Linux 2.6.28-rc5
[cris-mirror.git] / fs / ext2 / dir.c
blob9a0fc400f91cf606bdd9bfc7c9b14e136243a1c7
1 /*
2 * linux/fs/ext2/dir.c
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)
9 * from
11 * linux/fs/minix/dir.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * ext2 directory handling functions
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
20 * All code that works with directory layout had been switched to pagecache
21 * and moved here. AV
24 #include "ext2.h"
25 #include <linux/buffer_head.h>
26 #include <linux/pagemap.h>
27 #include <linux/swap.h>
29 typedef struct ext2_dir_entry_2 ext2_dirent;
31 static inline unsigned ext2_rec_len_from_disk(__le16 dlen)
33 unsigned len = le16_to_cpu(dlen);
35 if (len == EXT2_MAX_REC_LEN)
36 return 1 << 16;
37 return len;
40 static inline __le16 ext2_rec_len_to_disk(unsigned len)
42 if (len == (1 << 16))
43 return cpu_to_le16(EXT2_MAX_REC_LEN);
44 else
45 BUG_ON(len > (1 << 16));
46 return cpu_to_le16(len);
50 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
51 * more robust, but we have what we have
53 static inline unsigned ext2_chunk_size(struct inode *inode)
55 return inode->i_sb->s_blocksize;
58 static inline void ext2_put_page(struct page *page)
60 kunmap(page);
61 page_cache_release(page);
64 static inline unsigned long dir_pages(struct inode *inode)
66 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
70 * Return the offset into page `page_nr' of the last valid
71 * byte in that page, plus one.
73 static unsigned
74 ext2_last_byte(struct inode *inode, unsigned long page_nr)
76 unsigned last_byte = inode->i_size;
78 last_byte -= page_nr << PAGE_CACHE_SHIFT;
79 if (last_byte > PAGE_CACHE_SIZE)
80 last_byte = PAGE_CACHE_SIZE;
81 return last_byte;
84 static int ext2_commit_chunk(struct page *page, loff_t pos, unsigned len)
86 struct address_space *mapping = page->mapping;
87 struct inode *dir = mapping->host;
88 int err = 0;
90 dir->i_version++;
91 block_write_end(NULL, mapping, pos, len, len, page, NULL);
93 if (pos+len > dir->i_size) {
94 i_size_write(dir, pos+len);
95 mark_inode_dirty(dir);
98 if (IS_DIRSYNC(dir))
99 err = write_one_page(page, 1);
100 else
101 unlock_page(page);
103 return err;
106 static void ext2_check_page(struct page *page, int quiet)
108 struct inode *dir = page->mapping->host;
109 struct super_block *sb = dir->i_sb;
110 unsigned chunk_size = ext2_chunk_size(dir);
111 char *kaddr = page_address(page);
112 u32 max_inumber = le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count);
113 unsigned offs, rec_len;
114 unsigned limit = PAGE_CACHE_SIZE;
115 ext2_dirent *p;
116 char *error;
118 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
119 limit = dir->i_size & ~PAGE_CACHE_MASK;
120 if (limit & (chunk_size - 1))
121 goto Ebadsize;
122 if (!limit)
123 goto out;
125 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
126 p = (ext2_dirent *)(kaddr + offs);
127 rec_len = ext2_rec_len_from_disk(p->rec_len);
129 if (rec_len < EXT2_DIR_REC_LEN(1))
130 goto Eshort;
131 if (rec_len & 3)
132 goto Ealign;
133 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
134 goto Enamelen;
135 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
136 goto Espan;
137 if (le32_to_cpu(p->inode) > max_inumber)
138 goto Einumber;
140 if (offs != limit)
141 goto Eend;
142 out:
143 SetPageChecked(page);
144 return;
146 /* Too bad, we had an error */
148 Ebadsize:
149 if (!quiet)
150 ext2_error(sb, __func__,
151 "size of directory #%lu is not a multiple "
152 "of chunk size", dir->i_ino);
153 goto fail;
154 Eshort:
155 error = "rec_len is smaller than minimal";
156 goto bad_entry;
157 Ealign:
158 error = "unaligned directory entry";
159 goto bad_entry;
160 Enamelen:
161 error = "rec_len is too small for name_len";
162 goto bad_entry;
163 Espan:
164 error = "directory entry across blocks";
165 goto bad_entry;
166 Einumber:
167 error = "inode out of bounds";
168 bad_entry:
169 if (!quiet)
170 ext2_error(sb, __func__, "bad entry in directory #%lu: : %s - "
171 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
172 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
173 (unsigned long) le32_to_cpu(p->inode),
174 rec_len, p->name_len);
175 goto fail;
176 Eend:
177 if (!quiet) {
178 p = (ext2_dirent *)(kaddr + offs);
179 ext2_error(sb, "ext2_check_page",
180 "entry in directory #%lu spans the page boundary"
181 "offset=%lu, inode=%lu",
182 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
183 (unsigned long) le32_to_cpu(p->inode));
185 fail:
186 SetPageChecked(page);
187 SetPageError(page);
190 static struct page * ext2_get_page(struct inode *dir, unsigned long n,
191 int quiet)
193 struct address_space *mapping = dir->i_mapping;
194 struct page *page = read_mapping_page(mapping, n, NULL);
195 if (!IS_ERR(page)) {
196 kmap(page);
197 if (!PageChecked(page))
198 ext2_check_page(page, quiet);
199 if (PageError(page))
200 goto fail;
202 return page;
204 fail:
205 ext2_put_page(page);
206 return ERR_PTR(-EIO);
210 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
212 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
214 static inline int ext2_match (int len, const char * const name,
215 struct ext2_dir_entry_2 * de)
217 if (len != de->name_len)
218 return 0;
219 if (!de->inode)
220 return 0;
221 return !memcmp(name, de->name, len);
225 * p is at least 6 bytes before the end of page
227 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
229 return (ext2_dirent *)((char *)p +
230 ext2_rec_len_from_disk(p->rec_len));
233 static inline unsigned
234 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
236 ext2_dirent *de = (ext2_dirent*)(base + offset);
237 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
238 while ((char*)p < (char*)de) {
239 if (p->rec_len == 0)
240 break;
241 p = ext2_next_entry(p);
243 return (char *)p - base;
246 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
247 [EXT2_FT_UNKNOWN] = DT_UNKNOWN,
248 [EXT2_FT_REG_FILE] = DT_REG,
249 [EXT2_FT_DIR] = DT_DIR,
250 [EXT2_FT_CHRDEV] = DT_CHR,
251 [EXT2_FT_BLKDEV] = DT_BLK,
252 [EXT2_FT_FIFO] = DT_FIFO,
253 [EXT2_FT_SOCK] = DT_SOCK,
254 [EXT2_FT_SYMLINK] = DT_LNK,
257 #define S_SHIFT 12
258 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
259 [S_IFREG >> S_SHIFT] = EXT2_FT_REG_FILE,
260 [S_IFDIR >> S_SHIFT] = EXT2_FT_DIR,
261 [S_IFCHR >> S_SHIFT] = EXT2_FT_CHRDEV,
262 [S_IFBLK >> S_SHIFT] = EXT2_FT_BLKDEV,
263 [S_IFIFO >> S_SHIFT] = EXT2_FT_FIFO,
264 [S_IFSOCK >> S_SHIFT] = EXT2_FT_SOCK,
265 [S_IFLNK >> S_SHIFT] = EXT2_FT_SYMLINK,
268 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
270 mode_t mode = inode->i_mode;
271 if (EXT2_HAS_INCOMPAT_FEATURE(inode->i_sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
272 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
273 else
274 de->file_type = 0;
277 static int
278 ext2_readdir (struct file * filp, void * dirent, filldir_t filldir)
280 loff_t pos = filp->f_pos;
281 struct inode *inode = filp->f_path.dentry->d_inode;
282 struct super_block *sb = inode->i_sb;
283 unsigned int offset = pos & ~PAGE_CACHE_MASK;
284 unsigned long n = pos >> PAGE_CACHE_SHIFT;
285 unsigned long npages = dir_pages(inode);
286 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
287 unsigned char *types = NULL;
288 int need_revalidate = filp->f_version != inode->i_version;
290 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
291 return 0;
293 if (EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
294 types = ext2_filetype_table;
296 for ( ; n < npages; n++, offset = 0) {
297 char *kaddr, *limit;
298 ext2_dirent *de;
299 struct page *page = ext2_get_page(inode, n, 0);
301 if (IS_ERR(page)) {
302 ext2_error(sb, __func__,
303 "bad page in #%lu",
304 inode->i_ino);
305 filp->f_pos += PAGE_CACHE_SIZE - offset;
306 return PTR_ERR(page);
308 kaddr = page_address(page);
309 if (unlikely(need_revalidate)) {
310 if (offset) {
311 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
312 filp->f_pos = (n<<PAGE_CACHE_SHIFT) + offset;
314 filp->f_version = inode->i_version;
315 need_revalidate = 0;
317 de = (ext2_dirent *)(kaddr+offset);
318 limit = kaddr + ext2_last_byte(inode, n) - EXT2_DIR_REC_LEN(1);
319 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
320 if (de->rec_len == 0) {
321 ext2_error(sb, __func__,
322 "zero-length directory entry");
323 ext2_put_page(page);
324 return -EIO;
326 if (de->inode) {
327 int over;
328 unsigned char d_type = DT_UNKNOWN;
330 if (types && de->file_type < EXT2_FT_MAX)
331 d_type = types[de->file_type];
333 offset = (char *)de - kaddr;
334 over = filldir(dirent, de->name, de->name_len,
335 (n<<PAGE_CACHE_SHIFT) | offset,
336 le32_to_cpu(de->inode), d_type);
337 if (over) {
338 ext2_put_page(page);
339 return 0;
342 filp->f_pos += ext2_rec_len_from_disk(de->rec_len);
344 ext2_put_page(page);
346 return 0;
350 * ext2_find_entry()
352 * finds an entry in the specified directory with the wanted name. It
353 * returns the page in which the entry was found, and the entry itself
354 * (as a parameter - res_dir). Page is returned mapped and unlocked.
355 * Entry is guaranteed to be valid.
357 struct ext2_dir_entry_2 *ext2_find_entry (struct inode * dir,
358 struct qstr *child, struct page ** res_page)
360 const char *name = child->name;
361 int namelen = child->len;
362 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
363 unsigned long start, n;
364 unsigned long npages = dir_pages(dir);
365 struct page *page = NULL;
366 struct ext2_inode_info *ei = EXT2_I(dir);
367 ext2_dirent * de;
368 int dir_has_error = 0;
370 if (npages == 0)
371 goto out;
373 /* OFFSET_CACHE */
374 *res_page = NULL;
376 start = ei->i_dir_start_lookup;
377 if (start >= npages)
378 start = 0;
379 n = start;
380 do {
381 char *kaddr;
382 page = ext2_get_page(dir, n, dir_has_error);
383 if (!IS_ERR(page)) {
384 kaddr = page_address(page);
385 de = (ext2_dirent *) kaddr;
386 kaddr += ext2_last_byte(dir, n) - reclen;
387 while ((char *) de <= kaddr) {
388 if (de->rec_len == 0) {
389 ext2_error(dir->i_sb, __func__,
390 "zero-length directory entry");
391 ext2_put_page(page);
392 goto out;
394 if (ext2_match (namelen, name, de))
395 goto found;
396 de = ext2_next_entry(de);
398 ext2_put_page(page);
399 } else
400 dir_has_error = 1;
402 if (++n >= npages)
403 n = 0;
404 /* next page is past the blocks we've got */
405 if (unlikely(n > (dir->i_blocks >> (PAGE_CACHE_SHIFT - 9)))) {
406 ext2_error(dir->i_sb, __func__,
407 "dir %lu size %lld exceeds block count %llu",
408 dir->i_ino, dir->i_size,
409 (unsigned long long)dir->i_blocks);
410 goto out;
412 } while (n != start);
413 out:
414 return NULL;
416 found:
417 *res_page = page;
418 ei->i_dir_start_lookup = n;
419 return de;
422 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
424 struct page *page = ext2_get_page(dir, 0, 0);
425 ext2_dirent *de = NULL;
427 if (!IS_ERR(page)) {
428 de = ext2_next_entry((ext2_dirent *) page_address(page));
429 *p = page;
431 return de;
434 ino_t ext2_inode_by_name(struct inode *dir, struct qstr *child)
436 ino_t res = 0;
437 struct ext2_dir_entry_2 *de;
438 struct page *page;
440 de = ext2_find_entry (dir, child, &page);
441 if (de) {
442 res = le32_to_cpu(de->inode);
443 ext2_put_page(page);
445 return res;
448 /* Releases the page */
449 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
450 struct page *page, struct inode *inode)
452 loff_t pos = page_offset(page) +
453 (char *) de - (char *) page_address(page);
454 unsigned len = ext2_rec_len_from_disk(de->rec_len);
455 int err;
457 lock_page(page);
458 err = __ext2_write_begin(NULL, page->mapping, pos, len,
459 AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
460 BUG_ON(err);
461 de->inode = cpu_to_le32(inode->i_ino);
462 ext2_set_de_type(de, inode);
463 err = ext2_commit_chunk(page, pos, len);
464 ext2_put_page(page);
465 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
466 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
467 mark_inode_dirty(dir);
471 * Parent is locked.
473 int ext2_add_link (struct dentry *dentry, struct inode *inode)
475 struct inode *dir = dentry->d_parent->d_inode;
476 const char *name = dentry->d_name.name;
477 int namelen = dentry->d_name.len;
478 unsigned chunk_size = ext2_chunk_size(dir);
479 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
480 unsigned short rec_len, name_len;
481 struct page *page = NULL;
482 ext2_dirent * de;
483 unsigned long npages = dir_pages(dir);
484 unsigned long n;
485 char *kaddr;
486 loff_t pos;
487 int err;
490 * We take care of directory expansion in the same loop.
491 * This code plays outside i_size, so it locks the page
492 * to protect that region.
494 for (n = 0; n <= npages; n++) {
495 char *dir_end;
497 page = ext2_get_page(dir, n, 0);
498 err = PTR_ERR(page);
499 if (IS_ERR(page))
500 goto out;
501 lock_page(page);
502 kaddr = page_address(page);
503 dir_end = kaddr + ext2_last_byte(dir, n);
504 de = (ext2_dirent *)kaddr;
505 kaddr += PAGE_CACHE_SIZE - reclen;
506 while ((char *)de <= kaddr) {
507 if ((char *)de == dir_end) {
508 /* We hit i_size */
509 name_len = 0;
510 rec_len = chunk_size;
511 de->rec_len = ext2_rec_len_to_disk(chunk_size);
512 de->inode = 0;
513 goto got_it;
515 if (de->rec_len == 0) {
516 ext2_error(dir->i_sb, __func__,
517 "zero-length directory entry");
518 err = -EIO;
519 goto out_unlock;
521 err = -EEXIST;
522 if (ext2_match (namelen, name, de))
523 goto out_unlock;
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)
527 goto got_it;
528 if (rec_len >= name_len + reclen)
529 goto got_it;
530 de = (ext2_dirent *) ((char *) de + rec_len);
532 unlock_page(page);
533 ext2_put_page(page);
535 BUG();
536 return -EINVAL;
538 got_it:
539 pos = page_offset(page) +
540 (char*)de - (char*)page_address(page);
541 err = __ext2_write_begin(NULL, page->mapping, pos, rec_len, 0,
542 &page, NULL);
543 if (err)
544 goto out_unlock;
545 if (de->inode) {
546 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
547 de1->rec_len = ext2_rec_len_to_disk(rec_len - name_len);
548 de->rec_len = ext2_rec_len_to_disk(name_len);
549 de = de1;
551 de->name_len = namelen;
552 memcpy(de->name, name, namelen);
553 de->inode = cpu_to_le32(inode->i_ino);
554 ext2_set_de_type (de, inode);
555 err = ext2_commit_chunk(page, pos, rec_len);
556 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
557 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
558 mark_inode_dirty(dir);
559 /* OFFSET_CACHE */
560 out_put:
561 ext2_put_page(page);
562 out:
563 return err;
564 out_unlock:
565 unlock_page(page);
566 goto out_put;
570 * ext2_delete_entry deletes a directory entry by merging it with the
571 * previous entry. Page is up-to-date. Releases the page.
573 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
575 struct address_space *mapping = page->mapping;
576 struct inode *inode = mapping->host;
577 char *kaddr = page_address(page);
578 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
579 unsigned to = ((char *)dir - kaddr) +
580 ext2_rec_len_from_disk(dir->rec_len);
581 loff_t pos;
582 ext2_dirent * pde = NULL;
583 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
584 int err;
586 while ((char*)de < (char*)dir) {
587 if (de->rec_len == 0) {
588 ext2_error(inode->i_sb, __func__,
589 "zero-length directory entry");
590 err = -EIO;
591 goto out;
593 pde = de;
594 de = ext2_next_entry(de);
596 if (pde)
597 from = (char*)pde - (char*)page_address(page);
598 pos = page_offset(page) + from;
599 lock_page(page);
600 err = __ext2_write_begin(NULL, page->mapping, pos, to - from, 0,
601 &page, NULL);
602 BUG_ON(err);
603 if (pde)
604 pde->rec_len = ext2_rec_len_to_disk(to - from);
605 dir->inode = 0;
606 err = ext2_commit_chunk(page, pos, to - from);
607 inode->i_ctime = inode->i_mtime = CURRENT_TIME_SEC;
608 EXT2_I(inode)->i_flags &= ~EXT2_BTREE_FL;
609 mark_inode_dirty(inode);
610 out:
611 ext2_put_page(page);
612 return err;
616 * Set the first fragment of directory.
618 int ext2_make_empty(struct inode *inode, struct inode *parent)
620 struct address_space *mapping = inode->i_mapping;
621 struct page *page = grab_cache_page(mapping, 0);
622 unsigned chunk_size = ext2_chunk_size(inode);
623 struct ext2_dir_entry_2 * de;
624 int err;
625 void *kaddr;
627 if (!page)
628 return -ENOMEM;
630 err = __ext2_write_begin(NULL, page->mapping, 0, chunk_size, 0,
631 &page, NULL);
632 if (err) {
633 unlock_page(page);
634 goto fail;
636 kaddr = kmap_atomic(page, KM_USER0);
637 memset(kaddr, 0, chunk_size);
638 de = (struct ext2_dir_entry_2 *)kaddr;
639 de->name_len = 1;
640 de->rec_len = ext2_rec_len_to_disk(EXT2_DIR_REC_LEN(1));
641 memcpy (de->name, ".\0\0", 4);
642 de->inode = cpu_to_le32(inode->i_ino);
643 ext2_set_de_type (de, inode);
645 de = (struct ext2_dir_entry_2 *)(kaddr + EXT2_DIR_REC_LEN(1));
646 de->name_len = 2;
647 de->rec_len = ext2_rec_len_to_disk(chunk_size - EXT2_DIR_REC_LEN(1));
648 de->inode = cpu_to_le32(parent->i_ino);
649 memcpy (de->name, "..\0", 4);
650 ext2_set_de_type (de, inode);
651 kunmap_atomic(kaddr, KM_USER0);
652 err = ext2_commit_chunk(page, 0, chunk_size);
653 fail:
654 page_cache_release(page);
655 return err;
659 * routine to check that the specified directory is empty (for rmdir)
661 int ext2_empty_dir (struct inode * inode)
663 struct page *page = NULL;
664 unsigned long i, npages = dir_pages(inode);
665 int dir_has_error = 0;
667 for (i = 0; i < npages; i++) {
668 char *kaddr;
669 ext2_dirent * de;
670 page = ext2_get_page(inode, i, dir_has_error);
672 if (IS_ERR(page)) {
673 dir_has_error = 1;
674 continue;
677 kaddr = page_address(page);
678 de = (ext2_dirent *)kaddr;
679 kaddr += ext2_last_byte(inode, i) - EXT2_DIR_REC_LEN(1);
681 while ((char *)de <= kaddr) {
682 if (de->rec_len == 0) {
683 ext2_error(inode->i_sb, __func__,
684 "zero-length directory entry");
685 printk("kaddr=%p, de=%p\n", kaddr, de);
686 goto not_empty;
688 if (de->inode != 0) {
689 /* check for . and .. */
690 if (de->name[0] != '.')
691 goto not_empty;
692 if (de->name_len > 2)
693 goto not_empty;
694 if (de->name_len < 2) {
695 if (de->inode !=
696 cpu_to_le32(inode->i_ino))
697 goto not_empty;
698 } else if (de->name[1] != '.')
699 goto not_empty;
701 de = ext2_next_entry(de);
703 ext2_put_page(page);
705 return 1;
707 not_empty:
708 ext2_put_page(page);
709 return 0;
712 const struct file_operations ext2_dir_operations = {
713 .llseek = generic_file_llseek,
714 .read = generic_read_dir,
715 .readdir = ext2_readdir,
716 .unlocked_ioctl = ext2_ioctl,
717 #ifdef CONFIG_COMPAT
718 .compat_ioctl = ext2_compat_ioctl,
719 #endif
720 .fsync = ext2_sync_file,