1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/dir.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * ext4 directory handling functions
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
21 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
26 #include <linux/buffer_head.h>
27 #include <linux/slab.h>
28 #include <linux/iversion.h>
29 #include <linux/unicode.h>
33 static int ext4_dx_readdir(struct file
*, struct dir_context
*);
36 * is_dx_dir() - check if a directory is using htree indexing
37 * @inode: directory inode
39 * Check if the given dir-inode refers to an htree-indexed directory
40 * (or a directory which could potentially get converted to use htree
43 * Return 1 if it is a dx dir, 0 if not
45 static int is_dx_dir(struct inode
*inode
)
47 struct super_block
*sb
= inode
->i_sb
;
49 if (ext4_has_feature_dir_index(inode
->i_sb
) &&
50 ((ext4_test_inode_flag(inode
, EXT4_INODE_INDEX
)) ||
51 ((inode
->i_size
>> sb
->s_blocksize_bits
) == 1) ||
52 ext4_has_inline_data(inode
)))
59 * Return 0 if the directory entry is OK, and 1 if there is a problem
61 * Note: this is the opposite of what ext2 and ext3 historically returned...
63 * bh passed here can be an inode block or a dir data block, depending
64 * on the inode inline data flag.
66 int __ext4_check_dir_entry(const char *function
, unsigned int line
,
67 struct inode
*dir
, struct file
*filp
,
68 struct ext4_dir_entry_2
*de
,
69 struct buffer_head
*bh
, char *buf
, int size
,
72 const char *error_msg
= NULL
;
73 const int rlen
= ext4_rec_len_from_disk(de
->rec_len
,
74 dir
->i_sb
->s_blocksize
);
75 const int next_offset
= ((char *) de
- buf
) + rlen
;
77 if (unlikely(rlen
< EXT4_DIR_REC_LEN(1)))
78 error_msg
= "rec_len is smaller than minimal";
79 else if (unlikely(rlen
% 4 != 0))
80 error_msg
= "rec_len % 4 != 0";
81 else if (unlikely(rlen
< EXT4_DIR_REC_LEN(de
->name_len
)))
82 error_msg
= "rec_len is too small for name_len";
83 else if (unlikely(next_offset
> size
))
84 error_msg
= "directory entry overrun";
85 else if (unlikely(next_offset
> size
- EXT4_DIR_REC_LEN(1) &&
87 error_msg
= "directory entry too close to block end";
88 else if (unlikely(le32_to_cpu(de
->inode
) >
89 le32_to_cpu(EXT4_SB(dir
->i_sb
)->s_es
->s_inodes_count
)))
90 error_msg
= "inode out of bounds";
95 ext4_error_file(filp
, function
, line
, bh
->b_blocknr
,
96 "bad entry in directory: %s - offset=%u, "
97 "inode=%u, rec_len=%d, name_len=%d, size=%d",
98 error_msg
, offset
, le32_to_cpu(de
->inode
),
99 rlen
, de
->name_len
, size
);
101 ext4_error_inode(dir
, function
, line
, bh
->b_blocknr
,
102 "bad entry in directory: %s - offset=%u, "
103 "inode=%u, rec_len=%d, name_len=%d, size=%d",
104 error_msg
, offset
, le32_to_cpu(de
->inode
),
105 rlen
, de
->name_len
, size
);
110 static int ext4_readdir(struct file
*file
, struct dir_context
*ctx
)
114 struct ext4_dir_entry_2
*de
;
116 struct inode
*inode
= file_inode(file
);
117 struct super_block
*sb
= inode
->i_sb
;
118 struct buffer_head
*bh
= NULL
;
119 struct fscrypt_str fstr
= FSTR_INIT(NULL
, 0);
121 if (IS_ENCRYPTED(inode
)) {
122 err
= fscrypt_get_encryption_info(inode
);
123 if (err
&& err
!= -ENOKEY
)
127 if (is_dx_dir(inode
)) {
128 err
= ext4_dx_readdir(file
, ctx
);
129 if (err
!= ERR_BAD_DX_DIR
) {
132 /* Can we just clear INDEX flag to ignore htree information? */
133 if (!ext4_has_metadata_csum(sb
)) {
135 * We don't set the inode dirty flag since it's not
136 * critical that it gets flushed back to the disk.
138 ext4_clear_inode_flag(inode
, EXT4_INODE_INDEX
);
142 if (ext4_has_inline_data(inode
)) {
143 int has_inline_data
= 1;
144 err
= ext4_read_inline_dir(file
, ctx
,
150 if (IS_ENCRYPTED(inode
)) {
151 err
= fscrypt_fname_alloc_buffer(inode
, EXT4_NAME_LEN
, &fstr
);
156 while (ctx
->pos
< inode
->i_size
) {
157 struct ext4_map_blocks map
;
159 if (fatal_signal_pending(current
)) {
164 offset
= ctx
->pos
& (sb
->s_blocksize
- 1);
165 map
.m_lblk
= ctx
->pos
>> EXT4_BLOCK_SIZE_BITS(sb
);
167 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
169 /* m_len should never be zero but let's avoid
170 * an infinite loop if it somehow is */
173 ctx
->pos
+= map
.m_len
* sb
->s_blocksize
;
177 pgoff_t index
= map
.m_pblk
>>
178 (PAGE_SHIFT
- inode
->i_blkbits
);
179 if (!ra_has_index(&file
->f_ra
, index
))
180 page_cache_sync_readahead(
181 sb
->s_bdev
->bd_inode
->i_mapping
,
184 file
->f_ra
.prev_pos
= (loff_t
)index
<< PAGE_SHIFT
;
185 bh
= ext4_bread(NULL
, inode
, map
.m_lblk
, 0);
194 /* corrupt size? Maybe no more blocks to read */
195 if (ctx
->pos
> inode
->i_blocks
<< 9)
197 ctx
->pos
+= sb
->s_blocksize
- offset
;
201 /* Check the checksum */
202 if (!buffer_verified(bh
) &&
203 !ext4_dirblock_csum_verify(inode
, bh
)) {
204 EXT4_ERROR_FILE(file
, 0, "directory fails checksum "
206 (unsigned long long)ctx
->pos
);
207 ctx
->pos
+= sb
->s_blocksize
- offset
;
212 set_buffer_verified(bh
);
214 /* If the dir block has changed since the last call to
215 * readdir(2), then we might be pointing to an invalid
216 * dirent right now. Scan from the start of the block
218 if (!inode_eq_iversion(inode
, file
->f_version
)) {
219 for (i
= 0; i
< sb
->s_blocksize
&& i
< offset
; ) {
220 de
= (struct ext4_dir_entry_2
*)
222 /* It's too expensive to do a full
223 * dirent test each time round this
224 * loop, but we do have to test at
225 * least that it is non-zero. A
226 * failure will be detected in the
227 * dirent test below. */
228 if (ext4_rec_len_from_disk(de
->rec_len
,
229 sb
->s_blocksize
) < EXT4_DIR_REC_LEN(1))
231 i
+= ext4_rec_len_from_disk(de
->rec_len
,
235 ctx
->pos
= (ctx
->pos
& ~(sb
->s_blocksize
- 1))
237 file
->f_version
= inode_query_iversion(inode
);
240 while (ctx
->pos
< inode
->i_size
241 && offset
< sb
->s_blocksize
) {
242 de
= (struct ext4_dir_entry_2
*) (bh
->b_data
+ offset
);
243 if (ext4_check_dir_entry(inode
, file
, de
, bh
,
244 bh
->b_data
, bh
->b_size
,
247 * On error, skip to the next block
249 ctx
->pos
= (ctx
->pos
|
250 (sb
->s_blocksize
- 1)) + 1;
253 offset
+= ext4_rec_len_from_disk(de
->rec_len
,
255 if (le32_to_cpu(de
->inode
)) {
256 if (!IS_ENCRYPTED(inode
)) {
257 if (!dir_emit(ctx
, de
->name
,
259 le32_to_cpu(de
->inode
),
260 get_dtype(sb
, de
->file_type
)))
263 int save_len
= fstr
.len
;
264 struct fscrypt_str de_name
=
268 /* Directory is encrypted */
269 err
= fscrypt_fname_disk_to_usr(inode
,
270 0, 0, &de_name
, &fstr
);
276 de_name
.name
, de_name
.len
,
277 le32_to_cpu(de
->inode
),
278 get_dtype(sb
, de
->file_type
)))
282 ctx
->pos
+= ext4_rec_len_from_disk(de
->rec_len
,
285 if ((ctx
->pos
< inode
->i_size
) && !dir_relax_shared(inode
))
294 fscrypt_fname_free_buffer(&fstr
);
299 static inline int is_32bit_api(void)
302 return in_compat_syscall();
304 return (BITS_PER_LONG
== 32);
309 * These functions convert from the major/minor hash to an f_pos
310 * value for dx directories
312 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
313 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
314 * directly on both 32-bit and 64-bit nodes, under such case, neither
315 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
317 static inline loff_t
hash2pos(struct file
*filp
, __u32 major
, __u32 minor
)
319 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
320 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
323 return ((__u64
)(major
>> 1) << 32) | (__u64
)minor
;
326 static inline __u32
pos2maj_hash(struct file
*filp
, loff_t pos
)
328 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
329 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
330 return (pos
<< 1) & 0xffffffff;
332 return ((pos
>> 32) << 1) & 0xffffffff;
335 static inline __u32
pos2min_hash(struct file
*filp
, loff_t pos
)
337 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
338 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
341 return pos
& 0xffffffff;
345 * Return 32- or 64-bit end-of-file for dx directories
347 static inline loff_t
ext4_get_htree_eof(struct file
*filp
)
349 if ((filp
->f_mode
& FMODE_32BITHASH
) ||
350 (!(filp
->f_mode
& FMODE_64BITHASH
) && is_32bit_api()))
351 return EXT4_HTREE_EOF_32BIT
;
353 return EXT4_HTREE_EOF_64BIT
;
358 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
359 * directories, where the "offset" is in terms of the filename hash
360 * value instead of the byte offset.
362 * Because we may return a 64-bit hash that is well beyond offset limits,
363 * we need to pass the max hash as the maximum allowable offset in
364 * the htree directory case.
366 * For non-htree, ext4_llseek already chooses the proper max offset.
368 static loff_t
ext4_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
370 struct inode
*inode
= file
->f_mapping
->host
;
371 int dx_dir
= is_dx_dir(inode
);
372 loff_t ret
, htree_max
= ext4_get_htree_eof(file
);
375 ret
= generic_file_llseek_size(file
, offset
, whence
,
376 htree_max
, htree_max
);
378 ret
= ext4_llseek(file
, offset
, whence
);
379 file
->f_version
= inode_peek_iversion(inode
) - 1;
384 * This structure holds the nodes of the red-black tree used to store
385 * the directory entry in hash order.
390 struct rb_node rb_hash
;
399 * This functoin implements a non-recursive way of freeing all of the
400 * nodes in the red-black tree.
402 static void free_rb_tree_fname(struct rb_root
*root
)
404 struct fname
*fname
, *next
;
406 rbtree_postorder_for_each_entry_safe(fname
, next
, root
, rb_hash
)
408 struct fname
*old
= fname
;
417 static struct dir_private_info
*ext4_htree_create_dir_info(struct file
*filp
,
420 struct dir_private_info
*p
;
422 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
425 p
->curr_hash
= pos2maj_hash(filp
, pos
);
426 p
->curr_minor_hash
= pos2min_hash(filp
, pos
);
430 void ext4_htree_free_dir_info(struct dir_private_info
*p
)
432 free_rb_tree_fname(&p
->root
);
437 * Given a directory entry, enter it into the fname rb tree.
439 * When filename encryption is enabled, the dirent will hold the
440 * encrypted filename, while the htree will hold decrypted filename.
441 * The decrypted filename is passed in via ent_name. parameter.
443 int ext4_htree_store_dirent(struct file
*dir_file
, __u32 hash
,
445 struct ext4_dir_entry_2
*dirent
,
446 struct fscrypt_str
*ent_name
)
448 struct rb_node
**p
, *parent
= NULL
;
449 struct fname
*fname
, *new_fn
;
450 struct dir_private_info
*info
;
453 info
= dir_file
->private_data
;
454 p
= &info
->root
.rb_node
;
456 /* Create and allocate the fname structure */
457 len
= sizeof(struct fname
) + ent_name
->len
+ 1;
458 new_fn
= kzalloc(len
, GFP_KERNEL
);
462 new_fn
->minor_hash
= minor_hash
;
463 new_fn
->inode
= le32_to_cpu(dirent
->inode
);
464 new_fn
->name_len
= ent_name
->len
;
465 new_fn
->file_type
= dirent
->file_type
;
466 memcpy(new_fn
->name
, ent_name
->name
, ent_name
->len
);
467 new_fn
->name
[ent_name
->len
] = 0;
471 fname
= rb_entry(parent
, struct fname
, rb_hash
);
474 * If the hash and minor hash match up, then we put
475 * them on a linked list. This rarely happens...
477 if ((new_fn
->hash
== fname
->hash
) &&
478 (new_fn
->minor_hash
== fname
->minor_hash
)) {
479 new_fn
->next
= fname
->next
;
480 fname
->next
= new_fn
;
484 if (new_fn
->hash
< fname
->hash
)
486 else if (new_fn
->hash
> fname
->hash
)
488 else if (new_fn
->minor_hash
< fname
->minor_hash
)
490 else /* if (new_fn->minor_hash > fname->minor_hash) */
494 rb_link_node(&new_fn
->rb_hash
, parent
, p
);
495 rb_insert_color(&new_fn
->rb_hash
, &info
->root
);
502 * This is a helper function for ext4_dx_readdir. It calls filldir
503 * for all entres on the fname linked list. (Normally there is only
504 * one entry on the linked list, unless there are 62 bit hash collisions.)
506 static int call_filldir(struct file
*file
, struct dir_context
*ctx
,
509 struct dir_private_info
*info
= file
->private_data
;
510 struct inode
*inode
= file_inode(file
);
511 struct super_block
*sb
= inode
->i_sb
;
514 ext4_msg(sb
, KERN_ERR
, "%s:%d: inode #%lu: comm %s: "
515 "called with null fname?!?", __func__
, __LINE__
,
516 inode
->i_ino
, current
->comm
);
519 ctx
->pos
= hash2pos(file
, fname
->hash
, fname
->minor_hash
);
521 if (!dir_emit(ctx
, fname
->name
,
524 get_dtype(sb
, fname
->file_type
))) {
525 info
->extra_fname
= fname
;
533 static int ext4_dx_readdir(struct file
*file
, struct dir_context
*ctx
)
535 struct dir_private_info
*info
= file
->private_data
;
536 struct inode
*inode
= file_inode(file
);
541 info
= ext4_htree_create_dir_info(file
, ctx
->pos
);
544 file
->private_data
= info
;
547 if (ctx
->pos
== ext4_get_htree_eof(file
))
550 /* Some one has messed with f_pos; reset the world */
551 if (info
->last_pos
!= ctx
->pos
) {
552 free_rb_tree_fname(&info
->root
);
553 info
->curr_node
= NULL
;
554 info
->extra_fname
= NULL
;
555 info
->curr_hash
= pos2maj_hash(file
, ctx
->pos
);
556 info
->curr_minor_hash
= pos2min_hash(file
, ctx
->pos
);
560 * If there are any leftover names on the hash collision
561 * chain, return them first.
563 if (info
->extra_fname
) {
564 if (call_filldir(file
, ctx
, info
->extra_fname
))
566 info
->extra_fname
= NULL
;
568 } else if (!info
->curr_node
)
569 info
->curr_node
= rb_first(&info
->root
);
573 * Fill the rbtree if we have no more entries,
574 * or the inode has changed since we last read in the
577 if ((!info
->curr_node
) ||
578 !inode_eq_iversion(inode
, file
->f_version
)) {
579 info
->curr_node
= NULL
;
580 free_rb_tree_fname(&info
->root
);
581 file
->f_version
= inode_query_iversion(inode
);
582 ret
= ext4_htree_fill_tree(file
, info
->curr_hash
,
583 info
->curr_minor_hash
,
588 ctx
->pos
= ext4_get_htree_eof(file
);
591 info
->curr_node
= rb_first(&info
->root
);
594 fname
= rb_entry(info
->curr_node
, struct fname
, rb_hash
);
595 info
->curr_hash
= fname
->hash
;
596 info
->curr_minor_hash
= fname
->minor_hash
;
597 if (call_filldir(file
, ctx
, fname
))
600 info
->curr_node
= rb_next(info
->curr_node
);
601 if (info
->curr_node
) {
602 fname
= rb_entry(info
->curr_node
, struct fname
,
604 info
->curr_hash
= fname
->hash
;
605 info
->curr_minor_hash
= fname
->minor_hash
;
607 if (info
->next_hash
== ~0) {
608 ctx
->pos
= ext4_get_htree_eof(file
);
611 info
->curr_hash
= info
->next_hash
;
612 info
->curr_minor_hash
= 0;
616 info
->last_pos
= ctx
->pos
;
620 static int ext4_dir_open(struct inode
* inode
, struct file
* filp
)
622 if (IS_ENCRYPTED(inode
))
623 return fscrypt_get_encryption_info(inode
) ? -EACCES
: 0;
627 static int ext4_release_dir(struct inode
*inode
, struct file
*filp
)
629 if (filp
->private_data
)
630 ext4_htree_free_dir_info(filp
->private_data
);
635 int ext4_check_all_de(struct inode
*dir
, struct buffer_head
*bh
, void *buf
,
638 struct ext4_dir_entry_2
*de
;
640 unsigned int offset
= 0;
643 de
= (struct ext4_dir_entry_2
*)buf
;
644 top
= buf
+ buf_size
;
645 while ((char *) de
< top
) {
646 if (ext4_check_dir_entry(dir
, NULL
, de
, bh
,
647 buf
, buf_size
, offset
))
648 return -EFSCORRUPTED
;
649 rlen
= ext4_rec_len_from_disk(de
->rec_len
, buf_size
);
650 de
= (struct ext4_dir_entry_2
*)((char *)de
+ rlen
);
653 if ((char *) de
> top
)
654 return -EFSCORRUPTED
;
659 const struct file_operations ext4_dir_operations
= {
660 .llseek
= ext4_dir_llseek
,
661 .read
= generic_read_dir
,
662 .iterate_shared
= ext4_readdir
,
663 .unlocked_ioctl
= ext4_ioctl
,
665 .compat_ioctl
= ext4_compat_ioctl
,
667 .fsync
= ext4_sync_file
,
668 .open
= ext4_dir_open
,
669 .release
= ext4_release_dir
,
672 #ifdef CONFIG_UNICODE
673 static int ext4_d_compare(const struct dentry
*dentry
, unsigned int len
,
674 const char *str
, const struct qstr
*name
)
676 struct qstr qstr
= {.name
= str
, .len
= len
};
677 const struct dentry
*parent
= READ_ONCE(dentry
->d_parent
);
678 const struct inode
*inode
= READ_ONCE(parent
->d_inode
);
680 if (!inode
|| !IS_CASEFOLDED(inode
) ||
681 !EXT4_SB(inode
->i_sb
)->s_encoding
) {
682 if (len
!= name
->len
)
684 return memcmp(str
, name
->name
, len
);
687 return ext4_ci_compare(inode
, name
, &qstr
, false);
690 static int ext4_d_hash(const struct dentry
*dentry
, struct qstr
*str
)
692 const struct ext4_sb_info
*sbi
= EXT4_SB(dentry
->d_sb
);
693 const struct unicode_map
*um
= sbi
->s_encoding
;
694 const struct inode
*inode
= READ_ONCE(dentry
->d_inode
);
698 if (!inode
|| !IS_CASEFOLDED(inode
) || !um
)
701 norm
= kmalloc(PATH_MAX
, GFP_ATOMIC
);
705 len
= utf8_casefold(um
, str
, norm
, PATH_MAX
);
707 if (ext4_has_strict_mode(sbi
))
711 str
->hash
= full_name_hash(dentry
, norm
, len
);
717 const struct dentry_operations ext4_dentry_ops
= {
718 .d_hash
= ext4_d_hash
,
719 .d_compare
= ext4_d_compare
,