blob: 5b5f52876b427fc848bb934dbf7e4a7cd8160d7d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext2/dir.c
3 *
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)
8 *
9 * from
10 *
11 * linux/fs/minix/dir.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * ext2 directory handling functions
16 *
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 *
20 * All code that works with directory layout had been switched to pagecache
21 * and moved here. AV
22 */
23
24#include "ext2.h"
25#include <linux/pagemap.h>
26#include <linux/smp_lock.h>
27
28typedef struct ext2_dir_entry_2 ext2_dirent;
29
30/*
31 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
32 * more robust, but we have what we have
33 */
34static inline unsigned ext2_chunk_size(struct inode *inode)
35{
36 return inode->i_sb->s_blocksize;
37}
38
39static inline void ext2_put_page(struct page *page)
40{
41 kunmap(page);
42 page_cache_release(page);
43}
44
45static inline unsigned long dir_pages(struct inode *inode)
46{
47 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
48}
49
50/*
51 * Return the offset into page `page_nr' of the last valid
52 * byte in that page, plus one.
53 */
54static unsigned
55ext2_last_byte(struct inode *inode, unsigned long page_nr)
56{
57 unsigned last_byte = inode->i_size;
58
59 last_byte -= page_nr << PAGE_CACHE_SHIFT;
60 if (last_byte > PAGE_CACHE_SIZE)
61 last_byte = PAGE_CACHE_SIZE;
62 return last_byte;
63}
64
65static int ext2_commit_chunk(struct page *page, unsigned from, unsigned to)
66{
67 struct inode *dir = page->mapping->host;
68 int err = 0;
69 dir->i_version++;
70 page->mapping->a_ops->commit_write(NULL, page, from, to);
71 if (IS_DIRSYNC(dir))
72 err = write_one_page(page, 1);
73 else
74 unlock_page(page);
75 return err;
76}
77
78static void ext2_check_page(struct page *page)
79{
80 struct inode *dir = page->mapping->host;
81 struct super_block *sb = dir->i_sb;
82 unsigned chunk_size = ext2_chunk_size(dir);
83 char *kaddr = page_address(page);
84 u32 max_inumber = le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count);
85 unsigned offs, rec_len;
86 unsigned limit = PAGE_CACHE_SIZE;
87 ext2_dirent *p;
88 char *error;
89
90 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
91 limit = dir->i_size & ~PAGE_CACHE_MASK;
92 if (limit & (chunk_size - 1))
93 goto Ebadsize;
94 if (!limit)
95 goto out;
96 }
97 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
98 p = (ext2_dirent *)(kaddr + offs);
99 rec_len = le16_to_cpu(p->rec_len);
100
101 if (rec_len < EXT2_DIR_REC_LEN(1))
102 goto Eshort;
103 if (rec_len & 3)
104 goto Ealign;
105 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
106 goto Enamelen;
107 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
108 goto Espan;
109 if (le32_to_cpu(p->inode) > max_inumber)
110 goto Einumber;
111 }
112 if (offs != limit)
113 goto Eend;
114out:
115 SetPageChecked(page);
116 return;
117
118 /* Too bad, we had an error */
119
120Ebadsize:
121 ext2_error(sb, "ext2_check_page",
122 "size of directory #%lu is not a multiple of chunk size",
123 dir->i_ino
124 );
125 goto fail;
126Eshort:
127 error = "rec_len is smaller than minimal";
128 goto bad_entry;
129Ealign:
130 error = "unaligned directory entry";
131 goto bad_entry;
132Enamelen:
133 error = "rec_len is too small for name_len";
134 goto bad_entry;
135Espan:
136 error = "directory entry across blocks";
137 goto bad_entry;
138Einumber:
139 error = "inode out of bounds";
140bad_entry:
141 ext2_error (sb, "ext2_check_page", "bad entry in directory #%lu: %s - "
142 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
143 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
144 (unsigned long) le32_to_cpu(p->inode),
145 rec_len, p->name_len);
146 goto fail;
147Eend:
148 p = (ext2_dirent *)(kaddr + offs);
149 ext2_error (sb, "ext2_check_page",
150 "entry in directory #%lu spans the page boundary"
151 "offset=%lu, inode=%lu",
152 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
153 (unsigned long) le32_to_cpu(p->inode));
154fail:
155 SetPageChecked(page);
156 SetPageError(page);
157}
158
159static struct page * ext2_get_page(struct inode *dir, unsigned long n)
160{
161 struct address_space *mapping = dir->i_mapping;
162 struct page *page = read_cache_page(mapping, n,
163 (filler_t*)mapping->a_ops->readpage, NULL);
164 if (!IS_ERR(page)) {
165 wait_on_page_locked(page);
166 kmap(page);
167 if (!PageUptodate(page))
168 goto fail;
169 if (!PageChecked(page))
170 ext2_check_page(page);
171 if (PageError(page))
172 goto fail;
173 }
174 return page;
175
176fail:
177 ext2_put_page(page);
178 return ERR_PTR(-EIO);
179}
180
181/*
182 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
183 *
184 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
185 */
186static inline int ext2_match (int len, const char * const name,
187 struct ext2_dir_entry_2 * de)
188{
189 if (len != de->name_len)
190 return 0;
191 if (!de->inode)
192 return 0;
193 return !memcmp(name, de->name, len);
194}
195
196/*
197 * p is at least 6 bytes before the end of page
198 */
199static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
200{
201 return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
202}
203
204static inline unsigned
205ext2_validate_entry(char *base, unsigned offset, unsigned mask)
206{
207 ext2_dirent *de = (ext2_dirent*)(base + offset);
208 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
209 while ((char*)p < (char*)de) {
210 if (p->rec_len == 0)
211 break;
212 p = ext2_next_entry(p);
213 }
214 return (char *)p - base;
215}
216
217static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
218 [EXT2_FT_UNKNOWN] = DT_UNKNOWN,
219 [EXT2_FT_REG_FILE] = DT_REG,
220 [EXT2_FT_DIR] = DT_DIR,
221 [EXT2_FT_CHRDEV] = DT_CHR,
222 [EXT2_FT_BLKDEV] = DT_BLK,
223 [EXT2_FT_FIFO] = DT_FIFO,
224 [EXT2_FT_SOCK] = DT_SOCK,
225 [EXT2_FT_SYMLINK] = DT_LNK,
226};
227
228#define S_SHIFT 12
229static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
230 [S_IFREG >> S_SHIFT] = EXT2_FT_REG_FILE,
231 [S_IFDIR >> S_SHIFT] = EXT2_FT_DIR,
232 [S_IFCHR >> S_SHIFT] = EXT2_FT_CHRDEV,
233 [S_IFBLK >> S_SHIFT] = EXT2_FT_BLKDEV,
234 [S_IFIFO >> S_SHIFT] = EXT2_FT_FIFO,
235 [S_IFSOCK >> S_SHIFT] = EXT2_FT_SOCK,
236 [S_IFLNK >> S_SHIFT] = EXT2_FT_SYMLINK,
237};
238
239static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
240{
241 mode_t mode = inode->i_mode;
242 if (EXT2_HAS_INCOMPAT_FEATURE(inode->i_sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
243 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
244 else
245 de->file_type = 0;
246}
247
248static int
249ext2_readdir (struct file * filp, void * dirent, filldir_t filldir)
250{
251 loff_t pos = filp->f_pos;
252 struct inode *inode = filp->f_dentry->d_inode;
253 struct super_block *sb = inode->i_sb;
254 unsigned int offset = pos & ~PAGE_CACHE_MASK;
255 unsigned long n = pos >> PAGE_CACHE_SHIFT;
256 unsigned long npages = dir_pages(inode);
257 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
258 unsigned char *types = NULL;
259 int need_revalidate = (filp->f_version != inode->i_version);
260 int ret;
261
262 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
263 goto success;
264
265 if (EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
266 types = ext2_filetype_table;
267
268 for ( ; n < npages; n++, offset = 0) {
269 char *kaddr, *limit;
270 ext2_dirent *de;
271 struct page *page = ext2_get_page(inode, n);
272
273 if (IS_ERR(page)) {
274 ext2_error(sb, __FUNCTION__,
275 "bad page in #%lu",
276 inode->i_ino);
277 filp->f_pos += PAGE_CACHE_SIZE - offset;
278 ret = -EIO;
279 goto done;
280 }
281 kaddr = page_address(page);
282 if (need_revalidate) {
283 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
284 need_revalidate = 0;
285 }
286 de = (ext2_dirent *)(kaddr+offset);
287 limit = kaddr + ext2_last_byte(inode, n) - EXT2_DIR_REC_LEN(1);
288 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
289 if (de->rec_len == 0) {
290 ext2_error(sb, __FUNCTION__,
291 "zero-length directory entry");
292 ret = -EIO;
293 ext2_put_page(page);
294 goto done;
295 }
296 if (de->inode) {
297 int over;
298 unsigned char d_type = DT_UNKNOWN;
299
300 if (types && de->file_type < EXT2_FT_MAX)
301 d_type = types[de->file_type];
302
303 offset = (char *)de - kaddr;
304 over = filldir(dirent, de->name, de->name_len,
305 (n<<PAGE_CACHE_SHIFT) | offset,
306 le32_to_cpu(de->inode), d_type);
307 if (over) {
308 ext2_put_page(page);
309 goto success;
310 }
311 }
312 filp->f_pos += le16_to_cpu(de->rec_len);
313 }
314 ext2_put_page(page);
315 }
316
317success:
318 ret = 0;
319done:
320 filp->f_version = inode->i_version;
321 return ret;
322}
323
324/*
325 * ext2_find_entry()
326 *
327 * finds an entry in the specified directory with the wanted name. It
328 * returns the page in which the entry was found, and the entry itself
329 * (as a parameter - res_dir). Page is returned mapped and unlocked.
330 * Entry is guaranteed to be valid.
331 */
332struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
333 struct dentry *dentry, struct page ** res_page)
334{
335 const char *name = dentry->d_name.name;
336 int namelen = dentry->d_name.len;
337 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
338 unsigned long start, n;
339 unsigned long npages = dir_pages(dir);
340 struct page *page = NULL;
341 struct ext2_inode_info *ei = EXT2_I(dir);
342 ext2_dirent * de;
343
344 if (npages == 0)
345 goto out;
346
347 /* OFFSET_CACHE */
348 *res_page = NULL;
349
350 start = ei->i_dir_start_lookup;
351 if (start >= npages)
352 start = 0;
353 n = start;
354 do {
355 char *kaddr;
356 page = ext2_get_page(dir, n);
357 if (!IS_ERR(page)) {
358 kaddr = page_address(page);
359 de = (ext2_dirent *) kaddr;
360 kaddr += ext2_last_byte(dir, n) - reclen;
361 while ((char *) de <= kaddr) {
362 if (de->rec_len == 0) {
363 ext2_error(dir->i_sb, __FUNCTION__,
364 "zero-length directory entry");
365 ext2_put_page(page);
366 goto out;
367 }
368 if (ext2_match (namelen, name, de))
369 goto found;
370 de = ext2_next_entry(de);
371 }
372 ext2_put_page(page);
373 }
374 if (++n >= npages)
375 n = 0;
376 } while (n != start);
377out:
378 return NULL;
379
380found:
381 *res_page = page;
382 ei->i_dir_start_lookup = n;
383 return de;
384}
385
386struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
387{
388 struct page *page = ext2_get_page(dir, 0);
389 ext2_dirent *de = NULL;
390
391 if (!IS_ERR(page)) {
392 de = ext2_next_entry((ext2_dirent *) page_address(page));
393 *p = page;
394 }
395 return de;
396}
397
398ino_t ext2_inode_by_name(struct inode * dir, struct dentry *dentry)
399{
400 ino_t res = 0;
401 struct ext2_dir_entry_2 * de;
402 struct page *page;
403
404 de = ext2_find_entry (dir, dentry, &page);
405 if (de) {
406 res = le32_to_cpu(de->inode);
407 kunmap(page);
408 page_cache_release(page);
409 }
410 return res;
411}
412
413/* Releases the page */
414void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
415 struct page *page, struct inode *inode)
416{
417 unsigned from = (char *) de - (char *) page_address(page);
418 unsigned to = from + le16_to_cpu(de->rec_len);
419 int err;
420
421 lock_page(page);
422 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
423 if (err)
424 BUG();
425 de->inode = cpu_to_le32(inode->i_ino);
426 ext2_set_de_type (de, inode);
427 err = ext2_commit_chunk(page, from, to);
428 ext2_put_page(page);
429 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
430 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
431 mark_inode_dirty(dir);
432}
433
434/*
435 * Parent is locked.
436 */
437int ext2_add_link (struct dentry *dentry, struct inode *inode)
438{
439 struct inode *dir = dentry->d_parent->d_inode;
440 const char *name = dentry->d_name.name;
441 int namelen = dentry->d_name.len;
442 unsigned chunk_size = ext2_chunk_size(dir);
443 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
444 unsigned short rec_len, name_len;
445 struct page *page = NULL;
446 ext2_dirent * de;
447 unsigned long npages = dir_pages(dir);
448 unsigned long n;
449 char *kaddr;
450 unsigned from, to;
451 int err;
452
453 /*
454 * We take care of directory expansion in the same loop.
455 * This code plays outside i_size, so it locks the page
456 * to protect that region.
457 */
458 for (n = 0; n <= npages; n++) {
459 char *dir_end;
460
461 page = ext2_get_page(dir, n);
462 err = PTR_ERR(page);
463 if (IS_ERR(page))
464 goto out;
465 lock_page(page);
466 kaddr = page_address(page);
467 dir_end = kaddr + ext2_last_byte(dir, n);
468 de = (ext2_dirent *)kaddr;
469 kaddr += PAGE_CACHE_SIZE - reclen;
470 while ((char *)de <= kaddr) {
471 if ((char *)de == dir_end) {
472 /* We hit i_size */
473 name_len = 0;
474 rec_len = chunk_size;
475 de->rec_len = cpu_to_le16(chunk_size);
476 de->inode = 0;
477 goto got_it;
478 }
479 if (de->rec_len == 0) {
480 ext2_error(dir->i_sb, __FUNCTION__,
481 "zero-length directory entry");
482 err = -EIO;
483 goto out_unlock;
484 }
485 err = -EEXIST;
486 if (ext2_match (namelen, name, de))
487 goto out_unlock;
488 name_len = EXT2_DIR_REC_LEN(de->name_len);
489 rec_len = le16_to_cpu(de->rec_len);
490 if (!de->inode && rec_len >= reclen)
491 goto got_it;
492 if (rec_len >= name_len + reclen)
493 goto got_it;
494 de = (ext2_dirent *) ((char *) de + rec_len);
495 }
496 unlock_page(page);
497 ext2_put_page(page);
498 }
499 BUG();
500 return -EINVAL;
501
502got_it:
503 from = (char*)de - (char*)page_address(page);
504 to = from + rec_len;
505 err = page->mapping->a_ops->prepare_write(NULL, page, from, to);
506 if (err)
507 goto out_unlock;
508 if (de->inode) {
509 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
510 de1->rec_len = cpu_to_le16(rec_len - name_len);
511 de->rec_len = cpu_to_le16(name_len);
512 de = de1;
513 }
514 de->name_len = namelen;
515 memcpy (de->name, name, namelen);
516 de->inode = cpu_to_le32(inode->i_ino);
517 ext2_set_de_type (de, inode);
518 err = ext2_commit_chunk(page, from, to);
519 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
520 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
521 mark_inode_dirty(dir);
522 /* OFFSET_CACHE */
523out_put:
524 ext2_put_page(page);
525out:
526 return err;
527out_unlock:
528 unlock_page(page);
529 goto out_put;
530}
531
532/*
533 * ext2_delete_entry deletes a directory entry by merging it with the
534 * previous entry. Page is up-to-date. Releases the page.
535 */
536int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
537{
538 struct address_space *mapping = page->mapping;
539 struct inode *inode = mapping->host;
540 char *kaddr = page_address(page);
541 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
542 unsigned to = ((char*)dir - kaddr) + le16_to_cpu(dir->rec_len);
543 ext2_dirent * pde = NULL;
544 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
545 int err;
546
547 while ((char*)de < (char*)dir) {
548 if (de->rec_len == 0) {
549 ext2_error(inode->i_sb, __FUNCTION__,
550 "zero-length directory entry");
551 err = -EIO;
552 goto out;
553 }
554 pde = de;
555 de = ext2_next_entry(de);
556 }
557 if (pde)
558 from = (char*)pde - (char*)page_address(page);
559 lock_page(page);
560 err = mapping->a_ops->prepare_write(NULL, page, from, to);
561 if (err)
562 BUG();
563 if (pde)
564 pde->rec_len = cpu_to_le16(to-from);
565 dir->inode = 0;
566 err = ext2_commit_chunk(page, from, to);
567 inode->i_ctime = inode->i_mtime = CURRENT_TIME_SEC;
568 EXT2_I(inode)->i_flags &= ~EXT2_BTREE_FL;
569 mark_inode_dirty(inode);
570out:
571 ext2_put_page(page);
572 return err;
573}
574
575/*
576 * Set the first fragment of directory.
577 */
578int ext2_make_empty(struct inode *inode, struct inode *parent)
579{
580 struct address_space *mapping = inode->i_mapping;
581 struct page *page = grab_cache_page(mapping, 0);
582 unsigned chunk_size = ext2_chunk_size(inode);
583 struct ext2_dir_entry_2 * de;
584 int err;
585 void *kaddr;
586
587 if (!page)
588 return -ENOMEM;
589 err = mapping->a_ops->prepare_write(NULL, page, 0, chunk_size);
590 if (err) {
591 unlock_page(page);
592 goto fail;
593 }
594 kaddr = kmap_atomic(page, KM_USER0);
595 memset(kaddr, 0, chunk_size);
596 de = (struct ext2_dir_entry_2 *)kaddr;
597 de->name_len = 1;
598 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(1));
599 memcpy (de->name, ".\0\0", 4);
600 de->inode = cpu_to_le32(inode->i_ino);
601 ext2_set_de_type (de, inode);
602
603 de = (struct ext2_dir_entry_2 *)(kaddr + EXT2_DIR_REC_LEN(1));
604 de->name_len = 2;
605 de->rec_len = cpu_to_le16(chunk_size - EXT2_DIR_REC_LEN(1));
606 de->inode = cpu_to_le32(parent->i_ino);
607 memcpy (de->name, "..\0", 4);
608 ext2_set_de_type (de, inode);
609 kunmap_atomic(kaddr, KM_USER0);
610 err = ext2_commit_chunk(page, 0, chunk_size);
611fail:
612 page_cache_release(page);
613 return err;
614}
615
616/*
617 * routine to check that the specified directory is empty (for rmdir)
618 */
619int ext2_empty_dir (struct inode * inode)
620{
621 struct page *page = NULL;
622 unsigned long i, npages = dir_pages(inode);
623
624 for (i = 0; i < npages; i++) {
625 char *kaddr;
626 ext2_dirent * de;
627 page = ext2_get_page(inode, i);
628
629 if (IS_ERR(page))
630 continue;
631
632 kaddr = page_address(page);
633 de = (ext2_dirent *)kaddr;
634 kaddr += ext2_last_byte(inode, i) - EXT2_DIR_REC_LEN(1);
635
636 while ((char *)de <= kaddr) {
637 if (de->rec_len == 0) {
638 ext2_error(inode->i_sb, __FUNCTION__,
639 "zero-length directory entry");
640 printk("kaddr=%p, de=%p\n", kaddr, de);
641 goto not_empty;
642 }
643 if (de->inode != 0) {
644 /* check for . and .. */
645 if (de->name[0] != '.')
646 goto not_empty;
647 if (de->name_len > 2)
648 goto not_empty;
649 if (de->name_len < 2) {
650 if (de->inode !=
651 cpu_to_le32(inode->i_ino))
652 goto not_empty;
653 } else if (de->name[1] != '.')
654 goto not_empty;
655 }
656 de = ext2_next_entry(de);
657 }
658 ext2_put_page(page);
659 }
660 return 1;
661
662not_empty:
663 ext2_put_page(page);
664 return 0;
665}
666
667struct file_operations ext2_dir_operations = {
668 .llseek = generic_file_llseek,
669 .read = generic_read_dir,
670 .readdir = ext2_readdir,
671 .ioctl = ext2_ioctl,
672 .fsync = ext2_sync_file,
673};