gro: Allow tunnel stacking in the case of FOU/GUE
[linux/fpc-iii.git] / fs / ext2 / dir.c
blob796b491e69785e4b703b7c3a0cbc7e97d3510a40
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;
32 * Tests against MAX_REC_LEN etc were put in place for 64k block
33 * sizes; if that is not possible on this arch, we can skip
34 * those tests and speed things up.
36 static inline unsigned ext2_rec_len_from_disk(__le16 dlen)
38 unsigned len = le16_to_cpu(dlen);
40 #if (PAGE_CACHE_SIZE >= 65536)
41 if (len == EXT2_MAX_REC_LEN)
42 return 1 << 16;
43 #endif
44 return len;
47 static inline __le16 ext2_rec_len_to_disk(unsigned len)
49 #if (PAGE_CACHE_SIZE >= 65536)
50 if (len == (1 << 16))
51 return cpu_to_le16(EXT2_MAX_REC_LEN);
52 else
53 BUG_ON(len > (1 << 16));
54 #endif
55 return cpu_to_le16(len);
59 * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
60 * more robust, but we have what we have
62 static inline unsigned ext2_chunk_size(struct inode *inode)
64 return inode->i_sb->s_blocksize;
67 static inline void ext2_put_page(struct page *page)
69 kunmap(page);
70 page_cache_release(page);
73 static inline unsigned long dir_pages(struct inode *inode)
75 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
79 * Return the offset into page `page_nr' of the last valid
80 * byte in that page, plus one.
82 static unsigned
83 ext2_last_byte(struct inode *inode, unsigned long page_nr)
85 unsigned last_byte = inode->i_size;
87 last_byte -= page_nr << PAGE_CACHE_SHIFT;
88 if (last_byte > PAGE_CACHE_SIZE)
89 last_byte = PAGE_CACHE_SIZE;
90 return last_byte;
93 static int ext2_commit_chunk(struct page *page, loff_t pos, unsigned len)
95 struct address_space *mapping = page->mapping;
96 struct inode *dir = mapping->host;
97 int err = 0;
99 dir->i_version++;
100 block_write_end(NULL, mapping, pos, len, len, page, NULL);
102 if (pos+len > dir->i_size) {
103 i_size_write(dir, pos+len);
104 mark_inode_dirty(dir);
107 if (IS_DIRSYNC(dir)) {
108 err = write_one_page(page, 1);
109 if (!err)
110 err = sync_inode_metadata(dir, 1);
111 } else {
112 unlock_page(page);
115 return err;
118 static void ext2_check_page(struct page *page, int quiet)
120 struct inode *dir = page->mapping->host;
121 struct super_block *sb = dir->i_sb;
122 unsigned chunk_size = ext2_chunk_size(dir);
123 char *kaddr = page_address(page);
124 u32 max_inumber = le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count);
125 unsigned offs, rec_len;
126 unsigned limit = PAGE_CACHE_SIZE;
127 ext2_dirent *p;
128 char *error;
130 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
131 limit = dir->i_size & ~PAGE_CACHE_MASK;
132 if (limit & (chunk_size - 1))
133 goto Ebadsize;
134 if (!limit)
135 goto out;
137 for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
138 p = (ext2_dirent *)(kaddr + offs);
139 rec_len = ext2_rec_len_from_disk(p->rec_len);
141 if (unlikely(rec_len < EXT2_DIR_REC_LEN(1)))
142 goto Eshort;
143 if (unlikely(rec_len & 3))
144 goto Ealign;
145 if (unlikely(rec_len < EXT2_DIR_REC_LEN(p->name_len)))
146 goto Enamelen;
147 if (unlikely(((offs + rec_len - 1) ^ offs) & ~(chunk_size-1)))
148 goto Espan;
149 if (unlikely(le32_to_cpu(p->inode) > max_inumber))
150 goto Einumber;
152 if (offs != limit)
153 goto Eend;
154 out:
155 SetPageChecked(page);
156 return;
158 /* Too bad, we had an error */
160 Ebadsize:
161 if (!quiet)
162 ext2_error(sb, __func__,
163 "size of directory #%lu is not a multiple "
164 "of chunk size", dir->i_ino);
165 goto fail;
166 Eshort:
167 error = "rec_len is smaller than minimal";
168 goto bad_entry;
169 Ealign:
170 error = "unaligned directory entry";
171 goto bad_entry;
172 Enamelen:
173 error = "rec_len is too small for name_len";
174 goto bad_entry;
175 Espan:
176 error = "directory entry across blocks";
177 goto bad_entry;
178 Einumber:
179 error = "inode out of bounds";
180 bad_entry:
181 if (!quiet)
182 ext2_error(sb, __func__, "bad entry in directory #%lu: : %s - "
183 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
184 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
185 (unsigned long) le32_to_cpu(p->inode),
186 rec_len, p->name_len);
187 goto fail;
188 Eend:
189 if (!quiet) {
190 p = (ext2_dirent *)(kaddr + offs);
191 ext2_error(sb, "ext2_check_page",
192 "entry in directory #%lu spans the page boundary"
193 "offset=%lu, inode=%lu",
194 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
195 (unsigned long) le32_to_cpu(p->inode));
197 fail:
198 SetPageChecked(page);
199 SetPageError(page);
202 static struct page * ext2_get_page(struct inode *dir, unsigned long n,
203 int quiet)
205 struct address_space *mapping = dir->i_mapping;
206 struct page *page = read_mapping_page(mapping, n, NULL);
207 if (!IS_ERR(page)) {
208 kmap(page);
209 if (!PageChecked(page))
210 ext2_check_page(page, quiet);
211 if (PageError(page))
212 goto fail;
214 return page;
216 fail:
217 ext2_put_page(page);
218 return ERR_PTR(-EIO);
222 * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
224 * len <= EXT2_NAME_LEN and de != NULL are guaranteed by caller.
226 static inline int ext2_match (int len, const char * const name,
227 struct ext2_dir_entry_2 * de)
229 if (len != de->name_len)
230 return 0;
231 if (!de->inode)
232 return 0;
233 return !memcmp(name, de->name, len);
237 * p is at least 6 bytes before the end of page
239 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
241 return (ext2_dirent *)((char *)p +
242 ext2_rec_len_from_disk(p->rec_len));
245 static inline unsigned
246 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
248 ext2_dirent *de = (ext2_dirent*)(base + offset);
249 ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
250 while ((char*)p < (char*)de) {
251 if (p->rec_len == 0)
252 break;
253 p = ext2_next_entry(p);
255 return (char *)p - base;
258 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
259 [EXT2_FT_UNKNOWN] = DT_UNKNOWN,
260 [EXT2_FT_REG_FILE] = DT_REG,
261 [EXT2_FT_DIR] = DT_DIR,
262 [EXT2_FT_CHRDEV] = DT_CHR,
263 [EXT2_FT_BLKDEV] = DT_BLK,
264 [EXT2_FT_FIFO] = DT_FIFO,
265 [EXT2_FT_SOCK] = DT_SOCK,
266 [EXT2_FT_SYMLINK] = DT_LNK,
269 #define S_SHIFT 12
270 static unsigned char ext2_type_by_mode[S_IFMT >> S_SHIFT] = {
271 [S_IFREG >> S_SHIFT] = EXT2_FT_REG_FILE,
272 [S_IFDIR >> S_SHIFT] = EXT2_FT_DIR,
273 [S_IFCHR >> S_SHIFT] = EXT2_FT_CHRDEV,
274 [S_IFBLK >> S_SHIFT] = EXT2_FT_BLKDEV,
275 [S_IFIFO >> S_SHIFT] = EXT2_FT_FIFO,
276 [S_IFSOCK >> S_SHIFT] = EXT2_FT_SOCK,
277 [S_IFLNK >> S_SHIFT] = EXT2_FT_SYMLINK,
280 static inline void ext2_set_de_type(ext2_dirent *de, struct inode *inode)
282 umode_t mode = inode->i_mode;
283 if (EXT2_HAS_INCOMPAT_FEATURE(inode->i_sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
284 de->file_type = ext2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
285 else
286 de->file_type = 0;
289 static int
290 ext2_readdir(struct file *file, struct dir_context *ctx)
292 loff_t pos = ctx->pos;
293 struct inode *inode = file_inode(file);
294 struct super_block *sb = inode->i_sb;
295 unsigned int offset = pos & ~PAGE_CACHE_MASK;
296 unsigned long n = pos >> PAGE_CACHE_SHIFT;
297 unsigned long npages = dir_pages(inode);
298 unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
299 unsigned char *types = NULL;
300 int need_revalidate = file->f_version != inode->i_version;
302 if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
303 return 0;
305 if (EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
306 types = ext2_filetype_table;
308 for ( ; n < npages; n++, offset = 0) {
309 char *kaddr, *limit;
310 ext2_dirent *de;
311 struct page *page = ext2_get_page(inode, n, 0);
313 if (IS_ERR(page)) {
314 ext2_error(sb, __func__,
315 "bad page in #%lu",
316 inode->i_ino);
317 ctx->pos += PAGE_CACHE_SIZE - offset;
318 return PTR_ERR(page);
320 kaddr = page_address(page);
321 if (unlikely(need_revalidate)) {
322 if (offset) {
323 offset = ext2_validate_entry(kaddr, offset, chunk_mask);
324 ctx->pos = (n<<PAGE_CACHE_SHIFT) + offset;
326 file->f_version = inode->i_version;
327 need_revalidate = 0;
329 de = (ext2_dirent *)(kaddr+offset);
330 limit = kaddr + ext2_last_byte(inode, n) - EXT2_DIR_REC_LEN(1);
331 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
332 if (de->rec_len == 0) {
333 ext2_error(sb, __func__,
334 "zero-length directory entry");
335 ext2_put_page(page);
336 return -EIO;
338 if (de->inode) {
339 unsigned char d_type = DT_UNKNOWN;
341 if (types && de->file_type < EXT2_FT_MAX)
342 d_type = types[de->file_type];
344 if (!dir_emit(ctx, de->name, de->name_len,
345 le32_to_cpu(de->inode),
346 d_type)) {
347 ext2_put_page(page);
348 return 0;
351 ctx->pos += ext2_rec_len_from_disk(de->rec_len);
353 ext2_put_page(page);
355 return 0;
359 * ext2_find_entry()
361 * finds an entry in the specified directory with the wanted name. It
362 * returns the page in which the entry was found (as a parameter - res_page),
363 * and the entry itself. Page is returned mapped and unlocked.
364 * Entry is guaranteed to be valid.
366 struct ext2_dir_entry_2 *ext2_find_entry (struct inode * dir,
367 struct qstr *child, struct page ** res_page)
369 const char *name = child->name;
370 int namelen = child->len;
371 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
372 unsigned long start, n;
373 unsigned long npages = dir_pages(dir);
374 struct page *page = NULL;
375 struct ext2_inode_info *ei = EXT2_I(dir);
376 ext2_dirent * de;
377 int dir_has_error = 0;
379 if (npages == 0)
380 goto out;
382 /* OFFSET_CACHE */
383 *res_page = NULL;
385 start = ei->i_dir_start_lookup;
386 if (start >= npages)
387 start = 0;
388 n = start;
389 do {
390 char *kaddr;
391 page = ext2_get_page(dir, n, dir_has_error);
392 if (!IS_ERR(page)) {
393 kaddr = page_address(page);
394 de = (ext2_dirent *) kaddr;
395 kaddr += ext2_last_byte(dir, n) - reclen;
396 while ((char *) de <= kaddr) {
397 if (de->rec_len == 0) {
398 ext2_error(dir->i_sb, __func__,
399 "zero-length directory entry");
400 ext2_put_page(page);
401 goto out;
403 if (ext2_match (namelen, name, de))
404 goto found;
405 de = ext2_next_entry(de);
407 ext2_put_page(page);
408 } else
409 dir_has_error = 1;
411 if (++n >= npages)
412 n = 0;
413 /* next page is past the blocks we've got */
414 if (unlikely(n > (dir->i_blocks >> (PAGE_CACHE_SHIFT - 9)))) {
415 ext2_error(dir->i_sb, __func__,
416 "dir %lu size %lld exceeds block count %llu",
417 dir->i_ino, dir->i_size,
418 (unsigned long long)dir->i_blocks);
419 goto out;
421 } while (n != start);
422 out:
423 return NULL;
425 found:
426 *res_page = page;
427 ei->i_dir_start_lookup = n;
428 return de;
431 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p)
433 struct page *page = ext2_get_page(dir, 0, 0);
434 ext2_dirent *de = NULL;
436 if (!IS_ERR(page)) {
437 de = ext2_next_entry((ext2_dirent *) page_address(page));
438 *p = page;
440 return de;
443 ino_t ext2_inode_by_name(struct inode *dir, struct qstr *child)
445 ino_t res = 0;
446 struct ext2_dir_entry_2 *de;
447 struct page *page;
449 de = ext2_find_entry (dir, child, &page);
450 if (de) {
451 res = le32_to_cpu(de->inode);
452 ext2_put_page(page);
454 return res;
457 static int ext2_prepare_chunk(struct page *page, loff_t pos, unsigned len)
459 return __block_write_begin(page, pos, len, ext2_get_block);
462 /* Releases the page */
463 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
464 struct page *page, struct inode *inode, int update_times)
466 loff_t pos = page_offset(page) +
467 (char *) de - (char *) page_address(page);
468 unsigned len = ext2_rec_len_from_disk(de->rec_len);
469 int err;
471 lock_page(page);
472 err = ext2_prepare_chunk(page, pos, len);
473 BUG_ON(err);
474 de->inode = cpu_to_le32(inode->i_ino);
475 ext2_set_de_type(de, inode);
476 err = ext2_commit_chunk(page, pos, len);
477 ext2_put_page(page);
478 if (update_times)
479 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
480 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
481 mark_inode_dirty(dir);
485 * Parent is locked.
487 int ext2_add_link (struct dentry *dentry, struct inode *inode)
489 struct inode *dir = d_inode(dentry->d_parent);
490 const char *name = dentry->d_name.name;
491 int namelen = dentry->d_name.len;
492 unsigned chunk_size = ext2_chunk_size(dir);
493 unsigned reclen = EXT2_DIR_REC_LEN(namelen);
494 unsigned short rec_len, name_len;
495 struct page *page = NULL;
496 ext2_dirent * de;
497 unsigned long npages = dir_pages(dir);
498 unsigned long n;
499 char *kaddr;
500 loff_t pos;
501 int err;
504 * We take care of directory expansion in the same loop.
505 * This code plays outside i_size, so it locks the page
506 * to protect that region.
508 for (n = 0; n <= npages; n++) {
509 char *dir_end;
511 page = ext2_get_page(dir, n, 0);
512 err = PTR_ERR(page);
513 if (IS_ERR(page))
514 goto out;
515 lock_page(page);
516 kaddr = page_address(page);
517 dir_end = kaddr + ext2_last_byte(dir, n);
518 de = (ext2_dirent *)kaddr;
519 kaddr += PAGE_CACHE_SIZE - reclen;
520 while ((char *)de <= kaddr) {
521 if ((char *)de == dir_end) {
522 /* We hit i_size */
523 name_len = 0;
524 rec_len = chunk_size;
525 de->rec_len = ext2_rec_len_to_disk(chunk_size);
526 de->inode = 0;
527 goto got_it;
529 if (de->rec_len == 0) {
530 ext2_error(dir->i_sb, __func__,
531 "zero-length directory entry");
532 err = -EIO;
533 goto out_unlock;
535 err = -EEXIST;
536 if (ext2_match (namelen, name, de))
537 goto out_unlock;
538 name_len = EXT2_DIR_REC_LEN(de->name_len);
539 rec_len = ext2_rec_len_from_disk(de->rec_len);
540 if (!de->inode && rec_len >= reclen)
541 goto got_it;
542 if (rec_len >= name_len + reclen)
543 goto got_it;
544 de = (ext2_dirent *) ((char *) de + rec_len);
546 unlock_page(page);
547 ext2_put_page(page);
549 BUG();
550 return -EINVAL;
552 got_it:
553 pos = page_offset(page) +
554 (char*)de - (char*)page_address(page);
555 err = ext2_prepare_chunk(page, pos, rec_len);
556 if (err)
557 goto out_unlock;
558 if (de->inode) {
559 ext2_dirent *de1 = (ext2_dirent *) ((char *) de + name_len);
560 de1->rec_len = ext2_rec_len_to_disk(rec_len - name_len);
561 de->rec_len = ext2_rec_len_to_disk(name_len);
562 de = de1;
564 de->name_len = namelen;
565 memcpy(de->name, name, namelen);
566 de->inode = cpu_to_le32(inode->i_ino);
567 ext2_set_de_type (de, inode);
568 err = ext2_commit_chunk(page, pos, rec_len);
569 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
570 EXT2_I(dir)->i_flags &= ~EXT2_BTREE_FL;
571 mark_inode_dirty(dir);
572 /* OFFSET_CACHE */
573 out_put:
574 ext2_put_page(page);
575 out:
576 return err;
577 out_unlock:
578 unlock_page(page);
579 goto out_put;
583 * ext2_delete_entry deletes a directory entry by merging it with the
584 * previous entry. Page is up-to-date. Releases the page.
586 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page )
588 struct inode *inode = page->mapping->host;
589 char *kaddr = page_address(page);
590 unsigned from = ((char*)dir - kaddr) & ~(ext2_chunk_size(inode)-1);
591 unsigned to = ((char *)dir - kaddr) +
592 ext2_rec_len_from_disk(dir->rec_len);
593 loff_t pos;
594 ext2_dirent * pde = NULL;
595 ext2_dirent * de = (ext2_dirent *) (kaddr + from);
596 int err;
598 while ((char*)de < (char*)dir) {
599 if (de->rec_len == 0) {
600 ext2_error(inode->i_sb, __func__,
601 "zero-length directory entry");
602 err = -EIO;
603 goto out;
605 pde = de;
606 de = ext2_next_entry(de);
608 if (pde)
609 from = (char*)pde - (char*)page_address(page);
610 pos = page_offset(page) + from;
611 lock_page(page);
612 err = ext2_prepare_chunk(page, pos, to - from);
613 BUG_ON(err);
614 if (pde)
615 pde->rec_len = ext2_rec_len_to_disk(to - from);
616 dir->inode = 0;
617 err = ext2_commit_chunk(page, pos, to - from);
618 inode->i_ctime = inode->i_mtime = CURRENT_TIME_SEC;
619 EXT2_I(inode)->i_flags &= ~EXT2_BTREE_FL;
620 mark_inode_dirty(inode);
621 out:
622 ext2_put_page(page);
623 return err;
627 * Set the first fragment of directory.
629 int ext2_make_empty(struct inode *inode, struct inode *parent)
631 struct page *page = grab_cache_page(inode->i_mapping, 0);
632 unsigned chunk_size = ext2_chunk_size(inode);
633 struct ext2_dir_entry_2 * de;
634 int err;
635 void *kaddr;
637 if (!page)
638 return -ENOMEM;
640 err = ext2_prepare_chunk(page, 0, chunk_size);
641 if (err) {
642 unlock_page(page);
643 goto fail;
645 kaddr = kmap_atomic(page);
646 memset(kaddr, 0, chunk_size);
647 de = (struct ext2_dir_entry_2 *)kaddr;
648 de->name_len = 1;
649 de->rec_len = ext2_rec_len_to_disk(EXT2_DIR_REC_LEN(1));
650 memcpy (de->name, ".\0\0", 4);
651 de->inode = cpu_to_le32(inode->i_ino);
652 ext2_set_de_type (de, inode);
654 de = (struct ext2_dir_entry_2 *)(kaddr + EXT2_DIR_REC_LEN(1));
655 de->name_len = 2;
656 de->rec_len = ext2_rec_len_to_disk(chunk_size - EXT2_DIR_REC_LEN(1));
657 de->inode = cpu_to_le32(parent->i_ino);
658 memcpy (de->name, "..\0", 4);
659 ext2_set_de_type (de, inode);
660 kunmap_atomic(kaddr);
661 err = ext2_commit_chunk(page, 0, chunk_size);
662 fail:
663 page_cache_release(page);
664 return err;
668 * routine to check that the specified directory is empty (for rmdir)
670 int ext2_empty_dir (struct inode * inode)
672 struct page *page = NULL;
673 unsigned long i, npages = dir_pages(inode);
674 int dir_has_error = 0;
676 for (i = 0; i < npages; i++) {
677 char *kaddr;
678 ext2_dirent * de;
679 page = ext2_get_page(inode, i, dir_has_error);
681 if (IS_ERR(page)) {
682 dir_has_error = 1;
683 continue;
686 kaddr = page_address(page);
687 de = (ext2_dirent *)kaddr;
688 kaddr += ext2_last_byte(inode, i) - EXT2_DIR_REC_LEN(1);
690 while ((char *)de <= kaddr) {
691 if (de->rec_len == 0) {
692 ext2_error(inode->i_sb, __func__,
693 "zero-length directory entry");
694 printk("kaddr=%p, de=%p\n", kaddr, de);
695 goto not_empty;
697 if (de->inode != 0) {
698 /* check for . and .. */
699 if (de->name[0] != '.')
700 goto not_empty;
701 if (de->name_len > 2)
702 goto not_empty;
703 if (de->name_len < 2) {
704 if (de->inode !=
705 cpu_to_le32(inode->i_ino))
706 goto not_empty;
707 } else if (de->name[1] != '.')
708 goto not_empty;
710 de = ext2_next_entry(de);
712 ext2_put_page(page);
714 return 1;
716 not_empty:
717 ext2_put_page(page);
718 return 0;
721 const struct file_operations ext2_dir_operations = {
722 .llseek = generic_file_llseek,
723 .read = generic_read_dir,
724 .iterate = ext2_readdir,
725 .unlocked_ioctl = ext2_ioctl,
726 #ifdef CONFIG_COMPAT
727 .compat_ioctl = ext2_compat_ioctl,
728 #endif
729 .fsync = ext2_fsync,