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/dir.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * ext4 directory handling functions
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
20 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
25 #include <linux/jbd2.h>
26 #include <linux/buffer_head.h>
27 #include <linux/slab.h>
28 #include <linux/rbtree.h>
31 static unsigned char ext4_filetype_table
[] = {
32 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
35 static int ext4_readdir(struct file
*, void *, filldir_t
);
36 static int ext4_dx_readdir(struct file
*filp
,
37 void *dirent
, filldir_t filldir
);
38 static int ext4_release_dir(struct inode
*inode
,
41 const struct file_operations ext4_dir_operations
= {
42 .llseek
= ext4_llseek
,
43 .read
= generic_read_dir
,
44 .readdir
= ext4_readdir
, /* we take BKL. needed?*/
45 .unlocked_ioctl
= ext4_ioctl
,
47 .compat_ioctl
= ext4_compat_ioctl
,
49 .fsync
= ext4_sync_file
,
50 .release
= ext4_release_dir
,
54 static unsigned char get_dtype(struct super_block
*sb
, int filetype
)
56 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FILETYPE
) ||
57 (filetype
>= EXT4_FT_MAX
))
60 return (ext4_filetype_table
[filetype
]);
64 * Return 0 if the directory entry is OK, and 1 if there is a problem
66 * Note: this is the opposite of what ext2 and ext3 historically returned...
68 int __ext4_check_dir_entry(const char *function
, unsigned int line
,
70 struct ext4_dir_entry_2
*de
,
71 struct buffer_head
*bh
,
74 const char *error_msg
= NULL
;
75 const int rlen
= ext4_rec_len_from_disk(de
->rec_len
,
76 dir
->i_sb
->s_blocksize
);
78 if (unlikely(rlen
< EXT4_DIR_REC_LEN(1)))
79 error_msg
= "rec_len is smaller than minimal";
80 else if (unlikely(rlen
% 4 != 0))
81 error_msg
= "rec_len % 4 != 0";
82 else if (unlikely(rlen
< EXT4_DIR_REC_LEN(de
->name_len
)))
83 error_msg
= "rec_len is too small for name_len";
84 else if (unlikely(((char *) de
- bh
->b_data
) + rlen
>
85 dir
->i_sb
->s_blocksize
))
86 error_msg
= "directory entry across blocks";
87 else if (unlikely(le32_to_cpu(de
->inode
) >
88 le32_to_cpu(EXT4_SB(dir
->i_sb
)->s_es
->s_inodes_count
)))
89 error_msg
= "inode out of bounds";
93 ext4_error_inode(dir
, function
, line
, bh
->b_blocknr
,
94 "bad entry in directory: %s - "
95 "offset=%u(%u), inode=%u, rec_len=%d, name_len=%d",
96 error_msg
, (unsigned) (offset
%bh
->b_size
), offset
,
97 le32_to_cpu(de
->inode
),
102 static int ext4_readdir(struct file
*filp
,
103 void *dirent
, filldir_t filldir
)
108 struct ext4_dir_entry_2
*de
;
109 struct super_block
*sb
;
111 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
113 int dir_has_error
= 0;
117 if (EXT4_HAS_COMPAT_FEATURE(inode
->i_sb
,
118 EXT4_FEATURE_COMPAT_DIR_INDEX
) &&
119 ((ext4_test_inode_flag(inode
, EXT4_INODE_INDEX
)) ||
120 ((inode
->i_size
>> sb
->s_blocksize_bits
) == 1))) {
121 err
= ext4_dx_readdir(filp
, dirent
, filldir
);
122 if (err
!= ERR_BAD_DX_DIR
) {
127 * We don't set the inode dirty flag since it's not
128 * critical that it get flushed back to the disk.
130 ext4_clear_inode_flag(filp
->f_path
.dentry
->d_inode
,
134 offset
= filp
->f_pos
& (sb
->s_blocksize
- 1);
136 while (!error
&& !stored
&& filp
->f_pos
< inode
->i_size
) {
137 struct ext4_map_blocks map
;
138 struct buffer_head
*bh
= NULL
;
140 map
.m_lblk
= filp
->f_pos
>> EXT4_BLOCK_SIZE_BITS(sb
);
142 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
144 pgoff_t index
= map
.m_pblk
>>
145 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
146 if (!ra_has_index(&filp
->f_ra
, index
))
147 page_cache_sync_readahead(
148 sb
->s_bdev
->bd_inode
->i_mapping
,
151 filp
->f_ra
.prev_pos
= (loff_t
)index
<< PAGE_CACHE_SHIFT
;
152 bh
= ext4_bread(NULL
, inode
, map
.m_lblk
, 0, &err
);
156 * We ignore I/O errors on directories so users have a chance
157 * of recovering data when there's a bad sector
160 if (!dir_has_error
) {
161 EXT4_ERROR_INODE(inode
, "directory "
162 "contains a hole at offset %Lu",
163 (unsigned long long) filp
->f_pos
);
166 /* corrupt size? Maybe no more blocks to read */
167 if (filp
->f_pos
> inode
->i_blocks
<< 9)
169 filp
->f_pos
+= sb
->s_blocksize
- offset
;
174 /* If the dir block has changed since the last call to
175 * readdir(2), then we might be pointing to an invalid
176 * dirent right now. Scan from the start of the block
178 if (filp
->f_version
!= inode
->i_version
) {
179 for (i
= 0; i
< sb
->s_blocksize
&& i
< offset
; ) {
180 de
= (struct ext4_dir_entry_2
*)
182 /* It's too expensive to do a full
183 * dirent test each time round this
184 * loop, but we do have to test at
185 * least that it is non-zero. A
186 * failure will be detected in the
187 * dirent test below. */
188 if (ext4_rec_len_from_disk(de
->rec_len
,
189 sb
->s_blocksize
) < EXT4_DIR_REC_LEN(1))
191 i
+= ext4_rec_len_from_disk(de
->rec_len
,
195 filp
->f_pos
= (filp
->f_pos
& ~(sb
->s_blocksize
- 1))
197 filp
->f_version
= inode
->i_version
;
200 while (!error
&& filp
->f_pos
< inode
->i_size
201 && offset
< sb
->s_blocksize
) {
202 de
= (struct ext4_dir_entry_2
*) (bh
->b_data
+ offset
);
203 if (ext4_check_dir_entry(inode
, de
,
206 * On error, skip the f_pos to the next block
208 filp
->f_pos
= (filp
->f_pos
|
209 (sb
->s_blocksize
- 1)) + 1;
214 offset
+= ext4_rec_len_from_disk(de
->rec_len
,
216 if (le32_to_cpu(de
->inode
)) {
217 /* We might block in the next section
218 * if the data destination is
219 * currently swapped out. So, use a
220 * version stamp to detect whether or
221 * not the directory has been modified
222 * during the copy operation.
224 u64 version
= filp
->f_version
;
226 error
= filldir(dirent
, de
->name
,
229 le32_to_cpu(de
->inode
),
230 get_dtype(sb
, de
->file_type
));
233 if (version
!= filp
->f_version
)
237 filp
->f_pos
+= ext4_rec_len_from_disk(de
->rec_len
,
248 * These functions convert from the major/minor hash to an f_pos
251 * Currently we only use major hash numer. This is unfortunate, but
252 * on 32-bit machines, the same VFS interface is used for lseek and
253 * llseek, so if we use the 64 bit offset, then the 32-bit versions of
254 * lseek/telldir/seekdir will blow out spectacularly, and from within
255 * the ext2 low-level routine, we don't know if we're being called by
256 * a 64-bit version of the system call or the 32-bit version of the
257 * system call. Worse yet, NFSv2 only allows for a 32-bit readdir
260 #define hash2pos(major, minor) (major >> 1)
261 #define pos2maj_hash(pos) ((pos << 1) & 0xffffffff)
262 #define pos2min_hash(pos) (0)
265 * This structure holds the nodes of the red-black tree used to store
266 * the directory entry in hash order.
271 struct rb_node rb_hash
;
280 * This functoin implements a non-recursive way of freeing all of the
281 * nodes in the red-black tree.
283 static void free_rb_tree_fname(struct rb_root
*root
)
285 struct rb_node
*n
= root
->rb_node
;
286 struct rb_node
*parent
;
290 /* Do the node's children first */
300 * The node has no children; free it, and then zero
301 * out parent's link to it. Finally go to the
302 * beginning of the loop and try to free the parent
305 parent
= rb_parent(n
);
306 fname
= rb_entry(n
, struct fname
, rb_hash
);
308 struct fname
*old
= fname
;
314 else if (parent
->rb_left
== n
)
315 parent
->rb_left
= NULL
;
316 else if (parent
->rb_right
== n
)
317 parent
->rb_right
= NULL
;
323 static struct dir_private_info
*ext4_htree_create_dir_info(loff_t pos
)
325 struct dir_private_info
*p
;
327 p
= kzalloc(sizeof(struct dir_private_info
), GFP_KERNEL
);
330 p
->curr_hash
= pos2maj_hash(pos
);
331 p
->curr_minor_hash
= pos2min_hash(pos
);
335 void ext4_htree_free_dir_info(struct dir_private_info
*p
)
337 free_rb_tree_fname(&p
->root
);
342 * Given a directory entry, enter it into the fname rb tree.
344 int ext4_htree_store_dirent(struct file
*dir_file
, __u32 hash
,
346 struct ext4_dir_entry_2
*dirent
)
348 struct rb_node
**p
, *parent
= NULL
;
349 struct fname
*fname
, *new_fn
;
350 struct dir_private_info
*info
;
353 info
= dir_file
->private_data
;
354 p
= &info
->root
.rb_node
;
356 /* Create and allocate the fname structure */
357 len
= sizeof(struct fname
) + dirent
->name_len
+ 1;
358 new_fn
= kzalloc(len
, GFP_KERNEL
);
362 new_fn
->minor_hash
= minor_hash
;
363 new_fn
->inode
= le32_to_cpu(dirent
->inode
);
364 new_fn
->name_len
= dirent
->name_len
;
365 new_fn
->file_type
= dirent
->file_type
;
366 memcpy(new_fn
->name
, dirent
->name
, dirent
->name_len
);
367 new_fn
->name
[dirent
->name_len
] = 0;
371 fname
= rb_entry(parent
, struct fname
, rb_hash
);
374 * If the hash and minor hash match up, then we put
375 * them on a linked list. This rarely happens...
377 if ((new_fn
->hash
== fname
->hash
) &&
378 (new_fn
->minor_hash
== fname
->minor_hash
)) {
379 new_fn
->next
= fname
->next
;
380 fname
->next
= new_fn
;
384 if (new_fn
->hash
< fname
->hash
)
386 else if (new_fn
->hash
> fname
->hash
)
388 else if (new_fn
->minor_hash
< fname
->minor_hash
)
390 else /* if (new_fn->minor_hash > fname->minor_hash) */
394 rb_link_node(&new_fn
->rb_hash
, parent
, p
);
395 rb_insert_color(&new_fn
->rb_hash
, &info
->root
);
402 * This is a helper function for ext4_dx_readdir. It calls filldir
403 * for all entres on the fname linked list. (Normally there is only
404 * one entry on the linked list, unless there are 62 bit hash collisions.)
406 static int call_filldir(struct file
*filp
, void *dirent
,
407 filldir_t filldir
, struct fname
*fname
)
409 struct dir_private_info
*info
= filp
->private_data
;
411 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
412 struct super_block
*sb
;
418 printk(KERN_ERR
"EXT4-fs: call_filldir: called with "
422 curr_pos
= hash2pos(fname
->hash
, fname
->minor_hash
);
424 error
= filldir(dirent
, fname
->name
,
425 fname
->name_len
, curr_pos
,
427 get_dtype(sb
, fname
->file_type
));
429 filp
->f_pos
= curr_pos
;
430 info
->extra_fname
= fname
;
438 static int ext4_dx_readdir(struct file
*filp
,
439 void *dirent
, filldir_t filldir
)
441 struct dir_private_info
*info
= filp
->private_data
;
442 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
447 info
= ext4_htree_create_dir_info(filp
->f_pos
);
450 filp
->private_data
= info
;
453 if (filp
->f_pos
== EXT4_HTREE_EOF
)
456 /* Some one has messed with f_pos; reset the world */
457 if (info
->last_pos
!= filp
->f_pos
) {
458 free_rb_tree_fname(&info
->root
);
459 info
->curr_node
= NULL
;
460 info
->extra_fname
= NULL
;
461 info
->curr_hash
= pos2maj_hash(filp
->f_pos
);
462 info
->curr_minor_hash
= pos2min_hash(filp
->f_pos
);
466 * If there are any leftover names on the hash collision
467 * chain, return them first.
469 if (info
->extra_fname
) {
470 if (call_filldir(filp
, dirent
, filldir
, info
->extra_fname
))
472 info
->extra_fname
= NULL
;
474 } else if (!info
->curr_node
)
475 info
->curr_node
= rb_first(&info
->root
);
479 * Fill the rbtree if we have no more entries,
480 * or the inode has changed since we last read in the
483 if ((!info
->curr_node
) ||
484 (filp
->f_version
!= inode
->i_version
)) {
485 info
->curr_node
= NULL
;
486 free_rb_tree_fname(&info
->root
);
487 filp
->f_version
= inode
->i_version
;
488 ret
= ext4_htree_fill_tree(filp
, info
->curr_hash
,
489 info
->curr_minor_hash
,
494 filp
->f_pos
= EXT4_HTREE_EOF
;
497 info
->curr_node
= rb_first(&info
->root
);
500 fname
= rb_entry(info
->curr_node
, struct fname
, rb_hash
);
501 info
->curr_hash
= fname
->hash
;
502 info
->curr_minor_hash
= fname
->minor_hash
;
503 if (call_filldir(filp
, dirent
, filldir
, fname
))
506 info
->curr_node
= rb_next(info
->curr_node
);
507 if (info
->curr_node
) {
508 fname
= rb_entry(info
->curr_node
, struct fname
,
510 info
->curr_hash
= fname
->hash
;
511 info
->curr_minor_hash
= fname
->minor_hash
;
513 if (info
->next_hash
== ~0) {
514 filp
->f_pos
= EXT4_HTREE_EOF
;
517 info
->curr_hash
= info
->next_hash
;
518 info
->curr_minor_hash
= 0;
522 info
->last_pos
= filp
->f_pos
;
526 static int ext4_release_dir(struct inode
*inode
, struct file
*filp
)
528 if (filp
->private_data
)
529 ext4_htree_free_dir_info(filp
->private_data
);