writeback: bdi_writeback_task() must set task state before calling schedule()
[linux-2.6/next.git] / fs / nilfs2 / dir.c
blob85c89dfc71f0985c12f0355a5e397272078c48cc
1 /*
2 * dir.c - NILFS directory entry operations
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
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.
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.
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
20 * Modified for NILFS by Amagai Yoshiji <amagai@osrg.net>
23 * linux/fs/ext2/dir.c
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)
30 * from
32 * linux/fs/minix/dir.c
34 * Copyright (C) 1991, 1992 Linus Torvalds
36 * ext2 directory handling functions
38 * Big-endian to little-endian byte-swapping/bitmaps by
39 * David S. Miller (davem@caip.rutgers.edu), 1995
41 * All code that works with directory layout had been switched to pagecache
42 * and moved here. AV
45 #include <linux/pagemap.h>
46 #include "nilfs.h"
47 #include "page.h"
50 * nilfs uses block-sized chunks. Arguably, sector-sized ones would be
51 * more robust, but we have what we have
53 static inline unsigned nilfs_chunk_size(struct inode *inode)
55 return inode->i_sb->s_blocksize;
58 static inline void nilfs_put_page(struct page *page)
60 kunmap(page);
61 page_cache_release(page);
64 static inline unsigned long dir_pages(struct inode *inode)
66 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
70 * Return the offset into page `page_nr' of the last valid
71 * byte in that page, plus one.
73 static unsigned nilfs_last_byte(struct inode *inode, unsigned long page_nr)
75 unsigned last_byte = inode->i_size;
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;
83 static int nilfs_prepare_chunk_uninterruptible(struct page *page,
84 struct address_space *mapping,
85 unsigned from, unsigned to)
87 loff_t pos = page_offset(page) + from;
88 return block_write_begin(NULL, mapping, pos, to - from,
89 AOP_FLAG_UNINTERRUPTIBLE, &page,
90 NULL, nilfs_get_block);
93 static int nilfs_prepare_chunk(struct page *page,
94 struct address_space *mapping,
95 unsigned from, unsigned to)
97 loff_t pos = page_offset(page) + from;
98 return block_write_begin(NULL, mapping, pos, to - from, 0, &page,
99 NULL, nilfs_get_block);
102 static void nilfs_commit_chunk(struct page *page,
103 struct address_space *mapping,
104 unsigned from, unsigned to)
106 struct inode *dir = mapping->host;
107 struct nilfs_sb_info *sbi = NILFS_SB(dir->i_sb);
108 loff_t pos = page_offset(page) + from;
109 unsigned len = to - from;
110 unsigned nr_dirty, copied;
111 int err;
113 nr_dirty = nilfs_page_count_clean_buffers(page, from, to);
114 copied = block_write_end(NULL, mapping, pos, len, len, page, NULL);
115 if (pos + copied > dir->i_size)
116 i_size_write(dir, pos + copied);
117 if (IS_DIRSYNC(dir))
118 nilfs_set_transaction_flag(NILFS_TI_SYNC);
119 err = nilfs_set_file_dirty(sbi, dir, nr_dirty);
120 WARN_ON(err); /* do not happen */
121 unlock_page(page);
124 static void nilfs_check_page(struct page *page)
126 struct inode *dir = page->mapping->host;
127 struct super_block *sb = dir->i_sb;
128 unsigned chunk_size = nilfs_chunk_size(dir);
129 char *kaddr = page_address(page);
130 unsigned offs, rec_len;
131 unsigned limit = PAGE_CACHE_SIZE;
132 struct nilfs_dir_entry *p;
133 char *error;
135 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
136 limit = dir->i_size & ~PAGE_CACHE_MASK;
137 if (limit & (chunk_size - 1))
138 goto Ebadsize;
139 if (!limit)
140 goto out;
142 for (offs = 0; offs <= limit - NILFS_DIR_REC_LEN(1); offs += rec_len) {
143 p = (struct nilfs_dir_entry *)(kaddr + offs);
144 rec_len = le16_to_cpu(p->rec_len);
146 if (rec_len < NILFS_DIR_REC_LEN(1))
147 goto Eshort;
148 if (rec_len & 3)
149 goto Ealign;
150 if (rec_len < NILFS_DIR_REC_LEN(p->name_len))
151 goto Enamelen;
152 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
153 goto Espan;
155 if (offs != limit)
156 goto Eend;
157 out:
158 SetPageChecked(page);
159 return;
161 /* Too bad, we had an error */
163 Ebadsize:
164 nilfs_error(sb, "nilfs_check_page",
165 "size of directory #%lu is not a multiple of chunk size",
166 dir->i_ino
168 goto fail;
169 Eshort:
170 error = "rec_len is smaller than minimal";
171 goto bad_entry;
172 Ealign:
173 error = "unaligned directory entry";
174 goto bad_entry;
175 Enamelen:
176 error = "rec_len is too small for name_len";
177 goto bad_entry;
178 Espan:
179 error = "directory entry across blocks";
180 bad_entry:
181 nilfs_error(sb, "nilfs_check_page", "bad entry in directory #%lu: %s - "
182 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
183 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
184 (unsigned long) le64_to_cpu(p->inode),
185 rec_len, p->name_len);
186 goto fail;
187 Eend:
188 p = (struct nilfs_dir_entry *)(kaddr + offs);
189 nilfs_error(sb, "nilfs_check_page",
190 "entry in directory #%lu spans the page boundary"
191 "offset=%lu, inode=%lu",
192 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
193 (unsigned long) le64_to_cpu(p->inode));
194 fail:
195 SetPageChecked(page);
196 SetPageError(page);
199 static struct page *nilfs_get_page(struct inode *dir, unsigned long n)
201 struct address_space *mapping = dir->i_mapping;
202 struct page *page = read_cache_page(mapping, n,
203 (filler_t *)mapping->a_ops->readpage, NULL);
204 if (!IS_ERR(page)) {
205 wait_on_page_locked(page);
206 kmap(page);
207 if (!PageUptodate(page))
208 goto fail;
209 if (!PageChecked(page))
210 nilfs_check_page(page);
211 if (PageError(page))
212 goto fail;
214 return page;
216 fail:
217 nilfs_put_page(page);
218 return ERR_PTR(-EIO);
222 * NOTE! unlike strncmp, nilfs_match returns 1 for success, 0 for failure.
224 * len <= NILFS_NAME_LEN and de != NULL are guaranteed by caller.
226 static int
227 nilfs_match(int len, const unsigned char *name, struct nilfs_dir_entry *de)
229 if (len != de->name_len)
230 return 0;
231 if (!de->inode)
232 return 0;
233 return !memcmp(name, de->name, len);
237 * p is at least 6 bytes before the end of page
239 static struct nilfs_dir_entry *nilfs_next_entry(struct nilfs_dir_entry *p)
241 return (struct nilfs_dir_entry *)((char *)p + le16_to_cpu(p->rec_len));
244 static unsigned char
245 nilfs_filetype_table[NILFS_FT_MAX] = {
246 [NILFS_FT_UNKNOWN] = DT_UNKNOWN,
247 [NILFS_FT_REG_FILE] = DT_REG,
248 [NILFS_FT_DIR] = DT_DIR,
249 [NILFS_FT_CHRDEV] = DT_CHR,
250 [NILFS_FT_BLKDEV] = DT_BLK,
251 [NILFS_FT_FIFO] = DT_FIFO,
252 [NILFS_FT_SOCK] = DT_SOCK,
253 [NILFS_FT_SYMLINK] = DT_LNK,
256 #define S_SHIFT 12
257 static unsigned char
258 nilfs_type_by_mode[S_IFMT >> S_SHIFT] = {
259 [S_IFREG >> S_SHIFT] = NILFS_FT_REG_FILE,
260 [S_IFDIR >> S_SHIFT] = NILFS_FT_DIR,
261 [S_IFCHR >> S_SHIFT] = NILFS_FT_CHRDEV,
262 [S_IFBLK >> S_SHIFT] = NILFS_FT_BLKDEV,
263 [S_IFIFO >> S_SHIFT] = NILFS_FT_FIFO,
264 [S_IFSOCK >> S_SHIFT] = NILFS_FT_SOCK,
265 [S_IFLNK >> S_SHIFT] = NILFS_FT_SYMLINK,
268 static void nilfs_set_de_type(struct nilfs_dir_entry *de, struct inode *inode)
270 mode_t mode = inode->i_mode;
272 de->file_type = nilfs_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
275 static int nilfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
277 loff_t pos = filp->f_pos;
278 struct inode *inode = filp->f_dentry->d_inode;
279 struct super_block *sb = inode->i_sb;
280 unsigned int offset = pos & ~PAGE_CACHE_MASK;
281 unsigned long n = pos >> PAGE_CACHE_SHIFT;
282 unsigned long npages = dir_pages(inode);
283 /* unsigned chunk_mask = ~(nilfs_chunk_size(inode)-1); */
284 unsigned char *types = NULL;
285 int ret;
287 if (pos > inode->i_size - NILFS_DIR_REC_LEN(1))
288 goto success;
290 types = nilfs_filetype_table;
292 for ( ; n < npages; n++, offset = 0) {
293 char *kaddr, *limit;
294 struct nilfs_dir_entry *de;
295 struct page *page = nilfs_get_page(inode, n);
297 if (IS_ERR(page)) {
298 nilfs_error(sb, __func__, "bad page in #%lu",
299 inode->i_ino);
300 filp->f_pos += PAGE_CACHE_SIZE - offset;
301 ret = -EIO;
302 goto done;
304 kaddr = page_address(page);
305 de = (struct nilfs_dir_entry *)(kaddr + offset);
306 limit = kaddr + nilfs_last_byte(inode, n) -
307 NILFS_DIR_REC_LEN(1);
308 for ( ; (char *)de <= limit; de = nilfs_next_entry(de)) {
309 if (de->rec_len == 0) {
310 nilfs_error(sb, __func__,
311 "zero-length directory entry");
312 ret = -EIO;
313 nilfs_put_page(page);
314 goto done;
316 if (de->inode) {
317 int over;
318 unsigned char d_type = DT_UNKNOWN;
320 if (types && de->file_type < NILFS_FT_MAX)
321 d_type = types[de->file_type];
323 offset = (char *)de - kaddr;
324 over = filldir(dirent, de->name, de->name_len,
325 (n<<PAGE_CACHE_SHIFT) | offset,
326 le64_to_cpu(de->inode), d_type);
327 if (over) {
328 nilfs_put_page(page);
329 goto success;
332 filp->f_pos += le16_to_cpu(de->rec_len);
334 nilfs_put_page(page);
337 success:
338 ret = 0;
339 done:
340 return ret;
344 * nilfs_find_entry()
346 * finds an entry in the specified directory with the wanted name. It
347 * returns the page in which the entry was found, and the entry itself
348 * (as a parameter - res_dir). Page is returned mapped and unlocked.
349 * Entry is guaranteed to be valid.
351 struct nilfs_dir_entry *
352 nilfs_find_entry(struct inode *dir, const struct qstr *qstr,
353 struct page **res_page)
355 const unsigned char *name = qstr->name;
356 int namelen = qstr->len;
357 unsigned reclen = NILFS_DIR_REC_LEN(namelen);
358 unsigned long start, n;
359 unsigned long npages = dir_pages(dir);
360 struct page *page = NULL;
361 struct nilfs_inode_info *ei = NILFS_I(dir);
362 struct nilfs_dir_entry *de;
364 if (npages == 0)
365 goto out;
367 /* OFFSET_CACHE */
368 *res_page = NULL;
370 start = ei->i_dir_start_lookup;
371 if (start >= npages)
372 start = 0;
373 n = start;
374 do {
375 char *kaddr;
376 page = nilfs_get_page(dir, n);
377 if (!IS_ERR(page)) {
378 kaddr = page_address(page);
379 de = (struct nilfs_dir_entry *)kaddr;
380 kaddr += nilfs_last_byte(dir, n) - reclen;
381 while ((char *) de <= kaddr) {
382 if (de->rec_len == 0) {
383 nilfs_error(dir->i_sb, __func__,
384 "zero-length directory entry");
385 nilfs_put_page(page);
386 goto out;
388 if (nilfs_match(namelen, name, de))
389 goto found;
390 de = nilfs_next_entry(de);
392 nilfs_put_page(page);
394 if (++n >= npages)
395 n = 0;
396 /* next page is past the blocks we've got */
397 if (unlikely(n > (dir->i_blocks >> (PAGE_CACHE_SHIFT - 9)))) {
398 nilfs_error(dir->i_sb, __func__,
399 "dir %lu size %lld exceeds block count %llu",
400 dir->i_ino, dir->i_size,
401 (unsigned long long)dir->i_blocks);
402 goto out;
404 } while (n != start);
405 out:
406 return NULL;
408 found:
409 *res_page = page;
410 ei->i_dir_start_lookup = n;
411 return de;
414 struct nilfs_dir_entry *nilfs_dotdot(struct inode *dir, struct page **p)
416 struct page *page = nilfs_get_page(dir, 0);
417 struct nilfs_dir_entry *de = NULL;
419 if (!IS_ERR(page)) {
420 de = nilfs_next_entry(
421 (struct nilfs_dir_entry *)page_address(page));
422 *p = page;
424 return de;
427 ino_t nilfs_inode_by_name(struct inode *dir, const struct qstr *qstr)
429 ino_t res = 0;
430 struct nilfs_dir_entry *de;
431 struct page *page;
433 de = nilfs_find_entry(dir, qstr, &page);
434 if (de) {
435 res = le64_to_cpu(de->inode);
436 kunmap(page);
437 page_cache_release(page);
439 return res;
442 /* Releases the page */
443 void nilfs_set_link(struct inode *dir, struct nilfs_dir_entry *de,
444 struct page *page, struct inode *inode)
446 unsigned from = (char *) de - (char *) page_address(page);
447 unsigned to = from + le16_to_cpu(de->rec_len);
448 struct address_space *mapping = page->mapping;
449 int err;
451 lock_page(page);
452 err = nilfs_prepare_chunk_uninterruptible(page, mapping, from, to);
453 BUG_ON(err);
454 de->inode = cpu_to_le64(inode->i_ino);
455 nilfs_set_de_type(de, inode);
456 nilfs_commit_chunk(page, mapping, from, to);
457 nilfs_put_page(page);
458 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
459 /* NILFS_I(dir)->i_flags &= ~NILFS_BTREE_FL; */
463 * Parent is locked.
465 int nilfs_add_link(struct dentry *dentry, struct inode *inode)
467 struct inode *dir = dentry->d_parent->d_inode;
468 const unsigned char *name = dentry->d_name.name;
469 int namelen = dentry->d_name.len;
470 unsigned chunk_size = nilfs_chunk_size(dir);
471 unsigned reclen = NILFS_DIR_REC_LEN(namelen);
472 unsigned short rec_len, name_len;
473 struct page *page = NULL;
474 struct nilfs_dir_entry *de;
475 unsigned long npages = dir_pages(dir);
476 unsigned long n;
477 char *kaddr;
478 unsigned from, to;
479 int err;
482 * We take care of directory expansion in the same loop.
483 * This code plays outside i_size, so it locks the page
484 * to protect that region.
486 for (n = 0; n <= npages; n++) {
487 char *dir_end;
489 page = nilfs_get_page(dir, n);
490 err = PTR_ERR(page);
491 if (IS_ERR(page))
492 goto out;
493 lock_page(page);
494 kaddr = page_address(page);
495 dir_end = kaddr + nilfs_last_byte(dir, n);
496 de = (struct nilfs_dir_entry *)kaddr;
497 kaddr += PAGE_CACHE_SIZE - reclen;
498 while ((char *)de <= kaddr) {
499 if ((char *)de == dir_end) {
500 /* We hit i_size */
501 name_len = 0;
502 rec_len = chunk_size;
503 de->rec_len = cpu_to_le16(chunk_size);
504 de->inode = 0;
505 goto got_it;
507 if (de->rec_len == 0) {
508 nilfs_error(dir->i_sb, __func__,
509 "zero-length directory entry");
510 err = -EIO;
511 goto out_unlock;
513 err = -EEXIST;
514 if (nilfs_match(namelen, name, de))
515 goto out_unlock;
516 name_len = NILFS_DIR_REC_LEN(de->name_len);
517 rec_len = le16_to_cpu(de->rec_len);
518 if (!de->inode && rec_len >= reclen)
519 goto got_it;
520 if (rec_len >= name_len + reclen)
521 goto got_it;
522 de = (struct nilfs_dir_entry *)((char *)de + rec_len);
524 unlock_page(page);
525 nilfs_put_page(page);
527 BUG();
528 return -EINVAL;
530 got_it:
531 from = (char *)de - (char *)page_address(page);
532 to = from + rec_len;
533 err = nilfs_prepare_chunk(page, page->mapping, from, to);
534 if (err)
535 goto out_unlock;
536 if (de->inode) {
537 struct nilfs_dir_entry *de1;
539 de1 = (struct nilfs_dir_entry *)((char *)de + name_len);
540 de1->rec_len = cpu_to_le16(rec_len - name_len);
541 de->rec_len = cpu_to_le16(name_len);
542 de = de1;
544 de->name_len = namelen;
545 memcpy(de->name, name, namelen);
546 de->inode = cpu_to_le64(inode->i_ino);
547 nilfs_set_de_type(de, inode);
548 nilfs_commit_chunk(page, page->mapping, from, to);
549 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
550 /* NILFS_I(dir)->i_flags &= ~NILFS_BTREE_FL; */
551 nilfs_mark_inode_dirty(dir);
552 /* OFFSET_CACHE */
553 out_put:
554 nilfs_put_page(page);
555 out:
556 return err;
557 out_unlock:
558 unlock_page(page);
559 goto out_put;
563 * nilfs_delete_entry deletes a directory entry by merging it with the
564 * previous entry. Page is up-to-date. Releases the page.
566 int nilfs_delete_entry(struct nilfs_dir_entry *dir, struct page *page)
568 struct address_space *mapping = page->mapping;
569 struct inode *inode = mapping->host;
570 char *kaddr = page_address(page);
571 unsigned from = ((char *)dir - kaddr) & ~(nilfs_chunk_size(inode) - 1);
572 unsigned to = ((char *)dir - kaddr) + le16_to_cpu(dir->rec_len);
573 struct nilfs_dir_entry *pde = NULL;
574 struct nilfs_dir_entry *de = (struct nilfs_dir_entry *)(kaddr + from);
575 int err;
577 while ((char *)de < (char *)dir) {
578 if (de->rec_len == 0) {
579 nilfs_error(inode->i_sb, __func__,
580 "zero-length directory entry");
581 err = -EIO;
582 goto out;
584 pde = de;
585 de = nilfs_next_entry(de);
587 if (pde)
588 from = (char *)pde - (char *)page_address(page);
589 lock_page(page);
590 err = nilfs_prepare_chunk(page, mapping, from, to);
591 BUG_ON(err);
592 if (pde)
593 pde->rec_len = cpu_to_le16(to - from);
594 dir->inode = 0;
595 nilfs_commit_chunk(page, mapping, from, to);
596 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
597 /* NILFS_I(inode)->i_flags &= ~NILFS_BTREE_FL; */
598 out:
599 nilfs_put_page(page);
600 return err;
604 * Set the first fragment of directory.
606 int nilfs_make_empty(struct inode *inode, struct inode *parent)
608 struct address_space *mapping = inode->i_mapping;
609 struct page *page = grab_cache_page(mapping, 0);
610 unsigned chunk_size = nilfs_chunk_size(inode);
611 struct nilfs_dir_entry *de;
612 int err;
613 void *kaddr;
615 if (!page)
616 return -ENOMEM;
618 err = nilfs_prepare_chunk(page, mapping, 0, chunk_size);
619 if (unlikely(err)) {
620 unlock_page(page);
621 goto fail;
623 kaddr = kmap_atomic(page, KM_USER0);
624 memset(kaddr, 0, chunk_size);
625 de = (struct nilfs_dir_entry *)kaddr;
626 de->name_len = 1;
627 de->rec_len = cpu_to_le16(NILFS_DIR_REC_LEN(1));
628 memcpy(de->name, ".\0\0", 4);
629 de->inode = cpu_to_le64(inode->i_ino);
630 nilfs_set_de_type(de, inode);
632 de = (struct nilfs_dir_entry *)(kaddr + NILFS_DIR_REC_LEN(1));
633 de->name_len = 2;
634 de->rec_len = cpu_to_le16(chunk_size - NILFS_DIR_REC_LEN(1));
635 de->inode = cpu_to_le64(parent->i_ino);
636 memcpy(de->name, "..\0", 4);
637 nilfs_set_de_type(de, inode);
638 kunmap_atomic(kaddr, KM_USER0);
639 nilfs_commit_chunk(page, mapping, 0, chunk_size);
640 fail:
641 page_cache_release(page);
642 return err;
646 * routine to check that the specified directory is empty (for rmdir)
648 int nilfs_empty_dir(struct inode *inode)
650 struct page *page = NULL;
651 unsigned long i, npages = dir_pages(inode);
653 for (i = 0; i < npages; i++) {
654 char *kaddr;
655 struct nilfs_dir_entry *de;
657 page = nilfs_get_page(inode, i);
658 if (IS_ERR(page))
659 continue;
661 kaddr = page_address(page);
662 de = (struct nilfs_dir_entry *)kaddr;
663 kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1);
665 while ((char *)de <= kaddr) {
666 if (de->rec_len == 0) {
667 nilfs_error(inode->i_sb, __func__,
668 "zero-length directory entry "
669 "(kaddr=%p, de=%p)\n", kaddr, de);
670 goto not_empty;
672 if (de->inode != 0) {
673 /* check for . and .. */
674 if (de->name[0] != '.')
675 goto not_empty;
676 if (de->name_len > 2)
677 goto not_empty;
678 if (de->name_len < 2) {
679 if (de->inode !=
680 cpu_to_le64(inode->i_ino))
681 goto not_empty;
682 } else if (de->name[1] != '.')
683 goto not_empty;
685 de = nilfs_next_entry(de);
687 nilfs_put_page(page);
689 return 1;
691 not_empty:
692 nilfs_put_page(page);
693 return 0;
696 const struct file_operations nilfs_dir_operations = {
697 .llseek = generic_file_llseek,
698 .read = generic_read_dir,
699 .readdir = nilfs_readdir,
700 .unlocked_ioctl = nilfs_ioctl,
701 #ifdef CONFIG_COMPAT
702 .compat_ioctl = nilfs_ioctl,
703 #endif /* CONFIG_COMPAT */
704 .fsync = nilfs_sync_file,