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 page *page, loff_t pos, unsigned len)
86 struct address_space *mapping = page->mapping;
87 struct inode *dir = mapping->host;
89 inode_inc_iversion(dir);
90 block_write_end(NULL, mapping, pos, len, len, page, 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_page(struct page *page, int quiet, char *kaddr)
101 struct inode *dir = page->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 = PAGE_SIZE;
110 if ((dir->i_size >> PAGE_SHIFT) == page->index) {
111 limit = dir->i_size & ~PAGE_MASK;
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 SetPageChecked(page);
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=%lu, inode=%lu, rec_len=%d, name_len=%d",
164 dir->i_ino, error, (page->index<<PAGE_SHIFT)+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_page",
172 "entry in directory #%lu spans the page boundary"
173 "offset=%lu, inode=%lu",
174 dir->i_ino, (page->index<<PAGE_SHIFT)+offs,
175 (unsigned long) le32_to_cpu(p->inode));
183 * Calls to ext2_get_page()/ext2_put_page() must be nested according to the
184 * rules documented in kmap_local_page()/kunmap_local().
186 * NOTE: ext2_find_entry() and ext2_dotdot() act as a call to ext2_get_page()
187 * and should be treated as a call to ext2_get_page() for nesting purposes.
189 static void *ext2_get_page(struct inode *dir, unsigned long n,
190 int quiet, struct page **page)
192 struct address_space *mapping = dir->i_mapping;
193 struct folio *folio = read_mapping_folio(mapping, n, NULL);
197 return ERR_CAST(folio);
198 page_addr = kmap_local_folio(folio, n & (folio_nr_pages(folio) - 1));
199 if (unlikely(!folio_test_checked(folio))) {
200 if (!ext2_check_page(&folio->page, quiet, page_addr))
203 *page = &folio->page;
207 ext2_put_page(&folio->page, page_addr);
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, file->f_version);
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_page(inode, n, 0, &page);
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 file->f_version = 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 ext2_put_page(page, 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 ext2_put_page(page, de);
318 ctx->pos += ext2_rec_len_from_disk(de->rec_len);
320 ext2_put_page(page, 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 ext2_put_page() should be called on *res_page.
335 * NOTE: Calls to ext2_get_page()/ext2_put_page() must be nested according to
336 * the rules documented in kmap_local_page()/kunmap_local().
338 * ext2_find_entry() and ext2_dotdot() act as a call to ext2_get_page() and
339 * should be treated as a call to ext2_get_page() for nesting purposes.
341 struct ext2_dir_entry_2 *ext2_find_entry (struct inode *dir,
342 const struct qstr *child, struct page **res_page,
343 void **res_page_addr)
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 page *page = NULL;
351 struct ext2_inode_info *ei = EXT2_I(dir);
360 *res_page_addr = NULL;
362 start = ei->i_dir_start_lookup;
369 page_addr = ext2_get_page(dir, n, 0, &page);
370 if (IS_ERR(page_addr))
371 return ERR_CAST(page_addr);
374 de = (ext2_dirent *) kaddr;
375 kaddr += ext2_last_byte(dir, n) - reclen;
376 while ((char *) de <= kaddr) {
377 if (de->rec_len == 0) {
378 ext2_error(dir->i_sb, __func__,
379 "zero-length directory entry");
380 ext2_put_page(page, de);
383 if (ext2_match(namelen, name, de))
385 de = ext2_next_entry(de);
387 ext2_put_page(page, kaddr);
391 /* next page is past the blocks we've got */
392 if (unlikely(n > (dir->i_blocks >> (PAGE_SHIFT - 9)))) {
393 ext2_error(dir->i_sb, __func__,
394 "dir %lu size %lld exceeds block count %llu",
395 dir->i_ino, dir->i_size,
396 (unsigned long long)dir->i_blocks);
399 } while (n != start);
401 return ERR_PTR(-ENOENT);
405 *res_page_addr = page_addr;
406 ei->i_dir_start_lookup = n;
411 * Return the '..' directory entry and the page in which the entry was found
412 * (as a parameter - p).
414 * On Success ext2_put_page() should be called on *p.
416 * NOTE: Calls to ext2_get_page()/ext2_put_page() must be nested according to
417 * the rules documented in kmap_local_page()/kunmap_local().
419 * ext2_find_entry() and ext2_dotdot() act as a call to ext2_get_page() and
420 * should be treated as a call to ext2_get_page() for nesting purposes.
422 struct ext2_dir_entry_2 *ext2_dotdot(struct inode *dir, struct page **p,
425 void *page_addr = ext2_get_page(dir, 0, 0, p);
426 ext2_dirent *de = NULL;
428 if (!IS_ERR(page_addr)) {
429 de = ext2_next_entry((ext2_dirent *) page_addr);
435 int ext2_inode_by_name(struct inode *dir, const struct qstr *child, ino_t *ino)
437 struct ext2_dir_entry_2 *de;
441 de = ext2_find_entry(dir, child, &page, &page_addr);
445 *ino = le32_to_cpu(de->inode);
446 ext2_put_page(page, de);
450 static int ext2_prepare_chunk(struct page *page, loff_t pos, unsigned len)
452 return __block_write_begin(page, pos, len, ext2_get_block);
456 static int ext2_handle_dirsync(struct inode *dir)
460 err = filemap_write_and_wait(dir->i_mapping);
462 err = sync_inode_metadata(dir, 1);
466 int ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
467 struct page *page, struct inode *inode, bool update_times)
469 loff_t pos = page_offset(page) + offset_in_page(de);
470 unsigned len = ext2_rec_len_from_disk(de->rec_len);
474 err = ext2_prepare_chunk(page, pos, len);
479 de->inode = cpu_to_le32(inode->i_ino);
480 ext2_set_de_type(de, inode);
481 ext2_commit_chunk(page, pos, len);
483 dir->i_mtime = dir->i_ctime = current_time(dir);
484 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
485 mark_inode_dirty(dir);
486 return ext2_handle_dirsync(dir);
492 int ext2_add_link (struct dentry *dentry, struct inode *inode)
494 struct inode *dir = d_inode(dentry->d_parent);
495 const char *name = dentry->d_name.name;
496 int namelen = dentry->d_name.len;
497 unsigned chunk_size = ext2_chunk_size(dir);
498 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
499 unsigned short rec_len, name_len;
500 struct page *page = NULL;
502 unsigned long npages = dir_pages(dir);
508 * We take care of directory expansion in the same loop.
509 * This code plays outside i_size, so it locks the page
510 * to protect that region.
512 for (n = 0; n <= npages; n++) {
516 kaddr = ext2_get_page(dir, n, 0, &page);
518 return PTR_ERR(kaddr);
520 dir_end = kaddr + ext2_last_byte(dir, n);
521 de = (ext2_dirent *)kaddr;
522 kaddr += PAGE_SIZE - reclen;
523 while ((char *)de <= kaddr) {
524 if ((char *)de == dir_end) {
527 rec_len = chunk_size;
528 de->rec_len = ext2_rec_len_to_disk(chunk_size);
532 if (de->rec_len == 0) {
533 ext2_error(dir->i_sb, __func__,
534 "zero-length directory entry");
539 if (ext2_match (namelen, name, de))
541 name_len = EXT2_DIR_REC_LEN(de->name_len);
542 rec_len = ext2_rec_len_from_disk(de->rec_len);
543 if (!de->inode && rec_len >= reclen)
545 if (rec_len >= name_len + reclen)
547 de = (ext2_dirent *) ((char *) de + rec_len);
550 ext2_put_page(page, kaddr);
556 pos = page_offset(page) + offset_in_page(de);
557 err = ext2_prepare_chunk(page, pos, rec_len);
561 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
562 de1->rec_len = ext2_rec_len_to_disk(rec_len - name_len);
563 de->rec_len = ext2_rec_len_to_disk(name_len);
566 de->name_len = namelen;
567 memcpy(de->name, name, namelen);
568 de->inode = cpu_to_le32(inode->i_ino);
569 ext2_set_de_type (de, inode);
570 ext2_commit_chunk(page, pos, rec_len);
571 dir->i_mtime = dir->i_ctime = current_time(dir);
572 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
573 mark_inode_dirty(dir);
574 err = ext2_handle_dirsync(dir);
577 ext2_put_page(page, de);
585 * ext2_delete_entry deletes a directory entry by merging it with the
586 * previous entry. Page is up-to-date.
588 int ext2_delete_entry(struct ext2_dir_entry_2 *dir, struct page *page)
590 struct inode *inode = page->mapping->host;
591 char *kaddr = (char *)((unsigned long)dir & PAGE_MASK);
592 unsigned from = offset_in_page(dir) & ~(ext2_chunk_size(inode)-1);
593 unsigned to = offset_in_page(dir) +
594 ext2_rec_len_from_disk(dir->rec_len);
596 ext2_dirent *pde = NULL;
597 ext2_dirent *de = (ext2_dirent *)(kaddr + from);
600 while ((char*)de < (char*)dir) {
601 if (de->rec_len == 0) {
602 ext2_error(inode->i_sb, __func__,
603 "zero-length directory entry");
607 de = ext2_next_entry(de);
610 from = offset_in_page(pde);
611 pos = page_offset(page) + from;
613 err = ext2_prepare_chunk(page, pos, to - from);
619 pde->rec_len = ext2_rec_len_to_disk(to - from);
621 ext2_commit_chunk(page, pos, to - from);
622 inode->i_ctime = inode->i_mtime = current_time(inode);
623 EXT2_I(inode)->i_flags &= ~EXT2_BTREE_FL;
624 mark_inode_dirty(inode);
625 return ext2_handle_dirsync(inode);
629 * Set the first fragment of directory.
631 int ext2_make_empty(struct inode *inode, struct inode *parent)
633 struct page *page = grab_cache_page(inode->i_mapping, 0);
634 unsigned chunk_size = ext2_chunk_size(inode);
635 struct ext2_dir_entry_2 * de;
642 err = ext2_prepare_chunk(page, 0, chunk_size);
647 kaddr = kmap_local_page(page);
648 memset(kaddr, 0, chunk_size);
649 de = (struct ext2_dir_entry_2 *)kaddr;
651 de->rec_len = ext2_rec_len_to_disk(EXT2_DIR_REC_LEN(1));
652 memcpy (de->name, ".\0\0", 4);
653 de->inode = cpu_to_le32(inode->i_ino);
654 ext2_set_de_type (de, inode);
656 de = (struct ext2_dir_entry_2 *)(kaddr + EXT2_DIR_REC_LEN(1));
658 de->rec_len = ext2_rec_len_to_disk(chunk_size - EXT2_DIR_REC_LEN(1));
659 de->inode = cpu_to_le32(parent->i_ino);
660 memcpy (de->name, "..\0", 4);
661 ext2_set_de_type (de, inode);
663 ext2_commit_chunk(page, 0, chunk_size);
664 err = ext2_handle_dirsync(inode);
671 * routine to check that the specified directory is empty (for rmdir)
673 int ext2_empty_dir (struct inode * inode)
677 unsigned long i, npages = dir_pages(inode);
679 for (i = 0; i < npages; i++) {
682 kaddr = ext2_get_page(inode, i, 0, &page);
686 de = (ext2_dirent *)kaddr;
687 kaddr += ext2_last_byte(inode, i) - EXT2_DIR_REC_LEN(1);
689 while ((char *)de <= kaddr) {
690 if (de->rec_len == 0) {
691 ext2_error(inode->i_sb, __func__,
692 "zero-length directory entry");
693 printk("kaddr=%p, de=%p\n", kaddr, de);
696 if (de->inode != 0) {
697 /* check for . and .. */
698 if (de->name[0] != '.')
700 if (de->name_len > 2)
702 if (de->name_len < 2) {
704 cpu_to_le32(inode->i_ino))
706 } else if (de->name[1] != '.')
709 de = ext2_next_entry(de);
711 ext2_put_page(page, kaddr);
716 ext2_put_page(page, kaddr);
720 const struct file_operations ext2_dir_operations = {
721 .llseek = generic_file_llseek,
722 .read = generic_read_dir,
723 .iterate_shared = ext2_readdir,
724 .unlocked_ioctl = ext2_ioctl,
726 .compat_ioctl = ext2_compat_ioctl,