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/pagemap.h>
15#include <linux/mpage.h>
16#include <linux/vfs.h>
17#include <linux/seq_file.h>
18#include <linux/parser.h>
19#include <linux/uio.h>
20#include <linux/blkdev.h>
21#include <linux/backing-dev.h>
22#include <asm/unaligned.h>
23#include "fat.h"
24
25#ifndef CONFIG_FAT_DEFAULT_IOCHARSET
26/* if user don't select VFAT, this is undefined. */
27#define CONFIG_FAT_DEFAULT_IOCHARSET	""
28#endif
29
30#define KB_IN_SECTORS 2
31
32/*
33 * A deserialized copy of the on-disk structure laid out in struct
34 * fat_boot_sector.
35 */
36struct fat_bios_param_block {
37	u16	fat_sector_size;
38	u8	fat_sec_per_clus;
39	u16	fat_reserved;
40	u8	fat_fats;
41	u16	fat_dir_entries;
42	u16	fat_sectors;
43	u16	fat_fat_length;
44	u32	fat_total_sect;
45
46	u8	fat16_state;
47	u32	fat16_vol_id;
48
49	u32	fat32_length;
50	u32	fat32_root_cluster;
51	u16	fat32_info_sector;
52	u8	fat32_state;
53	u32	fat32_vol_id;
54};
55
56static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
57static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
58
59static struct fat_floppy_defaults {
60	unsigned nr_sectors;
61	unsigned sec_per_clus;
62	unsigned dir_entries;
63	unsigned media;
64	unsigned fat_length;
65} floppy_defaults[] = {
66{
67	.nr_sectors = 160 * KB_IN_SECTORS,
68	.sec_per_clus = 1,
69	.dir_entries = 64,
70	.media = 0xFE,
71	.fat_length = 1,
72},
73{
74	.nr_sectors = 180 * KB_IN_SECTORS,
75	.sec_per_clus = 1,
76	.dir_entries = 64,
77	.media = 0xFC,
78	.fat_length = 2,
79},
80{
81	.nr_sectors = 320 * KB_IN_SECTORS,
82	.sec_per_clus = 2,
83	.dir_entries = 112,
84	.media = 0xFF,
85	.fat_length = 1,
86},
87{
88	.nr_sectors = 360 * KB_IN_SECTORS,
89	.sec_per_clus = 2,
90	.dir_entries = 112,
91	.media = 0xFD,
92	.fat_length = 2,
93},
94};
95
96static int fat_add_cluster(struct inode *inode)
97{
98	int err, cluster;
99
100	err = fat_alloc_clusters(inode, &cluster, 1);
101	if (err)
102		return err;
103	/* FIXME: this cluster should be added after data of this
104	 * cluster is writed */
105	err = fat_chain_add(inode, cluster, 1);
106	if (err)
107		fat_free_clusters(inode, cluster);
108	return err;
109}
110
111static inline int __fat_get_block(struct inode *inode, sector_t iblock,
112				  unsigned long *max_blocks,
113				  struct buffer_head *bh_result, int create)
114{
115	struct super_block *sb = inode->i_sb;
116	struct msdos_sb_info *sbi = MSDOS_SB(sb);
117	unsigned long mapped_blocks;
118	sector_t phys;
119	int err, offset;
120
121	err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
122	if (err)
123		return err;
124	if (phys) {
125		map_bh(bh_result, sb, phys);
126		*max_blocks = min(mapped_blocks, *max_blocks);
127		return 0;
128	}
129	if (!create)
130		return 0;
131
132	if (iblock != MSDOS_I(inode)->mmu_private >> sb->s_blocksize_bits) {
133		fat_fs_error(sb, "corrupted file size (i_pos %lld, %lld)",
134			MSDOS_I(inode)->i_pos, MSDOS_I(inode)->mmu_private);
135		return -EIO;
136	}
137
138	offset = (unsigned long)iblock & (sbi->sec_per_clus - 1);
139	if (!offset) {
140		/* TODO: multiple cluster allocation would be desirable. */
141		err = fat_add_cluster(inode);
142		if (err)
143			return err;
144	}
145	/* available blocks on this cluster */
146	mapped_blocks = sbi->sec_per_clus - offset;
147
148	*max_blocks = min(mapped_blocks, *max_blocks);
149	MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
150
151	err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
152	if (err)
153		return err;
154
155	BUG_ON(!phys);
156	BUG_ON(*max_blocks != mapped_blocks);
157	set_buffer_new(bh_result);
158	map_bh(bh_result, sb, phys);
159
160	return 0;
161}
162
163static int fat_get_block(struct inode *inode, sector_t iblock,
164			 struct buffer_head *bh_result, int create)
165{
166	struct super_block *sb = inode->i_sb;
167	unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
168	int err;
169
170	err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
171	if (err)
172		return err;
173	bh_result->b_size = max_blocks << sb->s_blocksize_bits;
174	return 0;
175}
176
177static int fat_writepage(struct page *page, struct writeback_control *wbc)
178{
179	return block_write_full_page(page, fat_get_block, wbc);
180}
181
182static int fat_writepages(struct address_space *mapping,
183			  struct writeback_control *wbc)
184{
185	return mpage_writepages(mapping, wbc, fat_get_block);
186}
187
188static int fat_readpage(struct file *file, struct page *page)
189{
190	return mpage_readpage(page, fat_get_block);
191}
192
193static int fat_readpages(struct file *file, struct address_space *mapping,
194			 struct list_head *pages, unsigned nr_pages)
195{
196	return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
197}
198
199static void fat_write_failed(struct address_space *mapping, loff_t to)
200{
201	struct inode *inode = mapping->host;
202
203	if (to > inode->i_size) {
204		truncate_pagecache(inode, inode->i_size);
205		fat_truncate_blocks(inode, inode->i_size);
206	}
207}
208
209static int fat_write_begin(struct file *file, struct address_space *mapping,
210			loff_t pos, unsigned len, unsigned flags,
211			struct page **pagep, void **fsdata)
212{
213	int err;
214
215	*pagep = NULL;
216	err = cont_write_begin(file, mapping, pos, len, flags,
217				pagep, fsdata, fat_get_block,
218				&MSDOS_I(mapping->host)->mmu_private);
219	if (err < 0)
220		fat_write_failed(mapping, pos + len);
221	return err;
222}
223
224static int fat_write_end(struct file *file, struct address_space *mapping,
225			loff_t pos, unsigned len, unsigned copied,
226			struct page *pagep, void *fsdata)
227{
228	struct inode *inode = mapping->host;
229	int err;
230	err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
231	if (err < len)
232		fat_write_failed(mapping, pos + len);
233	if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
234		inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
235		MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
236		mark_inode_dirty(inode);
237	}
238	return err;
239}
240
241static ssize_t fat_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
242			     loff_t offset)
243{
244	struct file *file = iocb->ki_filp;
245	struct address_space *mapping = file->f_mapping;
246	struct inode *inode = mapping->host;
247	size_t count = iov_iter_count(iter);
248	ssize_t ret;
249
250	if (iov_iter_rw(iter) == WRITE) {
251		/*
252		 * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
253		 * so we need to update the ->mmu_private to block boundary.
254		 *
255		 * But we must fill the remaining area or hole by nul for
256		 * updating ->mmu_private.
257		 *
258		 * Return 0, and fallback to normal buffered write.
259		 */
260		loff_t size = offset + count;
261		if (MSDOS_I(inode)->mmu_private < size)
262			return 0;
263	}
264
265	/*
266	 * FAT need to use the DIO_LOCKING for avoiding the race
267	 * condition of fat_get_block() and ->truncate().
268	 */
269	ret = blockdev_direct_IO(iocb, inode, iter, offset, fat_get_block);
270	if (ret < 0 && iov_iter_rw(iter) == WRITE)
271		fat_write_failed(mapping, offset + count);
272
273	return ret;
274}
275
276static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
277{
278	sector_t blocknr;
279
280	/* fat_get_cluster() assumes the requested blocknr isn't truncated. */
281	down_read(&MSDOS_I(mapping->host)->truncate_lock);
282	blocknr = generic_block_bmap(mapping, block, fat_get_block);
283	up_read(&MSDOS_I(mapping->host)->truncate_lock);
284
285	return blocknr;
286}
287
288/*
289 * fat_block_truncate_page() zeroes out a mapping from file offset `from'
290 * up to the end of the block which corresponds to `from'.
291 * This is required during truncate to physically zeroout the tail end
292 * of that block so it doesn't yield old data if the file is later grown.
293 * Also, avoid causing failure from fsx for cases of "data past EOF"
294 */
295int fat_block_truncate_page(struct inode *inode, loff_t from)
296{
297	return block_truncate_page(inode->i_mapping, from, fat_get_block);
298}
299
300static const struct address_space_operations fat_aops = {
301	.readpage	= fat_readpage,
302	.readpages	= fat_readpages,
303	.writepage	= fat_writepage,
304	.writepages	= fat_writepages,
305	.write_begin	= fat_write_begin,
306	.write_end	= fat_write_end,
307	.direct_IO	= fat_direct_IO,
308	.bmap		= _fat_bmap
309};
310
311/*
312 * New FAT inode stuff. We do the following:
313 *	a) i_ino is constant and has nothing with on-disk location.
314 *	b) FAT manages its own cache of directory entries.
315 *	c) *This* cache is indexed by on-disk location.
316 *	d) inode has an associated directory entry, all right, but
317 *		it may be unhashed.
318 *	e) currently entries are stored within struct inode. That should
319 *		change.
320 *	f) we deal with races in the following way:
321 *		1. readdir() and lookup() do FAT-dir-cache lookup.
322 *		2. rename() unhashes the F-d-c entry and rehashes it in
323 *			a new place.
324 *		3. unlink() and rmdir() unhash F-d-c entry.
325 *		4. fat_write_inode() checks whether the thing is unhashed.
326 *			If it is we silently return. If it isn't we do bread(),
327 *			check if the location is still valid and retry if it
328 *			isn't. Otherwise we do changes.
329 *		5. Spinlock is used to protect hash/unhash/location check/lookup
330 *		6. fat_evict_inode() unhashes the F-d-c entry.
331 *		7. lookup() and readdir() do igrab() if they find a F-d-c entry
332 *			and consider negative result as cache miss.
333 */
334
335static void fat_hash_init(struct super_block *sb)
336{
337	struct msdos_sb_info *sbi = MSDOS_SB(sb);
338	int i;
339
340	spin_lock_init(&sbi->inode_hash_lock);
341	for (i = 0; i < FAT_HASH_SIZE; i++)
342		INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
343}
344
345static inline unsigned long fat_hash(loff_t i_pos)
346{
347	return hash_32(i_pos, FAT_HASH_BITS);
348}
349
350static void dir_hash_init(struct super_block *sb)
351{
352	struct msdos_sb_info *sbi = MSDOS_SB(sb);
353	int i;
354
355	spin_lock_init(&sbi->dir_hash_lock);
356	for (i = 0; i < FAT_HASH_SIZE; i++)
357		INIT_HLIST_HEAD(&sbi->dir_hashtable[i]);
358}
359
360void fat_attach(struct inode *inode, loff_t i_pos)
361{
362	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
363
364	if (inode->i_ino != MSDOS_ROOT_INO) {
365		struct hlist_head *head =   sbi->inode_hashtable
366					  + fat_hash(i_pos);
367
368		spin_lock(&sbi->inode_hash_lock);
369		MSDOS_I(inode)->i_pos = i_pos;
370		hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
371		spin_unlock(&sbi->inode_hash_lock);
372	}
373
374	/* If NFS support is enabled, cache the mapping of start cluster
375	 * to directory inode. This is used during reconnection of
376	 * dentries to the filesystem root.
377	 */
378	if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
379		struct hlist_head *d_head = sbi->dir_hashtable;
380		d_head += fat_dir_hash(MSDOS_I(inode)->i_logstart);
381
382		spin_lock(&sbi->dir_hash_lock);
383		hlist_add_head(&MSDOS_I(inode)->i_dir_hash, d_head);
384		spin_unlock(&sbi->dir_hash_lock);
385	}
386}
387EXPORT_SYMBOL_GPL(fat_attach);
388
389void fat_detach(struct inode *inode)
390{
391	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
392	spin_lock(&sbi->inode_hash_lock);
393	MSDOS_I(inode)->i_pos = 0;
394	hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
395	spin_unlock(&sbi->inode_hash_lock);
396
397	if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
398		spin_lock(&sbi->dir_hash_lock);
399		hlist_del_init(&MSDOS_I(inode)->i_dir_hash);
400		spin_unlock(&sbi->dir_hash_lock);
401	}
402}
403EXPORT_SYMBOL_GPL(fat_detach);
404
405struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
406{
407	struct msdos_sb_info *sbi = MSDOS_SB(sb);
408	struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
409	struct msdos_inode_info *i;
410	struct inode *inode = NULL;
411
412	spin_lock(&sbi->inode_hash_lock);
413	hlist_for_each_entry(i, head, i_fat_hash) {
414		BUG_ON(i->vfs_inode.i_sb != sb);
415		if (i->i_pos != i_pos)
416			continue;
417		inode = igrab(&i->vfs_inode);
418		if (inode)
419			break;
420	}
421	spin_unlock(&sbi->inode_hash_lock);
422	return inode;
423}
424
425static int is_exec(unsigned char *extension)
426{
427	unsigned char exe_extensions[] = "EXECOMBAT", *walk;
428
429	for (walk = exe_extensions; *walk; walk += 3)
430		if (!strncmp(extension, walk, 3))
431			return 1;
432	return 0;
433}
434
435static int fat_calc_dir_size(struct inode *inode)
436{
437	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
438	int ret, fclus, dclus;
439
440	inode->i_size = 0;
441	if (MSDOS_I(inode)->i_start == 0)
442		return 0;
443
444	ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
445	if (ret < 0)
446		return ret;
447	inode->i_size = (fclus + 1) << sbi->cluster_bits;
448
449	return 0;
450}
451
452/* doesn't deal with root inode */
453int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
454{
455	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
456	int error;
457
458	MSDOS_I(inode)->i_pos = 0;
459	inode->i_uid = sbi->options.fs_uid;
460	inode->i_gid = sbi->options.fs_gid;
461	inode->i_version++;
462	inode->i_generation = get_seconds();
463
464	if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
465		inode->i_generation &= ~1;
466		inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
467		inode->i_op = sbi->dir_ops;
468		inode->i_fop = &fat_dir_operations;
469
470		MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
471		MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
472		error = fat_calc_dir_size(inode);
473		if (error < 0)
474			return error;
475		MSDOS_I(inode)->mmu_private = inode->i_size;
476
477		set_nlink(inode, fat_subdirs(inode));
478	} else { /* not a directory */
479		inode->i_generation |= 1;
480		inode->i_mode = fat_make_mode(sbi, de->attr,
481			((sbi->options.showexec && !is_exec(de->name + 8))
482			 ? S_IRUGO|S_IWUGO : S_IRWXUGO));
483		MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
484
485		MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
486		inode->i_size = le32_to_cpu(de->size);
487		inode->i_op = &fat_file_inode_operations;
488		inode->i_fop = &fat_file_operations;
489		inode->i_mapping->a_ops = &fat_aops;
490		MSDOS_I(inode)->mmu_private = inode->i_size;
491	}
492	if (de->attr & ATTR_SYS) {
493		if (sbi->options.sys_immutable)
494			inode->i_flags |= S_IMMUTABLE;
495	}
496	fat_save_attrs(inode, de->attr);
497
498	inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
499			   & ~((loff_t)sbi->cluster_size - 1)) >> 9;
500
501	fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
502	if (sbi->options.isvfat) {
503		fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
504				  de->cdate, de->ctime_cs);
505		fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
506	} else
507		inode->i_ctime = inode->i_atime = inode->i_mtime;
508
509	return 0;
510}
511
512static inline void fat_lock_build_inode(struct msdos_sb_info *sbi)
513{
514	if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
515		mutex_lock(&sbi->nfs_build_inode_lock);
516}
517
518static inline void fat_unlock_build_inode(struct msdos_sb_info *sbi)
519{
520	if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
521		mutex_unlock(&sbi->nfs_build_inode_lock);
522}
523
524struct inode *fat_build_inode(struct super_block *sb,
525			struct msdos_dir_entry *de, loff_t i_pos)
526{
527	struct inode *inode;
528	int err;
529
530	fat_lock_build_inode(MSDOS_SB(sb));
531	inode = fat_iget(sb, i_pos);
532	if (inode)
533		goto out;
534	inode = new_inode(sb);
535	if (!inode) {
536		inode = ERR_PTR(-ENOMEM);
537		goto out;
538	}
539	inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
540	inode->i_version = 1;
541	err = fat_fill_inode(inode, de);
542	if (err) {
543		iput(inode);
544		inode = ERR_PTR(err);
545		goto out;
546	}
547	fat_attach(inode, i_pos);
548	insert_inode_hash(inode);
549out:
550	fat_unlock_build_inode(MSDOS_SB(sb));
551	return inode;
552}
553
554EXPORT_SYMBOL_GPL(fat_build_inode);
555
556static void fat_evict_inode(struct inode *inode)
557{
558	truncate_inode_pages_final(&inode->i_data);
559	if (!inode->i_nlink) {
560		inode->i_size = 0;
561		fat_truncate_blocks(inode, 0);
562	}
563	invalidate_inode_buffers(inode);
564	clear_inode(inode);
565	fat_cache_inval_inode(inode);
566	fat_detach(inode);
567}
568
569static void fat_set_state(struct super_block *sb,
570			unsigned int set, unsigned int force)
571{
572	struct buffer_head *bh;
573	struct fat_boot_sector *b;
574	struct msdos_sb_info *sbi = MSDOS_SB(sb);
575
576	/* do not change any thing if mounted read only */
577	if ((sb->s_flags & MS_RDONLY) && !force)
578		return;
579
580	/* do not change state if fs was dirty */
581	if (sbi->dirty) {
582		/* warn only on set (mount). */
583		if (set)
584			fat_msg(sb, KERN_WARNING, "Volume was not properly "
585				"unmounted. Some data may be corrupt. "
586				"Please run fsck.");
587		return;
588	}
589
590	bh = sb_bread(sb, 0);
591	if (bh == NULL) {
592		fat_msg(sb, KERN_ERR, "unable to read boot sector "
593			"to mark fs as dirty");
594		return;
595	}
596
597	b = (struct fat_boot_sector *) bh->b_data;
598
599	if (sbi->fat_bits == 32) {
600		if (set)
601			b->fat32.state |= FAT_STATE_DIRTY;
602		else
603			b->fat32.state &= ~FAT_STATE_DIRTY;
604	} else /* fat 16 and 12 */ {
605		if (set)
606			b->fat16.state |= FAT_STATE_DIRTY;
607		else
608			b->fat16.state &= ~FAT_STATE_DIRTY;
609	}
610
611	mark_buffer_dirty(bh);
612	sync_dirty_buffer(bh);
613	brelse(bh);
614}
615
616static void delayed_free(struct rcu_head *p)
617{
618	struct msdos_sb_info *sbi = container_of(p, struct msdos_sb_info, rcu);
619	unload_nls(sbi->nls_disk);
620	unload_nls(sbi->nls_io);
621	if (sbi->options.iocharset != fat_default_iocharset)
622		kfree(sbi->options.iocharset);
623	kfree(sbi);
624}
625
626static void fat_put_super(struct super_block *sb)
627{
628	struct msdos_sb_info *sbi = MSDOS_SB(sb);
629
630	fat_set_state(sb, 0, 0);
631
632	iput(sbi->fsinfo_inode);
633	iput(sbi->fat_inode);
634
635	call_rcu(&sbi->rcu, delayed_free);
636}
637
638static struct kmem_cache *fat_inode_cachep;
639
640static struct inode *fat_alloc_inode(struct super_block *sb)
641{
642	struct msdos_inode_info *ei;
643	ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
644	if (!ei)
645		return NULL;
646
647	init_rwsem(&ei->truncate_lock);
648	return &ei->vfs_inode;
649}
650
651static void fat_i_callback(struct rcu_head *head)
652{
653	struct inode *inode = container_of(head, struct inode, i_rcu);
654	kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
655}
656
657static void fat_destroy_inode(struct inode *inode)
658{
659	call_rcu(&inode->i_rcu, fat_i_callback);
660}
661
662static void init_once(void *foo)
663{
664	struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
665
666	spin_lock_init(&ei->cache_lru_lock);
667	ei->nr_caches = 0;
668	ei->cache_valid_id = FAT_CACHE_VALID + 1;
669	INIT_LIST_HEAD(&ei->cache_lru);
670	INIT_HLIST_NODE(&ei->i_fat_hash);
671	INIT_HLIST_NODE(&ei->i_dir_hash);
672	inode_init_once(&ei->vfs_inode);
673}
674
675static int __init fat_init_inodecache(void)
676{
677	fat_inode_cachep = kmem_cache_create("fat_inode_cache",
678					     sizeof(struct msdos_inode_info),
679					     0, (SLAB_RECLAIM_ACCOUNT|
680						SLAB_MEM_SPREAD),
681					     init_once);
682	if (fat_inode_cachep == NULL)
683		return -ENOMEM;
684	return 0;
685}
686
687static void __exit fat_destroy_inodecache(void)
688{
689	/*
690	 * Make sure all delayed rcu free inodes are flushed before we
691	 * destroy cache.
692	 */
693	rcu_barrier();
694	kmem_cache_destroy(fat_inode_cachep);
695}
696
697static int fat_remount(struct super_block *sb, int *flags, char *data)
698{
699	int new_rdonly;
700	struct msdos_sb_info *sbi = MSDOS_SB(sb);
701	*flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
702
703	sync_filesystem(sb);
704
705	/* make sure we update state on remount. */
706	new_rdonly = *flags & MS_RDONLY;
707	if (new_rdonly != (sb->s_flags & MS_RDONLY)) {
708		if (new_rdonly)
709			fat_set_state(sb, 0, 0);
710		else
711			fat_set_state(sb, 1, 1);
712	}
713	return 0;
714}
715
716static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
717{
718	struct super_block *sb = dentry->d_sb;
719	struct msdos_sb_info *sbi = MSDOS_SB(sb);
720	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
721
722	/* If the count of free cluster is still unknown, counts it here. */
723	if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
724		int err = fat_count_free_clusters(dentry->d_sb);
725		if (err)
726			return err;
727	}
728
729	buf->f_type = dentry->d_sb->s_magic;
730	buf->f_bsize = sbi->cluster_size;
731	buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
732	buf->f_bfree = sbi->free_clusters;
733	buf->f_bavail = sbi->free_clusters;
734	buf->f_fsid.val[0] = (u32)id;
735	buf->f_fsid.val[1] = (u32)(id >> 32);
736	buf->f_namelen =
737		(sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
738
739	return 0;
740}
741
742static int __fat_write_inode(struct inode *inode, int wait)
743{
744	struct super_block *sb = inode->i_sb;
745	struct msdos_sb_info *sbi = MSDOS_SB(sb);
746	struct buffer_head *bh;
747	struct msdos_dir_entry *raw_entry;
748	loff_t i_pos;
749	sector_t blocknr;
750	int err, offset;
751
752	if (inode->i_ino == MSDOS_ROOT_INO)
753		return 0;
754
755retry:
756	i_pos = fat_i_pos_read(sbi, inode);
757	if (!i_pos)
758		return 0;
759
760	fat_get_blknr_offset(sbi, i_pos, &blocknr, &offset);
761	bh = sb_bread(sb, blocknr);
762	if (!bh) {
763		fat_msg(sb, KERN_ERR, "unable to read inode block "
764		       "for updating (i_pos %lld)", i_pos);
765		return -EIO;
766	}
767	spin_lock(&sbi->inode_hash_lock);
768	if (i_pos != MSDOS_I(inode)->i_pos) {
769		spin_unlock(&sbi->inode_hash_lock);
770		brelse(bh);
771		goto retry;
772	}
773
774	raw_entry = &((struct msdos_dir_entry *) (bh->b_data))[offset];
775	if (S_ISDIR(inode->i_mode))
776		raw_entry->size = 0;
777	else
778		raw_entry->size = cpu_to_le32(inode->i_size);
779	raw_entry->attr = fat_make_attrs(inode);
780	fat_set_start(raw_entry, MSDOS_I(inode)->i_logstart);
781	fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
782			  &raw_entry->date, NULL);
783	if (sbi->options.isvfat) {
784		__le16 atime;
785		fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
786				  &raw_entry->cdate, &raw_entry->ctime_cs);
787		fat_time_unix2fat(sbi, &inode->i_atime, &atime,
788				  &raw_entry->adate, NULL);
789	}
790	spin_unlock(&sbi->inode_hash_lock);
791	mark_buffer_dirty(bh);
792	err = 0;
793	if (wait)
794		err = sync_dirty_buffer(bh);
795	brelse(bh);
796	return err;
797}
798
799static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
800{
801	int err;
802
803	if (inode->i_ino == MSDOS_FSINFO_INO) {
804		struct super_block *sb = inode->i_sb;
805
806		mutex_lock(&MSDOS_SB(sb)->s_lock);
807		err = fat_clusters_flush(sb);
808		mutex_unlock(&MSDOS_SB(sb)->s_lock);
809	} else
810		err = __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
811
812	return err;
813}
814
815int fat_sync_inode(struct inode *inode)
816{
817	return __fat_write_inode(inode, 1);
818}
819
820EXPORT_SYMBOL_GPL(fat_sync_inode);
821
822static int fat_show_options(struct seq_file *m, struct dentry *root);
823static const struct super_operations fat_sops = {
824	.alloc_inode	= fat_alloc_inode,
825	.destroy_inode	= fat_destroy_inode,
826	.write_inode	= fat_write_inode,
827	.evict_inode	= fat_evict_inode,
828	.put_super	= fat_put_super,
829	.statfs		= fat_statfs,
830	.remount_fs	= fat_remount,
831
832	.show_options	= fat_show_options,
833};
834
835static int fat_show_options(struct seq_file *m, struct dentry *root)
836{
837	struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
838	struct fat_mount_options *opts = &sbi->options;
839	int isvfat = opts->isvfat;
840
841	if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
842		seq_printf(m, ",uid=%u",
843				from_kuid_munged(&init_user_ns, opts->fs_uid));
844	if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
845		seq_printf(m, ",gid=%u",
846				from_kgid_munged(&init_user_ns, opts->fs_gid));
847	seq_printf(m, ",fmask=%04o", opts->fs_fmask);
848	seq_printf(m, ",dmask=%04o", opts->fs_dmask);
849	if (opts->allow_utime)
850		seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
851	if (sbi->nls_disk)
852		/* strip "cp" prefix from displayed option */
853		seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
854	if (isvfat) {
855		if (sbi->nls_io)
856			seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
857
858		switch (opts->shortname) {
859		case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
860			seq_puts(m, ",shortname=win95");
861			break;
862		case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
863			seq_puts(m, ",shortname=winnt");
864			break;
865		case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
866			seq_puts(m, ",shortname=mixed");
867			break;
868		case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
869			seq_puts(m, ",shortname=lower");
870			break;
871		default:
872			seq_puts(m, ",shortname=unknown");
873			break;
874		}
875	}
876	if (opts->name_check != 'n')
877		seq_printf(m, ",check=%c", opts->name_check);
878	if (opts->usefree)
879		seq_puts(m, ",usefree");
880	if (opts->quiet)
881		seq_puts(m, ",quiet");
882	if (opts->showexec)
883		seq_puts(m, ",showexec");
884	if (opts->sys_immutable)
885		seq_puts(m, ",sys_immutable");
886	if (!isvfat) {
887		if (opts->dotsOK)
888			seq_puts(m, ",dotsOK=yes");
889		if (opts->nocase)
890			seq_puts(m, ",nocase");
891	} else {
892		if (opts->utf8)
893			seq_puts(m, ",utf8");
894		if (opts->unicode_xlate)
895			seq_puts(m, ",uni_xlate");
896		if (!opts->numtail)
897			seq_puts(m, ",nonumtail");
898		if (opts->rodir)
899			seq_puts(m, ",rodir");
900	}
901	if (opts->flush)
902		seq_puts(m, ",flush");
903	if (opts->tz_set) {
904		if (opts->time_offset)
905			seq_printf(m, ",time_offset=%d", opts->time_offset);
906		else
907			seq_puts(m, ",tz=UTC");
908	}
909	if (opts->errors == FAT_ERRORS_CONT)
910		seq_puts(m, ",errors=continue");
911	else if (opts->errors == FAT_ERRORS_PANIC)
912		seq_puts(m, ",errors=panic");
913	else
914		seq_puts(m, ",errors=remount-ro");
915	if (opts->nfs == FAT_NFS_NOSTALE_RO)
916		seq_puts(m, ",nfs=nostale_ro");
917	else if (opts->nfs)
918		seq_puts(m, ",nfs=stale_rw");
919	if (opts->discard)
920		seq_puts(m, ",discard");
921	if (opts->dos1xfloppy)
922		seq_puts(m, ",dos1xfloppy");
923
924	return 0;
925}
926
927enum {
928	Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
929	Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
930	Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
931	Opt_immutable, Opt_dots, Opt_nodots,
932	Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
933	Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
934	Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
935	Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
936	Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
937	Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
938};
939
940static const match_table_t fat_tokens = {
941	{Opt_check_r, "check=relaxed"},
942	{Opt_check_s, "check=strict"},
943	{Opt_check_n, "check=normal"},
944	{Opt_check_r, "check=r"},
945	{Opt_check_s, "check=s"},
946	{Opt_check_n, "check=n"},
947	{Opt_uid, "uid=%u"},
948	{Opt_gid, "gid=%u"},
949	{Opt_umask, "umask=%o"},
950	{Opt_dmask, "dmask=%o"},
951	{Opt_fmask, "fmask=%o"},
952	{Opt_allow_utime, "allow_utime=%o"},
953	{Opt_codepage, "codepage=%u"},
954	{Opt_usefree, "usefree"},
955	{Opt_nocase, "nocase"},
956	{Opt_quiet, "quiet"},
957	{Opt_showexec, "showexec"},
958	{Opt_debug, "debug"},
959	{Opt_immutable, "sys_immutable"},
960	{Opt_flush, "flush"},
961	{Opt_tz_utc, "tz=UTC"},
962	{Opt_time_offset, "time_offset=%d"},
963	{Opt_err_cont, "errors=continue"},
964	{Opt_err_panic, "errors=panic"},
965	{Opt_err_ro, "errors=remount-ro"},
966	{Opt_discard, "discard"},
967	{Opt_nfs_stale_rw, "nfs"},
968	{Opt_nfs_stale_rw, "nfs=stale_rw"},
969	{Opt_nfs_nostale_ro, "nfs=nostale_ro"},
970	{Opt_dos1xfloppy, "dos1xfloppy"},
971	{Opt_obsolete, "conv=binary"},
972	{Opt_obsolete, "conv=text"},
973	{Opt_obsolete, "conv=auto"},
974	{Opt_obsolete, "conv=b"},
975	{Opt_obsolete, "conv=t"},
976	{Opt_obsolete, "conv=a"},
977	{Opt_obsolete, "fat=%u"},
978	{Opt_obsolete, "blocksize=%u"},
979	{Opt_obsolete, "cvf_format=%20s"},
980	{Opt_obsolete, "cvf_options=%100s"},
981	{Opt_obsolete, "posix"},
982	{Opt_err, NULL},
983};
984static const match_table_t msdos_tokens = {
985	{Opt_nodots, "nodots"},
986	{Opt_nodots, "dotsOK=no"},
987	{Opt_dots, "dots"},
988	{Opt_dots, "dotsOK=yes"},
989	{Opt_err, NULL}
990};
991static const match_table_t vfat_tokens = {
992	{Opt_charset, "iocharset=%s"},
993	{Opt_shortname_lower, "shortname=lower"},
994	{Opt_shortname_win95, "shortname=win95"},
995	{Opt_shortname_winnt, "shortname=winnt"},
996	{Opt_shortname_mixed, "shortname=mixed"},
997	{Opt_utf8_no, "utf8=0"},		/* 0 or no or false */
998	{Opt_utf8_no, "utf8=no"},
999	{Opt_utf8_no, "utf8=false"},
1000	{Opt_utf8_yes, "utf8=1"},		/* empty or 1 or yes or true */
1001	{Opt_utf8_yes, "utf8=yes"},
1002	{Opt_utf8_yes, "utf8=true"},
1003	{Opt_utf8_yes, "utf8"},
1004	{Opt_uni_xl_no, "uni_xlate=0"},		/* 0 or no or false */
1005	{Opt_uni_xl_no, "uni_xlate=no"},
1006	{Opt_uni_xl_no, "uni_xlate=false"},
1007	{Opt_uni_xl_yes, "uni_xlate=1"},	/* empty or 1 or yes or true */
1008	{Opt_uni_xl_yes, "uni_xlate=yes"},
1009	{Opt_uni_xl_yes, "uni_xlate=true"},
1010	{Opt_uni_xl_yes, "uni_xlate"},
1011	{Opt_nonumtail_no, "nonumtail=0"},	/* 0 or no or false */
1012	{Opt_nonumtail_no, "nonumtail=no"},
1013	{Opt_nonumtail_no, "nonumtail=false"},
1014	{Opt_nonumtail_yes, "nonumtail=1"},	/* empty or 1 or yes or true */
1015	{Opt_nonumtail_yes, "nonumtail=yes"},
1016	{Opt_nonumtail_yes, "nonumtail=true"},
1017	{Opt_nonumtail_yes, "nonumtail"},
1018	{Opt_rodir, "rodir"},
1019	{Opt_err, NULL}
1020};
1021
1022static int parse_options(struct super_block *sb, char *options, int is_vfat,
1023			 int silent, int *debug, struct fat_mount_options *opts)
1024{
1025	char *p;
1026	substring_t args[MAX_OPT_ARGS];
1027	int option;
1028	char *iocharset;
1029
1030	opts->isvfat = is_vfat;
1031
1032	opts->fs_uid = current_uid();
1033	opts->fs_gid = current_gid();
1034	opts->fs_fmask = opts->fs_dmask = current_umask();
1035	opts->allow_utime = -1;
1036	opts->codepage = fat_default_codepage;
1037	opts->iocharset = fat_default_iocharset;
1038	if (is_vfat) {
1039		opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1040		opts->rodir = 0;
1041	} else {
1042		opts->shortname = 0;
1043		opts->rodir = 1;
1044	}
1045	opts->name_check = 'n';
1046	opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
1047	opts->utf8 = opts->unicode_xlate = 0;
1048	opts->numtail = 1;
1049	opts->usefree = opts->nocase = 0;
1050	opts->tz_set = 0;
1051	opts->nfs = 0;
1052	opts->errors = FAT_ERRORS_RO;
1053	*debug = 0;
1054
1055	if (!options)
1056		goto out;
1057
1058	while ((p = strsep(&options, ",")) != NULL) {
1059		int token;
1060		if (!*p)
1061			continue;
1062
1063		token = match_token(p, fat_tokens, args);
1064		if (token == Opt_err) {
1065			if (is_vfat)
1066				token = match_token(p, vfat_tokens, args);
1067			else
1068				token = match_token(p, msdos_tokens, args);
1069		}
1070		switch (token) {
1071		case Opt_check_s:
1072			opts->name_check = 's';
1073			break;
1074		case Opt_check_r:
1075			opts->name_check = 'r';
1076			break;
1077		case Opt_check_n:
1078			opts->name_check = 'n';
1079			break;
1080		case Opt_usefree:
1081			opts->usefree = 1;
1082			break;
1083		case Opt_nocase:
1084			if (!is_vfat)
1085				opts->nocase = 1;
1086			else {
1087				/* for backward compatibility */
1088				opts->shortname = VFAT_SFN_DISPLAY_WIN95
1089					| VFAT_SFN_CREATE_WIN95;
1090			}
1091			break;
1092		case Opt_quiet:
1093			opts->quiet = 1;
1094			break;
1095		case Opt_showexec:
1096			opts->showexec = 1;
1097			break;
1098		case Opt_debug:
1099			*debug = 1;
1100			break;
1101		case Opt_immutable:
1102			opts->sys_immutable = 1;
1103			break;
1104		case Opt_uid:
1105			if (match_int(&args[0], &option))
1106				return -EINVAL;
1107			opts->fs_uid = make_kuid(current_user_ns(), option);
1108			if (!uid_valid(opts->fs_uid))
1109				return -EINVAL;
1110			break;
1111		case Opt_gid:
1112			if (match_int(&args[0], &option))
1113				return -EINVAL;
1114			opts->fs_gid = make_kgid(current_user_ns(), option);
1115			if (!gid_valid(opts->fs_gid))
1116				return -EINVAL;
1117			break;
1118		case Opt_umask:
1119			if (match_octal(&args[0], &option))
1120				return -EINVAL;
1121			opts->fs_fmask = opts->fs_dmask = option;
1122			break;
1123		case Opt_dmask:
1124			if (match_octal(&args[0], &option))
1125				return -EINVAL;
1126			opts->fs_dmask = option;
1127			break;
1128		case Opt_fmask:
1129			if (match_octal(&args[0], &option))
1130				return -EINVAL;
1131			opts->fs_fmask = option;
1132			break;
1133		case Opt_allow_utime:
1134			if (match_octal(&args[0], &option))
1135				return -EINVAL;
1136			opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1137			break;
1138		case Opt_codepage:
1139			if (match_int(&args[0], &option))
1140				return -EINVAL;
1141			opts->codepage = option;
1142			break;
1143		case Opt_flush:
1144			opts->flush = 1;
1145			break;
1146		case Opt_time_offset:
1147			if (match_int(&args[0], &option))
1148				return -EINVAL;
1149			if (option < -12 * 60 || option > 12 * 60)
1150				return -EINVAL;
1151			opts->tz_set = 1;
1152			opts->time_offset = option;
1153			break;
1154		case Opt_tz_utc:
1155			opts->tz_set = 1;
1156			opts->time_offset = 0;
1157			break;
1158		case Opt_err_cont:
1159			opts->errors = FAT_ERRORS_CONT;
1160			break;
1161		case Opt_err_panic:
1162			opts->errors = FAT_ERRORS_PANIC;
1163			break;
1164		case Opt_err_ro:
1165			opts->errors = FAT_ERRORS_RO;
1166			break;
1167		case Opt_nfs_stale_rw:
1168			opts->nfs = FAT_NFS_STALE_RW;
1169			break;
1170		case Opt_nfs_nostale_ro:
1171			opts->nfs = FAT_NFS_NOSTALE_RO;
1172			break;
1173		case Opt_dos1xfloppy:
1174			opts->dos1xfloppy = 1;
1175			break;
1176
1177		/* msdos specific */
1178		case Opt_dots:
1179			opts->dotsOK = 1;
1180			break;
1181		case Opt_nodots:
1182			opts->dotsOK = 0;
1183			break;
1184
1185		/* vfat specific */
1186		case Opt_charset:
1187			if (opts->iocharset != fat_default_iocharset)
1188				kfree(opts->iocharset);
1189			iocharset = match_strdup(&args[0]);
1190			if (!iocharset)
1191				return -ENOMEM;
1192			opts->iocharset = iocharset;
1193			break;
1194		case Opt_shortname_lower:
1195			opts->shortname = VFAT_SFN_DISPLAY_LOWER
1196					| VFAT_SFN_CREATE_WIN95;
1197			break;
1198		case Opt_shortname_win95:
1199			opts->shortname = VFAT_SFN_DISPLAY_WIN95
1200					| VFAT_SFN_CREATE_WIN95;
1201			break;
1202		case Opt_shortname_winnt:
1203			opts->shortname = VFAT_SFN_DISPLAY_WINNT
1204					| VFAT_SFN_CREATE_WINNT;
1205			break;
1206		case Opt_shortname_mixed:
1207			opts->shortname = VFAT_SFN_DISPLAY_WINNT
1208					| VFAT_SFN_CREATE_WIN95;
1209			break;
1210		case Opt_utf8_no:		/* 0 or no or false */
1211			opts->utf8 = 0;
1212			break;
1213		case Opt_utf8_yes:		/* empty or 1 or yes or true */
1214			opts->utf8 = 1;
1215			break;
1216		case Opt_uni_xl_no:		/* 0 or no or false */
1217			opts->unicode_xlate = 0;
1218			break;
1219		case Opt_uni_xl_yes:		/* empty or 1 or yes or true */
1220			opts->unicode_xlate = 1;
1221			break;
1222		case Opt_nonumtail_no:		/* 0 or no or false */
1223			opts->numtail = 1;	/* negated option */
1224			break;
1225		case Opt_nonumtail_yes:		/* empty or 1 or yes or true */
1226			opts->numtail = 0;	/* negated option */
1227			break;
1228		case Opt_rodir:
1229			opts->rodir = 1;
1230			break;
1231		case Opt_discard:
1232			opts->discard = 1;
1233			break;
1234
1235		/* obsolete mount options */
1236		case Opt_obsolete:
1237			fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1238			       "not supported now", p);
1239			break;
1240		/* unknown option */
1241		default:
1242			if (!silent) {
1243				fat_msg(sb, KERN_ERR,
1244				       "Unrecognized mount option \"%s\" "
1245				       "or missing value", p);
1246			}
1247			return -EINVAL;
1248		}
1249	}
1250
1251out:
1252	/* UTF-8 doesn't provide FAT semantics */
1253	if (!strcmp(opts->iocharset, "utf8")) {
1254		fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1255		       " for FAT filesystems, filesystem will be "
1256		       "case sensitive!");
1257	}
1258
1259	/* If user doesn't specify allow_utime, it's initialized from dmask. */
1260	if (opts->allow_utime == (unsigned short)-1)
1261		opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1262	if (opts->unicode_xlate)
1263		opts->utf8 = 0;
1264	if (opts->nfs == FAT_NFS_NOSTALE_RO) {
1265		sb->s_flags |= MS_RDONLY;
1266		sb->s_export_op = &fat_export_ops_nostale;
1267	}
1268
1269	return 0;
1270}
1271
1272static int fat_read_root(struct inode *inode)
1273{
1274	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
1275	int error;
1276
1277	MSDOS_I(inode)->i_pos = MSDOS_ROOT_INO;
1278	inode->i_uid = sbi->options.fs_uid;
1279	inode->i_gid = sbi->options.fs_gid;
1280	inode->i_version++;
1281	inode->i_generation = 0;
1282	inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1283	inode->i_op = sbi->dir_ops;
1284	inode->i_fop = &fat_dir_operations;
1285	if (sbi->fat_bits == 32) {
1286		MSDOS_I(inode)->i_start = sbi->root_cluster;
1287		error = fat_calc_dir_size(inode);
1288		if (error < 0)
1289			return error;
1290	} else {
1291		MSDOS_I(inode)->i_start = 0;
1292		inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1293	}
1294	inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1295			   & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1296	MSDOS_I(inode)->i_logstart = 0;
1297	MSDOS_I(inode)->mmu_private = inode->i_size;
1298
1299	fat_save_attrs(inode, ATTR_DIR);
1300	inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1301	inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1302	set_nlink(inode, fat_subdirs(inode)+2);
1303
1304	return 0;
1305}
1306
1307static unsigned long calc_fat_clusters(struct super_block *sb)
1308{
1309	struct msdos_sb_info *sbi = MSDOS_SB(sb);
1310
1311	/* Divide first to avoid overflow */
1312	if (sbi->fat_bits != 12) {
1313		unsigned long ent_per_sec = sb->s_blocksize * 8 / sbi->fat_bits;
1314		return ent_per_sec * sbi->fat_length;
1315	}
1316
1317	return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1318}
1319
1320static bool fat_bpb_is_zero(struct fat_boot_sector *b)
1321{
1322	if (get_unaligned_le16(&b->sector_size))
1323		return false;
1324	if (b->sec_per_clus)
1325		return false;
1326	if (b->reserved)
1327		return false;
1328	if (b->fats)
1329		return false;
1330	if (get_unaligned_le16(&b->dir_entries))
1331		return false;
1332	if (get_unaligned_le16(&b->sectors))
1333		return false;
1334	if (b->media)
1335		return false;
1336	if (b->fat_length)
1337		return false;
1338	if (b->secs_track)
1339		return false;
1340	if (b->heads)
1341		return false;
1342	return true;
1343}
1344
1345static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
1346	int silent, struct fat_bios_param_block *bpb)
1347{
1348	int error = -EINVAL;
1349
1350	/* Read in BPB ... */
1351	memset(bpb, 0, sizeof(*bpb));
1352	bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
1353	bpb->fat_sec_per_clus = b->sec_per_clus;
1354	bpb->fat_reserved = le16_to_cpu(b->reserved);
1355	bpb->fat_fats = b->fats;
1356	bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
1357	bpb->fat_sectors = get_unaligned_le16(&b->sectors);
1358	bpb->fat_fat_length = le16_to_cpu(b->fat_length);
1359	bpb->fat_total_sect = le32_to_cpu(b->total_sect);
1360
1361	bpb->fat16_state = b->fat16.state;
1362	bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
1363
1364	bpb->fat32_length = le32_to_cpu(b->fat32.length);
1365	bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
1366	bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
1367	bpb->fat32_state = b->fat32.state;
1368	bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
1369
1370	/* Validate this looks like a FAT filesystem BPB */
1371	if (!bpb->fat_reserved) {
1372		if (!silent)
1373			fat_msg(sb, KERN_ERR,
1374				"bogus number of reserved sectors");
1375		goto out;
1376	}
1377	if (!bpb->fat_fats) {
1378		if (!silent)
1379			fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1380		goto out;
1381	}
1382
1383	/*
1384	 * Earlier we checked here that b->secs_track and b->head are nonzero,
1385	 * but it turns out valid FAT filesystems can have zero there.
1386	 */
1387
1388	if (!fat_valid_media(b->media)) {
1389		if (!silent)
1390			fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1391				(unsigned)b->media);
1392		goto out;
1393	}
1394
1395	if (!is_power_of_2(bpb->fat_sector_size)
1396	    || (bpb->fat_sector_size < 512)
1397	    || (bpb->fat_sector_size > 4096)) {
1398		if (!silent)
1399			fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1400			       (unsigned)bpb->fat_sector_size);
1401		goto out;
1402	}
1403
1404	if (!is_power_of_2(bpb->fat_sec_per_clus)) {
1405		if (!silent)
1406			fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1407				(unsigned)bpb->fat_sec_per_clus);
1408		goto out;
1409	}
1410
1411	error = 0;
1412
1413out:
1414	return error;
1415}
1416
1417static int fat_read_static_bpb(struct super_block *sb,
1418	struct fat_boot_sector *b, int silent,
1419	struct fat_bios_param_block *bpb)
1420{
1421	static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
1422
1423	struct fat_floppy_defaults *fdefaults = NULL;
1424	int error = -EINVAL;
1425	sector_t bd_sects;
1426	unsigned i;
1427
1428	bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
1429
1430	/* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
1431	if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
1432		if (!silent)
1433			fat_msg(sb, KERN_ERR,
1434				"%s; no bootstrapping code", notdos1x);
1435		goto out;
1436	}
1437
1438	/*
1439	 * If any value in this region is non-zero, it isn't archaic
1440	 * DOS.
1441	 */
1442	if (!fat_bpb_is_zero(b)) {
1443		if (!silent)
1444			fat_msg(sb, KERN_ERR,
1445				"%s; DOS 2.x BPB is non-zero", notdos1x);
1446		goto out;
1447	}
1448
1449	for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
1450		if (floppy_defaults[i].nr_sectors == bd_sects) {
1451			fdefaults = &floppy_defaults[i];
1452			break;
1453		}
1454	}
1455
1456	if (fdefaults == NULL) {
1457		if (!silent)
1458			fat_msg(sb, KERN_WARNING,
1459				"This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
1460				(u64)bd_sects);
1461		goto out;
1462	}
1463
1464	if (!silent)
1465		fat_msg(sb, KERN_INFO,
1466			"This looks like a DOS 1.x volume; assuming default BPB values");
1467
1468	memset(bpb, 0, sizeof(*bpb));
1469	bpb->fat_sector_size = SECTOR_SIZE;
1470	bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
1471	bpb->fat_reserved = 1;
1472	bpb->fat_fats = 2;
1473	bpb->fat_dir_entries = fdefaults->dir_entries;
1474	bpb->fat_sectors = fdefaults->nr_sectors;
1475	bpb->fat_fat_length = fdefaults->fat_length;
1476
1477	error = 0;
1478
1479out:
1480	return error;
1481}
1482
1483/*
1484 * Read the super block of an MS-DOS FS.
1485 */
1486int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1487		   void (*setup)(struct super_block *))
1488{
1489	struct inode *root_inode = NULL, *fat_inode = NULL;
1490	struct inode *fsinfo_inode = NULL;
1491	struct buffer_head *bh;
1492	struct fat_bios_param_block bpb;
1493	struct msdos_sb_info *sbi;
1494	u16 logical_sector_size;
1495	u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1496	int debug;
1497	long error;
1498	char buf[50];
1499
1500	/*
1501	 * GFP_KERNEL is ok here, because while we do hold the
1502	 * supeblock lock, memory pressure can't call back into
1503	 * the filesystem, since we're only just about to mount
1504	 * it and have no inodes etc active!
1505	 */
1506	sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1507	if (!sbi)
1508		return -ENOMEM;
1509	sb->s_fs_info = sbi;
1510
1511	sb->s_flags |= MS_NODIRATIME;
1512	sb->s_magic = MSDOS_SUPER_MAGIC;
1513	sb->s_op = &fat_sops;
1514	sb->s_export_op = &fat_export_ops;
1515	mutex_init(&sbi->nfs_build_inode_lock);
1516	ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1517			     DEFAULT_RATELIMIT_BURST);
1518
1519	error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1520	if (error)
1521		goto out_fail;
1522
1523	setup(sb); /* flavour-specific stuff that needs options */
1524
1525	error = -EIO;
1526	sb_min_blocksize(sb, 512);
1527	bh = sb_bread(sb, 0);
1528	if (bh == NULL) {
1529		fat_msg(sb, KERN_ERR, "unable to read boot sector");
1530		goto out_fail;
1531	}
1532
1533	error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
1534		&bpb);
1535	if (error == -EINVAL && sbi->options.dos1xfloppy)
1536		error = fat_read_static_bpb(sb,
1537			(struct fat_boot_sector *)bh->b_data, silent, &bpb);
1538	brelse(bh);
1539
1540	if (error == -EINVAL)
1541		goto out_invalid;
1542	else if (error)
1543		goto out_fail;
1544
1545	logical_sector_size = bpb.fat_sector_size;
1546	sbi->sec_per_clus = bpb.fat_sec_per_clus;
1547
1548	error = -EIO;
1549	if (logical_sector_size < sb->s_blocksize) {
1550		fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1551		       " (logical sector size = %u)", logical_sector_size);
1552		goto out_fail;
1553	}
1554
1555	if (logical_sector_size > sb->s_blocksize) {
1556		struct buffer_head *bh_resize;
1557
1558		if (!sb_set_blocksize(sb, logical_sector_size)) {
1559			fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1560			       logical_sector_size);
1561			goto out_fail;
1562		}
1563
1564		/* Verify that the larger boot sector is fully readable */
1565		bh_resize = sb_bread(sb, 0);
1566		if (bh_resize == NULL) {
1567			fat_msg(sb, KERN_ERR, "unable to read boot sector"
1568			       " (logical sector size = %lu)",
1569			       sb->s_blocksize);
1570			goto out_fail;
1571		}
1572		brelse(bh_resize);
1573	}
1574
1575	mutex_init(&sbi->s_lock);
1576	sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1577	sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1578	sbi->fats = bpb.fat_fats;
1579	sbi->fat_bits = 0;		/* Don't know yet */
1580	sbi->fat_start = bpb.fat_reserved;
1581	sbi->fat_length = bpb.fat_fat_length;
1582	sbi->root_cluster = 0;
1583	sbi->free_clusters = -1;	/* Don't know yet */
1584	sbi->free_clus_valid = 0;
1585	sbi->prev_free = FAT_START_ENT;
1586	sb->s_maxbytes = 0xffffffff;
1587
1588	if (!sbi->fat_length && bpb.fat32_length) {
1589		struct fat_boot_fsinfo *fsinfo;
1590		struct buffer_head *fsinfo_bh;
1591
1592		/* Must be FAT32 */
1593		sbi->fat_bits = 32;
1594		sbi->fat_length = bpb.fat32_length;
1595		sbi->root_cluster = bpb.fat32_root_cluster;
1596
1597		/* MC - if info_sector is 0, don't multiply by 0 */
1598		sbi->fsinfo_sector = bpb.fat32_info_sector;
1599		if (sbi->fsinfo_sector == 0)
1600			sbi->fsinfo_sector = 1;
1601
1602		fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1603		if (fsinfo_bh == NULL) {
1604			fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1605			       " (sector = %lu)", sbi->fsinfo_sector);
1606			goto out_fail;
1607		}
1608
1609		fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1610		if (!IS_FSINFO(fsinfo)) {
1611			fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1612			       "0x%08x, 0x%08x (sector = %lu)",
1613			       le32_to_cpu(fsinfo->signature1),
1614			       le32_to_cpu(fsinfo->signature2),
1615			       sbi->fsinfo_sector);
1616		} else {
1617			if (sbi->options.usefree)
1618				sbi->free_clus_valid = 1;
1619			sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1620			sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1621		}
1622
1623		brelse(fsinfo_bh);
1624	}
1625
1626	/* interpret volume ID as a little endian 32 bit integer */
1627	if (sbi->fat_bits == 32)
1628		sbi->vol_id = bpb.fat32_vol_id;
1629	else /* fat 16 or 12 */
1630		sbi->vol_id = bpb.fat16_vol_id;
1631
1632	sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1633	sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1634
1635	sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1636	sbi->dir_entries = bpb.fat_dir_entries;
1637	if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1638		if (!silent)
1639			fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
1640			       " (%u)", sbi->dir_entries);
1641		goto out_invalid;
1642	}
1643
1644	rootdir_sectors = sbi->dir_entries
1645		* sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1646	sbi->data_start = sbi->dir_start + rootdir_sectors;
1647	total_sectors = bpb.fat_sectors;
1648	if (total_sectors == 0)
1649		total_sectors = bpb.fat_total_sect;
1650
1651	total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1652
1653	if (sbi->fat_bits != 32)
1654		sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1655
1656	/* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1657	if (sbi->fat_bits == 32)
1658		sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
1659	else /* fat 16 or 12 */
1660		sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
1661
1662	/* check that FAT table does not overflow */
1663	fat_clusters = calc_fat_clusters(sb);
1664	total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1665	if (total_clusters > MAX_FAT(sb)) {
1666		if (!silent)
1667			fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1668			       total_clusters);
1669		goto out_invalid;
1670	}
1671
1672	sbi->max_cluster = total_clusters + FAT_START_ENT;
1673	/* check the free_clusters, it's not necessarily correct */
1674	if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1675		sbi->free_clusters = -1;
1676	/* check the prev_free, it's not necessarily correct */
1677	sbi->prev_free %= sbi->max_cluster;
1678	if (sbi->prev_free < FAT_START_ENT)
1679		sbi->prev_free = FAT_START_ENT;
1680
1681	/* set up enough so that it can read an inode */
1682	fat_hash_init(sb);
1683	dir_hash_init(sb);
1684	fat_ent_access_init(sb);
1685
1686	/*
1687	 * The low byte of FAT's first entry must have same value with
1688	 * media-field.  But in real world, too many devices is
1689	 * writing wrong value.  So, removed that validity check.
1690	 *
1691	 * if (FAT_FIRST_ENT(sb, media) != first)
1692	 */
1693
1694	error = -EINVAL;
1695	sprintf(buf, "cp%d", sbi->options.codepage);
1696	sbi->nls_disk = load_nls(buf);
1697	if (!sbi->nls_disk) {
1698		fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1699		goto out_fail;
1700	}
1701
1702	/* FIXME: utf8 is using iocharset for upper/lower conversion */
1703	if (sbi->options.isvfat) {
1704		sbi->nls_io = load_nls(sbi->options.iocharset);
1705		if (!sbi->nls_io) {
1706			fat_msg(sb, KERN_ERR, "IO charset %s not found",
1707			       sbi->options.iocharset);
1708			goto out_fail;
1709		}
1710	}
1711
1712	error = -ENOMEM;
1713	fat_inode = new_inode(sb);
1714	if (!fat_inode)
1715		goto out_fail;
1716	MSDOS_I(fat_inode)->i_pos = 0;
1717	sbi->fat_inode = fat_inode;
1718
1719	fsinfo_inode = new_inode(sb);
1720	if (!fsinfo_inode)
1721		goto out_fail;
1722	fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1723	sbi->fsinfo_inode = fsinfo_inode;
1724	insert_inode_hash(fsinfo_inode);
1725
1726	root_inode = new_inode(sb);
1727	if (!root_inode)
1728		goto out_fail;
1729	root_inode->i_ino = MSDOS_ROOT_INO;
1730	root_inode->i_version = 1;
1731	error = fat_read_root(root_inode);
1732	if (error < 0) {
1733		iput(root_inode);
1734		goto out_fail;
1735	}
1736	error = -ENOMEM;
1737	insert_inode_hash(root_inode);
1738	fat_attach(root_inode, 0);
1739	sb->s_root = d_make_root(root_inode);
1740	if (!sb->s_root) {
1741		fat_msg(sb, KERN_ERR, "get root inode failed");
1742		goto out_fail;
1743	}
1744
1745	if (sbi->options.discard) {
1746		struct request_queue *q = bdev_get_queue(sb->s_bdev);
1747		if (!blk_queue_discard(q))
1748			fat_msg(sb, KERN_WARNING,
1749					"mounting with \"discard\" option, but "
1750					"the device does not support discard");
1751	}
1752
1753	fat_set_state(sb, 1, 0);
1754	return 0;
1755
1756out_invalid:
1757	error = -EINVAL;
1758	if (!silent)
1759		fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1760
1761out_fail:
1762	if (fsinfo_inode)
1763		iput(fsinfo_inode);
1764	if (fat_inode)
1765		iput(fat_inode);
1766	unload_nls(sbi->nls_io);
1767	unload_nls(sbi->nls_disk);
1768	if (sbi->options.iocharset != fat_default_iocharset)
1769		kfree(sbi->options.iocharset);
1770	sb->s_fs_info = NULL;
1771	kfree(sbi);
1772	return error;
1773}
1774
1775EXPORT_SYMBOL_GPL(fat_fill_super);
1776
1777/*
1778 * helper function for fat_flush_inodes.  This writes both the inode
1779 * and the file data blocks, waiting for in flight data blocks before
1780 * the start of the call.  It does not wait for any io started
1781 * during the call
1782 */
1783static int writeback_inode(struct inode *inode)
1784{
1785
1786	int ret;
1787
1788	/* if we used wait=1, sync_inode_metadata waits for the io for the
1789	* inode to finish.  So wait=0 is sent down to sync_inode_metadata
1790	* and filemap_fdatawrite is used for the data blocks
1791	*/
1792	ret = sync_inode_metadata(inode, 0);
1793	if (!ret)
1794		ret = filemap_fdatawrite(inode->i_mapping);
1795	return ret;
1796}
1797
1798/*
1799 * write data and metadata corresponding to i1 and i2.  The io is
1800 * started but we do not wait for any of it to finish.
1801 *
1802 * filemap_flush is used for the block device, so if there is a dirty
1803 * page for a block already in flight, we will not wait and start the
1804 * io over again
1805 */
1806int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1807{
1808	int ret = 0;
1809	if (!MSDOS_SB(sb)->options.flush)
1810		return 0;
1811	if (i1)
1812		ret = writeback_inode(i1);
1813	if (!ret && i2)
1814		ret = writeback_inode(i2);
1815	if (!ret) {
1816		struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1817		ret = filemap_flush(mapping);
1818	}
1819	return ret;
1820}
1821EXPORT_SYMBOL_GPL(fat_flush_inodes);
1822
1823static int __init init_fat_fs(void)
1824{
1825	int err;
1826
1827	err = fat_cache_init();
1828	if (err)
1829		return err;
1830
1831	err = fat_init_inodecache();
1832	if (err)
1833		goto failed;
1834
1835	return 0;
1836
1837failed:
1838	fat_cache_destroy();
1839	return err;
1840}
1841
1842static void __exit exit_fat_fs(void)
1843{
1844	fat_cache_destroy();
1845	fat_destroy_inodecache();
1846}
1847
1848module_init(init_fat_fs)
1849module_exit(exit_fat_fs)
1850
1851MODULE_LICENSE("GPL");
1852