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/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
37 #include "ext4_jbd2.h"
42 #include <trace/events/ext4.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 static struct buffer_head
*ext4_append(handle_t
*handle
,
54 struct buffer_head
*bh
;
57 if (unlikely(EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
&&
58 ((inode
->i_size
>> 10) >=
59 EXT4_SB(inode
->i_sb
)->s_max_dir_size_kb
)))
60 return ERR_PTR(-ENOSPC
);
62 *block
= inode
->i_size
>> inode
->i_sb
->s_blocksize_bits
;
64 bh
= ext4_bread(handle
, inode
, *block
, 1);
67 inode
->i_size
+= inode
->i_sb
->s_blocksize
;
68 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
69 BUFFER_TRACE(bh
, "get_write_access");
70 err
= ext4_journal_get_write_access(handle
, bh
);
73 ext4_std_error(inode
->i_sb
, err
);
79 static int ext4_dx_csum_verify(struct inode
*inode
,
80 struct ext4_dir_entry
*dirent
);
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __LINE__)
89 static struct buffer_head
*__ext4_read_dirblock(struct inode
*inode
,
94 struct buffer_head
*bh
;
95 struct ext4_dir_entry
*dirent
;
98 bh
= ext4_bread(NULL
, inode
, block
, 0);
100 __ext4_warning(inode
->i_sb
, __func__
, line
,
101 "error %ld reading directory block "
102 "(ino %lu, block %lu)", PTR_ERR(bh
), inode
->i_ino
,
103 (unsigned long) block
);
108 ext4_error_inode(inode
, __func__
, line
, block
, "Directory hole found");
109 return ERR_PTR(-EIO
);
111 dirent
= (struct ext4_dir_entry
*) bh
->b_data
;
112 /* Determine whether or not we have an index block */
116 else if (ext4_rec_len_from_disk(dirent
->rec_len
,
117 inode
->i_sb
->s_blocksize
) ==
118 inode
->i_sb
->s_blocksize
)
121 if (!is_dx_block
&& type
== INDEX
) {
122 ext4_error_inode(inode
, __func__
, line
, block
,
123 "directory leaf block found instead of index block");
124 return ERR_PTR(-EIO
);
126 if (!ext4_has_metadata_csum(inode
->i_sb
) ||
131 * An empty leaf block can get mistaken for a index block; for
132 * this reason, we can only check the index checksum when the
133 * caller is sure it should be an index block.
135 if (is_dx_block
&& type
== INDEX
) {
136 if (ext4_dx_csum_verify(inode
, dirent
))
137 set_buffer_verified(bh
);
139 ext4_error_inode(inode
, __func__
, line
, block
,
140 "Directory index failed checksum");
142 return ERR_PTR(-EIO
);
146 if (ext4_dirent_csum_verify(inode
, dirent
))
147 set_buffer_verified(bh
);
149 ext4_error_inode(inode
, __func__
, line
, block
,
150 "Directory block failed checksum");
152 return ERR_PTR(-EIO
);
159 #define assert(test) J_ASSERT(test)
163 #define dxtrace(command) command
165 #define dxtrace(command)
189 * dx_root_info is laid out so that if it should somehow get overlaid by a
190 * dirent the two low bits of the hash version will be zero. Therefore, the
191 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
196 struct fake_dirent dot
;
198 struct fake_dirent dotdot
;
202 __le32 reserved_zero
;
204 u8 info_length
; /* 8 */
209 struct dx_entry entries
[0];
214 struct fake_dirent fake
;
215 struct dx_entry entries
[0];
221 struct buffer_head
*bh
;
222 struct dx_entry
*entries
;
234 * This goes at the end of each htree block.
238 __le32 dt_checksum
; /* crc32c(uuid+inum+dirblock) */
241 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
);
242 static void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
);
243 static inline unsigned dx_get_hash(struct dx_entry
*entry
);
244 static void dx_set_hash(struct dx_entry
*entry
, unsigned value
);
245 static unsigned dx_get_count(struct dx_entry
*entries
);
246 static unsigned dx_get_limit(struct dx_entry
*entries
);
247 static void dx_set_count(struct dx_entry
*entries
, unsigned value
);
248 static void dx_set_limit(struct dx_entry
*entries
, unsigned value
);
249 static unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
);
250 static unsigned dx_node_limit(struct inode
*dir
);
251 static struct dx_frame
*dx_probe(const struct qstr
*d_name
,
253 struct dx_hash_info
*hinfo
,
254 struct dx_frame
*frame
);
255 static void dx_release(struct dx_frame
*frames
);
256 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
257 struct dx_hash_info
*hinfo
, struct dx_map_entry map
[]);
258 static void dx_sort_map(struct dx_map_entry
*map
, unsigned count
);
259 static struct ext4_dir_entry_2
*dx_move_dirents(char *from
, char *to
,
260 struct dx_map_entry
*offsets
, int count
, unsigned blocksize
);
261 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
);
262 static void dx_insert_block(struct dx_frame
*frame
,
263 u32 hash
, ext4_lblk_t block
);
264 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
265 struct dx_frame
*frame
,
266 struct dx_frame
*frames
,
268 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
,
269 const struct qstr
*d_name
,
270 struct ext4_dir_entry_2
**res_dir
);
271 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
272 struct inode
*inode
);
274 /* checksumming functions */
275 void initialize_dirent_tail(struct ext4_dir_entry_tail
*t
,
276 unsigned int blocksize
)
278 memset(t
, 0, sizeof(struct ext4_dir_entry_tail
));
279 t
->det_rec_len
= ext4_rec_len_to_disk(
280 sizeof(struct ext4_dir_entry_tail
), blocksize
);
281 t
->det_reserved_ft
= EXT4_FT_DIR_CSUM
;
284 /* Walk through a dirent block to find a checksum "dirent" at the tail */
285 static struct ext4_dir_entry_tail
*get_dirent_tail(struct inode
*inode
,
286 struct ext4_dir_entry
*de
)
288 struct ext4_dir_entry_tail
*t
;
291 struct ext4_dir_entry
*d
, *top
;
294 top
= (struct ext4_dir_entry
*)(((void *)de
) +
295 (EXT4_BLOCK_SIZE(inode
->i_sb
) -
296 sizeof(struct ext4_dir_entry_tail
)));
297 while (d
< top
&& d
->rec_len
)
298 d
= (struct ext4_dir_entry
*)(((void *)d
) +
299 le16_to_cpu(d
->rec_len
));
304 t
= (struct ext4_dir_entry_tail
*)d
;
306 t
= EXT4_DIRENT_TAIL(de
, EXT4_BLOCK_SIZE(inode
->i_sb
));
309 if (t
->det_reserved_zero1
||
310 le16_to_cpu(t
->det_rec_len
) != sizeof(struct ext4_dir_entry_tail
) ||
311 t
->det_reserved_zero2
||
312 t
->det_reserved_ft
!= EXT4_FT_DIR_CSUM
)
318 static __le32
ext4_dirent_csum(struct inode
*inode
,
319 struct ext4_dir_entry
*dirent
, int size
)
321 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
322 struct ext4_inode_info
*ei
= EXT4_I(inode
);
325 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
326 return cpu_to_le32(csum
);
329 static void warn_no_space_for_csum(struct inode
*inode
)
331 ext4_warning(inode
->i_sb
, "no space in directory inode %lu leaf for "
332 "checksum. Please run e2fsck -D.", inode
->i_ino
);
335 int ext4_dirent_csum_verify(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
337 struct ext4_dir_entry_tail
*t
;
339 if (!ext4_has_metadata_csum(inode
->i_sb
))
342 t
= get_dirent_tail(inode
, dirent
);
344 warn_no_space_for_csum(inode
);
348 if (t
->det_checksum
!= ext4_dirent_csum(inode
, dirent
,
349 (void *)t
- (void *)dirent
))
355 static void ext4_dirent_csum_set(struct inode
*inode
,
356 struct ext4_dir_entry
*dirent
)
358 struct ext4_dir_entry_tail
*t
;
360 if (!ext4_has_metadata_csum(inode
->i_sb
))
363 t
= get_dirent_tail(inode
, dirent
);
365 warn_no_space_for_csum(inode
);
369 t
->det_checksum
= ext4_dirent_csum(inode
, dirent
,
370 (void *)t
- (void *)dirent
);
373 int ext4_handle_dirty_dirent_node(handle_t
*handle
,
375 struct buffer_head
*bh
)
377 ext4_dirent_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
378 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
381 static struct dx_countlimit
*get_dx_countlimit(struct inode
*inode
,
382 struct ext4_dir_entry
*dirent
,
385 struct ext4_dir_entry
*dp
;
386 struct dx_root_info
*root
;
389 if (le16_to_cpu(dirent
->rec_len
) == EXT4_BLOCK_SIZE(inode
->i_sb
))
391 else if (le16_to_cpu(dirent
->rec_len
) == 12) {
392 dp
= (struct ext4_dir_entry
*)(((void *)dirent
) + 12);
393 if (le16_to_cpu(dp
->rec_len
) !=
394 EXT4_BLOCK_SIZE(inode
->i_sb
) - 12)
396 root
= (struct dx_root_info
*)(((void *)dp
+ 12));
397 if (root
->reserved_zero
||
398 root
->info_length
!= sizeof(struct dx_root_info
))
405 *offset
= count_offset
;
406 return (struct dx_countlimit
*)(((void *)dirent
) + count_offset
);
409 static __le32
ext4_dx_csum(struct inode
*inode
, struct ext4_dir_entry
*dirent
,
410 int count_offset
, int count
, struct dx_tail
*t
)
412 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
413 struct ext4_inode_info
*ei
= EXT4_I(inode
);
418 size
= count_offset
+ (count
* sizeof(struct dx_entry
));
419 save_csum
= t
->dt_checksum
;
421 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)dirent
, size
);
422 csum
= ext4_chksum(sbi
, csum
, (__u8
*)t
, sizeof(struct dx_tail
));
423 t
->dt_checksum
= save_csum
;
425 return cpu_to_le32(csum
);
428 static int ext4_dx_csum_verify(struct inode
*inode
,
429 struct ext4_dir_entry
*dirent
)
431 struct dx_countlimit
*c
;
433 int count_offset
, limit
, count
;
435 if (!ext4_has_metadata_csum(inode
->i_sb
))
438 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
440 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
443 limit
= le16_to_cpu(c
->limit
);
444 count
= le16_to_cpu(c
->count
);
445 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
446 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
447 warn_no_space_for_csum(inode
);
450 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
452 if (t
->dt_checksum
!= ext4_dx_csum(inode
, dirent
, count_offset
,
458 static void ext4_dx_csum_set(struct inode
*inode
, struct ext4_dir_entry
*dirent
)
460 struct dx_countlimit
*c
;
462 int count_offset
, limit
, count
;
464 if (!ext4_has_metadata_csum(inode
->i_sb
))
467 c
= get_dx_countlimit(inode
, dirent
, &count_offset
);
469 EXT4_ERROR_INODE(inode
, "dir seems corrupt? Run e2fsck -D.");
472 limit
= le16_to_cpu(c
->limit
);
473 count
= le16_to_cpu(c
->count
);
474 if (count_offset
+ (limit
* sizeof(struct dx_entry
)) >
475 EXT4_BLOCK_SIZE(inode
->i_sb
) - sizeof(struct dx_tail
)) {
476 warn_no_space_for_csum(inode
);
479 t
= (struct dx_tail
*)(((struct dx_entry
*)c
) + limit
);
481 t
->dt_checksum
= ext4_dx_csum(inode
, dirent
, count_offset
, count
, t
);
484 static inline int ext4_handle_dirty_dx_node(handle_t
*handle
,
486 struct buffer_head
*bh
)
488 ext4_dx_csum_set(inode
, (struct ext4_dir_entry
*)bh
->b_data
);
489 return ext4_handle_dirty_metadata(handle
, inode
, bh
);
493 * p is at least 6 bytes before the end of page
495 static inline struct ext4_dir_entry_2
*
496 ext4_next_entry(struct ext4_dir_entry_2
*p
, unsigned long blocksize
)
498 return (struct ext4_dir_entry_2
*)((char *)p
+
499 ext4_rec_len_from_disk(p
->rec_len
, blocksize
));
503 * Future: use high four bits of block for coalesce-on-delete flags
504 * Mask them off for now.
507 static inline ext4_lblk_t
dx_get_block(struct dx_entry
*entry
)
509 return le32_to_cpu(entry
->block
) & 0x00ffffff;
512 static inline void dx_set_block(struct dx_entry
*entry
, ext4_lblk_t value
)
514 entry
->block
= cpu_to_le32(value
);
517 static inline unsigned dx_get_hash(struct dx_entry
*entry
)
519 return le32_to_cpu(entry
->hash
);
522 static inline void dx_set_hash(struct dx_entry
*entry
, unsigned value
)
524 entry
->hash
= cpu_to_le32(value
);
527 static inline unsigned dx_get_count(struct dx_entry
*entries
)
529 return le16_to_cpu(((struct dx_countlimit
*) entries
)->count
);
532 static inline unsigned dx_get_limit(struct dx_entry
*entries
)
534 return le16_to_cpu(((struct dx_countlimit
*) entries
)->limit
);
537 static inline void dx_set_count(struct dx_entry
*entries
, unsigned value
)
539 ((struct dx_countlimit
*) entries
)->count
= cpu_to_le16(value
);
542 static inline void dx_set_limit(struct dx_entry
*entries
, unsigned value
)
544 ((struct dx_countlimit
*) entries
)->limit
= cpu_to_le16(value
);
547 static inline unsigned dx_root_limit(struct inode
*dir
, unsigned infosize
)
549 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(1) -
550 EXT4_DIR_REC_LEN(2) - infosize
;
552 if (ext4_has_metadata_csum(dir
->i_sb
))
553 entry_space
-= sizeof(struct dx_tail
);
554 return entry_space
/ sizeof(struct dx_entry
);
557 static inline unsigned dx_node_limit(struct inode
*dir
)
559 unsigned entry_space
= dir
->i_sb
->s_blocksize
- EXT4_DIR_REC_LEN(0);
561 if (ext4_has_metadata_csum(dir
->i_sb
))
562 entry_space
-= sizeof(struct dx_tail
);
563 return entry_space
/ sizeof(struct dx_entry
);
570 static void dx_show_index(char * label
, struct dx_entry
*entries
)
572 int i
, n
= dx_get_count (entries
);
573 printk(KERN_DEBUG
"%s index ", label
);
574 for (i
= 0; i
< n
; i
++) {
575 printk("%x->%lu ", i
? dx_get_hash(entries
+ i
) :
576 0, (unsigned long)dx_get_block(entries
+ i
));
588 static struct stats
dx_show_leaf(struct dx_hash_info
*hinfo
, struct ext4_dir_entry_2
*de
,
589 int size
, int show_names
)
591 unsigned names
= 0, space
= 0;
592 char *base
= (char *) de
;
593 struct dx_hash_info h
= *hinfo
;
596 while ((char *) de
< base
+ size
)
602 int len
= de
->name_len
;
603 char *name
= de
->name
;
604 while (len
--) printk("%c", *name
++);
605 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
606 printk(":%x.%u ", h
.hash
,
607 (unsigned) ((char *) de
- base
));
609 space
+= EXT4_DIR_REC_LEN(de
->name_len
);
612 de
= ext4_next_entry(de
, size
);
614 printk("(%i)\n", names
);
615 return (struct stats
) { names
, space
, 1 };
618 struct stats
dx_show_entries(struct dx_hash_info
*hinfo
, struct inode
*dir
,
619 struct dx_entry
*entries
, int levels
)
621 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
622 unsigned count
= dx_get_count(entries
), names
= 0, space
= 0, i
;
624 struct buffer_head
*bh
;
626 printk("%i indexed blocks...\n", count
);
627 for (i
= 0; i
< count
; i
++, entries
++)
629 ext4_lblk_t block
= dx_get_block(entries
);
630 ext4_lblk_t hash
= i
? dx_get_hash(entries
): 0;
631 u32 range
= i
< count
- 1? (dx_get_hash(entries
+ 1) - hash
): ~hash
;
633 printk("%s%3u:%03u hash %8x/%8x ",levels
?"":" ", i
, block
, hash
, range
);
634 bh
= ext4_bread(NULL
,dir
, block
, 0);
635 if (!bh
|| IS_ERR(bh
))
638 dx_show_entries(hinfo
, dir
, ((struct dx_node
*) bh
->b_data
)->entries
, levels
- 1):
639 dx_show_leaf(hinfo
, (struct ext4_dir_entry_2
*) bh
->b_data
, blocksize
, 0);
640 names
+= stats
.names
;
641 space
+= stats
.space
;
642 bcount
+= stats
.bcount
;
646 printk(KERN_DEBUG
"%snames %u, fullness %u (%u%%)\n",
647 levels
? "" : " ", names
, space
/bcount
,
648 (space
/bcount
)*100/blocksize
);
649 return (struct stats
) { names
, space
, bcount
};
651 #endif /* DX_DEBUG */
654 * Probe for a directory leaf block to search.
656 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
657 * error in the directory index, and the caller should fall back to
658 * searching the directory normally. The callers of dx_probe **MUST**
659 * check for this error code, and make sure it never gets reflected
662 static struct dx_frame
*
663 dx_probe(const struct qstr
*d_name
, struct inode
*dir
,
664 struct dx_hash_info
*hinfo
, struct dx_frame
*frame_in
)
666 unsigned count
, indirect
;
667 struct dx_entry
*at
, *entries
, *p
, *q
, *m
;
668 struct dx_root
*root
;
669 struct dx_frame
*frame
= frame_in
;
670 struct dx_frame
*ret_err
= ERR_PTR(ERR_BAD_DX_DIR
);
673 frame
->bh
= ext4_read_dirblock(dir
, 0, INDEX
);
674 if (IS_ERR(frame
->bh
))
675 return (struct dx_frame
*) frame
->bh
;
677 root
= (struct dx_root
*) frame
->bh
->b_data
;
678 if (root
->info
.hash_version
!= DX_HASH_TEA
&&
679 root
->info
.hash_version
!= DX_HASH_HALF_MD4
&&
680 root
->info
.hash_version
!= DX_HASH_LEGACY
) {
681 ext4_warning(dir
->i_sb
, "Unrecognised inode hash code %d",
682 root
->info
.hash_version
);
685 hinfo
->hash_version
= root
->info
.hash_version
;
686 if (hinfo
->hash_version
<= DX_HASH_TEA
)
687 hinfo
->hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
688 hinfo
->seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
690 ext4fs_dirhash(d_name
->name
, d_name
->len
, hinfo
);
693 if (root
->info
.unused_flags
& 1) {
694 ext4_warning(dir
->i_sb
, "Unimplemented inode hash flags: %#06x",
695 root
->info
.unused_flags
);
699 if ((indirect
= root
->info
.indirect_levels
) > 1) {
700 ext4_warning(dir
->i_sb
, "Unimplemented inode hash depth: %#06x",
701 root
->info
.indirect_levels
);
705 entries
= (struct dx_entry
*) (((char *)&root
->info
) +
706 root
->info
.info_length
);
708 if (dx_get_limit(entries
) != dx_root_limit(dir
,
709 root
->info
.info_length
)) {
710 ext4_warning(dir
->i_sb
, "dx entry: limit != root limit");
714 dxtrace(printk("Look up %x", hash
));
716 count
= dx_get_count(entries
);
717 if (!count
|| count
> dx_get_limit(entries
)) {
718 ext4_warning(dir
->i_sb
,
719 "dx entry: no count or count > limit");
724 q
= entries
+ count
- 1;
727 dxtrace(printk("."));
728 if (dx_get_hash(m
) > hash
)
734 if (0) { // linear search cross check
735 unsigned n
= count
- 1;
739 dxtrace(printk(","));
740 if (dx_get_hash(++at
) > hash
)
746 assert (at
== p
- 1);
750 dxtrace(printk(" %x->%u\n", at
== entries
? 0: dx_get_hash(at
), dx_get_block(at
)));
751 frame
->entries
= entries
;
756 frame
->bh
= ext4_read_dirblock(dir
, dx_get_block(at
), INDEX
);
757 if (IS_ERR(frame
->bh
)) {
758 ret_err
= (struct dx_frame
*) frame
->bh
;
762 entries
= ((struct dx_node
*) frame
->bh
->b_data
)->entries
;
764 if (dx_get_limit(entries
) != dx_node_limit (dir
)) {
765 ext4_warning(dir
->i_sb
,
766 "dx entry: limit != node limit");
771 while (frame
>= frame_in
) {
775 if (ret_err
== ERR_PTR(ERR_BAD_DX_DIR
))
776 ext4_warning(dir
->i_sb
,
777 "Corrupt dir inode %lu, running e2fsck is "
778 "recommended.", dir
->i_ino
);
782 static void dx_release (struct dx_frame
*frames
)
784 if (frames
[0].bh
== NULL
)
787 if (((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
)
788 brelse(frames
[1].bh
);
789 brelse(frames
[0].bh
);
793 * This function increments the frame pointer to search the next leaf
794 * block, and reads in the necessary intervening nodes if the search
795 * should be necessary. Whether or not the search is necessary is
796 * controlled by the hash parameter. If the hash value is even, then
797 * the search is only continued if the next block starts with that
798 * hash value. This is used if we are searching for a specific file.
800 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
802 * This function returns 1 if the caller should continue to search,
803 * or 0 if it should not. If there is an error reading one of the
804 * index blocks, it will a negative error code.
806 * If start_hash is non-null, it will be filled in with the starting
807 * hash of the next page.
809 static int ext4_htree_next_block(struct inode
*dir
, __u32 hash
,
810 struct dx_frame
*frame
,
811 struct dx_frame
*frames
,
815 struct buffer_head
*bh
;
821 * Find the next leaf page by incrementing the frame pointer.
822 * If we run out of entries in the interior node, loop around and
823 * increment pointer in the parent node. When we break out of
824 * this loop, num_frames indicates the number of interior
825 * nodes need to be read.
828 if (++(p
->at
) < p
->entries
+ dx_get_count(p
->entries
))
837 * If the hash is 1, then continue only if the next page has a
838 * continuation hash of any value. This is used for readdir
839 * handling. Otherwise, check to see if the hash matches the
840 * desired contiuation hash. If it doesn't, return since
841 * there's no point to read in the successive index pages.
843 bhash
= dx_get_hash(p
->at
);
846 if ((hash
& 1) == 0) {
847 if ((bhash
& ~1) != hash
)
851 * If the hash is HASH_NB_ALWAYS, we always go to the next
852 * block so no check is necessary
854 while (num_frames
--) {
855 bh
= ext4_read_dirblock(dir
, dx_get_block(p
->at
), INDEX
);
861 p
->at
= p
->entries
= ((struct dx_node
*) bh
->b_data
)->entries
;
868 * This function fills a red-black tree with information from a
869 * directory block. It returns the number directory entries loaded
870 * into the tree. If there is an error it is returned in err.
872 static int htree_dirblock_to_tree(struct file
*dir_file
,
873 struct inode
*dir
, ext4_lblk_t block
,
874 struct dx_hash_info
*hinfo
,
875 __u32 start_hash
, __u32 start_minor_hash
)
877 struct buffer_head
*bh
;
878 struct ext4_dir_entry_2
*de
, *top
;
879 int err
= 0, count
= 0;
880 struct ext4_str tmp_str
;
882 dxtrace(printk(KERN_INFO
"In htree dirblock_to_tree: block %lu\n",
883 (unsigned long)block
));
884 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
888 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
889 top
= (struct ext4_dir_entry_2
*) ((char *) de
+
890 dir
->i_sb
->s_blocksize
-
891 EXT4_DIR_REC_LEN(0));
892 for (; de
< top
; de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
)) {
893 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
894 bh
->b_data
, bh
->b_size
,
895 (block
<<EXT4_BLOCK_SIZE_BITS(dir
->i_sb
))
896 + ((char *)de
- bh
->b_data
))) {
897 /* silently ignore the rest of the block */
900 ext4fs_dirhash(de
->name
, de
->name_len
, hinfo
);
901 if ((hinfo
->hash
< start_hash
) ||
902 ((hinfo
->hash
== start_hash
) &&
903 (hinfo
->minor_hash
< start_minor_hash
)))
907 tmp_str
.name
= de
->name
;
908 tmp_str
.len
= de
->name_len
;
909 err
= ext4_htree_store_dirent(dir_file
,
910 hinfo
->hash
, hinfo
->minor_hash
, de
, &tmp_str
);
923 * This function fills a red-black tree with information from a
924 * directory. We start scanning the directory in hash order, starting
925 * at start_hash and start_minor_hash.
927 * This function returns the number of entries inserted into the tree,
928 * or a negative error code.
930 int ext4_htree_fill_tree(struct file
*dir_file
, __u32 start_hash
,
931 __u32 start_minor_hash
, __u32
*next_hash
)
933 struct dx_hash_info hinfo
;
934 struct ext4_dir_entry_2
*de
;
935 struct dx_frame frames
[2], *frame
;
941 struct ext4_str tmp_str
;
943 dxtrace(printk(KERN_DEBUG
"In htree_fill_tree, start hash: %x:%x\n",
944 start_hash
, start_minor_hash
));
945 dir
= file_inode(dir_file
);
946 if (!(ext4_test_inode_flag(dir
, EXT4_INODE_INDEX
))) {
947 hinfo
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
948 if (hinfo
.hash_version
<= DX_HASH_TEA
)
949 hinfo
.hash_version
+=
950 EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
951 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
952 if (ext4_has_inline_data(dir
)) {
953 int has_inline_data
= 1;
954 count
= htree_inlinedir_to_tree(dir_file
, dir
, 0,
958 if (has_inline_data
) {
963 count
= htree_dirblock_to_tree(dir_file
, dir
, 0, &hinfo
,
964 start_hash
, start_minor_hash
);
968 hinfo
.hash
= start_hash
;
969 hinfo
.minor_hash
= 0;
970 frame
= dx_probe(NULL
, dir
, &hinfo
, frames
);
972 return PTR_ERR(frame
);
974 /* Add '.' and '..' from the htree header */
975 if (!start_hash
&& !start_minor_hash
) {
976 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
977 tmp_str
.name
= de
->name
;
978 tmp_str
.len
= de
->name_len
;
979 err
= ext4_htree_store_dirent(dir_file
, 0, 0,
985 if (start_hash
< 2 || (start_hash
==2 && start_minor_hash
==0)) {
986 de
= (struct ext4_dir_entry_2
*) frames
[0].bh
->b_data
;
987 de
= ext4_next_entry(de
, dir
->i_sb
->s_blocksize
);
988 tmp_str
.name
= de
->name
;
989 tmp_str
.len
= de
->name_len
;
990 err
= ext4_htree_store_dirent(dir_file
, 2, 0,
998 block
= dx_get_block(frame
->at
);
999 ret
= htree_dirblock_to_tree(dir_file
, dir
, block
, &hinfo
,
1000 start_hash
, start_minor_hash
);
1007 ret
= ext4_htree_next_block(dir
, HASH_NB_ALWAYS
,
1008 frame
, frames
, &hashval
);
1009 *next_hash
= hashval
;
1015 * Stop if: (a) there are no more entries, or
1016 * (b) we have inserted at least one entry and the
1017 * next hash value is not a continuation
1020 (count
&& ((hashval
& 1) == 0)))
1024 dxtrace(printk(KERN_DEBUG
"Fill tree: returned %d entries, "
1025 "next hash: %x\n", count
, *next_hash
));
1032 static inline int search_dirblock(struct buffer_head
*bh
,
1034 const struct qstr
*d_name
,
1035 unsigned int offset
,
1036 struct ext4_dir_entry_2
**res_dir
)
1038 return search_dir(bh
, bh
->b_data
, dir
->i_sb
->s_blocksize
, dir
,
1039 d_name
, offset
, res_dir
);
1043 * Directory block splitting, compacting
1047 * Create map of hash values, offsets, and sizes, stored at end of block.
1048 * Returns number of entries mapped.
1050 static int dx_make_map(struct ext4_dir_entry_2
*de
, unsigned blocksize
,
1051 struct dx_hash_info
*hinfo
,
1052 struct dx_map_entry
*map_tail
)
1055 char *base
= (char *) de
;
1056 struct dx_hash_info h
= *hinfo
;
1058 while ((char *) de
< base
+ blocksize
) {
1059 if (de
->name_len
&& de
->inode
) {
1060 ext4fs_dirhash(de
->name
, de
->name_len
, &h
);
1062 map_tail
->hash
= h
.hash
;
1063 map_tail
->offs
= ((char *) de
- base
)>>2;
1064 map_tail
->size
= le16_to_cpu(de
->rec_len
);
1068 /* XXX: do we need to check rec_len == 0 case? -Chris */
1069 de
= ext4_next_entry(de
, blocksize
);
1074 /* Sort map by hash value */
1075 static void dx_sort_map (struct dx_map_entry
*map
, unsigned count
)
1077 struct dx_map_entry
*p
, *q
, *top
= map
+ count
- 1;
1079 /* Combsort until bubble sort doesn't suck */
1081 count
= count
*10/13;
1082 if (count
- 9 < 2) /* 9, 10 -> 11 */
1084 for (p
= top
, q
= p
- count
; q
>= map
; p
--, q
--)
1085 if (p
->hash
< q
->hash
)
1088 /* Garden variety bubble sort */
1093 if (q
[1].hash
>= q
[0].hash
)
1101 static void dx_insert_block(struct dx_frame
*frame
, u32 hash
, ext4_lblk_t block
)
1103 struct dx_entry
*entries
= frame
->entries
;
1104 struct dx_entry
*old
= frame
->at
, *new = old
+ 1;
1105 int count
= dx_get_count(entries
);
1107 assert(count
< dx_get_limit(entries
));
1108 assert(old
< entries
+ count
);
1109 memmove(new + 1, new, (char *)(entries
+ count
) - (char *)(new));
1110 dx_set_hash(new, hash
);
1111 dx_set_block(new, block
);
1112 dx_set_count(entries
, count
+ 1);
1116 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1118 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1119 * `de != NULL' is guaranteed by caller.
1121 static inline int ext4_match (int len
, const char * const name
,
1122 struct ext4_dir_entry_2
* de
)
1124 if (len
!= de
->name_len
)
1128 return !memcmp(name
, de
->name
, len
);
1132 * Returns 0 if not found, -1 on failure, and 1 on success
1134 int search_dir(struct buffer_head
*bh
,
1138 const struct qstr
*d_name
,
1139 unsigned int offset
,
1140 struct ext4_dir_entry_2
**res_dir
)
1142 struct ext4_dir_entry_2
* de
;
1145 const char *name
= d_name
->name
;
1146 int namelen
= d_name
->len
;
1148 de
= (struct ext4_dir_entry_2
*)search_buf
;
1149 dlimit
= search_buf
+ buf_size
;
1150 while ((char *) de
< dlimit
) {
1151 /* this code is executed quadratically often */
1152 /* do minimal checking `by hand' */
1154 if ((char *) de
+ namelen
<= dlimit
&&
1155 ext4_match (namelen
, name
, de
)) {
1156 /* found a match - just to be sure, do a full check */
1157 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
, bh
->b_data
,
1158 bh
->b_size
, offset
))
1163 /* prevent looping on a bad block */
1164 de_len
= ext4_rec_len_from_disk(de
->rec_len
,
1165 dir
->i_sb
->s_blocksize
);
1169 de
= (struct ext4_dir_entry_2
*) ((char *) de
+ de_len
);
1174 static int is_dx_internal_node(struct inode
*dir
, ext4_lblk_t block
,
1175 struct ext4_dir_entry
*de
)
1177 struct super_block
*sb
= dir
->i_sb
;
1183 if (de
->inode
== 0 &&
1184 ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) ==
1193 * finds an entry in the specified directory with the wanted name. It
1194 * returns the cache buffer in which the entry was found, and the entry
1195 * itself (as a parameter - res_dir). It does NOT read the inode of the
1196 * entry - you'll have to do that yourself if you want to.
1198 * The returned buffer_head has ->b_count elevated. The caller is expected
1199 * to brelse() it when appropriate.
1201 static struct buffer_head
* ext4_find_entry (struct inode
*dir
,
1202 const struct qstr
*d_name
,
1203 struct ext4_dir_entry_2
**res_dir
,
1206 struct super_block
*sb
;
1207 struct buffer_head
*bh_use
[NAMEI_RA_SIZE
];
1208 struct buffer_head
*bh
, *ret
= NULL
;
1209 ext4_lblk_t start
, block
, b
;
1210 const u8
*name
= d_name
->name
;
1211 int ra_max
= 0; /* Number of bh's in the readahead
1213 int ra_ptr
= 0; /* Current index into readahead
1216 ext4_lblk_t nblocks
;
1221 namelen
= d_name
->len
;
1222 if (namelen
> EXT4_NAME_LEN
)
1225 if (ext4_has_inline_data(dir
)) {
1226 int has_inline_data
= 1;
1227 ret
= ext4_find_inline_entry(dir
, d_name
, res_dir
,
1229 if (has_inline_data
) {
1236 if ((namelen
<= 2) && (name
[0] == '.') &&
1237 (name
[1] == '.' || name
[1] == '\0')) {
1239 * "." or ".." will only be in the first block
1240 * NFS may look up ".."; "." should be handled by the VFS
1247 bh
= ext4_dx_find_entry(dir
, d_name
, res_dir
);
1249 * On success, or if the error was file not found,
1250 * return. Otherwise, fall back to doing a search the
1251 * old fashioned way.
1253 if (!IS_ERR(bh
) || PTR_ERR(bh
) != ERR_BAD_DX_DIR
)
1255 dxtrace(printk(KERN_DEBUG
"ext4_find_entry: dx failed, "
1258 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1259 start
= EXT4_I(dir
)->i_dir_start_lookup
;
1260 if (start
>= nblocks
)
1266 * We deal with the read-ahead logic here.
1268 if (ra_ptr
>= ra_max
) {
1269 /* Refill the readahead buffer */
1272 for (ra_max
= 0; ra_max
< NAMEI_RA_SIZE
; ra_max
++) {
1274 * Terminate if we reach the end of the
1275 * directory and must wrap, or if our
1276 * search has finished at this block.
1278 if (b
>= nblocks
|| (num
&& block
== start
)) {
1279 bh_use
[ra_max
] = NULL
;
1283 bh
= ext4_getblk(NULL
, dir
, b
++, 0);
1284 if (unlikely(IS_ERR(bh
))) {
1289 bh_use
[ra_max
] = bh
;
1291 ll_rw_block(READ
| REQ_META
| REQ_PRIO
,
1295 if ((bh
= bh_use
[ra_ptr
++]) == NULL
)
1298 if (!buffer_uptodate(bh
)) {
1299 /* read error, skip block & hope for the best */
1300 EXT4_ERROR_INODE(dir
, "reading directory lblock %lu",
1301 (unsigned long) block
);
1305 if (!buffer_verified(bh
) &&
1306 !is_dx_internal_node(dir
, block
,
1307 (struct ext4_dir_entry
*)bh
->b_data
) &&
1308 !ext4_dirent_csum_verify(dir
,
1309 (struct ext4_dir_entry
*)bh
->b_data
)) {
1310 EXT4_ERROR_INODE(dir
, "checksumming directory "
1311 "block %lu", (unsigned long)block
);
1315 set_buffer_verified(bh
);
1316 i
= search_dirblock(bh
, dir
, d_name
,
1317 block
<< EXT4_BLOCK_SIZE_BITS(sb
), res_dir
);
1319 EXT4_I(dir
)->i_dir_start_lookup
= block
;
1321 goto cleanup_and_exit
;
1325 goto cleanup_and_exit
;
1328 if (++block
>= nblocks
)
1330 } while (block
!= start
);
1333 * If the directory has grown while we were searching, then
1334 * search the last part of the directory before giving up.
1337 nblocks
= dir
->i_size
>> EXT4_BLOCK_SIZE_BITS(sb
);
1338 if (block
< nblocks
) {
1344 /* Clean up the read-ahead blocks */
1345 for (; ra_ptr
< ra_max
; ra_ptr
++)
1346 brelse(bh_use
[ra_ptr
]);
1350 static struct buffer_head
* ext4_dx_find_entry(struct inode
*dir
, const struct qstr
*d_name
,
1351 struct ext4_dir_entry_2
**res_dir
)
1353 struct super_block
* sb
= dir
->i_sb
;
1354 struct dx_hash_info hinfo
;
1355 struct dx_frame frames
[2], *frame
;
1356 struct buffer_head
*bh
;
1360 frame
= dx_probe(d_name
, dir
, &hinfo
, frames
);
1362 return (struct buffer_head
*) frame
;
1364 block
= dx_get_block(frame
->at
);
1365 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1369 retval
= search_dirblock(bh
, dir
, d_name
,
1370 block
<< EXT4_BLOCK_SIZE_BITS(sb
),
1376 bh
= ERR_PTR(ERR_BAD_DX_DIR
);
1380 /* Check to see if we should continue to search */
1381 retval
= ext4_htree_next_block(dir
, hinfo
.hash
, frame
,
1385 "error %d reading index page in directory #%lu",
1386 retval
, dir
->i_ino
);
1387 bh
= ERR_PTR(retval
);
1390 } while (retval
== 1);
1394 dxtrace(printk(KERN_DEBUG
"%s not found\n", d_name
->name
));
1400 static struct dentry
*ext4_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
1402 struct inode
*inode
;
1403 struct ext4_dir_entry_2
*de
;
1404 struct buffer_head
*bh
;
1406 if (dentry
->d_name
.len
> EXT4_NAME_LEN
)
1407 return ERR_PTR(-ENAMETOOLONG
);
1409 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
1411 return (struct dentry
*) bh
;
1414 __u32 ino
= le32_to_cpu(de
->inode
);
1416 if (!ext4_valid_inum(dir
->i_sb
, ino
)) {
1417 EXT4_ERROR_INODE(dir
, "bad inode number: %u", ino
);
1418 return ERR_PTR(-EIO
);
1420 if (unlikely(ino
== dir
->i_ino
)) {
1421 EXT4_ERROR_INODE(dir
, "'%pd' linked to parent dir",
1423 return ERR_PTR(-EIO
);
1425 inode
= ext4_iget_normal(dir
->i_sb
, ino
);
1426 if (inode
== ERR_PTR(-ESTALE
)) {
1427 EXT4_ERROR_INODE(dir
,
1428 "deleted inode referenced: %u",
1430 return ERR_PTR(-EIO
);
1432 if (!IS_ERR(inode
) && ext4_encrypted_inode(dir
) &&
1433 (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
1434 S_ISLNK(inode
->i_mode
)) &&
1435 !ext4_is_child_context_consistent_with_parent(dir
,
1438 ext4_warning(inode
->i_sb
,
1439 "Inconsistent encryption contexts: %lu/%lu\n",
1440 (unsigned long) dir
->i_ino
,
1441 (unsigned long) inode
->i_ino
);
1442 return ERR_PTR(-EPERM
);
1445 return d_splice_alias(inode
, dentry
);
1449 struct dentry
*ext4_get_parent(struct dentry
*child
)
1452 static const struct qstr dotdot
= QSTR_INIT("..", 2);
1453 struct ext4_dir_entry_2
* de
;
1454 struct buffer_head
*bh
;
1456 bh
= ext4_find_entry(child
->d_inode
, &dotdot
, &de
, NULL
);
1458 return (struct dentry
*) bh
;
1460 return ERR_PTR(-ENOENT
);
1461 ino
= le32_to_cpu(de
->inode
);
1464 if (!ext4_valid_inum(child
->d_inode
->i_sb
, ino
)) {
1465 EXT4_ERROR_INODE(child
->d_inode
,
1466 "bad parent inode number: %u", ino
);
1467 return ERR_PTR(-EIO
);
1470 return d_obtain_alias(ext4_iget_normal(child
->d_inode
->i_sb
, ino
));
1474 * Move count entries from end of map between two memory locations.
1475 * Returns pointer to last entry moved.
1477 static struct ext4_dir_entry_2
*
1478 dx_move_dirents(char *from
, char *to
, struct dx_map_entry
*map
, int count
,
1481 unsigned rec_len
= 0;
1484 struct ext4_dir_entry_2
*de
= (struct ext4_dir_entry_2
*)
1485 (from
+ (map
->offs
<<2));
1486 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1487 memcpy (to
, de
, rec_len
);
1488 ((struct ext4_dir_entry_2
*) to
)->rec_len
=
1489 ext4_rec_len_to_disk(rec_len
, blocksize
);
1494 return (struct ext4_dir_entry_2
*) (to
- rec_len
);
1498 * Compact each dir entry in the range to the minimal rec_len.
1499 * Returns pointer to last entry in range.
1501 static struct ext4_dir_entry_2
* dx_pack_dirents(char *base
, unsigned blocksize
)
1503 struct ext4_dir_entry_2
*next
, *to
, *prev
, *de
= (struct ext4_dir_entry_2
*) base
;
1504 unsigned rec_len
= 0;
1507 while ((char*)de
< base
+ blocksize
) {
1508 next
= ext4_next_entry(de
, blocksize
);
1509 if (de
->inode
&& de
->name_len
) {
1510 rec_len
= EXT4_DIR_REC_LEN(de
->name_len
);
1512 memmove(to
, de
, rec_len
);
1513 to
->rec_len
= ext4_rec_len_to_disk(rec_len
, blocksize
);
1515 to
= (struct ext4_dir_entry_2
*) (((char *) to
) + rec_len
);
1523 * Split a full leaf block to make room for a new dir entry.
1524 * Allocate a new block, and move entries so that they are approx. equally full.
1525 * Returns pointer to de in block into which the new entry will be inserted.
1527 static struct ext4_dir_entry_2
*do_split(handle_t
*handle
, struct inode
*dir
,
1528 struct buffer_head
**bh
,struct dx_frame
*frame
,
1529 struct dx_hash_info
*hinfo
)
1531 unsigned blocksize
= dir
->i_sb
->s_blocksize
;
1532 unsigned count
, continued
;
1533 struct buffer_head
*bh2
;
1534 ext4_lblk_t newblock
;
1536 struct dx_map_entry
*map
;
1537 char *data1
= (*bh
)->b_data
, *data2
;
1538 unsigned split
, move
, size
;
1539 struct ext4_dir_entry_2
*de
= NULL
, *de2
;
1540 struct ext4_dir_entry_tail
*t
;
1544 if (ext4_has_metadata_csum(dir
->i_sb
))
1545 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1547 bh2
= ext4_append(handle
, dir
, &newblock
);
1551 return (struct ext4_dir_entry_2
*) bh2
;
1554 BUFFER_TRACE(*bh
, "get_write_access");
1555 err
= ext4_journal_get_write_access(handle
, *bh
);
1559 BUFFER_TRACE(frame
->bh
, "get_write_access");
1560 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
1564 data2
= bh2
->b_data
;
1566 /* create map in the end of data2 block */
1567 map
= (struct dx_map_entry
*) (data2
+ blocksize
);
1568 count
= dx_make_map((struct ext4_dir_entry_2
*) data1
,
1569 blocksize
, hinfo
, map
);
1571 dx_sort_map(map
, count
);
1572 /* Split the existing block in the middle, size-wise */
1575 for (i
= count
-1; i
>= 0; i
--) {
1576 /* is more than half of this entry in 2nd half of the block? */
1577 if (size
+ map
[i
].size
/2 > blocksize
/2)
1579 size
+= map
[i
].size
;
1582 /* map index at which we will split */
1583 split
= count
- move
;
1584 hash2
= map
[split
].hash
;
1585 continued
= hash2
== map
[split
- 1].hash
;
1586 dxtrace(printk(KERN_INFO
"Split block %lu at %x, %i/%i\n",
1587 (unsigned long)dx_get_block(frame
->at
),
1588 hash2
, split
, count
-split
));
1590 /* Fancy dance to stay within two buffers */
1591 de2
= dx_move_dirents(data1
, data2
, map
+ split
, count
- split
, blocksize
);
1592 de
= dx_pack_dirents(data1
, blocksize
);
1593 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1596 de2
->rec_len
= ext4_rec_len_to_disk(data2
+ (blocksize
- csum_size
) -
1600 t
= EXT4_DIRENT_TAIL(data2
, blocksize
);
1601 initialize_dirent_tail(t
, blocksize
);
1603 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1604 initialize_dirent_tail(t
, blocksize
);
1607 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data1
, blocksize
, 1));
1608 dxtrace(dx_show_leaf (hinfo
, (struct ext4_dir_entry_2
*) data2
, blocksize
, 1));
1610 /* Which block gets the new entry? */
1611 if (hinfo
->hash
>= hash2
) {
1615 dx_insert_block(frame
, hash2
+ continued
, newblock
);
1616 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh2
);
1619 err
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1623 dxtrace(dx_show_index("frame", frame
->entries
));
1630 ext4_std_error(dir
->i_sb
, err
);
1631 return ERR_PTR(err
);
1634 int ext4_find_dest_de(struct inode
*dir
, struct inode
*inode
,
1635 struct buffer_head
*bh
,
1636 void *buf
, int buf_size
,
1637 const char *name
, int namelen
,
1638 struct ext4_dir_entry_2
**dest_de
)
1640 struct ext4_dir_entry_2
*de
;
1641 unsigned short reclen
= EXT4_DIR_REC_LEN(namelen
);
1643 unsigned int offset
= 0;
1646 de
= (struct ext4_dir_entry_2
*)buf
;
1647 top
= buf
+ buf_size
- reclen
;
1648 while ((char *) de
<= top
) {
1649 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
1650 buf
, buf_size
, offset
))
1652 if (ext4_match(namelen
, name
, de
))
1654 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1655 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1656 if ((de
->inode
? rlen
- nlen
: rlen
) >= reclen
)
1658 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
1661 if ((char *) de
> top
)
1668 int ext4_insert_dentry(struct inode
*dir
,
1669 struct inode
*inode
,
1670 struct ext4_dir_entry_2
*de
,
1672 const struct qstr
*iname
,
1673 const char *name
, int namelen
)
1677 struct ext4_fname_crypto_ctx
*ctx
= NULL
;
1678 struct ext4_str fname_crypto_str
= {.name
= NULL
, .len
= 0};
1679 struct ext4_str tmp_str
;
1682 ctx
= ext4_get_fname_crypto_ctx(dir
, EXT4_NAME_LEN
);
1685 /* By default, the input name would be written to the disk */
1686 tmp_str
.name
= (unsigned char *)name
;
1687 tmp_str
.len
= namelen
;
1689 /* Directory is encrypted */
1690 res
= ext4_fname_crypto_alloc_buffer(ctx
, EXT4_NAME_LEN
,
1693 ext4_put_fname_crypto_ctx(&ctx
);
1696 res
= ext4_fname_usr_to_disk(ctx
, iname
, &fname_crypto_str
);
1698 ext4_put_fname_crypto_ctx(&ctx
);
1699 ext4_fname_crypto_free_buffer(&fname_crypto_str
);
1702 tmp_str
.name
= fname_crypto_str
.name
;
1703 tmp_str
.len
= fname_crypto_str
.len
;
1706 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
1707 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
1709 struct ext4_dir_entry_2
*de1
=
1710 (struct ext4_dir_entry_2
*)((char *)de
+ nlen
);
1711 de1
->rec_len
= ext4_rec_len_to_disk(rlen
- nlen
, buf_size
);
1712 de
->rec_len
= ext4_rec_len_to_disk(nlen
, buf_size
);
1715 de
->file_type
= EXT4_FT_UNKNOWN
;
1716 de
->inode
= cpu_to_le32(inode
->i_ino
);
1717 ext4_set_de_type(inode
->i_sb
, de
, inode
->i_mode
);
1718 de
->name_len
= tmp_str
.len
;
1720 memcpy(de
->name
, tmp_str
.name
, tmp_str
.len
);
1721 ext4_put_fname_crypto_ctx(&ctx
);
1722 ext4_fname_crypto_free_buffer(&fname_crypto_str
);
1727 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1728 * it points to a directory entry which is guaranteed to be large
1729 * enough for new directory entry. If de is NULL, then
1730 * add_dirent_to_buf will attempt search the directory block for
1731 * space. It will return -ENOSPC if no space is available, and -EIO
1732 * and -EEXIST if directory entry already exists.
1734 static int add_dirent_to_buf(handle_t
*handle
, struct dentry
*dentry
,
1735 struct inode
*inode
, struct ext4_dir_entry_2
*de
,
1736 struct buffer_head
*bh
)
1738 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1739 const char *name
= dentry
->d_name
.name
;
1740 int namelen
= dentry
->d_name
.len
;
1741 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
1745 if (ext4_has_metadata_csum(inode
->i_sb
))
1746 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1749 err
= ext4_find_dest_de(dir
, inode
,
1750 bh
, bh
->b_data
, blocksize
- csum_size
,
1751 name
, namelen
, &de
);
1755 BUFFER_TRACE(bh
, "get_write_access");
1756 err
= ext4_journal_get_write_access(handle
, bh
);
1758 ext4_std_error(dir
->i_sb
, err
);
1762 /* By now the buffer is marked for journaling. Due to crypto operations,
1763 * the following function call may fail */
1764 err
= ext4_insert_dentry(dir
, inode
, de
, blocksize
, &dentry
->d_name
,
1770 * XXX shouldn't update any times until successful
1771 * completion of syscall, but too many callers depend
1774 * XXX similarly, too many callers depend on
1775 * ext4_new_inode() setting the times, but error
1776 * recovery deletes the inode, so the worst that can
1777 * happen is that the times are slightly out of date
1778 * and/or different from the directory change time.
1780 dir
->i_mtime
= dir
->i_ctime
= ext4_current_time(dir
);
1781 ext4_update_dx_flag(dir
);
1783 ext4_mark_inode_dirty(handle
, dir
);
1784 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1785 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1787 ext4_std_error(dir
->i_sb
, err
);
1792 * This converts a one block unindexed directory to a 3 block indexed
1793 * directory, and adds the dentry to the indexed directory.
1795 static int make_indexed_dir(handle_t
*handle
, struct dentry
*dentry
,
1796 struct inode
*inode
, struct buffer_head
*bh
)
1798 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1799 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1800 struct ext4_fname_crypto_ctx
*ctx
= NULL
;
1803 const char *name
= dentry
->d_name
.name
;
1804 int namelen
= dentry
->d_name
.len
;
1806 struct buffer_head
*bh2
;
1807 struct dx_root
*root
;
1808 struct dx_frame frames
[2], *frame
;
1809 struct dx_entry
*entries
;
1810 struct ext4_dir_entry_2
*de
, *de2
;
1811 struct ext4_dir_entry_tail
*t
;
1816 struct dx_hash_info hinfo
;
1818 struct fake_dirent
*fde
;
1821 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1822 ctx
= ext4_get_fname_crypto_ctx(dir
, EXT4_NAME_LEN
);
1824 return PTR_ERR(ctx
);
1827 if (ext4_has_metadata_csum(inode
->i_sb
))
1828 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1830 blocksize
= dir
->i_sb
->s_blocksize
;
1831 dxtrace(printk(KERN_DEBUG
"Creating index: inode %lu\n", dir
->i_ino
));
1832 BUFFER_TRACE(bh
, "get_write_access");
1833 retval
= ext4_journal_get_write_access(handle
, bh
);
1835 ext4_std_error(dir
->i_sb
, retval
);
1839 root
= (struct dx_root
*) bh
->b_data
;
1841 /* The 0th block becomes the root, move the dirents out */
1842 fde
= &root
->dotdot
;
1843 de
= (struct ext4_dir_entry_2
*)((char *)fde
+
1844 ext4_rec_len_from_disk(fde
->rec_len
, blocksize
));
1845 if ((char *) de
>= (((char *) root
) + blocksize
)) {
1846 EXT4_ERROR_INODE(dir
, "invalid rec_len for '..'");
1850 len
= ((char *) root
) + (blocksize
- csum_size
) - (char *) de
;
1852 /* Allocate new block for the 0th block's dirents */
1853 bh2
= ext4_append(handle
, dir
, &block
);
1856 return PTR_ERR(bh2
);
1858 ext4_set_inode_flag(dir
, EXT4_INODE_INDEX
);
1859 data1
= bh2
->b_data
;
1861 memcpy (data1
, de
, len
);
1862 de
= (struct ext4_dir_entry_2
*) data1
;
1864 while ((char *)(de2
= ext4_next_entry(de
, blocksize
)) < top
)
1866 de
->rec_len
= ext4_rec_len_to_disk(data1
+ (blocksize
- csum_size
) -
1871 t
= EXT4_DIRENT_TAIL(data1
, blocksize
);
1872 initialize_dirent_tail(t
, blocksize
);
1875 /* Initialize the root; the dot dirents already exist */
1876 de
= (struct ext4_dir_entry_2
*) (&root
->dotdot
);
1877 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- EXT4_DIR_REC_LEN(2),
1879 memset (&root
->info
, 0, sizeof(root
->info
));
1880 root
->info
.info_length
= sizeof(root
->info
);
1881 root
->info
.hash_version
= EXT4_SB(dir
->i_sb
)->s_def_hash_version
;
1882 entries
= root
->entries
;
1883 dx_set_block(entries
, 1);
1884 dx_set_count(entries
, 1);
1885 dx_set_limit(entries
, dx_root_limit(dir
, sizeof(root
->info
)));
1887 /* Initialize as for dx_probe */
1888 hinfo
.hash_version
= root
->info
.hash_version
;
1889 if (hinfo
.hash_version
<= DX_HASH_TEA
)
1890 hinfo
.hash_version
+= EXT4_SB(dir
->i_sb
)->s_hash_unsigned
;
1891 hinfo
.seed
= EXT4_SB(dir
->i_sb
)->s_hash_seed
;
1892 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1893 res
= ext4_fname_usr_to_hash(ctx
, &dentry
->d_name
, &hinfo
);
1895 ext4_put_fname_crypto_ctx(&ctx
);
1896 ext4_mark_inode_dirty(handle
, dir
);
1900 ext4_put_fname_crypto_ctx(&ctx
);
1902 ext4fs_dirhash(name
, namelen
, &hinfo
);
1904 memset(frames
, 0, sizeof(frames
));
1906 frame
->entries
= entries
;
1907 frame
->at
= entries
;
1911 retval
= ext4_handle_dirty_dx_node(handle
, dir
, frame
->bh
);
1914 retval
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
1918 de
= do_split(handle
,dir
, &bh
, frame
, &hinfo
);
1920 retval
= PTR_ERR(de
);
1925 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
1930 * Even if the block split failed, we have to properly write
1931 * out all the changes we did so far. Otherwise we can end up
1932 * with corrupted filesystem.
1934 ext4_mark_inode_dirty(handle
, dir
);
1942 * adds a file entry to the specified directory, using the same
1943 * semantics as ext4_find_entry(). It returns NULL if it failed.
1945 * NOTE!! The inode part of 'de' is left at 0 - which means you
1946 * may not sleep between calling this and putting something into
1947 * the entry, as someone else might have used it while you slept.
1949 static int ext4_add_entry(handle_t
*handle
, struct dentry
*dentry
,
1950 struct inode
*inode
)
1952 struct inode
*dir
= dentry
->d_parent
->d_inode
;
1953 struct buffer_head
*bh
= NULL
;
1954 struct ext4_dir_entry_2
*de
;
1955 struct ext4_dir_entry_tail
*t
;
1956 struct super_block
*sb
;
1960 ext4_lblk_t block
, blocks
;
1963 if (ext4_has_metadata_csum(inode
->i_sb
))
1964 csum_size
= sizeof(struct ext4_dir_entry_tail
);
1967 blocksize
= sb
->s_blocksize
;
1968 if (!dentry
->d_name
.len
)
1971 if (ext4_has_inline_data(dir
)) {
1972 retval
= ext4_try_add_inline_entry(handle
, dentry
, inode
);
1982 retval
= ext4_dx_add_entry(handle
, dentry
, inode
);
1983 if (!retval
|| (retval
!= ERR_BAD_DX_DIR
))
1985 ext4_clear_inode_flag(dir
, EXT4_INODE_INDEX
);
1987 ext4_mark_inode_dirty(handle
, dir
);
1989 blocks
= dir
->i_size
>> sb
->s_blocksize_bits
;
1990 for (block
= 0; block
< blocks
; block
++) {
1991 bh
= ext4_read_dirblock(dir
, block
, DIRENT
);
1995 retval
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
1996 if (retval
!= -ENOSPC
)
1999 if (blocks
== 1 && !dx_fallback
&&
2000 EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_DIR_INDEX
)) {
2001 retval
= make_indexed_dir(handle
, dentry
, inode
, bh
);
2002 bh
= NULL
; /* make_indexed_dir releases bh */
2007 bh
= ext4_append(handle
, dir
, &block
);
2010 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2012 de
->rec_len
= ext4_rec_len_to_disk(blocksize
- csum_size
, blocksize
);
2015 t
= EXT4_DIRENT_TAIL(bh
->b_data
, blocksize
);
2016 initialize_dirent_tail(t
, blocksize
);
2019 retval
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
2023 ext4_set_inode_state(inode
, EXT4_STATE_NEWENTRY
);
2028 * Returns 0 for success, or a negative error value
2030 static int ext4_dx_add_entry(handle_t
*handle
, struct dentry
*dentry
,
2031 struct inode
*inode
)
2033 struct dx_frame frames
[2], *frame
;
2034 struct dx_entry
*entries
, *at
;
2035 struct dx_hash_info hinfo
;
2036 struct buffer_head
*bh
;
2037 struct inode
*dir
= dentry
->d_parent
->d_inode
;
2038 struct super_block
*sb
= dir
->i_sb
;
2039 struct ext4_dir_entry_2
*de
;
2042 frame
= dx_probe(&dentry
->d_name
, dir
, &hinfo
, frames
);
2044 return PTR_ERR(frame
);
2045 entries
= frame
->entries
;
2047 bh
= ext4_read_dirblock(dir
, dx_get_block(frame
->at
), DIRENT
);
2054 BUFFER_TRACE(bh
, "get_write_access");
2055 err
= ext4_journal_get_write_access(handle
, bh
);
2059 err
= add_dirent_to_buf(handle
, dentry
, inode
, NULL
, bh
);
2063 /* Block full, should compress but for now just split */
2064 dxtrace(printk(KERN_DEBUG
"using %u of %u node entries\n",
2065 dx_get_count(entries
), dx_get_limit(entries
)));
2066 /* Need to split index? */
2067 if (dx_get_count(entries
) == dx_get_limit(entries
)) {
2068 ext4_lblk_t newblock
;
2069 unsigned icount
= dx_get_count(entries
);
2070 int levels
= frame
- frames
;
2071 struct dx_entry
*entries2
;
2072 struct dx_node
*node2
;
2073 struct buffer_head
*bh2
;
2075 if (levels
&& (dx_get_count(frames
->entries
) ==
2076 dx_get_limit(frames
->entries
))) {
2077 ext4_warning(sb
, "Directory index full!");
2081 bh2
= ext4_append(handle
, dir
, &newblock
);
2086 node2
= (struct dx_node
*)(bh2
->b_data
);
2087 entries2
= node2
->entries
;
2088 memset(&node2
->fake
, 0, sizeof(struct fake_dirent
));
2089 node2
->fake
.rec_len
= ext4_rec_len_to_disk(sb
->s_blocksize
,
2091 BUFFER_TRACE(frame
->bh
, "get_write_access");
2092 err
= ext4_journal_get_write_access(handle
, frame
->bh
);
2096 unsigned icount1
= icount
/2, icount2
= icount
- icount1
;
2097 unsigned hash2
= dx_get_hash(entries
+ icount1
);
2098 dxtrace(printk(KERN_DEBUG
"Split index %i/%i\n",
2101 BUFFER_TRACE(frame
->bh
, "get_write_access"); /* index root */
2102 err
= ext4_journal_get_write_access(handle
,
2107 memcpy((char *) entries2
, (char *) (entries
+ icount1
),
2108 icount2
* sizeof(struct dx_entry
));
2109 dx_set_count(entries
, icount1
);
2110 dx_set_count(entries2
, icount2
);
2111 dx_set_limit(entries2
, dx_node_limit(dir
));
2113 /* Which index block gets the new entry? */
2114 if (at
- entries
>= icount1
) {
2115 frame
->at
= at
= at
- entries
- icount1
+ entries2
;
2116 frame
->entries
= entries
= entries2
;
2117 swap(frame
->bh
, bh2
);
2119 dx_insert_block(frames
+ 0, hash2
, newblock
);
2120 dxtrace(dx_show_index("node", frames
[1].entries
));
2121 dxtrace(dx_show_index("node",
2122 ((struct dx_node
*) bh2
->b_data
)->entries
));
2123 err
= ext4_handle_dirty_dx_node(handle
, dir
, bh2
);
2128 dxtrace(printk(KERN_DEBUG
2129 "Creating second level index...\n"));
2130 memcpy((char *) entries2
, (char *) entries
,
2131 icount
* sizeof(struct dx_entry
));
2132 dx_set_limit(entries2
, dx_node_limit(dir
));
2135 dx_set_count(entries
, 1);
2136 dx_set_block(entries
+ 0, newblock
);
2137 ((struct dx_root
*) frames
[0].bh
->b_data
)->info
.indirect_levels
= 1;
2139 /* Add new access path frame */
2141 frame
->at
= at
= at
- entries
+ entries2
;
2142 frame
->entries
= entries
= entries2
;
2144 err
= ext4_journal_get_write_access(handle
,
2149 err
= ext4_handle_dirty_dx_node(handle
, dir
, frames
[0].bh
);
2151 ext4_std_error(inode
->i_sb
, err
);
2155 de
= do_split(handle
, dir
, &bh
, frame
, &hinfo
);
2160 err
= add_dirent_to_buf(handle
, dentry
, inode
, de
, bh
);
2164 ext4_std_error(dir
->i_sb
, err
);
2172 * ext4_generic_delete_entry deletes a directory entry by merging it
2173 * with the previous entry
2175 int ext4_generic_delete_entry(handle_t
*handle
,
2177 struct ext4_dir_entry_2
*de_del
,
2178 struct buffer_head
*bh
,
2183 struct ext4_dir_entry_2
*de
, *pde
;
2184 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2189 de
= (struct ext4_dir_entry_2
*)entry_buf
;
2190 while (i
< buf_size
- csum_size
) {
2191 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
2192 bh
->b_data
, bh
->b_size
, i
))
2196 pde
->rec_len
= ext4_rec_len_to_disk(
2197 ext4_rec_len_from_disk(pde
->rec_len
,
2199 ext4_rec_len_from_disk(de
->rec_len
,
2207 i
+= ext4_rec_len_from_disk(de
->rec_len
, blocksize
);
2209 de
= ext4_next_entry(de
, blocksize
);
2214 static int ext4_delete_entry(handle_t
*handle
,
2216 struct ext4_dir_entry_2
*de_del
,
2217 struct buffer_head
*bh
)
2219 int err
, csum_size
= 0;
2221 if (ext4_has_inline_data(dir
)) {
2222 int has_inline_data
= 1;
2223 err
= ext4_delete_inline_entry(handle
, dir
, de_del
, bh
,
2225 if (has_inline_data
)
2229 if (ext4_has_metadata_csum(dir
->i_sb
))
2230 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2232 BUFFER_TRACE(bh
, "get_write_access");
2233 err
= ext4_journal_get_write_access(handle
, bh
);
2237 err
= ext4_generic_delete_entry(handle
, dir
, de_del
,
2239 dir
->i_sb
->s_blocksize
, csum_size
);
2243 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
2244 err
= ext4_handle_dirty_dirent_node(handle
, dir
, bh
);
2251 ext4_std_error(dir
->i_sb
, err
);
2256 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2257 * since this indicates that nlinks count was previously 1.
2259 static void ext4_inc_count(handle_t
*handle
, struct inode
*inode
)
2262 if (is_dx(inode
) && inode
->i_nlink
> 1) {
2263 /* limit is 16-bit i_links_count */
2264 if (inode
->i_nlink
>= EXT4_LINK_MAX
|| inode
->i_nlink
== 2) {
2265 set_nlink(inode
, 1);
2266 EXT4_SET_RO_COMPAT_FEATURE(inode
->i_sb
,
2267 EXT4_FEATURE_RO_COMPAT_DIR_NLINK
);
2273 * If a directory had nlink == 1, then we should let it be 1. This indicates
2274 * directory has >EXT4_LINK_MAX subdirs.
2276 static void ext4_dec_count(handle_t
*handle
, struct inode
*inode
)
2278 if (!S_ISDIR(inode
->i_mode
) || inode
->i_nlink
> 2)
2283 static int ext4_add_nondir(handle_t
*handle
,
2284 struct dentry
*dentry
, struct inode
*inode
)
2286 int err
= ext4_add_entry(handle
, dentry
, inode
);
2288 ext4_mark_inode_dirty(handle
, inode
);
2289 unlock_new_inode(inode
);
2290 d_instantiate(dentry
, inode
);
2294 unlock_new_inode(inode
);
2300 * By the time this is called, we already have created
2301 * the directory cache entry for the new file, but it
2302 * is so far negative - it has no inode.
2304 * If the create succeeds, we fill in the inode information
2305 * with d_instantiate().
2307 static int ext4_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2311 struct inode
*inode
;
2312 int err
, credits
, retries
= 0;
2314 dquot_initialize(dir
);
2316 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2317 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2319 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2320 NULL
, EXT4_HT_DIR
, credits
);
2321 handle
= ext4_journal_current_handle();
2322 err
= PTR_ERR(inode
);
2323 if (!IS_ERR(inode
)) {
2324 inode
->i_op
= &ext4_file_inode_operations
;
2325 if (test_opt(inode
->i_sb
, DAX
))
2326 inode
->i_fop
= &ext4_dax_file_operations
;
2328 inode
->i_fop
= &ext4_file_operations
;
2329 ext4_set_aops(inode
);
2331 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2332 if (!err
&& ext4_encrypted_inode(dir
)) {
2333 err
= ext4_inherit_context(dir
, inode
);
2336 unlock_new_inode(inode
);
2342 err
= ext4_add_nondir(handle
, dentry
, inode
);
2343 if (!err
&& IS_DIRSYNC(dir
))
2344 ext4_handle_sync(handle
);
2347 ext4_journal_stop(handle
);
2348 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2353 static int ext4_mknod(struct inode
*dir
, struct dentry
*dentry
,
2354 umode_t mode
, dev_t rdev
)
2357 struct inode
*inode
;
2358 int err
, credits
, retries
= 0;
2360 if (!new_valid_dev(rdev
))
2363 dquot_initialize(dir
);
2365 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2366 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2368 inode
= ext4_new_inode_start_handle(dir
, mode
, &dentry
->d_name
, 0,
2369 NULL
, EXT4_HT_DIR
, credits
);
2370 handle
= ext4_journal_current_handle();
2371 err
= PTR_ERR(inode
);
2372 if (!IS_ERR(inode
)) {
2373 init_special_inode(inode
, inode
->i_mode
, rdev
);
2374 inode
->i_op
= &ext4_special_inode_operations
;
2375 err
= ext4_add_nondir(handle
, dentry
, inode
);
2376 if (!err
&& IS_DIRSYNC(dir
))
2377 ext4_handle_sync(handle
);
2380 ext4_journal_stop(handle
);
2381 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2386 static int ext4_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2389 struct inode
*inode
;
2390 int err
, retries
= 0;
2392 dquot_initialize(dir
);
2395 inode
= ext4_new_inode_start_handle(dir
, mode
,
2398 EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2399 4 + EXT4_XATTR_TRANS_BLOCKS
);
2400 handle
= ext4_journal_current_handle();
2401 err
= PTR_ERR(inode
);
2402 if (!IS_ERR(inode
)) {
2403 inode
->i_op
= &ext4_file_inode_operations
;
2404 if (test_opt(inode
->i_sb
, DAX
))
2405 inode
->i_fop
= &ext4_dax_file_operations
;
2407 inode
->i_fop
= &ext4_file_operations
;
2408 ext4_set_aops(inode
);
2409 d_tmpfile(dentry
, inode
);
2410 err
= ext4_orphan_add(handle
, inode
);
2412 goto err_unlock_inode
;
2413 mark_inode_dirty(inode
);
2414 unlock_new_inode(inode
);
2417 ext4_journal_stop(handle
);
2418 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2422 ext4_journal_stop(handle
);
2423 unlock_new_inode(inode
);
2427 struct ext4_dir_entry_2
*ext4_init_dot_dotdot(struct inode
*inode
,
2428 struct ext4_dir_entry_2
*de
,
2429 int blocksize
, int csum_size
,
2430 unsigned int parent_ino
, int dotdot_real_len
)
2432 de
->inode
= cpu_to_le32(inode
->i_ino
);
2434 de
->rec_len
= ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de
->name_len
),
2436 strcpy(de
->name
, ".");
2437 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2439 de
= ext4_next_entry(de
, blocksize
);
2440 de
->inode
= cpu_to_le32(parent_ino
);
2442 if (!dotdot_real_len
)
2443 de
->rec_len
= ext4_rec_len_to_disk(blocksize
-
2444 (csum_size
+ EXT4_DIR_REC_LEN(1)),
2447 de
->rec_len
= ext4_rec_len_to_disk(
2448 EXT4_DIR_REC_LEN(de
->name_len
), blocksize
);
2449 strcpy(de
->name
, "..");
2450 ext4_set_de_type(inode
->i_sb
, de
, S_IFDIR
);
2452 return ext4_next_entry(de
, blocksize
);
2455 static int ext4_init_new_dir(handle_t
*handle
, struct inode
*dir
,
2456 struct inode
*inode
)
2458 struct buffer_head
*dir_block
= NULL
;
2459 struct ext4_dir_entry_2
*de
;
2460 struct ext4_dir_entry_tail
*t
;
2461 ext4_lblk_t block
= 0;
2462 unsigned int blocksize
= dir
->i_sb
->s_blocksize
;
2466 if (ext4_has_metadata_csum(dir
->i_sb
))
2467 csum_size
= sizeof(struct ext4_dir_entry_tail
);
2469 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2470 err
= ext4_try_create_inline_dir(handle
, dir
, inode
);
2471 if (err
< 0 && err
!= -ENOSPC
)
2478 dir_block
= ext4_append(handle
, inode
, &block
);
2479 if (IS_ERR(dir_block
))
2480 return PTR_ERR(dir_block
);
2481 de
= (struct ext4_dir_entry_2
*)dir_block
->b_data
;
2482 ext4_init_dot_dotdot(inode
, de
, blocksize
, csum_size
, dir
->i_ino
, 0);
2483 set_nlink(inode
, 2);
2485 t
= EXT4_DIRENT_TAIL(dir_block
->b_data
, blocksize
);
2486 initialize_dirent_tail(t
, blocksize
);
2489 BUFFER_TRACE(dir_block
, "call ext4_handle_dirty_metadata");
2490 err
= ext4_handle_dirty_dirent_node(handle
, inode
, dir_block
);
2493 set_buffer_verified(dir_block
);
2499 static int ext4_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2502 struct inode
*inode
;
2503 int err
, credits
, retries
= 0;
2505 if (EXT4_DIR_LINK_MAX(dir
))
2508 dquot_initialize(dir
);
2510 credits
= (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2511 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3);
2513 inode
= ext4_new_inode_start_handle(dir
, S_IFDIR
| mode
,
2515 0, NULL
, EXT4_HT_DIR
, credits
);
2516 handle
= ext4_journal_current_handle();
2517 err
= PTR_ERR(inode
);
2521 inode
->i_op
= &ext4_dir_inode_operations
;
2522 inode
->i_fop
= &ext4_dir_operations
;
2523 err
= ext4_init_new_dir(handle
, dir
, inode
);
2525 goto out_clear_inode
;
2526 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2527 if (ext4_encrypted_inode(dir
)) {
2528 err
= ext4_inherit_context(dir
, inode
);
2530 goto out_clear_inode
;
2533 err
= ext4_mark_inode_dirty(handle
, inode
);
2535 err
= ext4_add_entry(handle
, dentry
, inode
);
2539 unlock_new_inode(inode
);
2540 ext4_mark_inode_dirty(handle
, inode
);
2544 ext4_inc_count(handle
, dir
);
2545 ext4_update_dx_flag(dir
);
2546 err
= ext4_mark_inode_dirty(handle
, dir
);
2548 goto out_clear_inode
;
2549 unlock_new_inode(inode
);
2550 d_instantiate(dentry
, inode
);
2551 if (IS_DIRSYNC(dir
))
2552 ext4_handle_sync(handle
);
2556 ext4_journal_stop(handle
);
2557 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
2563 * routine to check that the specified directory is empty (for rmdir)
2565 int ext4_empty_dir(struct inode
*inode
)
2567 unsigned int offset
;
2568 struct buffer_head
*bh
;
2569 struct ext4_dir_entry_2
*de
, *de1
;
2570 struct super_block
*sb
;
2573 if (ext4_has_inline_data(inode
)) {
2574 int has_inline_data
= 1;
2576 err
= empty_inline_dir(inode
, &has_inline_data
);
2577 if (has_inline_data
)
2582 if (inode
->i_size
< EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2583 EXT4_ERROR_INODE(inode
, "invalid size");
2586 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
2590 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2591 de1
= ext4_next_entry(de
, sb
->s_blocksize
);
2592 if (le32_to_cpu(de
->inode
) != inode
->i_ino
||
2593 !le32_to_cpu(de1
->inode
) ||
2594 strcmp(".", de
->name
) ||
2595 strcmp("..", de1
->name
)) {
2596 ext4_warning(inode
->i_sb
,
2597 "bad directory (dir #%lu) - no `.' or `..'",
2602 offset
= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
) +
2603 ext4_rec_len_from_disk(de1
->rec_len
, sb
->s_blocksize
);
2604 de
= ext4_next_entry(de1
, sb
->s_blocksize
);
2605 while (offset
< inode
->i_size
) {
2606 if ((void *) de
>= (void *) (bh
->b_data
+sb
->s_blocksize
)) {
2607 unsigned int lblock
;
2610 lblock
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
2611 bh
= ext4_read_dirblock(inode
, lblock
, EITHER
);
2614 de
= (struct ext4_dir_entry_2
*) bh
->b_data
;
2616 if (ext4_check_dir_entry(inode
, NULL
, de
, bh
,
2617 bh
->b_data
, bh
->b_size
, offset
)) {
2618 de
= (struct ext4_dir_entry_2
*)(bh
->b_data
+
2620 offset
= (offset
| (sb
->s_blocksize
- 1)) + 1;
2623 if (le32_to_cpu(de
->inode
)) {
2627 offset
+= ext4_rec_len_from_disk(de
->rec_len
, sb
->s_blocksize
);
2628 de
= ext4_next_entry(de
, sb
->s_blocksize
);
2635 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2636 * such inodes, starting at the superblock, in case we crash before the
2637 * file is closed/deleted, or in case the inode truncate spans multiple
2638 * transactions and the last transaction is not recovered after a crash.
2640 * At filesystem recovery time, we walk this list deleting unlinked
2641 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2643 * Orphan list manipulation functions must be called under i_mutex unless
2644 * we are just creating the inode or deleting it.
2646 int ext4_orphan_add(handle_t
*handle
, struct inode
*inode
)
2648 struct super_block
*sb
= inode
->i_sb
;
2649 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2650 struct ext4_iloc iloc
;
2654 if (!sbi
->s_journal
|| is_bad_inode(inode
))
2657 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2658 !mutex_is_locked(&inode
->i_mutex
));
2660 * Exit early if inode already is on orphan list. This is a big speedup
2661 * since we don't have to contend on the global s_orphan_lock.
2663 if (!list_empty(&EXT4_I(inode
)->i_orphan
))
2667 * Orphan handling is only valid for files with data blocks
2668 * being truncated, or files being unlinked. Note that we either
2669 * hold i_mutex, or the inode can not be referenced from outside,
2670 * so i_nlink should not be bumped due to race
2672 J_ASSERT((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2673 S_ISLNK(inode
->i_mode
)) || inode
->i_nlink
== 0);
2675 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2676 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2680 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2684 mutex_lock(&sbi
->s_orphan_lock
);
2686 * Due to previous errors inode may be already a part of on-disk
2687 * orphan list. If so skip on-disk list modification.
2689 if (!NEXT_ORPHAN(inode
) || NEXT_ORPHAN(inode
) >
2690 (le32_to_cpu(sbi
->s_es
->s_inodes_count
))) {
2691 /* Insert this inode at the head of the on-disk orphan list */
2692 NEXT_ORPHAN(inode
) = le32_to_cpu(sbi
->s_es
->s_last_orphan
);
2693 sbi
->s_es
->s_last_orphan
= cpu_to_le32(inode
->i_ino
);
2696 list_add(&EXT4_I(inode
)->i_orphan
, &sbi
->s_orphan
);
2697 mutex_unlock(&sbi
->s_orphan_lock
);
2700 err
= ext4_handle_dirty_super(handle
, sb
);
2701 rc
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2706 * We have to remove inode from in-memory list if
2707 * addition to on disk orphan list failed. Stray orphan
2708 * list entries can cause panics at unmount time.
2710 mutex_lock(&sbi
->s_orphan_lock
);
2711 list_del(&EXT4_I(inode
)->i_orphan
);
2712 mutex_unlock(&sbi
->s_orphan_lock
);
2715 jbd_debug(4, "superblock will point to %lu\n", inode
->i_ino
);
2716 jbd_debug(4, "orphan inode %lu will point to %d\n",
2717 inode
->i_ino
, NEXT_ORPHAN(inode
));
2719 ext4_std_error(sb
, err
);
2724 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2725 * of such inodes stored on disk, because it is finally being cleaned up.
2727 int ext4_orphan_del(handle_t
*handle
, struct inode
*inode
)
2729 struct list_head
*prev
;
2730 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2731 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2733 struct ext4_iloc iloc
;
2736 if (!sbi
->s_journal
&& !(sbi
->s_mount_state
& EXT4_ORPHAN_FS
))
2739 WARN_ON_ONCE(!(inode
->i_state
& (I_NEW
| I_FREEING
)) &&
2740 !mutex_is_locked(&inode
->i_mutex
));
2741 /* Do this quick check before taking global s_orphan_lock. */
2742 if (list_empty(&ei
->i_orphan
))
2746 /* Grab inode buffer early before taking global s_orphan_lock */
2747 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
2750 mutex_lock(&sbi
->s_orphan_lock
);
2751 jbd_debug(4, "remove inode %lu from orphan list\n", inode
->i_ino
);
2753 prev
= ei
->i_orphan
.prev
;
2754 list_del_init(&ei
->i_orphan
);
2756 /* If we're on an error path, we may not have a valid
2757 * transaction handle with which to update the orphan list on
2758 * disk, but we still need to remove the inode from the linked
2759 * list in memory. */
2760 if (!handle
|| err
) {
2761 mutex_unlock(&sbi
->s_orphan_lock
);
2765 ino_next
= NEXT_ORPHAN(inode
);
2766 if (prev
== &sbi
->s_orphan
) {
2767 jbd_debug(4, "superblock will point to %u\n", ino_next
);
2768 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
2769 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
2771 mutex_unlock(&sbi
->s_orphan_lock
);
2774 sbi
->s_es
->s_last_orphan
= cpu_to_le32(ino_next
);
2775 mutex_unlock(&sbi
->s_orphan_lock
);
2776 err
= ext4_handle_dirty_super(handle
, inode
->i_sb
);
2778 struct ext4_iloc iloc2
;
2779 struct inode
*i_prev
=
2780 &list_entry(prev
, struct ext4_inode_info
, i_orphan
)->vfs_inode
;
2782 jbd_debug(4, "orphan inode %lu will point to %u\n",
2783 i_prev
->i_ino
, ino_next
);
2784 err
= ext4_reserve_inode_write(handle
, i_prev
, &iloc2
);
2786 mutex_unlock(&sbi
->s_orphan_lock
);
2789 NEXT_ORPHAN(i_prev
) = ino_next
;
2790 err
= ext4_mark_iloc_dirty(handle
, i_prev
, &iloc2
);
2791 mutex_unlock(&sbi
->s_orphan_lock
);
2795 NEXT_ORPHAN(inode
) = 0;
2796 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
2798 ext4_std_error(inode
->i_sb
, err
);
2806 static int ext4_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2809 struct inode
*inode
;
2810 struct buffer_head
*bh
;
2811 struct ext4_dir_entry_2
*de
;
2812 handle_t
*handle
= NULL
;
2814 /* Initialize quotas before so that eventual writes go in
2815 * separate transaction */
2816 dquot_initialize(dir
);
2817 dquot_initialize(dentry
->d_inode
);
2820 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2826 inode
= dentry
->d_inode
;
2829 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2832 retval
= -ENOTEMPTY
;
2833 if (!ext4_empty_dir(inode
))
2836 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2837 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2838 if (IS_ERR(handle
)) {
2839 retval
= PTR_ERR(handle
);
2844 if (IS_DIRSYNC(dir
))
2845 ext4_handle_sync(handle
);
2847 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2850 if (!EXT4_DIR_LINK_EMPTY(inode
))
2851 ext4_warning(inode
->i_sb
,
2852 "empty directory has too many links (%d)",
2856 /* There's no need to set i_disksize: the fact that i_nlink is
2857 * zero will ensure that the right thing happens during any
2860 ext4_orphan_add(handle
, inode
);
2861 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(inode
);
2862 ext4_mark_inode_dirty(handle
, inode
);
2863 ext4_dec_count(handle
, dir
);
2864 ext4_update_dx_flag(dir
);
2865 ext4_mark_inode_dirty(handle
, dir
);
2870 ext4_journal_stop(handle
);
2874 static int ext4_unlink(struct inode
*dir
, struct dentry
*dentry
)
2877 struct inode
*inode
;
2878 struct buffer_head
*bh
;
2879 struct ext4_dir_entry_2
*de
;
2880 handle_t
*handle
= NULL
;
2882 trace_ext4_unlink_enter(dir
, dentry
);
2883 /* Initialize quotas before so that eventual writes go
2884 * in separate transaction */
2885 dquot_initialize(dir
);
2886 dquot_initialize(dentry
->d_inode
);
2889 bh
= ext4_find_entry(dir
, &dentry
->d_name
, &de
, NULL
);
2895 inode
= dentry
->d_inode
;
2898 if (le32_to_cpu(de
->inode
) != inode
->i_ino
)
2901 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
2902 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
));
2903 if (IS_ERR(handle
)) {
2904 retval
= PTR_ERR(handle
);
2909 if (IS_DIRSYNC(dir
))
2910 ext4_handle_sync(handle
);
2912 if (!inode
->i_nlink
) {
2913 ext4_warning(inode
->i_sb
,
2914 "Deleting nonexistent file (%lu), %d",
2915 inode
->i_ino
, inode
->i_nlink
);
2916 set_nlink(inode
, 1);
2918 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
2921 dir
->i_ctime
= dir
->i_mtime
= ext4_current_time(dir
);
2922 ext4_update_dx_flag(dir
);
2923 ext4_mark_inode_dirty(handle
, dir
);
2925 if (!inode
->i_nlink
)
2926 ext4_orphan_add(handle
, inode
);
2927 inode
->i_ctime
= ext4_current_time(inode
);
2928 ext4_mark_inode_dirty(handle
, inode
);
2933 ext4_journal_stop(handle
);
2934 trace_ext4_unlink_exit(dentry
, retval
);
2938 static int ext4_symlink(struct inode
*dir
,
2939 struct dentry
*dentry
, const char *symname
)
2942 struct inode
*inode
;
2943 int l
, err
, retries
= 0;
2946 l
= strlen(symname
)+1;
2947 if (l
> dir
->i_sb
->s_blocksize
)
2948 return -ENAMETOOLONG
;
2950 dquot_initialize(dir
);
2952 if (l
> EXT4_N_BLOCKS
* 4) {
2954 * For non-fast symlinks, we just allocate inode and put it on
2955 * orphan list in the first transaction => we need bitmap,
2956 * group descriptor, sb, inode block, quota blocks, and
2957 * possibly selinux xattr blocks.
2959 credits
= 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir
->i_sb
) +
2960 EXT4_XATTR_TRANS_BLOCKS
;
2963 * Fast symlink. We have to add entry to directory
2964 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2965 * allocate new inode (bitmap, group descriptor, inode block,
2966 * quota blocks, sb is already counted in previous macros).
2968 credits
= EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
2969 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 3;
2972 inode
= ext4_new_inode_start_handle(dir
, S_IFLNK
|S_IRWXUGO
,
2973 &dentry
->d_name
, 0, NULL
,
2974 EXT4_HT_DIR
, credits
);
2975 handle
= ext4_journal_current_handle();
2976 err
= PTR_ERR(inode
);
2980 if (l
> EXT4_N_BLOCKS
* 4) {
2981 inode
->i_op
= &ext4_symlink_inode_operations
;
2982 ext4_set_aops(inode
);
2984 * We cannot call page_symlink() with transaction started
2985 * because it calls into ext4_write_begin() which can wait
2986 * for transaction commit if we are running out of space
2987 * and thus we deadlock. So we have to stop transaction now
2988 * and restart it when symlink contents is written.
2990 * To keep fs consistent in case of crash, we have to put inode
2991 * to orphan list in the mean time.
2994 err
= ext4_orphan_add(handle
, inode
);
2995 ext4_journal_stop(handle
);
2997 goto err_drop_inode
;
2998 err
= __page_symlink(inode
, symname
, l
, 1);
3000 goto err_drop_inode
;
3002 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3003 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3005 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3006 EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3007 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 1);
3008 if (IS_ERR(handle
)) {
3009 err
= PTR_ERR(handle
);
3010 goto err_drop_inode
;
3012 set_nlink(inode
, 1);
3013 err
= ext4_orphan_del(handle
, inode
);
3015 ext4_journal_stop(handle
);
3017 goto err_drop_inode
;
3020 /* clear the extent format for fast symlink */
3021 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
3022 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3023 memcpy((char *)&EXT4_I(inode
)->i_data
, symname
, l
);
3024 inode
->i_size
= l
-1;
3026 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3027 err
= ext4_add_nondir(handle
, dentry
, inode
);
3028 if (!err
&& IS_DIRSYNC(dir
))
3029 ext4_handle_sync(handle
);
3033 ext4_journal_stop(handle
);
3034 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3038 unlock_new_inode(inode
);
3043 static int ext4_link(struct dentry
*old_dentry
,
3044 struct inode
*dir
, struct dentry
*dentry
)
3047 struct inode
*inode
= old_dentry
->d_inode
;
3048 int err
, retries
= 0;
3050 if (inode
->i_nlink
>= EXT4_LINK_MAX
)
3052 if (ext4_encrypted_inode(dir
) &&
3053 !ext4_is_child_context_consistent_with_parent(dir
, inode
))
3055 dquot_initialize(dir
);
3058 handle
= ext4_journal_start(dir
, EXT4_HT_DIR
,
3059 (EXT4_DATA_TRANS_BLOCKS(dir
->i_sb
) +
3060 EXT4_INDEX_EXTRA_TRANS_BLOCKS
) + 1);
3062 return PTR_ERR(handle
);
3064 if (IS_DIRSYNC(dir
))
3065 ext4_handle_sync(handle
);
3067 inode
->i_ctime
= ext4_current_time(inode
);
3068 ext4_inc_count(handle
, inode
);
3071 err
= ext4_add_entry(handle
, dentry
, inode
);
3073 ext4_mark_inode_dirty(handle
, inode
);
3074 /* this can happen only for tmpfile being
3075 * linked the first time
3077 if (inode
->i_nlink
== 1)
3078 ext4_orphan_del(handle
, inode
);
3079 d_instantiate(dentry
, inode
);
3084 ext4_journal_stop(handle
);
3085 if (err
== -ENOSPC
&& ext4_should_retry_alloc(dir
->i_sb
, &retries
))
3092 * Try to find buffer head where contains the parent block.
3093 * It should be the inode block if it is inlined or the 1st block
3094 * if it is a normal dir.
3096 static struct buffer_head
*ext4_get_first_dir_block(handle_t
*handle
,
3097 struct inode
*inode
,
3099 struct ext4_dir_entry_2
**parent_de
,
3102 struct buffer_head
*bh
;
3104 if (!ext4_has_inline_data(inode
)) {
3105 bh
= ext4_read_dirblock(inode
, 0, EITHER
);
3107 *retval
= PTR_ERR(bh
);
3110 *parent_de
= ext4_next_entry(
3111 (struct ext4_dir_entry_2
*)bh
->b_data
,
3112 inode
->i_sb
->s_blocksize
);
3117 return ext4_get_first_inline_block(inode
, parent_de
, retval
);
3120 struct ext4_renament
{
3122 struct dentry
*dentry
;
3123 struct inode
*inode
;
3125 int dir_nlink_delta
;
3127 /* entry for "dentry" */
3128 struct buffer_head
*bh
;
3129 struct ext4_dir_entry_2
*de
;
3132 /* entry for ".." in inode if it's a directory */
3133 struct buffer_head
*dir_bh
;
3134 struct ext4_dir_entry_2
*parent_de
;
3138 static int ext4_rename_dir_prepare(handle_t
*handle
, struct ext4_renament
*ent
)
3142 ent
->dir_bh
= ext4_get_first_dir_block(handle
, ent
->inode
,
3143 &retval
, &ent
->parent_de
,
3147 if (le32_to_cpu(ent
->parent_de
->inode
) != ent
->dir
->i_ino
)
3149 BUFFER_TRACE(ent
->dir_bh
, "get_write_access");
3150 return ext4_journal_get_write_access(handle
, ent
->dir_bh
);
3153 static int ext4_rename_dir_finish(handle_t
*handle
, struct ext4_renament
*ent
,
3158 ent
->parent_de
->inode
= cpu_to_le32(dir_ino
);
3159 BUFFER_TRACE(ent
->dir_bh
, "call ext4_handle_dirty_metadata");
3160 if (!ent
->dir_inlined
) {
3161 if (is_dx(ent
->inode
)) {
3162 retval
= ext4_handle_dirty_dx_node(handle
,
3166 retval
= ext4_handle_dirty_dirent_node(handle
,
3171 retval
= ext4_mark_inode_dirty(handle
, ent
->inode
);
3174 ext4_std_error(ent
->dir
->i_sb
, retval
);
3180 static int ext4_setent(handle_t
*handle
, struct ext4_renament
*ent
,
3181 unsigned ino
, unsigned file_type
)
3185 BUFFER_TRACE(ent
->bh
, "get write access");
3186 retval
= ext4_journal_get_write_access(handle
, ent
->bh
);
3189 ent
->de
->inode
= cpu_to_le32(ino
);
3190 if (EXT4_HAS_INCOMPAT_FEATURE(ent
->dir
->i_sb
,
3191 EXT4_FEATURE_INCOMPAT_FILETYPE
))
3192 ent
->de
->file_type
= file_type
;
3193 ent
->dir
->i_version
++;
3194 ent
->dir
->i_ctime
= ent
->dir
->i_mtime
=
3195 ext4_current_time(ent
->dir
);
3196 ext4_mark_inode_dirty(handle
, ent
->dir
);
3197 BUFFER_TRACE(ent
->bh
, "call ext4_handle_dirty_metadata");
3198 if (!ent
->inlined
) {
3199 retval
= ext4_handle_dirty_dirent_node(handle
,
3201 if (unlikely(retval
)) {
3202 ext4_std_error(ent
->dir
->i_sb
, retval
);
3212 static int ext4_find_delete_entry(handle_t
*handle
, struct inode
*dir
,
3213 const struct qstr
*d_name
)
3215 int retval
= -ENOENT
;
3216 struct buffer_head
*bh
;
3217 struct ext4_dir_entry_2
*de
;
3219 bh
= ext4_find_entry(dir
, d_name
, &de
, NULL
);
3223 retval
= ext4_delete_entry(handle
, dir
, de
, bh
);
3229 static void ext4_rename_delete(handle_t
*handle
, struct ext4_renament
*ent
,
3234 * ent->de could have moved from under us during htree split, so make
3235 * sure that we are deleting the right entry. We might also be pointing
3236 * to a stale entry in the unused part of ent->bh so just checking inum
3237 * and the name isn't enough.
3239 if (le32_to_cpu(ent
->de
->inode
) != ent
->inode
->i_ino
||
3240 ent
->de
->name_len
!= ent
->dentry
->d_name
.len
||
3241 strncmp(ent
->de
->name
, ent
->dentry
->d_name
.name
,
3242 ent
->de
->name_len
) ||
3244 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3245 &ent
->dentry
->d_name
);
3247 retval
= ext4_delete_entry(handle
, ent
->dir
, ent
->de
, ent
->bh
);
3248 if (retval
== -ENOENT
) {
3249 retval
= ext4_find_delete_entry(handle
, ent
->dir
,
3250 &ent
->dentry
->d_name
);
3255 ext4_warning(ent
->dir
->i_sb
,
3256 "Deleting old file (%lu), %d, error=%d",
3257 ent
->dir
->i_ino
, ent
->dir
->i_nlink
, retval
);
3261 static void ext4_update_dir_count(handle_t
*handle
, struct ext4_renament
*ent
)
3263 if (ent
->dir_nlink_delta
) {
3264 if (ent
->dir_nlink_delta
== -1)
3265 ext4_dec_count(handle
, ent
->dir
);
3267 ext4_inc_count(handle
, ent
->dir
);
3268 ext4_mark_inode_dirty(handle
, ent
->dir
);
3272 static struct inode
*ext4_whiteout_for_rename(struct ext4_renament
*ent
,
3273 int credits
, handle_t
**h
)
3280 * for inode block, sb block, group summaries,
3283 credits
+= (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent
->dir
->i_sb
) +
3284 EXT4_XATTR_TRANS_BLOCKS
+ 4);
3286 wh
= ext4_new_inode_start_handle(ent
->dir
, S_IFCHR
| WHITEOUT_MODE
,
3287 &ent
->dentry
->d_name
, 0, NULL
,
3288 EXT4_HT_DIR
, credits
);
3290 handle
= ext4_journal_current_handle();
3293 ext4_journal_stop(handle
);
3294 if (PTR_ERR(wh
) == -ENOSPC
&&
3295 ext4_should_retry_alloc(ent
->dir
->i_sb
, &retries
))
3299 init_special_inode(wh
, wh
->i_mode
, WHITEOUT_DEV
);
3300 wh
->i_op
= &ext4_special_inode_operations
;
3306 * Anybody can rename anything with this: the permission checks are left to the
3307 * higher-level routines.
3309 * n.b. old_{dentry,inode) refers to the source dentry/inode
3310 * while new_{dentry,inode) refers to the destination dentry/inode
3311 * This comes from rename(const char *oldpath, const char *newpath)
3313 static int ext4_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3314 struct inode
*new_dir
, struct dentry
*new_dentry
,
3317 handle_t
*handle
= NULL
;
3318 struct ext4_renament old
= {
3320 .dentry
= old_dentry
,
3321 .inode
= old_dentry
->d_inode
,
3323 struct ext4_renament
new = {
3325 .dentry
= new_dentry
,
3326 .inode
= new_dentry
->d_inode
,
3330 struct inode
*whiteout
= NULL
;
3334 dquot_initialize(old
.dir
);
3335 dquot_initialize(new.dir
);
3337 /* Initialize quotas before so that eventual writes go
3338 * in separate transaction */
3340 dquot_initialize(new.inode
);
3342 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
, &old
.de
, NULL
);
3344 return PTR_ERR(old
.bh
);
3346 * Check for inode number is _not_ due to possible IO errors.
3347 * We might rmdir the source, keep it as pwd of some process
3348 * and merrily kill the link to whatever was created under the
3349 * same name. Goodbye sticky bit ;-<
3352 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3355 if ((old
.dir
!= new.dir
) &&
3356 ext4_encrypted_inode(new.dir
) &&
3357 !ext4_is_child_context_consistent_with_parent(new.dir
,
3363 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3364 &new.de
, &new.inlined
);
3365 if (IS_ERR(new.bh
)) {
3366 retval
= PTR_ERR(new.bh
);
3376 if (new.inode
&& !test_opt(new.dir
->i_sb
, NO_AUTO_DA_ALLOC
))
3377 ext4_alloc_da_blocks(old
.inode
);
3379 credits
= (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3380 EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2);
3381 if (!(flags
& RENAME_WHITEOUT
)) {
3382 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
, credits
);
3383 if (IS_ERR(handle
)) {
3384 retval
= PTR_ERR(handle
);
3389 whiteout
= ext4_whiteout_for_rename(&old
, credits
, &handle
);
3390 if (IS_ERR(whiteout
)) {
3391 retval
= PTR_ERR(whiteout
);
3397 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3398 ext4_handle_sync(handle
);
3400 if (S_ISDIR(old
.inode
->i_mode
)) {
3402 retval
= -ENOTEMPTY
;
3403 if (!ext4_empty_dir(new.inode
))
3407 if (new.dir
!= old
.dir
&& EXT4_DIR_LINK_MAX(new.dir
))
3410 retval
= ext4_rename_dir_prepare(handle
, &old
);
3415 * If we're renaming a file within an inline_data dir and adding or
3416 * setting the new dirent causes a conversion from inline_data to
3417 * extents/blockmap, we need to force the dirent delete code to
3418 * re-read the directory, or else we end up trying to delete a dirent
3419 * from what is now the extent tree root (or a block map).
3421 force_reread
= (new.dir
->i_ino
== old
.dir
->i_ino
&&
3422 ext4_test_inode_flag(new.dir
, EXT4_INODE_INLINE_DATA
));
3424 old_file_type
= old
.de
->file_type
;
3427 * Do this before adding a new entry, so the old entry is sure
3428 * to be still pointing to the valid old entry.
3430 retval
= ext4_setent(handle
, &old
, whiteout
->i_ino
,
3434 ext4_mark_inode_dirty(handle
, whiteout
);
3437 retval
= ext4_add_entry(handle
, new.dentry
, old
.inode
);
3441 retval
= ext4_setent(handle
, &new,
3442 old
.inode
->i_ino
, old_file_type
);
3447 force_reread
= !ext4_test_inode_flag(new.dir
,
3448 EXT4_INODE_INLINE_DATA
);
3451 * Like most other Unix systems, set the ctime for inodes on a
3454 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3455 ext4_mark_inode_dirty(handle
, old
.inode
);
3461 ext4_rename_delete(handle
, &old
, force_reread
);
3465 ext4_dec_count(handle
, new.inode
);
3466 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3468 old
.dir
->i_ctime
= old
.dir
->i_mtime
= ext4_current_time(old
.dir
);
3469 ext4_update_dx_flag(old
.dir
);
3471 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3475 ext4_dec_count(handle
, old
.dir
);
3477 /* checked ext4_empty_dir above, can't have another
3478 * parent, ext4_dec_count() won't work for many-linked
3480 clear_nlink(new.inode
);
3482 ext4_inc_count(handle
, new.dir
);
3483 ext4_update_dx_flag(new.dir
);
3484 ext4_mark_inode_dirty(handle
, new.dir
);
3487 ext4_mark_inode_dirty(handle
, old
.dir
);
3489 ext4_mark_inode_dirty(handle
, new.inode
);
3490 if (!new.inode
->i_nlink
)
3491 ext4_orphan_add(handle
, new.inode
);
3501 drop_nlink(whiteout
);
3502 unlock_new_inode(whiteout
);
3506 ext4_journal_stop(handle
);
3510 static int ext4_cross_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3511 struct inode
*new_dir
, struct dentry
*new_dentry
)
3513 handle_t
*handle
= NULL
;
3514 struct ext4_renament old
= {
3516 .dentry
= old_dentry
,
3517 .inode
= old_dentry
->d_inode
,
3519 struct ext4_renament
new = {
3521 .dentry
= new_dentry
,
3522 .inode
= new_dentry
->d_inode
,
3527 dquot_initialize(old
.dir
);
3528 dquot_initialize(new.dir
);
3530 old
.bh
= ext4_find_entry(old
.dir
, &old
.dentry
->d_name
,
3531 &old
.de
, &old
.inlined
);
3533 return PTR_ERR(old
.bh
);
3535 * Check for inode number is _not_ due to possible IO errors.
3536 * We might rmdir the source, keep it as pwd of some process
3537 * and merrily kill the link to whatever was created under the
3538 * same name. Goodbye sticky bit ;-<
3541 if (!old
.bh
|| le32_to_cpu(old
.de
->inode
) != old
.inode
->i_ino
)
3544 new.bh
= ext4_find_entry(new.dir
, &new.dentry
->d_name
,
3545 &new.de
, &new.inlined
);
3546 if (IS_ERR(new.bh
)) {
3547 retval
= PTR_ERR(new.bh
);
3552 /* RENAME_EXCHANGE case: old *and* new must both exist */
3553 if (!new.bh
|| le32_to_cpu(new.de
->inode
) != new.inode
->i_ino
)
3556 handle
= ext4_journal_start(old
.dir
, EXT4_HT_DIR
,
3557 (2 * EXT4_DATA_TRANS_BLOCKS(old
.dir
->i_sb
) +
3558 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS
+ 2));
3559 if (IS_ERR(handle
)) {
3560 retval
= PTR_ERR(handle
);
3565 if (IS_DIRSYNC(old
.dir
) || IS_DIRSYNC(new.dir
))
3566 ext4_handle_sync(handle
);
3568 if (S_ISDIR(old
.inode
->i_mode
)) {
3570 retval
= ext4_rename_dir_prepare(handle
, &old
);
3574 if (S_ISDIR(new.inode
->i_mode
)) {
3576 retval
= ext4_rename_dir_prepare(handle
, &new);
3582 * Other than the special case of overwriting a directory, parents'
3583 * nlink only needs to be modified if this is a cross directory rename.
3585 if (old
.dir
!= new.dir
&& old
.is_dir
!= new.is_dir
) {
3586 old
.dir_nlink_delta
= old
.is_dir
? -1 : 1;
3587 new.dir_nlink_delta
= -old
.dir_nlink_delta
;
3589 if ((old
.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(old
.dir
)) ||
3590 (new.dir_nlink_delta
> 0 && EXT4_DIR_LINK_MAX(new.dir
)))
3594 new_file_type
= new.de
->file_type
;
3595 retval
= ext4_setent(handle
, &new, old
.inode
->i_ino
, old
.de
->file_type
);
3599 retval
= ext4_setent(handle
, &old
, new.inode
->i_ino
, new_file_type
);
3604 * Like most other Unix systems, set the ctime for inodes on a
3607 old
.inode
->i_ctime
= ext4_current_time(old
.inode
);
3608 new.inode
->i_ctime
= ext4_current_time(new.inode
);
3609 ext4_mark_inode_dirty(handle
, old
.inode
);
3610 ext4_mark_inode_dirty(handle
, new.inode
);
3613 retval
= ext4_rename_dir_finish(handle
, &old
, new.dir
->i_ino
);
3618 retval
= ext4_rename_dir_finish(handle
, &new, old
.dir
->i_ino
);
3622 ext4_update_dir_count(handle
, &old
);
3623 ext4_update_dir_count(handle
, &new);
3632 ext4_journal_stop(handle
);
3636 static int ext4_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
3637 struct inode
*new_dir
, struct dentry
*new_dentry
,
3640 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3643 if (flags
& RENAME_EXCHANGE
) {
3644 return ext4_cross_rename(old_dir
, old_dentry
,
3645 new_dir
, new_dentry
);
3648 return ext4_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
3652 * directories can handle most operations...
3654 const struct inode_operations ext4_dir_inode_operations
= {
3655 .create
= ext4_create
,
3656 .lookup
= ext4_lookup
,
3658 .unlink
= ext4_unlink
,
3659 .symlink
= ext4_symlink
,
3660 .mkdir
= ext4_mkdir
,
3661 .rmdir
= ext4_rmdir
,
3662 .mknod
= ext4_mknod
,
3663 .tmpfile
= ext4_tmpfile
,
3664 .rename2
= ext4_rename2
,
3665 .setattr
= ext4_setattr
,
3666 .setxattr
= generic_setxattr
,
3667 .getxattr
= generic_getxattr
,
3668 .listxattr
= ext4_listxattr
,
3669 .removexattr
= generic_removexattr
,
3670 .get_acl
= ext4_get_acl
,
3671 .set_acl
= ext4_set_acl
,
3672 .fiemap
= ext4_fiemap
,
3675 const struct inode_operations ext4_special_inode_operations
= {
3676 .setattr
= ext4_setattr
,
3677 .setxattr
= generic_setxattr
,
3678 .getxattr
= generic_getxattr
,
3679 .listxattr
= ext4_listxattr
,
3680 .removexattr
= generic_removexattr
,
3681 .get_acl
= ext4_get_acl
,
3682 .set_acl
= ext4_set_acl
,