXT4_I(inode
);
329 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
330 return cpu_to_le32(csum
);
333 static void warn_no_space_for_csum(struct inode
*inode
)
335 ext4_warning(inode
->i_sb
, "no space in directory inode %lu leaf for "
336 "checksum. Please run e2fsck -D.", inode
->i_ino
);
339 int ext4_dirent_csum_verify(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
341 struct ext4_dir_entry_tail
*t
;
343 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
344 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
347 t
= get_dirent_tail(inode
, dirent
);
349 warn_no_space_for_csum(inode
);
353 if (t
->det_checksum
!= ext4_dirent_csum(inode
, dirent
,
354 (void *)t
- (void *)dirent
))
360 static void ext4_dirent_csum_set(struct inode
*inode
,
361 struct ext4_dir_entry
*dirent
)
363 struct ext4_dir_entry_tail
*t
;
365 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
366 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
369 t
= get_dirent_tail(inode
, dirent
);
371 warn_no_space_for_csum(inode
);
375 t
->det_checksum
= ext4_dirent_csum(inode
, dirent
,
376 (void *)t
- (void *)dirent
);
379 int ext4_handle_dirty_dirent_node(handle_t
*handle
,
381 struct buffer_head
*bh
)
383 ext4_dirent_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
384 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
387 static struct dx_countlimit
*get_dx_countlimit(struct inode
*inode
,
388 struct ext4_dir_entry
*dirent
,
391 struct ext4_dir_entry
*dp
;
392 struct dx_root_info
*root
;
395 if (le16_to_cpu(dirent
->rec_len
) == EXT4_BLOCK_SIZE(inode
->i_sb
))
397 else if (le16_to_cpu(dirent
->rec_len
) == 12) {
398 dp
= (struct ext4_dir_entry
*)(((void *)dirent
) + 12);
399 if (le16_to_cpu(dp
->rec_len
) !=
400 EXT4_BLOCK_SIZE(inode
->i_sb
) - 12)
402 root
= (struct dx_root_info
*)(((void *)dp
+ 12));
403 if (root
->reserved_zero
||
404 root
->info_length
!= sizeof(struct dx_root_info
))
411 *offset
= count_offset
;
412 return (struct dx_countlimit
*)(((void *)dirent
) + count_offset
);
415 static __le32
ext4_dx_csum(struct inode
*inode
, struct ext4_dir_entry
*dirent
,
416 int count_offset
, int count
, struct dx_tail
*t
)
418 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
419 struct ext4_inode_info
*ei
= EXT4_I(inode
);
424 size
= count_offset
+ (count
* sizeof(struct dx_entry
));
425 save_csum
= t
->dt_checksum
;
427 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
428 csum
= ext4_chksum(sbi
, csum
, (__u8
*)t
, sizeof(struct dx_tail
));
429 t
->dt_checksum
= save_csum
;
431 return cpu_to_le32(csum
);
434 static int ext4_dx_csum_verify(struct inode
*inode
,
435 struct ext4_dir_entry
*dirent
)
437 struct dx_countlimit
*c
;
439 int count_offset
, limit
, count
;
441 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
442 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
445 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
447 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
450 limit
= le16_to_cpu(c
->limit
);
451 count
= le16_to_cpu(c
->count
);
452 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
453 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
454 warn_no_space_for_csum(inode
);
457 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
459 if (t
->dt_checksum
!= ext4_dx_csum(inode
, dirent
, count_offset
,
465 static void ext4_dx_csum_set(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
467 struct dx_countlimit
*c
;
469 int count_offset
, limit
, count
;
471 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
472 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
475 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
477 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
480 limit
= le16_to_cpu(c
->limit
);
481 count
= le16_to_cpu(c
->count
);
482 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
483 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
484 warn_no_space_for_csum(inode
);
487 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
489 t
->dt_checksum
= ext4_dx_csum(inode
, dirent
, count_offset
, count
, t
);
492 static inline int ext4_handle_dirty_dx_node(handle_t
*handle
,
494 struct buffer_head
*bh
)
496 ext4_dx_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
497 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
501 * p is at least 6 bytes before the end of page
503 static inline struct ext4_dir_entry_2
*
504 ext4_next_entry(struct ext4_dir_entry_2
*p
, unsigned long blocksize
)
506 return (struct ext4_dir_entry_2
*)((char *)p
+
507 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
511 * Future: use high four bits of block for coalesce-on-delete flags
512 * Mask them off for now.
515 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
517 return le32_to_cpu(entry
->block
) & 0x00ffffff;
520 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
522 entry
->block
= cpu_to_le32(value
);
525 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
527 return le32_to_cpu(entry
->hash
);
530 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
532 entry
->hash
= cpu_to_le32(value
);
535 static inline unsigned dx_get_count(struct dx_entry
*entries
)
537 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
540 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
542 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
545 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
547 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
550 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
552 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
555 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
557 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
558 EXT4_DIR_REC_LEN(2) - infosize
;
560 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
561 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
562 entry_space
-= sizeof(struct dx_tail
);
563 return entry_space
/ sizeof(struct dx_entry
);
566 static inline unsigned dx_node_limit(struct inode
*dir
)
568 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
570 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
571 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
572 entry_space
-= sizeof(struct dx_tail
);
573 return entry_space
/ sizeof(struct dx_entry
);
580 static void dx_show_index(char * label
, struct dx_entry
*entries
)
582 int i
, n
= dx_get_count (entries
);
583 printk(KERN_DEBUG
"%s index ", label
);
584 for (i
= 0; i
< n
; i
++) {
585 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
586 0, (unsigned long)dx_get_block(entries
+ i
));
598 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
599 int size
, int show_names
)
601 unsigned names
= 0, space
= 0;
602 char *base
= (char *) de
;
603 struct dx_hash_info h
= *hinfo
;
606 while ((char *) de
< base
+ size
)
612 int len
= de
->name_len
;
613 char *name
= de
->name
;
614 while (len
--) printk("%c", *name
++);
615 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
616 printk(":%x.%u ", h
.hash
,
617 (unsigned) ((char *) de
- base
));
619 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
622 de
= ext4_next_entry(de
, size
);
624 printk("(%i)\n", names
);
625 return (struct stats
) { names
, space
, 1 };
628 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
629 struct dx_entry
*entries
, int levels
)
631 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
632 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
634 struct buffer_head
*bh
;
636 printk("%i indexed blocks...\n", count
);
637 for (i
= 0; i
< count
; i
++, entries
++)
639 ext4_lblk_t block
= dx_get_block(entries
);
640 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
641 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
643 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
644 if (!(bh
= ext4_bread (NULL
,dir
, block
, 0,&err
))) continue;
646 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
647 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
648 names
+= stats
.names
;
649 space
+= stats
.space
;
650 bcount
+= stats
.bcount
;
654 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
655 levels
? "" : " ", names
, space
/bcount
,
656 (space
/bcount
)*100/blocksize
);
657 return (struct stats
) { names
, space
, bcount
};
659 #endif /* DX_DEBUG */
662 * Probe for a directory leaf block to search.
664 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
665 * error in the directory index, and the caller should fall back to
666 * searching the directory normally. The callers of dx_probe **MUST**
667 * check for this error code, and make sure it never gets reflected
670 static struct dx_frame
*
671 dx_probe(const struct qstr
*d_name
, struct inode
*dir
,
672 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
, int *err
)
674 unsigned count
, indirect
;
675 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
676 struct dx_root
*root
;
677 struct buffer_head
*bh
;
678 struct dx_frame
*frame
= frame_in
;
682 bh
= ext4_read_dirblock(dir
, 0, INDEX
);
687 root
= (struct dx_root
*) bh
->b_data
;
688 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
689 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
690 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
691 ext4_warning(dir
->i_sb
, "Unrecognised inode hash code %d",
692 root
->info
.hash_version
);
694 *err
= ERR_BAD_DX_DIR
;
697 hinfo
->hash_version
= root
->info
.hash_version
;
698 if (hinfo
->hash_version
<= DX_HASH_TEA
)
699 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
700 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
702 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
705 if (root
->info
.unused_flags
& 1) {
706 ext4_warning(dir
->i_sb
, "Unimplemented inode hash flags: %#06x",
707 root
->info
.unused_flags
);
709 *err
= ERR_BAD_DX_DIR
;
713 if ((indirect
= root
->info
.indirect_levels
) > 1) {
714 ext4_warning(dir
->i_sb
, "Unimplemented inode hash depth: %#06x",
715 root
->info
.indirect_levels
);
717 *err
= ERR_BAD_DX_DIR
;
721 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
722 root
->info
.info_length
);
724 if (dx_get_limit(entries
) != dx_root_limit(dir
,
725 root
->info
.info_length
)) {
726 ext4_warning(dir
->i_sb
, "dx entry: limit != root limit");
728 *err
= ERR_BAD_DX_DIR
;
732 dxtrace(printk("Look up %x", hash
));
735 count
= dx_get_count(entries
);
736 if (!count
|| count
> dx_get_limit(entries
)) {
737 ext4_warning(dir
->i_sb
,
738 "dx entry: no count or count > limit");
740 *err
= ERR_BAD_DX_DIR
;
745 q
= entries
+ count
- 1;
749 dxtrace(printk("."));
750 if (dx_get_hash(m
) > hash
)
756 if (0) // linear search cross check
758 unsigned n
= count
- 1;
762 dxtrace(printk(","));
763 if (dx_get_hash(++at
) > hash
)
769 assert (at
== p
- 1);
773 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
775 frame
->entries
= entries
;
777 if (!indirect
--) return frame
;
778 bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
783 entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
785 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
786 ext4_warning(dir
->i_sb
,
787 "dx entry: limit != node limit");
789 *err
= ERR_BAD_DX_DIR
;
796 while (frame
>= frame_in
) {
801 if (*err
== ERR_BAD_DX_DIR
)
802 ext4_warning(dir
->i_sb
,
803 "Corrupt dir inode %lu, running e2fsck is "
804 "recommended.", dir
->i_ino
);
808 static void dx_release (struct dx_frame
*frames
)
810 if (frames
[0].bh
== NULL
)
813 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
814 brelse(frames
[1].bh
);
815 brelse(frames
[0].bh
);
819 * This function increments the frame pointer to search the next leaf
820 * block, and reads in the necessary intervening nodes if the search
821 * should be necessary. Whether or not the search is necessary is
822 * controlled by the hash parameter. If the hash value is even, then
823 * the search is only continued if the next block starts with that
824 * hash value. This is used if we are searching for a specific file.
826 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
828 * This function returns 1 if the caller should continue to search,
829 * or 0 if it should not. If there is an error reading one of the
830 * index blocks, it will a negative error code.
832 * If start_hash is non-null, it will be filled in with the starting
833 * hash of the next page.
835 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
836 struct dx_frame
*frame
,
837 struct dx_frame
*frames
,
841 struct buffer_head
*bh
;
847 * Find the next leaf page by incrementing the frame pointer.
848 * If we run out of entries in the interior node, loop around and
849 * increment pointer in the parent node. When we break out of
850 * this loop, num_frames indicates the number of interior
851 * nodes need to be read.
854 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
863 * If the hash is 1, then continue only if the next page has a
864 * continuation hash of any value. This is used for readdir
865 * handling. Otherwise, check to see if the hash matches the
866 * desired contiuation hash. If it doesn't, return since
867 * there's no point to read in the successive index pages.
869 bhash
= dx_get_hash(p
->at
);
872 if ((hash
& 1) == 0) {
873 if ((bhash
& ~1) != hash
)
877 * If the hash is HASH_NB_ALWAYS, we always go to the next
878 * block so no check is necessary
880 while (num_frames
--) {
881 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
887 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
894 * This function fills a red-black tree with information from a
895 * directory block. It returns the number directory entries loaded
896 * into the tree. If there is an error it is returned in err.
898 static int htree_dirblock_to_tree(struct file
*dir_file
,
899 struct inode
*dir
, ext4_lblk_t block
,
900 struct dx_hash_info
*hinfo
,
901 __u32 start_hash
, __u32 start_minor_hash
)
903 struct buffer_head
*bh
;
904 struct ext4_dir_entry_2
*de
, *top
;
905 int err
= 0, count
= 0;
907 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
908 (unsigned long)block
));
909 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
913 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
914 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
915 dir
->i_sb
->s_blocksize
-
916 EXT4_DIR_REC_LEN(0));
917 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
918 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
919 bh
->b_data
, bh
->b_size
,
920 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
921 + ((char *)de
- bh
->b_data
))) {
922 /* silently ignore the rest of the block */
925 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
926 if ((hinfo
->hash
< start_hash
) ||
927 ((hinfo
->hash
== start_hash
) &&
928 (hinfo
->minor_hash
< start_minor_hash
)))
932 if ((err
= ext4_htree_store_dirent(dir_file
,
933 hinfo
->hash
, hinfo
->minor_hash
, de
)) != 0) {
945 * This function fills a red-black tree with information from a
946 * directory. We start scanning the directory in hash order, starting
947 * at start_hash and start_minor_hash.
949 * This function returns the number of entries inserted into the tree,
950 * or a negative error code.
952 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
953 __u32 start_minor_hash
, __u32
*next_hash
)
955 struct dx_hash_info hinfo
;
956 struct ext4_dir_entry_2
*de
;
957 struct dx_frame frames
[2], *frame
;
964 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
965 start_hash
, start_minor_hash
));
966 dir
= file_inode(dir_file
);
967 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
968 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
969 if (hinfo
.hash_version
<= DX_HASH_TEA
)
970 hinfo
.hash_version
+=
971 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
972 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
973 if (ext4_has_inline_data(dir
)) {
974 int has_inline_data
= 1;
975 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
979 if (has_inline_data
) {
984 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
985 start_hash
, start_minor_hash
);
989 hinfo
.hash
= start_hash
;
990 hinfo
.minor_hash
= 0;
991 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
, &err
);
995 /* Add '.' and '..' from the htree header */
996 if (!start_hash
&& !start_minor_hash
) {
997 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
998 if ((err
= ext4_htree_store_dirent(dir_file
, 0, 0, de
)) != 0)
1002 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
1003 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
1004 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
1005 if ((err
= ext4_htree_store_dirent(dir_file
, 2, 0, de
)) != 0)
1011 block
= dx_get_block(frame
->at
);
1012 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
1013 start_hash
, start_minor_hash
);
1020 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
1021 frame
, frames
, &hashval
);
1022 *next_hash
= hashval
;
1028 * Stop if: (a) there are no more entries, or
1029 * (b) we have inserted at least one entry and the
1030 * next hash value is not a continuation
1033 (count
&& ((hashval
& 1) == 0)))
1037 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1038 "next hash: %x\n", count
, *next_hash
));
1045 static inline int search_dirblock(struct buffer_head
*bh
,
1047 const struct qstr
*d_name
,
1048 unsigned int offset
,
1049 struct ext4_dir_entry_2
**res_dir
)
1051 return search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1052 d_name
, offset
, res_dir
);
1056 * Directory block splitting, compacting
1060 * Create map of hash values, offsets, and sizes, stored at end of block.
1061 * Returns number of entries mapped.
1063 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
1064 struct dx_hash_info
*hinfo
,
1065 struct dx_map_entry
*map_tail
)
1068 char *base
= (char *) de
;
1069 struct dx_hash_info h
= *hinfo
;
1071 while ((char *) de
< base
+ blocksize
) {
1072 if (de
->name_len
&& de
->inode
) {
1073 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1075 map_tail
->hash
= h
.hash
;
1076 map_tail
->offs
= ((char *) de
- base
)>>2;
1077 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1081 /* XXX: do we need to check rec_len == 0 case? -Chris */
1082 de
= ext4_next_entry(de
, blocksize
);
1087 /* Sort map by hash value */
1088 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1090 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1092 /* Combsort until bubble sort doesn't suck */
1094 count
= count
*10/13;
1095 if (count
- 9 < 2) /* 9, 10 -> 11 */
1097 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1098 if (p
->hash
< q
->hash
)
1101 /* Garden variety bubble sort */
1106 if (q
[1].hash
>= q
[0].hash
)
1114 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1116 struct dx_entry
*entries
= frame
->entries
;
1117 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1118 int count
= dx_get_count(entries
);
1120 assert(count
< dx_get_limit(entries
));
1121 assert(old
< entries
+ count
);
1122 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1123 dx_set_hash(new, hash
);
1124 dx_set_block(new, block
);
1125 dx_set_count(entries
, count
+ 1);
1129 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1131 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1132 * `de != NULL' is guaranteed by caller.
1134 static inline int ext4_match (int len
, const char * const name
,
1135 struct ext4_dir_entry_2
* de
)
1137 if (len
!= de
->name_len
)
1141 return !memcmp(name
, de
->name
, len
);
1145 * Returns 0 if not found, -1 on failure, and 1 on success
1147 int search_dir(struct buffer_head
*bh
,
1151 const struct qstr
*d_name
,
1152 unsigned int offset
,
1153 struct ext4_dir_entry_2
**res_dir
)
1155 struct ext4_dir_entry_2
* de
;
1158 const char *name
= d_name
->name
;
1159 int namelen
= d_name
->len
;
1161 de
= (struct ext4_dir_entry_2
*)search_buf
;
1162 dlimit
= search_buf
+ buf_size
;
1163 while ((char *) de
< dlimit
) {
1164 /* this code is executed quadratically often */
1165 /* do minimal checking `by hand' */
1167 if ((char *) de
+ namelen
<= dlimit
&&
1168 ext4_match (namelen
, name
, de
)) {
1169 /* found a match - just to be sure, do a full check */
1170 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, bh
->b_data
,
1171 bh
->b_size
, offset
))
1176 /* prevent looping on a bad block */
1177 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1178 dir
->i_sb
->s_blocksize
);
1182 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1187 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1188 struct ext4_dir_entry
*de
)
1190 struct super_block
*sb
= dir
->i_sb
;
1196 if (de
->inode
== 0 &&
1197 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1206 * finds an entry in the specified directory with the wanted name. It
1207 * returns the cache buffer in which the entry was found, and the entry
1208 * itself (as a parameter - res_dir). It does NOT read the inode of the
1209 * entry - you'll have to do that yourself if you want to.
1211 * The returned buffer_head has ->b_count elevated. The caller is expected
1212 * to brelse() it when appropriate.
1214 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1215 const struct qstr
*d_name
,
1216 struct ext4_dir_entry_2
**res_dir
,
1219 struct super_block
*sb
;
1220 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1221 struct buffer_head
*bh
, *ret
= NULL
;
1222 ext4_lblk_t start
, block
, b
;
1223 const u8
*name
= d_name
->name
;
1224 int ra_max
= 0; /* Number of bh's in the readahead
1226 int ra_ptr
= 0; /* Current index into readahead
1229 ext4_lblk_t nblocks
;
1235 namelen
= d_name
->len
;
1236 if (namelen
> EXT4_NAME_LEN
)
1239 if (ext4_has_inline_data(dir
)) {
1240 int has_inline_data
= 1;
1241 ret
= ext4_find_inline_entry(dir
, d_name
, res_dir
,
1243 if (has_inline_data
) {
1250 if ((namelen
<= 2) && (name
[0] == '.') &&
1251 (name
[1] == '.' || name
[1] == '\0')) {
1253 * "." or ".." will only be in the first block
1254 * NFS may look up ".."; "." should be handled by the VFS
1261 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
, &err
);
1263 * On success, or if the error was file not found,
1264 * return. Otherwise, fall back to doing a search the
1265 * old fashioned way.
1269 if (err
&& err
!= ERR_BAD_DX_DIR
)
1270 return ERR_PTR(err
);
1273 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1276 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1277 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1278 if (start
>= nblocks
)
1284 * We deal with the read-ahead logic here.
1286 if (ra_ptr
>= ra_max
) {
1287 /* Refill the readahead buffer */
1290 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1292 * Terminate if we reach the end of the
1293 * directory and must wrap, or if our
1294 * search has finished at this block.
1296 if (b
>= nblocks
|| (num
&& block
== start
)) {
1297 bh_use
[ra_max
] = NULL
;
1301 bh
= ext4_getblk(NULL
, dir
, b
++, 0, &err
);
1302 if (unlikely(err
)) {
1304 return ERR_PTR(err
);
1307 bh_use
[ra_max
] = bh
;
1309 ll_rw_block(READ
| REQ_META
| REQ_PRIO
,
1313 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1316 if (!buffer_uptodate(bh
)) {
1317 /* read error, skip block & hope for the best */
1318 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1319 (unsigned long) block
);
1323 if (!buffer_verified(bh
) &&
1324 !is_dx_internal_node(dir
, block
,
1325 (struct ext4_dir_entry
*)bh
->b_data
) &&
1326 !ext4_dirent_csum_verify(dir
,
1327 (struct ext4_dir_entry
*)bh
->b_data
)) {
1328 EXT4_ERROR_INODE(dir
, "checksumming directory "
1329 "block %lu", (unsigned long)block
);
1333 set_buffer_verified(bh
);
1334 i
= search_dirblock(bh
, dir
, d_name
,
1335 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1337 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1339 goto cleanup_and_exit
;
1343 goto cleanup_and_exit
;
1346 if (++block
>= nblocks
)
1348 } while (block
!= start
);
1351 * If the directory has grown while we were searching, then
1352 * search the last part of the directory before giving up.
1355 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1356 if (block
< nblocks
) {
1362 /* Clean up the read-ahead blocks */
1363 for (; ra_ptr
< ra_max
; ra_ptr
++)
1364 brelse(bh_use
[ra_ptr
]);
1368 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
1369 struct ext4_dir_entry_2
**res_dir
, int *err
)
1371 struct super_block
* sb
= dir
->i_sb
;
1372 struct dx_hash_info hinfo
;
1373 struct dx_frame frames
[2], *frame
;
1374 struct buffer_head
*bh
;
1378 if (!(frame
= dx_probe(d_name
, dir
, &hinfo
, frames
, err
)))
1381 block
= dx_get_block(frame
->at
);
1382 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1387 retval
= search_dirblock(bh
, dir
, d_name
,
1388 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1390 if (retval
== 1) { /* Success! */
1396 *err
= ERR_BAD_DX_DIR
;
1400 /* Check to see if we should continue to search */
1401 retval
= ext4_htree_next_block(dir
, hinfo
.hash
, frame
,
1405 "error reading index page in directory #%lu",
1410 } while (retval
== 1);
1414 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1415 dx_release (frames
);
1419 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1421 struct inode
*inode
;
1422 struct ext4_dir_entry_2
*de
;
1423 struct buffer_head
*bh
;
1425 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1426 return ERR_PTR(-ENAMETOOLONG
);
1428 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1430 return (struct dentry
*) bh
;
1433 __u32 ino
= le32_to_cpu(de
->inode
);
1435 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1436 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1437 return ERR_PTR(-EIO
);
1439 if (unlikely(ino
== dir
->i_ino
)) {
1440 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1442 return ERR_PTR(-EIO
);
1444 inode
= ext4_iget(dir
->i_sb
, ino
);
1445 if (inode
== ERR_PTR(-ESTALE
)) {
1446 EXT4_ERROR_INODE(dir
,
1447 "deleted inode referenced: %u",
1449 return ERR_PTR(-EIO
);
1452 return d_splice_alias(inode
, dentry
);
1456 struct dentry
*ext4_get_parent(struct dentry
*child
)
1459 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1460 struct ext4_dir_entry_2
* de
;
1461 struct buffer_head
*bh
;
1463 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
, NULL
);
1465 return (struct dentry
*) bh
;
1467 return ERR_PTR(-ENOENT
);
1468 ino
= le32_to_cpu(de
->inode
);
1471 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1472 EXT4_ERROR_INODE(child
->d_inode
,
1473 "bad parent inode number: %u", ino
);
1474 return ERR_PTR(-EIO
);
1477 return d_obtain_alias(ext4_iget(child
->d_inode
->i_sb
, ino
));
1481 * Move count entries from end of map between two memory locations.
1482 * Returns pointer to last entry moved.
1484 static struct ext4_dir_entry_2
*
1485 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1488 unsigned rec_len
= 0;
1491 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1492 (from
+ (map
->offs
<<2));
1493 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1494 memcpy (to
, de
, rec_len
);
1495 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1496 ext4_rec_len_to_disk(rec_len
, blocksize
);
1501 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1505 * Compact each dir entry in the range to the minimal rec_len.
1506 * Returns pointer to last entry in range.
1508 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1510 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1511 unsigned rec_len
= 0;
1514 while ((char*)de
< base
+ blocksize
) {
1515 next
= ext4_next_entry(de
, blocksize
);
1516 if (de
->inode
&& de
->name_len
) {
1517 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1519 memmove(to
, de
, rec_len
);
1520 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1522 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1530 * Split a full leaf block to make room for a new dir entry.
1531 * Allocate a new block, and move entries so that they are approx. equally full.
1532 * Returns pointer to de in block into which the new entry will be inserted.
1534 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1535 struct buffer_head
**bh
,struct dx_frame
*frame
,
1536 struct dx_hash_info
*hinfo
, int *error
)
1538 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1539 unsigned count
, continued
;
1540 struct buffer_head
*bh2
;
1541 ext4_lblk_t newblock
;
1543 struct dx_map_entry
*map
;
1544 char *data1
= (*bh
)->b_data
, *data2
;
1545 unsigned split
, move
, size
;
1546 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1547 struct ext4_dir_entry_tail
*t
;
1551 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
1552 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1553 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1555 bh2
= ext4_append(handle
, dir
, &newblock
);
1559 *error
= PTR_ERR(bh2
);
1563 BUFFER_TRACE(*bh
, "get_write_access");
1564 err
= ext4_journal_get_write_access(handle
, *bh
);
1568 BUFFER_TRACE(frame
->bh
, "get_write_access");
1569 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1573 data2
= bh2
->b_data
;
1575 /* create map in the end of data2 block */
1576 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1577 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1578 blocksize
, hinfo
, map
);
1580 dx_sort_map(map
, count
);
1581 /* Split the existing block in the middle, size-wise */
1584 for (i
= count
-1; i
>= 0; i
--) {
1585 /* is more than half of this entry in 2nd half of the block? */
1586 if (size
+ map
[i
].size
/2 > blocksize
/2)
1588 size
+= map
[i
].size
;
1591 /* map index at which we will split */
1592 split
= count
- move
;
1593 hash2
= map
[split
].hash
;
1594 continued
= hash2
== map
[split
- 1].hash
;
1595 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1596 (unsigned long)dx_get_block(frame
->at
),
1597 hash2
, split
, count
-split
));
1599 /* Fancy dance to stay within two buffers */
1600 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
, blocksize
);
1601 de
= dx_pack_dirents(data1
, blocksize
);
1602 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1605 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1609 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1610 initialize_dirent_tail(t
, blocksize
);
1612 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1613 initialize_dirent_tail(t
, blocksize
);
1616 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1617 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1619 /* Which block gets the new entry? */
1620 if (hinfo
->hash
>= hash2
)
1625 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1626 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1629 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1633 dxtrace(dx_show_index("frame", frame
->entries
));
1640 ext4_std_error(dir
->i_sb
, err
);
1645 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1646 struct buffer_head
*bh
,
1647 void *buf
, int buf_size
,
1648 const char *name
, int namelen
,
1649 struct ext4_dir_entry_2
**dest_de
)
1651 struct ext4_dir_entry_2
*de
;
1652 unsigned short reclen
= EXT4_DIR_REC_LEN(namelen
);
1654 unsigned int offset
= 0;
1657 de
= (struct ext4_dir_entry_2
*)buf
;
1658 top
= buf
+ buf_size
- reclen
;
1659 while ((char *) de
<= top
) {
1660 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1661 buf
, buf_size
, offset
))
1663 if (ext4_match(namelen
, name
, de
))
1665 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1666 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1667 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1669 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1672 if ((char *) de
> top
)
1679 void ext4_insert_dentry(struct inode
*inode
,
1680 struct ext4_dir_entry_2
*de
,
1682 const char *name
, int namelen
)
1687 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1688 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1690 struct ext4_dir_entry_2
*de1
=
1691 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1692 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1693 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1696 de
->file_type
= EXT4_FT_UNKNOWN
;
1697 de
->inode
= cpu_to_le32(inode
->i_ino
);
1698 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1699 de
->name_len
= namelen
;
1700 memcpy(de
->name
, name
, namelen
);
1703 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1704 * it points to a directory entry which is guaranteed to be large
1705 * enough for new directory entry. If de is NULL, then
1706 * add_dirent_to_buf will attempt search the directory block for
1707 * space. It will return -ENOSPC if no space is available, and -EIO
1708 * and -EEXIST if directory entry already exists.
1710 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1711 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1712 struct buffer_head
*bh
)
1714 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1715 const char *name
= dentry
->d_name
.name
;
1716 int namelen
= dentry
->d_name
.len
;
1717 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1721 if (EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
1722 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1723 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1726 err
= ext4_find_dest_de(dir
, inode
,
1727 bh
, bh
->b_data
, blocksize
- csum_size
,
1728 name
, namelen
, &de
);
1732 BUFFER_TRACE(bh
, "get_write_access");
1733 err
= ext4_journal_get_write_access(handle
, bh
);
1735 ext4_std_error(dir
->i_sb
, err
);
1739 /* By now the buffer is marked for journaling */
1740 ext4_insert_dentry(inode
, de
, blocksize
, name
, namelen
);
1743 * XXX shouldn't update any times until successful
1744 * completion of syscall, but too many callers depend
1747 * XXX similarly, too many callers depend on
1748 * ext4_new_inode() setting the times, but error
1749 * recovery deletes the inode, so the worst that can
1750 * happen is that the times are slightly out of date
1751 * and/or different from the directory change time.
1753 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1754 ext4_update_dx_flag(dir
);
1756 ext4_mark_inode_dirty(handle
, dir
);
1757 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1758 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1760 ext4_std_error(dir
->i_sb
, err
);
1765 * This converts a one block unindexed directory to a 3 block indexed
1766 * directory, and adds the dentry to the indexed directory.
1768 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1769 struct inode
*inode
, struct buffer_head
*bh
)
1771 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1772 const char *name
= dentry
->d_name
.name
;
1773 int namelen
= dentry
->d_name
.len
;
1774 struct buffer_head
*bh2
;
1775 struct dx_root
*root
;
1776 struct dx_frame frames
[2], *frame
;
1777 struct dx_entry
*entries
;
1778 struct ext4_dir_entry_2
*de
, *de2
;
1779 struct ext4_dir_entry_tail
*t
;
1784 struct dx_hash_info hinfo
;
1786 struct fake_dirent
*fde
;
1789 if (EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
1790 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1791 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1793 blocksize
= dir
->i_sb
->s_blocksize
;
1794 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1795 BUFFER_TRACE(bh
, "get_write_access");
1796 retval
= ext4_journal_get_write_access(handle
, bh
);
1798 ext4_std_error(dir
->i_sb
, retval
);
1802 root
= (struct dx_root
*) bh
->b_data
;
1804 /* The 0th block becomes the root, move the dirents out */
1805 fde
= &root
->dotdot
;
1806 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1807 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1808 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1809 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1813 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1815 /* Allocate new block for the 0th block's dirents */
1816 bh2
= ext4_append(handle
, dir
, &block
);
1819 return PTR_ERR(bh2
);
1821 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1822 data1
= bh2
->b_data
;
1824 memcpy (data1
, de
, len
);
1825 de
= (struct ext4_dir_entry_2
*) data1
;
1827 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1829 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1834 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1835 initialize_dirent_tail(t
, blocksize
);
1838 /* Initialize the root; the dot dirents already exist */
1839 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1840 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1842 memset (&root
->info
, 0, sizeof(root
->info
));
1843 root
->info
.info_length
= sizeof(root
->info
);
1844 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1845 entries
= root
->entries
;
1846 dx_set_block(entries
, 1);
1847 dx_set_count(entries
, 1);
1848 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1850 /* Initialize as for dx_probe */
1851 hinfo
.hash_version
= root
->info
.hash_version
;
1852 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1853 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1854 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1855 ext4fs_dirhash(name
, namelen
, &hinfo
);
1857 frame
->entries
= entries
;
1858 frame
->at
= entries
;
1862 ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1863 ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1865 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
, &retval
);
1868 * Even if the block split failed, we have to properly write
1869 * out all the changes we did so far. Otherwise we can end up
1870 * with corrupted filesystem.
1872 ext4_mark_inode_dirty(handle
, dir
);
1878 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1886 * adds a file entry to the specified directory, using the same
1887 * semantics as ext4_find_entry(). It returns NULL if it failed.
1889 * NOTE!! The inode part of 'de' is left at 0 - which means you
1890 * may not sleep between calling this and putting something into
1891 * the entry, as someone else might have used it while you slept.
1893 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1894 struct inode
*inode
)
1896 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1897 struct buffer_head
*bh
;
1898 struct ext4_dir_entry_2
*de
;
1899 struct ext4_dir_entry_tail
*t
;
1900 struct super_block
*sb
;
1904 ext4_lblk_t block
, blocks
;
1907 if (EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
1908 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1909 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1912 blocksize
= sb
->s_blocksize
;
1913 if (!dentry
->d_name
.len
)
1916 if (ext4_has_inline_data(dir
)) {
1917 retval
= ext4_try_add_inline_entry(handle
, dentry
, inode
);
1927 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1928 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1930 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
1932 ext4_mark_inode_dirty(handle
, dir
);
1934 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1935 for (block
= 0; block
< blocks
; block
++) {
1936 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1940 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1941 if (retval
!= -ENOSPC
) {
1946 if (blocks
== 1 && !dx_fallback
&&
1947 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
))
1948 return make_indexed_dir(handle
, dentry
, inode
, bh
);
1951 bh
= ext4_append(handle
, dir
, &block
);
1954 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
1956 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
1959 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
1960 initialize_dirent_tail(t
, blocksize
);
1963 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1966 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
1971 * Returns 0 for success, or a negative error value
1973 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1974 struct inode
*inode
)
1976 struct dx_frame frames
[2], *frame
;
1977 struct dx_entry
*entries
, *at
;
1978 struct dx_hash_info hinfo
;
1979 struct buffer_head
*bh
;
1980 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1981 struct super_block
*sb
= dir
->i_sb
;
1982 struct ext4_dir_entry_2
*de
;
1985 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
, &err
);
1988 entries
= frame
->entries
;
1990 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
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
);
2029 node2
= (struct dx_node
*)(bh2
->b_data
);
2030 entries2
= node2
->entries
;
2031 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2032 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2034 BUFFER_TRACE(frame
->bh
, "get_write_access");
2035 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2039 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2040 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2041 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2044 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2045 err
= ext4_journal_get_write_access(handle
,
2050 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2051 icount2
* sizeof(struct dx_entry
));
2052 dx_set_count(entries
, icount1
);
2053 dx_set_count(entries2
, icount2
);
2054 dx_set_limit(entries2
, dx_node_limit(dir
));
2056 /* Which index block gets the new entry? */
2057 if (at
- entries
>= icount1
) {
2058 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2059 frame
->entries
= entries
= entries2
;
2060 swap(frame
->bh
, bh2
);
2062 dx_insert_block(frames
+ 0, hash2
, newblock
);
2063 dxtrace(dx_show_index("node", frames
[1].entries
));
2064 dxtrace(dx_show_index("node",
2065 ((struct dx_node
*) bh2
->b_data
)->entries
));
2066 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2071 dxtrace(printk(KERN_DEBUG
2072 "Creating second level index...\n"));
2073 memcpy((char *) entries2
, (char *) entries
,
2074 icount
* sizeof(struct dx_entry
));
2075 dx_set_limit(entries2
, dx_node_limit(dir
));
2078 dx_set_count(entries
, 1);
2079 dx_set_block(entries
+ 0, newblock
);
2080 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2082 /* Add new access path frame */
2084 frame
->at
= at
= at
- entries
+ entries2
;
2085 frame
->entries
= entries
= entries2
;
2087 err
= ext4_journal_get_write_access(handle
,
2092 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2094 ext4_std_error(inode
->i_sb
, err
);
2098 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
, &err
);
2101 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
2105 ext4_std_error(dir
->i_sb
, err
);
2113 * ext4_generic_delete_entry deletes a directory entry by merging it
2114 * with the previous entry
2116 int ext4_generic_delete_entry(handle_t
*handle
,
2118 struct ext4_dir_entry_2
*de_del
,
2119 struct buffer_head
*bh
,
2124 struct ext4_dir_entry_2
*de
, *pde
;
2125 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2130 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2131 while (i
< buf_size
- csum_size
) {
2132 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2133 bh
->b_data
, bh
->b_size
, i
))
2137 pde
->rec_len
= ext4_rec_len_to_disk(
2138 ext4_rec_len_from_disk(pde
->rec_len
,
2140 ext4_rec_len_from_disk(de
->rec_len
,
2148 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2150 de
= ext4_next_entry(de
, blocksize
);
2155 static int ext4_delete_entry(handle_t
*handle
,
2157 struct ext4_dir_entry_2
*de_del
,
2158 struct buffer_head
*bh
)
2160 int err
, csum_size
= 0;
2162 if (ext4_has_inline_data(dir
)) {
2163 int has_inline_data
= 1;
2164 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2166 if (has_inline_data
)
2170 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
2171 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
2172 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2174 BUFFER_TRACE(bh
, "get_write_access");
2175 err
= ext4_journal_get_write_access(handle
, bh
);
2179 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2181 dir
->i_sb
->s_blocksize
, csum_size
);
2185 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2186 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2193 ext4_std_error(dir
->i_sb
, err
);
2198 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2199 * since this indicates that nlinks count was previously 1.
2201 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2204 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2205 /* limit is 16-bit i_links_count */
2206 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2207 set_nlink(inode
, 1);
2208 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
2209 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
2215 * If a directory had nlink == 1, then we should let it be 1. This indicates
2216 * directory has >EXT4_LINK_MAX subdirs.
2218 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2220 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2225 static int ext4_add_nondir(handle_t
*handle
,
2226 struct dentry
*dentry
, struct inode
*inode
)
2228 int err
= ext4_add_entry(handle
, dentry
, inode
);
2230 ext4_mark_inode_dirty(handle
, inode
);
2231 unlock_new_inode(inode
);
2232 d_instantiate(dentry
, inode
);
2236 unlock_new_inode(inode
);
2242 * By the time this is called, we already have created
2243 * the directory cache entry for the new file, but it
2244 * is so far negative - it has no inode.
2246 * If the create succeeds, we fill in the inode information
2247 * with d_instantiate().
2249 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2253 struct inode
*inode
;
2254 int err
, credits
, retries
= 0;
2256 dquot_initialize(dir
);
2258 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2259 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2261 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2262 NULL
, EXT4_HT_DIR
, credits
);
2263 handle
= ext4_journal_current_handle();
2264 err
= PTR_ERR(inode
);
2265 if (!IS_ERR(inode
)) {
2266 inode
->i_op
= &ext4_file_inode_operations
;
2267 inode
->i_fop
= &ext4_file_operations
;
2268 ext4_set_aops(inode
);
2269 err
= ext4_add_nondir(handle
, dentry
, inode
);
2270 if (!err
&& IS_DIRSYNC(dir
))
2271 ext4_handle_sync(handle
);
2274 ext4_journal_stop(handle
);
2275 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2280 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2281 umode_t mode
, dev_t rdev
)
2284 struct inode
*inode
;
2285 int err
, credits
, retries
= 0;
2287 if (!new_valid_dev(rdev
))
2290 dquot_initialize(dir
);
2292 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2293 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2295 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2296 NULL
, EXT4_HT_DIR
, credits
);
2297 handle
= ext4_journal_current_handle();
2298 err
= PTR_ERR(inode
);
2299 if (!IS_ERR(inode
)) {
2300 init_special_inode(inode
, inode
->i_mode
, rdev
);
2301 inode
->i_op
= &ext4_special_inode_operations
;
2302 err
= ext4_add_nondir(handle
, dentry
, inode
);
2303 if (!err
&& IS_DIRSYNC(dir
))
2304 ext4_handle_sync(handle
);
2307 ext4_journal_stop(handle
);
2308 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2313 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2316 struct inode
*inode
;
2317 int err
, retries
= 0;
2319 dquot_initialize(dir
);
2322 inode
= ext4_new_inode_start_handle(dir
, mode
,
2325 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2326 4 + EXT4_XATTR_TRANS_BLOCKS
);
2327 handle
= ext4_journal_current_handle();
2328 err
= PTR_ERR(inode
);
2329 if (!IS_ERR(inode
)) {
2330 inode
->i_op
= &ext4_file_inode_operations
;
2331 inode
->i_fop
= &ext4_file_operations
;
2332 ext4_set_aops(inode
);
2333 d_tmpfile(dentry
, inode
);
2334 err
= ext4_orphan_add(handle
, inode
);
2336 goto err_unlock_inode
;
2337 mark_inode_dirty(inode
);
2338 unlock_new_inode(inode
);
2341 ext4_journal_stop(handle
);
2342 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2346 ext4_journal_stop(handle
);
2347 unlock_new_inode(inode
);
2351 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2352 struct ext4_dir_entry_2
*de
,
2353 int blocksize
, int csum_size
,
2354 unsigned int parent_ino
, int dotdot_real_len
)
2356 de
->inode
= cpu_to_le32(inode
->i_ino
);
2358 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2360 strcpy(de
->name
, ".");
2361 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2363 de
= ext4_next_entry(de
, blocksize
);
2364 de
->inode
= cpu_to_le32(parent_ino
);
2366 if (!dotdot_real_len
)
2367 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2368 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2371 de
->rec_len
= ext4_rec_len_to_disk(
2372 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2373 strcpy(de
->name
, "..");
2374 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2376 return ext4_next_entry(de
, blocksize
);
2379 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2380 struct inode
*inode
)
2382 struct buffer_head
*dir_block
= NULL
;
2383 struct ext4_dir_entry_2
*de
;
2384 struct ext4_dir_entry_tail
*t
;
2385 ext4_lblk_t block
= 0;
2386 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2390 if (EXT4_HAS_RO_COMPAT_FEATURE(dir
->i_sb
,
2391 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
2392 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2394 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2395 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2396 if (err
< 0 && err
!= -ENOSPC
)
2403 dir_block
= ext4_append(handle
, inode
, &block
);
2404 if (IS_ERR(dir_block
))
2405 return PTR_ERR(dir_block
);
2406 BUFFER_TRACE(dir_block
, "get_write_access");
2407 err
= ext4_journal_get_write_access(handle
, dir_block
);
2410 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2411 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2412 set_nlink(inode
, 2);
2414 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2415 initialize_dirent_tail(t
, blocksize
);
2418 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2419 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2422 set_buffer_verified(dir_block
);
2428 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2431 struct inode
*inode
;
2432 int err
, credits
, retries
= 0;
2434 if (EXT4_DIR_LINK_MAX(dir
))
2437 dquot_initialize(dir
);
2439 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2440 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2442 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2444 0, NULL
, EXT4_HT_DIR
, credits
);
2445 handle
= ext4_journal_current_handle();
2446 err
= PTR_ERR(inode
);
2450 inode
->i_op
= &ext4_dir_inode_operations
;
2451 inode
->i_fop
= &ext4_dir_operations
;
2452 err
= ext4_init_new_dir(handle
, dir
, inode
);
2454 goto out_clear_inode
;
2455 err
= ext4_mark_inode_dirty(handle
, inode
);
2457 err
= ext4_add_entry(handle
, dentry
, inode
);
2461 unlock_new_inode(inode
);
2462 ext4_mark_inode_dirty(handle
, inode
);
2466 ext4_inc_count(handle
, dir
);
2467 ext4_update_dx_flag(dir
);
2468 err
= ext4_mark_inode_dirty(handle
, dir
);
2470 goto out_clear_inode
;
2471 unlock_new_inode(inode
);
2472 d_instantiate(dentry
, inode
);
2473 if (IS_DIRSYNC(dir
))
2474 ext4_handle_sync(handle
);
2478 ext4_journal_stop(handle
);
2479 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2485 * routine to check that the specified directory is empty (for rmdir)
2487 static int empty_dir(struct inode
*inode
)
2489 unsigned int offset
;
2490 struct buffer_head
*bh
;
2491 struct ext4_dir_entry_2
*de
, *de1
;
2492 struct super_block
*sb
;
2495 if (ext4_has_inline_data(inode
)) {
2496 int has_inline_data
= 1;
2498 err
= empty_inline_dir(inode
, &has_inline_data
);
2499 if (has_inline_data
)
2504 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2505 EXT4_ERROR_INODE(inode
, "invalid size");
2508 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2512 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2513 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2514 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2515 !le32_to_cpu(de1
->inode
) ||
2516 strcmp(".", de
->name
) ||
2517 strcmp("..", de1
->name
)) {
2518 ext4_warning(inode
->i_sb
,
2519 "bad directory (dir #%lu) - no `.' or `..'",
2524 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2525 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2526 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2527 while (offset
< inode
->i_size
) {
2528 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2529 unsigned int lblock
;
2532 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2533 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2536 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2538 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2539 bh
->b_data
, bh
->b_size
, offset
)) {
2540 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2542 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2545 if (le32_to_cpu(de
->inode
)) {
2549 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2550 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2557 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2558 * such inodes, starting at the superblock, in case we crash before the
2559 * file is closed/deleted, or in case the inode truncate spans multiple
2560 * transactions and the last transaction is not recovered after a crash.
2562 * At filesystem recovery time, we walk this list deleting unlinked
2563 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2565 * Orphan list manipulation functions must be called under i_mutex unless
2566 * we are just creating the inode or deleting it.
2568 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2570 struct super_block
*sb
= inode
->i_sb
;
2571 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2572 struct ext4_iloc iloc
;
2576 if (!sbi
->s_journal
)
2579 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2580 !mutex_is_locked(&inode
->i_mutex
));
2582 * Exit early if inode already is on orphan list. This is a big speedup
2583 * since we don't have to contend on the global s_orphan_lock.
2585 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2589 * Orphan handling is only valid for files with data blocks
2590 * being truncated, or files being unlinked. Note that we either
2591 * hold i_mutex, or the inode can not be referenced from outside,
2592 * so i_nlink should not be bumped due to race
2594 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2595 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2597 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2598 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2602 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2606 mutex_lock(&sbi
->s_orphan_lock
);
2608 * Due to previous errors inode may be already a part of on-disk
2609 * orphan list. If so skip on-disk list modification.
2611 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2612 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2613 /* Insert this inode at the head of the on-disk orphan list */
2614 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2615 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2618 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2619 mutex_unlock(&sbi
->s_orphan_lock
);
2622 err
= ext4_handle_dirty_super(handle
, sb
);
2623 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2628 * We have to remove inode from in-memory list if
2629 * addition to on disk orphan list failed. Stray orphan
2630 * list entries can cause panics at unmount time.
2632 mutex_lock(&sbi
->s_orphan_lock
);
2633 list_del(&EXT4_I(inode
)->i_orphan
);
2634 mutex_unlock(&sbi
->s_orphan_lock
);
2637 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2638 jbd_debug(4, "orphan inode %lu will point to %d\n",
2639 inode
->i_ino
, NEXT_ORPHAN(inode
));
2641 ext4_std_error(sb
, err
);
2646 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2647 * of such inodes stored on disk, because it is finally being cleaned up.
2649 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2651 struct list_head
*prev
;
2652 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2653 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2655 struct ext4_iloc iloc
;
2658 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2661 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2662 !mutex_is_locked(&inode
->i_mutex
));
2663 /* Do this quick check before taking global s_orphan_lock. */
2664 if (list_empty(&ei
->i_orphan
))
2668 /* Grab inode buffer early before taking global s_orphan_lock */
2669 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2672 mutex_lock(&sbi
->s_orphan_lock
);
2673 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2675 prev
= ei
->i_orphan
.prev
;
2676 list_del_init(&ei
->i_orphan
);
2678 /* If we're on an error path, we may not have a valid
2679 * transaction handle with which to update the orphan list on
2680 * disk, but we still need to remove the inode from the linked
2681 * list in memory. */
2682 if (!handle
|| err
) {
2683 mutex_unlock(&sbi
->s_orphan_lock
);
2687 ino_next
= NEXT_ORPHAN(inode
);
2688 if (prev
== &sbi
->s_orphan
) {
2689 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2690 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2691 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2693 mutex_unlock(&sbi
->s_orphan_lock
);
2696 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2697 mutex_unlock(&sbi
->s_orphan_lock
);
2698 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2700 struct ext4_iloc iloc2
;
2701 struct inode
*i_prev
=
2702 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2704 jbd_debug(4, "orphan inode %lu will point to %u\n",
2705 i_prev
->i_ino
, ino_next
);
2706 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2708 mutex_unlock(&sbi
->s_orphan_lock
);
2711 NEXT_ORPHAN(i_prev
) = ino_next
;
2712 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2713 mutex_unlock(&sbi
->s_orphan_lock
);
2717 NEXT_ORPHAN(inode
) = 0;
2718 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2720 ext4_std_error(inode
->i_sb
, err
);
2728 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2731 struct inode
*inode
;
2732 struct buffer_head
*bh
;
2733 struct ext4_dir_entry_2
*de
;
2734 handle_t
*handle
= NULL
;
2736 /* Initialize quotas before so that eventual writes go in
2737 * separate transaction */
2738 dquot_initialize(dir
);
2739 dquot_initialize(dentry
->d_inode
);
2742 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2748 inode
= dentry
->d_inode
;
2751 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2754 retval
= -ENOTEMPTY
;
2755 if (!empty_dir(inode
))
2758 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2759 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2760 if (IS_ERR(handle
)) {
2761 retval
= PTR_ERR(handle
);
2766 if (IS_DIRSYNC(dir
))
2767 ext4_handle_sync(handle
);
2769 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2772 if (!EXT4_DIR_LINK_EMPTY(inode
))
2773 ext4_warning(inode
->i_sb
,
2774 "empty directory has too many links (%d)",
2778 /* There's no need to set i_disksize: the fact that i_nlink is
2779 * zero will ensure that the right thing happens during any
2782 ext4_orphan_add(handle
, inode
);
2783 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2784 ext4_mark_inode_dirty(handle
, inode
);
2785 ext4_dec_count(handle
, dir
);
2786 ext4_update_dx_flag(dir
);
2787 ext4_mark_inode_dirty(handle
, dir
);
2792 ext4_journal_stop(handle
);
2796 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2799 struct inode
*inode
;
2800 struct buffer_head
*bh
;
2801 struct ext4_dir_entry_2
*de
;
2802 handle_t
*handle
= NULL
;
2804 trace_ext4_unlink_enter(dir
, dentry
);
2805 /* Initialize quotas before so that eventual writes go
2806 * in separate transaction */
2807 dquot_initialize(dir
);
2808 dquot_initialize(dentry
->d_inode
);
2811 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2817 inode
= dentry
->d_inode
;
2820 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2823 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2824 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2825 if (IS_ERR(handle
)) {
2826 retval
= PTR_ERR(handle
);
2831 if (IS_DIRSYNC(dir
))
2832 ext4_handle_sync(handle
);
2834 if (!inode
->i_nlink
) {
2835 ext4_warning(inode
->i_sb
,
2836 "Deleting nonexistent file (%lu), %d",
2837 inode
->i_ino
, inode
->i_nlink
);
2838 set_nlink(inode
, 1);
2840 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2843 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2844 ext4_update_dx_flag(dir
);
2845 ext4_mark_inode_dirty(handle
, dir
);
2847 if (!inode
->i_nlink
)
2848 ext4_orphan_add(handle
, inode
);
2849 inode
->i_ctime
= ext4_current_time(inode
);
2850 ext4_mark_inode_dirty(handle
, inode
);
2856 ext4_journal_stop(handle
);
2857 trace_ext4_unlink_exit(dentry
, retval
);
2861 static int ext4_symlink(struct inode
*dir
,
2862 struct dentry
*dentry
, const char *symname
)
2865 struct inode
*inode
;
2866 int l
, err
, retries
= 0;
2869 l
= strlen(symname
)+1;
2870 if (l
> dir
->i_sb
->s_blocksize
)
2871 return -ENAMETOOLONG
;
2873 dquot_initialize(dir
);
2875 if (l
> EXT4_N_BLOCKS
* 4) {
2877 * For non-fast symlinks, we just allocate inode and put it on
2878 * orphan list in the first transaction => we need bitmap,
2879 * group descriptor, sb, inode block, quota blocks, and
2880 * possibly selinux xattr blocks.
2882 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2883 EXT4_XATTR_TRANS_BLOCKS
;
2886 * Fast symlink. We have to add entry to directory
2887 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2888 * allocate new inode (bitmap, group descriptor, inode block,
2889 * quota blocks, sb is already counted in previous macros).
2891 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2892 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
2895 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
2896 &dentry
->d_name
, 0, NULL
,
2897 EXT4_HT_DIR
, credits
);
2898 handle
= ext4_journal_current_handle();
2899 err
= PTR_ERR(inode
);
2903 if (l
> EXT4_N_BLOCKS
* 4) {
2904 inode
->i_op
= &ext4_symlink_inode_operations
;
2905 ext4_set_aops(inode
);
2907 * We cannot call page_symlink() with transaction started
2908 * because it calls into ext4_write_begin() which can wait
2909 * for transaction commit if we are running out of space
2910 * and thus we deadlock. So we have to stop transaction now
2911 * and restart it when symlink contents is written.
2913 * To keep fs consistent in case of crash, we have to put inode
2914 * to orphan list in the mean time.
2917 err
= ext4_orphan_add(handle
, inode
);
2918 ext4_journal_stop(handle
);
2920 goto err_drop_inode
;
2921 err
= __page_symlink(inode
, symname
, l
, 1);
2923 goto err_drop_inode
;
2925 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2926 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2928 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2929 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2930 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
2931 if (IS_ERR(handle
)) {
2932 err
= PTR_ERR(handle
);
2933 goto err_drop_inode
;
2935 set_nlink(inode
, 1);
2936 err
= ext4_orphan_del(handle
, inode
);
2938 ext4_journal_stop(handle
);
2940 goto err_drop_inode
;
2943 /* clear the extent format for fast symlink */
2944 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
2945 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2946 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
2947 inode
->i_size
= l
-1;
2949 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2950 err
= ext4_add_nondir(handle
, dentry
, inode
);
2951 if (!err
&& IS_DIRSYNC(dir
))
2952 ext4_handle_sync(handle
);
2956 ext4_journal_stop(handle
);
2957 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2961 unlock_new_inode(inode
);
2966 static int ext4_link(struct dentry
*old_dentry
,
2967 struct inode
*dir
, struct dentry
*dentry
)
2970 struct inode
*inode
= old_dentry
->d_inode
;
2971 int err
, retries
= 0;
2973 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
2976 dquot_initialize(dir
);
2979 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2980 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2981 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
2983 return PTR_ERR(handle
);
2985 if (IS_DIRSYNC(dir
))
2986 ext4_handle_sync(handle
);
2988 inode
->i_ctime
= ext4_current_time(inode
);
2989 ext4_inc_count(handle
, inode
);
2992 err
= ext4_add_entry(handle
, dentry
, inode
);
2994 ext4_mark_inode_dirty(handle
, inode
);
2995 /* this can happen only for tmpfile being
2996 * linked the first time
2998 if (inode
->i_nlink
== 1)
2999 ext4_orphan_del(handle
, inode
);
3000 d_instantiate(dentry
, inode
);
3005 ext4_journal_stop(handle
);
3006 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3013 * Try to find buffer head where contains the parent block.
3014 * It should be the inode block if it is inlined or the 1st block
3015 * if it is a normal dir.
3017 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
3018 struct inode
*inode
,
3020 struct ext4_dir_entry_2
**parent_de
,
3023 struct buffer_head
*bh
;
3025 if (!ext4_has_inline_data(inode
)) {
3026 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
3028 *retval
= PTR_ERR(bh
);
3031 *parent_de
= ext4_next_entry(
3032 (struct ext4_dir_entry_2
*)bh
->b_data
,
3033 inode
->i_sb
->s_blocksize
);
3038 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3041 struct ext4_renament
{
3043 struct dentry
*dentry
;
3044 struct inode
*inode
;
3046 int dir_nlink_delta
;
3048 /* entry for "dentry" */
3049 struct buffer_head
*bh
;
3050 struct ext4_dir_entry_2
*de
;
3053 /* entry for ".." in inode if it's a directory */
3054 struct buffer_head
*dir_bh
;
3055 struct ext4_dir_entry_2
*parent_de
;
3059 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3063 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3064 &retval
, &ent
->parent_de
,
3068 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3070 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3071 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3074 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3079 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3080 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3081 if (!ent
->dir_inlined
) {
3082 if (is_dx(ent
->inode
)) {
3083 retval
= ext4_handle_dirty_dx_node(handle
,
3087 retval
= ext4_handle_dirty_dirent_node(handle
,
3092 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3095 ext4_std_error(ent
->dir
->i_sb
, retval
);
3101 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3102 unsigned ino
, unsigned file_type
)
3106 BUFFER_TRACE(ent
->bh
, "get write access");
3107 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3110 ent
->de
->inode
= cpu_to_le32(ino
);
3111 if (EXT4_HAS_INCOMPAT_FEATURE(ent
->dir
->i_sb
,
3112 EXT4_FEATURE_INCOMPAT_FILETYPE
))
3113 ent
->de
->file_type
= file_type
;
3114 ent
->dir
->i_version
++;
3115 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3116 ext4_current_time(ent
->dir
);
3117 ext4_mark_inode_dirty(handle
, ent
->dir
);
3118 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3119 if (!ent
->inlined
) {
3120 retval
= ext4_handle_dirty_dirent_node(handle
,
3122 if (unlikely(retval
)) {
3123 ext4_std_error(ent
->dir
->i_sb
, retval
);
3133 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3134 const struct qstr
*d_name
)
3136 int retval
= -ENOENT
;
3137 struct buffer_head
*bh
;
3138 struct ext4_dir_entry_2
*de
;
3140 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3144 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3150 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3155 * ent->de could have moved from under us during htree split, so make
3156 * sure that we are deleting the right entry. We might also be pointing
3157 * to a stale entry in the unused part of ent->bh so just checking inum
3158 * and the name isn't enough.
3160 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3161 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3162 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3163 ent
->de
->name_len
) ||
3165 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3166 &ent
->dentry
->d_name
);
3168 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3169 if (retval
== -ENOENT
) {
3170 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3171 &ent
->dentry
->d_name
);
3176 ext4_warning(ent
->dir
->i_sb
,
3177 "Deleting old file (%lu), %d, error=%d",
3178 ent
->dir
->i_ino
, ent
->dir
->i_nlink
, retval
);
3182 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3184 if (ent
->dir_nlink_delta
) {
3185 if (ent
->dir_nlink_delta
== -1)
3186 ext4_dec_count(handle
, ent
->dir
);
3188 ext4_inc_count(handle
, ent
->dir
);
3189 ext4_mark_inode_dirty(handle
, ent
->dir
);
3194 * Anybody can rename anything with this: the permission checks are left to the
3195 * higher-level routines.
3197 * n.b. old_{dentry,inode) refers to the source dentry/inode
3198 * while new_{dentry,inode) refers to the destination dentry/inode
3199 * This comes from rename(const char *oldpath, const char *newpath)
3201 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3202 struct inode
*new_dir
, struct dentry
*new_dentry
)
3204 handle_t
*handle
= NULL
;
3205 struct ext4_renament old
= {
3207 .dentry
= old_dentry
,
3208 .inode
= old_dentry
->d_inode
,
3210 struct ext4_renament
new = {
3212 .dentry
= new_dentry
,
3213 .inode
= new_dentry
->d_inode
,
3218 dquot_initialize(old
.dir
);
3219 dquot_initialize(new.dir
);
3221 /* Initialize quotas before so that eventual writes go
3222 * in separate transaction */
3224 dquot_initialize(new.inode
);
3226 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3228 return PTR_ERR(old
.bh
);
3230 * Check for inode number is _not_ due to possible IO errors.
3231 * We might rmdir the source, keep it as pwd of some process
3232 * and merrily kill the link to whatever was created under the
3233 * same name. Goodbye sticky bit ;-<
3236 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3239 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3240 &new.de
, &new.inlined
);
3241 if (IS_ERR(new.bh
)) {
3242 retval
= PTR_ERR(new.bh
);
3252 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3253 ext4_alloc_da_blocks(old
.inode
);
3255 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3256 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3257 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3259 return PTR_ERR(handle
);
3261 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3262 ext4_handle_sync(handle
);
3264 if (S_ISDIR(old
.inode
->i_mode
)) {
3266 retval
= -ENOTEMPTY
;
3267 if (!empty_dir(new.inode
))
3271 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3274 retval
= ext4_rename_dir_prepare(handle
, &old
);
3279 * If we're renaming a file within an inline_data dir and adding or
3280 * setting the new dirent causes a conversion from inline_data to
3281 * extents/blockmap, we need to force the dirent delete code to
3282 * re-read the directory, or else we end up trying to delete a dirent
3283 * from what is now the extent tree root (or a block map).
3285 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3286 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3288 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3292 retval
= ext4_setent(handle
, &new,
3293 old
.inode
->i_ino
, old
.de
->file_type
);
3298 force_reread
= !ext4_test_inode_flag(new.dir
,
3299 EXT4_INODE_INLINE_DATA
);
3302 * Like most other Unix systems, set the ctime for inodes on a
3305 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3306 ext4_mark_inode_dirty(handle
, old
.inode
);
3311 ext4_rename_delete(handle
, &old
, force_reread
);
3314 ext4_dec_count(handle
, new.inode
);
3315 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3317 old
.dir
->i_ctime
= old
.dir
->i_mtime
= ext4_current_time(old
.dir
);
3318 ext4_update_dx_flag(old
.dir
);
3320 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3324 ext4_dec_count(handle
, old
.dir
);
3326 /* checked empty_dir above, can't have another parent,
3327 * ext4_dec_count() won't work for many-linked dirs */
3328 clear_nlink(new.inode
);
3330 ext4_inc_count(handle
, new.dir
);
3331 ext4_update_dx_flag(new.dir
);
3332 ext4_mark_inode_dirty(handle
, new.dir
);
3335 ext4_mark_inode_dirty(handle
, old
.dir
);
3337 ext4_mark_inode_dirty(handle
, new.inode
);
3338 if (!new.inode
->i_nlink
)
3339 ext4_orphan_add(handle
, new.inode
);
3348 ext4_journal_stop(handle
);
3352 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3353 struct inode
*new_dir
, struct dentry
*new_dentry
)
3355 handle_t
*handle
= NULL
;
3356 struct ext4_renament old
= {
3358 .dentry
= old_dentry
,
3359 .inode
= old_dentry
->d_inode
,
3361 struct ext4_renament
new = {
3363 .dentry
= new_dentry
,
3364 .inode
= new_dentry
->d_inode
,
3369 dquot_initialize(old
.dir
);
3370 dquot_initialize(new.dir
);
3372 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3373 &old
.de
, &old
.inlined
);
3375 return PTR_ERR(old
.bh
);
3377 * Check for inode number is _not_ due to possible IO errors.
3378 * We might rmdir the source, keep it as pwd of some process
3379 * and merrily kill the link to whatever was created under the
3380 * same name. Goodbye sticky bit ;-<
3383 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3386 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3387 &new.de
, &new.inlined
);
3388 if (IS_ERR(new.bh
)) {
3389 retval
= PTR_ERR(new.bh
);
3394 /* RENAME_EXCHANGE case: old *and* new must both exist */
3395 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3398 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3399 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3400 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3402 return PTR_ERR(handle
);
3404 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3405 ext4_handle_sync(handle
);
3407 if (S_ISDIR(old
.inode
->i_mode
)) {
3409 retval
= ext4_rename_dir_prepare(handle
, &old
);
3413 if (S_ISDIR(new.inode
->i_mode
)) {
3415 retval
= ext4_rename_dir_prepare(handle
, &new);
3421 * Other than the special case of overwriting a directory, parents'
3422 * nlink only needs to be modified if this is a cross directory rename.
3424 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3425 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3426 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3428 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3429 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3433 new_file_type
= new.de
->file_type
;
3434 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3438 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3443 * Like most other Unix systems, set the ctime for inodes on a
3446 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3447 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3448 ext4_mark_inode_dirty(handle
, old
.inode
);
3449 ext4_mark_inode_dirty(handle
, new.inode
);
3452 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3457 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3461 ext4_update_dir_count(handle
, &old
);
3462 ext4_update_dir_count(handle
, &new);
3471 ext4_journal_stop(handle
);
3475 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3476 struct inode
*new_dir
, struct dentry
*new_dentry
,
3479 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
))
3482 if (flags
& RENAME_EXCHANGE
) {
3483 return ext4_cross_rename(old_dir
, old_dentry
,
3484 new_dir
, new_dentry
);
3487 * Existence checking was done by the VFS, otherwise "RENAME_NOREPLACE"
3488 * is equivalent to regular rename.
3490 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3494 * directories can handle most operations...
3496 const struct inode_operations ext4_dir_inode_operations
= {
3497 .create
= ext4_create
,
3498 .lookup
= ext4_lookup
,
3500 .unlink
= ext4_unlink
,
3501 .symlink
= ext4_symlink
,
3502 .mkdir
= ext4_mkdir
,
3503 .rmdir
= ext4_rmdir
,
3504 .mknod
= ext4_mknod
,
3505 .tmpfile
= ext4_tmpfile
,
3506 .rename2
= ext4_rename2
,
3507 .setattr
= ext4_setattr
,
3508 .setxattr
= generic_setxattr
,
3509 .getxattr
= generic_getxattr
,
3510 .listxattr
= ext4_listxattr
,
3511 .removexattr
= generic_removexattr
,
3512 .get_acl
= ext4_get_acl
,
3513 .set_acl
= ext4_set_acl
,
3514 .fiemap
= ext4_fiemap
,
3517 const struct inode_operations ext4_special_inode_operations
= {
3518 .setattr
= ext4_setattr
,
3519 .setxattr
= generic_setxattr
,
3520 .getxattr
= generic_getxattr
,
3521 .listxattr
= ext4_listxattr
,
3522 .removexattr
= generic_removexattr
,
3523 .get_acl
= ext4_get_acl
,
3524 .set_acl
= ext4_set_acl
,