Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / fat / inode.c
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 <asm/unaligned.h>
22 #include "fat.h"
23
24 #ifndef CONFIG_FAT_DEFAULT_IOCHARSET
25 /* if user don't select VFAT, this is undefined. */
26 #define CONFIG_FAT_DEFAULT_IOCHARSET    ""
27 #endif
28
29 #define KB_IN_SECTORS 2
30
31 /*
32  * A deserialized copy of the on-disk structure laid out in struct
33  * fat_boot_sector.
34  */
35 struct fat_bios_param_block {
36         u16     fat_sector_size;
37         u8      fat_sec_per_clus;
38         u16     fat_reserved;
39         u8      fat_fats;
40         u16     fat_dir_entries;
41         u16     fat_sectors;
42         u16     fat_fat_length;
43         u32     fat_total_sect;
44
45         u8      fat16_state;
46         u32     fat16_vol_id;
47
48         u32     fat32_length;
49         u32     fat32_root_cluster;
50         u16     fat32_info_sector;
51         u8      fat32_state;
52         u32     fat32_vol_id;
53 };
54
55 static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
56 static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
57
58 static struct fat_floppy_defaults {
59         unsigned nr_sectors;
60         unsigned sec_per_clus;
61         unsigned dir_entries;
62         unsigned media;
63         unsigned fat_length;
64 } floppy_defaults[] = {
65 {
66         .nr_sectors = 160 * KB_IN_SECTORS,
67         .sec_per_clus = 1,
68         .dir_entries = 64,
69         .media = 0xFE,
70         .fat_length = 1,
71 },
72 {
73         .nr_sectors = 180 * KB_IN_SECTORS,
74         .sec_per_clus = 1,
75         .dir_entries = 64,
76         .media = 0xFC,
77         .fat_length = 2,
78 },
79 {
80         .nr_sectors = 320 * KB_IN_SECTORS,
81         .sec_per_clus = 2,
82         .dir_entries = 112,
83         .media = 0xFF,
84         .fat_length = 1,
85 },
86 {
87         .nr_sectors = 360 * KB_IN_SECTORS,
88         .sec_per_clus = 2,
89         .dir_entries = 112,
90         .media = 0xFD,
91         .fat_length = 2,
92 },
93 };
94
95 static int fat_add_cluster(struct inode *inode)
96 {
97         int err, cluster;
98
99         err = fat_alloc_clusters(inode, &cluster, 1);
100         if (err)
101                 return err;
102         /* FIXME: this cluster should be added after data of this
103          * cluster is writed */
104         err = fat_chain_add(inode, cluster, 1);
105         if (err)
106                 fat_free_clusters(inode, cluster);
107         return err;
108 }
109
110 static inline int __fat_get_block(struct inode *inode, sector_t iblock,
111                                   unsigned long *max_blocks,
112                                   struct buffer_head *bh_result, int create)
113 {
114         struct super_block *sb = inode->i_sb;
115         struct msdos_sb_info *sbi = MSDOS_SB(sb);
116         unsigned long mapped_blocks;
117         sector_t phys;
118         int err, offset;
119
120         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
121         if (err)
122                 return err;
123         if (phys) {
124                 map_bh(bh_result, sb, phys);
125                 *max_blocks = min(mapped_blocks, *max_blocks);
126                 return 0;
127         }
128         if (!create)
129                 return 0;
130
131         if (iblock != MSDOS_I(inode)->mmu_private >> sb->s_blocksize_bits) {
132                 fat_fs_error(sb, "corrupted file size (i_pos %lld, %lld)",
133                         MSDOS_I(inode)->i_pos, MSDOS_I(inode)->mmu_private);
134                 return -EIO;
135         }
136
137         offset = (unsigned long)iblock & (sbi->sec_per_clus - 1);
138         if (!offset) {
139                 /* TODO: multiple cluster allocation would be desirable. */
140                 err = fat_add_cluster(inode);
141                 if (err)
142                         return err;
143         }
144         /* available blocks on this cluster */
145         mapped_blocks = sbi->sec_per_clus - offset;
146
147         *max_blocks = min(mapped_blocks, *max_blocks);
148         MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
149
150         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
151         if (err)
152                 return err;
153
154         BUG_ON(!phys);
155         BUG_ON(*max_blocks != mapped_blocks);
156         set_buffer_new(bh_result);
157         map_bh(bh_result, sb, phys);
158
159         return 0;
160 }
161
162 static int fat_get_block(struct inode *inode, sector_t iblock,
163                          struct buffer_head *bh_result, int create)
164 {
165         struct super_block *sb = inode->i_sb;
166         unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
167         int err;
168
169         err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
170         if (err)
171                 return err;
172         bh_result->b_size = max_blocks << sb->s_blocksize_bits;
173         return 0;
174 }
175
176 static int fat_writepage(struct page *page, struct writeback_control *wbc)
177 {
178         return block_write_full_page(page, fat_get_block, wbc);
179 }
180
181 static int fat_writepages(struct address_space *mapping,
182                           struct writeback_control *wbc)
183 {
184         return mpage_writepages(mapping, wbc, fat_get_block);
185 }
186
187 static int fat_readpage(struct file *file, struct page *page)
188 {
189         return mpage_readpage(page, fat_get_block);
190 }
191
192 static int fat_readpages(struct file *file, struct address_space *mapping,
193                          struct list_head *pages, unsigned nr_pages)
194 {
195         return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
196 }
197
198 static void fat_write_failed(struct address_space *mapping, loff_t to)
199 {
200         struct inode *inode = mapping->host;
201
202         if (to > inode->i_size) {
203                 truncate_pagecache(inode, inode->i_size);
204                 fat_truncate_blocks(inode, inode->i_size);
205         }
206 }
207
208 static int fat_write_begin(struct file *file, struct address_space *mapping,
209                         loff_t pos, unsigned len, unsigned flags,
210                         struct page **pagep, void **fsdata)
211 {
212         int err;
213
214         *pagep = NULL;
215         err = cont_write_begin(file, mapping, pos, len, flags,
216                                 pagep, fsdata, fat_get_block,
217                                 &MSDOS_I(mapping->host)->mmu_private);
218         if (err < 0)
219                 fat_write_failed(mapping, pos + len);
220         return err;
221 }
222
223 static int fat_write_end(struct file *file, struct address_space *mapping,
224                         loff_t pos, unsigned len, unsigned copied,
225                         struct page *pagep, void *fsdata)
226 {
227         struct inode *inode = mapping->host;
228         int err;
229         err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
230         if (err < len)
231                 fat_write_failed(mapping, pos + len);
232         if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
233                 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
234                 MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
235                 mark_inode_dirty(inode);
236         }
237         return err;
238 }
239
240 static ssize_t fat_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
241                              loff_t offset)
242 {
243         struct file *file = iocb->ki_filp;
244         struct address_space *mapping = file->f_mapping;
245         struct inode *inode = mapping->host;
246         size_t count = iov_iter_count(iter);
247         ssize_t ret;
248
249         if (iov_iter_rw(iter) == WRITE) {
250                 /*
251                  * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
252                  * so we need to update the ->mmu_private to block boundary.
253                  *
254                  * But we must fill the remaining area or hole by nul for
255                  * updating ->mmu_private.
256                  *
257                  * Return 0, and fallback to normal buffered write.
258                  */
259                 loff_t size = offset + count;
260                 if (MSDOS_I(inode)->mmu_private < size)
261                         return 0;
262         }
263
264         /*
265          * FAT need to use the DIO_LOCKING for avoiding the race
266          * condition of fat_get_block() and ->truncate().
267          */
268         ret = blockdev_direct_IO(iocb, inode, iter, offset, fat_get_block);
269         if (ret < 0 && iov_iter_rw(iter) == WRITE)
270                 fat_write_failed(mapping, offset + count);
271
272         return ret;
273 }
274
275 static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
276 {
277         sector_t blocknr;
278
279         /* fat_get_cluster() assumes the requested blocknr isn't truncated. */
280         down_read(&MSDOS_I(mapping->host)->truncate_lock);
281         blocknr = generic_block_bmap(mapping, block, fat_get_block);
282         up_read(&MSDOS_I(mapping->host)->truncate_lock);
283
284         return blocknr;
285 }
286
287 /*
288  * fat_block_truncate_page() zeroes out a mapping from file offset `from'
289  * up to the end of the block which corresponds to `from'.
290  * This is required during truncate to physically zeroout the tail end
291  * of that block so it doesn't yield old data if the file is later grown.
292  * Also, avoid causing failure from fsx for cases of "data past EOF"
293  */
294 int fat_block_truncate_page(struct inode *inode, loff_t from)
295 {
296         return block_truncate_page(inode->i_mapping, from, fat_get_block);
297 }
298
299 static const struct address_space_operations fat_aops = {
300         .readpage       = fat_readpage,
301         .readpages      = fat_readpages,
302         .writepage      = fat_writepage,
303         .writepages     = fat_writepages,
304         .write_begin    = fat_write_begin,
305         .write_end      = fat_write_end,
306         .direct_IO      = fat_direct_IO,
307         .bmap           = _fat_bmap
308 };
309
310 /*
311  * New FAT inode stuff. We do the following:
312  *      a) i_ino is constant and has nothing with on-disk location.
313  *      b) FAT manages its own cache of directory entries.
314  *      c) *This* cache is indexed by on-disk location.
315  *      d) inode has an associated directory entry, all right, but
316  *              it may be unhashed.
317  *      e) currently entries are stored within struct inode. That should
318  *              change.
319  *      f) we deal with races in the following way:
320  *              1. readdir() and lookup() do FAT-dir-cache lookup.
321  *              2. rename() unhashes the F-d-c entry and rehashes it in
322  *                      a new place.
323  *              3. unlink() and rmdir() unhash F-d-c entry.
324  *              4. fat_write_inode() checks whether the thing is unhashed.
325  *                      If it is we silently return. If it isn't we do bread(),
326  *                      check if the location is still valid and retry if it
327  *                      isn't. Otherwise we do changes.
328  *              5. Spinlock is used to protect hash/unhash/location check/lookup
329  *              6. fat_evict_inode() unhashes the F-d-c entry.
330  *              7. lookup() and readdir() do igrab() if they find a F-d-c entry
331  *                      and consider negative result as cache miss.
332  */
333
334 static void fat_hash_init(struct super_block *sb)
335 {
336         struct msdos_sb_info *sbi = MSDOS_SB(sb);
337         int i;
338
339         spin_lock_init(&sbi->inode_hash_lock);
340         for (i = 0; i < FAT_HASH_SIZE; i++)
341                 INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
342 }
343
344 static inline unsigned long fat_hash(loff_t i_pos)
345 {
346         return hash_32(i_pos, FAT_HASH_BITS);
347 }
348
349 static void dir_hash_init(struct super_block *sb)
350 {
351         struct msdos_sb_info *sbi = MSDOS_SB(sb);
352         int i;
353
354         spin_lock_init(&sbi->dir_hash_lock);
355         for (i = 0; i < FAT_HASH_SIZE; i++)
356                 INIT_HLIST_HEAD(&sbi->dir_hashtable[i]);
357 }
358
359 void fat_attach(struct inode *inode, loff_t i_pos)
360 {
361         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
362
363         if (inode->i_ino != MSDOS_ROOT_INO) {
364                 struct hlist_head *head =   sbi->inode_hashtable
365                                           + fat_hash(i_pos);
366
367                 spin_lock(&sbi->inode_hash_lock);
368                 MSDOS_I(inode)->i_pos = i_pos;
369                 hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
370                 spin_unlock(&sbi->inode_hash_lock);
371         }
372
373         /* If NFS support is enabled, cache the mapping of start cluster
374          * to directory inode. This is used during reconnection of
375          * dentries to the filesystem root.
376          */
377         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
378                 struct hlist_head *d_head = sbi->dir_hashtable;
379                 d_head += fat_dir_hash(MSDOS_I(inode)->i_logstart);
380
381                 spin_lock(&sbi->dir_hash_lock);
382                 hlist_add_head(&MSDOS_I(inode)->i_dir_hash, d_head);
383                 spin_unlock(&sbi->dir_hash_lock);
384         }
385 }
386 EXPORT_SYMBOL_GPL(fat_attach);
387
388 void fat_detach(struct inode *inode)
389 {
390         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
391         spin_lock(&sbi->inode_hash_lock);
392         MSDOS_I(inode)->i_pos = 0;
393         hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
394         spin_unlock(&sbi->inode_hash_lock);
395
396         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
397                 spin_lock(&sbi->dir_hash_lock);
398                 hlist_del_init(&MSDOS_I(inode)->i_dir_hash);
399                 spin_unlock(&sbi->dir_hash_lock);
400         }
401 }
402 EXPORT_SYMBOL_GPL(fat_detach);
403
404 struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
405 {
406         struct msdos_sb_info *sbi = MSDOS_SB(sb);
407         struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
408         struct msdos_inode_info *i;
409         struct inode *inode = NULL;
410
411         spin_lock(&sbi->inode_hash_lock);
412         hlist_for_each_entry(i, head, i_fat_hash) {
413                 BUG_ON(i->vfs_inode.i_sb != sb);
414                 if (i->i_pos != i_pos)
415                         continue;
416                 inode = igrab(&i->vfs_inode);
417                 if (inode)
418                         break;
419         }
420         spin_unlock(&sbi->inode_hash_lock);
421         return inode;
422 }
423
424 static int is_exec(unsigned char *extension)
425 {
426         unsigned char exe_extensions[] = "EXECOMBAT", *walk;
427
428         for (walk = exe_extensions; *walk; walk += 3)
429                 if (!strncmp(extension, walk, 3))
430                         return 1;
431         return 0;
432 }
433
434 static int fat_calc_dir_size(struct inode *inode)
435 {
436         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
437         int ret, fclus, dclus;
438
439         inode->i_size = 0;
440         if (MSDOS_I(inode)->i_start == 0)
441                 return 0;
442
443         ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
444         if (ret < 0)
445                 return ret;
446         inode->i_size = (fclus + 1) << sbi->cluster_bits;
447
448         return 0;
449 }
450
451 /* doesn't deal with root inode */
452 int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
453 {
454         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
455         int error;
456
457         MSDOS_I(inode)->i_pos = 0;
458         inode->i_uid = sbi->options.fs_uid;
459         inode->i_gid = sbi->options.fs_gid;
460         inode->i_version++;
461         inode->i_generation = get_seconds();
462
463         if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
464                 inode->i_generation &= ~1;
465                 inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
466                 inode->i_op = sbi->dir_ops;
467                 inode->i_fop = &fat_dir_operations;
468
469                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
470                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
471                 error = fat_calc_dir_size(inode);
472                 if (error < 0)
473                         return error;
474                 MSDOS_I(inode)->mmu_private = inode->i_size;
475
476                 set_nlink(inode, fat_subdirs(inode));
477         } else { /* not a directory */
478                 inode->i_generation |= 1;
479                 inode->i_mode = fat_make_mode(sbi, de->attr,
480                         ((sbi->options.showexec && !is_exec(de->name + 8))
481                          ? S_IRUGO|S_IWUGO : S_IRWXUGO));
482                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
483
484                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
485                 inode->i_size = le32_to_cpu(de->size);
486                 inode->i_op = &fat_file_inode_operations;
487                 inode->i_fop = &fat_file_operations;
488                 inode->i_mapping->a_ops = &fat_aops;
489                 MSDOS_I(inode)->mmu_private = inode->i_size;
490         }
491         if (de->attr & ATTR_SYS) {
492                 if (sbi->options.sys_immutable)
493                         inode->i_flags |= S_IMMUTABLE;
494         }
495         fat_save_attrs(inode, de->attr);
496
497         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
498                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
499
500         fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
501         if (sbi->options.isvfat) {
502                 fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
503                                   de->cdate, de->ctime_cs);
504                 fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
505         } else
506                 inode->i_ctime = inode->i_atime = inode->i_mtime;
507
508         return 0;
509 }
510
511 static inline void fat_lock_build_inode(struct msdos_sb_info *sbi)
512 {
513         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
514                 mutex_lock(&sbi->nfs_build_inode_lock);
515 }
516
517 static inline void fat_unlock_build_inode(struct msdos_sb_info *sbi)
518 {
519         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
520                 mutex_unlock(&sbi->nfs_build_inode_lock);
521 }
522
523 struct inode *fat_build_inode(struct super_block *sb,
524                         struct msdos_dir_entry *de, loff_t i_pos)
525 {
526         struct inode *inode;
527         int err;
528
529         fat_lock_build_inode(MSDOS_SB(sb));
530         inode = fat_iget(sb, i_pos);
531         if (inode)
532                 goto out;
533         inode = new_inode(sb);
534         if (!inode) {
535                 inode = ERR_PTR(-ENOMEM);
536                 goto out;
537         }
538         inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
539         inode->i_version = 1;
540         err = fat_fill_inode(inode, de);
541         if (err) {
542                 iput(inode);
543                 inode = ERR_PTR(err);
544                 goto out;
545         }
546         fat_attach(inode, i_pos);
547         insert_inode_hash(inode);
548 out:
549         fat_unlock_build_inode(MSDOS_SB(sb));
550         return inode;
551 }
552
553 EXPORT_SYMBOL_GPL(fat_build_inode);
554
555 static void fat_evict_inode(struct inode *inode)
556 {
557         truncate_inode_pages_final(&inode->i_data);
558         if (!inode->i_nlink) {
559                 inode->i_size = 0;
560                 fat_truncate_blocks(inode, 0);
561         }
562         invalidate_inode_buffers(inode);
563         clear_inode(inode);
564         fat_cache_inval_inode(inode);
565         fat_detach(inode);
566 }
567
568 static void fat_set_state(struct super_block *sb,
569                         unsigned int set, unsigned int force)
570 {
571         struct buffer_head *bh;
572         struct fat_boot_sector *b;
573         struct msdos_sb_info *sbi = MSDOS_SB(sb);
574
575         /* do not change any thing if mounted read only */
576         if ((sb->s_flags & MS_RDONLY) && !force)
577                 return;
578
579         /* do not change state if fs was dirty */
580         if (sbi->dirty) {
581                 /* warn only on set (mount). */
582                 if (set)
583                         fat_msg(sb, KERN_WARNING, "Volume was not properly "
584                                 "unmounted. Some data may be corrupt. "
585                                 "Please run fsck.");
586                 return;
587         }
588
589         bh = sb_bread(sb, 0);
590         if (bh == NULL) {
591                 fat_msg(sb, KERN_ERR, "unable to read boot sector "
592                         "to mark fs as dirty");
593                 return;
594         }
595
596         b = (struct fat_boot_sector *) bh->b_data;
597
598         if (sbi->fat_bits == 32) {
599                 if (set)
600                         b->fat32.state |= FAT_STATE_DIRTY;
601                 else
602                         b->fat32.state &= ~FAT_STATE_DIRTY;
603         } else /* fat 16 and 12 */ {
604                 if (set)
605                         b->fat16.state |= FAT_STATE_DIRTY;
606                 else
607                         b->fat16.state &= ~FAT_STATE_DIRTY;
608         }
609
610         mark_buffer_dirty(bh);
611         sync_dirty_buffer(bh);
612         brelse(bh);
613 }
614
615 static void delayed_free(struct rcu_head *p)
616 {
617         struct msdos_sb_info *sbi = container_of(p, struct msdos_sb_info, rcu);
618         unload_nls(sbi->nls_disk);
619         unload_nls(sbi->nls_io);
620         if (sbi->options.iocharset != fat_default_iocharset)
621                 kfree(sbi->options.iocharset);
622         kfree(sbi);
623 }
624
625 static void fat_put_super(struct super_block *sb)
626 {
627         struct msdos_sb_info *sbi = MSDOS_SB(sb);
628
629         fat_set_state(sb, 0, 0);
630
631         iput(sbi->fsinfo_inode);
632         iput(sbi->fat_inode);
633
634         call_rcu(&sbi->rcu, delayed_free);
635 }
636
637 static struct kmem_cache *fat_inode_cachep;
638
639 static struct inode *fat_alloc_inode(struct super_block *sb)
640 {
641         struct msdos_inode_info *ei;
642         ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
643         if (!ei)
644                 return NULL;
645
646         init_rwsem(&ei->truncate_lock);
647         return &ei->vfs_inode;
648 }
649
650 static void fat_i_callback(struct rcu_head *head)
651 {
652         struct inode *inode = container_of(head, struct inode, i_rcu);
653         kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
654 }
655
656 static void fat_destroy_inode(struct inode *inode)
657 {
658         call_rcu(&inode->i_rcu, fat_i_callback);
659 }
660
661 static void init_once(void *foo)
662 {
663         struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
664
665         spin_lock_init(&ei->cache_lru_lock);
666         ei->nr_caches = 0;
667         ei->cache_valid_id = FAT_CACHE_VALID + 1;
668         INIT_LIST_HEAD(&ei->cache_lru);
669         INIT_HLIST_NODE(&ei->i_fat_hash);
670         INIT_HLIST_NODE(&ei->i_dir_hash);
671         inode_init_once(&ei->vfs_inode);
672 }
673
674 static int __init fat_init_inodecache(void)
675 {
676         fat_inode_cachep = kmem_cache_create("fat_inode_cache",
677                                              sizeof(struct msdos_inode_info),
678                                              0, (SLAB_RECLAIM_ACCOUNT|
679                                                 SLAB_MEM_SPREAD),
680                                              init_once);
681         if (fat_inode_cachep == NULL)
682                 return -ENOMEM;
683         return 0;
684 }
685
686 static void __exit fat_destroy_inodecache(void)
687 {
688         /*
689          * Make sure all delayed rcu free inodes are flushed before we
690          * destroy cache.
691          */
692         rcu_barrier();
693         kmem_cache_destroy(fat_inode_cachep);
694 }
695
696 static int fat_remount(struct super_block *sb, int *flags, char *data)
697 {
698         int new_rdonly;
699         struct msdos_sb_info *sbi = MSDOS_SB(sb);
700         *flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
701
702         sync_filesystem(sb);
703
704         /* make sure we update state on remount. */
705         new_rdonly = *flags & MS_RDONLY;
706         if (new_rdonly != (sb->s_flags & MS_RDONLY)) {
707                 if (new_rdonly)
708                         fat_set_state(sb, 0, 0);
709                 else
710                         fat_set_state(sb, 1, 1);
711         }
712         return 0;
713 }
714
715 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
716 {
717         struct super_block *sb = dentry->d_sb;
718         struct msdos_sb_info *sbi = MSDOS_SB(sb);
719         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
720
721         /* If the count of free cluster is still unknown, counts it here. */
722         if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
723                 int err = fat_count_free_clusters(dentry->d_sb);
724                 if (err)
725                         return err;
726         }
727
728         buf->f_type = dentry->d_sb->s_magic;
729         buf->f_bsize = sbi->cluster_size;
730         buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
731         buf->f_bfree = sbi->free_clusters;
732         buf->f_bavail = sbi->free_clusters;
733         buf->f_fsid.val[0] = (u32)id;
734         buf->f_fsid.val[1] = (u32)(id >> 32);
735         buf->f_namelen =
736                 (sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
737
738         return 0;
739 }
740
741 static int __fat_write_inode(struct inode *inode, int wait)
742 {
743         struct super_block *sb = inode->i_sb;
744         struct msdos_sb_info *sbi = MSDOS_SB(sb);
745         struct buffer_head *bh;
746         struct msdos_dir_entry *raw_entry;
747         loff_t i_pos;
748         sector_t blocknr;
749         int err, offset;
750
751         if (inode->i_ino == MSDOS_ROOT_INO)
752                 return 0;
753
754 retry:
755         i_pos = fat_i_pos_read(sbi, inode);
756         if (!i_pos)
757                 return 0;
758
759         fat_get_blknr_offset(sbi, i_pos, &blocknr, &offset);
760         bh = sb_bread(sb, blocknr);
761         if (!bh) {
762                 fat_msg(sb, KERN_ERR, "unable to read inode block "
763                        "for updating (i_pos %lld)", i_pos);
764                 return -EIO;
765         }
766         spin_lock(&sbi->inode_hash_lock);
767         if (i_pos != MSDOS_I(inode)->i_pos) {
768                 spin_unlock(&sbi->inode_hash_lock);
769                 brelse(bh);
770                 goto retry;
771         }
772
773         raw_entry = &((struct msdos_dir_entry *) (bh->b_data))[offset];
774         if (S_ISDIR(inode->i_mode))
775                 raw_entry->size = 0;
776         else
777                 raw_entry->size = cpu_to_le32(inode->i_size);
778         raw_entry->attr = fat_make_attrs(inode);
779         fat_set_start(raw_entry, MSDOS_I(inode)->i_logstart);
780         fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
781                           &raw_entry->date, NULL);
782         if (sbi->options.isvfat) {
783                 __le16 atime;
784                 fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
785                                   &raw_entry->cdate, &raw_entry->ctime_cs);
786                 fat_time_unix2fat(sbi, &inode->i_atime, &atime,
787                                   &raw_entry->adate, NULL);
788         }
789         spin_unlock(&sbi->inode_hash_lock);
790         mark_buffer_dirty(bh);
791         err = 0;
792         if (wait)
793                 err = sync_dirty_buffer(bh);
794         brelse(bh);
795         return err;
796 }
797
798 static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
799 {
800         int err;
801
802         if (inode->i_ino == MSDOS_FSINFO_INO) {
803                 struct super_block *sb = inode->i_sb;
804
805                 mutex_lock(&MSDOS_SB(sb)->s_lock);
806                 err = fat_clusters_flush(sb);
807                 mutex_unlock(&MSDOS_SB(sb)->s_lock);
808         } else
809                 err = __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
810
811         return err;
812 }
813
814 int fat_sync_inode(struct inode *inode)
815 {
816         return __fat_write_inode(inode, 1);
817 }
818
819 EXPORT_SYMBOL_GPL(fat_sync_inode);
820
821 static int fat_show_options(struct seq_file *m, struct dentry *root);
822 static const struct super_operations fat_sops = {
823         .alloc_inode    = fat_alloc_inode,
824         .destroy_inode  = fat_destroy_inode,
825         .write_inode    = fat_write_inode,
826         .evict_inode    = fat_evict_inode,
827         .put_super      = fat_put_super,
828         .statfs         = fat_statfs,
829         .remount_fs     = fat_remount,
830
831         .show_options   = fat_show_options,
832 };
833
834 static int fat_show_options(struct seq_file *m, struct dentry *root)
835 {
836         struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
837         struct fat_mount_options *opts = &sbi->options;
838         int isvfat = opts->isvfat;
839
840         if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
841                 seq_printf(m, ",uid=%u",
842                                 from_kuid_munged(&init_user_ns, opts->fs_uid));
843         if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
844                 seq_printf(m, ",gid=%u",
845                                 from_kgid_munged(&init_user_ns, opts->fs_gid));
846         seq_printf(m, ",fmask=%04o", opts->fs_fmask);
847         seq_printf(m, ",dmask=%04o", opts->fs_dmask);
848         if (opts->allow_utime)
849                 seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
850         if (sbi->nls_disk)
851                 /* strip "cp" prefix from displayed option */
852                 seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
853         if (isvfat) {
854                 if (sbi->nls_io)
855                         seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
856
857                 switch (opts->shortname) {
858                 case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
859                         seq_puts(m, ",shortname=win95");
860                         break;
861                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
862                         seq_puts(m, ",shortname=winnt");
863                         break;
864                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
865                         seq_puts(m, ",shortname=mixed");
866                         break;
867                 case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
868                         seq_puts(m, ",shortname=lower");
869                         break;
870                 default:
871                         seq_puts(m, ",shortname=unknown");
872                         break;
873                 }
874         }
875         if (opts->name_check != 'n')
876                 seq_printf(m, ",check=%c", opts->name_check);
877         if (opts->usefree)
878                 seq_puts(m, ",usefree");
879         if (opts->quiet)
880                 seq_puts(m, ",quiet");
881         if (opts->showexec)
882                 seq_puts(m, ",showexec");
883         if (opts->sys_immutable)
884                 seq_puts(m, ",sys_immutable");
885         if (!isvfat) {
886                 if (opts->dotsOK)
887                         seq_puts(m, ",dotsOK=yes");
888                 if (opts->nocase)
889                         seq_puts(m, ",nocase");
890         } else {
891                 if (opts->utf8)
892                         seq_puts(m, ",utf8");
893                 if (opts->unicode_xlate)
894                         seq_puts(m, ",uni_xlate");
895                 if (!opts->numtail)
896                         seq_puts(m, ",nonumtail");
897                 if (opts->rodir)
898                         seq_puts(m, ",rodir");
899         }
900         if (opts->flush)
901                 seq_puts(m, ",flush");
902         if (opts->tz_set) {
903                 if (opts->time_offset)
904                         seq_printf(m, ",time_offset=%d", opts->time_offset);
905                 else
906                         seq_puts(m, ",tz=UTC");
907         }
908         if (opts->errors == FAT_ERRORS_CONT)
909                 seq_puts(m, ",errors=continue");
910         else if (opts->errors == FAT_ERRORS_PANIC)
911                 seq_puts(m, ",errors=panic");
912         else
913                 seq_puts(m, ",errors=remount-ro");
914         if (opts->nfs == FAT_NFS_NOSTALE_RO)
915                 seq_puts(m, ",nfs=nostale_ro");
916         else if (opts->nfs)
917                 seq_puts(m, ",nfs=stale_rw");
918         if (opts->discard)
919                 seq_puts(m, ",discard");
920         if (opts->dos1xfloppy)
921                 seq_puts(m, ",dos1xfloppy");
922
923         return 0;
924 }
925
926 enum {
927         Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
928         Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
929         Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
930         Opt_immutable, Opt_dots, Opt_nodots,
931         Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
932         Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
933         Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
934         Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
935         Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
936         Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
937 };
938
939 static const match_table_t fat_tokens = {
940         {Opt_check_r, "check=relaxed"},
941         {Opt_check_s, "check=strict"},
942         {Opt_check_n, "check=normal"},
943         {Opt_check_r, "check=r"},
944         {Opt_check_s, "check=s"},
945         {Opt_check_n, "check=n"},
946         {Opt_uid, "uid=%u"},
947         {Opt_gid, "gid=%u"},
948         {Opt_umask, "umask=%o"},
949         {Opt_dmask, "dmask=%o"},
950         {Opt_fmask, "fmask=%o"},
951         {Opt_allow_utime, "allow_utime=%o"},
952         {Opt_codepage, "codepage=%u"},
953         {Opt_usefree, "usefree"},
954         {Opt_nocase, "nocase"},
955         {Opt_quiet, "quiet"},
956         {Opt_showexec, "showexec"},
957         {Opt_debug, "debug"},
958         {Opt_immutable, "sys_immutable"},
959         {Opt_flush, "flush"},
960         {Opt_tz_utc, "tz=UTC"},
961         {Opt_time_offset, "time_offset=%d"},
962         {Opt_err_cont, "errors=continue"},
963         {Opt_err_panic, "errors=panic"},
964         {Opt_err_ro, "errors=remount-ro"},
965         {Opt_discard, "discard"},
966         {Opt_nfs_stale_rw, "nfs"},
967         {Opt_nfs_stale_rw, "nfs=stale_rw"},
968         {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
969         {Opt_dos1xfloppy, "dos1xfloppy"},
970         {Opt_obsolete, "conv=binary"},
971         {Opt_obsolete, "conv=text"},
972         {Opt_obsolete, "conv=auto"},
973         {Opt_obsolete, "conv=b"},
974         {Opt_obsolete, "conv=t"},
975         {Opt_obsolete, "conv=a"},
976         {Opt_obsolete, "fat=%u"},
977         {Opt_obsolete, "blocksize=%u"},
978         {Opt_obsolete, "cvf_format=%20s"},
979         {Opt_obsolete, "cvf_options=%100s"},
980         {Opt_obsolete, "posix"},
981         {Opt_err, NULL},
982 };
983 static const match_table_t msdos_tokens = {
984         {Opt_nodots, "nodots"},
985         {Opt_nodots, "dotsOK=no"},
986         {Opt_dots, "dots"},
987         {Opt_dots, "dotsOK=yes"},
988         {Opt_err, NULL}
989 };
990 static const match_table_t vfat_tokens = {
991         {Opt_charset, "iocharset=%s"},
992         {Opt_shortname_lower, "shortname=lower"},
993         {Opt_shortname_win95, "shortname=win95"},
994         {Opt_shortname_winnt, "shortname=winnt"},
995         {Opt_shortname_mixed, "shortname=mixed"},
996         {Opt_utf8_no, "utf8=0"},                /* 0 or no or false */
997         {Opt_utf8_no, "utf8=no"},
998         {Opt_utf8_no, "utf8=false"},
999         {Opt_utf8_yes, "utf8=1"},               /* empty or 1 or yes or true */
1000         {Opt_utf8_yes, "utf8=yes"},
1001         {Opt_utf8_yes, "utf8=true"},
1002         {Opt_utf8_yes, "utf8"},
1003         {Opt_uni_xl_no, "uni_xlate=0"},         /* 0 or no or false */
1004         {Opt_uni_xl_no, "uni_xlate=no"},
1005         {Opt_uni_xl_no, "uni_xlate=false"},
1006         {Opt_uni_xl_yes, "uni_xlate=1"},        /* empty or 1 or yes or true */
1007         {Opt_uni_xl_yes, "uni_xlate=yes"},
1008         {Opt_uni_xl_yes, "uni_xlate=true"},
1009         {Opt_uni_xl_yes, "uni_xlate"},
1010         {Opt_nonumtail_no, "nonumtail=0"},      /* 0 or no or false */
1011         {Opt_nonumtail_no, "nonumtail=no"},
1012         {Opt_nonumtail_no, "nonumtail=false"},
1013         {Opt_nonumtail_yes, "nonumtail=1"},     /* empty or 1 or yes or true */
1014         {Opt_nonumtail_yes, "nonumtail=yes"},
1015         {Opt_nonumtail_yes, "nonumtail=true"},
1016         {Opt_nonumtail_yes, "nonumtail"},
1017         {Opt_rodir, "rodir"},
1018         {Opt_err, NULL}
1019 };
1020
1021 static int parse_options(struct super_block *sb, char *options, int is_vfat,
1022                          int silent, int *debug, struct fat_mount_options *opts)
1023 {
1024         char *p;
1025         substring_t args[MAX_OPT_ARGS];
1026         int option;
1027         char *iocharset;
1028
1029         opts->isvfat = is_vfat;
1030
1031         opts->fs_uid = current_uid();
1032         opts->fs_gid = current_gid();
1033         opts->fs_fmask = opts->fs_dmask = current_umask();
1034         opts->allow_utime = -1;
1035         opts->codepage = fat_default_codepage;
1036         opts->iocharset = fat_default_iocharset;
1037         if (is_vfat) {
1038                 opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1039                 opts->rodir = 0;
1040         } else {
1041                 opts->shortname = 0;
1042                 opts->rodir = 1;
1043         }
1044         opts->name_check = 'n';
1045         opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
1046         opts->utf8 = opts->unicode_xlate = 0;
1047         opts->numtail = 1;
1048         opts->usefree = opts->nocase = 0;
1049         opts->tz_set = 0;
1050         opts->nfs = 0;
1051         opts->errors = FAT_ERRORS_RO;
1052         *debug = 0;
1053
1054         if (!options)
1055                 goto out;
1056
1057         while ((p = strsep(&options, ",")) != NULL) {
1058                 int token;
1059                 if (!*p)
1060                         continue;
1061
1062                 token = match_token(p, fat_tokens, args);
1063                 if (token == Opt_err) {
1064                         if (is_vfat)
1065                                 token = match_token(p, vfat_tokens, args);
1066                         else
1067                                 token = match_token(p, msdos_tokens, args);
1068                 }
1069                 switch (token) {
1070                 case Opt_check_s:
1071                         opts->name_check = 's';
1072                         break;
1073                 case Opt_check_r:
1074                         opts->name_check = 'r';
1075                         break;
1076                 case Opt_check_n:
1077                         opts->name_check = 'n';
1078                         break;
1079                 case Opt_usefree:
1080                         opts->usefree = 1;
1081                         break;
1082                 case Opt_nocase:
1083                         if (!is_vfat)
1084                                 opts->nocase = 1;
1085                         else {
1086                                 /* for backward compatibility */
1087                                 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1088                                         | VFAT_SFN_CREATE_WIN95;
1089                         }
1090                         break;
1091                 case Opt_quiet:
1092                         opts->quiet = 1;
1093                         break;
1094                 case Opt_showexec:
1095                         opts->showexec = 1;
1096                         break;
1097                 case Opt_debug:
1098                         *debug = 1;
1099                         break;
1100                 case Opt_immutable:
1101                         opts->sys_immutable = 1;
1102                         break;
1103                 case Opt_uid:
1104                         if (match_int(&args[0], &option))
1105                                 return -EINVAL;
1106                         opts->fs_uid = make_kuid(current_user_ns(), option);
1107                         if (!uid_valid(opts->fs_uid))
1108                                 return -EINVAL;
1109                         break;
1110                 case Opt_gid:
1111                         if (match_int(&args[0], &option))
1112                                 return -EINVAL;
1113                         opts->fs_gid = make_kgid(current_user_ns(), option);
1114                         if (!gid_valid(opts->fs_gid))
1115                                 return -EINVAL;
1116                         break;
1117                 case Opt_umask:
1118                         if (match_octal(&args[0], &option))
1119                                 return -EINVAL;
1120                         opts->fs_fmask = opts->fs_dmask = option;
1121                         break;
1122                 case Opt_dmask:
1123                         if (match_octal(&args[0], &option))
1124                                 return -EINVAL;
1125                         opts->fs_dmask = option;
1126                         break;
1127                 case Opt_fmask:
1128                         if (match_octal(&args[0], &option))
1129                                 return -EINVAL;
1130                         opts->fs_fmask = option;
1131                         break;
1132                 case Opt_allow_utime:
1133                         if (match_octal(&args[0], &option))
1134                                 return -EINVAL;
1135                         opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1136                         break;
1137                 case Opt_codepage:
1138                         if (match_int(&args[0], &option))
1139                                 return -EINVAL;
1140                         opts->codepage = option;
1141                         break;
1142                 case Opt_flush:
1143                         opts->flush = 1;
1144                         break;
1145                 case Opt_time_offset:
1146                         if (match_int(&args[0], &option))
1147                                 return -EINVAL;
1148                         if (option < -12 * 60 || option > 12 * 60)
1149                                 return -EINVAL;
1150                         opts->tz_set = 1;
1151                         opts->time_offset = option;
1152                         break;
1153                 case Opt_tz_utc:
1154                         opts->tz_set = 1;
1155                         opts->time_offset = 0;
1156                         break;
1157                 case Opt_err_cont:
1158                         opts->errors = FAT_ERRORS_CONT;
1159                         break;
1160                 case Opt_err_panic:
1161                         opts->errors = FAT_ERRORS_PANIC;
1162                         break;
1163                 case Opt_err_ro:
1164                         opts->errors = FAT_ERRORS_RO;
1165                         break;
1166                 case Opt_nfs_stale_rw:
1167                         opts->nfs = FAT_NFS_STALE_RW;
1168                         break;
1169                 case Opt_nfs_nostale_ro:
1170                         opts->nfs = FAT_NFS_NOSTALE_RO;
1171                         break;
1172                 case Opt_dos1xfloppy:
1173                         opts->dos1xfloppy = 1;
1174                         break;
1175
1176                 /* msdos specific */
1177                 case Opt_dots:
1178                         opts->dotsOK = 1;
1179                         break;
1180                 case Opt_nodots:
1181                         opts->dotsOK = 0;
1182                         break;
1183
1184                 /* vfat specific */
1185                 case Opt_charset:
1186                         if (opts->iocharset != fat_default_iocharset)
1187                                 kfree(opts->iocharset);
1188                         iocharset = match_strdup(&args[0]);
1189                         if (!iocharset)
1190                                 return -ENOMEM;
1191                         opts->iocharset = iocharset;
1192                         break;
1193                 case Opt_shortname_lower:
1194                         opts->shortname = VFAT_SFN_DISPLAY_LOWER
1195                                         | VFAT_SFN_CREATE_WIN95;
1196                         break;
1197                 case Opt_shortname_win95:
1198                         opts->shortname = VFAT_SFN_DISPLAY_WIN95
1199                                         | VFAT_SFN_CREATE_WIN95;
1200                         break;
1201                 case Opt_shortname_winnt:
1202                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1203                                         | VFAT_SFN_CREATE_WINNT;
1204                         break;
1205                 case Opt_shortname_mixed:
1206                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1207                                         | VFAT_SFN_CREATE_WIN95;
1208                         break;
1209                 case Opt_utf8_no:               /* 0 or no or false */
1210                         opts->utf8 = 0;
1211                         break;
1212                 case Opt_utf8_yes:              /* empty or 1 or yes or true */
1213                         opts->utf8 = 1;
1214                         break;
1215                 case Opt_uni_xl_no:             /* 0 or no or false */
1216                         opts->unicode_xlate = 0;
1217                         break;
1218                 case Opt_uni_xl_yes:            /* empty or 1 or yes or true */
1219                         opts->unicode_xlate = 1;
1220                         break;
1221                 case Opt_nonumtail_no:          /* 0 or no or false */
1222                         opts->numtail = 1;      /* negated option */
1223                         break;
1224                 case Opt_nonumtail_yes:         /* empty or 1 or yes or true */
1225                         opts->numtail = 0;      /* negated option */
1226                         break;
1227                 case Opt_rodir:
1228                         opts->rodir = 1;
1229                         break;
1230                 case Opt_discard:
1231                         opts->discard = 1;
1232                         break;
1233
1234                 /* obsolete mount options */
1235                 case Opt_obsolete:
1236                         fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1237                                "not supported now", p);
1238                         break;
1239                 /* unknown option */
1240                 default:
1241                         if (!silent) {
1242                                 fat_msg(sb, KERN_ERR,
1243                                        "Unrecognized mount option \"%s\" "
1244                                        "or missing value", p);
1245                         }
1246                         return -EINVAL;
1247                 }
1248         }
1249
1250 out:
1251         /* UTF-8 doesn't provide FAT semantics */
1252         if (!strcmp(opts->iocharset, "utf8")) {
1253                 fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1254                        " for FAT filesystems, filesystem will be "
1255                        "case sensitive!");
1256         }
1257
1258         /* If user doesn't specify allow_utime, it's initialized from dmask. */
1259         if (opts->allow_utime == (unsigned short)-1)
1260                 opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1261         if (opts->unicode_xlate)
1262                 opts->utf8 = 0;
1263         if (opts->nfs == FAT_NFS_NOSTALE_RO) {
1264                 sb->s_flags |= MS_RDONLY;
1265                 sb->s_export_op = &fat_export_ops_nostale;
1266         }
1267
1268         return 0;
1269 }
1270
1271 static int fat_read_root(struct inode *inode)
1272 {
1273         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
1274         int error;
1275
1276         MSDOS_I(inode)->i_pos = MSDOS_ROOT_INO;
1277         inode->i_uid = sbi->options.fs_uid;
1278         inode->i_gid = sbi->options.fs_gid;
1279         inode->i_version++;
1280         inode->i_generation = 0;
1281         inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1282         inode->i_op = sbi->dir_ops;
1283         inode->i_fop = &fat_dir_operations;
1284         if (sbi->fat_bits == 32) {
1285                 MSDOS_I(inode)->i_start = sbi->root_cluster;
1286                 error = fat_calc_dir_size(inode);
1287                 if (error < 0)
1288                         return error;
1289         } else {
1290                 MSDOS_I(inode)->i_start = 0;
1291                 inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1292         }
1293         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1294                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1295         MSDOS_I(inode)->i_logstart = 0;
1296         MSDOS_I(inode)->mmu_private = inode->i_size;
1297
1298         fat_save_attrs(inode, ATTR_DIR);
1299         inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1300         inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1301         set_nlink(inode, fat_subdirs(inode)+2);
1302
1303         return 0;
1304 }
1305
1306 static unsigned long calc_fat_clusters(struct super_block *sb)
1307 {
1308         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1309
1310         /* Divide first to avoid overflow */
1311         if (sbi->fat_bits != 12) {
1312                 unsigned long ent_per_sec = sb->s_blocksize * 8 / sbi->fat_bits;
1313                 return ent_per_sec * sbi->fat_length;
1314         }
1315
1316         return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1317 }
1318
1319 static bool fat_bpb_is_zero(struct fat_boot_sector *b)
1320 {
1321         if (get_unaligned_le16(&b->sector_size))
1322                 return false;
1323         if (b->sec_per_clus)
1324                 return false;
1325         if (b->reserved)
1326                 return false;
1327         if (b->fats)
1328                 return false;
1329         if (get_unaligned_le16(&b->dir_entries))
1330                 return false;
1331         if (get_unaligned_le16(&b->sectors))
1332                 return false;
1333         if (b->media)
1334                 return false;
1335         if (b->fat_length)
1336                 return false;
1337         if (b->secs_track)
1338                 return false;
1339         if (b->heads)
1340                 return false;
1341         return true;
1342 }
1343
1344 static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
1345         int silent, struct fat_bios_param_block *bpb)
1346 {
1347         int error = -EINVAL;
1348
1349         /* Read in BPB ... */
1350         memset(bpb, 0, sizeof(*bpb));
1351         bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
1352         bpb->fat_sec_per_clus = b->sec_per_clus;
1353         bpb->fat_reserved = le16_to_cpu(b->reserved);
1354         bpb->fat_fats = b->fats;
1355         bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
1356         bpb->fat_sectors = get_unaligned_le16(&b->sectors);
1357         bpb->fat_fat_length = le16_to_cpu(b->fat_length);
1358         bpb->fat_total_sect = le32_to_cpu(b->total_sect);
1359
1360         bpb->fat16_state = b->fat16.state;
1361         bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
1362
1363         bpb->fat32_length = le32_to_cpu(b->fat32.length);
1364         bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
1365         bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
1366         bpb->fat32_state = b->fat32.state;
1367         bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
1368
1369         /* Validate this looks like a FAT filesystem BPB */
1370         if (!bpb->fat_reserved) {
1371                 if (!silent)
1372                         fat_msg(sb, KERN_ERR,
1373                                 "bogus number of reserved sectors");
1374                 goto out;
1375         }
1376         if (!bpb->fat_fats) {
1377                 if (!silent)
1378                         fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1379                 goto out;
1380         }
1381
1382         /*
1383          * Earlier we checked here that b->secs_track and b->head are nonzero,
1384          * but it turns out valid FAT filesystems can have zero there.
1385          */
1386
1387         if (!fat_valid_media(b->media)) {
1388                 if (!silent)
1389                         fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1390                                 (unsigned)b->media);
1391                 goto out;
1392         }
1393
1394         if (!is_power_of_2(bpb->fat_sector_size)
1395             || (bpb->fat_sector_size < 512)
1396             || (bpb->fat_sector_size > 4096)) {
1397                 if (!silent)
1398                         fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1399                                (unsigned)bpb->fat_sector_size);
1400                 goto out;
1401         }
1402
1403         if (!is_power_of_2(bpb->fat_sec_per_clus)) {
1404                 if (!silent)
1405                         fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1406                                 (unsigned)bpb->fat_sec_per_clus);
1407                 goto out;
1408         }
1409
1410         error = 0;
1411
1412 out:
1413         return error;
1414 }
1415
1416 static int fat_read_static_bpb(struct super_block *sb,
1417         struct fat_boot_sector *b, int silent,
1418         struct fat_bios_param_block *bpb)
1419 {
1420         static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
1421
1422         struct fat_floppy_defaults *fdefaults = NULL;
1423         int error = -EINVAL;
1424         sector_t bd_sects;
1425         unsigned i;
1426
1427         bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
1428
1429         /* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
1430         if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
1431                 if (!silent)
1432                         fat_msg(sb, KERN_ERR,
1433                                 "%s; no bootstrapping code", notdos1x);
1434                 goto out;
1435         }
1436
1437         /*
1438          * If any value in this region is non-zero, it isn't archaic
1439          * DOS.
1440          */
1441         if (!fat_bpb_is_zero(b)) {
1442                 if (!silent)
1443                         fat_msg(sb, KERN_ERR,
1444                                 "%s; DOS 2.x BPB is non-zero", notdos1x);
1445                 goto out;
1446         }
1447
1448         for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
1449                 if (floppy_defaults[i].nr_sectors == bd_sects) {
1450                         fdefaults = &floppy_defaults[i];
1451                         break;
1452                 }
1453         }
1454
1455         if (fdefaults == NULL) {
1456                 if (!silent)
1457                         fat_msg(sb, KERN_WARNING,
1458                                 "This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
1459                                 (u64)bd_sects);
1460                 goto out;
1461         }
1462
1463         if (!silent)
1464                 fat_msg(sb, KERN_INFO,
1465                         "This looks like a DOS 1.x volume; assuming default BPB values");
1466
1467         memset(bpb, 0, sizeof(*bpb));
1468         bpb->fat_sector_size = SECTOR_SIZE;
1469         bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
1470         bpb->fat_reserved = 1;
1471         bpb->fat_fats = 2;
1472         bpb->fat_dir_entries = fdefaults->dir_entries;
1473         bpb->fat_sectors = fdefaults->nr_sectors;
1474         bpb->fat_fat_length = fdefaults->fat_length;
1475
1476         error = 0;
1477
1478 out:
1479         return error;
1480 }
1481
1482 /*
1483  * Read the super block of an MS-DOS FS.
1484  */
1485 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1486                    void (*setup)(struct super_block *))
1487 {
1488         struct inode *root_inode = NULL, *fat_inode = NULL;
1489         struct inode *fsinfo_inode = NULL;
1490         struct buffer_head *bh;
1491         struct fat_bios_param_block bpb;
1492         struct msdos_sb_info *sbi;
1493         u16 logical_sector_size;
1494         u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1495         int debug;
1496         long error;
1497         char buf[50];
1498
1499         /*
1500          * GFP_KERNEL is ok here, because while we do hold the
1501          * supeblock lock, memory pressure can't call back into
1502          * the filesystem, since we're only just about to mount
1503          * it and have no inodes etc active!
1504          */
1505         sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1506         if (!sbi)
1507                 return -ENOMEM;
1508         sb->s_fs_info = sbi;
1509
1510         sb->s_flags |= MS_NODIRATIME;
1511         sb->s_magic = MSDOS_SUPER_MAGIC;
1512         sb->s_op = &fat_sops;
1513         sb->s_export_op = &fat_export_ops;
1514         mutex_init(&sbi->nfs_build_inode_lock);
1515         ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1516                              DEFAULT_RATELIMIT_BURST);
1517
1518         error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1519         if (error)
1520                 goto out_fail;
1521
1522         setup(sb); /* flavour-specific stuff that needs options */
1523
1524         error = -EIO;
1525         sb_min_blocksize(sb, 512);
1526         bh = sb_bread(sb, 0);
1527         if (bh == NULL) {
1528                 fat_msg(sb, KERN_ERR, "unable to read boot sector");
1529                 goto out_fail;
1530         }
1531
1532         error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
1533                 &bpb);
1534         if (error == -EINVAL && sbi->options.dos1xfloppy)
1535                 error = fat_read_static_bpb(sb,
1536                         (struct fat_boot_sector *)bh->b_data, silent, &bpb);
1537         brelse(bh);
1538
1539         if (error == -EINVAL)
1540                 goto out_invalid;
1541         else if (error)
1542                 goto out_fail;
1543
1544         logical_sector_size = bpb.fat_sector_size;
1545         sbi->sec_per_clus = bpb.fat_sec_per_clus;
1546
1547         error = -EIO;
1548         if (logical_sector_size < sb->s_blocksize) {
1549                 fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1550                        " (logical sector size = %u)", logical_sector_size);
1551                 goto out_fail;
1552         }
1553
1554         if (logical_sector_size > sb->s_blocksize) {
1555                 struct buffer_head *bh_resize;
1556
1557                 if (!sb_set_blocksize(sb, logical_sector_size)) {
1558                         fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1559                                logical_sector_size);
1560                         goto out_fail;
1561                 }
1562
1563                 /* Verify that the larger boot sector is fully readable */
1564                 bh_resize = sb_bread(sb, 0);
1565                 if (bh_resize == NULL) {
1566                         fat_msg(sb, KERN_ERR, "unable to read boot sector"
1567                                " (logical sector size = %lu)",
1568                                sb->s_blocksize);
1569                         goto out_fail;
1570                 }
1571                 brelse(bh_resize);
1572         }
1573
1574         mutex_init(&sbi->s_lock);
1575         sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1576         sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1577         sbi->fats = bpb.fat_fats;
1578         sbi->fat_bits = 0;              /* Don't know yet */
1579         sbi->fat_start = bpb.fat_reserved;
1580         sbi->fat_length = bpb.fat_fat_length;
1581         sbi->root_cluster = 0;
1582         sbi->free_clusters = -1;        /* Don't know yet */
1583         sbi->free_clus_valid = 0;
1584         sbi->prev_free = FAT_START_ENT;
1585         sb->s_maxbytes = 0xffffffff;
1586
1587         if (!sbi->fat_length && bpb.fat32_length) {
1588                 struct fat_boot_fsinfo *fsinfo;
1589                 struct buffer_head *fsinfo_bh;
1590
1591                 /* Must be FAT32 */
1592                 sbi->fat_bits = 32;
1593                 sbi->fat_length = bpb.fat32_length;
1594                 sbi->root_cluster = bpb.fat32_root_cluster;
1595
1596                 /* MC - if info_sector is 0, don't multiply by 0 */
1597                 sbi->fsinfo_sector = bpb.fat32_info_sector;
1598                 if (sbi->fsinfo_sector == 0)
1599                         sbi->fsinfo_sector = 1;
1600
1601                 fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1602                 if (fsinfo_bh == NULL) {
1603                         fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1604                                " (sector = %lu)", sbi->fsinfo_sector);
1605                         goto out_fail;
1606                 }
1607
1608                 fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1609                 if (!IS_FSINFO(fsinfo)) {
1610                         fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1611                                "0x%08x, 0x%08x (sector = %lu)",
1612                                le32_to_cpu(fsinfo->signature1),
1613                                le32_to_cpu(fsinfo->signature2),
1614                                sbi->fsinfo_sector);
1615                 } else {
1616                         if (sbi->options.usefree)
1617                                 sbi->free_clus_valid = 1;
1618                         sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1619                         sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1620                 }
1621
1622                 brelse(fsinfo_bh);
1623         }
1624
1625         /* interpret volume ID as a little endian 32 bit integer */
1626         if (sbi->fat_bits == 32)
1627                 sbi->vol_id = bpb.fat32_vol_id;
1628         else /* fat 16 or 12 */
1629                 sbi->vol_id = bpb.fat16_vol_id;
1630
1631         sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1632         sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1633
1634         sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1635         sbi->dir_entries = bpb.fat_dir_entries;
1636         if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1637                 if (!silent)
1638                         fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
1639                                " (%u)", sbi->dir_entries);
1640                 goto out_invalid;
1641         }
1642
1643         rootdir_sectors = sbi->dir_entries
1644                 * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1645         sbi->data_start = sbi->dir_start + rootdir_sectors;
1646         total_sectors = bpb.fat_sectors;
1647         if (total_sectors == 0)
1648                 total_sectors = bpb.fat_total_sect;
1649
1650         total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1651
1652         if (sbi->fat_bits != 32)
1653                 sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1654
1655         /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1656         if (sbi->fat_bits == 32)
1657                 sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
1658         else /* fat 16 or 12 */
1659                 sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
1660
1661         /* check that FAT table does not overflow */
1662         fat_clusters = calc_fat_clusters(sb);
1663         total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1664         if (total_clusters > MAX_FAT(sb)) {
1665                 if (!silent)
1666                         fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1667                                total_clusters);
1668                 goto out_invalid;
1669         }
1670
1671         sbi->max_cluster = total_clusters + FAT_START_ENT;
1672         /* check the free_clusters, it's not necessarily correct */
1673         if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1674                 sbi->free_clusters = -1;
1675         /* check the prev_free, it's not necessarily correct */
1676         sbi->prev_free %= sbi->max_cluster;
1677         if (sbi->prev_free < FAT_START_ENT)
1678                 sbi->prev_free = FAT_START_ENT;
1679
1680         /* set up enough so that it can read an inode */
1681         fat_hash_init(sb);
1682         dir_hash_init(sb);
1683         fat_ent_access_init(sb);
1684
1685         /*
1686          * The low byte of FAT's first entry must have same value with
1687          * media-field.  But in real world, too many devices is
1688          * writing wrong value.  So, removed that validity check.
1689          *
1690          * if (FAT_FIRST_ENT(sb, media) != first)
1691          */
1692
1693         error = -EINVAL;
1694         sprintf(buf, "cp%d", sbi->options.codepage);
1695         sbi->nls_disk = load_nls(buf);
1696         if (!sbi->nls_disk) {
1697                 fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1698                 goto out_fail;
1699         }
1700
1701         /* FIXME: utf8 is using iocharset for upper/lower conversion */
1702         if (sbi->options.isvfat) {
1703                 sbi->nls_io = load_nls(sbi->options.iocharset);
1704                 if (!sbi->nls_io) {
1705                         fat_msg(sb, KERN_ERR, "IO charset %s not found",
1706                                sbi->options.iocharset);
1707                         goto out_fail;
1708                 }
1709         }
1710
1711         error = -ENOMEM;
1712         fat_inode = new_inode(sb);
1713         if (!fat_inode)
1714                 goto out_fail;
1715         MSDOS_I(fat_inode)->i_pos = 0;
1716         sbi->fat_inode = fat_inode;
1717
1718         fsinfo_inode = new_inode(sb);
1719         if (!fsinfo_inode)
1720                 goto out_fail;
1721         fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1722         sbi->fsinfo_inode = fsinfo_inode;
1723         insert_inode_hash(fsinfo_inode);
1724
1725         root_inode = new_inode(sb);
1726         if (!root_inode)
1727                 goto out_fail;
1728         root_inode->i_ino = MSDOS_ROOT_INO;
1729         root_inode->i_version = 1;
1730         error = fat_read_root(root_inode);
1731         if (error < 0) {
1732                 iput(root_inode);
1733                 goto out_fail;
1734         }
1735         error = -ENOMEM;
1736         insert_inode_hash(root_inode);
1737         fat_attach(root_inode, 0);
1738         sb->s_root = d_make_root(root_inode);
1739         if (!sb->s_root) {
1740                 fat_msg(sb, KERN_ERR, "get root inode failed");
1741                 goto out_fail;
1742         }
1743
1744         if (sbi->options.discard) {
1745                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1746                 if (!blk_queue_discard(q))
1747                         fat_msg(sb, KERN_WARNING,
1748                                         "mounting with \"discard\" option, but "
1749                                         "the device does not support discard");
1750         }
1751
1752         fat_set_state(sb, 1, 0);
1753         return 0;
1754
1755 out_invalid:
1756         error = -EINVAL;
1757         if (!silent)
1758                 fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1759
1760 out_fail:
1761         if (fsinfo_inode)
1762                 iput(fsinfo_inode);
1763         if (fat_inode)
1764                 iput(fat_inode);
1765         unload_nls(sbi->nls_io);
1766         unload_nls(sbi->nls_disk);
1767         if (sbi->options.iocharset != fat_default_iocharset)
1768                 kfree(sbi->options.iocharset);
1769         sb->s_fs_info = NULL;
1770         kfree(sbi);
1771         return error;
1772 }
1773
1774 EXPORT_SYMBOL_GPL(fat_fill_super);
1775
1776 /*
1777  * helper function for fat_flush_inodes.  This writes both the inode
1778  * and the file data blocks, waiting for in flight data blocks before
1779  * the start of the call.  It does not wait for any io started
1780  * during the call
1781  */
1782 static int writeback_inode(struct inode *inode)
1783 {
1784
1785         int ret;
1786
1787         /* if we used wait=1, sync_inode_metadata waits for the io for the
1788         * inode to finish.  So wait=0 is sent down to sync_inode_metadata
1789         * and filemap_fdatawrite is used for the data blocks
1790         */
1791         ret = sync_inode_metadata(inode, 0);
1792         if (!ret)
1793                 ret = filemap_fdatawrite(inode->i_mapping);
1794         return ret;
1795 }
1796
1797 /*
1798  * write data and metadata corresponding to i1 and i2.  The io is
1799  * started but we do not wait for any of it to finish.
1800  *
1801  * filemap_flush is used for the block device, so if there is a dirty
1802  * page for a block already in flight, we will not wait and start the
1803  * io over again
1804  */
1805 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1806 {
1807         int ret = 0;
1808         if (!MSDOS_SB(sb)->options.flush)
1809                 return 0;
1810         if (i1)
1811                 ret = writeback_inode(i1);
1812         if (!ret && i2)
1813                 ret = writeback_inode(i2);
1814         if (!ret) {
1815                 struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1816                 ret = filemap_flush(mapping);
1817         }
1818         return ret;
1819 }
1820 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1821
1822 static int __init init_fat_fs(void)
1823 {
1824         int err;
1825
1826         err = fat_cache_init();
1827         if (err)
1828                 return err;
1829
1830         err = fat_init_inodecache();
1831         if (err)
1832                 goto failed;
1833
1834         return 0;
1835
1836 failed:
1837         fat_cache_destroy();
1838         return err;
1839 }
1840
1841 static void __exit exit_fat_fs(void)
1842 {
1843         fat_cache_destroy();
1844         fat_destroy_inodecache();
1845 }
1846
1847 module_init(init_fat_fs)
1848 module_exit(exit_fat_fs)
1849
1850 MODULE_LICENSE("GPL");