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
64/*
65 * Return the offset into page `page_nr' of the last valid
66 * byte in that page, plus one.
67 */
68static unsigned nilfs_last_byte(struct inode *inode, unsigned long page_nr)
69{
70	unsigned last_byte = inode->i_size;
71
72	last_byte -= page_nr << PAGE_CACHE_SHIFT;
73	if (last_byte > PAGE_CACHE_SIZE)
74		last_byte = PAGE_CACHE_SIZE;
75	return last_byte;
76}
77
78static int nilfs_prepare_chunk(struct page *page, unsigned from, unsigned to)
79{
80	loff_t pos = page_offset(page) + from;
81	return __block_write_begin(page, pos, to - from, nilfs_get_block);
82}
83
84static void nilfs_commit_chunk(struct page *page,
85			       struct address_space *mapping,
86			       unsigned from, unsigned to)
87{
88	struct inode *dir = mapping->host;
89	loff_t pos = page_offset(page) + from;
90	unsigned len = to - from;
91	unsigned nr_dirty, copied;
92	int err;
93
94	nr_dirty = nilfs_page_count_clean_buffers(page, from, to);
95	copied = block_write_end(NULL, mapping, pos, len, len, page, NULL);
96	if (pos + copied > dir->i_size)
97		i_size_write(dir, pos + copied);
98	if (IS_DIRSYNC(dir))
99		nilfs_set_transaction_flag(NILFS_TI_SYNC);
100	err = nilfs_set_file_dirty(dir, nr_dirty);
101	WARN_ON(err); /* do not happen */
102	unlock_page(page);
103}
104
105static void nilfs_check_page(struct page *page)
106{
107	struct inode *dir = page->mapping->host;
108	struct super_block *sb = dir->i_sb;
109	unsigned chunk_size = nilfs_chunk_size(dir);
110	char *kaddr = page_address(page);
111	unsigned offs, rec_len;
112	unsigned limit = PAGE_CACHE_SIZE;
113	struct nilfs_dir_entry *p;
114	char *error;
115
116	if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
117		limit = dir->i_size & ~PAGE_CACHE_MASK;
118		if (limit & (chunk_size - 1))
119			goto Ebadsize;
120		if (!limit)
121			goto out;
122	}
123	for (offs = 0; offs <= limit - NILFS_DIR_REC_LEN(1); offs += rec_len) {
124		p = (struct nilfs_dir_entry *)(kaddr + offs);
125		rec_len = nilfs_rec_len_from_disk(p->rec_len);
126
127		if (rec_len < NILFS_DIR_REC_LEN(1))
128			goto Eshort;
129		if (rec_len & 3)
130			goto Ealign;
131		if (rec_len < NILFS_DIR_REC_LEN(p->name_len))
132			goto Enamelen;
133		if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
134			goto Espan;
135	}
136	if (offs != limit)
137		goto Eend;
138out:
139	SetPageChecked(page);
140	return;
141
142	/* Too bad, we had an error */
143
144Ebadsize:
145	nilfs_error(sb, "nilfs_check_page",
146		    "size of directory #%lu is not a multiple of chunk size",
147		    dir->i_ino
148	);
149	goto fail;
150Eshort:
151	error = "rec_len is smaller than minimal";
152	goto bad_entry;
153Ealign:
154	error = "unaligned directory entry";
155	goto bad_entry;
156Enamelen:
157	error = "rec_len is too small for name_len";
158	goto bad_entry;
159Espan:
160	error = "directory entry across blocks";
161bad_entry:
162	nilfs_error(sb, "nilfs_check_page", "bad entry in directory #%lu: %s - "
163		    "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
164		    dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT)+offs,
165		    (unsigned long) le64_to_cpu(p->inode),
166		    rec_len, p->name_len);
167	goto fail;
168Eend:
169	p = (struct nilfs_dir_entry *)(kaddr + offs);
170	nilfs_error(sb, "nilfs_check_page",
171		    "entry in directory #%lu spans the page boundary"
172		    "offset=%lu, inode=%lu",
173		    dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
174		    (unsigned long) le64_to_cpu(p->inode));
175fail:
176	SetPageChecked(page);
177	SetPageError(page);
178}
179
180static struct page *nilfs_get_page(struct inode *dir, unsigned long n)
181{
182	struct address_space *mapping = dir->i_mapping;
183	struct page *page = read_mapping_page(mapping, n, NULL);
184
185	if (!IS_ERR(page)) {
186		kmap(page);
187		if (!PageChecked(page))
188			nilfs_check_page(page);
189		if (PageError(page))
190			goto fail;
191	}
192	return page;
193
194fail:
195	nilfs_put_page(page);
196	return ERR_PTR(-EIO);
197}
198
199/*
200 * NOTE! unlike strncmp, nilfs_match returns 1 for success, 0 for failure.
201 *
202 * len <= NILFS_NAME_LEN and de != NULL are guaranteed by caller.
203 */
204static int
205nilfs_match(int len, const unsigned char *name, struct nilfs_dir_entry *de)
206{
207	if (len != de->name_len)
208		return 0;
209	if (!de->inode)
210		return 0;
211	return !memcmp(name, de->name, len);
212}
213
214/*
215 * p is at least 6 bytes before the end of page
216 */
217static struct nilfs_dir_entry *nilfs_next_entry(struct nilfs_dir_entry *p)
218{
219	return (struct nilfs_dir_entry *)((char *)p +
220					  nilfs_rec_len_from_disk(p->rec_len));
221}
222
223static unsigned char
224nilfs_filetype_table[NILFS_FT_MAX] = {
225	[NILFS_FT_UNKNOWN]	= DT_UNKNOWN,
226	[NILFS_FT_REG_FILE]	= DT_REG,
227	[NILFS_FT_DIR]		= DT_DIR,
228	[NILFS_FT_CHRDEV]	= DT_CHR,
229	[NILFS_FT_BLKDEV]	= DT_BLK,
230	[NILFS_FT_FIFO]		= DT_FIFO,
231	[NILFS_FT_SOCK]		= DT_SOCK,
232	[NILFS_FT_SYMLINK]	= DT_LNK,
233};
234
235#define S_SHIFT 12
236static unsigned char
237nilfs_type_by_mode[S_IFMT >> S_SHIFT] = {
238	[S_IFREG >> S_SHIFT]	= NILFS_FT_REG_FILE,
239	[S_IFDIR >> S_SHIFT]	= NILFS_FT_DIR,
240	[S_IFCHR >> S_SHIFT]	= NILFS_FT_CHRDEV,
241	[S_IFBLK >> S_SHIFT]	= NILFS_FT_BLKDEV,
242	[S_IFIFO >> S_SHIFT]	= NILFS_FT_FIFO,
243	[S_IFSOCK >> S_SHIFT]	= NILFS_FT_SOCK,
244	[S_IFLNK >> S_SHIFT]	= NILFS_FT_SYMLINK,
245};
246
247static void nilfs_set_de_type(struct nilfs_dir_entry *de, struct inode *inode)
248{
249	umode_t mode = inode->i_mode;
250
251	de->file_type = nilfs_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
252}
253
254static int nilfs_readdir(struct file *file, struct dir_context *ctx)
255{
256	loff_t pos = ctx->pos;
257	struct inode *inode = file_inode(file);
258	struct super_block *sb = inode->i_sb;
259	unsigned int offset = pos & ~PAGE_CACHE_MASK;
260	unsigned long n = pos >> PAGE_CACHE_SHIFT;
261	unsigned long npages = dir_pages(inode);
262/*	unsigned chunk_mask = ~(nilfs_chunk_size(inode)-1); */
263
264	if (pos > inode->i_size - NILFS_DIR_REC_LEN(1))
265		return 0;
266
267	for ( ; n < npages; n++, offset = 0) {
268		char *kaddr, *limit;
269		struct nilfs_dir_entry *de;
270		struct page *page = nilfs_get_page(inode, n);
271
272		if (IS_ERR(page)) {
273			nilfs_error(sb, __func__, "bad page in #%lu",
274				    inode->i_ino);
275			ctx->pos += PAGE_CACHE_SIZE - offset;
276			return -EIO;
277		}
278		kaddr = page_address(page);
279		de = (struct nilfs_dir_entry *)(kaddr + offset);
280		limit = kaddr + nilfs_last_byte(inode, n) -
281			NILFS_DIR_REC_LEN(1);
282		for ( ; (char *)de <= limit; de = nilfs_next_entry(de)) {
283			if (de->rec_len == 0) {
284				nilfs_error(sb, __func__,
285					    "zero-length directory entry");
286				nilfs_put_page(page);
287				return -EIO;
288			}
289			if (de->inode) {
290				unsigned char t;
291
292				if (de->file_type < NILFS_FT_MAX)
293					t = nilfs_filetype_table[de->file_type];
294				else
295					t = DT_UNKNOWN;
296
297				if (!dir_emit(ctx, de->name, de->name_len,
298						le64_to_cpu(de->inode), t)) {
299					nilfs_put_page(page);
300					return 0;
301				}
302			}
303			ctx->pos += nilfs_rec_len_from_disk(de->rec_len);
304		}
305		nilfs_put_page(page);
306	}
307	return 0;
308}
309
310/*
311 *	nilfs_find_entry()
312 *
313 * finds an entry in the specified directory with the wanted name. It
314 * returns the page in which the entry was found, and the entry itself
315 * (as a parameter - res_dir). Page is returned mapped and unlocked.
316 * Entry is guaranteed to be valid.
317 */
318struct nilfs_dir_entry *
319nilfs_find_entry(struct inode *dir, const struct qstr *qstr,
320		 struct page **res_page)
321{
322	const unsigned char *name = qstr->name;
323	int namelen = qstr->len;
324	unsigned reclen = NILFS_DIR_REC_LEN(namelen);
325	unsigned long start, n;
326	unsigned long npages = dir_pages(dir);
327	struct page *page = NULL;
328	struct nilfs_inode_info *ei = NILFS_I(dir);
329	struct nilfs_dir_entry *de;
330
331	if (npages == 0)
332		goto out;
333
334	/* OFFSET_CACHE */
335	*res_page = NULL;
336
337	start = ei->i_dir_start_lookup;
338	if (start >= npages)
339		start = 0;
340	n = start;
341	do {
342		char *kaddr;
343		page = nilfs_get_page(dir, n);
344		if (!IS_ERR(page)) {
345			kaddr = page_address(page);
346			de = (struct nilfs_dir_entry *)kaddr;
347			kaddr += nilfs_last_byte(dir, n) - reclen;
348			while ((char *) de <= kaddr) {
349				if (de->rec_len == 0) {
350					nilfs_error(dir->i_sb, __func__,
351						"zero-length directory entry");
352					nilfs_put_page(page);
353					goto out;
354				}
355				if (nilfs_match(namelen, name, de))
356					goto found;
357				de = nilfs_next_entry(de);
358			}
359			nilfs_put_page(page);
360		}
361		if (++n >= npages)
362			n = 0;
363		/* next page is past the blocks we've got */
364		if (unlikely(n > (dir->i_blocks >> (PAGE_CACHE_SHIFT - 9)))) {
365			nilfs_error(dir->i_sb, __func__,
366			       "dir %lu size %lld exceeds block count %llu",
367			       dir->i_ino, dir->i_size,
368			       (unsigned long long)dir->i_blocks);
369			goto out;
370		}
371	} while (n != start);
372out:
373	return NULL;
374
375found:
376	*res_page = page;
377	ei->i_dir_start_lookup = n;
378	return de;
379}
380
381struct nilfs_dir_entry *nilfs_dotdot(struct inode *dir, struct page **p)
382{
383	struct page *page = nilfs_get_page(dir, 0);
384	struct nilfs_dir_entry *de = NULL;
385
386	if (!IS_ERR(page)) {
387		de = nilfs_next_entry(
388			(struct nilfs_dir_entry *)page_address(page));
389		*p = page;
390	}
391	return de;
392}
393
394ino_t nilfs_inode_by_name(struct inode *dir, const struct qstr *qstr)
395{
396	ino_t res = 0;
397	struct nilfs_dir_entry *de;
398	struct page *page;
399
400	de = nilfs_find_entry(dir, qstr, &page);
401	if (de) {
402		res = le64_to_cpu(de->inode);
403		kunmap(page);
404		page_cache_release(page);
405	}
406	return res;
407}
408
409/* Releases the page */
410void nilfs_set_link(struct inode *dir, struct nilfs_dir_entry *de,
411		    struct page *page, struct inode *inode)
412{
413	unsigned from = (char *) de - (char *) page_address(page);
414	unsigned to = from + nilfs_rec_len_from_disk(de->rec_len);
415	struct address_space *mapping = page->mapping;
416	int err;
417
418	lock_page(page);
419	err = nilfs_prepare_chunk(page, from, to);
420	BUG_ON(err);
421	de->inode = cpu_to_le64(inode->i_ino);
422	nilfs_set_de_type(de, inode);
423	nilfs_commit_chunk(page, mapping, from, to);
424	nilfs_put_page(page);
425	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
426}
427
428/*
429 *	Parent is locked.
430 */
431int nilfs_add_link(struct dentry *dentry, struct inode *inode)
432{
433	struct inode *dir = d_inode(dentry->d_parent);
434	const unsigned char *name = dentry->d_name.name;
435	int namelen = dentry->d_name.len;
436	unsigned chunk_size = nilfs_chunk_size(dir);
437	unsigned reclen = NILFS_DIR_REC_LEN(namelen);
438	unsigned short rec_len, name_len;
439	struct page *page = NULL;
440	struct nilfs_dir_entry *de;
441	unsigned long npages = dir_pages(dir);
442	unsigned long n;
443	char *kaddr;
444	unsigned from, to;
445	int err;
446
447	/*
448	 * We take care of directory expansion in the same loop.
449	 * This code plays outside i_size, so it locks the page
450	 * to protect that region.
451	 */
452	for (n = 0; n <= npages; n++) {
453		char *dir_end;
454
455		page = nilfs_get_page(dir, n);
456		err = PTR_ERR(page);
457		if (IS_ERR(page))
458			goto out;
459		lock_page(page);
460		kaddr = page_address(page);
461		dir_end = kaddr + nilfs_last_byte(dir, n);
462		de = (struct nilfs_dir_entry *)kaddr;
463		kaddr += PAGE_CACHE_SIZE - reclen;
464		while ((char *)de <= kaddr) {
465			if ((char *)de == dir_end) {
466				/* We hit i_size */
467				name_len = 0;
468				rec_len = chunk_size;
469				de->rec_len = nilfs_rec_len_to_disk(chunk_size);
470				de->inode = 0;
471				goto got_it;
472			}
473			if (de->rec_len == 0) {
474				nilfs_error(dir->i_sb, __func__,
475					    "zero-length directory entry");
476				err = -EIO;
477				goto out_unlock;
478			}
479			err = -EEXIST;
480			if (nilfs_match(namelen, name, de))
481				goto out_unlock;
482			name_len = NILFS_DIR_REC_LEN(de->name_len);
483			rec_len = nilfs_rec_len_from_disk(de->rec_len);
484			if (!de->inode && rec_len >= reclen)
485				goto got_it;
486			if (rec_len >= name_len + reclen)
487				goto got_it;
488			de = (struct nilfs_dir_entry *)((char *)de + rec_len);
489		}
490		unlock_page(page);
491		nilfs_put_page(page);
492	}
493	BUG();
494	return -EINVAL;
495
496got_it:
497	from = (char *)de - (char *)page_address(page);
498	to = from + rec_len;
499	err = nilfs_prepare_chunk(page, from, to);
500	if (err)
501		goto out_unlock;
502	if (de->inode) {
503		struct nilfs_dir_entry *de1;
504
505		de1 = (struct nilfs_dir_entry *)((char *)de + name_len);
506		de1->rec_len = nilfs_rec_len_to_disk(rec_len - name_len);
507		de->rec_len = nilfs_rec_len_to_disk(name_len);
508		de = de1;
509	}
510	de->name_len = namelen;
511	memcpy(de->name, name, namelen);
512	de->inode = cpu_to_le64(inode->i_ino);
513	nilfs_set_de_type(de, inode);
514	nilfs_commit_chunk(page, page->mapping, from, to);
515	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
516	nilfs_mark_inode_dirty(dir);
517	/* OFFSET_CACHE */
518out_put:
519	nilfs_put_page(page);
520out:
521	return err;
522out_unlock:
523	unlock_page(page);
524	goto out_put;
525}
526
527/*
528 * nilfs_delete_entry deletes a directory entry by merging it with the
529 * previous entry. Page is up-to-date. Releases the page.
530 */
531int nilfs_delete_entry(struct nilfs_dir_entry *dir, struct page *page)
532{
533	struct address_space *mapping = page->mapping;
534	struct inode *inode = mapping->host;
535	char *kaddr = page_address(page);
536	unsigned from = ((char *)dir - kaddr) & ~(nilfs_chunk_size(inode) - 1);
537	unsigned to = ((char *)dir - kaddr) +
538		nilfs_rec_len_from_disk(dir->rec_len);
539	struct nilfs_dir_entry *pde = NULL;
540	struct nilfs_dir_entry *de = (struct nilfs_dir_entry *)(kaddr + from);
541	int err;
542
543	while ((char *)de < (char *)dir) {
544		if (de->rec_len == 0) {
545			nilfs_error(inode->i_sb, __func__,
546				    "zero-length directory entry");
547			err = -EIO;
548			goto out;
549		}
550		pde = de;
551		de = nilfs_next_entry(de);
552	}
553	if (pde)
554		from = (char *)pde - (char *)page_address(page);
555	lock_page(page);
556	err = nilfs_prepare_chunk(page, from, to);
557	BUG_ON(err);
558	if (pde)
559		pde->rec_len = nilfs_rec_len_to_disk(to - from);
560	dir->inode = 0;
561	nilfs_commit_chunk(page, mapping, from, to);
562	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
563out:
564	nilfs_put_page(page);
565	return err;
566}
567
568/*
569 * Set the first fragment of directory.
570 */
571int nilfs_make_empty(struct inode *inode, struct inode *parent)
572{
573	struct address_space *mapping = inode->i_mapping;
574	struct page *page = grab_cache_page(mapping, 0);
575	unsigned chunk_size = nilfs_chunk_size(inode);
576	struct nilfs_dir_entry *de;
577	int err;
578	void *kaddr;
579
580	if (!page)
581		return -ENOMEM;
582
583	err = nilfs_prepare_chunk(page, 0, chunk_size);
584	if (unlikely(err)) {
585		unlock_page(page);
586		goto fail;
587	}
588	kaddr = kmap_atomic(page);
589	memset(kaddr, 0, chunk_size);
590	de = (struct nilfs_dir_entry *)kaddr;
591	de->name_len = 1;
592	de->rec_len = nilfs_rec_len_to_disk(NILFS_DIR_REC_LEN(1));
593	memcpy(de->name, ".\0\0", 4);
594	de->inode = cpu_to_le64(inode->i_ino);
595	nilfs_set_de_type(de, inode);
596
597	de = (struct nilfs_dir_entry *)(kaddr + NILFS_DIR_REC_LEN(1));
598	de->name_len = 2;
599	de->rec_len = nilfs_rec_len_to_disk(chunk_size - NILFS_DIR_REC_LEN(1));
600	de->inode = cpu_to_le64(parent->i_ino);
601	memcpy(de->name, "..\0", 4);
602	nilfs_set_de_type(de, inode);
603	kunmap_atomic(kaddr);
604	nilfs_commit_chunk(page, mapping, 0, chunk_size);
605fail:
606	page_cache_release(page);
607	return err;
608}
609
610/*
611 * routine to check that the specified directory is empty (for rmdir)
612 */
613int nilfs_empty_dir(struct inode *inode)
614{
615	struct page *page = NULL;
616	unsigned long i, npages = dir_pages(inode);
617
618	for (i = 0; i < npages; i++) {
619		char *kaddr;
620		struct nilfs_dir_entry *de;
621
622		page = nilfs_get_page(inode, i);
623		if (IS_ERR(page))
624			continue;
625
626		kaddr = page_address(page);
627		de = (struct nilfs_dir_entry *)kaddr;
628		kaddr += nilfs_last_byte(inode, i) - NILFS_DIR_REC_LEN(1);
629
630		while ((char *)de <= kaddr) {
631			if (de->rec_len == 0) {
632				nilfs_error(inode->i_sb, __func__,
633					    "zero-length directory entry "
634					    "(kaddr=%p, de=%p)\n", kaddr, de);
635				goto not_empty;
636			}
637			if (de->inode != 0) {
638				/* check for . and .. */
639				if (de->name[0] != '.')
640					goto not_empty;
641				if (de->name_len > 2)
642					goto not_empty;
643				if (de->name_len < 2) {
644					if (de->inode !=
645					    cpu_to_le64(inode->i_ino))
646						goto not_empty;
647				} else if (de->name[1] != '.')
648					goto not_empty;
649			}
650			de = nilfs_next_entry(de);
651		}
652		nilfs_put_page(page);
653	}
654	return 1;
655
656not_empty:
657	nilfs_put_page(page);
658	return 0;
659}
660
661const struct file_operations nilfs_dir_operations = {
662	.llseek		= generic_file_llseek,
663	.read		= generic_read_dir,
664	.iterate	= nilfs_readdir,
665	.unlocked_ioctl	= nilfs_ioctl,
666#ifdef CONFIG_COMPAT
667	.compat_ioctl	= nilfs_compat_ioctl,
668#endif	/* CONFIG_COMPAT */
669	.fsync		= nilfs_sync_file,
670
671};
672