]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - fs/fat/inode.c
Merge tag 'v3.16-rc1' into i2c/for-next
[karo-tx-linux.git] / 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/init.h>
15 #include <linux/time.h>
16 #include <linux/slab.h>
17 #include <linux/seq_file.h>
18 #include <linux/pagemap.h>
19 #include <linux/mpage.h>
20 #include <linux/buffer_head.h>
21 #include <linux/mount.h>
22 #include <linux/aio.h>
23 #include <linux/vfs.h>
24 #include <linux/parser.h>
25 #include <linux/uio.h>
26 #include <linux/writeback.h>
27 #include <linux/log2.h>
28 #include <linux/hash.h>
29 #include <linux/blkdev.h>
30 #include <asm/unaligned.h>
31 #include "fat.h"
32
33 #ifndef CONFIG_FAT_DEFAULT_IOCHARSET
34 /* if user don't select VFAT, this is undefined. */
35 #define CONFIG_FAT_DEFAULT_IOCHARSET    ""
36 #endif
37
38 #define KB_IN_SECTORS 2
39
40 /*
41  * A deserialized copy of the on-disk structure laid out in struct
42  * fat_boot_sector.
43  */
44 struct fat_bios_param_block {
45         u16     fat_sector_size;
46         u8      fat_sec_per_clus;
47         u16     fat_reserved;
48         u8      fat_fats;
49         u16     fat_dir_entries;
50         u16     fat_sectors;
51         u16     fat_fat_length;
52         u32     fat_total_sect;
53
54         u8      fat16_state;
55         u32     fat16_vol_id;
56
57         u32     fat32_length;
58         u32     fat32_root_cluster;
59         u16     fat32_info_sector;
60         u8      fat32_state;
61         u32     fat32_vol_id;
62 };
63
64 static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
65 static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
66
67 static struct fat_floppy_defaults {
68         unsigned nr_sectors;
69         unsigned sec_per_clus;
70         unsigned dir_entries;
71         unsigned media;
72         unsigned fat_length;
73 } floppy_defaults[] = {
74 {
75         .nr_sectors = 160 * KB_IN_SECTORS,
76         .sec_per_clus = 1,
77         .dir_entries = 64,
78         .media = 0xFE,
79         .fat_length = 1,
80 },
81 {
82         .nr_sectors = 180 * KB_IN_SECTORS,
83         .sec_per_clus = 1,
84         .dir_entries = 64,
85         .media = 0xFC,
86         .fat_length = 2,
87 },
88 {
89         .nr_sectors = 320 * KB_IN_SECTORS,
90         .sec_per_clus = 2,
91         .dir_entries = 112,
92         .media = 0xFF,
93         .fat_length = 1,
94 },
95 {
96         .nr_sectors = 360 * KB_IN_SECTORS,
97         .sec_per_clus = 2,
98         .dir_entries = 112,
99         .media = 0xFD,
100         .fat_length = 2,
101 },
102 };
103
104 static int fat_add_cluster(struct inode *inode)
105 {
106         int err, cluster;
107
108         err = fat_alloc_clusters(inode, &cluster, 1);
109         if (err)
110                 return err;
111         /* FIXME: this cluster should be added after data of this
112          * cluster is writed */
113         err = fat_chain_add(inode, cluster, 1);
114         if (err)
115                 fat_free_clusters(inode, cluster);
116         return err;
117 }
118
119 static inline int __fat_get_block(struct inode *inode, sector_t iblock,
120                                   unsigned long *max_blocks,
121                                   struct buffer_head *bh_result, int create)
122 {
123         struct super_block *sb = inode->i_sb;
124         struct msdos_sb_info *sbi = MSDOS_SB(sb);
125         unsigned long mapped_blocks;
126         sector_t phys;
127         int err, offset;
128
129         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
130         if (err)
131                 return err;
132         if (phys) {
133                 map_bh(bh_result, sb, phys);
134                 *max_blocks = min(mapped_blocks, *max_blocks);
135                 return 0;
136         }
137         if (!create)
138                 return 0;
139
140         if (iblock != MSDOS_I(inode)->mmu_private >> sb->s_blocksize_bits) {
141                 fat_fs_error(sb, "corrupted file size (i_pos %lld, %lld)",
142                         MSDOS_I(inode)->i_pos, MSDOS_I(inode)->mmu_private);
143                 return -EIO;
144         }
145
146         offset = (unsigned long)iblock & (sbi->sec_per_clus - 1);
147         if (!offset) {
148                 /* TODO: multiple cluster allocation would be desirable. */
149                 err = fat_add_cluster(inode);
150                 if (err)
151                         return err;
152         }
153         /* available blocks on this cluster */
154         mapped_blocks = sbi->sec_per_clus - offset;
155
156         *max_blocks = min(mapped_blocks, *max_blocks);
157         MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
158
159         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
160         if (err)
161                 return err;
162
163         BUG_ON(!phys);
164         BUG_ON(*max_blocks != mapped_blocks);
165         set_buffer_new(bh_result);
166         map_bh(bh_result, sb, phys);
167
168         return 0;
169 }
170
171 static int fat_get_block(struct inode *inode, sector_t iblock,
172                          struct buffer_head *bh_result, int create)
173 {
174         struct super_block *sb = inode->i_sb;
175         unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
176         int err;
177
178         err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
179         if (err)
180                 return err;
181         bh_result->b_size = max_blocks << sb->s_blocksize_bits;
182         return 0;
183 }
184
185 static int fat_writepage(struct page *page, struct writeback_control *wbc)
186 {
187         return block_write_full_page(page, fat_get_block, wbc);
188 }
189
190 static int fat_writepages(struct address_space *mapping,
191                           struct writeback_control *wbc)
192 {
193         return mpage_writepages(mapping, wbc, fat_get_block);
194 }
195
196 static int fat_readpage(struct file *file, struct page *page)
197 {
198         return mpage_readpage(page, fat_get_block);
199 }
200
201 static int fat_readpages(struct file *file, struct address_space *mapping,
202                          struct list_head *pages, unsigned nr_pages)
203 {
204         return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
205 }
206
207 static void fat_write_failed(struct address_space *mapping, loff_t to)
208 {
209         struct inode *inode = mapping->host;
210
211         if (to > inode->i_size) {
212                 truncate_pagecache(inode, inode->i_size);
213                 fat_truncate_blocks(inode, inode->i_size);
214         }
215 }
216
217 static int fat_write_begin(struct file *file, struct address_space *mapping,
218                         loff_t pos, unsigned len, unsigned flags,
219                         struct page **pagep, void **fsdata)
220 {
221         int err;
222
223         *pagep = NULL;
224         err = cont_write_begin(file, mapping, pos, len, flags,
225                                 pagep, fsdata, fat_get_block,
226                                 &MSDOS_I(mapping->host)->mmu_private);
227         if (err < 0)
228                 fat_write_failed(mapping, pos + len);
229         return err;
230 }
231
232 static int fat_write_end(struct file *file, struct address_space *mapping,
233                         loff_t pos, unsigned len, unsigned copied,
234                         struct page *pagep, void *fsdata)
235 {
236         struct inode *inode = mapping->host;
237         int err;
238         err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
239         if (err < len)
240                 fat_write_failed(mapping, pos + len);
241         if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
242                 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
243                 MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
244                 mark_inode_dirty(inode);
245         }
246         return err;
247 }
248
249 static ssize_t fat_direct_IO(int rw, struct kiocb *iocb,
250                              struct iov_iter *iter,
251                              loff_t offset)
252 {
253         struct file *file = iocb->ki_filp;
254         struct address_space *mapping = file->f_mapping;
255         struct inode *inode = mapping->host;
256         size_t count = iov_iter_count(iter);
257         ssize_t ret;
258
259         if (rw == WRITE) {
260                 /*
261                  * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
262                  * so we need to update the ->mmu_private to block boundary.
263                  *
264                  * But we must fill the remaining area or hole by nul for
265                  * updating ->mmu_private.
266                  *
267                  * Return 0, and fallback to normal buffered write.
268                  */
269                 loff_t size = offset + count;
270                 if (MSDOS_I(inode)->mmu_private < size)
271                         return 0;
272         }
273
274         /*
275          * FAT need to use the DIO_LOCKING for avoiding the race
276          * condition of fat_get_block() and ->truncate().
277          */
278         ret = blockdev_direct_IO(rw, iocb, inode, iter, offset, fat_get_block);
279         if (ret < 0 && (rw & WRITE))
280                 fat_write_failed(mapping, offset + count);
281
282         return ret;
283 }
284
285 static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
286 {
287         sector_t blocknr;
288
289         /* fat_get_cluster() assumes the requested blocknr isn't truncated. */
290         down_read(&MSDOS_I(mapping->host)->truncate_lock);
291         blocknr = generic_block_bmap(mapping, block, fat_get_block);
292         up_read(&MSDOS_I(mapping->host)->truncate_lock);
293
294         return blocknr;
295 }
296
297 static const struct address_space_operations fat_aops = {
298         .readpage       = fat_readpage,
299         .readpages      = fat_readpages,
300         .writepage      = fat_writepage,
301         .writepages     = fat_writepages,
302         .write_begin    = fat_write_begin,
303         .write_end      = fat_write_end,
304         .direct_IO      = fat_direct_IO,
305         .bmap           = _fat_bmap
306 };
307
308 /*
309  * New FAT inode stuff. We do the following:
310  *      a) i_ino is constant and has nothing with on-disk location.
311  *      b) FAT manages its own cache of directory entries.
312  *      c) *This* cache is indexed by on-disk location.
313  *      d) inode has an associated directory entry, all right, but
314  *              it may be unhashed.
315  *      e) currently entries are stored within struct inode. That should
316  *              change.
317  *      f) we deal with races in the following way:
318  *              1. readdir() and lookup() do FAT-dir-cache lookup.
319  *              2. rename() unhashes the F-d-c entry and rehashes it in
320  *                      a new place.
321  *              3. unlink() and rmdir() unhash F-d-c entry.
322  *              4. fat_write_inode() checks whether the thing is unhashed.
323  *                      If it is we silently return. If it isn't we do bread(),
324  *                      check if the location is still valid and retry if it
325  *                      isn't. Otherwise we do changes.
326  *              5. Spinlock is used to protect hash/unhash/location check/lookup
327  *              6. fat_evict_inode() unhashes the F-d-c entry.
328  *              7. lookup() and readdir() do igrab() if they find a F-d-c entry
329  *                      and consider negative result as cache miss.
330  */
331
332 static void fat_hash_init(struct super_block *sb)
333 {
334         struct msdos_sb_info *sbi = MSDOS_SB(sb);
335         int i;
336
337         spin_lock_init(&sbi->inode_hash_lock);
338         for (i = 0; i < FAT_HASH_SIZE; i++)
339                 INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
340 }
341
342 static inline unsigned long fat_hash(loff_t i_pos)
343 {
344         return hash_32(i_pos, FAT_HASH_BITS);
345 }
346
347 static void dir_hash_init(struct super_block *sb)
348 {
349         struct msdos_sb_info *sbi = MSDOS_SB(sb);
350         int i;
351
352         spin_lock_init(&sbi->dir_hash_lock);
353         for (i = 0; i < FAT_HASH_SIZE; i++)
354                 INIT_HLIST_HEAD(&sbi->dir_hashtable[i]);
355 }
356
357 void fat_attach(struct inode *inode, loff_t i_pos)
358 {
359         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
360
361         if (inode->i_ino != MSDOS_ROOT_INO) {
362                 struct hlist_head *head =   sbi->inode_hashtable
363                                           + fat_hash(i_pos);
364
365                 spin_lock(&sbi->inode_hash_lock);
366                 MSDOS_I(inode)->i_pos = i_pos;
367                 hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
368                 spin_unlock(&sbi->inode_hash_lock);
369         }
370
371         /* If NFS support is enabled, cache the mapping of start cluster
372          * to directory inode. This is used during reconnection of
373          * dentries to the filesystem root.
374          */
375         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
376                 struct hlist_head *d_head = sbi->dir_hashtable;
377                 d_head += fat_dir_hash(MSDOS_I(inode)->i_logstart);
378
379                 spin_lock(&sbi->dir_hash_lock);
380                 hlist_add_head(&MSDOS_I(inode)->i_dir_hash, d_head);
381                 spin_unlock(&sbi->dir_hash_lock);
382         }
383 }
384 EXPORT_SYMBOL_GPL(fat_attach);
385
386 void fat_detach(struct inode *inode)
387 {
388         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
389         spin_lock(&sbi->inode_hash_lock);
390         MSDOS_I(inode)->i_pos = 0;
391         hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
392         spin_unlock(&sbi->inode_hash_lock);
393
394         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
395                 spin_lock(&sbi->dir_hash_lock);
396                 hlist_del_init(&MSDOS_I(inode)->i_dir_hash);
397                 spin_unlock(&sbi->dir_hash_lock);
398         }
399 }
400 EXPORT_SYMBOL_GPL(fat_detach);
401
402 struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
403 {
404         struct msdos_sb_info *sbi = MSDOS_SB(sb);
405         struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
406         struct msdos_inode_info *i;
407         struct inode *inode = NULL;
408
409         spin_lock(&sbi->inode_hash_lock);
410         hlist_for_each_entry(i, head, i_fat_hash) {
411                 BUG_ON(i->vfs_inode.i_sb != sb);
412                 if (i->i_pos != i_pos)
413                         continue;
414                 inode = igrab(&i->vfs_inode);
415                 if (inode)
416                         break;
417         }
418         spin_unlock(&sbi->inode_hash_lock);
419         return inode;
420 }
421
422 static int is_exec(unsigned char *extension)
423 {
424         unsigned char exe_extensions[] = "EXECOMBAT", *walk;
425
426         for (walk = exe_extensions; *walk; walk += 3)
427                 if (!strncmp(extension, walk, 3))
428                         return 1;
429         return 0;
430 }
431
432 static int fat_calc_dir_size(struct inode *inode)
433 {
434         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
435         int ret, fclus, dclus;
436
437         inode->i_size = 0;
438         if (MSDOS_I(inode)->i_start == 0)
439                 return 0;
440
441         ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
442         if (ret < 0)
443                 return ret;
444         inode->i_size = (fclus + 1) << sbi->cluster_bits;
445
446         return 0;
447 }
448
449 /* doesn't deal with root inode */
450 int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
451 {
452         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
453         int error;
454
455         MSDOS_I(inode)->i_pos = 0;
456         inode->i_uid = sbi->options.fs_uid;
457         inode->i_gid = sbi->options.fs_gid;
458         inode->i_version++;
459         inode->i_generation = get_seconds();
460
461         if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
462                 inode->i_generation &= ~1;
463                 inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
464                 inode->i_op = sbi->dir_ops;
465                 inode->i_fop = &fat_dir_operations;
466
467                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
468                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
469                 error = fat_calc_dir_size(inode);
470                 if (error < 0)
471                         return error;
472                 MSDOS_I(inode)->mmu_private = inode->i_size;
473
474                 set_nlink(inode, fat_subdirs(inode));
475         } else { /* not a directory */
476                 inode->i_generation |= 1;
477                 inode->i_mode = fat_make_mode(sbi, de->attr,
478                         ((sbi->options.showexec && !is_exec(de->name + 8))
479                          ? S_IRUGO|S_IWUGO : S_IRWXUGO));
480                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
481
482                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
483                 inode->i_size = le32_to_cpu(de->size);
484                 inode->i_op = &fat_file_inode_operations;
485                 inode->i_fop = &fat_file_operations;
486                 inode->i_mapping->a_ops = &fat_aops;
487                 MSDOS_I(inode)->mmu_private = inode->i_size;
488         }
489         if (de->attr & ATTR_SYS) {
490                 if (sbi->options.sys_immutable)
491                         inode->i_flags |= S_IMMUTABLE;
492         }
493         fat_save_attrs(inode, de->attr);
494
495         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
496                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
497
498         fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
499         if (sbi->options.isvfat) {
500                 fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
501                                   de->cdate, de->ctime_cs);
502                 fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
503         } else
504                 inode->i_ctime = inode->i_atime = inode->i_mtime;
505
506         return 0;
507 }
508
509 static inline void fat_lock_build_inode(struct msdos_sb_info *sbi)
510 {
511         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
512                 mutex_lock(&sbi->nfs_build_inode_lock);
513 }
514
515 static inline void fat_unlock_build_inode(struct msdos_sb_info *sbi)
516 {
517         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
518                 mutex_unlock(&sbi->nfs_build_inode_lock);
519 }
520
521 struct inode *fat_build_inode(struct super_block *sb,
522                         struct msdos_dir_entry *de, loff_t i_pos)
523 {
524         struct inode *inode;
525         int err;
526
527         fat_lock_build_inode(MSDOS_SB(sb));
528         inode = fat_iget(sb, i_pos);
529         if (inode)
530                 goto out;
531         inode = new_inode(sb);
532         if (!inode) {
533                 inode = ERR_PTR(-ENOMEM);
534                 goto out;
535         }
536         inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
537         inode->i_version = 1;
538         err = fat_fill_inode(inode, de);
539         if (err) {
540                 iput(inode);
541                 inode = ERR_PTR(err);
542                 goto out;
543         }
544         fat_attach(inode, i_pos);
545         insert_inode_hash(inode);
546 out:
547         fat_unlock_build_inode(MSDOS_SB(sb));
548         return inode;
549 }
550
551 EXPORT_SYMBOL_GPL(fat_build_inode);
552
553 static void fat_evict_inode(struct inode *inode)
554 {
555         truncate_inode_pages_final(&inode->i_data);
556         if (!inode->i_nlink) {
557                 inode->i_size = 0;
558                 fat_truncate_blocks(inode, 0);
559         }
560         invalidate_inode_buffers(inode);
561         clear_inode(inode);
562         fat_cache_inval_inode(inode);
563         fat_detach(inode);
564 }
565
566 static void fat_set_state(struct super_block *sb,
567                         unsigned int set, unsigned int force)
568 {
569         struct buffer_head *bh;
570         struct fat_boot_sector *b;
571         struct msdos_sb_info *sbi = sb->s_fs_info;
572
573         /* do not change any thing if mounted read only */
574         if ((sb->s_flags & MS_RDONLY) && !force)
575                 return;
576
577         /* do not change state if fs was dirty */
578         if (sbi->dirty) {
579                 /* warn only on set (mount). */
580                 if (set)
581                         fat_msg(sb, KERN_WARNING, "Volume was not properly "
582                                 "unmounted. Some data may be corrupt. "
583                                 "Please run fsck.");
584                 return;
585         }
586
587         bh = sb_bread(sb, 0);
588         if (bh == NULL) {
589                 fat_msg(sb, KERN_ERR, "unable to read boot sector "
590                         "to mark fs as dirty");
591                 return;
592         }
593
594         b = (struct fat_boot_sector *) bh->b_data;
595
596         if (sbi->fat_bits == 32) {
597                 if (set)
598                         b->fat32.state |= FAT_STATE_DIRTY;
599                 else
600                         b->fat32.state &= ~FAT_STATE_DIRTY;
601         } else /* fat 16 and 12 */ {
602                 if (set)
603                         b->fat16.state |= FAT_STATE_DIRTY;
604                 else
605                         b->fat16.state &= ~FAT_STATE_DIRTY;
606         }
607
608         mark_buffer_dirty(bh);
609         sync_dirty_buffer(bh);
610         brelse(bh);
611 }
612
613 static void delayed_free(struct rcu_head *p)
614 {
615         struct msdos_sb_info *sbi = container_of(p, struct msdos_sb_info, rcu);
616         unload_nls(sbi->nls_disk);
617         unload_nls(sbi->nls_io);
618         if (sbi->options.iocharset != fat_default_iocharset)
619                 kfree(sbi->options.iocharset);
620         kfree(sbi);
621 }
622
623 static void fat_put_super(struct super_block *sb)
624 {
625         struct msdos_sb_info *sbi = MSDOS_SB(sb);
626
627         fat_set_state(sb, 0, 0);
628
629         iput(sbi->fsinfo_inode);
630         iput(sbi->fat_inode);
631
632         call_rcu(&sbi->rcu, delayed_free);
633 }
634
635 static struct kmem_cache *fat_inode_cachep;
636
637 static struct inode *fat_alloc_inode(struct super_block *sb)
638 {
639         struct msdos_inode_info *ei;
640         ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
641         if (!ei)
642                 return NULL;
643
644         init_rwsem(&ei->truncate_lock);
645         return &ei->vfs_inode;
646 }
647
648 static void fat_i_callback(struct rcu_head *head)
649 {
650         struct inode *inode = container_of(head, struct inode, i_rcu);
651         kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
652 }
653
654 static void fat_destroy_inode(struct inode *inode)
655 {
656         call_rcu(&inode->i_rcu, fat_i_callback);
657 }
658
659 static void init_once(void *foo)
660 {
661         struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
662
663         spin_lock_init(&ei->cache_lru_lock);
664         ei->nr_caches = 0;
665         ei->cache_valid_id = FAT_CACHE_VALID + 1;
666         INIT_LIST_HEAD(&ei->cache_lru);
667         INIT_HLIST_NODE(&ei->i_fat_hash);
668         INIT_HLIST_NODE(&ei->i_dir_hash);
669         inode_init_once(&ei->vfs_inode);
670 }
671
672 static int __init fat_init_inodecache(void)
673 {
674         fat_inode_cachep = kmem_cache_create("fat_inode_cache",
675                                              sizeof(struct msdos_inode_info),
676                                              0, (SLAB_RECLAIM_ACCOUNT|
677                                                 SLAB_MEM_SPREAD),
678                                              init_once);
679         if (fat_inode_cachep == NULL)
680                 return -ENOMEM;
681         return 0;
682 }
683
684 static void __exit fat_destroy_inodecache(void)
685 {
686         /*
687          * Make sure all delayed rcu free inodes are flushed before we
688          * destroy cache.
689          */
690         rcu_barrier();
691         kmem_cache_destroy(fat_inode_cachep);
692 }
693
694 static int fat_remount(struct super_block *sb, int *flags, char *data)
695 {
696         int new_rdonly;
697         struct msdos_sb_info *sbi = MSDOS_SB(sb);
698         *flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
699
700         sync_filesystem(sb);
701
702         /* make sure we update state on remount. */
703         new_rdonly = *flags & MS_RDONLY;
704         if (new_rdonly != (sb->s_flags & MS_RDONLY)) {
705                 if (new_rdonly)
706                         fat_set_state(sb, 0, 0);
707                 else
708                         fat_set_state(sb, 1, 1);
709         }
710         return 0;
711 }
712
713 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
714 {
715         struct super_block *sb = dentry->d_sb;
716         struct msdos_sb_info *sbi = MSDOS_SB(sb);
717         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
718
719         /* If the count of free cluster is still unknown, counts it here. */
720         if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
721                 int err = fat_count_free_clusters(dentry->d_sb);
722                 if (err)
723                         return err;
724         }
725
726         buf->f_type = dentry->d_sb->s_magic;
727         buf->f_bsize = sbi->cluster_size;
728         buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
729         buf->f_bfree = sbi->free_clusters;
730         buf->f_bavail = sbi->free_clusters;
731         buf->f_fsid.val[0] = (u32)id;
732         buf->f_fsid.val[1] = (u32)(id >> 32);
733         buf->f_namelen =
734                 (sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
735
736         return 0;
737 }
738
739 static int __fat_write_inode(struct inode *inode, int wait)
740 {
741         struct super_block *sb = inode->i_sb;
742         struct msdos_sb_info *sbi = MSDOS_SB(sb);
743         struct buffer_head *bh;
744         struct msdos_dir_entry *raw_entry;
745         loff_t i_pos;
746         sector_t blocknr;
747         int err, offset;
748
749         if (inode->i_ino == MSDOS_ROOT_INO)
750                 return 0;
751
752 retry:
753         i_pos = fat_i_pos_read(sbi, inode);
754         if (!i_pos)
755                 return 0;
756
757         fat_get_blknr_offset(sbi, i_pos, &blocknr, &offset);
758         bh = sb_bread(sb, blocknr);
759         if (!bh) {
760                 fat_msg(sb, KERN_ERR, "unable to read inode block "
761                        "for updating (i_pos %lld)", i_pos);
762                 return -EIO;
763         }
764         spin_lock(&sbi->inode_hash_lock);
765         if (i_pos != MSDOS_I(inode)->i_pos) {
766                 spin_unlock(&sbi->inode_hash_lock);
767                 brelse(bh);
768                 goto retry;
769         }
770
771         raw_entry = &((struct msdos_dir_entry *) (bh->b_data))[offset];
772         if (S_ISDIR(inode->i_mode))
773                 raw_entry->size = 0;
774         else
775                 raw_entry->size = cpu_to_le32(inode->i_size);
776         raw_entry->attr = fat_make_attrs(inode);
777         fat_set_start(raw_entry, MSDOS_I(inode)->i_logstart);
778         fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
779                           &raw_entry->date, NULL);
780         if (sbi->options.isvfat) {
781                 __le16 atime;
782                 fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
783                                   &raw_entry->cdate, &raw_entry->ctime_cs);
784                 fat_time_unix2fat(sbi, &inode->i_atime, &atime,
785                                   &raw_entry->adate, NULL);
786         }
787         spin_unlock(&sbi->inode_hash_lock);
788         mark_buffer_dirty(bh);
789         err = 0;
790         if (wait)
791                 err = sync_dirty_buffer(bh);
792         brelse(bh);
793         return err;
794 }
795
796 static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
797 {
798         int err;
799
800         if (inode->i_ino == MSDOS_FSINFO_INO) {
801                 struct super_block *sb = inode->i_sb;
802
803                 mutex_lock(&MSDOS_SB(sb)->s_lock);
804                 err = fat_clusters_flush(sb);
805                 mutex_unlock(&MSDOS_SB(sb)->s_lock);
806         } else
807                 err = __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
808
809         return err;
810 }
811
812 int fat_sync_inode(struct inode *inode)
813 {
814         return __fat_write_inode(inode, 1);
815 }
816
817 EXPORT_SYMBOL_GPL(fat_sync_inode);
818
819 static int fat_show_options(struct seq_file *m, struct dentry *root);
820 static const struct super_operations fat_sops = {
821         .alloc_inode    = fat_alloc_inode,
822         .destroy_inode  = fat_destroy_inode,
823         .write_inode    = fat_write_inode,
824         .evict_inode    = fat_evict_inode,
825         .put_super      = fat_put_super,
826         .statfs         = fat_statfs,
827         .remount_fs     = fat_remount,
828
829         .show_options   = fat_show_options,
830 };
831
832 static int fat_show_options(struct seq_file *m, struct dentry *root)
833 {
834         struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
835         struct fat_mount_options *opts = &sbi->options;
836         int isvfat = opts->isvfat;
837
838         if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
839                 seq_printf(m, ",uid=%u",
840                                 from_kuid_munged(&init_user_ns, opts->fs_uid));
841         if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
842                 seq_printf(m, ",gid=%u",
843                                 from_kgid_munged(&init_user_ns, opts->fs_gid));
844         seq_printf(m, ",fmask=%04o", opts->fs_fmask);
845         seq_printf(m, ",dmask=%04o", opts->fs_dmask);
846         if (opts->allow_utime)
847                 seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
848         if (sbi->nls_disk)
849                 /* strip "cp" prefix from displayed option */
850                 seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
851         if (isvfat) {
852                 if (sbi->nls_io)
853                         seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
854
855                 switch (opts->shortname) {
856                 case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
857                         seq_puts(m, ",shortname=win95");
858                         break;
859                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
860                         seq_puts(m, ",shortname=winnt");
861                         break;
862                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
863                         seq_puts(m, ",shortname=mixed");
864                         break;
865                 case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
866                         seq_puts(m, ",shortname=lower");
867                         break;
868                 default:
869                         seq_puts(m, ",shortname=unknown");
870                         break;
871                 }
872         }
873         if (opts->name_check != 'n')
874                 seq_printf(m, ",check=%c", opts->name_check);
875         if (opts->usefree)
876                 seq_puts(m, ",usefree");
877         if (opts->quiet)
878                 seq_puts(m, ",quiet");
879         if (opts->showexec)
880                 seq_puts(m, ",showexec");
881         if (opts->sys_immutable)
882                 seq_puts(m, ",sys_immutable");
883         if (!isvfat) {
884                 if (opts->dotsOK)
885                         seq_puts(m, ",dotsOK=yes");
886                 if (opts->nocase)
887                         seq_puts(m, ",nocase");
888         } else {
889                 if (opts->utf8)
890                         seq_puts(m, ",utf8");
891                 if (opts->unicode_xlate)
892                         seq_puts(m, ",uni_xlate");
893                 if (!opts->numtail)
894                         seq_puts(m, ",nonumtail");
895                 if (opts->rodir)
896                         seq_puts(m, ",rodir");
897         }
898         if (opts->flush)
899                 seq_puts(m, ",flush");
900         if (opts->tz_set) {
901                 if (opts->time_offset)
902                         seq_printf(m, ",time_offset=%d", opts->time_offset);
903                 else
904                         seq_puts(m, ",tz=UTC");
905         }
906         if (opts->errors == FAT_ERRORS_CONT)
907                 seq_puts(m, ",errors=continue");
908         else if (opts->errors == FAT_ERRORS_PANIC)
909                 seq_puts(m, ",errors=panic");
910         else
911                 seq_puts(m, ",errors=remount-ro");
912         if (opts->nfs == FAT_NFS_NOSTALE_RO)
913                 seq_puts(m, ",nfs=nostale_ro");
914         else if (opts->nfs)
915                 seq_puts(m, ",nfs=stale_rw");
916         if (opts->discard)
917                 seq_puts(m, ",discard");
918         if (opts->dos1xfloppy)
919                 seq_puts(m, ",dos1xfloppy");
920
921         return 0;
922 }
923
924 enum {
925         Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
926         Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
927         Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
928         Opt_immutable, Opt_dots, Opt_nodots,
929         Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
930         Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
931         Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
932         Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
933         Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
934         Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
935 };
936
937 static const match_table_t fat_tokens = {
938         {Opt_check_r, "check=relaxed"},
939         {Opt_check_s, "check=strict"},
940         {Opt_check_n, "check=normal"},
941         {Opt_check_r, "check=r"},
942         {Opt_check_s, "check=s"},
943         {Opt_check_n, "check=n"},
944         {Opt_uid, "uid=%u"},
945         {Opt_gid, "gid=%u"},
946         {Opt_umask, "umask=%o"},
947         {Opt_dmask, "dmask=%o"},
948         {Opt_fmask, "fmask=%o"},
949         {Opt_allow_utime, "allow_utime=%o"},
950         {Opt_codepage, "codepage=%u"},
951         {Opt_usefree, "usefree"},
952         {Opt_nocase, "nocase"},
953         {Opt_quiet, "quiet"},
954         {Opt_showexec, "showexec"},
955         {Opt_debug, "debug"},
956         {Opt_immutable, "sys_immutable"},
957         {Opt_flush, "flush"},
958         {Opt_tz_utc, "tz=UTC"},
959         {Opt_time_offset, "time_offset=%d"},
960         {Opt_err_cont, "errors=continue"},
961         {Opt_err_panic, "errors=panic"},
962         {Opt_err_ro, "errors=remount-ro"},
963         {Opt_discard, "discard"},
964         {Opt_nfs_stale_rw, "nfs"},
965         {Opt_nfs_stale_rw, "nfs=stale_rw"},
966         {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
967         {Opt_dos1xfloppy, "dos1xfloppy"},
968         {Opt_obsolete, "conv=binary"},
969         {Opt_obsolete, "conv=text"},
970         {Opt_obsolete, "conv=auto"},
971         {Opt_obsolete, "conv=b"},
972         {Opt_obsolete, "conv=t"},
973         {Opt_obsolete, "conv=a"},
974         {Opt_obsolete, "fat=%u"},
975         {Opt_obsolete, "blocksize=%u"},
976         {Opt_obsolete, "cvf_format=%20s"},
977         {Opt_obsolete, "cvf_options=%100s"},
978         {Opt_obsolete, "posix"},
979         {Opt_err, NULL},
980 };
981 static const match_table_t msdos_tokens = {
982         {Opt_nodots, "nodots"},
983         {Opt_nodots, "dotsOK=no"},
984         {Opt_dots, "dots"},
985         {Opt_dots, "dotsOK=yes"},
986         {Opt_err, NULL}
987 };
988 static const match_table_t vfat_tokens = {
989         {Opt_charset, "iocharset=%s"},
990         {Opt_shortname_lower, "shortname=lower"},
991         {Opt_shortname_win95, "shortname=win95"},
992         {Opt_shortname_winnt, "shortname=winnt"},
993         {Opt_shortname_mixed, "shortname=mixed"},
994         {Opt_utf8_no, "utf8=0"},                /* 0 or no or false */
995         {Opt_utf8_no, "utf8=no"},
996         {Opt_utf8_no, "utf8=false"},
997         {Opt_utf8_yes, "utf8=1"},               /* empty or 1 or yes or true */
998         {Opt_utf8_yes, "utf8=yes"},
999         {Opt_utf8_yes, "utf8=true"},
1000         {Opt_utf8_yes, "utf8"},
1001         {Opt_uni_xl_no, "uni_xlate=0"},         /* 0 or no or false */
1002         {Opt_uni_xl_no, "uni_xlate=no"},
1003         {Opt_uni_xl_no, "uni_xlate=false"},
1004         {Opt_uni_xl_yes, "uni_xlate=1"},        /* empty or 1 or yes or true */
1005         {Opt_uni_xl_yes, "uni_xlate=yes"},
1006         {Opt_uni_xl_yes, "uni_xlate=true"},
1007         {Opt_uni_xl_yes, "uni_xlate"},
1008         {Opt_nonumtail_no, "nonumtail=0"},      /* 0 or no or false */
1009         {Opt_nonumtail_no, "nonumtail=no"},
1010         {Opt_nonumtail_no, "nonumtail=false"},
1011         {Opt_nonumtail_yes, "nonumtail=1"},     /* empty or 1 or yes or true */
1012         {Opt_nonumtail_yes, "nonumtail=yes"},
1013         {Opt_nonumtail_yes, "nonumtail=true"},
1014         {Opt_nonumtail_yes, "nonumtail"},
1015         {Opt_rodir, "rodir"},
1016         {Opt_err, NULL}
1017 };
1018
1019 static int parse_options(struct super_block *sb, char *options, int is_vfat,
1020                          int silent, int *debug, struct fat_mount_options *opts)
1021 {
1022         char *p;
1023         substring_t args[MAX_OPT_ARGS];
1024         int option;
1025         char *iocharset;
1026
1027         opts->isvfat = is_vfat;
1028
1029         opts->fs_uid = current_uid();
1030         opts->fs_gid = current_gid();
1031         opts->fs_fmask = opts->fs_dmask = current_umask();
1032         opts->allow_utime = -1;
1033         opts->codepage = fat_default_codepage;
1034         opts->iocharset = fat_default_iocharset;
1035         if (is_vfat) {
1036                 opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1037                 opts->rodir = 0;
1038         } else {
1039                 opts->shortname = 0;
1040                 opts->rodir = 1;
1041         }
1042         opts->name_check = 'n';
1043         opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
1044         opts->utf8 = opts->unicode_xlate = 0;
1045         opts->numtail = 1;
1046         opts->usefree = opts->nocase = 0;
1047         opts->tz_set = 0;
1048         opts->nfs = 0;
1049         opts->errors = FAT_ERRORS_RO;
1050         *debug = 0;
1051
1052         if (!options)
1053                 goto out;
1054
1055         while ((p = strsep(&options, ",")) != NULL) {
1056                 int token;
1057                 if (!*p)
1058                         continue;
1059
1060                 token = match_token(p, fat_tokens, args);
1061                 if (token == Opt_err) {
1062                         if (is_vfat)
1063                                 token = match_token(p, vfat_tokens, args);
1064                         else
1065                                 token = match_token(p, msdos_tokens, args);
1066                 }
1067                 switch (token) {
1068                 case Opt_check_s:
1069                         opts->name_check = 's';
1070                         break;
1071                 case Opt_check_r:
1072                         opts->name_check = 'r';
1073                         break;
1074                 case Opt_check_n:
1075                         opts->name_check = 'n';
1076                         break;
1077                 case Opt_usefree:
1078                         opts->usefree = 1;
1079                         break;
1080                 case Opt_nocase:
1081                         if (!is_vfat)
1082                                 opts->nocase = 1;
1083                         else {
1084                                 /* for backward compatibility */
1085                                 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1086                                         | VFAT_SFN_CREATE_WIN95;
1087                         }
1088                         break;
1089                 case Opt_quiet:
1090                         opts->quiet = 1;
1091                         break;
1092                 case Opt_showexec:
1093                         opts->showexec = 1;
1094                         break;
1095                 case Opt_debug:
1096                         *debug = 1;
1097                         break;
1098                 case Opt_immutable:
1099                         opts->sys_immutable = 1;
1100                         break;
1101                 case Opt_uid:
1102                         if (match_int(&args[0], &option))
1103                                 return -EINVAL;
1104                         opts->fs_uid = make_kuid(current_user_ns(), option);
1105                         if (!uid_valid(opts->fs_uid))
1106                                 return -EINVAL;
1107                         break;
1108                 case Opt_gid:
1109                         if (match_int(&args[0], &option))
1110                                 return -EINVAL;
1111                         opts->fs_gid = make_kgid(current_user_ns(), option);
1112                         if (!gid_valid(opts->fs_gid))
1113                                 return -EINVAL;
1114                         break;
1115                 case Opt_umask:
1116                         if (match_octal(&args[0], &option))
1117                                 return -EINVAL;
1118                         opts->fs_fmask = opts->fs_dmask = option;
1119                         break;
1120                 case Opt_dmask:
1121                         if (match_octal(&args[0], &option))
1122                                 return -EINVAL;
1123                         opts->fs_dmask = option;
1124                         break;
1125                 case Opt_fmask:
1126                         if (match_octal(&args[0], &option))
1127                                 return -EINVAL;
1128                         opts->fs_fmask = option;
1129                         break;
1130                 case Opt_allow_utime:
1131                         if (match_octal(&args[0], &option))
1132                                 return -EINVAL;
1133                         opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1134                         break;
1135                 case Opt_codepage:
1136                         if (match_int(&args[0], &option))
1137                                 return -EINVAL;
1138                         opts->codepage = option;
1139                         break;
1140                 case Opt_flush:
1141                         opts->flush = 1;
1142                         break;
1143                 case Opt_time_offset:
1144                         if (match_int(&args[0], &option))
1145                                 return -EINVAL;
1146                         if (option < -12 * 60 || option > 12 * 60)
1147                                 return -EINVAL;
1148                         opts->tz_set = 1;
1149                         opts->time_offset = option;
1150                         break;
1151                 case Opt_tz_utc:
1152                         opts->tz_set = 1;
1153                         opts->time_offset = 0;
1154                         break;
1155                 case Opt_err_cont:
1156                         opts->errors = FAT_ERRORS_CONT;
1157                         break;
1158                 case Opt_err_panic:
1159                         opts->errors = FAT_ERRORS_PANIC;
1160                         break;
1161                 case Opt_err_ro:
1162                         opts->errors = FAT_ERRORS_RO;
1163                         break;
1164                 case Opt_nfs_stale_rw:
1165                         opts->nfs = FAT_NFS_STALE_RW;
1166                         break;
1167                 case Opt_nfs_nostale_ro:
1168                         opts->nfs = FAT_NFS_NOSTALE_RO;
1169                         break;
1170                 case Opt_dos1xfloppy:
1171                         opts->dos1xfloppy = 1;
1172                         break;
1173
1174                 /* msdos specific */
1175                 case Opt_dots:
1176                         opts->dotsOK = 1;
1177                         break;
1178                 case Opt_nodots:
1179                         opts->dotsOK = 0;
1180                         break;
1181
1182                 /* vfat specific */
1183                 case Opt_charset:
1184                         if (opts->iocharset != fat_default_iocharset)
1185                                 kfree(opts->iocharset);
1186                         iocharset = match_strdup(&args[0]);
1187                         if (!iocharset)
1188                                 return -ENOMEM;
1189                         opts->iocharset = iocharset;
1190                         break;
1191                 case Opt_shortname_lower:
1192                         opts->shortname = VFAT_SFN_DISPLAY_LOWER
1193                                         | VFAT_SFN_CREATE_WIN95;
1194                         break;
1195                 case Opt_shortname_win95:
1196                         opts->shortname = VFAT_SFN_DISPLAY_WIN95
1197                                         | VFAT_SFN_CREATE_WIN95;
1198                         break;
1199                 case Opt_shortname_winnt:
1200                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1201                                         | VFAT_SFN_CREATE_WINNT;
1202                         break;
1203                 case Opt_shortname_mixed:
1204                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1205                                         | VFAT_SFN_CREATE_WIN95;
1206                         break;
1207                 case Opt_utf8_no:               /* 0 or no or false */
1208                         opts->utf8 = 0;
1209                         break;
1210                 case Opt_utf8_yes:              /* empty or 1 or yes or true */
1211                         opts->utf8 = 1;
1212                         break;
1213                 case Opt_uni_xl_no:             /* 0 or no or false */
1214                         opts->unicode_xlate = 0;
1215                         break;
1216                 case Opt_uni_xl_yes:            /* empty or 1 or yes or true */
1217                         opts->unicode_xlate = 1;
1218                         break;
1219                 case Opt_nonumtail_no:          /* 0 or no or false */
1220                         opts->numtail = 1;      /* negated option */
1221                         break;
1222                 case Opt_nonumtail_yes:         /* empty or 1 or yes or true */
1223                         opts->numtail = 0;      /* negated option */
1224                         break;
1225                 case Opt_rodir:
1226                         opts->rodir = 1;
1227                         break;
1228                 case Opt_discard:
1229                         opts->discard = 1;
1230                         break;
1231
1232                 /* obsolete mount options */
1233                 case Opt_obsolete:
1234                         fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1235                                "not supported now", p);
1236                         break;
1237                 /* unknown option */
1238                 default:
1239                         if (!silent) {
1240                                 fat_msg(sb, KERN_ERR,
1241                                        "Unrecognized mount option \"%s\" "
1242                                        "or missing value", p);
1243                         }
1244                         return -EINVAL;
1245                 }
1246         }
1247
1248 out:
1249         /* UTF-8 doesn't provide FAT semantics */
1250         if (!strcmp(opts->iocharset, "utf8")) {
1251                 fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1252                        " for FAT filesystems, filesystem will be "
1253                        "case sensitive!");
1254         }
1255
1256         /* If user doesn't specify allow_utime, it's initialized from dmask. */
1257         if (opts->allow_utime == (unsigned short)-1)
1258                 opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1259         if (opts->unicode_xlate)
1260                 opts->utf8 = 0;
1261         if (opts->nfs == FAT_NFS_NOSTALE_RO) {
1262                 sb->s_flags |= MS_RDONLY;
1263                 sb->s_export_op = &fat_export_ops_nostale;
1264         }
1265
1266         return 0;
1267 }
1268
1269 static int fat_read_root(struct inode *inode)
1270 {
1271         struct super_block *sb = inode->i_sb;
1272         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1273         int error;
1274
1275         MSDOS_I(inode)->i_pos = MSDOS_ROOT_INO;
1276         inode->i_uid = sbi->options.fs_uid;
1277         inode->i_gid = sbi->options.fs_gid;
1278         inode->i_version++;
1279         inode->i_generation = 0;
1280         inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1281         inode->i_op = sbi->dir_ops;
1282         inode->i_fop = &fat_dir_operations;
1283         if (sbi->fat_bits == 32) {
1284                 MSDOS_I(inode)->i_start = sbi->root_cluster;
1285                 error = fat_calc_dir_size(inode);
1286                 if (error < 0)
1287                         return error;
1288         } else {
1289                 MSDOS_I(inode)->i_start = 0;
1290                 inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1291         }
1292         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1293                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1294         MSDOS_I(inode)->i_logstart = 0;
1295         MSDOS_I(inode)->mmu_private = inode->i_size;
1296
1297         fat_save_attrs(inode, ATTR_DIR);
1298         inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1299         inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1300         set_nlink(inode, fat_subdirs(inode)+2);
1301
1302         return 0;
1303 }
1304
1305 static unsigned long calc_fat_clusters(struct super_block *sb)
1306 {
1307         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1308
1309         /* Divide first to avoid overflow */
1310         if (sbi->fat_bits != 12) {
1311                 unsigned long ent_per_sec = sb->s_blocksize * 8 / sbi->fat_bits;
1312                 return ent_per_sec * sbi->fat_length;
1313         }
1314
1315         return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1316 }
1317
1318 static bool fat_bpb_is_zero(struct fat_boot_sector *b)
1319 {
1320         if (get_unaligned_le16(&b->sector_size))
1321                 return false;
1322         if (b->sec_per_clus)
1323                 return false;
1324         if (b->reserved)
1325                 return false;
1326         if (b->fats)
1327                 return false;
1328         if (get_unaligned_le16(&b->dir_entries))
1329                 return false;
1330         if (get_unaligned_le16(&b->sectors))
1331                 return false;
1332         if (b->media)
1333                 return false;
1334         if (b->fat_length)
1335                 return false;
1336         if (b->secs_track)
1337                 return false;
1338         if (b->heads)
1339                 return false;
1340         return true;
1341 }
1342
1343 static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
1344         int silent, struct fat_bios_param_block *bpb)
1345 {
1346         int error = -EINVAL;
1347
1348         /* Read in BPB ... */
1349         memset(bpb, 0, sizeof(*bpb));
1350         bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
1351         bpb->fat_sec_per_clus = b->sec_per_clus;
1352         bpb->fat_reserved = le16_to_cpu(b->reserved);
1353         bpb->fat_fats = b->fats;
1354         bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
1355         bpb->fat_sectors = get_unaligned_le16(&b->sectors);
1356         bpb->fat_fat_length = le16_to_cpu(b->fat_length);
1357         bpb->fat_total_sect = le32_to_cpu(b->total_sect);
1358
1359         bpb->fat16_state = b->fat16.state;
1360         bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
1361
1362         bpb->fat32_length = le32_to_cpu(b->fat32.length);
1363         bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
1364         bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
1365         bpb->fat32_state = b->fat32.state;
1366         bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
1367
1368         /* Validate this looks like a FAT filesystem BPB */
1369         if (!bpb->fat_reserved) {
1370                 if (!silent)
1371                         fat_msg(sb, KERN_ERR,
1372                                 "bogus number of reserved sectors");
1373                 goto out;
1374         }
1375         if (!bpb->fat_fats) {
1376                 if (!silent)
1377                         fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1378                 goto out;
1379         }
1380
1381         /*
1382          * Earlier we checked here that b->secs_track and b->head are nonzero,
1383          * but it turns out valid FAT filesystems can have zero there.
1384          */
1385
1386         if (!fat_valid_media(b->media)) {
1387                 if (!silent)
1388                         fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1389                                 (unsigned)b->media);
1390                 goto out;
1391         }
1392
1393         if (!is_power_of_2(bpb->fat_sector_size)
1394             || (bpb->fat_sector_size < 512)
1395             || (bpb->fat_sector_size > 4096)) {
1396                 if (!silent)
1397                         fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1398                                (unsigned)bpb->fat_sector_size);
1399                 goto out;
1400         }
1401
1402         if (!is_power_of_2(bpb->fat_sec_per_clus)) {
1403                 if (!silent)
1404                         fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1405                                 (unsigned)bpb->fat_sec_per_clus);
1406                 goto out;
1407         }
1408
1409         error = 0;
1410
1411 out:
1412         return error;
1413 }
1414
1415 static int fat_read_static_bpb(struct super_block *sb,
1416         struct fat_boot_sector *b, int silent,
1417         struct fat_bios_param_block *bpb)
1418 {
1419         static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
1420
1421         struct fat_floppy_defaults *fdefaults = NULL;
1422         int error = -EINVAL;
1423         sector_t bd_sects;
1424         unsigned i;
1425
1426         bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
1427
1428         /* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
1429         if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
1430                 if (!silent)
1431                         fat_msg(sb, KERN_ERR,
1432                                 "%s; no bootstrapping code", notdos1x);
1433                 goto out;
1434         }
1435
1436         /*
1437          * If any value in this region is non-zero, it isn't archaic
1438          * DOS.
1439          */
1440         if (!fat_bpb_is_zero(b)) {
1441                 if (!silent)
1442                         fat_msg(sb, KERN_ERR,
1443                                 "%s; DOS 2.x BPB is non-zero", notdos1x);
1444                 goto out;
1445         }
1446
1447         for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
1448                 if (floppy_defaults[i].nr_sectors == bd_sects) {
1449                         fdefaults = &floppy_defaults[i];
1450                         break;
1451                 }
1452         }
1453
1454         if (fdefaults == NULL) {
1455                 if (!silent)
1456                         fat_msg(sb, KERN_WARNING,
1457                                 "This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
1458                                 (u64)bd_sects);
1459                 goto out;
1460         }
1461
1462         if (!silent)
1463                 fat_msg(sb, KERN_INFO,
1464                         "This looks like a DOS 1.x volume; assuming default BPB values");
1465
1466         memset(bpb, 0, sizeof(*bpb));
1467         bpb->fat_sector_size = SECTOR_SIZE;
1468         bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
1469         bpb->fat_reserved = 1;
1470         bpb->fat_fats = 2;
1471         bpb->fat_dir_entries = fdefaults->dir_entries;
1472         bpb->fat_sectors = fdefaults->nr_sectors;
1473         bpb->fat_fat_length = fdefaults->fat_length;
1474
1475         error = 0;
1476
1477 out:
1478         return error;
1479 }
1480
1481 /*
1482  * Read the super block of an MS-DOS FS.
1483  */
1484 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1485                    void (*setup)(struct super_block *))
1486 {
1487         struct inode *root_inode = NULL, *fat_inode = NULL;
1488         struct inode *fsinfo_inode = NULL;
1489         struct buffer_head *bh;
1490         struct fat_bios_param_block bpb;
1491         struct msdos_sb_info *sbi;
1492         u16 logical_sector_size;
1493         u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1494         int debug;
1495         long error;
1496         char buf[50];
1497
1498         /*
1499          * GFP_KERNEL is ok here, because while we do hold the
1500          * supeblock lock, memory pressure can't call back into
1501          * the filesystem, since we're only just about to mount
1502          * it and have no inodes etc active!
1503          */
1504         sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1505         if (!sbi)
1506                 return -ENOMEM;
1507         sb->s_fs_info = sbi;
1508
1509         sb->s_flags |= MS_NODIRATIME;
1510         sb->s_magic = MSDOS_SUPER_MAGIC;
1511         sb->s_op = &fat_sops;
1512         sb->s_export_op = &fat_export_ops;
1513         mutex_init(&sbi->nfs_build_inode_lock);
1514         ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1515                              DEFAULT_RATELIMIT_BURST);
1516
1517         error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1518         if (error)
1519                 goto out_fail;
1520
1521         setup(sb); /* flavour-specific stuff that needs options */
1522
1523         error = -EIO;
1524         sb_min_blocksize(sb, 512);
1525         bh = sb_bread(sb, 0);
1526         if (bh == NULL) {
1527                 fat_msg(sb, KERN_ERR, "unable to read boot sector");
1528                 goto out_fail;
1529         }
1530
1531         error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
1532                 &bpb);
1533         if (error == -EINVAL && sbi->options.dos1xfloppy)
1534                 error = fat_read_static_bpb(sb,
1535                         (struct fat_boot_sector *)bh->b_data, silent, &bpb);
1536         brelse(bh);
1537
1538         if (error == -EINVAL)
1539                 goto out_invalid;
1540         else if (error)
1541                 goto out_fail;
1542
1543         logical_sector_size = bpb.fat_sector_size;
1544         sbi->sec_per_clus = bpb.fat_sec_per_clus;
1545
1546         error = -EIO;
1547         if (logical_sector_size < sb->s_blocksize) {
1548                 fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1549                        " (logical sector size = %u)", logical_sector_size);
1550                 goto out_fail;
1551         }
1552
1553         if (logical_sector_size > sb->s_blocksize) {
1554                 struct buffer_head *bh_resize;
1555
1556                 if (!sb_set_blocksize(sb, logical_sector_size)) {
1557                         fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1558                                logical_sector_size);
1559                         goto out_fail;
1560                 }
1561
1562                 /* Verify that the larger boot sector is fully readable */
1563                 bh_resize = sb_bread(sb, 0);
1564                 if (bh_resize == NULL) {
1565                         fat_msg(sb, KERN_ERR, "unable to read boot sector"
1566                                " (logical sector size = %lu)",
1567                                sb->s_blocksize);
1568                         goto out_fail;
1569                 }
1570                 brelse(bh_resize);
1571         }
1572
1573         mutex_init(&sbi->s_lock);
1574         sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1575         sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1576         sbi->fats = bpb.fat_fats;
1577         sbi->fat_bits = 0;              /* Don't know yet */
1578         sbi->fat_start = bpb.fat_reserved;
1579         sbi->fat_length = bpb.fat_fat_length;
1580         sbi->root_cluster = 0;
1581         sbi->free_clusters = -1;        /* Don't know yet */
1582         sbi->free_clus_valid = 0;
1583         sbi->prev_free = FAT_START_ENT;
1584         sb->s_maxbytes = 0xffffffff;
1585
1586         if (!sbi->fat_length && bpb.fat32_length) {
1587                 struct fat_boot_fsinfo *fsinfo;
1588                 struct buffer_head *fsinfo_bh;
1589
1590                 /* Must be FAT32 */
1591                 sbi->fat_bits = 32;
1592                 sbi->fat_length = bpb.fat32_length;
1593                 sbi->root_cluster = bpb.fat32_root_cluster;
1594
1595                 /* MC - if info_sector is 0, don't multiply by 0 */
1596                 sbi->fsinfo_sector = bpb.fat32_info_sector;
1597                 if (sbi->fsinfo_sector == 0)
1598                         sbi->fsinfo_sector = 1;
1599
1600                 fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1601                 if (fsinfo_bh == NULL) {
1602                         fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1603                                " (sector = %lu)", sbi->fsinfo_sector);
1604                         goto out_fail;
1605                 }
1606
1607                 fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1608                 if (!IS_FSINFO(fsinfo)) {
1609                         fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1610                                "0x%08x, 0x%08x (sector = %lu)",
1611                                le32_to_cpu(fsinfo->signature1),
1612                                le32_to_cpu(fsinfo->signature2),
1613                                sbi->fsinfo_sector);
1614                 } else {
1615                         if (sbi->options.usefree)
1616                                 sbi->free_clus_valid = 1;
1617                         sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1618                         sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1619                 }
1620
1621                 brelse(fsinfo_bh);
1622         }
1623
1624         /* interpret volume ID as a little endian 32 bit integer */
1625         if (sbi->fat_bits == 32)
1626                 sbi->vol_id = bpb.fat32_vol_id;
1627         else /* fat 16 or 12 */
1628                 sbi->vol_id = bpb.fat16_vol_id;
1629
1630         sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1631         sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1632
1633         sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1634         sbi->dir_entries = bpb.fat_dir_entries;
1635         if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1636                 if (!silent)
1637                         fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
1638                                " (%u)", sbi->dir_entries);
1639                 goto out_invalid;
1640         }
1641
1642         rootdir_sectors = sbi->dir_entries
1643                 * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1644         sbi->data_start = sbi->dir_start + rootdir_sectors;
1645         total_sectors = bpb.fat_sectors;
1646         if (total_sectors == 0)
1647                 total_sectors = bpb.fat_total_sect;
1648
1649         total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1650
1651         if (sbi->fat_bits != 32)
1652                 sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1653
1654         /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1655         if (sbi->fat_bits == 32)
1656                 sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
1657         else /* fat 16 or 12 */
1658                 sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
1659
1660         /* check that FAT table does not overflow */
1661         fat_clusters = calc_fat_clusters(sb);
1662         total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1663         if (total_clusters > MAX_FAT(sb)) {
1664                 if (!silent)
1665                         fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1666                                total_clusters);
1667                 goto out_invalid;
1668         }
1669
1670         sbi->max_cluster = total_clusters + FAT_START_ENT;
1671         /* check the free_clusters, it's not necessarily correct */
1672         if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1673                 sbi->free_clusters = -1;
1674         /* check the prev_free, it's not necessarily correct */
1675         sbi->prev_free %= sbi->max_cluster;
1676         if (sbi->prev_free < FAT_START_ENT)
1677                 sbi->prev_free = FAT_START_ENT;
1678
1679         /* set up enough so that it can read an inode */
1680         fat_hash_init(sb);
1681         dir_hash_init(sb);
1682         fat_ent_access_init(sb);
1683
1684         /*
1685          * The low byte of FAT's first entry must have same value with
1686          * media-field.  But in real world, too many devices is
1687          * writing wrong value.  So, removed that validity check.
1688          *
1689          * if (FAT_FIRST_ENT(sb, media) != first)
1690          */
1691
1692         error = -EINVAL;
1693         sprintf(buf, "cp%d", sbi->options.codepage);
1694         sbi->nls_disk = load_nls(buf);
1695         if (!sbi->nls_disk) {
1696                 fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1697                 goto out_fail;
1698         }
1699
1700         /* FIXME: utf8 is using iocharset for upper/lower conversion */
1701         if (sbi->options.isvfat) {
1702                 sbi->nls_io = load_nls(sbi->options.iocharset);
1703                 if (!sbi->nls_io) {
1704                         fat_msg(sb, KERN_ERR, "IO charset %s not found",
1705                                sbi->options.iocharset);
1706                         goto out_fail;
1707                 }
1708         }
1709
1710         error = -ENOMEM;
1711         fat_inode = new_inode(sb);
1712         if (!fat_inode)
1713                 goto out_fail;
1714         MSDOS_I(fat_inode)->i_pos = 0;
1715         sbi->fat_inode = fat_inode;
1716
1717         fsinfo_inode = new_inode(sb);
1718         if (!fsinfo_inode)
1719                 goto out_fail;
1720         fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1721         sbi->fsinfo_inode = fsinfo_inode;
1722         insert_inode_hash(fsinfo_inode);
1723
1724         root_inode = new_inode(sb);
1725         if (!root_inode)
1726                 goto out_fail;
1727         root_inode->i_ino = MSDOS_ROOT_INO;
1728         root_inode->i_version = 1;
1729         error = fat_read_root(root_inode);
1730         if (error < 0) {
1731                 iput(root_inode);
1732                 goto out_fail;
1733         }
1734         error = -ENOMEM;
1735         insert_inode_hash(root_inode);
1736         fat_attach(root_inode, 0);
1737         sb->s_root = d_make_root(root_inode);
1738         if (!sb->s_root) {
1739                 fat_msg(sb, KERN_ERR, "get root inode failed");
1740                 goto out_fail;
1741         }
1742
1743         if (sbi->options.discard) {
1744                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1745                 if (!blk_queue_discard(q))
1746                         fat_msg(sb, KERN_WARNING,
1747                                         "mounting with \"discard\" option, but "
1748                                         "the device does not support discard");
1749         }
1750
1751         fat_set_state(sb, 1, 0);
1752         return 0;
1753
1754 out_invalid:
1755         error = -EINVAL;
1756         if (!silent)
1757                 fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1758
1759 out_fail:
1760         if (fsinfo_inode)
1761                 iput(fsinfo_inode);
1762         if (fat_inode)
1763                 iput(fat_inode);
1764         unload_nls(sbi->nls_io);
1765         unload_nls(sbi->nls_disk);
1766         if (sbi->options.iocharset != fat_default_iocharset)
1767                 kfree(sbi->options.iocharset);
1768         sb->s_fs_info = NULL;
1769         kfree(sbi);
1770         return error;
1771 }
1772
1773 EXPORT_SYMBOL_GPL(fat_fill_super);
1774
1775 /*
1776  * helper function for fat_flush_inodes.  This writes both the inode
1777  * and the file data blocks, waiting for in flight data blocks before
1778  * the start of the call.  It does not wait for any io started
1779  * during the call
1780  */
1781 static int writeback_inode(struct inode *inode)
1782 {
1783
1784         int ret;
1785
1786         /* if we used wait=1, sync_inode_metadata waits for the io for the
1787         * inode to finish.  So wait=0 is sent down to sync_inode_metadata
1788         * and filemap_fdatawrite is used for the data blocks
1789         */
1790         ret = sync_inode_metadata(inode, 0);
1791         if (!ret)
1792                 ret = filemap_fdatawrite(inode->i_mapping);
1793         return ret;
1794 }
1795
1796 /*
1797  * write data and metadata corresponding to i1 and i2.  The io is
1798  * started but we do not wait for any of it to finish.
1799  *
1800  * filemap_flush is used for the block device, so if there is a dirty
1801  * page for a block already in flight, we will not wait and start the
1802  * io over again
1803  */
1804 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1805 {
1806         int ret = 0;
1807         if (!MSDOS_SB(sb)->options.flush)
1808                 return 0;
1809         if (i1)
1810                 ret = writeback_inode(i1);
1811         if (!ret && i2)
1812                 ret = writeback_inode(i2);
1813         if (!ret) {
1814                 struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1815                 ret = filemap_flush(mapping);
1816         }
1817         return ret;
1818 }
1819 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1820
1821 static int __init init_fat_fs(void)
1822 {
1823         int err;
1824
1825         err = fat_cache_init();
1826         if (err)
1827                 return err;
1828
1829         err = fat_init_inodecache();
1830         if (err)
1831                 goto failed;
1832
1833         return 0;
1834
1835 failed:
1836         fat_cache_destroy();
1837         return err;
1838 }
1839
1840 static void __exit exit_fat_fs(void)
1841 {
1842         fat_cache_destroy();
1843         fat_destroy_inodecache();
1844 }
1845
1846 module_init(init_fat_fs)
1847 module_exit(exit_fat_fs)
1848
1849 MODULE_LICENSE("GPL");