dir.c

  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(pos, len, len, folio);
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};