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// 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 u64 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, 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, from, to);
99 copied = block_write_end(NULL, mapping, pos, len, len, folio, 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 if (unlikely(p->inode &&
139 NILFS_PRIVATE_INODE(le64_to_cpu(p->inode))))
140 goto Einumber;
141 }
142 if (offs != limit)
143 goto Eend;
144out:
145 folio_set_checked(folio);
146 return true;
147
148 /* Too bad, we had an error */
149
150Ebadsize:
151 nilfs_error(sb,
152 "size of directory #%lu is not a multiple of chunk size",
153 dir->i_ino);
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";
166 goto bad_entry;
167Einumber:
168 error = "disallowed inode number";
169bad_entry:
170 nilfs_error(sb,
171 "bad entry in directory #%lu: %s - offset=%lu, inode=%lu, rec_len=%zd, name_len=%d",
172 dir->i_ino, error, (folio->index << PAGE_SHIFT) + offs,
173 (unsigned long)le64_to_cpu(p->inode),
174 rec_len, p->name_len);
175 goto fail;
176Eend:
177 p = (struct nilfs_dir_entry *)(kaddr + offs);
178 nilfs_error(sb,
179 "entry in directory #%lu spans the page boundary offset=%lu, inode=%lu",
180 dir->i_ino, (folio->index << PAGE_SHIFT) + offs,
181 (unsigned long)le64_to_cpu(p->inode));
182fail:
183 return false;
184}
185
186static void *nilfs_get_folio(struct inode *dir, unsigned long n,
187 struct folio **foliop)
188{
189 struct address_space *mapping = dir->i_mapping;
190 struct folio *folio = read_mapping_folio(mapping, n, NULL);
191 void *kaddr;
192
193 if (IS_ERR(folio))
194 return folio;
195
196 kaddr = kmap_local_folio(folio, 0);
197 if (unlikely(!folio_test_checked(folio))) {
198 if (!nilfs_check_folio(folio, kaddr))
199 goto fail;
200 }
201
202 *foliop = folio;
203 return kaddr;
204
205fail:
206 folio_release_kmap(folio, kaddr);
207 return ERR_PTR(-EIO);
208}
209
210/*
211 * NOTE! unlike strncmp, nilfs_match returns 1 for success, 0 for failure.
212 *
213 * len <= NILFS_NAME_LEN and de != NULL are guaranteed by caller.
214 */
215static int
216nilfs_match(int len, const unsigned char *name, struct nilfs_dir_entry *de)
217{
218 if (len != de->name_len)
219 return 0;
220 if (!de->inode)
221 return 0;
222 return !memcmp(name, de->name, len);
223}
224
225/*
226 * p is at least 6 bytes before the end of page
227 */
228static struct nilfs_dir_entry *nilfs_next_entry(struct nilfs_dir_entry *p)
229{
230 return (struct nilfs_dir_entry *)((char *)p +
231 nilfs_rec_len_from_disk(p->rec_len));
232}
233
234static int nilfs_readdir(struct file *file, struct dir_context *ctx)
235{
236 loff_t pos = ctx->pos;
237 struct inode *inode = file_inode(file);
238 struct super_block *sb = inode->i_sb;
239 unsigned int offset = pos & ~PAGE_MASK;
240 unsigned long n = pos >> PAGE_SHIFT;
241 unsigned long npages = dir_pages(inode);
242
243 if (pos > inode->i_size - NILFS_DIR_REC_LEN(1))
244 return 0;
245
246 for ( ; n < npages; n++, offset = 0) {
247 char *kaddr, *limit;
248 struct nilfs_dir_entry *de;
249 struct folio *folio;
250
251 kaddr = nilfs_get_folio(inode, n, &folio);
252 if (IS_ERR(kaddr)) {
253 nilfs_error(sb, "bad page in #%lu", inode->i_ino);
254 ctx->pos += PAGE_SIZE - offset;
255 return -EIO;
256 }
257 de = (struct nilfs_dir_entry *)(kaddr + offset);
258 limit = kaddr + nilfs_last_byte(inode, n) -
259 NILFS_DIR_REC_LEN(1);
260 for ( ; (char *)de <= limit; de = nilfs_next_entry(de)) {
261 if (de->rec_len == 0) {
262 nilfs_error(sb, "zero-length directory entry");
263 folio_release_kmap(folio, kaddr);
264 return -EIO;
265 }
266 if (de->inode) {
267 unsigned char t;
268
269 t = fs_ftype_to_dtype(de->file_type);
270
271 if (!dir_emit(ctx, de->name, de->name_len,
272 le64_to_cpu(de->inode), t)) {
273 folio_release_kmap(folio, kaddr);
274 return 0;
275 }
276 }
277 ctx->pos += nilfs_rec_len_from_disk(de->rec_len);
278 }
279 folio_release_kmap(folio, kaddr);
280 }
281 return 0;
282}
283
284/*
285 * nilfs_find_entry()
286 *
287 * Finds an entry in the specified directory with the wanted name. It
288 * returns the folio in which the entry was found, and the entry itself.
289 * The folio is mapped and unlocked. When the caller is finished with
290 * the entry, it should call folio_release_kmap().
291 *
292 * On failure, returns an error pointer and the caller should ignore foliop.
293 */
294struct nilfs_dir_entry *nilfs_find_entry(struct inode *dir,
295 const struct qstr *qstr, struct folio **foliop)
296{
297 const unsigned char *name = qstr->name;
298 int namelen = qstr->len;
299 unsigned int reclen = NILFS_DIR_REC_LEN(namelen);
300 unsigned long start, n;
301 unsigned long npages = dir_pages(dir);
302 struct nilfs_inode_info *ei = NILFS_I(dir);
303 struct nilfs_dir_entry *de;
304
305 if (npages == 0)
306 goto out;
307
308 start = ei->i_dir_start_lookup;
309 if (start >= npages)
310 start = 0;
311 n = start;
312 do {
313 char *kaddr = nilfs_get_folio(dir, n, foliop);
314
315 if (IS_ERR(kaddr))
316 return ERR_CAST(kaddr);
317
318 de = (struct nilfs_dir_entry *)kaddr;
319 kaddr += nilfs_last_byte(dir, n) - reclen;
320 while ((char *)de <= kaddr) {
321 if (de->rec_len == 0) {
322 nilfs_error(dir->i_sb,
323 "zero-length directory entry");
324 folio_release_kmap(*foliop, kaddr);
325 goto out;
326 }
327 if (nilfs_match(namelen, name, de))
328 goto found;
329 de = nilfs_next_entry(de);
330 }
331 folio_release_kmap(*foliop, kaddr);
332
333 if (++n >= npages)
334 n = 0;
335 /* next folio is past the blocks we've got */
336 if (unlikely(n > (dir->i_blocks >> (PAGE_SHIFT - 9)))) {
337 nilfs_error(dir->i_sb,
338 "dir %lu size %lld exceeds block count %llu",
339 dir->i_ino, dir->i_size,
340 (unsigned long long)dir->i_blocks);
341 goto out;
342 }
343 } while (n != start);
344out:
345 return ERR_PTR(-ENOENT);
346
347found:
348 ei->i_dir_start_lookup = n;
349 return de;
350}
351
352struct nilfs_dir_entry *nilfs_dotdot(struct inode *dir, struct folio **foliop)
353{
354 struct folio *folio;
355 struct nilfs_dir_entry *de, *next_de;
356 size_t limit;
357 char *msg;
358
359 de = nilfs_get_folio(dir, 0, &folio);
360 if (IS_ERR(de))
361 return NULL;
362
363 limit = nilfs_last_byte(dir, 0); /* is a multiple of chunk size */
364 if (unlikely(!limit || le64_to_cpu(de->inode) != dir->i_ino ||
365 !nilfs_match(1, ".", de))) {
366 msg = "missing '.'";
367 goto fail;
368 }
369
370 next_de = nilfs_next_entry(de);
371 /*
372 * If "next_de" has not reached the end of the chunk, there is
373 * at least one more record. Check whether it matches "..".
374 */
375 if (unlikely((char *)next_de == (char *)de + nilfs_chunk_size(dir) ||
376 !nilfs_match(2, "..", next_de))) {
377 msg = "missing '..'";
378 goto fail;
379 }
380 *foliop = folio;
381 return next_de;
382
383fail:
384 nilfs_error(dir->i_sb, "directory #%lu %s", dir->i_ino, msg);
385 folio_release_kmap(folio, de);
386 return NULL;
387}
388
389int nilfs_inode_by_name(struct inode *dir, const struct qstr *qstr, ino_t *ino)
390{
391 struct nilfs_dir_entry *de;
392 struct folio *folio;
393
394 de = nilfs_find_entry(dir, qstr, &folio);
395 if (IS_ERR(de))
396 return PTR_ERR(de);
397
398 *ino = le64_to_cpu(de->inode);
399 folio_release_kmap(folio, de);
400 return 0;
401}
402
403int nilfs_set_link(struct inode *dir, struct nilfs_dir_entry *de,
404 struct folio *folio, struct inode *inode)
405{
406 size_t from = offset_in_folio(folio, de);
407 size_t to = from + nilfs_rec_len_from_disk(de->rec_len);
408 struct address_space *mapping = folio->mapping;
409 int err;
410
411 folio_lock(folio);
412 err = nilfs_prepare_chunk(folio, from, to);
413 if (unlikely(err)) {
414 folio_unlock(folio);
415 return err;
416 }
417 de->inode = cpu_to_le64(inode->i_ino);
418 de->file_type = fs_umode_to_ftype(inode->i_mode);
419 nilfs_commit_chunk(folio, mapping, from, to);
420 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
421 return 0;
422}
423
424/*
425 * Parent is locked.
426 */
427int nilfs_add_link(struct dentry *dentry, struct inode *inode)
428{
429 struct inode *dir = d_inode(dentry->d_parent);
430 const unsigned char *name = dentry->d_name.name;
431 int namelen = dentry->d_name.len;
432 unsigned int chunk_size = nilfs_chunk_size(dir);
433 unsigned int reclen = NILFS_DIR_REC_LEN(namelen);
434 unsigned short rec_len, name_len;
435 struct folio *folio = NULL;
436 struct nilfs_dir_entry *de;
437 unsigned long npages = dir_pages(dir);
438 unsigned long n;
439 size_t from, to;
440 int err;
441
442 /*
443 * We take care of directory expansion in the same loop.
444 * This code plays outside i_size, so it locks the folio
445 * to protect that region.
446 */
447 for (n = 0; n <= npages; n++) {
448 char *kaddr = nilfs_get_folio(dir, n, &folio);
449 char *dir_end;
450
451 if (IS_ERR(kaddr))
452 return PTR_ERR(kaddr);
453 folio_lock(folio);
454 dir_end = kaddr + nilfs_last_byte(dir, n);
455 de = (struct nilfs_dir_entry *)kaddr;
456 kaddr += folio_size(folio) - reclen;
457 while ((char *)de <= kaddr) {
458 if ((char *)de == dir_end) {
459 /* We hit i_size */
460 name_len = 0;
461 rec_len = chunk_size;
462 de->rec_len = nilfs_rec_len_to_disk(chunk_size);
463 de->inode = 0;
464 goto got_it;
465 }
466 if (de->rec_len == 0) {
467 nilfs_error(dir->i_sb,
468 "zero-length directory entry");
469 err = -EIO;
470 goto out_unlock;
471 }
472 err = -EEXIST;
473 if (nilfs_match(namelen, name, de))
474 goto out_unlock;
475 name_len = NILFS_DIR_REC_LEN(de->name_len);
476 rec_len = nilfs_rec_len_from_disk(de->rec_len);
477 if (!de->inode && rec_len >= reclen)
478 goto got_it;
479 if (rec_len >= name_len + reclen)
480 goto got_it;
481 de = (struct nilfs_dir_entry *)((char *)de + rec_len);
482 }
483 folio_unlock(folio);
484 folio_release_kmap(folio, kaddr);
485 }
486 BUG();
487 return -EINVAL;
488
489got_it:
490 from = offset_in_folio(folio, de);
491 to = from + rec_len;
492 err = nilfs_prepare_chunk(folio, from, to);
493 if (err)
494 goto out_unlock;
495 if (de->inode) {
496 struct nilfs_dir_entry *de1;
497
498 de1 = (struct nilfs_dir_entry *)((char *)de + name_len);
499 de1->rec_len = nilfs_rec_len_to_disk(rec_len - name_len);
500 de->rec_len = nilfs_rec_len_to_disk(name_len);
501 de = de1;
502 }
503 de->name_len = namelen;
504 memcpy(de->name, name, namelen);
505 de->inode = cpu_to_le64(inode->i_ino);
506 de->file_type = fs_umode_to_ftype(inode->i_mode);
507 nilfs_commit_chunk(folio, folio->mapping, from, to);
508 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
509 nilfs_mark_inode_dirty(dir);
510 /* OFFSET_CACHE */
511out_put:
512 folio_release_kmap(folio, de);
513 return err;
514out_unlock:
515 folio_unlock(folio);
516 goto out_put;
517}
518
519/*
520 * nilfs_delete_entry deletes a directory entry by merging it with the
521 * previous entry. Folio is up-to-date.
522 */
523int nilfs_delete_entry(struct nilfs_dir_entry *dir, struct folio *folio)
524{
525 struct address_space *mapping = folio->mapping;
526 struct inode *inode = mapping->host;
527 char *kaddr = (char *)((unsigned long)dir & ~(folio_size(folio) - 1));
528 size_t from, to;
529 struct nilfs_dir_entry *de, *pde = NULL;
530 int err;
531
532 from = ((char *)dir - kaddr) & ~(nilfs_chunk_size(inode) - 1);
533 to = ((char *)dir - kaddr) + nilfs_rec_len_from_disk(dir->rec_len);
534 de = (struct nilfs_dir_entry *)(kaddr + from);
535
536 while ((char *)de < (char *)dir) {
537 if (de->rec_len == 0) {
538 nilfs_error(inode->i_sb,
539 "zero-length directory entry");
540 err = -EIO;
541 goto out;
542 }
543 pde = de;
544 de = nilfs_next_entry(de);
545 }
546 if (pde)
547 from = (char *)pde - kaddr;
548 folio_lock(folio);
549 err = nilfs_prepare_chunk(folio, from, to);
550 if (unlikely(err)) {
551 folio_unlock(folio);
552 goto out;
553 }
554 if (pde)
555 pde->rec_len = nilfs_rec_len_to_disk(to - from);
556 dir->inode = 0;
557 nilfs_commit_chunk(folio, mapping, from, to);
558 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
559out:
560 return err;
561}
562
563/*
564 * Set the first fragment of directory.
565 */
566int nilfs_make_empty(struct inode *inode, struct inode *parent)
567{
568 struct address_space *mapping = inode->i_mapping;
569 struct folio *folio = filemap_grab_folio(mapping, 0);
570 unsigned int chunk_size = nilfs_chunk_size(inode);
571 struct nilfs_dir_entry *de;
572 int err;
573 void *kaddr;
574
575 if (IS_ERR(folio))
576 return PTR_ERR(folio);
577
578 err = nilfs_prepare_chunk(folio, 0, chunk_size);
579 if (unlikely(err)) {
580 folio_unlock(folio);
581 goto fail;
582 }
583 kaddr = kmap_local_folio(folio, 0);
584 memset(kaddr, 0, chunk_size);
585 de = (struct nilfs_dir_entry *)kaddr;
586 de->name_len = 1;
587 de->rec_len = nilfs_rec_len_to_disk(NILFS_DIR_REC_LEN(1));
588 memcpy(de->name, ".\0\0", 4);
589 de->inode = cpu_to_le64(inode->i_ino);
590 de->file_type = fs_umode_to_ftype(inode->i_mode);
591
592 de = (struct nilfs_dir_entry *)(kaddr + NILFS_DIR_REC_LEN(1));
593 de->name_len = 2;
594 de->rec_len = nilfs_rec_len_to_disk(chunk_size - NILFS_DIR_REC_LEN(1));
595 de->inode = cpu_to_le64(parent->i_ino);
596 memcpy(de->name, "..\0", 4);
597 de->file_type = fs_umode_to_ftype(inode->i_mode);
598 kunmap_local(kaddr);
599 nilfs_commit_chunk(folio, mapping, 0, chunk_size);
600fail:
601 folio_put(folio);
602 return err;
603}
604
605/*
606 * routine to check that the specified directory is empty (for rmdir)
607 */
608int nilfs_empty_dir(struct inode *inode)
609{
610 struct folio *folio = NULL;
611 char *kaddr;
612 unsigned long i, npages = dir_pages(inode);
613
614 for (i = 0; i < npages; i++) {
615 struct nilfs_dir_entry *de;
616
617 kaddr = nilfs_get_folio(inode, i, &folio);
618 if (IS_ERR(kaddr))
619 return 0;
620
621 de = (struct nilfs_dir_entry *)kaddr;
622 kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1);
623
624 while ((char *)de <= kaddr) {
625 if (de->rec_len == 0) {
626 nilfs_error(inode->i_sb,
627 "zero-length directory entry (kaddr=%p, de=%p)",
628 kaddr, de);
629 goto not_empty;
630 }
631 if (de->inode != 0) {
632 /* check for . and .. */
633 if (de->name[0] != '.')
634 goto not_empty;
635 if (de->name_len > 2)
636 goto not_empty;
637 if (de->name_len < 2) {
638 if (de->inode !=
639 cpu_to_le64(inode->i_ino))
640 goto not_empty;
641 } else if (de->name[1] != '.')
642 goto not_empty;
643 }
644 de = nilfs_next_entry(de);
645 }
646 folio_release_kmap(folio, kaddr);
647 }
648 return 1;
649
650not_empty:
651 folio_release_kmap(folio, kaddr);
652 return 0;
653}
654
655const struct file_operations nilfs_dir_operations = {
656 .llseek = generic_file_llseek,
657 .read = generic_read_dir,
658 .iterate_shared = nilfs_readdir,
659 .unlocked_ioctl = nilfs_ioctl,
660#ifdef CONFIG_COMPAT
661 .compat_ioctl = nilfs_compat_ioctl,
662#endif /* CONFIG_COMPAT */
663 .fsync = nilfs_sync_file,
664
665};