xref: /openbmc/linux/fs/fat/inode.c (revision 81d67439)
1 /*
2  *  linux/fs/fat/inode.c
3  *
4  *  Written 1992,1993 by Werner Almesberger
5  *  VFAT extensions by Gordon Chaffee, merged with msdos fs by Henrik Storner
6  *  Rewritten for the constant inumbers support by Al Viro
7  *
8  *  Fixes:
9  *
10  *	Max Cohan: Fixed invalid FSINFO offset when info_sector is 0
11  */
12 
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/time.h>
16 #include <linux/slab.h>
17 #include <linux/seq_file.h>
18 #include <linux/pagemap.h>
19 #include <linux/mpage.h>
20 #include <linux/buffer_head.h>
21 #include <linux/exportfs.h>
22 #include <linux/mount.h>
23 #include <linux/vfs.h>
24 #include <linux/parser.h>
25 #include <linux/uio.h>
26 #include <linux/writeback.h>
27 #include <linux/log2.h>
28 #include <linux/hash.h>
29 #include <asm/unaligned.h>
30 #include "fat.h"
31 
32 #ifndef CONFIG_FAT_DEFAULT_IOCHARSET
33 /* if user don't select VFAT, this is undefined. */
34 #define CONFIG_FAT_DEFAULT_IOCHARSET	""
35 #endif
36 
37 static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
38 static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
39 
40 
41 static int fat_add_cluster(struct inode *inode)
42 {
43 	int err, cluster;
44 
45 	err = fat_alloc_clusters(inode, &cluster, 1);
46 	if (err)
47 		return err;
48 	/* FIXME: this cluster should be added after data of this
49 	 * cluster is writed */
50 	err = fat_chain_add(inode, cluster, 1);
51 	if (err)
52 		fat_free_clusters(inode, cluster);
53 	return err;
54 }
55 
56 static inline int __fat_get_block(struct inode *inode, sector_t iblock,
57 				  unsigned long *max_blocks,
58 				  struct buffer_head *bh_result, int create)
59 {
60 	struct super_block *sb = inode->i_sb;
61 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
62 	unsigned long mapped_blocks;
63 	sector_t phys;
64 	int err, offset;
65 
66 	err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
67 	if (err)
68 		return err;
69 	if (phys) {
70 		map_bh(bh_result, sb, phys);
71 		*max_blocks = min(mapped_blocks, *max_blocks);
72 		return 0;
73 	}
74 	if (!create)
75 		return 0;
76 
77 	if (iblock != MSDOS_I(inode)->mmu_private >> sb->s_blocksize_bits) {
78 		fat_fs_error(sb, "corrupted file size (i_pos %lld, %lld)",
79 			MSDOS_I(inode)->i_pos, MSDOS_I(inode)->mmu_private);
80 		return -EIO;
81 	}
82 
83 	offset = (unsigned long)iblock & (sbi->sec_per_clus - 1);
84 	if (!offset) {
85 		/* TODO: multiple cluster allocation would be desirable. */
86 		err = fat_add_cluster(inode);
87 		if (err)
88 			return err;
89 	}
90 	/* available blocks on this cluster */
91 	mapped_blocks = sbi->sec_per_clus - offset;
92 
93 	*max_blocks = min(mapped_blocks, *max_blocks);
94 	MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
95 
96 	err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
97 	if (err)
98 		return err;
99 
100 	BUG_ON(!phys);
101 	BUG_ON(*max_blocks != mapped_blocks);
102 	set_buffer_new(bh_result);
103 	map_bh(bh_result, sb, phys);
104 
105 	return 0;
106 }
107 
108 static int fat_get_block(struct inode *inode, sector_t iblock,
109 			 struct buffer_head *bh_result, int create)
110 {
111 	struct super_block *sb = inode->i_sb;
112 	unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
113 	int err;
114 
115 	err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
116 	if (err)
117 		return err;
118 	bh_result->b_size = max_blocks << sb->s_blocksize_bits;
119 	return 0;
120 }
121 
122 static int fat_writepage(struct page *page, struct writeback_control *wbc)
123 {
124 	return block_write_full_page(page, fat_get_block, wbc);
125 }
126 
127 static int fat_writepages(struct address_space *mapping,
128 			  struct writeback_control *wbc)
129 {
130 	return mpage_writepages(mapping, wbc, fat_get_block);
131 }
132 
133 static int fat_readpage(struct file *file, struct page *page)
134 {
135 	return mpage_readpage(page, fat_get_block);
136 }
137 
138 static int fat_readpages(struct file *file, struct address_space *mapping,
139 			 struct list_head *pages, unsigned nr_pages)
140 {
141 	return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
142 }
143 
144 static void fat_write_failed(struct address_space *mapping, loff_t to)
145 {
146 	struct inode *inode = mapping->host;
147 
148 	if (to > inode->i_size) {
149 		truncate_pagecache(inode, to, inode->i_size);
150 		fat_truncate_blocks(inode, inode->i_size);
151 	}
152 }
153 
154 static int fat_write_begin(struct file *file, struct address_space *mapping,
155 			loff_t pos, unsigned len, unsigned flags,
156 			struct page **pagep, void **fsdata)
157 {
158 	int err;
159 
160 	*pagep = NULL;
161 	err = cont_write_begin(file, mapping, pos, len, flags,
162 				pagep, fsdata, fat_get_block,
163 				&MSDOS_I(mapping->host)->mmu_private);
164 	if (err < 0)
165 		fat_write_failed(mapping, pos + len);
166 	return err;
167 }
168 
169 static int fat_write_end(struct file *file, struct address_space *mapping,
170 			loff_t pos, unsigned len, unsigned copied,
171 			struct page *pagep, void *fsdata)
172 {
173 	struct inode *inode = mapping->host;
174 	int err;
175 	err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
176 	if (err < len)
177 		fat_write_failed(mapping, pos + len);
178 	if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
179 		inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
180 		MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
181 		mark_inode_dirty(inode);
182 	}
183 	return err;
184 }
185 
186 static ssize_t fat_direct_IO(int rw, struct kiocb *iocb,
187 			     const struct iovec *iov,
188 			     loff_t offset, unsigned long nr_segs)
189 {
190 	struct file *file = iocb->ki_filp;
191 	struct address_space *mapping = file->f_mapping;
192 	struct inode *inode = mapping->host;
193 	ssize_t ret;
194 
195 	if (rw == WRITE) {
196 		/*
197 		 * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
198 		 * so we need to update the ->mmu_private to block boundary.
199 		 *
200 		 * But we must fill the remaining area or hole by nul for
201 		 * updating ->mmu_private.
202 		 *
203 		 * Return 0, and fallback to normal buffered write.
204 		 */
205 		loff_t size = offset + iov_length(iov, nr_segs);
206 		if (MSDOS_I(inode)->mmu_private < size)
207 			return 0;
208 	}
209 
210 	/*
211 	 * FAT need to use the DIO_LOCKING for avoiding the race
212 	 * condition of fat_get_block() and ->truncate().
213 	 */
214 	ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
215 				 fat_get_block);
216 	if (ret < 0 && (rw & WRITE))
217 		fat_write_failed(mapping, offset + iov_length(iov, nr_segs));
218 
219 	return ret;
220 }
221 
222 static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
223 {
224 	sector_t blocknr;
225 
226 	/* fat_get_cluster() assumes the requested blocknr isn't truncated. */
227 	down_read(&MSDOS_I(mapping->host)->truncate_lock);
228 	blocknr = generic_block_bmap(mapping, block, fat_get_block);
229 	up_read(&MSDOS_I(mapping->host)->truncate_lock);
230 
231 	return blocknr;
232 }
233 
234 static const struct address_space_operations fat_aops = {
235 	.readpage	= fat_readpage,
236 	.readpages	= fat_readpages,
237 	.writepage	= fat_writepage,
238 	.writepages	= fat_writepages,
239 	.write_begin	= fat_write_begin,
240 	.write_end	= fat_write_end,
241 	.direct_IO	= fat_direct_IO,
242 	.bmap		= _fat_bmap
243 };
244 
245 /*
246  * New FAT inode stuff. We do the following:
247  *	a) i_ino is constant and has nothing with on-disk location.
248  *	b) FAT manages its own cache of directory entries.
249  *	c) *This* cache is indexed by on-disk location.
250  *	d) inode has an associated directory entry, all right, but
251  *		it may be unhashed.
252  *	e) currently entries are stored within struct inode. That should
253  *		change.
254  *	f) we deal with races in the following way:
255  *		1. readdir() and lookup() do FAT-dir-cache lookup.
256  *		2. rename() unhashes the F-d-c entry and rehashes it in
257  *			a new place.
258  *		3. unlink() and rmdir() unhash F-d-c entry.
259  *		4. fat_write_inode() checks whether the thing is unhashed.
260  *			If it is we silently return. If it isn't we do bread(),
261  *			check if the location is still valid and retry if it
262  *			isn't. Otherwise we do changes.
263  *		5. Spinlock is used to protect hash/unhash/location check/lookup
264  *		6. fat_evict_inode() unhashes the F-d-c entry.
265  *		7. lookup() and readdir() do igrab() if they find a F-d-c entry
266  *			and consider negative result as cache miss.
267  */
268 
269 static void fat_hash_init(struct super_block *sb)
270 {
271 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
272 	int i;
273 
274 	spin_lock_init(&sbi->inode_hash_lock);
275 	for (i = 0; i < FAT_HASH_SIZE; i++)
276 		INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
277 }
278 
279 static inline unsigned long fat_hash(loff_t i_pos)
280 {
281 	return hash_32(i_pos, FAT_HASH_BITS);
282 }
283 
284 void fat_attach(struct inode *inode, loff_t i_pos)
285 {
286 	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
287 	struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
288 
289 	spin_lock(&sbi->inode_hash_lock);
290 	MSDOS_I(inode)->i_pos = i_pos;
291 	hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
292 	spin_unlock(&sbi->inode_hash_lock);
293 }
294 EXPORT_SYMBOL_GPL(fat_attach);
295 
296 void fat_detach(struct inode *inode)
297 {
298 	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
299 	spin_lock(&sbi->inode_hash_lock);
300 	MSDOS_I(inode)->i_pos = 0;
301 	hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
302 	spin_unlock(&sbi->inode_hash_lock);
303 }
304 EXPORT_SYMBOL_GPL(fat_detach);
305 
306 struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
307 {
308 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
309 	struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
310 	struct hlist_node *_p;
311 	struct msdos_inode_info *i;
312 	struct inode *inode = NULL;
313 
314 	spin_lock(&sbi->inode_hash_lock);
315 	hlist_for_each_entry(i, _p, head, i_fat_hash) {
316 		BUG_ON(i->vfs_inode.i_sb != sb);
317 		if (i->i_pos != i_pos)
318 			continue;
319 		inode = igrab(&i->vfs_inode);
320 		if (inode)
321 			break;
322 	}
323 	spin_unlock(&sbi->inode_hash_lock);
324 	return inode;
325 }
326 
327 static int is_exec(unsigned char *extension)
328 {
329 	unsigned char *exe_extensions = "EXECOMBAT", *walk;
330 
331 	for (walk = exe_extensions; *walk; walk += 3)
332 		if (!strncmp(extension, walk, 3))
333 			return 1;
334 	return 0;
335 }
336 
337 static int fat_calc_dir_size(struct inode *inode)
338 {
339 	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
340 	int ret, fclus, dclus;
341 
342 	inode->i_size = 0;
343 	if (MSDOS_I(inode)->i_start == 0)
344 		return 0;
345 
346 	ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
347 	if (ret < 0)
348 		return ret;
349 	inode->i_size = (fclus + 1) << sbi->cluster_bits;
350 
351 	return 0;
352 }
353 
354 /* doesn't deal with root inode */
355 static int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
356 {
357 	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
358 	int error;
359 
360 	MSDOS_I(inode)->i_pos = 0;
361 	inode->i_uid = sbi->options.fs_uid;
362 	inode->i_gid = sbi->options.fs_gid;
363 	inode->i_version++;
364 	inode->i_generation = get_seconds();
365 
366 	if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
367 		inode->i_generation &= ~1;
368 		inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
369 		inode->i_op = sbi->dir_ops;
370 		inode->i_fop = &fat_dir_operations;
371 
372 		MSDOS_I(inode)->i_start = le16_to_cpu(de->start);
373 		if (sbi->fat_bits == 32)
374 			MSDOS_I(inode)->i_start |= (le16_to_cpu(de->starthi) << 16);
375 
376 		MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
377 		error = fat_calc_dir_size(inode);
378 		if (error < 0)
379 			return error;
380 		MSDOS_I(inode)->mmu_private = inode->i_size;
381 
382 		inode->i_nlink = fat_subdirs(inode);
383 	} else { /* not a directory */
384 		inode->i_generation |= 1;
385 		inode->i_mode = fat_make_mode(sbi, de->attr,
386 			((sbi->options.showexec && !is_exec(de->name + 8))
387 			 ? S_IRUGO|S_IWUGO : S_IRWXUGO));
388 		MSDOS_I(inode)->i_start = le16_to_cpu(de->start);
389 		if (sbi->fat_bits == 32)
390 			MSDOS_I(inode)->i_start |= (le16_to_cpu(de->starthi) << 16);
391 
392 		MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
393 		inode->i_size = le32_to_cpu(de->size);
394 		inode->i_op = &fat_file_inode_operations;
395 		inode->i_fop = &fat_file_operations;
396 		inode->i_mapping->a_ops = &fat_aops;
397 		MSDOS_I(inode)->mmu_private = inode->i_size;
398 	}
399 	if (de->attr & ATTR_SYS) {
400 		if (sbi->options.sys_immutable)
401 			inode->i_flags |= S_IMMUTABLE;
402 	}
403 	fat_save_attrs(inode, de->attr);
404 
405 	inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
406 			   & ~((loff_t)sbi->cluster_size - 1)) >> 9;
407 
408 	fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
409 	if (sbi->options.isvfat) {
410 		fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
411 				  de->cdate, de->ctime_cs);
412 		fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
413 	} else
414 		inode->i_ctime = inode->i_atime = inode->i_mtime;
415 
416 	return 0;
417 }
418 
419 struct inode *fat_build_inode(struct super_block *sb,
420 			struct msdos_dir_entry *de, loff_t i_pos)
421 {
422 	struct inode *inode;
423 	int err;
424 
425 	inode = fat_iget(sb, i_pos);
426 	if (inode)
427 		goto out;
428 	inode = new_inode(sb);
429 	if (!inode) {
430 		inode = ERR_PTR(-ENOMEM);
431 		goto out;
432 	}
433 	inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
434 	inode->i_version = 1;
435 	err = fat_fill_inode(inode, de);
436 	if (err) {
437 		iput(inode);
438 		inode = ERR_PTR(err);
439 		goto out;
440 	}
441 	fat_attach(inode, i_pos);
442 	insert_inode_hash(inode);
443 out:
444 	return inode;
445 }
446 
447 EXPORT_SYMBOL_GPL(fat_build_inode);
448 
449 static void fat_evict_inode(struct inode *inode)
450 {
451 	truncate_inode_pages(&inode->i_data, 0);
452 	if (!inode->i_nlink) {
453 		inode->i_size = 0;
454 		fat_truncate_blocks(inode, 0);
455 	}
456 	invalidate_inode_buffers(inode);
457 	end_writeback(inode);
458 	fat_cache_inval_inode(inode);
459 	fat_detach(inode);
460 }
461 
462 static void fat_write_super(struct super_block *sb)
463 {
464 	lock_super(sb);
465 	sb->s_dirt = 0;
466 
467 	if (!(sb->s_flags & MS_RDONLY))
468 		fat_clusters_flush(sb);
469 	unlock_super(sb);
470 }
471 
472 static int fat_sync_fs(struct super_block *sb, int wait)
473 {
474 	int err = 0;
475 
476 	if (sb->s_dirt) {
477 		lock_super(sb);
478 		sb->s_dirt = 0;
479 		err = fat_clusters_flush(sb);
480 		unlock_super(sb);
481 	}
482 
483 	return err;
484 }
485 
486 static void fat_put_super(struct super_block *sb)
487 {
488 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
489 
490 	if (sb->s_dirt)
491 		fat_write_super(sb);
492 
493 	iput(sbi->fat_inode);
494 
495 	unload_nls(sbi->nls_disk);
496 	unload_nls(sbi->nls_io);
497 
498 	if (sbi->options.iocharset != fat_default_iocharset)
499 		kfree(sbi->options.iocharset);
500 
501 	sb->s_fs_info = NULL;
502 	kfree(sbi);
503 }
504 
505 static struct kmem_cache *fat_inode_cachep;
506 
507 static struct inode *fat_alloc_inode(struct super_block *sb)
508 {
509 	struct msdos_inode_info *ei;
510 	ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
511 	if (!ei)
512 		return NULL;
513 
514 	init_rwsem(&ei->truncate_lock);
515 	return &ei->vfs_inode;
516 }
517 
518 static void fat_i_callback(struct rcu_head *head)
519 {
520 	struct inode *inode = container_of(head, struct inode, i_rcu);
521 	INIT_LIST_HEAD(&inode->i_dentry);
522 	kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
523 }
524 
525 static void fat_destroy_inode(struct inode *inode)
526 {
527 	call_rcu(&inode->i_rcu, fat_i_callback);
528 }
529 
530 static void init_once(void *foo)
531 {
532 	struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
533 
534 	spin_lock_init(&ei->cache_lru_lock);
535 	ei->nr_caches = 0;
536 	ei->cache_valid_id = FAT_CACHE_VALID + 1;
537 	INIT_LIST_HEAD(&ei->cache_lru);
538 	INIT_HLIST_NODE(&ei->i_fat_hash);
539 	inode_init_once(&ei->vfs_inode);
540 }
541 
542 static int __init fat_init_inodecache(void)
543 {
544 	fat_inode_cachep = kmem_cache_create("fat_inode_cache",
545 					     sizeof(struct msdos_inode_info),
546 					     0, (SLAB_RECLAIM_ACCOUNT|
547 						SLAB_MEM_SPREAD),
548 					     init_once);
549 	if (fat_inode_cachep == NULL)
550 		return -ENOMEM;
551 	return 0;
552 }
553 
554 static void __exit fat_destroy_inodecache(void)
555 {
556 	kmem_cache_destroy(fat_inode_cachep);
557 }
558 
559 static int fat_remount(struct super_block *sb, int *flags, char *data)
560 {
561 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
562 	*flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
563 	return 0;
564 }
565 
566 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
567 {
568 	struct super_block *sb = dentry->d_sb;
569 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
570 	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
571 
572 	/* If the count of free cluster is still unknown, counts it here. */
573 	if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
574 		int err = fat_count_free_clusters(dentry->d_sb);
575 		if (err)
576 			return err;
577 	}
578 
579 	buf->f_type = dentry->d_sb->s_magic;
580 	buf->f_bsize = sbi->cluster_size;
581 	buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
582 	buf->f_bfree = sbi->free_clusters;
583 	buf->f_bavail = sbi->free_clusters;
584 	buf->f_fsid.val[0] = (u32)id;
585 	buf->f_fsid.val[1] = (u32)(id >> 32);
586 	buf->f_namelen =
587 		(sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
588 
589 	return 0;
590 }
591 
592 static inline loff_t fat_i_pos_read(struct msdos_sb_info *sbi,
593 				    struct inode *inode)
594 {
595 	loff_t i_pos;
596 #if BITS_PER_LONG == 32
597 	spin_lock(&sbi->inode_hash_lock);
598 #endif
599 	i_pos = MSDOS_I(inode)->i_pos;
600 #if BITS_PER_LONG == 32
601 	spin_unlock(&sbi->inode_hash_lock);
602 #endif
603 	return i_pos;
604 }
605 
606 static int __fat_write_inode(struct inode *inode, int wait)
607 {
608 	struct super_block *sb = inode->i_sb;
609 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
610 	struct buffer_head *bh;
611 	struct msdos_dir_entry *raw_entry;
612 	loff_t i_pos;
613 	int err;
614 
615 	if (inode->i_ino == MSDOS_ROOT_INO)
616 		return 0;
617 
618 retry:
619 	i_pos = fat_i_pos_read(sbi, inode);
620 	if (!i_pos)
621 		return 0;
622 
623 	bh = sb_bread(sb, i_pos >> sbi->dir_per_block_bits);
624 	if (!bh) {
625 		fat_msg(sb, KERN_ERR, "unable to read inode block "
626 		       "for updating (i_pos %lld)", i_pos);
627 		return -EIO;
628 	}
629 	spin_lock(&sbi->inode_hash_lock);
630 	if (i_pos != MSDOS_I(inode)->i_pos) {
631 		spin_unlock(&sbi->inode_hash_lock);
632 		brelse(bh);
633 		goto retry;
634 	}
635 
636 	raw_entry = &((struct msdos_dir_entry *) (bh->b_data))
637 	    [i_pos & (sbi->dir_per_block - 1)];
638 	if (S_ISDIR(inode->i_mode))
639 		raw_entry->size = 0;
640 	else
641 		raw_entry->size = cpu_to_le32(inode->i_size);
642 	raw_entry->attr = fat_make_attrs(inode);
643 	raw_entry->start = cpu_to_le16(MSDOS_I(inode)->i_logstart);
644 	raw_entry->starthi = cpu_to_le16(MSDOS_I(inode)->i_logstart >> 16);
645 	fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
646 			  &raw_entry->date, NULL);
647 	if (sbi->options.isvfat) {
648 		__le16 atime;
649 		fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
650 				  &raw_entry->cdate, &raw_entry->ctime_cs);
651 		fat_time_unix2fat(sbi, &inode->i_atime, &atime,
652 				  &raw_entry->adate, NULL);
653 	}
654 	spin_unlock(&sbi->inode_hash_lock);
655 	mark_buffer_dirty(bh);
656 	err = 0;
657 	if (wait)
658 		err = sync_dirty_buffer(bh);
659 	brelse(bh);
660 	return err;
661 }
662 
663 static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
664 {
665 	return __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
666 }
667 
668 int fat_sync_inode(struct inode *inode)
669 {
670 	return __fat_write_inode(inode, 1);
671 }
672 
673 EXPORT_SYMBOL_GPL(fat_sync_inode);
674 
675 static int fat_show_options(struct seq_file *m, struct vfsmount *mnt);
676 static const struct super_operations fat_sops = {
677 	.alloc_inode	= fat_alloc_inode,
678 	.destroy_inode	= fat_destroy_inode,
679 	.write_inode	= fat_write_inode,
680 	.evict_inode	= fat_evict_inode,
681 	.put_super	= fat_put_super,
682 	.write_super	= fat_write_super,
683 	.sync_fs	= fat_sync_fs,
684 	.statfs		= fat_statfs,
685 	.remount_fs	= fat_remount,
686 
687 	.show_options	= fat_show_options,
688 };
689 
690 /*
691  * a FAT file handle with fhtype 3 is
692  *  0/  i_ino - for fast, reliable lookup if still in the cache
693  *  1/  i_generation - to see if i_ino is still valid
694  *          bit 0 == 0 iff directory
695  *  2/  i_pos(8-39) - if ino has changed, but still in cache
696  *  3/  i_pos(4-7)|i_logstart - to semi-verify inode found at i_pos
697  *  4/  i_pos(0-3)|parent->i_logstart - maybe used to hunt for the file on disc
698  *
699  * Hack for NFSv2: Maximum FAT entry number is 28bits and maximum
700  * i_pos is 40bits (blocknr(32) + dir offset(8)), so two 4bits
701  * of i_logstart is used to store the directory entry offset.
702  */
703 
704 static struct dentry *fat_fh_to_dentry(struct super_block *sb,
705 		struct fid *fid, int fh_len, int fh_type)
706 {
707 	struct inode *inode = NULL;
708 	u32 *fh = fid->raw;
709 
710 	if (fh_len < 5 || fh_type != 3)
711 		return NULL;
712 
713 	inode = ilookup(sb, fh[0]);
714 	if (!inode || inode->i_generation != fh[1]) {
715 		if (inode)
716 			iput(inode);
717 		inode = NULL;
718 	}
719 	if (!inode) {
720 		loff_t i_pos;
721 		int i_logstart = fh[3] & 0x0fffffff;
722 
723 		i_pos = (loff_t)fh[2] << 8;
724 		i_pos |= ((fh[3] >> 24) & 0xf0) | (fh[4] >> 28);
725 
726 		/* try 2 - see if i_pos is in F-d-c
727 		 * require i_logstart to be the same
728 		 * Will fail if you truncate and then re-write
729 		 */
730 
731 		inode = fat_iget(sb, i_pos);
732 		if (inode && MSDOS_I(inode)->i_logstart != i_logstart) {
733 			iput(inode);
734 			inode = NULL;
735 		}
736 	}
737 
738 	/*
739 	 * For now, do nothing if the inode is not found.
740 	 *
741 	 * What we could do is:
742 	 *
743 	 *	- follow the file starting at fh[4], and record the ".." entry,
744 	 *	  and the name of the fh[2] entry.
745 	 *	- then follow the ".." file finding the next step up.
746 	 *
747 	 * This way we build a path to the root of the tree. If this works, we
748 	 * lookup the path and so get this inode into the cache.  Finally try
749 	 * the fat_iget lookup again.  If that fails, then we are totally out
750 	 * of luck.  But all that is for another day
751 	 */
752 	return d_obtain_alias(inode);
753 }
754 
755 static int
756 fat_encode_fh(struct dentry *de, __u32 *fh, int *lenp, int connectable)
757 {
758 	int len = *lenp;
759 	struct inode *inode =  de->d_inode;
760 	u32 ipos_h, ipos_m, ipos_l;
761 
762 	if (len < 5) {
763 		*lenp = 5;
764 		return 255; /* no room */
765 	}
766 
767 	ipos_h = MSDOS_I(inode)->i_pos >> 8;
768 	ipos_m = (MSDOS_I(inode)->i_pos & 0xf0) << 24;
769 	ipos_l = (MSDOS_I(inode)->i_pos & 0x0f) << 28;
770 	*lenp = 5;
771 	fh[0] = inode->i_ino;
772 	fh[1] = inode->i_generation;
773 	fh[2] = ipos_h;
774 	fh[3] = ipos_m | MSDOS_I(inode)->i_logstart;
775 	spin_lock(&de->d_lock);
776 	fh[4] = ipos_l | MSDOS_I(de->d_parent->d_inode)->i_logstart;
777 	spin_unlock(&de->d_lock);
778 	return 3;
779 }
780 
781 static struct dentry *fat_get_parent(struct dentry *child)
782 {
783 	struct super_block *sb = child->d_sb;
784 	struct buffer_head *bh;
785 	struct msdos_dir_entry *de;
786 	loff_t i_pos;
787 	struct dentry *parent;
788 	struct inode *inode;
789 	int err;
790 
791 	lock_super(sb);
792 
793 	err = fat_get_dotdot_entry(child->d_inode, &bh, &de, &i_pos);
794 	if (err) {
795 		parent = ERR_PTR(err);
796 		goto out;
797 	}
798 	inode = fat_build_inode(sb, de, i_pos);
799 	brelse(bh);
800 
801 	parent = d_obtain_alias(inode);
802 out:
803 	unlock_super(sb);
804 
805 	return parent;
806 }
807 
808 static const struct export_operations fat_export_ops = {
809 	.encode_fh	= fat_encode_fh,
810 	.fh_to_dentry	= fat_fh_to_dentry,
811 	.get_parent	= fat_get_parent,
812 };
813 
814 static int fat_show_options(struct seq_file *m, struct vfsmount *mnt)
815 {
816 	struct msdos_sb_info *sbi = MSDOS_SB(mnt->mnt_sb);
817 	struct fat_mount_options *opts = &sbi->options;
818 	int isvfat = opts->isvfat;
819 
820 	if (opts->fs_uid != 0)
821 		seq_printf(m, ",uid=%u", opts->fs_uid);
822 	if (opts->fs_gid != 0)
823 		seq_printf(m, ",gid=%u", opts->fs_gid);
824 	seq_printf(m, ",fmask=%04o", opts->fs_fmask);
825 	seq_printf(m, ",dmask=%04o", opts->fs_dmask);
826 	if (opts->allow_utime)
827 		seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
828 	if (sbi->nls_disk)
829 		seq_printf(m, ",codepage=%s", sbi->nls_disk->charset);
830 	if (isvfat) {
831 		if (sbi->nls_io)
832 			seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
833 
834 		switch (opts->shortname) {
835 		case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
836 			seq_puts(m, ",shortname=win95");
837 			break;
838 		case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
839 			seq_puts(m, ",shortname=winnt");
840 			break;
841 		case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
842 			seq_puts(m, ",shortname=mixed");
843 			break;
844 		case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
845 			seq_puts(m, ",shortname=lower");
846 			break;
847 		default:
848 			seq_puts(m, ",shortname=unknown");
849 			break;
850 		}
851 	}
852 	if (opts->name_check != 'n')
853 		seq_printf(m, ",check=%c", opts->name_check);
854 	if (opts->usefree)
855 		seq_puts(m, ",usefree");
856 	if (opts->quiet)
857 		seq_puts(m, ",quiet");
858 	if (opts->showexec)
859 		seq_puts(m, ",showexec");
860 	if (opts->sys_immutable)
861 		seq_puts(m, ",sys_immutable");
862 	if (!isvfat) {
863 		if (opts->dotsOK)
864 			seq_puts(m, ",dotsOK=yes");
865 		if (opts->nocase)
866 			seq_puts(m, ",nocase");
867 	} else {
868 		if (opts->utf8)
869 			seq_puts(m, ",utf8");
870 		if (opts->unicode_xlate)
871 			seq_puts(m, ",uni_xlate");
872 		if (!opts->numtail)
873 			seq_puts(m, ",nonumtail");
874 		if (opts->rodir)
875 			seq_puts(m, ",rodir");
876 	}
877 	if (opts->flush)
878 		seq_puts(m, ",flush");
879 	if (opts->tz_utc)
880 		seq_puts(m, ",tz=UTC");
881 	if (opts->errors == FAT_ERRORS_CONT)
882 		seq_puts(m, ",errors=continue");
883 	else if (opts->errors == FAT_ERRORS_PANIC)
884 		seq_puts(m, ",errors=panic");
885 	else
886 		seq_puts(m, ",errors=remount-ro");
887 	if (opts->discard)
888 		seq_puts(m, ",discard");
889 
890 	return 0;
891 }
892 
893 enum {
894 	Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
895 	Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
896 	Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
897 	Opt_immutable, Opt_dots, Opt_nodots,
898 	Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
899 	Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
900 	Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
901 	Opt_obsolate, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
902 	Opt_err_panic, Opt_err_ro, Opt_discard, Opt_err,
903 };
904 
905 static const match_table_t fat_tokens = {
906 	{Opt_check_r, "check=relaxed"},
907 	{Opt_check_s, "check=strict"},
908 	{Opt_check_n, "check=normal"},
909 	{Opt_check_r, "check=r"},
910 	{Opt_check_s, "check=s"},
911 	{Opt_check_n, "check=n"},
912 	{Opt_uid, "uid=%u"},
913 	{Opt_gid, "gid=%u"},
914 	{Opt_umask, "umask=%o"},
915 	{Opt_dmask, "dmask=%o"},
916 	{Opt_fmask, "fmask=%o"},
917 	{Opt_allow_utime, "allow_utime=%o"},
918 	{Opt_codepage, "codepage=%u"},
919 	{Opt_usefree, "usefree"},
920 	{Opt_nocase, "nocase"},
921 	{Opt_quiet, "quiet"},
922 	{Opt_showexec, "showexec"},
923 	{Opt_debug, "debug"},
924 	{Opt_immutable, "sys_immutable"},
925 	{Opt_flush, "flush"},
926 	{Opt_tz_utc, "tz=UTC"},
927 	{Opt_err_cont, "errors=continue"},
928 	{Opt_err_panic, "errors=panic"},
929 	{Opt_err_ro, "errors=remount-ro"},
930 	{Opt_discard, "discard"},
931 	{Opt_obsolate, "conv=binary"},
932 	{Opt_obsolate, "conv=text"},
933 	{Opt_obsolate, "conv=auto"},
934 	{Opt_obsolate, "conv=b"},
935 	{Opt_obsolate, "conv=t"},
936 	{Opt_obsolate, "conv=a"},
937 	{Opt_obsolate, "fat=%u"},
938 	{Opt_obsolate, "blocksize=%u"},
939 	{Opt_obsolate, "cvf_format=%20s"},
940 	{Opt_obsolate, "cvf_options=%100s"},
941 	{Opt_obsolate, "posix"},
942 	{Opt_err, NULL},
943 };
944 static const match_table_t msdos_tokens = {
945 	{Opt_nodots, "nodots"},
946 	{Opt_nodots, "dotsOK=no"},
947 	{Opt_dots, "dots"},
948 	{Opt_dots, "dotsOK=yes"},
949 	{Opt_err, NULL}
950 };
951 static const match_table_t vfat_tokens = {
952 	{Opt_charset, "iocharset=%s"},
953 	{Opt_shortname_lower, "shortname=lower"},
954 	{Opt_shortname_win95, "shortname=win95"},
955 	{Opt_shortname_winnt, "shortname=winnt"},
956 	{Opt_shortname_mixed, "shortname=mixed"},
957 	{Opt_utf8_no, "utf8=0"},		/* 0 or no or false */
958 	{Opt_utf8_no, "utf8=no"},
959 	{Opt_utf8_no, "utf8=false"},
960 	{Opt_utf8_yes, "utf8=1"},		/* empty or 1 or yes or true */
961 	{Opt_utf8_yes, "utf8=yes"},
962 	{Opt_utf8_yes, "utf8=true"},
963 	{Opt_utf8_yes, "utf8"},
964 	{Opt_uni_xl_no, "uni_xlate=0"},		/* 0 or no or false */
965 	{Opt_uni_xl_no, "uni_xlate=no"},
966 	{Opt_uni_xl_no, "uni_xlate=false"},
967 	{Opt_uni_xl_yes, "uni_xlate=1"},	/* empty or 1 or yes or true */
968 	{Opt_uni_xl_yes, "uni_xlate=yes"},
969 	{Opt_uni_xl_yes, "uni_xlate=true"},
970 	{Opt_uni_xl_yes, "uni_xlate"},
971 	{Opt_nonumtail_no, "nonumtail=0"},	/* 0 or no or false */
972 	{Opt_nonumtail_no, "nonumtail=no"},
973 	{Opt_nonumtail_no, "nonumtail=false"},
974 	{Opt_nonumtail_yes, "nonumtail=1"},	/* empty or 1 or yes or true */
975 	{Opt_nonumtail_yes, "nonumtail=yes"},
976 	{Opt_nonumtail_yes, "nonumtail=true"},
977 	{Opt_nonumtail_yes, "nonumtail"},
978 	{Opt_rodir, "rodir"},
979 	{Opt_err, NULL}
980 };
981 
982 static int parse_options(struct super_block *sb, char *options, int is_vfat,
983 			 int silent, int *debug, struct fat_mount_options *opts)
984 {
985 	char *p;
986 	substring_t args[MAX_OPT_ARGS];
987 	int option;
988 	char *iocharset;
989 
990 	opts->isvfat = is_vfat;
991 
992 	opts->fs_uid = current_uid();
993 	opts->fs_gid = current_gid();
994 	opts->fs_fmask = opts->fs_dmask = current_umask();
995 	opts->allow_utime = -1;
996 	opts->codepage = fat_default_codepage;
997 	opts->iocharset = fat_default_iocharset;
998 	if (is_vfat) {
999 		opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1000 		opts->rodir = 0;
1001 	} else {
1002 		opts->shortname = 0;
1003 		opts->rodir = 1;
1004 	}
1005 	opts->name_check = 'n';
1006 	opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
1007 	opts->utf8 = opts->unicode_xlate = 0;
1008 	opts->numtail = 1;
1009 	opts->usefree = opts->nocase = 0;
1010 	opts->tz_utc = 0;
1011 	opts->errors = FAT_ERRORS_RO;
1012 	*debug = 0;
1013 
1014 	if (!options)
1015 		goto out;
1016 
1017 	while ((p = strsep(&options, ",")) != NULL) {
1018 		int token;
1019 		if (!*p)
1020 			continue;
1021 
1022 		token = match_token(p, fat_tokens, args);
1023 		if (token == Opt_err) {
1024 			if (is_vfat)
1025 				token = match_token(p, vfat_tokens, args);
1026 			else
1027 				token = match_token(p, msdos_tokens, args);
1028 		}
1029 		switch (token) {
1030 		case Opt_check_s:
1031 			opts->name_check = 's';
1032 			break;
1033 		case Opt_check_r:
1034 			opts->name_check = 'r';
1035 			break;
1036 		case Opt_check_n:
1037 			opts->name_check = 'n';
1038 			break;
1039 		case Opt_usefree:
1040 			opts->usefree = 1;
1041 			break;
1042 		case Opt_nocase:
1043 			if (!is_vfat)
1044 				opts->nocase = 1;
1045 			else {
1046 				/* for backward compatibility */
1047 				opts->shortname = VFAT_SFN_DISPLAY_WIN95
1048 					| VFAT_SFN_CREATE_WIN95;
1049 			}
1050 			break;
1051 		case Opt_quiet:
1052 			opts->quiet = 1;
1053 			break;
1054 		case Opt_showexec:
1055 			opts->showexec = 1;
1056 			break;
1057 		case Opt_debug:
1058 			*debug = 1;
1059 			break;
1060 		case Opt_immutable:
1061 			opts->sys_immutable = 1;
1062 			break;
1063 		case Opt_uid:
1064 			if (match_int(&args[0], &option))
1065 				return 0;
1066 			opts->fs_uid = option;
1067 			break;
1068 		case Opt_gid:
1069 			if (match_int(&args[0], &option))
1070 				return 0;
1071 			opts->fs_gid = option;
1072 			break;
1073 		case Opt_umask:
1074 			if (match_octal(&args[0], &option))
1075 				return 0;
1076 			opts->fs_fmask = opts->fs_dmask = option;
1077 			break;
1078 		case Opt_dmask:
1079 			if (match_octal(&args[0], &option))
1080 				return 0;
1081 			opts->fs_dmask = option;
1082 			break;
1083 		case Opt_fmask:
1084 			if (match_octal(&args[0], &option))
1085 				return 0;
1086 			opts->fs_fmask = option;
1087 			break;
1088 		case Opt_allow_utime:
1089 			if (match_octal(&args[0], &option))
1090 				return 0;
1091 			opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1092 			break;
1093 		case Opt_codepage:
1094 			if (match_int(&args[0], &option))
1095 				return 0;
1096 			opts->codepage = option;
1097 			break;
1098 		case Opt_flush:
1099 			opts->flush = 1;
1100 			break;
1101 		case Opt_tz_utc:
1102 			opts->tz_utc = 1;
1103 			break;
1104 		case Opt_err_cont:
1105 			opts->errors = FAT_ERRORS_CONT;
1106 			break;
1107 		case Opt_err_panic:
1108 			opts->errors = FAT_ERRORS_PANIC;
1109 			break;
1110 		case Opt_err_ro:
1111 			opts->errors = FAT_ERRORS_RO;
1112 			break;
1113 
1114 		/* msdos specific */
1115 		case Opt_dots:
1116 			opts->dotsOK = 1;
1117 			break;
1118 		case Opt_nodots:
1119 			opts->dotsOK = 0;
1120 			break;
1121 
1122 		/* vfat specific */
1123 		case Opt_charset:
1124 			if (opts->iocharset != fat_default_iocharset)
1125 				kfree(opts->iocharset);
1126 			iocharset = match_strdup(&args[0]);
1127 			if (!iocharset)
1128 				return -ENOMEM;
1129 			opts->iocharset = iocharset;
1130 			break;
1131 		case Opt_shortname_lower:
1132 			opts->shortname = VFAT_SFN_DISPLAY_LOWER
1133 					| VFAT_SFN_CREATE_WIN95;
1134 			break;
1135 		case Opt_shortname_win95:
1136 			opts->shortname = VFAT_SFN_DISPLAY_WIN95
1137 					| VFAT_SFN_CREATE_WIN95;
1138 			break;
1139 		case Opt_shortname_winnt:
1140 			opts->shortname = VFAT_SFN_DISPLAY_WINNT
1141 					| VFAT_SFN_CREATE_WINNT;
1142 			break;
1143 		case Opt_shortname_mixed:
1144 			opts->shortname = VFAT_SFN_DISPLAY_WINNT
1145 					| VFAT_SFN_CREATE_WIN95;
1146 			break;
1147 		case Opt_utf8_no:		/* 0 or no or false */
1148 			opts->utf8 = 0;
1149 			break;
1150 		case Opt_utf8_yes:		/* empty or 1 or yes or true */
1151 			opts->utf8 = 1;
1152 			break;
1153 		case Opt_uni_xl_no:		/* 0 or no or false */
1154 			opts->unicode_xlate = 0;
1155 			break;
1156 		case Opt_uni_xl_yes:		/* empty or 1 or yes or true */
1157 			opts->unicode_xlate = 1;
1158 			break;
1159 		case Opt_nonumtail_no:		/* 0 or no or false */
1160 			opts->numtail = 1;	/* negated option */
1161 			break;
1162 		case Opt_nonumtail_yes:		/* empty or 1 or yes or true */
1163 			opts->numtail = 0;	/* negated option */
1164 			break;
1165 		case Opt_rodir:
1166 			opts->rodir = 1;
1167 			break;
1168 		case Opt_discard:
1169 			opts->discard = 1;
1170 			break;
1171 
1172 		/* obsolete mount options */
1173 		case Opt_obsolate:
1174 			fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1175 			       "not supported now", p);
1176 			break;
1177 		/* unknown option */
1178 		default:
1179 			if (!silent) {
1180 				fat_msg(sb, KERN_ERR,
1181 				       "Unrecognized mount option \"%s\" "
1182 				       "or missing value", p);
1183 			}
1184 			return -EINVAL;
1185 		}
1186 	}
1187 
1188 out:
1189 	/* UTF-8 doesn't provide FAT semantics */
1190 	if (!strcmp(opts->iocharset, "utf8")) {
1191 		fat_msg(sb, KERN_ERR, "utf8 is not a recommended IO charset"
1192 		       " for FAT filesystems, filesystem will be "
1193 		       "case sensitive!\n");
1194 	}
1195 
1196 	/* If user doesn't specify allow_utime, it's initialized from dmask. */
1197 	if (opts->allow_utime == (unsigned short)-1)
1198 		opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1199 	if (opts->unicode_xlate)
1200 		opts->utf8 = 0;
1201 
1202 	return 0;
1203 }
1204 
1205 static int fat_read_root(struct inode *inode)
1206 {
1207 	struct super_block *sb = inode->i_sb;
1208 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
1209 	int error;
1210 
1211 	MSDOS_I(inode)->i_pos = 0;
1212 	inode->i_uid = sbi->options.fs_uid;
1213 	inode->i_gid = sbi->options.fs_gid;
1214 	inode->i_version++;
1215 	inode->i_generation = 0;
1216 	inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1217 	inode->i_op = sbi->dir_ops;
1218 	inode->i_fop = &fat_dir_operations;
1219 	if (sbi->fat_bits == 32) {
1220 		MSDOS_I(inode)->i_start = sbi->root_cluster;
1221 		error = fat_calc_dir_size(inode);
1222 		if (error < 0)
1223 			return error;
1224 	} else {
1225 		MSDOS_I(inode)->i_start = 0;
1226 		inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1227 	}
1228 	inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1229 			   & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1230 	MSDOS_I(inode)->i_logstart = 0;
1231 	MSDOS_I(inode)->mmu_private = inode->i_size;
1232 
1233 	fat_save_attrs(inode, ATTR_DIR);
1234 	inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1235 	inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1236 	inode->i_nlink = fat_subdirs(inode)+2;
1237 
1238 	return 0;
1239 }
1240 
1241 /*
1242  * Read the super block of an MS-DOS FS.
1243  */
1244 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1245 		   void (*setup)(struct super_block *))
1246 {
1247 	struct inode *root_inode = NULL, *fat_inode = NULL;
1248 	struct buffer_head *bh;
1249 	struct fat_boot_sector *b;
1250 	struct msdos_sb_info *sbi;
1251 	u16 logical_sector_size;
1252 	u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1253 	int debug;
1254 	unsigned int media;
1255 	long error;
1256 	char buf[50];
1257 
1258 	/*
1259 	 * GFP_KERNEL is ok here, because while we do hold the
1260 	 * supeblock lock, memory pressure can't call back into
1261 	 * the filesystem, since we're only just about to mount
1262 	 * it and have no inodes etc active!
1263 	 */
1264 	sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1265 	if (!sbi)
1266 		return -ENOMEM;
1267 	sb->s_fs_info = sbi;
1268 
1269 	sb->s_flags |= MS_NODIRATIME;
1270 	sb->s_magic = MSDOS_SUPER_MAGIC;
1271 	sb->s_op = &fat_sops;
1272 	sb->s_export_op = &fat_export_ops;
1273 	ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1274 			     DEFAULT_RATELIMIT_BURST);
1275 
1276 	error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1277 	if (error)
1278 		goto out_fail;
1279 
1280 	setup(sb); /* flavour-specific stuff that needs options */
1281 
1282 	error = -EIO;
1283 	sb_min_blocksize(sb, 512);
1284 	bh = sb_bread(sb, 0);
1285 	if (bh == NULL) {
1286 		fat_msg(sb, KERN_ERR, "unable to read boot sector");
1287 		goto out_fail;
1288 	}
1289 
1290 	b = (struct fat_boot_sector *) bh->b_data;
1291 	if (!b->reserved) {
1292 		if (!silent)
1293 			fat_msg(sb, KERN_ERR, "bogus number of reserved sectors");
1294 		brelse(bh);
1295 		goto out_invalid;
1296 	}
1297 	if (!b->fats) {
1298 		if (!silent)
1299 			fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1300 		brelse(bh);
1301 		goto out_invalid;
1302 	}
1303 
1304 	/*
1305 	 * Earlier we checked here that b->secs_track and b->head are nonzero,
1306 	 * but it turns out valid FAT filesystems can have zero there.
1307 	 */
1308 
1309 	media = b->media;
1310 	if (!fat_valid_media(media)) {
1311 		if (!silent)
1312 			fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1313 			       media);
1314 		brelse(bh);
1315 		goto out_invalid;
1316 	}
1317 	logical_sector_size = get_unaligned_le16(&b->sector_size);
1318 	if (!is_power_of_2(logical_sector_size)
1319 	    || (logical_sector_size < 512)
1320 	    || (logical_sector_size > 4096)) {
1321 		if (!silent)
1322 			fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1323 			       logical_sector_size);
1324 		brelse(bh);
1325 		goto out_invalid;
1326 	}
1327 	sbi->sec_per_clus = b->sec_per_clus;
1328 	if (!is_power_of_2(sbi->sec_per_clus)) {
1329 		if (!silent)
1330 			fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1331 			       sbi->sec_per_clus);
1332 		brelse(bh);
1333 		goto out_invalid;
1334 	}
1335 
1336 	if (logical_sector_size < sb->s_blocksize) {
1337 		fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1338 		       " (logical sector size = %u)", logical_sector_size);
1339 		brelse(bh);
1340 		goto out_fail;
1341 	}
1342 	if (logical_sector_size > sb->s_blocksize) {
1343 		brelse(bh);
1344 
1345 		if (!sb_set_blocksize(sb, logical_sector_size)) {
1346 			fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1347 			       logical_sector_size);
1348 			goto out_fail;
1349 		}
1350 		bh = sb_bread(sb, 0);
1351 		if (bh == NULL) {
1352 			fat_msg(sb, KERN_ERR, "unable to read boot sector"
1353 			       " (logical sector size = %lu)",
1354 			       sb->s_blocksize);
1355 			goto out_fail;
1356 		}
1357 		b = (struct fat_boot_sector *) bh->b_data;
1358 	}
1359 
1360 	sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1361 	sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1362 	sbi->fats = b->fats;
1363 	sbi->fat_bits = 0;		/* Don't know yet */
1364 	sbi->fat_start = le16_to_cpu(b->reserved);
1365 	sbi->fat_length = le16_to_cpu(b->fat_length);
1366 	sbi->root_cluster = 0;
1367 	sbi->free_clusters = -1;	/* Don't know yet */
1368 	sbi->free_clus_valid = 0;
1369 	sbi->prev_free = FAT_START_ENT;
1370 
1371 	if (!sbi->fat_length && b->fat32_length) {
1372 		struct fat_boot_fsinfo *fsinfo;
1373 		struct buffer_head *fsinfo_bh;
1374 
1375 		/* Must be FAT32 */
1376 		sbi->fat_bits = 32;
1377 		sbi->fat_length = le32_to_cpu(b->fat32_length);
1378 		sbi->root_cluster = le32_to_cpu(b->root_cluster);
1379 
1380 		sb->s_maxbytes = 0xffffffff;
1381 
1382 		/* MC - if info_sector is 0, don't multiply by 0 */
1383 		sbi->fsinfo_sector = le16_to_cpu(b->info_sector);
1384 		if (sbi->fsinfo_sector == 0)
1385 			sbi->fsinfo_sector = 1;
1386 
1387 		fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1388 		if (fsinfo_bh == NULL) {
1389 			fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1390 			       " (sector = %lu)", sbi->fsinfo_sector);
1391 			brelse(bh);
1392 			goto out_fail;
1393 		}
1394 
1395 		fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1396 		if (!IS_FSINFO(fsinfo)) {
1397 			fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1398 			       "0x%08x, 0x%08x (sector = %lu)",
1399 			       le32_to_cpu(fsinfo->signature1),
1400 			       le32_to_cpu(fsinfo->signature2),
1401 			       sbi->fsinfo_sector);
1402 		} else {
1403 			if (sbi->options.usefree)
1404 				sbi->free_clus_valid = 1;
1405 			sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1406 			sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1407 		}
1408 
1409 		brelse(fsinfo_bh);
1410 	}
1411 
1412 	sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1413 	sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1414 
1415 	sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1416 	sbi->dir_entries = get_unaligned_le16(&b->dir_entries);
1417 	if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1418 		if (!silent)
1419 			fat_msg(sb, KERN_ERR, "bogus directroy-entries per block"
1420 			       " (%u)", sbi->dir_entries);
1421 		brelse(bh);
1422 		goto out_invalid;
1423 	}
1424 
1425 	rootdir_sectors = sbi->dir_entries
1426 		* sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1427 	sbi->data_start = sbi->dir_start + rootdir_sectors;
1428 	total_sectors = get_unaligned_le16(&b->sectors);
1429 	if (total_sectors == 0)
1430 		total_sectors = le32_to_cpu(b->total_sect);
1431 
1432 	total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1433 
1434 	if (sbi->fat_bits != 32)
1435 		sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1436 
1437 	/* check that FAT table does not overflow */
1438 	fat_clusters = sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1439 	total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1440 	if (total_clusters > MAX_FAT(sb)) {
1441 		if (!silent)
1442 			fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1443 			       total_clusters);
1444 		brelse(bh);
1445 		goto out_invalid;
1446 	}
1447 
1448 	sbi->max_cluster = total_clusters + FAT_START_ENT;
1449 	/* check the free_clusters, it's not necessarily correct */
1450 	if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1451 		sbi->free_clusters = -1;
1452 	/* check the prev_free, it's not necessarily correct */
1453 	sbi->prev_free %= sbi->max_cluster;
1454 	if (sbi->prev_free < FAT_START_ENT)
1455 		sbi->prev_free = FAT_START_ENT;
1456 
1457 	brelse(bh);
1458 
1459 	/* set up enough so that it can read an inode */
1460 	fat_hash_init(sb);
1461 	fat_ent_access_init(sb);
1462 
1463 	/*
1464 	 * The low byte of FAT's first entry must have same value with
1465 	 * media-field.  But in real world, too many devices is
1466 	 * writing wrong value.  So, removed that validity check.
1467 	 *
1468 	 * if (FAT_FIRST_ENT(sb, media) != first)
1469 	 */
1470 
1471 	error = -EINVAL;
1472 	sprintf(buf, "cp%d", sbi->options.codepage);
1473 	sbi->nls_disk = load_nls(buf);
1474 	if (!sbi->nls_disk) {
1475 		fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1476 		goto out_fail;
1477 	}
1478 
1479 	/* FIXME: utf8 is using iocharset for upper/lower conversion */
1480 	if (sbi->options.isvfat) {
1481 		sbi->nls_io = load_nls(sbi->options.iocharset);
1482 		if (!sbi->nls_io) {
1483 			fat_msg(sb, KERN_ERR, "IO charset %s not found",
1484 			       sbi->options.iocharset);
1485 			goto out_fail;
1486 		}
1487 	}
1488 
1489 	error = -ENOMEM;
1490 	fat_inode = new_inode(sb);
1491 	if (!fat_inode)
1492 		goto out_fail;
1493 	MSDOS_I(fat_inode)->i_pos = 0;
1494 	sbi->fat_inode = fat_inode;
1495 	root_inode = new_inode(sb);
1496 	if (!root_inode)
1497 		goto out_fail;
1498 	root_inode->i_ino = MSDOS_ROOT_INO;
1499 	root_inode->i_version = 1;
1500 	error = fat_read_root(root_inode);
1501 	if (error < 0)
1502 		goto out_fail;
1503 	error = -ENOMEM;
1504 	insert_inode_hash(root_inode);
1505 	sb->s_root = d_alloc_root(root_inode);
1506 	if (!sb->s_root) {
1507 		fat_msg(sb, KERN_ERR, "get root inode failed");
1508 		goto out_fail;
1509 	}
1510 
1511 	return 0;
1512 
1513 out_invalid:
1514 	error = -EINVAL;
1515 	if (!silent)
1516 		fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1517 
1518 out_fail:
1519 	if (fat_inode)
1520 		iput(fat_inode);
1521 	if (root_inode)
1522 		iput(root_inode);
1523 	unload_nls(sbi->nls_io);
1524 	unload_nls(sbi->nls_disk);
1525 	if (sbi->options.iocharset != fat_default_iocharset)
1526 		kfree(sbi->options.iocharset);
1527 	sb->s_fs_info = NULL;
1528 	kfree(sbi);
1529 	return error;
1530 }
1531 
1532 EXPORT_SYMBOL_GPL(fat_fill_super);
1533 
1534 /*
1535  * helper function for fat_flush_inodes.  This writes both the inode
1536  * and the file data blocks, waiting for in flight data blocks before
1537  * the start of the call.  It does not wait for any io started
1538  * during the call
1539  */
1540 static int writeback_inode(struct inode *inode)
1541 {
1542 
1543 	int ret;
1544 	struct address_space *mapping = inode->i_mapping;
1545 	struct writeback_control wbc = {
1546 	       .sync_mode = WB_SYNC_NONE,
1547 	      .nr_to_write = 0,
1548 	};
1549 	/* if we used WB_SYNC_ALL, sync_inode waits for the io for the
1550 	* inode to finish.  So WB_SYNC_NONE is sent down to sync_inode
1551 	* and filemap_fdatawrite is used for the data blocks
1552 	*/
1553 	ret = sync_inode(inode, &wbc);
1554 	if (!ret)
1555 	       ret = filemap_fdatawrite(mapping);
1556 	return ret;
1557 }
1558 
1559 /*
1560  * write data and metadata corresponding to i1 and i2.  The io is
1561  * started but we do not wait for any of it to finish.
1562  *
1563  * filemap_flush is used for the block device, so if there is a dirty
1564  * page for a block already in flight, we will not wait and start the
1565  * io over again
1566  */
1567 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1568 {
1569 	int ret = 0;
1570 	if (!MSDOS_SB(sb)->options.flush)
1571 		return 0;
1572 	if (i1)
1573 		ret = writeback_inode(i1);
1574 	if (!ret && i2)
1575 		ret = writeback_inode(i2);
1576 	if (!ret) {
1577 		struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1578 		ret = filemap_flush(mapping);
1579 	}
1580 	return ret;
1581 }
1582 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1583 
1584 static int __init init_fat_fs(void)
1585 {
1586 	int err;
1587 
1588 	err = fat_cache_init();
1589 	if (err)
1590 		return err;
1591 
1592 	err = fat_init_inodecache();
1593 	if (err)
1594 		goto failed;
1595 
1596 	return 0;
1597 
1598 failed:
1599 	fat_cache_destroy();
1600 	return err;
1601 }
1602 
1603 static void __exit exit_fat_fs(void)
1604 {
1605 	fat_cache_destroy();
1606 	fat_destroy_inodecache();
1607 }
1608 
1609 module_init(init_fat_fs)
1610 module_exit(exit_fat_fs)
1611 
1612 MODULE_LICENSE("GPL");
1613