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 chould potentially get coverted 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
,
137 offset
= ctx
->pos
& (sb
->s_blocksize
- 1);
139 while (ctx
->pos
< inode
->i_size
) {
140 struct ext4_map_blocks map
;
141 struct buffer_head
*bh
= NULL
;
143 map
.m_lblk
= ctx
->pos
>> EXT4_BLOCK_SIZE_BITS(sb
);
145 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
147 pgoff_t index
= map
.m_pblk
>>
148 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
149 if (!ra_has_index(&file
->f_ra
, index
))
150 page_cache_sync_readahead(
151 sb
->s_bdev
->bd_inode
->i_mapping
,
154 file
->f_ra
.prev_pos
= (loff_t
)index
<< PAGE_CACHE_SHIFT
;
155 bh
= ext4_bread(NULL
, inode
, map
.m_lblk
, 0, &err
);
159 * We ignore I/O errors on directories so users have a chance
160 * of recovering data when there's a bad sector
163 if (!dir_has_error
) {
164 EXT4_ERROR_FILE(file
, 0,
165 "directory contains a "
166 "hole at offset %llu",
167 (unsigned long long) ctx
->pos
);
170 /* corrupt size? Maybe no more blocks to read */
171 if (ctx
->pos
> inode
->i_blocks
<< 9)
173 ctx
->pos
+= sb
->s_blocksize
- offset
;
177 /* Check the checksum */
178 if (!buffer_verified(bh
) &&
179 !ext4_dirent_csum_verify(inode
,
180 (struct ext4_dir_entry
*)bh
->b_data
)) {
181 EXT4_ERROR_FILE(file
, 0, "directory fails checksum "
183 (unsigned long long)ctx
->pos
);
184 ctx
->pos
+= sb
->s_blocksize
- offset
;
188 set_buffer_verified(bh
);
190 /* If the dir block has changed since the last call to
191 * readdir(2), then we might be pointing to an invalid
192 * dirent right now. Scan from the start of the block
194 if (file
->f_version
!= inode
->i_version
) {
195 for (i
= 0; i
< sb
->s_blocksize
&& i
< offset
; ) {
196 de
= (struct ext4_dir_entry_2
*)
198 /* It's too expensive to do a full
199 * dirent test each time round this
200 * loop, but we do have to test at
201 * least that it is non-zero. A
202 * failure will be detected in the
203 * dirent test below. */
204 if (ext4_rec_len_from_disk(de
->rec_len
,
205 sb
->s_blocksize
) < EXT4_DIR_REC_LEN(1))
207 i
+= ext4_rec_len_from_disk(de
->rec_len
,
211 ctx
->pos
= (ctx
->pos
& ~(sb
->s_blocksize
- 1))
213 file
->f_version
= inode
->i_version
;
216 while (ctx
->pos
< inode
->i_size
217 && offset
< sb
->s_blocksize
) {
218 de
= (struct ext4_dir_entry_2
*) (bh
->b_data
+ offset
);
219 if (ext4_check_dir_entry(inode
, file
, de
, bh
,
220 bh
->b_data
, bh
->b_size
,
223 * On error, skip to the next block
225 ctx
->pos
= (ctx
->pos
|
226 (sb
->s_blocksize
- 1)) + 1;
229 offset
+= ext4_rec_len_from_disk(de
->rec_len
,
231 if (le32_to_cpu(de
->inode
)) {
232 if (!dir_emit(ctx
, de
->name
,
234 le32_to_cpu(de
->inode
),
235 get_dtype(sb
, de
->file_type
))) {
240 ctx
->pos
+= ext4_rec_len_from_disk(de
->rec_len
,
245 if (ctx
->pos
< inode
->i_size
) {
246 if (!dir_relax(inode
))
253 static inline int is_32bit_api(void)
256 return is_compat_task();
258 return (BITS_PER_LONG
== 32);
263 * These functions convert from the major/minor hash to an f_pos
264 * value for dx directories
266 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
267 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
268 * directly on both 32-bit and 64-bit nodes, under such case, neither
269 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
271 static inline loff_t
hash2pos(struct file
*filp
, __u32 major
, __u32 minor
)
273 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
274 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
277 return ((__u64
)(major
>> 1) << 32) | (__u64
)minor
;
280 static inline __u32
pos2maj_hash(struct file
*filp
, loff_t pos
)
282 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
283 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
284 return (pos
<< 1) & 0xffffffff;
286 return ((pos
>> 32) << 1) & 0xffffffff;
289 static inline __u32
pos2min_hash(struct file
*filp
, loff_t pos
)
291 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
292 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
295 return pos
& 0xffffffff;
299 * Return 32- or 64-bit end-of-file for dx directories
301 static inline loff_t
ext4_get_htree_eof(struct file
*filp
)
303 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
304 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
305 return EXT4_HTREE_EOF_32BIT
;
307 return EXT4_HTREE_EOF_64BIT
;
312 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
313 * directories, where the "offset" is in terms of the filename hash
314 * value instead of the byte offset.
316 * Because we may return a 64-bit hash that is well beyond offset limits,
317 * we need to pass the max hash as the maximum allowable offset in
318 * the htree directory case.
320 * For non-htree, ext4_llseek already chooses the proper max offset.
322 static loff_t
ext4_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
324 struct inode
*inode
= file
->f_mapping
->host
;
325 int dx_dir
= is_dx_dir(inode
);
326 loff_t htree_max
= ext4_get_htree_eof(file
);
329 return generic_file_llseek_size(file
, offset
, whence
,
330 htree_max
, htree_max
);
332 return ext4_llseek(file
, offset
, whence
);
336 * This structure holds the nodes of the red-black tree used to store
337 * the directory entry in hash order.
342 struct rb_node rb_hash
;
351 * This functoin implements a non-recursive way of freeing all of the
352 * nodes in the red-black tree.
354 static void free_rb_tree_fname(struct rb_root
*root
)
356 struct rb_node
*n
= root
->rb_node
;
357 struct rb_node
*parent
;
361 /* Do the node's children first */
371 * The node has no children; free it, and then zero
372 * out parent's link to it. Finally go to the
373 * beginning of the loop and try to free the parent
376 parent
= rb_parent(n
);
377 fname
= rb_entry(n
, struct fname
, rb_hash
);
379 struct fname
*old
= fname
;
385 else if (parent
->rb_left
== n
)
386 parent
->rb_left
= NULL
;
387 else if (parent
->rb_right
== n
)
388 parent
->rb_right
= NULL
;
394 static struct dir_private_info
*ext4_htree_create_dir_info(struct file
*filp
,
397 struct dir_private_info
*p
;
399 p
= kzalloc(sizeof(struct dir_private_info
), GFP_KERNEL
);
402 p
->curr_hash
= pos2maj_hash(filp
, pos
);
403 p
->curr_minor_hash
= pos2min_hash(filp
, pos
);
407 void ext4_htree_free_dir_info(struct dir_private_info
*p
)
409 free_rb_tree_fname(&p
->root
);
414 * Given a directory entry, enter it into the fname rb tree.
416 int ext4_htree_store_dirent(struct file
*dir_file
, __u32 hash
,
418 struct ext4_dir_entry_2
*dirent
)
420 struct rb_node
**p
, *parent
= NULL
;
421 struct fname
*fname
, *new_fn
;
422 struct dir_private_info
*info
;
425 info
= dir_file
->private_data
;
426 p
= &info
->root
.rb_node
;
428 /* Create and allocate the fname structure */
429 len
= sizeof(struct fname
) + dirent
->name_len
+ 1;
430 new_fn
= kzalloc(len
, GFP_KERNEL
);
434 new_fn
->minor_hash
= minor_hash
;
435 new_fn
->inode
= le32_to_cpu(dirent
->inode
);
436 new_fn
->name_len
= dirent
->name_len
;
437 new_fn
->file_type
= dirent
->file_type
;
438 memcpy(new_fn
->name
, dirent
->name
, dirent
->name_len
);
439 new_fn
->name
[dirent
->name_len
] = 0;
443 fname
= rb_entry(parent
, struct fname
, rb_hash
);
446 * If the hash and minor hash match up, then we put
447 * them on a linked list. This rarely happens...
449 if ((new_fn
->hash
== fname
->hash
) &&
450 (new_fn
->minor_hash
== fname
->minor_hash
)) {
451 new_fn
->next
= fname
->next
;
452 fname
->next
= new_fn
;
456 if (new_fn
->hash
< fname
->hash
)
458 else if (new_fn
->hash
> fname
->hash
)
460 else if (new_fn
->minor_hash
< fname
->minor_hash
)
462 else /* if (new_fn->minor_hash > fname->minor_hash) */
466 rb_link_node(&new_fn
->rb_hash
, parent
, p
);
467 rb_insert_color(&new_fn
->rb_hash
, &info
->root
);
474 * This is a helper function for ext4_dx_readdir. It calls filldir
475 * for all entres on the fname linked list. (Normally there is only
476 * one entry on the linked list, unless there are 62 bit hash collisions.)
478 static int call_filldir(struct file
*file
, struct dir_context
*ctx
,
481 struct dir_private_info
*info
= file
->private_data
;
482 struct inode
*inode
= file_inode(file
);
483 struct super_block
*sb
= inode
->i_sb
;
486 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: comm %s: "
487 "called with null fname?!?", __func__
, __LINE__
,
488 inode
->i_ino
, current
->comm
);
491 ctx
->pos
= hash2pos(file
, fname
->hash
, fname
->minor_hash
);
493 if (!dir_emit(ctx
, fname
->name
,
496 get_dtype(sb
, fname
->file_type
))) {
497 info
->extra_fname
= fname
;
505 static int ext4_dx_readdir(struct file
*file
, struct dir_context
*ctx
)
507 struct dir_private_info
*info
= file
->private_data
;
508 struct inode
*inode
= file_inode(file
);
513 info
= ext4_htree_create_dir_info(file
, ctx
->pos
);
516 file
->private_data
= info
;
519 if (ctx
->pos
== ext4_get_htree_eof(file
))
522 /* Some one has messed with f_pos; reset the world */
523 if (info
->last_pos
!= ctx
->pos
) {
524 free_rb_tree_fname(&info
->root
);
525 info
->curr_node
= NULL
;
526 info
->extra_fname
= NULL
;
527 info
->curr_hash
= pos2maj_hash(file
, ctx
->pos
);
528 info
->curr_minor_hash
= pos2min_hash(file
, ctx
->pos
);
532 * If there are any leftover names on the hash collision
533 * chain, return them first.
535 if (info
->extra_fname
) {
536 if (call_filldir(file
, ctx
, info
->extra_fname
))
538 info
->extra_fname
= NULL
;
540 } else if (!info
->curr_node
)
541 info
->curr_node
= rb_first(&info
->root
);
545 * Fill the rbtree if we have no more entries,
546 * or the inode has changed since we last read in the
549 if ((!info
->curr_node
) ||
550 (file
->f_version
!= inode
->i_version
)) {
551 info
->curr_node
= NULL
;
552 free_rb_tree_fname(&info
->root
);
553 file
->f_version
= inode
->i_version
;
554 ret
= ext4_htree_fill_tree(file
, info
->curr_hash
,
555 info
->curr_minor_hash
,
560 ctx
->pos
= ext4_get_htree_eof(file
);
563 info
->curr_node
= rb_first(&info
->root
);
566 fname
= rb_entry(info
->curr_node
, struct fname
, rb_hash
);
567 info
->curr_hash
= fname
->hash
;
568 info
->curr_minor_hash
= fname
->minor_hash
;
569 if (call_filldir(file
, ctx
, fname
))
572 info
->curr_node
= rb_next(info
->curr_node
);
573 if (info
->curr_node
) {
574 fname
= rb_entry(info
->curr_node
, struct fname
,
576 info
->curr_hash
= fname
->hash
;
577 info
->curr_minor_hash
= fname
->minor_hash
;
579 if (info
->next_hash
== ~0) {
580 ctx
->pos
= ext4_get_htree_eof(file
);
583 info
->curr_hash
= info
->next_hash
;
584 info
->curr_minor_hash
= 0;
588 info
->last_pos
= ctx
->pos
;
592 static int ext4_release_dir(struct inode
*inode
, struct file
*filp
)
594 if (filp
->private_data
)
595 ext4_htree_free_dir_info(filp
->private_data
);
600 const struct file_operations ext4_dir_operations
= {
601 .llseek
= ext4_dir_llseek
,
602 .read
= generic_read_dir
,
603 .iterate
= ext4_readdir
,
604 .unlocked_ioctl
= ext4_ioctl
,
606 .compat_ioctl
= ext4_compat_ioctl
,
608 .fsync
= ext4_sync_file
,
609 .release
= ext4_release_dir
,