Loading...
1/*
2 * dir.c - NILFS directory entry operations
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * Modified for NILFS by Amagai Yoshiji.
17 */
18/*
19 * linux/fs/ext2/dir.c
20 *
21 * Copyright (C) 1992, 1993, 1994, 1995
22 * Remy Card (card@masi.ibp.fr)
23 * Laboratoire MASI - Institut Blaise Pascal
24 * Universite Pierre et Marie Curie (Paris VI)
25 *
26 * from
27 *
28 * linux/fs/minix/dir.c
29 *
30 * Copyright (C) 1991, 1992 Linus Torvalds
31 *
32 * ext2 directory handling functions
33 *
34 * Big-endian to little-endian byte-swapping/bitmaps by
35 * David S. Miller (davem@caip.rutgers.edu), 1995
36 *
37 * All code that works with directory layout had been switched to pagecache
38 * and moved here. AV
39 */
40
41#include <linux/pagemap.h>
42#include "nilfs.h"
43#include "page.h"
44
45static inline unsigned int nilfs_rec_len_from_disk(__le16 dlen)
46{
47 unsigned int len = le16_to_cpu(dlen);
48
49#if (PAGE_SIZE >= 65536)
50 if (len == NILFS_MAX_REC_LEN)
51 return 1 << 16;
52#endif
53 return len;
54}
55
56static inline __le16 nilfs_rec_len_to_disk(unsigned int len)
57{
58#if (PAGE_SIZE >= 65536)
59 if (len == (1 << 16))
60 return cpu_to_le16(NILFS_MAX_REC_LEN);
61
62 BUG_ON(len > (1 << 16));
63#endif
64 return cpu_to_le16(len);
65}
66
67/*
68 * nilfs uses block-sized chunks. Arguably, sector-sized ones would be
69 * more robust, but we have what we have
70 */
71static inline unsigned int nilfs_chunk_size(struct inode *inode)
72{
73 return inode->i_sb->s_blocksize;
74}
75
76static inline void nilfs_put_page(struct page *page)
77{
78 kunmap(page);
79 put_page(page);
80}
81
82/*
83 * Return the offset into page `page_nr' of the last valid
84 * byte in that page, plus one.
85 */
86static unsigned int nilfs_last_byte(struct inode *inode, unsigned long page_nr)
87{
88 unsigned int last_byte = inode->i_size;
89
90 last_byte -= page_nr << PAGE_SHIFT;
91 if (last_byte > PAGE_SIZE)
92 last_byte = PAGE_SIZE;
93 return last_byte;
94}
95
96static int nilfs_prepare_chunk(struct page *page, unsigned int from,
97 unsigned int to)
98{
99 loff_t pos = page_offset(page) + from;
100
101 return __block_write_begin(page, pos, to - from, nilfs_get_block);
102}
103
104static void nilfs_commit_chunk(struct page *page,
105 struct address_space *mapping,
106 unsigned int from, unsigned int to)
107{
108 struct inode *dir = mapping->host;
109 loff_t pos = page_offset(page) + from;
110 unsigned int len = to - from;
111 unsigned int nr_dirty, copied;
112 int err;
113
114 nr_dirty = nilfs_page_count_clean_buffers(page, from, to);
115 copied = block_write_end(NULL, mapping, pos, len, len, page, NULL);
116 if (pos + copied > dir->i_size)
117 i_size_write(dir, pos + copied);
118 if (IS_DIRSYNC(dir))
119 nilfs_set_transaction_flag(NILFS_TI_SYNC);
120 err = nilfs_set_file_dirty(dir, nr_dirty);
121 WARN_ON(err); /* do not happen */
122 unlock_page(page);
123}
124
125static bool nilfs_check_page(struct page *page)
126{
127 struct inode *dir = page->mapping->host;
128 struct super_block *sb = dir->i_sb;
129 unsigned int chunk_size = nilfs_chunk_size(dir);
130 char *kaddr = page_address(page);
131 unsigned int offs, rec_len;
132 unsigned int limit = PAGE_SIZE;
133 struct nilfs_dir_entry *p;
134 char *error;
135
136 if ((dir->i_size >> PAGE_SHIFT) == page->index) {
137 limit = dir->i_size & ~PAGE_MASK;
138 if (limit & (chunk_size - 1))
139 goto Ebadsize;
140 if (!limit)
141 goto out;
142 }
143 for (offs = 0; offs <= limit - NILFS_DIR_REC_LEN(1); offs += rec_len) {
144 p = (struct nilfs_dir_entry *)(kaddr + offs);
145 rec_len = nilfs_rec_len_from_disk(p->rec_len);
146
147 if (rec_len < NILFS_DIR_REC_LEN(1))
148 goto Eshort;
149 if (rec_len & 3)
150 goto Ealign;
151 if (rec_len < NILFS_DIR_REC_LEN(p->name_len))
152 goto Enamelen;
153 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
154 goto Espan;
155 }
156 if (offs != limit)
157 goto Eend;
158out:
159 SetPageChecked(page);
160 return true;
161
162 /* Too bad, we had an error */
163
164Ebadsize:
165 nilfs_error(sb,
166 "size of directory #%lu is not a multiple of chunk size",
167 dir->i_ino);
168 goto fail;
169Eshort:
170 error = "rec_len is smaller than minimal";
171 goto bad_entry;
172Ealign:
173 error = "unaligned directory entry";
174 goto bad_entry;
175Enamelen:
176 error = "rec_len is too small for name_len";
177 goto bad_entry;
178Espan:
179 error = "directory entry across blocks";
180bad_entry:
181 nilfs_error(sb,
182 "bad entry in directory #%lu: %s - offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
183 dir->i_ino, error, (page->index << PAGE_SHIFT) + offs,
184 (unsigned long)le64_to_cpu(p->inode),
185 rec_len, p->name_len);
186 goto fail;
187Eend:
188 p = (struct nilfs_dir_entry *)(kaddr + offs);
189 nilfs_error(sb,
190 "entry in directory #%lu spans the page boundary offset=%lu, inode=%lu",
191 dir->i_ino, (page->index << PAGE_SHIFT) + offs,
192 (unsigned long)le64_to_cpu(p->inode));
193fail:
194 SetPageError(page);
195 return false;
196}
197
198static struct page *nilfs_get_page(struct inode *dir, unsigned long n)
199{
200 struct address_space *mapping = dir->i_mapping;
201 struct page *page = read_mapping_page(mapping, n, NULL);
202
203 if (!IS_ERR(page)) {
204 kmap(page);
205 if (unlikely(!PageChecked(page))) {
206 if (PageError(page) || !nilfs_check_page(page))
207 goto fail;
208 }
209 }
210 return page;
211
212fail:
213 nilfs_put_page(page);
214 return ERR_PTR(-EIO);
215}
216
217/*
218 * NOTE! unlike strncmp, nilfs_match returns 1 for success, 0 for failure.
219 *
220 * len <= NILFS_NAME_LEN and de != NULL are guaranteed by caller.
221 */
222static int
223nilfs_match(int len, const unsigned char *name, struct nilfs_dir_entry *de)
224{
225 if (len != de->name_len)
226 return 0;
227 if (!de->inode)
228 return 0;
229 return !memcmp(name, de->name, len);
230}
231
232/*
233 * p is at least 6 bytes before the end of page
234 */
235static struct nilfs_dir_entry *nilfs_next_entry(struct nilfs_dir_entry *p)
236{
237 return (struct nilfs_dir_entry *)((char *)p +
238 nilfs_rec_len_from_disk(p->rec_len));
239}
240
241static unsigned char
242nilfs_filetype_table[NILFS_FT_MAX] = {
243 [NILFS_FT_UNKNOWN] = DT_UNKNOWN,
244 [NILFS_FT_REG_FILE] = DT_REG,
245 [NILFS_FT_DIR] = DT_DIR,
246 [NILFS_FT_CHRDEV] = DT_CHR,
247 [NILFS_FT_BLKDEV] = DT_BLK,
248 [NILFS_FT_FIFO] = DT_FIFO,
249 [NILFS_FT_SOCK] = DT_SOCK,
250 [NILFS_FT_SYMLINK] = DT_LNK,
251};
252
253#define S_SHIFT 12
254static unsigned char
255nilfs_type_by_mode[S_IFMT >> S_SHIFT] = {
256 [S_IFREG >> S_SHIFT] = NILFS_FT_REG_FILE,
257 [S_IFDIR >> S_SHIFT] = NILFS_FT_DIR,
258 [S_IFCHR >> S_SHIFT] = NILFS_FT_CHRDEV,
259 [S_IFBLK >> S_SHIFT] = NILFS_FT_BLKDEV,
260 [S_IFIFO >> S_SHIFT] = NILFS_FT_FIFO,
261 [S_IFSOCK >> S_SHIFT] = NILFS_FT_SOCK,
262 [S_IFLNK >> S_SHIFT] = NILFS_FT_SYMLINK,
263};
264
265static void nilfs_set_de_type(struct nilfs_dir_entry *de, struct inode *inode)
266{
267 umode_t mode = inode->i_mode;
268
269 de->file_type = nilfs_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
270}
271
272static int nilfs_readdir(struct file *file, struct dir_context *ctx)
273{
274 loff_t pos = ctx->pos;
275 struct inode *inode = file_inode(file);
276 struct super_block *sb = inode->i_sb;
277 unsigned int offset = pos & ~PAGE_MASK;
278 unsigned long n = pos >> PAGE_SHIFT;
279 unsigned long npages = dir_pages(inode);
280
281 if (pos > inode->i_size - NILFS_DIR_REC_LEN(1))
282 return 0;
283
284 for ( ; n < npages; n++, offset = 0) {
285 char *kaddr, *limit;
286 struct nilfs_dir_entry *de;
287 struct page *page = nilfs_get_page(inode, n);
288
289 if (IS_ERR(page)) {
290 nilfs_error(sb, "bad page in #%lu", inode->i_ino);
291 ctx->pos += PAGE_SIZE - offset;
292 return -EIO;
293 }
294 kaddr = page_address(page);
295 de = (struct nilfs_dir_entry *)(kaddr + offset);
296 limit = kaddr + nilfs_last_byte(inode, n) -
297 NILFS_DIR_REC_LEN(1);
298 for ( ; (char *)de <= limit; de = nilfs_next_entry(de)) {
299 if (de->rec_len == 0) {
300 nilfs_error(sb, "zero-length directory entry");
301 nilfs_put_page(page);
302 return -EIO;
303 }
304 if (de->inode) {
305 unsigned char t;
306
307 if (de->file_type < NILFS_FT_MAX)
308 t = nilfs_filetype_table[de->file_type];
309 else
310 t = DT_UNKNOWN;
311
312 if (!dir_emit(ctx, de->name, de->name_len,
313 le64_to_cpu(de->inode), t)) {
314 nilfs_put_page(page);
315 return 0;
316 }
317 }
318 ctx->pos += nilfs_rec_len_from_disk(de->rec_len);
319 }
320 nilfs_put_page(page);
321 }
322 return 0;
323}
324
325/*
326 * nilfs_find_entry()
327 *
328 * finds an entry in the specified directory with the wanted name. It
329 * returns the page in which the entry was found, and the entry itself
330 * (as a parameter - res_dir). Page is returned mapped and unlocked.
331 * Entry is guaranteed to be valid.
332 */
333struct nilfs_dir_entry *
334nilfs_find_entry(struct inode *dir, const struct qstr *qstr,
335 struct page **res_page)
336{
337 const unsigned char *name = qstr->name;
338 int namelen = qstr->len;
339 unsigned int reclen = NILFS_DIR_REC_LEN(namelen);
340 unsigned long start, n;
341 unsigned long npages = dir_pages(dir);
342 struct page *page = NULL;
343 struct nilfs_inode_info *ei = NILFS_I(dir);
344 struct nilfs_dir_entry *de;
345
346 if (npages == 0)
347 goto out;
348
349 /* OFFSET_CACHE */
350 *res_page = NULL;
351
352 start = ei->i_dir_start_lookup;
353 if (start >= npages)
354 start = 0;
355 n = start;
356 do {
357 char *kaddr;
358
359 page = nilfs_get_page(dir, n);
360 if (!IS_ERR(page)) {
361 kaddr = page_address(page);
362 de = (struct nilfs_dir_entry *)kaddr;
363 kaddr += nilfs_last_byte(dir, n) - reclen;
364 while ((char *) de <= kaddr) {
365 if (de->rec_len == 0) {
366 nilfs_error(dir->i_sb,
367 "zero-length directory entry");
368 nilfs_put_page(page);
369 goto out;
370 }
371 if (nilfs_match(namelen, name, de))
372 goto found;
373 de = nilfs_next_entry(de);
374 }
375 nilfs_put_page(page);
376 }
377 if (++n >= npages)
378 n = 0;
379 /* next page is past the blocks we've got */
380 if (unlikely(n > (dir->i_blocks >> (PAGE_SHIFT - 9)))) {
381 nilfs_error(dir->i_sb,
382 "dir %lu size %lld exceeds block count %llu",
383 dir->i_ino, dir->i_size,
384 (unsigned long long)dir->i_blocks);
385 goto out;
386 }
387 } while (n != start);
388out:
389 return NULL;
390
391found:
392 *res_page = page;
393 ei->i_dir_start_lookup = n;
394 return de;
395}
396
397struct nilfs_dir_entry *nilfs_dotdot(struct inode *dir, struct page **p)
398{
399 struct page *page = nilfs_get_page(dir, 0);
400 struct nilfs_dir_entry *de = NULL;
401
402 if (!IS_ERR(page)) {
403 de = nilfs_next_entry(
404 (struct nilfs_dir_entry *)page_address(page));
405 *p = page;
406 }
407 return de;
408}
409
410ino_t nilfs_inode_by_name(struct inode *dir, const struct qstr *qstr)
411{
412 ino_t res = 0;
413 struct nilfs_dir_entry *de;
414 struct page *page;
415
416 de = nilfs_find_entry(dir, qstr, &page);
417 if (de) {
418 res = le64_to_cpu(de->inode);
419 kunmap(page);
420 put_page(page);
421 }
422 return res;
423}
424
425/* Releases the page */
426void nilfs_set_link(struct inode *dir, struct nilfs_dir_entry *de,
427 struct page *page, struct inode *inode)
428{
429 unsigned int from = (char *)de - (char *)page_address(page);
430 unsigned int to = from + nilfs_rec_len_from_disk(de->rec_len);
431 struct address_space *mapping = page->mapping;
432 int err;
433
434 lock_page(page);
435 err = nilfs_prepare_chunk(page, from, to);
436 BUG_ON(err);
437 de->inode = cpu_to_le64(inode->i_ino);
438 nilfs_set_de_type(de, inode);
439 nilfs_commit_chunk(page, mapping, from, to);
440 nilfs_put_page(page);
441 dir->i_mtime = dir->i_ctime = current_time(dir);
442}
443
444/*
445 * Parent is locked.
446 */
447int nilfs_add_link(struct dentry *dentry, struct inode *inode)
448{
449 struct inode *dir = d_inode(dentry->d_parent);
450 const unsigned char *name = dentry->d_name.name;
451 int namelen = dentry->d_name.len;
452 unsigned int chunk_size = nilfs_chunk_size(dir);
453 unsigned int reclen = NILFS_DIR_REC_LEN(namelen);
454 unsigned short rec_len, name_len;
455 struct page *page = NULL;
456 struct nilfs_dir_entry *de;
457 unsigned long npages = dir_pages(dir);
458 unsigned long n;
459 char *kaddr;
460 unsigned int from, to;
461 int err;
462
463 /*
464 * We take care of directory expansion in the same loop.
465 * This code plays outside i_size, so it locks the page
466 * to protect that region.
467 */
468 for (n = 0; n <= npages; n++) {
469 char *dir_end;
470
471 page = nilfs_get_page(dir, n);
472 err = PTR_ERR(page);
473 if (IS_ERR(page))
474 goto out;
475 lock_page(page);
476 kaddr = page_address(page);
477 dir_end = kaddr + nilfs_last_byte(dir, n);
478 de = (struct nilfs_dir_entry *)kaddr;
479 kaddr += PAGE_SIZE - reclen;
480 while ((char *)de <= kaddr) {
481 if ((char *)de == dir_end) {
482 /* We hit i_size */
483 name_len = 0;
484 rec_len = chunk_size;
485 de->rec_len = nilfs_rec_len_to_disk(chunk_size);
486 de->inode = 0;
487 goto got_it;
488 }
489 if (de->rec_len == 0) {
490 nilfs_error(dir->i_sb,
491 "zero-length directory entry");
492 err = -EIO;
493 goto out_unlock;
494 }
495 err = -EEXIST;
496 if (nilfs_match(namelen, name, de))
497 goto out_unlock;
498 name_len = NILFS_DIR_REC_LEN(de->name_len);
499 rec_len = nilfs_rec_len_from_disk(de->rec_len);
500 if (!de->inode && rec_len >= reclen)
501 goto got_it;
502 if (rec_len >= name_len + reclen)
503 goto got_it;
504 de = (struct nilfs_dir_entry *)((char *)de + rec_len);
505 }
506 unlock_page(page);
507 nilfs_put_page(page);
508 }
509 BUG();
510 return -EINVAL;
511
512got_it:
513 from = (char *)de - (char *)page_address(page);
514 to = from + rec_len;
515 err = nilfs_prepare_chunk(page, from, to);
516 if (err)
517 goto out_unlock;
518 if (de->inode) {
519 struct nilfs_dir_entry *de1;
520
521 de1 = (struct nilfs_dir_entry *)((char *)de + name_len);
522 de1->rec_len = nilfs_rec_len_to_disk(rec_len - name_len);
523 de->rec_len = nilfs_rec_len_to_disk(name_len);
524 de = de1;
525 }
526 de->name_len = namelen;
527 memcpy(de->name, name, namelen);
528 de->inode = cpu_to_le64(inode->i_ino);
529 nilfs_set_de_type(de, inode);
530 nilfs_commit_chunk(page, page->mapping, from, to);
531 dir->i_mtime = dir->i_ctime = current_time(dir);
532 nilfs_mark_inode_dirty(dir);
533 /* OFFSET_CACHE */
534out_put:
535 nilfs_put_page(page);
536out:
537 return err;
538out_unlock:
539 unlock_page(page);
540 goto out_put;
541}
542
543/*
544 * nilfs_delete_entry deletes a directory entry by merging it with the
545 * previous entry. Page is up-to-date. Releases the page.
546 */
547int nilfs_delete_entry(struct nilfs_dir_entry *dir, struct page *page)
548{
549 struct address_space *mapping = page->mapping;
550 struct inode *inode = mapping->host;
551 char *kaddr = page_address(page);
552 unsigned int from, to;
553 struct nilfs_dir_entry *de, *pde = NULL;
554 int err;
555
556 from = ((char *)dir - kaddr) & ~(nilfs_chunk_size(inode) - 1);
557 to = ((char *)dir - kaddr) + nilfs_rec_len_from_disk(dir->rec_len);
558 de = (struct nilfs_dir_entry *)(kaddr + from);
559
560 while ((char *)de < (char *)dir) {
561 if (de->rec_len == 0) {
562 nilfs_error(inode->i_sb,
563 "zero-length directory entry");
564 err = -EIO;
565 goto out;
566 }
567 pde = de;
568 de = nilfs_next_entry(de);
569 }
570 if (pde)
571 from = (char *)pde - (char *)page_address(page);
572 lock_page(page);
573 err = nilfs_prepare_chunk(page, from, to);
574 BUG_ON(err);
575 if (pde)
576 pde->rec_len = nilfs_rec_len_to_disk(to - from);
577 dir->inode = 0;
578 nilfs_commit_chunk(page, mapping, from, to);
579 inode->i_ctime = inode->i_mtime = current_time(inode);
580out:
581 nilfs_put_page(page);
582 return err;
583}
584
585/*
586 * Set the first fragment of directory.
587 */
588int nilfs_make_empty(struct inode *inode, struct inode *parent)
589{
590 struct address_space *mapping = inode->i_mapping;
591 struct page *page = grab_cache_page(mapping, 0);
592 unsigned int chunk_size = nilfs_chunk_size(inode);
593 struct nilfs_dir_entry *de;
594 int err;
595 void *kaddr;
596
597 if (!page)
598 return -ENOMEM;
599
600 err = nilfs_prepare_chunk(page, 0, chunk_size);
601 if (unlikely(err)) {
602 unlock_page(page);
603 goto fail;
604 }
605 kaddr = kmap_atomic(page);
606 memset(kaddr, 0, chunk_size);
607 de = (struct nilfs_dir_entry *)kaddr;
608 de->name_len = 1;
609 de->rec_len = nilfs_rec_len_to_disk(NILFS_DIR_REC_LEN(1));
610 memcpy(de->name, ".\0\0", 4);
611 de->inode = cpu_to_le64(inode->i_ino);
612 nilfs_set_de_type(de, inode);
613
614 de = (struct nilfs_dir_entry *)(kaddr + NILFS_DIR_REC_LEN(1));
615 de->name_len = 2;
616 de->rec_len = nilfs_rec_len_to_disk(chunk_size - NILFS_DIR_REC_LEN(1));
617 de->inode = cpu_to_le64(parent->i_ino);
618 memcpy(de->name, "..\0", 4);
619 nilfs_set_de_type(de, inode);
620 kunmap_atomic(kaddr);
621 nilfs_commit_chunk(page, mapping, 0, chunk_size);
622fail:
623 put_page(page);
624 return err;
625}
626
627/*
628 * routine to check that the specified directory is empty (for rmdir)
629 */
630int nilfs_empty_dir(struct inode *inode)
631{
632 struct page *page = NULL;
633 unsigned long i, npages = dir_pages(inode);
634
635 for (i = 0; i < npages; i++) {
636 char *kaddr;
637 struct nilfs_dir_entry *de;
638
639 page = nilfs_get_page(inode, i);
640 if (IS_ERR(page))
641 continue;
642
643 kaddr = page_address(page);
644 de = (struct nilfs_dir_entry *)kaddr;
645 kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1);
646
647 while ((char *)de <= kaddr) {
648 if (de->rec_len == 0) {
649 nilfs_error(inode->i_sb,
650 "zero-length directory entry (kaddr=%p, de=%p)",
651 kaddr, de);
652 goto not_empty;
653 }
654 if (de->inode != 0) {
655 /* check for . and .. */
656 if (de->name[0] != '.')
657 goto not_empty;
658 if (de->name_len > 2)
659 goto not_empty;
660 if (de->name_len < 2) {
661 if (de->inode !=
662 cpu_to_le64(inode->i_ino))
663 goto not_empty;
664 } else if (de->name[1] != '.')
665 goto not_empty;
666 }
667 de = nilfs_next_entry(de);
668 }
669 nilfs_put_page(page);
670 }
671 return 1;
672
673not_empty:
674 nilfs_put_page(page);
675 return 0;
676}
677
678const struct file_operations nilfs_dir_operations = {
679 .llseek = generic_file_llseek,
680 .read = generic_read_dir,
681 .iterate_shared = nilfs_readdir,
682 .unlocked_ioctl = nilfs_ioctl,
683#ifdef CONFIG_COMPAT
684 .compat_ioctl = nilfs_compat_ioctl,
685#endif /* CONFIG_COMPAT */
686 .fsync = nilfs_sync_file,
687
688};
1/*
2 * dir.c - NILFS directory entry operations
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Modified for NILFS by Amagai Yoshiji <amagai@osrg.net>
21 */
22/*
23 * linux/fs/ext2/dir.c
24 *
25 * Copyright (C) 1992, 1993, 1994, 1995
26 * Remy Card (card@masi.ibp.fr)
27 * Laboratoire MASI - Institut Blaise Pascal
28 * Universite Pierre et Marie Curie (Paris VI)
29 *
30 * from
31 *
32 * linux/fs/minix/dir.c
33 *
34 * Copyright (C) 1991, 1992 Linus Torvalds
35 *
36 * ext2 directory handling functions
37 *
38 * Big-endian to little-endian byte-swapping/bitmaps by
39 * David S. Miller (davem@caip.rutgers.edu), 1995
40 *
41 * All code that works with directory layout had been switched to pagecache
42 * and moved here. AV
43 */
44
45#include <linux/pagemap.h>
46#include "nilfs.h"
47#include "page.h"
48
49/*
50 * nilfs uses block-sized chunks. Arguably, sector-sized ones would be
51 * more robust, but we have what we have
52 */
53static inline unsigned nilfs_chunk_size(struct inode *inode)
54{
55 return inode->i_sb->s_blocksize;
56}
57
58static inline void nilfs_put_page(struct page *page)
59{
60 kunmap(page);
61 page_cache_release(page);
62}
63
64static inline unsigned long dir_pages(struct inode *inode)
65{
66 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
67}
68
69/*
70 * Return the offset into page `page_nr' of the last valid
71 * byte in that page, plus one.
72 */
73static unsigned nilfs_last_byte(struct inode *inode, unsigned long page_nr)
74{
75 unsigned last_byte = inode->i_size;
76
77 last_byte -= page_nr << PAGE_CACHE_SHIFT;
78 if (last_byte > PAGE_CACHE_SIZE)
79 last_byte = PAGE_CACHE_SIZE;
80 return last_byte;
81}
82
83static int nilfs_prepare_chunk(struct page *page, unsigned from, unsigned to)
84{
85 loff_t pos = page_offset(page) + from;
86 return __block_write_begin(page, pos, to - from, nilfs_get_block);
87}
88
89static void nilfs_commit_chunk(struct page *page,
90 struct address_space *mapping,
91 unsigned from, unsigned to)
92{
93 struct inode *dir = mapping->host;
94 loff_t pos = page_offset(page) + from;
95 unsigned len = to - from;
96 unsigned nr_dirty, copied;
97 int err;
98
99 nr_dirty = nilfs_page_count_clean_buffers(page, from, to);
100 copied = block_write_end(NULL, mapping, pos, len, len, page, NULL);
101 if (pos + copied > dir->i_size)
102 i_size_write(dir, pos + copied);
103 if (IS_DIRSYNC(dir))
104 nilfs_set_transaction_flag(NILFS_TI_SYNC);
105 err = nilfs_set_file_dirty(dir, nr_dirty);
106 WARN_ON(err); /* do not happen */
107 unlock_page(page);
108}
109
110static void nilfs_check_page(struct page *page)
111{
112 struct inode *dir = page->mapping->host;
113 struct super_block *sb = dir->i_sb;
114 unsigned chunk_size = nilfs_chunk_size(dir);
115 char *kaddr = page_address(page);
116 unsigned offs, rec_len;
117 unsigned limit = PAGE_CACHE_SIZE;
118 struct nilfs_dir_entry *p;
119 char *error;
120
121 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
122 limit = dir->i_size & ~PAGE_CACHE_MASK;
123 if (limit & (chunk_size - 1))
124 goto Ebadsize;
125 if (!limit)
126 goto out;
127 }
128 for (offs = 0; offs <= limit - NILFS_DIR_REC_LEN(1); offs += rec_len) {
129 p = (struct nilfs_dir_entry *)(kaddr + offs);
130 rec_len = nilfs_rec_len_from_disk(p->rec_len);
131
132 if (rec_len < NILFS_DIR_REC_LEN(1))
133 goto Eshort;
134 if (rec_len & 3)
135 goto Ealign;
136 if (rec_len < NILFS_DIR_REC_LEN(p->name_len))
137 goto Enamelen;
138 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
139 goto Espan;
140 }
141 if (offs != limit)
142 goto Eend;
143out:
144 SetPageChecked(page);
145 return;
146
147 /* Too bad, we had an error */
148
149Ebadsize:
150 nilfs_error(sb, "nilfs_check_page",
151 "size of directory #%lu is not a multiple of chunk size",
152 dir->i_ino
153 );
154 goto fail;
155Eshort:
156 error = "rec_len is smaller than minimal";
157 goto bad_entry;
158Ealign:
159 error = "unaligned directory entry";
160 goto bad_entry;
161Enamelen:
162 error = "rec_len is too small for name_len";
163 goto bad_entry;
164Espan:
165 error = "directory entry across blocks";
166bad_entry:
167 nilfs_error(sb, "nilfs_check_page", "bad entry in directory #%lu: %s - "
168 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
169 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
170 (unsigned long) le64_to_cpu(p->inode),
171 rec_len, p->name_len);
172 goto fail;
173Eend:
174 p = (struct nilfs_dir_entry *)(kaddr + offs);
175 nilfs_error(sb, "nilfs_check_page",
176 "entry in directory #%lu spans the page boundary"
177 "offset=%lu, inode=%lu",
178 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
179 (unsigned long) le64_to_cpu(p->inode));
180fail:
181 SetPageChecked(page);
182 SetPageError(page);
183}
184
185static struct page *nilfs_get_page(struct inode *dir, unsigned long n)
186{
187 struct address_space *mapping = dir->i_mapping;
188 struct page *page = read_mapping_page(mapping, n, NULL);
189
190 if (!IS_ERR(page)) {
191 kmap(page);
192 if (!PageChecked(page))
193 nilfs_check_page(page);
194 if (PageError(page))
195 goto fail;
196 }
197 return page;
198
199fail:
200 nilfs_put_page(page);
201 return ERR_PTR(-EIO);
202}
203
204/*
205 * NOTE! unlike strncmp, nilfs_match returns 1 for success, 0 for failure.
206 *
207 * len <= NILFS_NAME_LEN and de != NULL are guaranteed by caller.
208 */
209static int
210nilfs_match(int len, const unsigned char *name, struct nilfs_dir_entry *de)
211{
212 if (len != de->name_len)
213 return 0;
214 if (!de->inode)
215 return 0;
216 return !memcmp(name, de->name, len);
217}
218
219/*
220 * p is at least 6 bytes before the end of page
221 */
222static struct nilfs_dir_entry *nilfs_next_entry(struct nilfs_dir_entry *p)
223{
224 return (struct nilfs_dir_entry *)((char *)p +
225 nilfs_rec_len_from_disk(p->rec_len));
226}
227
228static unsigned char
229nilfs_filetype_table[NILFS_FT_MAX] = {
230 [NILFS_FT_UNKNOWN] = DT_UNKNOWN,
231 [NILFS_FT_REG_FILE] = DT_REG,
232 [NILFS_FT_DIR] = DT_DIR,
233 [NILFS_FT_CHRDEV] = DT_CHR,
234 [NILFS_FT_BLKDEV] = DT_BLK,
235 [NILFS_FT_FIFO] = DT_FIFO,
236 [NILFS_FT_SOCK] = DT_SOCK,
237 [NILFS_FT_SYMLINK] = DT_LNK,
238};
239
240#define S_SHIFT 12
241static unsigned char
242nilfs_type_by_mode[S_IFMT >> S_SHIFT] = {
243 [S_IFREG >> S_SHIFT] = NILFS_FT_REG_FILE,
244 [S_IFDIR >> S_SHIFT] = NILFS_FT_DIR,
245 [S_IFCHR >> S_SHIFT] = NILFS_FT_CHRDEV,
246 [S_IFBLK >> S_SHIFT] = NILFS_FT_BLKDEV,
247 [S_IFIFO >> S_SHIFT] = NILFS_FT_FIFO,
248 [S_IFSOCK >> S_SHIFT] = NILFS_FT_SOCK,
249 [S_IFLNK >> S_SHIFT] = NILFS_FT_SYMLINK,
250};
251
252static void nilfs_set_de_type(struct nilfs_dir_entry *de, struct inode *inode)
253{
254 mode_t mode = inode->i_mode;
255
256 de->file_type = nilfs_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
257}
258
259static int nilfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
260{
261 loff_t pos = filp->f_pos;
262 struct inode *inode = filp->f_dentry->d_inode;
263 struct super_block *sb = inode->i_sb;
264 unsigned int offset = pos & ~PAGE_CACHE_MASK;
265 unsigned long n = pos >> PAGE_CACHE_SHIFT;
266 unsigned long npages = dir_pages(inode);
267/* unsigned chunk_mask = ~(nilfs_chunk_size(inode)-1); */
268 unsigned char *types = NULL;
269 int ret;
270
271 if (pos > inode->i_size - NILFS_DIR_REC_LEN(1))
272 goto success;
273
274 types = nilfs_filetype_table;
275
276 for ( ; n < npages; n++, offset = 0) {
277 char *kaddr, *limit;
278 struct nilfs_dir_entry *de;
279 struct page *page = nilfs_get_page(inode, n);
280
281 if (IS_ERR(page)) {
282 nilfs_error(sb, __func__, "bad page in #%lu",
283 inode->i_ino);
284 filp->f_pos += PAGE_CACHE_SIZE - offset;
285 ret = -EIO;
286 goto done;
287 }
288 kaddr = page_address(page);
289 de = (struct nilfs_dir_entry *)(kaddr + offset);
290 limit = kaddr + nilfs_last_byte(inode, n) -
291 NILFS_DIR_REC_LEN(1);
292 for ( ; (char *)de <= limit; de = nilfs_next_entry(de)) {
293 if (de->rec_len == 0) {
294 nilfs_error(sb, __func__,
295 "zero-length directory entry");
296 ret = -EIO;
297 nilfs_put_page(page);
298 goto done;
299 }
300 if (de->inode) {
301 int over;
302 unsigned char d_type = DT_UNKNOWN;
303
304 if (types && de->file_type < NILFS_FT_MAX)
305 d_type = types[de->file_type];
306
307 offset = (char *)de - kaddr;
308 over = filldir(dirent, de->name, de->name_len,
309 (n<<PAGE_CACHE_SHIFT) | offset,
310 le64_to_cpu(de->inode), d_type);
311 if (over) {
312 nilfs_put_page(page);
313 goto success;
314 }
315 }
316 filp->f_pos += nilfs_rec_len_from_disk(de->rec_len);
317 }
318 nilfs_put_page(page);
319 }
320
321success:
322 ret = 0;
323done:
324 return ret;
325}
326
327/*
328 * nilfs_find_entry()
329 *
330 * finds an entry in the specified directory with the wanted name. It
331 * returns the page in which the entry was found, and the entry itself
332 * (as a parameter - res_dir). Page is returned mapped and unlocked.
333 * Entry is guaranteed to be valid.
334 */
335struct nilfs_dir_entry *
336nilfs_find_entry(struct inode *dir, const struct qstr *qstr,
337 struct page **res_page)
338{
339 const unsigned char *name = qstr->name;
340 int namelen = qstr->len;
341 unsigned reclen = NILFS_DIR_REC_LEN(namelen);
342 unsigned long start, n;
343 unsigned long npages = dir_pages(dir);
344 struct page *page = NULL;
345 struct nilfs_inode_info *ei = NILFS_I(dir);
346 struct nilfs_dir_entry *de;
347
348 if (npages == 0)
349 goto out;
350
351 /* OFFSET_CACHE */
352 *res_page = NULL;
353
354 start = ei->i_dir_start_lookup;
355 if (start >= npages)
356 start = 0;
357 n = start;
358 do {
359 char *kaddr;
360 page = nilfs_get_page(dir, n);
361 if (!IS_ERR(page)) {
362 kaddr = page_address(page);
363 de = (struct nilfs_dir_entry *)kaddr;
364 kaddr += nilfs_last_byte(dir, n) - reclen;
365 while ((char *) de <= kaddr) {
366 if (de->rec_len == 0) {
367 nilfs_error(dir->i_sb, __func__,
368 "zero-length directory entry");
369 nilfs_put_page(page);
370 goto out;
371 }
372 if (nilfs_match(namelen, name, de))
373 goto found;
374 de = nilfs_next_entry(de);
375 }
376 nilfs_put_page(page);
377 }
378 if (++n >= npages)
379 n = 0;
380 /* next page is past the blocks we've got */
381 if (unlikely(n > (dir->i_blocks >> (PAGE_CACHE_SHIFT - 9)))) {
382 nilfs_error(dir->i_sb, __func__,
383 "dir %lu size %lld exceeds block count %llu",
384 dir->i_ino, dir->i_size,
385 (unsigned long long)dir->i_blocks);
386 goto out;
387 }
388 } while (n != start);
389out:
390 return NULL;
391
392found:
393 *res_page = page;
394 ei->i_dir_start_lookup = n;
395 return de;
396}
397
398struct nilfs_dir_entry *nilfs_dotdot(struct inode *dir, struct page **p)
399{
400 struct page *page = nilfs_get_page(dir, 0);
401 struct nilfs_dir_entry *de = NULL;
402
403 if (!IS_ERR(page)) {
404 de = nilfs_next_entry(
405 (struct nilfs_dir_entry *)page_address(page));
406 *p = page;
407 }
408 return de;
409}
410
411ino_t nilfs_inode_by_name(struct inode *dir, const struct qstr *qstr)
412{
413 ino_t res = 0;
414 struct nilfs_dir_entry *de;
415 struct page *page;
416
417 de = nilfs_find_entry(dir, qstr, &page);
418 if (de) {
419 res = le64_to_cpu(de->inode);
420 kunmap(page);
421 page_cache_release(page);
422 }
423 return res;
424}
425
426/* Releases the page */
427void nilfs_set_link(struct inode *dir, struct nilfs_dir_entry *de,
428 struct page *page, struct inode *inode)
429{
430 unsigned from = (char *) de - (char *) page_address(page);
431 unsigned to = from + nilfs_rec_len_from_disk(de->rec_len);
432 struct address_space *mapping = page->mapping;
433 int err;
434
435 lock_page(page);
436 err = nilfs_prepare_chunk(page, from, to);
437 BUG_ON(err);
438 de->inode = cpu_to_le64(inode->i_ino);
439 nilfs_set_de_type(de, inode);
440 nilfs_commit_chunk(page, mapping, from, to);
441 nilfs_put_page(page);
442 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
443}
444
445/*
446 * Parent is locked.
447 */
448int nilfs_add_link(struct dentry *dentry, struct inode *inode)
449{
450 struct inode *dir = dentry->d_parent->d_inode;
451 const unsigned char *name = dentry->d_name.name;
452 int namelen = dentry->d_name.len;
453 unsigned chunk_size = nilfs_chunk_size(dir);
454 unsigned reclen = NILFS_DIR_REC_LEN(namelen);
455 unsigned short rec_len, name_len;
456 struct page *page = NULL;
457 struct nilfs_dir_entry *de;
458 unsigned long npages = dir_pages(dir);
459 unsigned long n;
460 char *kaddr;
461 unsigned from, to;
462 int err;
463
464 /*
465 * We take care of directory expansion in the same loop.
466 * This code plays outside i_size, so it locks the page
467 * to protect that region.
468 */
469 for (n = 0; n <= npages; n++) {
470 char *dir_end;
471
472 page = nilfs_get_page(dir, n);
473 err = PTR_ERR(page);
474 if (IS_ERR(page))
475 goto out;
476 lock_page(page);
477 kaddr = page_address(page);
478 dir_end = kaddr + nilfs_last_byte(dir, n);
479 de = (struct nilfs_dir_entry *)kaddr;
480 kaddr += PAGE_CACHE_SIZE - reclen;
481 while ((char *)de <= kaddr) {
482 if ((char *)de == dir_end) {
483 /* We hit i_size */
484 name_len = 0;
485 rec_len = chunk_size;
486 de->rec_len = nilfs_rec_len_to_disk(chunk_size);
487 de->inode = 0;
488 goto got_it;
489 }
490 if (de->rec_len == 0) {
491 nilfs_error(dir->i_sb, __func__,
492 "zero-length directory entry");
493 err = -EIO;
494 goto out_unlock;
495 }
496 err = -EEXIST;
497 if (nilfs_match(namelen, name, de))
498 goto out_unlock;
499 name_len = NILFS_DIR_REC_LEN(de->name_len);
500 rec_len = nilfs_rec_len_from_disk(de->rec_len);
501 if (!de->inode && rec_len >= reclen)
502 goto got_it;
503 if (rec_len >= name_len + reclen)
504 goto got_it;
505 de = (struct nilfs_dir_entry *)((char *)de + rec_len);
506 }
507 unlock_page(page);
508 nilfs_put_page(page);
509 }
510 BUG();
511 return -EINVAL;
512
513got_it:
514 from = (char *)de - (char *)page_address(page);
515 to = from + rec_len;
516 err = nilfs_prepare_chunk(page, from, to);
517 if (err)
518 goto out_unlock;
519 if (de->inode) {
520 struct nilfs_dir_entry *de1;
521
522 de1 = (struct nilfs_dir_entry *)((char *)de + name_len);
523 de1->rec_len = nilfs_rec_len_to_disk(rec_len - name_len);
524 de->rec_len = nilfs_rec_len_to_disk(name_len);
525 de = de1;
526 }
527 de->name_len = namelen;
528 memcpy(de->name, name, namelen);
529 de->inode = cpu_to_le64(inode->i_ino);
530 nilfs_set_de_type(de, inode);
531 nilfs_commit_chunk(page, page->mapping, from, to);
532 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
533 nilfs_mark_inode_dirty(dir);
534 /* OFFSET_CACHE */
535out_put:
536 nilfs_put_page(page);
537out:
538 return err;
539out_unlock:
540 unlock_page(page);
541 goto out_put;
542}
543
544/*
545 * nilfs_delete_entry deletes a directory entry by merging it with the
546 * previous entry. Page is up-to-date. Releases the page.
547 */
548int nilfs_delete_entry(struct nilfs_dir_entry *dir, struct page *page)
549{
550 struct address_space *mapping = page->mapping;
551 struct inode *inode = mapping->host;
552 char *kaddr = page_address(page);
553 unsigned from = ((char *)dir - kaddr) & ~(nilfs_chunk_size(inode) - 1);
554 unsigned to = ((char *)dir - kaddr) +
555 nilfs_rec_len_from_disk(dir->rec_len);
556 struct nilfs_dir_entry *pde = NULL;
557 struct nilfs_dir_entry *de = (struct nilfs_dir_entry *)(kaddr + from);
558 int err;
559
560 while ((char *)de < (char *)dir) {
561 if (de->rec_len == 0) {
562 nilfs_error(inode->i_sb, __func__,
563 "zero-length directory entry");
564 err = -EIO;
565 goto out;
566 }
567 pde = de;
568 de = nilfs_next_entry(de);
569 }
570 if (pde)
571 from = (char *)pde - (char *)page_address(page);
572 lock_page(page);
573 err = nilfs_prepare_chunk(page, from, to);
574 BUG_ON(err);
575 if (pde)
576 pde->rec_len = nilfs_rec_len_to_disk(to - from);
577 dir->inode = 0;
578 nilfs_commit_chunk(page, mapping, from, to);
579 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
580out:
581 nilfs_put_page(page);
582 return err;
583}
584
585/*
586 * Set the first fragment of directory.
587 */
588int nilfs_make_empty(struct inode *inode, struct inode *parent)
589{
590 struct address_space *mapping = inode->i_mapping;
591 struct page *page = grab_cache_page(mapping, 0);
592 unsigned chunk_size = nilfs_chunk_size(inode);
593 struct nilfs_dir_entry *de;
594 int err;
595 void *kaddr;
596
597 if (!page)
598 return -ENOMEM;
599
600 err = nilfs_prepare_chunk(page, 0, chunk_size);
601 if (unlikely(err)) {
602 unlock_page(page);
603 goto fail;
604 }
605 kaddr = kmap_atomic(page, KM_USER0);
606 memset(kaddr, 0, chunk_size);
607 de = (struct nilfs_dir_entry *)kaddr;
608 de->name_len = 1;
609 de->rec_len = nilfs_rec_len_to_disk(NILFS_DIR_REC_LEN(1));
610 memcpy(de->name, ".\0\0", 4);
611 de->inode = cpu_to_le64(inode->i_ino);
612 nilfs_set_de_type(de, inode);
613
614 de = (struct nilfs_dir_entry *)(kaddr + NILFS_DIR_REC_LEN(1));
615 de->name_len = 2;
616 de->rec_len = nilfs_rec_len_to_disk(chunk_size - NILFS_DIR_REC_LEN(1));
617 de->inode = cpu_to_le64(parent->i_ino);
618 memcpy(de->name, "..\0", 4);
619 nilfs_set_de_type(de, inode);
620 kunmap_atomic(kaddr, KM_USER0);
621 nilfs_commit_chunk(page, mapping, 0, chunk_size);
622fail:
623 page_cache_release(page);
624 return err;
625}
626
627/*
628 * routine to check that the specified directory is empty (for rmdir)
629 */
630int nilfs_empty_dir(struct inode *inode)
631{
632 struct page *page = NULL;
633 unsigned long i, npages = dir_pages(inode);
634
635 for (i = 0; i < npages; i++) {
636 char *kaddr;
637 struct nilfs_dir_entry *de;
638
639 page = nilfs_get_page(inode, i);
640 if (IS_ERR(page))
641 continue;
642
643 kaddr = page_address(page);
644 de = (struct nilfs_dir_entry *)kaddr;
645 kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1);
646
647 while ((char *)de <= kaddr) {
648 if (de->rec_len == 0) {
649 nilfs_error(inode->i_sb, __func__,
650 "zero-length directory entry "
651 "(kaddr=%p, de=%p)\n", kaddr, de);
652 goto not_empty;
653 }
654 if (de->inode != 0) {
655 /* check for . and .. */
656 if (de->name[0] != '.')
657 goto not_empty;
658 if (de->name_len > 2)
659 goto not_empty;
660 if (de->name_len < 2) {
661 if (de->inode !=
662 cpu_to_le64(inode->i_ino))
663 goto not_empty;
664 } else if (de->name[1] != '.')
665 goto not_empty;
666 }
667 de = nilfs_next_entry(de);
668 }
669 nilfs_put_page(page);
670 }
671 return 1;
672
673not_empty:
674 nilfs_put_page(page);
675 return 0;
676}
677
678const struct file_operations nilfs_dir_operations = {
679 .llseek = generic_file_llseek,
680 .read = generic_read_dir,
681 .readdir = nilfs_readdir,
682 .unlocked_ioctl = nilfs_ioctl,
683#ifdef CONFIG_COMPAT
684 .compat_ioctl = nilfs_compat_ioctl,
685#endif /* CONFIG_COMPAT */
686 .fsync = nilfs_sync_file,
687
688};