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>
32 static int ext4_dx_readdir(struct file
*, struct dir_context
*);
35 * Check if the given dir-inode refers to an htree-indexed directory
36 * (or a directory which could potentially get converted to use htree
39 * Return 1 if it is a dx dir, 0 if not
41 static int is_dx_dir(struct inode
*inode
)
43 struct super_block
*sb
= inode
->i_sb
;
45 if (EXT4_HAS_COMPAT_FEATURE(inode
->i_sb
,
46 EXT4_FEATURE_COMPAT_DIR_INDEX
) &&
47 ((ext4_test_inode_flag(inode
, EXT4_INODE_INDEX
)) ||
48 ((inode
->i_size
>> sb
->s_blocksize_bits
) == 1) ||
49 ext4_has_inline_data(inode
)))
56 * Return 0 if the directory entry is OK, and 1 if there is a problem
58 * Note: this is the opposite of what ext2 and ext3 historically returned...
60 * bh passed here can be an inode block or a dir data block, depending
61 * on the inode inline data flag.
63 int __ext4_check_dir_entry(const char *function
, unsigned int line
,
64 struct inode
*dir
, struct file
*filp
,
65 struct ext4_dir_entry_2
*de
,
66 struct buffer_head
*bh
, char *buf
, int size
,
69 const char *error_msg
= NULL
;
70 const int rlen
= ext4_rec_len_from_disk(de
->rec_len
,
71 dir
->i_sb
->s_blocksize
);
73 if (unlikely(rlen
< EXT4_DIR_REC_LEN(1)))
74 error_msg
= "rec_len is smaller than minimal";
75 else if (unlikely(rlen
% 4 != 0))
76 error_msg
= "rec_len % 4 != 0";
77 else if (unlikely(rlen
< EXT4_DIR_REC_LEN(de
->name_len
)))
78 error_msg
= "rec_len is too small for name_len";
79 else if (unlikely(((char *) de
- buf
) + rlen
> size
))
80 error_msg
= "directory entry across range";
81 else if (unlikely(le32_to_cpu(de
->inode
) >
82 le32_to_cpu(EXT4_SB(dir
->i_sb
)->s_es
->s_inodes_count
)))
83 error_msg
= "inode out of bounds";
88 ext4_error_file(filp
, function
, line
, bh
->b_blocknr
,
89 "bad entry in directory: %s - offset=%u(%u), "
90 "inode=%u, rec_len=%d, name_len=%d",
91 error_msg
, (unsigned) (offset
% size
),
92 offset
, le32_to_cpu(de
->inode
),
95 ext4_error_inode(dir
, function
, line
, bh
->b_blocknr
,
96 "bad entry in directory: %s - offset=%u(%u), "
97 "inode=%u, rec_len=%d, name_len=%d",
98 error_msg
, (unsigned) (offset
% size
),
99 offset
, le32_to_cpu(de
->inode
),
105 static int ext4_readdir(struct file
*file
, struct dir_context
*ctx
)
109 struct ext4_dir_entry_2
*de
;
111 struct inode
*inode
= file_inode(file
);
112 struct super_block
*sb
= inode
->i_sb
;
113 int dir_has_error
= 0;
115 if (is_dx_dir(inode
)) {
116 err
= ext4_dx_readdir(file
, ctx
);
117 if (err
!= ERR_BAD_DX_DIR
) {
121 * We don't set the inode dirty flag since it's not
122 * critical that it get flushed back to the disk.
124 ext4_clear_inode_flag(file_inode(file
),
128 if (ext4_has_inline_data(inode
)) {
129 int has_inline_data
= 1;
130 int ret
= ext4_read_inline_dir(file
, ctx
,
136 offset
= ctx
->pos
& (sb
->s_blocksize
- 1);
138 while (ctx
->pos
< inode
->i_size
) {
139 struct ext4_map_blocks map
;
140 struct buffer_head
*bh
= NULL
;
142 map
.m_lblk
= ctx
->pos
>> EXT4_BLOCK_SIZE_BITS(sb
);
144 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
146 pgoff_t index
= map
.m_pblk
>>
147 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
148 if (!ra_has_index(&file
->f_ra
, index
))
149 page_cache_sync_readahead(
150 sb
->s_bdev
->bd_inode
->i_mapping
,
153 file
->f_ra
.prev_pos
= (loff_t
)index
<< PAGE_CACHE_SHIFT
;
154 bh
= ext4_bread(NULL
, inode
, map
.m_lblk
, 0, &err
);
158 * We ignore I/O errors on directories so users have a chance
159 * of recovering data when there's a bad sector
162 if (!dir_has_error
) {
163 EXT4_ERROR_FILE(file
, 0,
164 "directory contains a "
165 "hole at offset %llu",
166 (unsigned long long) ctx
->pos
);
169 /* corrupt size? Maybe no more blocks to read */
170 if (ctx
->pos
> inode
->i_blocks
<< 9)
172 ctx
->pos
+= sb
->s_blocksize
- offset
;
176 /* Check the checksum */
177 if (!buffer_verified(bh
) &&
178 !ext4_dirent_csum_verify(inode
,
179 (struct ext4_dir_entry
*)bh
->b_data
)) {
180 EXT4_ERROR_FILE(file
, 0, "directory fails checksum "
182 (unsigned long long)ctx
->pos
);
183 ctx
->pos
+= sb
->s_blocksize
- offset
;
187 set_buffer_verified(bh
);
189 /* If the dir block has changed since the last call to
190 * readdir(2), then we might be pointing to an invalid
191 * dirent right now. Scan from the start of the block
193 if (file
->f_version
!= inode
->i_version
) {
194 for (i
= 0; i
< sb
->s_blocksize
&& i
< offset
; ) {
195 de
= (struct ext4_dir_entry_2
*)
197 /* It's too expensive to do a full
198 * dirent test each time round this
199 * loop, but we do have to test at
200 * least that it is non-zero. A
201 * failure will be detected in the
202 * dirent test below. */
203 if (ext4_rec_len_from_disk(de
->rec_len
,
204 sb
->s_blocksize
) < EXT4_DIR_REC_LEN(1))
206 i
+= ext4_rec_len_from_disk(de
->rec_len
,
210 ctx
->pos
= (ctx
->pos
& ~(sb
->s_blocksize
- 1))
212 file
->f_version
= inode
->i_version
;
215 while (ctx
->pos
< inode
->i_size
216 && offset
< sb
->s_blocksize
) {
217 de
= (struct ext4_dir_entry_2
*) (bh
->b_data
+ offset
);
218 if (ext4_check_dir_entry(inode
, file
, de
, bh
,
219 bh
->b_data
, bh
->b_size
,
222 * On error, skip to the next block
224 ctx
->pos
= (ctx
->pos
|
225 (sb
->s_blocksize
- 1)) + 1;
228 offset
+= ext4_rec_len_from_disk(de
->rec_len
,
230 if (le32_to_cpu(de
->inode
)) {
231 if (!dir_emit(ctx
, de
->name
,
233 le32_to_cpu(de
->inode
),
234 get_dtype(sb
, de
->file_type
))) {
239 ctx
->pos
+= ext4_rec_len_from_disk(de
->rec_len
,
244 if (ctx
->pos
< inode
->i_size
) {
245 if (!dir_relax(inode
))
252 static inline int is_32bit_api(void)
255 return is_compat_task();
257 return (BITS_PER_LONG
== 32);
262 * These functions convert from the major/minor hash to an f_pos
263 * value for dx directories
265 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
266 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
267 * directly on both 32-bit and 64-bit nodes, under such case, neither
268 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
270 static inline loff_t
hash2pos(struct file
*filp
, __u32 major
, __u32 minor
)
272 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
273 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
276 return ((__u64
)(major
>> 1) << 32) | (__u64
)minor
;
279 static inline __u32
pos2maj_hash(struct file
*filp
, loff_t pos
)
281 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
282 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
283 return (pos
<< 1) & 0xffffffff;
285 return ((pos
>> 32) << 1) & 0xffffffff;
288 static inline __u32
pos2min_hash(struct file
*filp
, loff_t pos
)
290 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
291 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
294 return pos
& 0xffffffff;
298 * Return 32- or 64-bit end-of-file for dx directories
300 static inline loff_t
ext4_get_htree_eof(struct file
*filp
)
302 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
303 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
304 return EXT4_HTREE_EOF_32BIT
;
306 return EXT4_HTREE_EOF_64BIT
;
311 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
312 * directories, where the "offset" is in terms of the filename hash
313 * value instead of the byte offset.
315 * Because we may return a 64-bit hash that is well beyond offset limits,
316 * we need to pass the max hash as the maximum allowable offset in
317 * the htree directory case.
319 * For non-htree, ext4_llseek already chooses the proper max offset.
321 static loff_t
ext4_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
323 struct inode
*inode
= file
->f_mapping
->host
;
324 int dx_dir
= is_dx_dir(inode
);
325 loff_t htree_max
= ext4_get_htree_eof(file
);
328 return generic_file_llseek_size(file
, offset
, whence
,
329 htree_max
, htree_max
);
331 return ext4_llseek(file
, offset
, whence
);
335 * This structure holds the nodes of the red-black tree used to store
336 * the directory entry in hash order.
341 struct rb_node rb_hash
;
350 * This functoin implements a non-recursive way of freeing all of the
351 * nodes in the red-black tree.
353 static void free_rb_tree_fname(struct rb_root
*root
)
355 struct fname
*fname
, *next
;
357 rbtree_postorder_for_each_entry_safe(fname
, next
, root
, rb_hash
)
359 struct fname
*old
= fname
;
368 static struct dir_private_info
*ext4_htree_create_dir_info(struct file
*filp
,
371 struct dir_private_info
*p
;
373 p
= kzalloc(sizeof(struct dir_private_info
), GFP_KERNEL
);
376 p
->curr_hash
= pos2maj_hash(filp
, pos
);
377 p
->curr_minor_hash
= pos2min_hash(filp
, pos
);
381 void ext4_htree_free_dir_info(struct dir_private_info
*p
)
383 free_rb_tree_fname(&p
->root
);
388 * Given a directory entry, enter it into the fname rb tree.
390 int ext4_htree_store_dirent(struct file
*dir_file
, __u32 hash
,
392 struct ext4_dir_entry_2
*dirent
)
394 struct rb_node
**p
, *parent
= NULL
;
395 struct fname
*fname
, *new_fn
;
396 struct dir_private_info
*info
;
399 info
= dir_file
->private_data
;
400 p
= &info
->root
.rb_node
;
402 /* Create and allocate the fname structure */
403 len
= sizeof(struct fname
) + dirent
->name_len
+ 1;
404 new_fn
= kzalloc(len
, GFP_KERNEL
);
408 new_fn
->minor_hash
= minor_hash
;
409 new_fn
->inode
= le32_to_cpu(dirent
->inode
);
410 new_fn
->name_len
= dirent
->name_len
;
411 new_fn
->file_type
= dirent
->file_type
;
412 memcpy(new_fn
->name
, dirent
->name
, dirent
->name_len
);
413 new_fn
->name
[dirent
->name_len
] = 0;
417 fname
= rb_entry(parent
, struct fname
, rb_hash
);
420 * If the hash and minor hash match up, then we put
421 * them on a linked list. This rarely happens...
423 if ((new_fn
->hash
== fname
->hash
) &&
424 (new_fn
->minor_hash
== fname
->minor_hash
)) {
425 new_fn
->next
= fname
->next
;
426 fname
->next
= new_fn
;
430 if (new_fn
->hash
< fname
->hash
)
432 else if (new_fn
->hash
> fname
->hash
)
434 else if (new_fn
->minor_hash
< fname
->minor_hash
)
436 else /* if (new_fn->minor_hash > fname->minor_hash) */
440 rb_link_node(&new_fn
->rb_hash
, parent
, p
);
441 rb_insert_color(&new_fn
->rb_hash
, &info
->root
);
448 * This is a helper function for ext4_dx_readdir. It calls filldir
449 * for all entres on the fname linked list. (Normally there is only
450 * one entry on the linked list, unless there are 62 bit hash collisions.)
452 static int call_filldir(struct file
*file
, struct dir_context
*ctx
,
455 struct dir_private_info
*info
= file
->private_data
;
456 struct inode
*inode
= file_inode(file
);
457 struct super_block
*sb
= inode
->i_sb
;
460 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: comm %s: "
461 "called with null fname?!?", __func__
, __LINE__
,
462 inode
->i_ino
, current
->comm
);
465 ctx
->pos
= hash2pos(file
, fname
->hash
, fname
->minor_hash
);
467 if (!dir_emit(ctx
, fname
->name
,
470 get_dtype(sb
, fname
->file_type
))) {
471 info
->extra_fname
= fname
;
479 static int ext4_dx_readdir(struct file
*file
, struct dir_context
*ctx
)
481 struct dir_private_info
*info
= file
->private_data
;
482 struct inode
*inode
= file_inode(file
);
487 info
= ext4_htree_create_dir_info(file
, ctx
->pos
);
490 file
->private_data
= info
;
493 if (ctx
->pos
== ext4_get_htree_eof(file
))
496 /* Some one has messed with f_pos; reset the world */
497 if (info
->last_pos
!= ctx
->pos
) {
498 free_rb_tree_fname(&info
->root
);
499 info
->curr_node
= NULL
;
500 info
->extra_fname
= NULL
;
501 info
->curr_hash
= pos2maj_hash(file
, ctx
->pos
);
502 info
->curr_minor_hash
= pos2min_hash(file
, ctx
->pos
);
506 * If there are any leftover names on the hash collision
507 * chain, return them first.
509 if (info
->extra_fname
) {
510 if (call_filldir(file
, ctx
, info
->extra_fname
))
512 info
->extra_fname
= NULL
;
514 } else if (!info
->curr_node
)
515 info
->curr_node
= rb_first(&info
->root
);
519 * Fill the rbtree if we have no more entries,
520 * or the inode has changed since we last read in the
523 if ((!info
->curr_node
) ||
524 (file
->f_version
!= inode
->i_version
)) {
525 info
->curr_node
= NULL
;
526 free_rb_tree_fname(&info
->root
);
527 file
->f_version
= inode
->i_version
;
528 ret
= ext4_htree_fill_tree(file
, info
->curr_hash
,
529 info
->curr_minor_hash
,
534 ctx
->pos
= ext4_get_htree_eof(file
);
537 info
->curr_node
= rb_first(&info
->root
);
540 fname
= rb_entry(info
->curr_node
, struct fname
, rb_hash
);
541 info
->curr_hash
= fname
->hash
;
542 info
->curr_minor_hash
= fname
->minor_hash
;
543 if (call_filldir(file
, ctx
, fname
))
546 info
->curr_node
= rb_next(info
->curr_node
);
547 if (info
->curr_node
) {
548 fname
= rb_entry(info
->curr_node
, struct fname
,
550 info
->curr_hash
= fname
->hash
;
551 info
->curr_minor_hash
= fname
->minor_hash
;
553 if (info
->next_hash
== ~0) {
554 ctx
->pos
= ext4_get_htree_eof(file
);
557 info
->curr_hash
= info
->next_hash
;
558 info
->curr_minor_hash
= 0;
562 info
->last_pos
= ctx
->pos
;
566 static int ext4_release_dir(struct inode
*inode
, struct file
*filp
)
568 if (filp
->private_data
)
569 ext4_htree_free_dir_info(filp
->private_data
);
574 const struct file_operations ext4_dir_operations
= {
575 .llseek
= ext4_dir_llseek
,
576 .read
= generic_read_dir
,
577 .iterate
= ext4_readdir
,
578 .unlocked_ioctl
= ext4_ioctl
,
580 .compat_ioctl
= ext4_compat_ioctl
,
582 .fsync
= ext4_sync_file
,
583 .release
= ext4_release_dir
,