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 * ext2 directory handling functions
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
21 * All code that works with directory layout had been switched to pagecache
26 #include <linux/buffer_head.h>
27 #include <linux/pagemap.h>
28 #include <linux/swap.h>
29 #include <linux/iversion.h>
31 typedef struct ext2_dir_entry_2 ext2_dirent
;
34 * Tests against MAX_REC_LEN etc were put in place for 64k block
35 * sizes; if that is not possible on this arch, we can skip
36 * those tests and speed things up.
38 static inline unsigned ext2_rec_len_from_disk(__le16 dlen
)
40 unsigned len
= le16_to_cpu(dlen
);
42 #if (PAGE_SIZE >= 65536)
43 if (len
== EXT2_MAX_REC_LEN
)
49 static inline __le16
ext2_rec_len_to_disk(unsigned len
)
51 #if (PAGE_SIZE >= 65536)
53 return cpu_to_le16(EXT2_MAX_REC_LEN
);
55 BUG_ON(len
> (1 << 16));
57 return cpu_to_le16(len
);
61 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
62 * more robust, but we have what we have
64 static inline unsigned ext2_chunk_size(struct inode
*inode
)
66 return inode
->i_sb
->s_blocksize
;
70 * Return the offset into page `page_nr' of the last valid
71 * byte in that page, plus one.
74 ext2_last_byte(struct inode
*inode
, unsigned long page_nr
)
76 unsigned last_byte
= inode
->i_size
;
78 last_byte
-= page_nr
<< PAGE_SHIFT
;
79 if (last_byte
> PAGE_SIZE
)
80 last_byte
= PAGE_SIZE
;
84 static void ext2_commit_chunk(struct folio
*folio
, loff_t pos
, unsigned len
)
86 struct address_space
*mapping
= folio
->mapping
;
87 struct inode
*dir
= mapping
->host
;
89 inode_inc_iversion(dir
);
90 block_write_end(NULL
, mapping
, pos
, len
, len
, folio
, NULL
);
92 if (pos
+len
> dir
->i_size
) {
93 i_size_write(dir
, pos
+len
);
94 mark_inode_dirty(dir
);
99 static bool ext2_check_folio(struct folio
*folio
, int quiet
, char *kaddr
)
101 struct inode
*dir
= folio
->mapping
->host
;
102 struct super_block
*sb
= dir
->i_sb
;
103 unsigned chunk_size
= ext2_chunk_size(dir
);
104 u32 max_inumber
= le32_to_cpu(EXT2_SB(sb
)->s_es
->s_inodes_count
);
105 unsigned offs
, rec_len
;
106 unsigned limit
= folio_size(folio
);
110 if (dir
->i_size
< folio_pos(folio
) + limit
) {
111 limit
= offset_in_folio(folio
, dir
->i_size
);
112 if (limit
& (chunk_size
- 1))
117 for (offs
= 0; offs
<= limit
- EXT2_DIR_REC_LEN(1); offs
+= rec_len
) {
118 p
= (ext2_dirent
*)(kaddr
+ offs
);
119 rec_len
= ext2_rec_len_from_disk(p
->rec_len
);
121 if (unlikely(rec_len
< EXT2_DIR_REC_LEN(1)))
123 if (unlikely(rec_len
& 3))
125 if (unlikely(rec_len
< EXT2_DIR_REC_LEN(p
->name_len
)))
127 if (unlikely(((offs
+ rec_len
- 1) ^ offs
) & ~(chunk_size
-1)))
129 if (unlikely(le32_to_cpu(p
->inode
) > max_inumber
))
135 folio_set_checked(folio
);
138 /* Too bad, we had an error */
142 ext2_error(sb
, __func__
,
143 "size of directory #%lu is not a multiple "
144 "of chunk size", dir
->i_ino
);
147 error
= "rec_len is smaller than minimal";
150 error
= "unaligned directory entry";
153 error
= "rec_len is too small for name_len";
156 error
= "directory entry across blocks";
159 error
= "inode out of bounds";
162 ext2_error(sb
, __func__
, "bad entry in directory #%lu: : %s - "
163 "offset=%llu, inode=%lu, rec_len=%d, name_len=%d",
164 dir
->i_ino
, error
, folio_pos(folio
) + offs
,
165 (unsigned long) le32_to_cpu(p
->inode
),
166 rec_len
, p
->name_len
);
170 p
= (ext2_dirent
*)(kaddr
+ offs
);
171 ext2_error(sb
, "ext2_check_folio",
172 "entry in directory #%lu spans the page boundary"
173 "offset=%llu, inode=%lu",
174 dir
->i_ino
, folio_pos(folio
) + offs
,
175 (unsigned long) le32_to_cpu(p
->inode
));
182 * Calls to ext2_get_folio()/folio_release_kmap() must be nested according
183 * to the rules documented in kmap_local_folio()/kunmap_local().
185 * NOTE: ext2_find_entry() and ext2_dotdot() act as a call
186 * to folio_release_kmap() and should be treated as a call to
187 * folio_release_kmap() for nesting purposes.
189 static void *ext2_get_folio(struct inode
*dir
, unsigned long n
,
190 int quiet
, struct folio
**foliop
)
192 struct address_space
*mapping
= dir
->i_mapping
;
193 struct folio
*folio
= read_mapping_folio(mapping
, n
, NULL
);
197 return ERR_CAST(folio
);
198 kaddr
= kmap_local_folio(folio
, 0);
199 if (unlikely(!folio_test_checked(folio
))) {
200 if (!ext2_check_folio(folio
, quiet
, kaddr
))
207 folio_release_kmap(folio
, kaddr
);
208 return ERR_PTR(-EIO
);
212 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
214 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
216 static inline int ext2_match (int len
, const char * const name
,
217 struct ext2_dir_entry_2
* de
)
219 if (len
!= de
->name_len
)
223 return !memcmp(name
, de
->name
, len
);
227 * p is at least 6 bytes before the end of page
229 static inline ext2_dirent
*ext2_next_entry(ext2_dirent
*p
)
231 return (ext2_dirent
*)((char *)p
+
232 ext2_rec_len_from_disk(p
->rec_len
));
235 static inline unsigned
236 ext2_validate_entry(char *base
, unsigned offset
, unsigned mask
)
238 ext2_dirent
*de
= (ext2_dirent
*)(base
+ offset
);
239 ext2_dirent
*p
= (ext2_dirent
*)(base
+ (offset
&mask
));
240 while ((char*)p
< (char*)de
) {
243 p
= ext2_next_entry(p
);
245 return offset_in_page(p
);
248 static inline void ext2_set_de_type(ext2_dirent
*de
, struct inode
*inode
)
250 if (EXT2_HAS_INCOMPAT_FEATURE(inode
->i_sb
, EXT2_FEATURE_INCOMPAT_FILETYPE
))
251 de
->file_type
= fs_umode_to_ftype(inode
->i_mode
);
257 ext2_readdir(struct file
*file
, struct dir_context
*ctx
)
259 loff_t pos
= ctx
->pos
;
260 struct inode
*inode
= file_inode(file
);
261 struct super_block
*sb
= inode
->i_sb
;
262 unsigned int offset
= pos
& ~PAGE_MASK
;
263 unsigned long n
= pos
>> PAGE_SHIFT
;
264 unsigned long npages
= dir_pages(inode
);
265 unsigned chunk_mask
= ~(ext2_chunk_size(inode
)-1);
266 bool need_revalidate
= !inode_eq_iversion(inode
, *(u64
*)file
->private_data
);
269 if (pos
> inode
->i_size
- EXT2_DIR_REC_LEN(1))
273 EXT2_HAS_INCOMPAT_FEATURE(sb
, EXT2_FEATURE_INCOMPAT_FILETYPE
);
275 for ( ; n
< npages
; n
++, offset
= 0) {
278 char *kaddr
= ext2_get_folio(inode
, n
, 0, &folio
);
282 ext2_error(sb
, __func__
,
285 ctx
->pos
+= PAGE_SIZE
- offset
;
286 return PTR_ERR(kaddr
);
288 if (unlikely(need_revalidate
)) {
290 offset
= ext2_validate_entry(kaddr
, offset
, chunk_mask
);
291 ctx
->pos
= (n
<<PAGE_SHIFT
) + offset
;
293 *(u64
*)file
->private_data
= inode_query_iversion(inode
);
294 need_revalidate
= false;
296 de
= (ext2_dirent
*)(kaddr
+offset
);
297 limit
= kaddr
+ ext2_last_byte(inode
, n
) - EXT2_DIR_REC_LEN(1);
298 for ( ;(char*)de
<= limit
; de
= ext2_next_entry(de
)) {
299 if (de
->rec_len
== 0) {
300 ext2_error(sb
, __func__
,
301 "zero-length directory entry");
302 folio_release_kmap(folio
, de
);
306 unsigned char d_type
= DT_UNKNOWN
;
309 d_type
= fs_ftype_to_dtype(de
->file_type
);
311 if (!dir_emit(ctx
, de
->name
, de
->name_len
,
312 le32_to_cpu(de
->inode
),
314 folio_release_kmap(folio
, de
);
318 ctx
->pos
+= ext2_rec_len_from_disk(de
->rec_len
);
320 folio_release_kmap(folio
, kaddr
);
328 * finds an entry in the specified directory with the wanted name. It
329 * returns the page in which the entry was found (as a parameter - res_page),
330 * and the entry itself. Page is returned mapped and unlocked.
331 * Entry is guaranteed to be valid.
333 * On Success folio_release_kmap() should be called on *foliop.
335 * NOTE: Calls to ext2_get_folio()/folio_release_kmap() must be nested
336 * according to the rules documented in kmap_local_folio()/kunmap_local().
338 * ext2_find_entry() and ext2_dotdot() act as a call to ext2_get_folio()
339 * and should be treated as a call to ext2_get_folio() for nesting
342 struct ext2_dir_entry_2
*ext2_find_entry (struct inode
*dir
,
343 const struct qstr
*child
, struct folio
**foliop
)
345 const char *name
= child
->name
;
346 int namelen
= child
->len
;
347 unsigned reclen
= EXT2_DIR_REC_LEN(namelen
);
348 unsigned long start
, n
;
349 unsigned long npages
= dir_pages(dir
);
350 struct ext2_inode_info
*ei
= EXT2_I(dir
);
356 start
= ei
->i_dir_start_lookup
;
361 char *kaddr
= ext2_get_folio(dir
, n
, 0, foliop
);
363 return ERR_CAST(kaddr
);
365 de
= (ext2_dirent
*) kaddr
;
366 kaddr
+= ext2_last_byte(dir
, n
) - reclen
;
367 while ((char *) de
<= kaddr
) {
368 if (de
->rec_len
== 0) {
369 ext2_error(dir
->i_sb
, __func__
,
370 "zero-length directory entry");
371 folio_release_kmap(*foliop
, de
);
374 if (ext2_match(namelen
, name
, de
))
376 de
= ext2_next_entry(de
);
378 folio_release_kmap(*foliop
, kaddr
);
382 /* next folio is past the blocks we've got */
383 if (unlikely(n
> (dir
->i_blocks
>> (PAGE_SHIFT
- 9)))) {
384 ext2_error(dir
->i_sb
, __func__
,
385 "dir %lu size %lld exceeds block count %llu",
386 dir
->i_ino
, dir
->i_size
,
387 (unsigned long long)dir
->i_blocks
);
390 } while (n
!= start
);
392 return ERR_PTR(-ENOENT
);
395 ei
->i_dir_start_lookup
= n
;
400 * Return the '..' directory entry and the page in which the entry was found
401 * (as a parameter - p).
403 * On Success folio_release_kmap() should be called on *foliop.
405 * NOTE: Calls to ext2_get_folio()/folio_release_kmap() must be nested
406 * according to the rules documented in kmap_local_folio()/kunmap_local().
408 * ext2_find_entry() and ext2_dotdot() act as a call to ext2_get_folio()
409 * and should be treated as a call to ext2_get_folio() for nesting
412 struct ext2_dir_entry_2
*ext2_dotdot(struct inode
*dir
, struct folio
**foliop
)
414 ext2_dirent
*de
= ext2_get_folio(dir
, 0, 0, foliop
);
417 return ext2_next_entry(de
);
421 int ext2_inode_by_name(struct inode
*dir
, const struct qstr
*child
, ino_t
*ino
)
423 struct ext2_dir_entry_2
*de
;
426 de
= ext2_find_entry(dir
, child
, &folio
);
430 *ino
= le32_to_cpu(de
->inode
);
431 folio_release_kmap(folio
, de
);
435 static int ext2_prepare_chunk(struct folio
*folio
, loff_t pos
, unsigned len
)
437 return __block_write_begin(folio
, pos
, len
, ext2_get_block
);
440 static int ext2_handle_dirsync(struct inode
*dir
)
444 err
= filemap_write_and_wait(dir
->i_mapping
);
446 err
= sync_inode_metadata(dir
, 1);
450 int ext2_set_link(struct inode
*dir
, struct ext2_dir_entry_2
*de
,
451 struct folio
*folio
, struct inode
*inode
, bool update_times
)
453 loff_t pos
= folio_pos(folio
) + offset_in_folio(folio
, de
);
454 unsigned len
= ext2_rec_len_from_disk(de
->rec_len
);
458 err
= ext2_prepare_chunk(folio
, pos
, len
);
463 de
->inode
= cpu_to_le32(inode
->i_ino
);
464 ext2_set_de_type(de
, inode
);
465 ext2_commit_chunk(folio
, pos
, len
);
467 inode_set_mtime_to_ts(dir
, inode_set_ctime_current(dir
));
468 EXT2_I(dir
)->i_flags
&= ~EXT2_BTREE_FL
;
469 mark_inode_dirty(dir
);
470 return ext2_handle_dirsync(dir
);
476 int ext2_add_link (struct dentry
*dentry
, struct inode
*inode
)
478 struct inode
*dir
= d_inode(dentry
->d_parent
);
479 const char *name
= dentry
->d_name
.name
;
480 int namelen
= dentry
->d_name
.len
;
481 unsigned chunk_size
= ext2_chunk_size(dir
);
482 unsigned reclen
= EXT2_DIR_REC_LEN(namelen
);
483 unsigned short rec_len
, name_len
;
484 struct folio
*folio
= NULL
;
486 unsigned long npages
= dir_pages(dir
);
492 * We take care of directory expansion in the same loop.
493 * This code plays outside i_size, so it locks the folio
494 * to protect that region.
496 for (n
= 0; n
<= npages
; n
++) {
497 char *kaddr
= ext2_get_folio(dir
, n
, 0, &folio
);
501 return PTR_ERR(kaddr
);
503 dir_end
= kaddr
+ ext2_last_byte(dir
, n
);
504 de
= (ext2_dirent
*)kaddr
;
505 kaddr
+= folio_size(folio
) - reclen
;
506 while ((char *)de
<= kaddr
) {
507 if ((char *)de
== dir_end
) {
510 rec_len
= chunk_size
;
511 de
->rec_len
= ext2_rec_len_to_disk(chunk_size
);
515 if (de
->rec_len
== 0) {
516 ext2_error(dir
->i_sb
, __func__
,
517 "zero-length directory entry");
522 if (ext2_match (namelen
, name
, de
))
524 name_len
= EXT2_DIR_REC_LEN(de
->name_len
);
525 rec_len
= ext2_rec_len_from_disk(de
->rec_len
);
526 if (!de
->inode
&& rec_len
>= reclen
)
528 if (rec_len
>= name_len
+ reclen
)
530 de
= (ext2_dirent
*) ((char *) de
+ rec_len
);
533 folio_release_kmap(folio
, kaddr
);
539 pos
= folio_pos(folio
) + offset_in_folio(folio
, de
);
540 err
= ext2_prepare_chunk(folio
, pos
, rec_len
);
544 ext2_dirent
*de1
= (ext2_dirent
*) ((char *) de
+ name_len
);
545 de1
->rec_len
= ext2_rec_len_to_disk(rec_len
- name_len
);
546 de
->rec_len
= ext2_rec_len_to_disk(name_len
);
549 de
->name_len
= namelen
;
550 memcpy(de
->name
, name
, namelen
);
551 de
->inode
= cpu_to_le32(inode
->i_ino
);
552 ext2_set_de_type (de
, inode
);
553 ext2_commit_chunk(folio
, pos
, rec_len
);
554 inode_set_mtime_to_ts(dir
, inode_set_ctime_current(dir
));
555 EXT2_I(dir
)->i_flags
&= ~EXT2_BTREE_FL
;
556 mark_inode_dirty(dir
);
557 err
= ext2_handle_dirsync(dir
);
560 folio_release_kmap(folio
, de
);
568 * ext2_delete_entry deletes a directory entry by merging it with the
569 * previous entry. Page is up-to-date.
571 int ext2_delete_entry(struct ext2_dir_entry_2
*dir
, struct folio
*folio
)
573 struct inode
*inode
= folio
->mapping
->host
;
577 ext2_dirent
*de
, *pde
= NULL
;
580 from
= offset_in_folio(folio
, dir
);
581 to
= from
+ ext2_rec_len_from_disk(dir
->rec_len
);
582 kaddr
= (char *)dir
- from
;
583 from
&= ~(ext2_chunk_size(inode
)-1);
584 de
= (ext2_dirent
*)(kaddr
+ from
);
586 while ((char*)de
< (char*)dir
) {
587 if (de
->rec_len
== 0) {
588 ext2_error(inode
->i_sb
, __func__
,
589 "zero-length directory entry");
593 de
= ext2_next_entry(de
);
596 from
= offset_in_folio(folio
, pde
);
597 pos
= folio_pos(folio
) + from
;
599 err
= ext2_prepare_chunk(folio
, pos
, to
- from
);
605 pde
->rec_len
= ext2_rec_len_to_disk(to
- from
);
607 ext2_commit_chunk(folio
, pos
, to
- from
);
608 inode_set_mtime_to_ts(inode
, inode_set_ctime_current(inode
));
609 EXT2_I(inode
)->i_flags
&= ~EXT2_BTREE_FL
;
610 mark_inode_dirty(inode
);
611 return ext2_handle_dirsync(inode
);
615 * Set the first fragment of directory.
617 int ext2_make_empty(struct inode
*inode
, struct inode
*parent
)
619 struct folio
*folio
= filemap_grab_folio(inode
->i_mapping
, 0);
620 unsigned chunk_size
= ext2_chunk_size(inode
);
621 struct ext2_dir_entry_2
* de
;
626 return PTR_ERR(folio
);
628 err
= ext2_prepare_chunk(folio
, 0, chunk_size
);
633 kaddr
= kmap_local_folio(folio
, 0);
634 memset(kaddr
, 0, chunk_size
);
635 de
= (struct ext2_dir_entry_2
*)kaddr
;
637 de
->rec_len
= ext2_rec_len_to_disk(EXT2_DIR_REC_LEN(1));
638 memcpy (de
->name
, ".\0\0", 4);
639 de
->inode
= cpu_to_le32(inode
->i_ino
);
640 ext2_set_de_type (de
, inode
);
642 de
= (struct ext2_dir_entry_2
*)(kaddr
+ EXT2_DIR_REC_LEN(1));
644 de
->rec_len
= ext2_rec_len_to_disk(chunk_size
- EXT2_DIR_REC_LEN(1));
645 de
->inode
= cpu_to_le32(parent
->i_ino
);
646 memcpy (de
->name
, "..\0", 4);
647 ext2_set_de_type (de
, inode
);
649 ext2_commit_chunk(folio
, 0, chunk_size
);
650 err
= ext2_handle_dirsync(inode
);
657 * routine to check that the specified directory is empty (for rmdir)
659 int ext2_empty_dir(struct inode
*inode
)
663 unsigned long i
, npages
= dir_pages(inode
);
665 for (i
= 0; i
< npages
; i
++) {
668 kaddr
= ext2_get_folio(inode
, i
, 0, &folio
);
672 de
= (ext2_dirent
*)kaddr
;
673 kaddr
+= ext2_last_byte(inode
, i
) - EXT2_DIR_REC_LEN(1);
675 while ((char *)de
<= kaddr
) {
676 if (de
->rec_len
== 0) {
677 ext2_error(inode
->i_sb
, __func__
,
678 "zero-length directory entry");
679 printk("kaddr=%p, de=%p\n", kaddr
, de
);
682 if (de
->inode
!= 0) {
683 /* check for . and .. */
684 if (de
->name
[0] != '.')
686 if (de
->name_len
> 2)
688 if (de
->name_len
< 2) {
690 cpu_to_le32(inode
->i_ino
))
692 } else if (de
->name
[1] != '.')
695 de
= ext2_next_entry(de
);
697 folio_release_kmap(folio
, kaddr
);
702 folio_release_kmap(folio
, kaddr
);
706 static int ext2_dir_open(struct inode
*inode
, struct file
*file
)
708 file
->private_data
= kzalloc(sizeof(u64
), GFP_KERNEL
);
709 if (!file
->private_data
)
714 static int ext2_dir_release(struct inode
*inode
, struct file
*file
)
716 kfree(file
->private_data
);
720 static loff_t
ext2_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
722 return generic_llseek_cookie(file
, offset
, whence
,
723 (u64
*)file
->private_data
);
726 const struct file_operations ext2_dir_operations
= {
727 .open
= ext2_dir_open
,
728 .release
= ext2_dir_release
,
729 .llseek
= ext2_dir_llseek
,
730 .read
= generic_read_dir
,
731 .iterate_shared
= ext2_readdir
,
732 .unlocked_ioctl
= ext2_ioctl
,
734 .compat_ioctl
= ext2_compat_ioctl
,