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/buffer_head.h>
26 #include <linux/slab.h>
30 static int ext4_dx_readdir(struct file
*, struct dir_context
*);
33 * Check if the given dir-inode refers to an htree-indexed directory
34 * (or a directory which could potentially get converted to use htree
37 * Return 1 if it is a dx dir, 0 if not
39 static int is_dx_dir(struct inode
*inode
)
41 struct super_block
*sb
= inode
->i_sb
;
43 if (ext4_has_feature_dir_index(inode
->i_sb
) &&
44 ((ext4_test_inode_flag(inode
, EXT4_INODE_INDEX
)) ||
45 ((inode
->i_size
>> sb
->s_blocksize_bits
) == 1) ||
46 ext4_has_inline_data(inode
)))
53 * Return 0 if the directory entry is OK, and 1 if there is a problem
55 * Note: this is the opposite of what ext2 and ext3 historically returned...
57 * bh passed here can be an inode block or a dir data block, depending
58 * on the inode inline data flag.
60 int __ext4_check_dir_entry(const char *function
, unsigned int line
,
61 struct inode
*dir
, struct file
*filp
,
62 struct ext4_dir_entry_2
*de
,
63 struct buffer_head
*bh
, char *buf
, int size
,
66 const char *error_msg
= NULL
;
67 const int rlen
= ext4_rec_len_from_disk(de
->rec_len
,
68 dir
->i_sb
->s_blocksize
);
70 if (unlikely(rlen
< EXT4_DIR_REC_LEN(1)))
71 error_msg
= "rec_len is smaller than minimal";
72 else if (unlikely(rlen
% 4 != 0))
73 error_msg
= "rec_len % 4 != 0";
74 else if (unlikely(rlen
< EXT4_DIR_REC_LEN(de
->name_len
)))
75 error_msg
= "rec_len is too small for name_len";
76 else if (unlikely(((char *) de
- buf
) + rlen
> size
))
77 error_msg
= "directory entry overrun";
78 else if (unlikely(le32_to_cpu(de
->inode
) >
79 le32_to_cpu(EXT4_SB(dir
->i_sb
)->s_es
->s_inodes_count
)))
80 error_msg
= "inode out of bounds";
85 ext4_error_file(filp
, function
, line
, bh
->b_blocknr
,
86 "bad entry in directory: %s - offset=%u, "
87 "inode=%u, rec_len=%d, name_len=%d, size=%d",
88 error_msg
, offset
, le32_to_cpu(de
->inode
),
89 rlen
, de
->name_len
, size
);
91 ext4_error_inode(dir
, function
, line
, bh
->b_blocknr
,
92 "bad entry in directory: %s - offset=%u, "
93 "inode=%u, rec_len=%d, name_len=%d, size=%d",
94 error_msg
, offset
, le32_to_cpu(de
->inode
),
95 rlen
, de
->name_len
, size
);
100 static int ext4_readdir(struct file
*file
, struct dir_context
*ctx
)
104 struct ext4_dir_entry_2
*de
;
106 struct inode
*inode
= file_inode(file
);
107 struct super_block
*sb
= inode
->i_sb
;
108 struct buffer_head
*bh
= NULL
;
109 int dir_has_error
= 0;
110 struct ext4_str fname_crypto_str
= {.name
= NULL
, .len
= 0};
112 if (ext4_encrypted_inode(inode
)) {
113 err
= ext4_get_encryption_info(inode
);
114 if (err
&& err
!= -ENOKEY
)
118 if (is_dx_dir(inode
)) {
119 err
= ext4_dx_readdir(file
, ctx
);
120 if (err
!= ERR_BAD_DX_DIR
) {
124 * We don't set the inode dirty flag since it's not
125 * critical that it get flushed back to the disk.
127 ext4_clear_inode_flag(file_inode(file
),
131 if (ext4_has_inline_data(inode
)) {
132 int has_inline_data
= 1;
133 err
= ext4_read_inline_dir(file
, ctx
,
139 if (ext4_encrypted_inode(inode
)) {
140 err
= ext4_fname_crypto_alloc_buffer(inode
, EXT4_NAME_LEN
,
146 offset
= ctx
->pos
& (sb
->s_blocksize
- 1);
148 while (ctx
->pos
< inode
->i_size
) {
149 struct ext4_map_blocks map
;
151 map
.m_lblk
= ctx
->pos
>> EXT4_BLOCK_SIZE_BITS(sb
);
153 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
155 pgoff_t index
= map
.m_pblk
>>
156 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
157 if (!ra_has_index(&file
->f_ra
, index
))
158 page_cache_sync_readahead(
159 sb
->s_bdev
->bd_inode
->i_mapping
,
162 file
->f_ra
.prev_pos
= (loff_t
)index
<< PAGE_CACHE_SHIFT
;
163 bh
= ext4_bread(NULL
, inode
, map
.m_lblk
, 0);
169 if (!dir_has_error
) {
170 EXT4_ERROR_FILE(file
, 0,
171 "directory contains a "
172 "hole at offset %llu",
173 (unsigned long long) ctx
->pos
);
176 /* corrupt size? Maybe no more blocks to read */
177 if (ctx
->pos
> inode
->i_blocks
<< 9)
179 ctx
->pos
+= sb
->s_blocksize
- offset
;
183 /* Check the checksum */
184 if (!buffer_verified(bh
) &&
185 !ext4_dirent_csum_verify(inode
,
186 (struct ext4_dir_entry
*)bh
->b_data
)) {
187 EXT4_ERROR_FILE(file
, 0, "directory fails checksum "
189 (unsigned long long)ctx
->pos
);
190 ctx
->pos
+= sb
->s_blocksize
- offset
;
195 set_buffer_verified(bh
);
197 /* If the dir block has changed since the last call to
198 * readdir(2), then we might be pointing to an invalid
199 * dirent right now. Scan from the start of the block
201 if (file
->f_version
!= inode
->i_version
) {
202 for (i
= 0; i
< sb
->s_blocksize
&& i
< offset
; ) {
203 de
= (struct ext4_dir_entry_2
*)
205 /* It's too expensive to do a full
206 * dirent test each time round this
207 * loop, but we do have to test at
208 * least that it is non-zero. A
209 * failure will be detected in the
210 * dirent test below. */
211 if (ext4_rec_len_from_disk(de
->rec_len
,
212 sb
->s_blocksize
) < EXT4_DIR_REC_LEN(1))
214 i
+= ext4_rec_len_from_disk(de
->rec_len
,
218 ctx
->pos
= (ctx
->pos
& ~(sb
->s_blocksize
- 1))
220 file
->f_version
= inode
->i_version
;
223 while (ctx
->pos
< inode
->i_size
224 && offset
< sb
->s_blocksize
) {
225 de
= (struct ext4_dir_entry_2
*) (bh
->b_data
+ offset
);
226 if (ext4_check_dir_entry(inode
, file
, de
, bh
,
227 bh
->b_data
, bh
->b_size
,
230 * On error, skip to the next block
232 ctx
->pos
= (ctx
->pos
|
233 (sb
->s_blocksize
- 1)) + 1;
236 offset
+= ext4_rec_len_from_disk(de
->rec_len
,
238 if (le32_to_cpu(de
->inode
)) {
239 if (!ext4_encrypted_inode(inode
)) {
240 if (!dir_emit(ctx
, de
->name
,
242 le32_to_cpu(de
->inode
),
243 get_dtype(sb
, de
->file_type
)))
246 int save_len
= fname_crypto_str
.len
;
248 /* Directory is encrypted */
249 err
= ext4_fname_disk_to_usr(inode
,
250 NULL
, de
, &fname_crypto_str
);
251 fname_crypto_str
.len
= save_len
;
255 fname_crypto_str
.name
, err
,
256 le32_to_cpu(de
->inode
),
257 get_dtype(sb
, de
->file_type
)))
261 ctx
->pos
+= ext4_rec_len_from_disk(de
->rec_len
,
264 if ((ctx
->pos
< inode
->i_size
) && !dir_relax(inode
))
273 #ifdef CONFIG_EXT4_FS_ENCRYPTION
274 ext4_fname_crypto_free_buffer(&fname_crypto_str
);
280 static inline int is_32bit_api(void)
283 return is_compat_task();
285 return (BITS_PER_LONG
== 32);
290 * These functions convert from the major/minor hash to an f_pos
291 * value for dx directories
293 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
294 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
295 * directly on both 32-bit and 64-bit nodes, under such case, neither
296 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
298 static inline loff_t
hash2pos(struct file
*filp
, __u32 major
, __u32 minor
)
300 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
301 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
304 return ((__u64
)(major
>> 1) << 32) | (__u64
)minor
;
307 static inline __u32
pos2maj_hash(struct file
*filp
, loff_t pos
)
309 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
310 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
311 return (pos
<< 1) & 0xffffffff;
313 return ((pos
>> 32) << 1) & 0xffffffff;
316 static inline __u32
pos2min_hash(struct file
*filp
, loff_t pos
)
318 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
319 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
322 return pos
& 0xffffffff;
326 * Return 32- or 64-bit end-of-file for dx directories
328 static inline loff_t
ext4_get_htree_eof(struct file
*filp
)
330 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
331 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
332 return EXT4_HTREE_EOF_32BIT
;
334 return EXT4_HTREE_EOF_64BIT
;
339 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
340 * directories, where the "offset" is in terms of the filename hash
341 * value instead of the byte offset.
343 * Because we may return a 64-bit hash that is well beyond offset limits,
344 * we need to pass the max hash as the maximum allowable offset in
345 * the htree directory case.
347 * For non-htree, ext4_llseek already chooses the proper max offset.
349 static loff_t
ext4_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
351 struct inode
*inode
= file
->f_mapping
->host
;
352 int dx_dir
= is_dx_dir(inode
);
353 loff_t htree_max
= ext4_get_htree_eof(file
);
356 return generic_file_llseek_size(file
, offset
, whence
,
357 htree_max
, htree_max
);
359 return ext4_llseek(file
, offset
, whence
);
363 * This structure holds the nodes of the red-black tree used to store
364 * the directory entry in hash order.
369 struct rb_node rb_hash
;
378 * This functoin implements a non-recursive way of freeing all of the
379 * nodes in the red-black tree.
381 static void free_rb_tree_fname(struct rb_root
*root
)
383 struct fname
*fname
, *next
;
385 rbtree_postorder_for_each_entry_safe(fname
, next
, root
, rb_hash
)
387 struct fname
*old
= fname
;
396 static struct dir_private_info
*ext4_htree_create_dir_info(struct file
*filp
,
399 struct dir_private_info
*p
;
401 p
= kzalloc(sizeof(struct dir_private_info
), GFP_KERNEL
);
404 p
->curr_hash
= pos2maj_hash(filp
, pos
);
405 p
->curr_minor_hash
= pos2min_hash(filp
, pos
);
409 void ext4_htree_free_dir_info(struct dir_private_info
*p
)
411 free_rb_tree_fname(&p
->root
);
416 * Given a directory entry, enter it into the fname rb tree.
418 * When filename encryption is enabled, the dirent will hold the
419 * encrypted filename, while the htree will hold decrypted filename.
420 * The decrypted filename is passed in via ent_name. parameter.
422 int ext4_htree_store_dirent(struct file
*dir_file
, __u32 hash
,
424 struct ext4_dir_entry_2
*dirent
,
425 struct ext4_str
*ent_name
)
427 struct rb_node
**p
, *parent
= NULL
;
428 struct fname
*fname
, *new_fn
;
429 struct dir_private_info
*info
;
432 info
= dir_file
->private_data
;
433 p
= &info
->root
.rb_node
;
435 /* Create and allocate the fname structure */
436 len
= sizeof(struct fname
) + ent_name
->len
+ 1;
437 new_fn
= kzalloc(len
, GFP_KERNEL
);
441 new_fn
->minor_hash
= minor_hash
;
442 new_fn
->inode
= le32_to_cpu(dirent
->inode
);
443 new_fn
->name_len
= ent_name
->len
;
444 new_fn
->file_type
= dirent
->file_type
;
445 memcpy(new_fn
->name
, ent_name
->name
, ent_name
->len
);
446 new_fn
->name
[ent_name
->len
] = 0;
450 fname
= rb_entry(parent
, struct fname
, rb_hash
);
453 * If the hash and minor hash match up, then we put
454 * them on a linked list. This rarely happens...
456 if ((new_fn
->hash
== fname
->hash
) &&
457 (new_fn
->minor_hash
== fname
->minor_hash
)) {
458 new_fn
->next
= fname
->next
;
459 fname
->next
= new_fn
;
463 if (new_fn
->hash
< fname
->hash
)
465 else if (new_fn
->hash
> fname
->hash
)
467 else if (new_fn
->minor_hash
< fname
->minor_hash
)
469 else /* if (new_fn->minor_hash > fname->minor_hash) */
473 rb_link_node(&new_fn
->rb_hash
, parent
, p
);
474 rb_insert_color(&new_fn
->rb_hash
, &info
->root
);
481 * This is a helper function for ext4_dx_readdir. It calls filldir
482 * for all entres on the fname linked list. (Normally there is only
483 * one entry on the linked list, unless there are 62 bit hash collisions.)
485 static int call_filldir(struct file
*file
, struct dir_context
*ctx
,
488 struct dir_private_info
*info
= file
->private_data
;
489 struct inode
*inode
= file_inode(file
);
490 struct super_block
*sb
= inode
->i_sb
;
493 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: comm %s: "
494 "called with null fname?!?", __func__
, __LINE__
,
495 inode
->i_ino
, current
->comm
);
498 ctx
->pos
= hash2pos(file
, fname
->hash
, fname
->minor_hash
);
500 if (!dir_emit(ctx
, fname
->name
,
503 get_dtype(sb
, fname
->file_type
))) {
504 info
->extra_fname
= fname
;
512 static int ext4_dx_readdir(struct file
*file
, struct dir_context
*ctx
)
514 struct dir_private_info
*info
= file
->private_data
;
515 struct inode
*inode
= file_inode(file
);
520 info
= ext4_htree_create_dir_info(file
, ctx
->pos
);
523 file
->private_data
= info
;
526 if (ctx
->pos
== ext4_get_htree_eof(file
))
529 /* Some one has messed with f_pos; reset the world */
530 if (info
->last_pos
!= ctx
->pos
) {
531 free_rb_tree_fname(&info
->root
);
532 info
->curr_node
= NULL
;
533 info
->extra_fname
= NULL
;
534 info
->curr_hash
= pos2maj_hash(file
, ctx
->pos
);
535 info
->curr_minor_hash
= pos2min_hash(file
, ctx
->pos
);
539 * If there are any leftover names on the hash collision
540 * chain, return them first.
542 if (info
->extra_fname
) {
543 if (call_filldir(file
, ctx
, info
->extra_fname
))
545 info
->extra_fname
= NULL
;
547 } else if (!info
->curr_node
)
548 info
->curr_node
= rb_first(&info
->root
);
552 * Fill the rbtree if we have no more entries,
553 * or the inode has changed since we last read in the
556 if ((!info
->curr_node
) ||
557 (file
->f_version
!= inode
->i_version
)) {
558 info
->curr_node
= NULL
;
559 free_rb_tree_fname(&info
->root
);
560 file
->f_version
= inode
->i_version
;
561 ret
= ext4_htree_fill_tree(file
, info
->curr_hash
,
562 info
->curr_minor_hash
,
567 ctx
->pos
= ext4_get_htree_eof(file
);
570 info
->curr_node
= rb_first(&info
->root
);
573 fname
= rb_entry(info
->curr_node
, struct fname
, rb_hash
);
574 info
->curr_hash
= fname
->hash
;
575 info
->curr_minor_hash
= fname
->minor_hash
;
576 if (call_filldir(file
, ctx
, fname
))
579 info
->curr_node
= rb_next(info
->curr_node
);
580 if (info
->curr_node
) {
581 fname
= rb_entry(info
->curr_node
, struct fname
,
583 info
->curr_hash
= fname
->hash
;
584 info
->curr_minor_hash
= fname
->minor_hash
;
586 if (info
->next_hash
== ~0) {
587 ctx
->pos
= ext4_get_htree_eof(file
);
590 info
->curr_hash
= info
->next_hash
;
591 info
->curr_minor_hash
= 0;
595 info
->last_pos
= ctx
->pos
;
599 static int ext4_dir_open(struct inode
* inode
, struct file
* filp
)
601 if (ext4_encrypted_inode(inode
))
602 return ext4_get_encryption_info(inode
) ? -EACCES
: 0;
606 static int ext4_release_dir(struct inode
*inode
, struct file
*filp
)
608 if (filp
->private_data
)
609 ext4_htree_free_dir_info(filp
->private_data
);
614 int ext4_check_all_de(struct inode
*dir
, struct buffer_head
*bh
, void *buf
,
617 struct ext4_dir_entry_2
*de
;
619 unsigned int offset
= 0;
622 de
= (struct ext4_dir_entry_2
*)buf
;
623 top
= buf
+ buf_size
;
624 while ((char *) de
< top
) {
625 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
626 buf
, buf_size
, offset
))
627 return -EFSCORRUPTED
;
628 nlen
= EXT4_DIR_REC_LEN(de
->name_len
);
629 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
630 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
633 if ((char *) de
> top
)
634 return -EFSCORRUPTED
;
639 const struct file_operations ext4_dir_operations
= {
640 .llseek
= ext4_dir_llseek
,
641 .read
= generic_read_dir
,
642 .iterate
= ext4_readdir
,
643 .unlocked_ioctl
= ext4_ioctl
,
645 .compat_ioctl
= ext4_compat_ioctl
,
647 .fsync
= ext4_sync_file
,
648 .open
= ext4_dir_open
,
649 .release
= ext4_release_dir
,