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>
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)
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
45 #include <linux/pagemap.h>
46 #include <linux/smp_lock.h>
51 * nilfs uses block-sized chunks. Arguably, sector-sized ones would be
52 * more robust, but we have what we have
54 static inline unsigned nilfs_chunk_size(struct inode *inode)
56 return inode->i_sb->s_blocksize;
59 static inline void nilfs_put_page(struct page *page)
62 page_cache_release(page);
65 static inline unsigned long dir_pages(struct inode *inode)
67 return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
71 * Return the offset into page `page_nr' of the last valid
72 * byte in that page, plus one.
74 static unsigned nilfs_last_byte(struct inode *inode, unsigned long page_nr)
76 unsigned last_byte = inode->i_size;
78 last_byte -= page_nr << PAGE_CACHE_SHIFT;
79 if (last_byte > PAGE_CACHE_SIZE)
80 last_byte = PAGE_CACHE_SIZE;
84 static int nilfs_prepare_chunk_uninterruptible(struct page *page,
85 struct address_space *mapping,
86 unsigned from, unsigned to)
88 loff_t pos = page_offset(page) + from;
89 return block_write_begin(NULL, mapping, pos, to - from,
90 AOP_FLAG_UNINTERRUPTIBLE, &page,
91 NULL, nilfs_get_block);
94 static int nilfs_prepare_chunk(struct page *page,
95 struct address_space *mapping,
96 unsigned from, unsigned to)
98 loff_t pos = page_offset(page) + from;
99 return block_write_begin(NULL, mapping, pos, to - from, 0, &page,
100 NULL, nilfs_get_block);
103 static int nilfs_commit_chunk(struct page *page,
104 struct address_space *mapping,
105 unsigned from, unsigned to)
107 struct inode *dir = mapping->host;
108 struct nilfs_sb_info *sbi = NILFS_SB(dir->i_sb);
109 loff_t pos = page_offset(page) + from;
110 unsigned len = to - from;
111 unsigned nr_dirty, copied;
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 mark_inode_dirty(dir);
121 nilfs_set_transaction_flag(NILFS_TI_SYNC);
122 err = nilfs_set_file_dirty(sbi, dir, nr_dirty);
127 static void nilfs_check_page(struct page *page)
129 struct inode *dir = page->mapping->host;
130 struct super_block *sb = dir->i_sb;
131 unsigned chunk_size = nilfs_chunk_size(dir);
132 char *kaddr = page_address(page);
133 unsigned offs, rec_len;
134 unsigned limit = PAGE_CACHE_SIZE;
135 struct nilfs_dir_entry *p;
138 if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
139 limit = dir->i_size & ~PAGE_CACHE_MASK;
140 if (limit & (chunk_size - 1))
145 for (offs = 0; offs <= limit - NILFS_DIR_REC_LEN(1); offs += rec_len) {
146 p = (struct nilfs_dir_entry *)(kaddr + offs);
147 rec_len = le16_to_cpu(p->rec_len);
149 if (rec_len < NILFS_DIR_REC_LEN(1))
153 if (rec_len < NILFS_DIR_REC_LEN(p->name_len))
155 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
161 SetPageChecked(page);
164 /* Too bad, we had an error */
167 nilfs_error(sb, "nilfs_check_page",
168 "size of directory #%lu is not a multiple of chunk size",
173 error = "rec_len is smaller than minimal";
176 error = "unaligned directory entry";
179 error = "rec_len is too small for name_len";
182 error = "directory entry across blocks";
184 nilfs_error(sb, "nilfs_check_page", "bad entry in directory #%lu: %s - "
185 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
186 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
187 (unsigned long) le64_to_cpu(p->inode),
188 rec_len, p->name_len);
191 p = (struct nilfs_dir_entry *)(kaddr + offs);
192 nilfs_error(sb, "nilfs_check_page",
193 "entry in directory #%lu spans the page boundary"
194 "offset=%lu, inode=%lu",
195 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
196 (unsigned long) le64_to_cpu(p->inode));
198 SetPageChecked(page);
202 static struct page *nilfs_get_page(struct inode *dir, unsigned long n)
204 struct address_space *mapping = dir->i_mapping;
205 struct page *page = read_cache_page(mapping, n,
206 (filler_t *)mapping->a_ops->readpage, NULL);
208 wait_on_page_locked(page);
210 if (!PageUptodate(page))
212 if (!PageChecked(page))
213 nilfs_check_page(page);
220 nilfs_put_page(page);
221 return ERR_PTR(-EIO);
225 * NOTE! unlike strncmp, nilfs_match returns 1 for success, 0 for failure.
227 * len <= NILFS_NAME_LEN and de != NULL are guaranteed by caller.
230 nilfs_match(int len, const char * const name, struct nilfs_dir_entry *de)
232 if (len != de->name_len)
236 return !memcmp(name, de->name, len);
240 * p is at least 6 bytes before the end of page
242 static struct nilfs_dir_entry *nilfs_next_entry(struct nilfs_dir_entry *p)
244 return (struct nilfs_dir_entry *)((char *)p + le16_to_cpu(p->rec_len));
248 nilfs_filetype_table[NILFS_FT_MAX] = {
249 [NILFS_FT_UNKNOWN] = DT_UNKNOWN,
250 [NILFS_FT_REG_FILE] = DT_REG,
251 [NILFS_FT_DIR] = DT_DIR,
252 [NILFS_FT_CHRDEV] = DT_CHR,
253 [NILFS_FT_BLKDEV] = DT_BLK,
254 [NILFS_FT_FIFO] = DT_FIFO,
255 [NILFS_FT_SOCK] = DT_SOCK,
256 [NILFS_FT_SYMLINK] = DT_LNK,
261 nilfs_type_by_mode[S_IFMT >> S_SHIFT] = {
262 [S_IFREG >> S_SHIFT] = NILFS_FT_REG_FILE,
263 [S_IFDIR >> S_SHIFT] = NILFS_FT_DIR,
264 [S_IFCHR >> S_SHIFT] = NILFS_FT_CHRDEV,
265 [S_IFBLK >> S_SHIFT] = NILFS_FT_BLKDEV,
266 [S_IFIFO >> S_SHIFT] = NILFS_FT_FIFO,
267 [S_IFSOCK >> S_SHIFT] = NILFS_FT_SOCK,
268 [S_IFLNK >> S_SHIFT] = NILFS_FT_SYMLINK,
271 static void nilfs_set_de_type(struct nilfs_dir_entry *de, struct inode *inode)
273 mode_t mode = inode->i_mode;
275 de->file_type = nilfs_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
278 static int nilfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
280 loff_t pos = filp->f_pos;
281 struct inode *inode = filp->f_dentry->d_inode;
282 struct super_block *sb = inode->i_sb;
283 unsigned int offset = pos & ~PAGE_CACHE_MASK;
284 unsigned long n = pos >> PAGE_CACHE_SHIFT;
285 unsigned long npages = dir_pages(inode);
286 /* unsigned chunk_mask = ~(nilfs_chunk_size(inode)-1); */
287 unsigned char *types = NULL;
290 if (pos > inode->i_size - NILFS_DIR_REC_LEN(1))
293 types = nilfs_filetype_table;
295 for ( ; n < npages; n++, offset = 0) {
297 struct nilfs_dir_entry *de;
298 struct page *page = nilfs_get_page(inode, n);
301 nilfs_error(sb, __func__, "bad page in #%lu",
303 filp->f_pos += PAGE_CACHE_SIZE - offset;
307 kaddr = page_address(page);
308 de = (struct nilfs_dir_entry *)(kaddr + offset);
309 limit = kaddr + nilfs_last_byte(inode, n) -
310 NILFS_DIR_REC_LEN(1);
311 for ( ; (char *)de <= limit; de = nilfs_next_entry(de)) {
312 if (de->rec_len == 0) {
313 nilfs_error(sb, __func__,
314 "zero-length directory entry");
316 nilfs_put_page(page);
321 unsigned char d_type = DT_UNKNOWN;
323 if (types && de->file_type < NILFS_FT_MAX)
324 d_type = types[de->file_type];
326 offset = (char *)de - kaddr;
327 over = filldir(dirent, de->name, de->name_len,
328 (n<<PAGE_CACHE_SHIFT) | offset,
329 le64_to_cpu(de->inode), d_type);
331 nilfs_put_page(page);
335 filp->f_pos += le16_to_cpu(de->rec_len);
337 nilfs_put_page(page);
349 * finds an entry in the specified directory with the wanted name. It
350 * returns the page in which the entry was found, and the entry itself
351 * (as a parameter - res_dir). Page is returned mapped and unlocked.
352 * Entry is guaranteed to be valid.
354 struct nilfs_dir_entry *
355 nilfs_find_entry(struct inode *dir, struct dentry *dentry,
356 struct page **res_page)
358 const char *name = dentry->d_name.name;
359 int namelen = dentry->d_name.len;
360 unsigned reclen = NILFS_DIR_REC_LEN(namelen);
361 unsigned long start, n;
362 unsigned long npages = dir_pages(dir);
363 struct page *page = NULL;
364 struct nilfs_inode_info *ei = NILFS_I(dir);
365 struct nilfs_dir_entry *de;
373 start = ei->i_dir_start_lookup;
379 page = nilfs_get_page(dir, n);
381 kaddr = page_address(page);
382 de = (struct nilfs_dir_entry *)kaddr;
383 kaddr += nilfs_last_byte(dir, n) - reclen;
384 while ((char *) de <= kaddr) {
385 if (de->rec_len == 0) {
386 nilfs_error(dir->i_sb, __func__,
387 "zero-length directory entry");
388 nilfs_put_page(page);
391 if (nilfs_match(namelen, name, de))
393 de = nilfs_next_entry(de);
395 nilfs_put_page(page);
399 /* next page is past the blocks we've got */
400 if (unlikely(n > (dir->i_blocks >> (PAGE_CACHE_SHIFT - 9)))) {
401 nilfs_error(dir->i_sb, __func__,
402 "dir %lu size %lld exceeds block cout %llu",
403 dir->i_ino, dir->i_size,
404 (unsigned long long)dir->i_blocks);
407 } while (n != start);
413 ei->i_dir_start_lookup = n;
417 struct nilfs_dir_entry *nilfs_dotdot(struct inode *dir, struct page **p)
419 struct page *page = nilfs_get_page(dir, 0);
420 struct nilfs_dir_entry *de = NULL;
423 de = nilfs_next_entry(
424 (struct nilfs_dir_entry *)page_address(page));
430 ino_t nilfs_inode_by_name(struct inode *dir, struct dentry *dentry)
433 struct nilfs_dir_entry *de;
436 de = nilfs_find_entry(dir, dentry, &page);
438 res = le64_to_cpu(de->inode);
440 page_cache_release(page);
445 /* Releases the page */
446 void nilfs_set_link(struct inode *dir, struct nilfs_dir_entry *de,
447 struct page *page, struct inode *inode)
449 unsigned from = (char *) de - (char *) page_address(page);
450 unsigned to = from + le16_to_cpu(de->rec_len);
451 struct address_space *mapping = page->mapping;
455 err = nilfs_prepare_chunk_uninterruptible(page, mapping, from, to);
457 de->inode = cpu_to_le64(inode->i_ino);
458 nilfs_set_de_type(de, inode);
459 err = nilfs_commit_chunk(page, mapping, from, to);
460 nilfs_put_page(page);
461 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
462 /* NILFS_I(dir)->i_flags &= ~NILFS_BTREE_FL; */
463 mark_inode_dirty(dir);
469 int nilfs_add_link(struct dentry *dentry, struct inode *inode)
471 struct inode *dir = dentry->d_parent->d_inode;
472 const char *name = dentry->d_name.name;
473 int namelen = dentry->d_name.len;
474 unsigned chunk_size = nilfs_chunk_size(dir);
475 unsigned reclen = NILFS_DIR_REC_LEN(namelen);
476 unsigned short rec_len, name_len;
477 struct page *page = NULL;
478 struct nilfs_dir_entry *de;
479 unsigned long npages = dir_pages(dir);
486 * We take care of directory expansion in the same loop.
487 * This code plays outside i_size, so it locks the page
488 * to protect that region.
490 for (n = 0; n <= npages; n++) {
493 page = nilfs_get_page(dir, n);
498 kaddr = page_address(page);
499 dir_end = kaddr + nilfs_last_byte(dir, n);
500 de = (struct nilfs_dir_entry *)kaddr;
501 kaddr += PAGE_CACHE_SIZE - reclen;
502 while ((char *)de <= kaddr) {
503 if ((char *)de == dir_end) {
506 rec_len = chunk_size;
507 de->rec_len = cpu_to_le16(chunk_size);
511 if (de->rec_len == 0) {
512 nilfs_error(dir->i_sb, __func__,
513 "zero-length directory entry");
518 if (nilfs_match(namelen, name, de))
520 name_len = NILFS_DIR_REC_LEN(de->name_len);
521 rec_len = le16_to_cpu(de->rec_len);
522 if (!de->inode && rec_len >= reclen)
524 if (rec_len >= name_len + reclen)
526 de = (struct nilfs_dir_entry *)((char *)de + rec_len);
529 nilfs_put_page(page);
535 from = (char *)de - (char *)page_address(page);
537 err = nilfs_prepare_chunk(page, page->mapping, from, to);
541 struct nilfs_dir_entry *de1;
543 de1 = (struct nilfs_dir_entry *)((char *)de + name_len);
544 de1->rec_len = cpu_to_le16(rec_len - name_len);
545 de->rec_len = cpu_to_le16(name_len);
548 de->name_len = namelen;
549 memcpy(de->name, name, namelen);
550 de->inode = cpu_to_le64(inode->i_ino);
551 nilfs_set_de_type(de, inode);
552 err = nilfs_commit_chunk(page, page->mapping, from, to);
553 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
554 /* NILFS_I(dir)->i_flags &= ~NILFS_BTREE_FL; */
555 mark_inode_dirty(dir);
558 nilfs_put_page(page);
567 * nilfs_delete_entry deletes a directory entry by merging it with the
568 * previous entry. Page is up-to-date. Releases the page.
570 int nilfs_delete_entry(struct nilfs_dir_entry *dir, struct page *page)
572 struct address_space *mapping = page->mapping;
573 struct inode *inode = mapping->host;
574 char *kaddr = page_address(page);
575 unsigned from = ((char *)dir - kaddr) & ~(nilfs_chunk_size(inode) - 1);
576 unsigned to = ((char *)dir - kaddr) + le16_to_cpu(dir->rec_len);
577 struct nilfs_dir_entry *pde = NULL;
578 struct nilfs_dir_entry *de = (struct nilfs_dir_entry *)(kaddr + from);
581 while ((char *)de < (char *)dir) {
582 if (de->rec_len == 0) {
583 nilfs_error(inode->i_sb, __func__,
584 "zero-length directory entry");
589 de = nilfs_next_entry(de);
592 from = (char *)pde - (char *)page_address(page);
594 err = nilfs_prepare_chunk(page, mapping, from, to);
597 pde->rec_len = cpu_to_le16(to - from);
599 err = nilfs_commit_chunk(page, mapping, from, to);
600 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
601 /* NILFS_I(inode)->i_flags &= ~NILFS_BTREE_FL; */
602 mark_inode_dirty(inode);
604 nilfs_put_page(page);
609 * Set the first fragment of directory.
611 int nilfs_make_empty(struct inode *inode, struct inode *parent)
613 struct address_space *mapping = inode->i_mapping;
614 struct page *page = grab_cache_page(mapping, 0);
615 unsigned chunk_size = nilfs_chunk_size(inode);
616 struct nilfs_dir_entry *de;
623 err = nilfs_prepare_chunk(page, mapping, 0, chunk_size);
628 kaddr = kmap_atomic(page, KM_USER0);
629 memset(kaddr, 0, chunk_size);
630 de = (struct nilfs_dir_entry *)kaddr;
632 de->rec_len = cpu_to_le16(NILFS_DIR_REC_LEN(1));
633 memcpy(de->name, ".\0\0", 4);
634 de->inode = cpu_to_le64(inode->i_ino);
635 nilfs_set_de_type(de, inode);
637 de = (struct nilfs_dir_entry *)(kaddr + NILFS_DIR_REC_LEN(1));
639 de->rec_len = cpu_to_le16(chunk_size - NILFS_DIR_REC_LEN(1));
640 de->inode = cpu_to_le64(parent->i_ino);
641 memcpy(de->name, "..\0", 4);
642 nilfs_set_de_type(de, inode);
643 kunmap_atomic(kaddr, KM_USER0);
644 err = nilfs_commit_chunk(page, mapping, 0, chunk_size);
646 page_cache_release(page);
651 * routine to check that the specified directory is empty (for rmdir)
653 int nilfs_empty_dir(struct inode *inode)
655 struct page *page = NULL;
656 unsigned long i, npages = dir_pages(inode);
658 for (i = 0; i < npages; i++) {
660 struct nilfs_dir_entry *de;
662 page = nilfs_get_page(inode, i);
666 kaddr = page_address(page);
667 de = (struct nilfs_dir_entry *)kaddr;
668 kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1);
670 while ((char *)de <= kaddr) {
671 if (de->rec_len == 0) {
672 nilfs_error(inode->i_sb, __func__,
673 "zero-length directory entry "
674 "(kaddr=%p, de=%p)\n", kaddr, de);
677 if (de->inode != 0) {
678 /* check for . and .. */
679 if (de->name[0] != '.')
681 if (de->name_len > 2)
683 if (de->name_len < 2) {
685 cpu_to_le64(inode->i_ino))
687 } else if (de->name[1] != '.')
690 de = nilfs_next_entry(de);
692 nilfs_put_page(page);
697 nilfs_put_page(page);
701 struct file_operations nilfs_dir_operations = {
702 .llseek = generic_file_llseek,
703 .read = generic_read_dir,
704 .readdir = nilfs_readdir,
705 .unlocked_ioctl = nilfs_ioctl,
707 .compat_ioctl = nilfs_ioctl,
708 #endif /* CONFIG_COMPAT */
709 .fsync = nilfs_sync_file,