Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * NILFS directory entry operations
4 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Modified for NILFS by Amagai Yoshiji.
8 */
9/*
10 * linux/fs/ext2/dir.c
11 *
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise Pascal
15 * Universite Pierre et Marie Curie (Paris VI)
16 *
17 * from
18 *
19 * linux/fs/minix/dir.c
20 *
21 * Copyright (C) 1991, 1992 Linus Torvalds
22 *
23 * ext2 directory handling functions
24 *
25 * Big-endian to little-endian byte-swapping/bitmaps by
26 * David S. Miller (davem@caip.rutgers.edu), 1995
27 *
28 * All code that works with directory layout had been switched to pagecache
29 * and moved here. AV
30 */
31
32#include <linux/pagemap.h>
33#include "nilfs.h"
34#include "page.h"
35
36static inline unsigned int nilfs_rec_len_from_disk(__le16 dlen)
37{
38 unsigned int len = le16_to_cpu(dlen);
39
40#if (PAGE_SIZE >= 65536)
41 if (len == NILFS_MAX_REC_LEN)
42 return 1 << 16;
43#endif
44 return len;
45}
46
47static inline __le16 nilfs_rec_len_to_disk(unsigned int len)
48{
49#if (PAGE_SIZE >= 65536)
50 if (len == (1 << 16))
51 return cpu_to_le16(NILFS_MAX_REC_LEN);
52
53 BUG_ON(len > (1 << 16));
54#endif
55 return cpu_to_le16(len);
56}
57
58/*
59 * nilfs uses block-sized chunks. Arguably, sector-sized ones would be
60 * more robust, but we have what we have
61 */
62static inline unsigned int nilfs_chunk_size(struct inode *inode)
63{
64 return inode->i_sb->s_blocksize;
65}
66
67/*
68 * Return the offset into page `page_nr' of the last valid
69 * byte in that page, plus one.
70 */
71static unsigned int nilfs_last_byte(struct inode *inode, unsigned long page_nr)
72{
73 unsigned int last_byte = inode->i_size;
74
75 last_byte -= page_nr << PAGE_SHIFT;
76 if (last_byte > PAGE_SIZE)
77 last_byte = PAGE_SIZE;
78 return last_byte;
79}
80
81static int nilfs_prepare_chunk(struct folio *folio, unsigned int from,
82 unsigned int to)
83{
84 loff_t pos = folio_pos(folio) + from;
85
86 return __block_write_begin(&folio->page, pos, to - from, nilfs_get_block);
87}
88
89static void nilfs_commit_chunk(struct folio *folio,
90 struct address_space *mapping, size_t from, size_t to)
91{
92 struct inode *dir = mapping->host;
93 loff_t pos = folio_pos(folio) + from;
94 size_t copied, len = to - from;
95 unsigned int nr_dirty;
96 int err;
97
98 nr_dirty = nilfs_page_count_clean_buffers(&folio->page, from, to);
99 copied = block_write_end(NULL, mapping, pos, len, len, &folio->page, NULL);
100 if (pos + copied > dir->i_size)
101 i_size_write(dir, pos + copied);
102 if (IS_DIRSYNC(dir))
103 nilfs_set_transaction_flag(NILFS_TI_SYNC);
104 err = nilfs_set_file_dirty(dir, nr_dirty);
105 WARN_ON(err); /* do not happen */
106 folio_unlock(folio);
107}
108
109static bool nilfs_check_folio(struct folio *folio, char *kaddr)
110{
111 struct inode *dir = folio->mapping->host;
112 struct super_block *sb = dir->i_sb;
113 unsigned int chunk_size = nilfs_chunk_size(dir);
114 size_t offs, rec_len;
115 size_t limit = folio_size(folio);
116 struct nilfs_dir_entry *p;
117 char *error;
118
119 if (dir->i_size < folio_pos(folio) + limit) {
120 limit = dir->i_size - folio_pos(folio);
121 if (limit & (chunk_size - 1))
122 goto Ebadsize;
123 if (!limit)
124 goto out;
125 }
126 for (offs = 0; offs <= limit - NILFS_DIR_REC_LEN(1); offs += rec_len) {
127 p = (struct nilfs_dir_entry *)(kaddr + offs);
128 rec_len = nilfs_rec_len_from_disk(p->rec_len);
129
130 if (rec_len < NILFS_DIR_REC_LEN(1))
131 goto Eshort;
132 if (rec_len & 3)
133 goto Ealign;
134 if (rec_len < NILFS_DIR_REC_LEN(p->name_len))
135 goto Enamelen;
136 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
137 goto Espan;
138 }
139 if (offs != limit)
140 goto Eend;
141out:
142 folio_set_checked(folio);
143 return true;
144
145 /* Too bad, we had an error */
146
147Ebadsize:
148 nilfs_error(sb,
149 "size of directory #%lu is not a multiple of chunk size",
150 dir->i_ino);
151 goto fail;
152Eshort:
153 error = "rec_len is smaller than minimal";
154 goto bad_entry;
155Ealign:
156 error = "unaligned directory entry";
157 goto bad_entry;
158Enamelen:
159 error = "rec_len is too small for name_len";
160 goto bad_entry;
161Espan:
162 error = "directory entry across blocks";
163bad_entry:
164 nilfs_error(sb,
165 "bad entry in directory #%lu: %s - offset=%lu, inode=%lu, rec_len=%zd, name_len=%d",
166 dir->i_ino, error, (folio->index << PAGE_SHIFT) + offs,
167 (unsigned long)le64_to_cpu(p->inode),
168 rec_len, p->name_len);
169 goto fail;
170Eend:
171 p = (struct nilfs_dir_entry *)(kaddr + offs);
172 nilfs_error(sb,
173 "entry in directory #%lu spans the page boundary offset=%lu, inode=%lu",
174 dir->i_ino, (folio->index << PAGE_SHIFT) + offs,
175 (unsigned long)le64_to_cpu(p->inode));
176fail:
177 folio_set_error(folio);
178 return false;
179}
180
181static void *nilfs_get_folio(struct inode *dir, unsigned long n,
182 struct folio **foliop)
183{
184 struct address_space *mapping = dir->i_mapping;
185 struct folio *folio = read_mapping_folio(mapping, n, NULL);
186 void *kaddr;
187
188 if (IS_ERR(folio))
189 return folio;
190
191 kaddr = kmap_local_folio(folio, 0);
192 if (unlikely(!folio_test_checked(folio))) {
193 if (!nilfs_check_folio(folio, kaddr))
194 goto fail;
195 }
196
197 *foliop = folio;
198 return kaddr;
199
200fail:
201 folio_release_kmap(folio, kaddr);
202 return ERR_PTR(-EIO);
203}
204
205/*
206 * NOTE! unlike strncmp, nilfs_match returns 1 for success, 0 for failure.
207 *
208 * len <= NILFS_NAME_LEN and de != NULL are guaranteed by caller.
209 */
210static int
211nilfs_match(int len, const unsigned char *name, struct nilfs_dir_entry *de)
212{
213 if (len != de->name_len)
214 return 0;
215 if (!de->inode)
216 return 0;
217 return !memcmp(name, de->name, len);
218}
219
220/*
221 * p is at least 6 bytes before the end of page
222 */
223static struct nilfs_dir_entry *nilfs_next_entry(struct nilfs_dir_entry *p)
224{
225 return (struct nilfs_dir_entry *)((char *)p +
226 nilfs_rec_len_from_disk(p->rec_len));
227}
228
229static unsigned char
230nilfs_filetype_table[NILFS_FT_MAX] = {
231 [NILFS_FT_UNKNOWN] = DT_UNKNOWN,
232 [NILFS_FT_REG_FILE] = DT_REG,
233 [NILFS_FT_DIR] = DT_DIR,
234 [NILFS_FT_CHRDEV] = DT_CHR,
235 [NILFS_FT_BLKDEV] = DT_BLK,
236 [NILFS_FT_FIFO] = DT_FIFO,
237 [NILFS_FT_SOCK] = DT_SOCK,
238 [NILFS_FT_SYMLINK] = DT_LNK,
239};
240
241#define S_SHIFT 12
242static unsigned char
243nilfs_type_by_mode[S_IFMT >> S_SHIFT] = {
244 [S_IFREG >> S_SHIFT] = NILFS_FT_REG_FILE,
245 [S_IFDIR >> S_SHIFT] = NILFS_FT_DIR,
246 [S_IFCHR >> S_SHIFT] = NILFS_FT_CHRDEV,
247 [S_IFBLK >> S_SHIFT] = NILFS_FT_BLKDEV,
248 [S_IFIFO >> S_SHIFT] = NILFS_FT_FIFO,
249 [S_IFSOCK >> S_SHIFT] = NILFS_FT_SOCK,
250 [S_IFLNK >> S_SHIFT] = NILFS_FT_SYMLINK,
251};
252
253static void nilfs_set_de_type(struct nilfs_dir_entry *de, struct inode *inode)
254{
255 umode_t mode = inode->i_mode;
256
257 de->file_type = nilfs_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
258}
259
260static int nilfs_readdir(struct file *file, struct dir_context *ctx)
261{
262 loff_t pos = ctx->pos;
263 struct inode *inode = file_inode(file);
264 struct super_block *sb = inode->i_sb;
265 unsigned int offset = pos & ~PAGE_MASK;
266 unsigned long n = pos >> PAGE_SHIFT;
267 unsigned long npages = dir_pages(inode);
268
269 if (pos > inode->i_size - NILFS_DIR_REC_LEN(1))
270 return 0;
271
272 for ( ; n < npages; n++, offset = 0) {
273 char *kaddr, *limit;
274 struct nilfs_dir_entry *de;
275 struct folio *folio;
276
277 kaddr = nilfs_get_folio(inode, n, &folio);
278 if (IS_ERR(kaddr)) {
279 nilfs_error(sb, "bad page in #%lu", inode->i_ino);
280 ctx->pos += PAGE_SIZE - offset;
281 return -EIO;
282 }
283 de = (struct nilfs_dir_entry *)(kaddr + offset);
284 limit = kaddr + nilfs_last_byte(inode, n) -
285 NILFS_DIR_REC_LEN(1);
286 for ( ; (char *)de <= limit; de = nilfs_next_entry(de)) {
287 if (de->rec_len == 0) {
288 nilfs_error(sb, "zero-length directory entry");
289 folio_release_kmap(folio, kaddr);
290 return -EIO;
291 }
292 if (de->inode) {
293 unsigned char t;
294
295 if (de->file_type < NILFS_FT_MAX)
296 t = nilfs_filetype_table[de->file_type];
297 else
298 t = DT_UNKNOWN;
299
300 if (!dir_emit(ctx, de->name, de->name_len,
301 le64_to_cpu(de->inode), t)) {
302 folio_release_kmap(folio, kaddr);
303 return 0;
304 }
305 }
306 ctx->pos += nilfs_rec_len_from_disk(de->rec_len);
307 }
308 folio_release_kmap(folio, kaddr);
309 }
310 return 0;
311}
312
313/*
314 * nilfs_find_entry()
315 *
316 * Finds an entry in the specified directory with the wanted name. It
317 * returns the folio in which the entry was found, and the entry itself.
318 * The folio is mapped and unlocked. When the caller is finished with
319 * the entry, it should call folio_release_kmap().
320 *
321 * On failure, returns NULL and the caller should ignore foliop.
322 */
323struct nilfs_dir_entry *nilfs_find_entry(struct inode *dir,
324 const struct qstr *qstr, struct folio **foliop)
325{
326 const unsigned char *name = qstr->name;
327 int namelen = qstr->len;
328 unsigned int reclen = NILFS_DIR_REC_LEN(namelen);
329 unsigned long start, n;
330 unsigned long npages = dir_pages(dir);
331 struct nilfs_inode_info *ei = NILFS_I(dir);
332 struct nilfs_dir_entry *de;
333
334 if (npages == 0)
335 goto out;
336
337 start = ei->i_dir_start_lookup;
338 if (start >= npages)
339 start = 0;
340 n = start;
341 do {
342 char *kaddr = nilfs_get_folio(dir, n, foliop);
343
344 if (!IS_ERR(kaddr)) {
345 de = (struct nilfs_dir_entry *)kaddr;
346 kaddr += nilfs_last_byte(dir, n) - reclen;
347 while ((char *) de <= kaddr) {
348 if (de->rec_len == 0) {
349 nilfs_error(dir->i_sb,
350 "zero-length directory entry");
351 folio_release_kmap(*foliop, kaddr);
352 goto out;
353 }
354 if (nilfs_match(namelen, name, de))
355 goto found;
356 de = nilfs_next_entry(de);
357 }
358 folio_release_kmap(*foliop, kaddr);
359 }
360 if (++n >= npages)
361 n = 0;
362 /* next folio is past the blocks we've got */
363 if (unlikely(n > (dir->i_blocks >> (PAGE_SHIFT - 9)))) {
364 nilfs_error(dir->i_sb,
365 "dir %lu size %lld exceeds block count %llu",
366 dir->i_ino, dir->i_size,
367 (unsigned long long)dir->i_blocks);
368 goto out;
369 }
370 } while (n != start);
371out:
372 return NULL;
373
374found:
375 ei->i_dir_start_lookup = n;
376 return de;
377}
378
379struct nilfs_dir_entry *nilfs_dotdot(struct inode *dir, struct folio **foliop)
380{
381 struct nilfs_dir_entry *de = nilfs_get_folio(dir, 0, foliop);
382
383 if (IS_ERR(de))
384 return NULL;
385 return nilfs_next_entry(de);
386}
387
388ino_t nilfs_inode_by_name(struct inode *dir, const struct qstr *qstr)
389{
390 ino_t res = 0;
391 struct nilfs_dir_entry *de;
392 struct folio *folio;
393
394 de = nilfs_find_entry(dir, qstr, &folio);
395 if (de) {
396 res = le64_to_cpu(de->inode);
397 folio_release_kmap(folio, de);
398 }
399 return res;
400}
401
402void nilfs_set_link(struct inode *dir, struct nilfs_dir_entry *de,
403 struct folio *folio, struct inode *inode)
404{
405 size_t from = offset_in_folio(folio, de);
406 size_t to = from + nilfs_rec_len_from_disk(de->rec_len);
407 struct address_space *mapping = folio->mapping;
408 int err;
409
410 folio_lock(folio);
411 err = nilfs_prepare_chunk(folio, from, to);
412 BUG_ON(err);
413 de->inode = cpu_to_le64(inode->i_ino);
414 nilfs_set_de_type(de, inode);
415 nilfs_commit_chunk(folio, mapping, from, to);
416 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
417}
418
419/*
420 * Parent is locked.
421 */
422int nilfs_add_link(struct dentry *dentry, struct inode *inode)
423{
424 struct inode *dir = d_inode(dentry->d_parent);
425 const unsigned char *name = dentry->d_name.name;
426 int namelen = dentry->d_name.len;
427 unsigned int chunk_size = nilfs_chunk_size(dir);
428 unsigned int reclen = NILFS_DIR_REC_LEN(namelen);
429 unsigned short rec_len, name_len;
430 struct folio *folio = NULL;
431 struct nilfs_dir_entry *de;
432 unsigned long npages = dir_pages(dir);
433 unsigned long n;
434 size_t from, to;
435 int err;
436
437 /*
438 * We take care of directory expansion in the same loop.
439 * This code plays outside i_size, so it locks the folio
440 * to protect that region.
441 */
442 for (n = 0; n <= npages; n++) {
443 char *kaddr = nilfs_get_folio(dir, n, &folio);
444 char *dir_end;
445
446 if (IS_ERR(kaddr))
447 return PTR_ERR(kaddr);
448 folio_lock(folio);
449 dir_end = kaddr + nilfs_last_byte(dir, n);
450 de = (struct nilfs_dir_entry *)kaddr;
451 kaddr += folio_size(folio) - reclen;
452 while ((char *)de <= kaddr) {
453 if ((char *)de == dir_end) {
454 /* We hit i_size */
455 name_len = 0;
456 rec_len = chunk_size;
457 de->rec_len = nilfs_rec_len_to_disk(chunk_size);
458 de->inode = 0;
459 goto got_it;
460 }
461 if (de->rec_len == 0) {
462 nilfs_error(dir->i_sb,
463 "zero-length directory entry");
464 err = -EIO;
465 goto out_unlock;
466 }
467 err = -EEXIST;
468 if (nilfs_match(namelen, name, de))
469 goto out_unlock;
470 name_len = NILFS_DIR_REC_LEN(de->name_len);
471 rec_len = nilfs_rec_len_from_disk(de->rec_len);
472 if (!de->inode && rec_len >= reclen)
473 goto got_it;
474 if (rec_len >= name_len + reclen)
475 goto got_it;
476 de = (struct nilfs_dir_entry *)((char *)de + rec_len);
477 }
478 folio_unlock(folio);
479 folio_release_kmap(folio, kaddr);
480 }
481 BUG();
482 return -EINVAL;
483
484got_it:
485 from = offset_in_folio(folio, de);
486 to = from + rec_len;
487 err = nilfs_prepare_chunk(folio, from, to);
488 if (err)
489 goto out_unlock;
490 if (de->inode) {
491 struct nilfs_dir_entry *de1;
492
493 de1 = (struct nilfs_dir_entry *)((char *)de + name_len);
494 de1->rec_len = nilfs_rec_len_to_disk(rec_len - name_len);
495 de->rec_len = nilfs_rec_len_to_disk(name_len);
496 de = de1;
497 }
498 de->name_len = namelen;
499 memcpy(de->name, name, namelen);
500 de->inode = cpu_to_le64(inode->i_ino);
501 nilfs_set_de_type(de, inode);
502 nilfs_commit_chunk(folio, folio->mapping, from, to);
503 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
504 nilfs_mark_inode_dirty(dir);
505 /* OFFSET_CACHE */
506out_put:
507 folio_release_kmap(folio, de);
508 return err;
509out_unlock:
510 folio_unlock(folio);
511 goto out_put;
512}
513
514/*
515 * nilfs_delete_entry deletes a directory entry by merging it with the
516 * previous entry. Folio is up-to-date.
517 */
518int nilfs_delete_entry(struct nilfs_dir_entry *dir, struct folio *folio)
519{
520 struct address_space *mapping = folio->mapping;
521 struct inode *inode = mapping->host;
522 char *kaddr = (char *)((unsigned long)dir & ~(folio_size(folio) - 1));
523 size_t from, to;
524 struct nilfs_dir_entry *de, *pde = NULL;
525 int err;
526
527 from = ((char *)dir - kaddr) & ~(nilfs_chunk_size(inode) - 1);
528 to = ((char *)dir - kaddr) + nilfs_rec_len_from_disk(dir->rec_len);
529 de = (struct nilfs_dir_entry *)(kaddr + from);
530
531 while ((char *)de < (char *)dir) {
532 if (de->rec_len == 0) {
533 nilfs_error(inode->i_sb,
534 "zero-length directory entry");
535 err = -EIO;
536 goto out;
537 }
538 pde = de;
539 de = nilfs_next_entry(de);
540 }
541 if (pde)
542 from = (char *)pde - kaddr;
543 folio_lock(folio);
544 err = nilfs_prepare_chunk(folio, from, to);
545 BUG_ON(err);
546 if (pde)
547 pde->rec_len = nilfs_rec_len_to_disk(to - from);
548 dir->inode = 0;
549 nilfs_commit_chunk(folio, mapping, from, to);
550 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
551out:
552 return err;
553}
554
555/*
556 * Set the first fragment of directory.
557 */
558int nilfs_make_empty(struct inode *inode, struct inode *parent)
559{
560 struct address_space *mapping = inode->i_mapping;
561 struct folio *folio = filemap_grab_folio(mapping, 0);
562 unsigned int chunk_size = nilfs_chunk_size(inode);
563 struct nilfs_dir_entry *de;
564 int err;
565 void *kaddr;
566
567 if (IS_ERR(folio))
568 return PTR_ERR(folio);
569
570 err = nilfs_prepare_chunk(folio, 0, chunk_size);
571 if (unlikely(err)) {
572 folio_unlock(folio);
573 goto fail;
574 }
575 kaddr = kmap_local_folio(folio, 0);
576 memset(kaddr, 0, chunk_size);
577 de = (struct nilfs_dir_entry *)kaddr;
578 de->name_len = 1;
579 de->rec_len = nilfs_rec_len_to_disk(NILFS_DIR_REC_LEN(1));
580 memcpy(de->name, ".\0\0", 4);
581 de->inode = cpu_to_le64(inode->i_ino);
582 nilfs_set_de_type(de, inode);
583
584 de = (struct nilfs_dir_entry *)(kaddr + NILFS_DIR_REC_LEN(1));
585 de->name_len = 2;
586 de->rec_len = nilfs_rec_len_to_disk(chunk_size - NILFS_DIR_REC_LEN(1));
587 de->inode = cpu_to_le64(parent->i_ino);
588 memcpy(de->name, "..\0", 4);
589 nilfs_set_de_type(de, inode);
590 kunmap_local(kaddr);
591 nilfs_commit_chunk(folio, mapping, 0, chunk_size);
592fail:
593 folio_put(folio);
594 return err;
595}
596
597/*
598 * routine to check that the specified directory is empty (for rmdir)
599 */
600int nilfs_empty_dir(struct inode *inode)
601{
602 struct folio *folio = NULL;
603 char *kaddr;
604 unsigned long i, npages = dir_pages(inode);
605
606 for (i = 0; i < npages; i++) {
607 struct nilfs_dir_entry *de;
608
609 kaddr = nilfs_get_folio(inode, i, &folio);
610 if (IS_ERR(kaddr))
611 continue;
612
613 de = (struct nilfs_dir_entry *)kaddr;
614 kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1);
615
616 while ((char *)de <= kaddr) {
617 if (de->rec_len == 0) {
618 nilfs_error(inode->i_sb,
619 "zero-length directory entry (kaddr=%p, de=%p)",
620 kaddr, de);
621 goto not_empty;
622 }
623 if (de->inode != 0) {
624 /* check for . and .. */
625 if (de->name[0] != '.')
626 goto not_empty;
627 if (de->name_len > 2)
628 goto not_empty;
629 if (de->name_len < 2) {
630 if (de->inode !=
631 cpu_to_le64(inode->i_ino))
632 goto not_empty;
633 } else if (de->name[1] != '.')
634 goto not_empty;
635 }
636 de = nilfs_next_entry(de);
637 }
638 folio_release_kmap(folio, kaddr);
639 }
640 return 1;
641
642not_empty:
643 folio_release_kmap(folio, kaddr);
644 return 0;
645}
646
647const struct file_operations nilfs_dir_operations = {
648 .llseek = generic_file_llseek,
649 .read = generic_read_dir,
650 .iterate_shared = nilfs_readdir,
651 .unlocked_ioctl = nilfs_ioctl,
652#ifdef CONFIG_COMPAT
653 .compat_ioctl = nilfs_compat_ioctl,
654#endif /* CONFIG_COMPAT */
655 .fsync = nilfs_sync_file,
656
657};
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};