]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/fat/inode.c
5b2cb358748da160abaa6267c16c555af3998bff
[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         MSDOS_I(inode)->i_disksize = MSDOS_I(inode)->mmu_private;
159
160         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
161         if (err)
162                 return err;
163
164         BUG_ON(!phys);
165         BUG_ON(*max_blocks != mapped_blocks);
166         set_buffer_new(bh_result);
167         map_bh(bh_result, sb, phys);
168
169         return 0;
170 }
171
172 static int fat_get_block(struct inode *inode, sector_t iblock,
173                          struct buffer_head *bh_result, int create)
174 {
175         struct super_block *sb = inode->i_sb;
176         unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
177         int err;
178
179         err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
180         if (err)
181                 return err;
182         bh_result->b_size = max_blocks << sb->s_blocksize_bits;
183         return 0;
184 }
185
186 static int fat_writepage(struct page *page, struct writeback_control *wbc)
187 {
188         return block_write_full_page(page, fat_get_block, wbc);
189 }
190
191 static int fat_writepages(struct address_space *mapping,
192                           struct writeback_control *wbc)
193 {
194         return mpage_writepages(mapping, wbc, fat_get_block);
195 }
196
197 static int fat_readpage(struct file *file, struct page *page)
198 {
199         return mpage_readpage(page, fat_get_block);
200 }
201
202 static int fat_readpages(struct file *file, struct address_space *mapping,
203                          struct list_head *pages, unsigned nr_pages)
204 {
205         return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
206 }
207
208 static void fat_write_failed(struct address_space *mapping, loff_t to)
209 {
210         struct inode *inode = mapping->host;
211
212         if (to > inode->i_size) {
213                 truncate_pagecache(inode, inode->i_size);
214                 fat_truncate_blocks(inode, inode->i_size);
215         }
216 }
217
218 static int fat_write_begin(struct file *file, struct address_space *mapping,
219                         loff_t pos, unsigned len, unsigned flags,
220                         struct page **pagep, void **fsdata)
221 {
222         int err;
223
224         *pagep = NULL;
225         err = cont_write_begin(file, mapping, pos, len, flags,
226                                 pagep, fsdata, fat_get_block,
227                                 &MSDOS_I(mapping->host)->mmu_private);
228         if (err < 0)
229                 fat_write_failed(mapping, pos + len);
230         return err;
231 }
232
233 static int fat_write_end(struct file *file, struct address_space *mapping,
234                         loff_t pos, unsigned len, unsigned copied,
235                         struct page *pagep, void *fsdata)
236 {
237         struct inode *inode = mapping->host;
238         int err;
239         err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
240         if (err < len)
241                 fat_write_failed(mapping, pos + len);
242         if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
243                 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
244                 MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
245                 mark_inode_dirty(inode);
246         }
247         return err;
248 }
249
250 static ssize_t fat_direct_IO(int rw, struct kiocb *iocb,
251                              struct iov_iter *iter,
252                              loff_t offset)
253 {
254         struct file *file = iocb->ki_filp;
255         struct address_space *mapping = file->f_mapping;
256         struct inode *inode = mapping->host;
257         size_t count = iov_iter_count(iter);
258         ssize_t ret;
259
260         if (rw == WRITE) {
261                 /*
262                  * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
263                  * so we need to update the ->mmu_private to block boundary.
264                  *
265                  * But we must fill the remaining area or hole by nul for
266                  * updating ->mmu_private.
267                  *
268                  * Return 0, and fallback to normal buffered write.
269                  */
270                 loff_t size = offset + count;
271                 if (MSDOS_I(inode)->mmu_private < size)
272                         return 0;
273         }
274
275         /*
276          * FAT need to use the DIO_LOCKING for avoiding the race
277          * condition of fat_get_block() and ->truncate().
278          */
279         ret = blockdev_direct_IO(rw, iocb, inode, iter, offset, fat_get_block);
280         if (ret < 0 && (rw & WRITE))
281                 fat_write_failed(mapping, offset + count);
282
283         return ret;
284 }
285
286 static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
287 {
288         sector_t blocknr;
289
290         /* fat_get_cluster() assumes the requested blocknr isn't truncated. */
291         down_read(&MSDOS_I(mapping->host)->truncate_lock);
292         blocknr = generic_block_bmap(mapping, block, fat_get_block);
293         up_read(&MSDOS_I(mapping->host)->truncate_lock);
294
295         return blocknr;
296 }
297
298 static const struct address_space_operations fat_aops = {
299         .readpage       = fat_readpage,
300         .readpages      = fat_readpages,
301         .writepage      = fat_writepage,
302         .writepages     = fat_writepages,
303         .write_begin    = fat_write_begin,
304         .write_end      = fat_write_end,
305         .direct_IO      = fat_direct_IO,
306         .bmap           = _fat_bmap
307 };
308
309 /*
310  * New FAT inode stuff. We do the following:
311  *      a) i_ino is constant and has nothing with on-disk location.
312  *      b) FAT manages its own cache of directory entries.
313  *      c) *This* cache is indexed by on-disk location.
314  *      d) inode has an associated directory entry, all right, but
315  *              it may be unhashed.
316  *      e) currently entries are stored within struct inode. That should
317  *              change.
318  *      f) we deal with races in the following way:
319  *              1. readdir() and lookup() do FAT-dir-cache lookup.
320  *              2. rename() unhashes the F-d-c entry and rehashes it in
321  *                      a new place.
322  *              3. unlink() and rmdir() unhash F-d-c entry.
323  *              4. fat_write_inode() checks whether the thing is unhashed.
324  *                      If it is we silently return. If it isn't we do bread(),
325  *                      check if the location is still valid and retry if it
326  *                      isn't. Otherwise we do changes.
327  *              5. Spinlock is used to protect hash/unhash/location check/lookup
328  *              6. fat_evict_inode() unhashes the F-d-c entry.
329  *              7. lookup() and readdir() do igrab() if they find a F-d-c entry
330  *                      and consider negative result as cache miss.
331  */
332
333 static void fat_hash_init(struct super_block *sb)
334 {
335         struct msdos_sb_info *sbi = MSDOS_SB(sb);
336         int i;
337
338         spin_lock_init(&sbi->inode_hash_lock);
339         for (i = 0; i < FAT_HASH_SIZE; i++)
340                 INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
341 }
342
343 static inline unsigned long fat_hash(loff_t i_pos)
344 {
345         return hash_32(i_pos, FAT_HASH_BITS);
346 }
347
348 static void dir_hash_init(struct super_block *sb)
349 {
350         struct msdos_sb_info *sbi = MSDOS_SB(sb);
351         int i;
352
353         spin_lock_init(&sbi->dir_hash_lock);
354         for (i = 0; i < FAT_HASH_SIZE; i++)
355                 INIT_HLIST_HEAD(&sbi->dir_hashtable[i]);
356 }
357
358 void fat_attach(struct inode *inode, loff_t i_pos)
359 {
360         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
361
362         if (inode->i_ino != MSDOS_ROOT_INO) {
363                 struct hlist_head *head =   sbi->inode_hashtable
364                                           + fat_hash(i_pos);
365
366                 spin_lock(&sbi->inode_hash_lock);
367                 MSDOS_I(inode)->i_pos = i_pos;
368                 hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
369                 spin_unlock(&sbi->inode_hash_lock);
370         }
371
372         /* If NFS support is enabled, cache the mapping of start cluster
373          * to directory inode. This is used during reconnection of
374          * dentries to the filesystem root.
375          */
376         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
377                 struct hlist_head *d_head = sbi->dir_hashtable;
378                 d_head += fat_dir_hash(MSDOS_I(inode)->i_logstart);
379
380                 spin_lock(&sbi->dir_hash_lock);
381                 hlist_add_head(&MSDOS_I(inode)->i_dir_hash, d_head);
382                 spin_unlock(&sbi->dir_hash_lock);
383         }
384 }
385 EXPORT_SYMBOL_GPL(fat_attach);
386
387 void fat_detach(struct inode *inode)
388 {
389         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
390         spin_lock(&sbi->inode_hash_lock);
391         MSDOS_I(inode)->i_pos = 0;
392         hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
393         spin_unlock(&sbi->inode_hash_lock);
394
395         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
396                 spin_lock(&sbi->dir_hash_lock);
397                 hlist_del_init(&MSDOS_I(inode)->i_dir_hash);
398                 spin_unlock(&sbi->dir_hash_lock);
399         }
400 }
401 EXPORT_SYMBOL_GPL(fat_detach);
402
403 struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
404 {
405         struct msdos_sb_info *sbi = MSDOS_SB(sb);
406         struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
407         struct msdos_inode_info *i;
408         struct inode *inode = NULL;
409
410         spin_lock(&sbi->inode_hash_lock);
411         hlist_for_each_entry(i, head, i_fat_hash) {
412                 BUG_ON(i->vfs_inode.i_sb != sb);
413                 if (i->i_pos != i_pos)
414                         continue;
415                 inode = igrab(&i->vfs_inode);
416                 if (inode)
417                         break;
418         }
419         spin_unlock(&sbi->inode_hash_lock);
420         return inode;
421 }
422
423 static int is_exec(unsigned char *extension)
424 {
425         unsigned char exe_extensions[] = "EXECOMBAT", *walk;
426
427         for (walk = exe_extensions; *walk; walk += 3)
428                 if (!strncmp(extension, walk, 3))
429                         return 1;
430         return 0;
431 }
432
433 static int fat_calc_dir_size(struct inode *inode)
434 {
435         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
436         int ret, fclus, dclus;
437
438         inode->i_size = 0;
439         if (MSDOS_I(inode)->i_start == 0)
440                 return 0;
441
442         ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
443         if (ret < 0)
444                 return ret;
445         inode->i_size = (fclus + 1) << sbi->cluster_bits;
446
447         return 0;
448 }
449
450 /* doesn't deal with root inode */
451 int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
452 {
453         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
454         int error;
455
456         MSDOS_I(inode)->i_pos = 0;
457         inode->i_uid = sbi->options.fs_uid;
458         inode->i_gid = sbi->options.fs_gid;
459         inode->i_version++;
460         inode->i_generation = get_seconds();
461
462         if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
463                 inode->i_generation &= ~1;
464                 inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
465                 inode->i_op = sbi->dir_ops;
466                 inode->i_fop = &fat_dir_operations;
467
468                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
469                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
470                 error = fat_calc_dir_size(inode);
471                 if (error < 0)
472                         return error;
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         }
488
489         MSDOS_I(inode)->mmu_private = inode->i_size;
490         MSDOS_I(inode)->i_disksize = round_up(inode->i_size,
491                 inode->i_sb->s_blocksize);
492
493         if (de->attr & ATTR_SYS) {
494                 if (sbi->options.sys_immutable)
495                         inode->i_flags |= S_IMMUTABLE;
496         }
497         fat_save_attrs(inode, de->attr);
498
499         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
500                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
501
502         fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
503         if (sbi->options.isvfat) {
504                 fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
505                                   de->cdate, de->ctime_cs);
506                 fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
507         } else
508                 inode->i_ctime = inode->i_atime = inode->i_mtime;
509
510         return 0;
511 }
512
513 static inline void fat_lock_build_inode(struct msdos_sb_info *sbi)
514 {
515         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
516                 mutex_lock(&sbi->nfs_build_inode_lock);
517 }
518
519 static inline void fat_unlock_build_inode(struct msdos_sb_info *sbi)
520 {
521         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
522                 mutex_unlock(&sbi->nfs_build_inode_lock);
523 }
524
525 struct inode *fat_build_inode(struct super_block *sb,
526                         struct msdos_dir_entry *de, loff_t i_pos)
527 {
528         struct inode *inode;
529         int err;
530
531         fat_lock_build_inode(MSDOS_SB(sb));
532         inode = fat_iget(sb, i_pos);
533         if (inode)
534                 goto out;
535         inode = new_inode(sb);
536         if (!inode) {
537                 inode = ERR_PTR(-ENOMEM);
538                 goto out;
539         }
540         inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
541         inode->i_version = 1;
542         err = fat_fill_inode(inode, de);
543         if (err) {
544                 iput(inode);
545                 inode = ERR_PTR(err);
546                 goto out;
547         }
548         fat_attach(inode, i_pos);
549         insert_inode_hash(inode);
550 out:
551         fat_unlock_build_inode(MSDOS_SB(sb));
552         return inode;
553 }
554
555 EXPORT_SYMBOL_GPL(fat_build_inode);
556
557 static void fat_evict_inode(struct inode *inode)
558 {
559         truncate_inode_pages_final(&inode->i_data);
560         if (!inode->i_nlink) {
561                 inode->i_size = 0;
562                 fat_truncate_blocks(inode, 0);
563         }
564         invalidate_inode_buffers(inode);
565         clear_inode(inode);
566         fat_cache_inval_inode(inode);
567         fat_detach(inode);
568 }
569
570 static void fat_set_state(struct super_block *sb,
571                         unsigned int set, unsigned int force)
572 {
573         struct buffer_head *bh;
574         struct fat_boot_sector *b;
575         struct msdos_sb_info *sbi = sb->s_fs_info;
576
577         /* do not change any thing if mounted read only */
578         if ((sb->s_flags & MS_RDONLY) && !force)
579                 return;
580
581         /* do not change state if fs was dirty */
582         if (sbi->dirty) {
583                 /* warn only on set (mount). */
584                 if (set)
585                         fat_msg(sb, KERN_WARNING, "Volume was not properly "
586                                 "unmounted. Some data may be corrupt. "
587                                 "Please run fsck.");
588                 return;
589         }
590
591         bh = sb_bread(sb, 0);
592         if (bh == NULL) {
593                 fat_msg(sb, KERN_ERR, "unable to read boot sector "
594                         "to mark fs as dirty");
595                 return;
596         }
597
598         b = (struct fat_boot_sector *) bh->b_data;
599
600         if (sbi->fat_bits == 32) {
601                 if (set)
602                         b->fat32.state |= FAT_STATE_DIRTY;
603                 else
604                         b->fat32.state &= ~FAT_STATE_DIRTY;
605         } else /* fat 16 and 12 */ {
606                 if (set)
607                         b->fat16.state |= FAT_STATE_DIRTY;
608                 else
609                         b->fat16.state &= ~FAT_STATE_DIRTY;
610         }
611
612         mark_buffer_dirty(bh);
613         sync_dirty_buffer(bh);
614         brelse(bh);
615 }
616
617 static void delayed_free(struct rcu_head *p)
618 {
619         struct msdos_sb_info *sbi = container_of(p, struct msdos_sb_info, rcu);
620         unload_nls(sbi->nls_disk);
621         unload_nls(sbi->nls_io);
622         if (sbi->options.iocharset != fat_default_iocharset)
623                 kfree(sbi->options.iocharset);
624         kfree(sbi);
625 }
626
627 static void fat_put_super(struct super_block *sb)
628 {
629         struct msdos_sb_info *sbi = MSDOS_SB(sb);
630
631         fat_set_state(sb, 0, 0);
632
633         iput(sbi->fsinfo_inode);
634         iput(sbi->fat_inode);
635
636         call_rcu(&sbi->rcu, delayed_free);
637 }
638
639 static struct kmem_cache *fat_inode_cachep;
640
641 static struct inode *fat_alloc_inode(struct super_block *sb)
642 {
643         struct msdos_inode_info *ei;
644         ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
645         if (!ei)
646                 return NULL;
647
648         init_rwsem(&ei->truncate_lock);
649         return &ei->vfs_inode;
650 }
651
652 static void fat_i_callback(struct rcu_head *head)
653 {
654         struct inode *inode = container_of(head, struct inode, i_rcu);
655         kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
656 }
657
658 static void fat_destroy_inode(struct inode *inode)
659 {
660         call_rcu(&inode->i_rcu, fat_i_callback);
661 }
662
663 static void init_once(void *foo)
664 {
665         struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
666
667         spin_lock_init(&ei->cache_lru_lock);
668         ei->nr_caches = 0;
669         ei->cache_valid_id = FAT_CACHE_VALID + 1;
670         INIT_LIST_HEAD(&ei->cache_lru);
671         INIT_HLIST_NODE(&ei->i_fat_hash);
672         INIT_HLIST_NODE(&ei->i_dir_hash);
673         inode_init_once(&ei->vfs_inode);
674 }
675
676 static int __init fat_init_inodecache(void)
677 {
678         fat_inode_cachep = kmem_cache_create("fat_inode_cache",
679                                              sizeof(struct msdos_inode_info),
680                                              0, (SLAB_RECLAIM_ACCOUNT|
681                                                 SLAB_MEM_SPREAD),
682                                              init_once);
683         if (fat_inode_cachep == NULL)
684                 return -ENOMEM;
685         return 0;
686 }
687
688 static void __exit fat_destroy_inodecache(void)
689 {
690         /*
691          * Make sure all delayed rcu free inodes are flushed before we
692          * destroy cache.
693          */
694         rcu_barrier();
695         kmem_cache_destroy(fat_inode_cachep);
696 }
697
698 static int fat_remount(struct super_block *sb, int *flags, char *data)
699 {
700         int new_rdonly;
701         struct msdos_sb_info *sbi = MSDOS_SB(sb);
702         *flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
703
704         sync_filesystem(sb);
705
706         /* make sure we update state on remount. */
707         new_rdonly = *flags & MS_RDONLY;
708         if (new_rdonly != (sb->s_flags & MS_RDONLY)) {
709                 if (new_rdonly)
710                         fat_set_state(sb, 0, 0);
711                 else
712                         fat_set_state(sb, 1, 1);
713         }
714         return 0;
715 }
716
717 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
718 {
719         struct super_block *sb = dentry->d_sb;
720         struct msdos_sb_info *sbi = MSDOS_SB(sb);
721         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
722
723         /* If the count of free cluster is still unknown, counts it here. */
724         if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
725                 int err = fat_count_free_clusters(dentry->d_sb);
726                 if (err)
727                         return err;
728         }
729
730         buf->f_type = dentry->d_sb->s_magic;
731         buf->f_bsize = sbi->cluster_size;
732         buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
733         buf->f_bfree = sbi->free_clusters;
734         buf->f_bavail = sbi->free_clusters;
735         buf->f_fsid.val[0] = (u32)id;
736         buf->f_fsid.val[1] = (u32)(id >> 32);
737         buf->f_namelen =
738                 (sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
739
740         return 0;
741 }
742
743 static int __fat_write_inode(struct inode *inode, int wait)
744 {
745         struct super_block *sb = inode->i_sb;
746         struct msdos_sb_info *sbi = MSDOS_SB(sb);
747         struct buffer_head *bh;
748         struct msdos_dir_entry *raw_entry;
749         loff_t i_pos;
750         sector_t blocknr;
751         int err, offset;
752
753         if (inode->i_ino == MSDOS_ROOT_INO)
754                 return 0;
755
756 retry:
757         i_pos = fat_i_pos_read(sbi, inode);
758         if (!i_pos)
759                 return 0;
760
761         fat_get_blknr_offset(sbi, i_pos, &blocknr, &offset);
762         bh = sb_bread(sb, blocknr);
763         if (!bh) {
764                 fat_msg(sb, KERN_ERR, "unable to read inode block "
765                        "for updating (i_pos %lld)", i_pos);
766                 return -EIO;
767         }
768         spin_lock(&sbi->inode_hash_lock);
769         if (i_pos != MSDOS_I(inode)->i_pos) {
770                 spin_unlock(&sbi->inode_hash_lock);
771                 brelse(bh);
772                 goto retry;
773         }
774
775         raw_entry = &((struct msdos_dir_entry *) (bh->b_data))[offset];
776         if (S_ISDIR(inode->i_mode))
777                 raw_entry->size = 0;
778         else
779                 raw_entry->size = cpu_to_le32(inode->i_size);
780         raw_entry->attr = fat_make_attrs(inode);
781         fat_set_start(raw_entry, MSDOS_I(inode)->i_logstart);
782         fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
783                           &raw_entry->date, NULL);
784         if (sbi->options.isvfat) {
785                 __le16 atime;
786                 fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
787                                   &raw_entry->cdate, &raw_entry->ctime_cs);
788                 fat_time_unix2fat(sbi, &inode->i_atime, &atime,
789                                   &raw_entry->adate, NULL);
790         }
791         spin_unlock(&sbi->inode_hash_lock);
792         mark_buffer_dirty(bh);
793         err = 0;
794         if (wait)
795                 err = sync_dirty_buffer(bh);
796         brelse(bh);
797         return err;
798 }
799
800 static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
801 {
802         int err;
803
804         if (inode->i_ino == MSDOS_FSINFO_INO) {
805                 struct super_block *sb = inode->i_sb;
806
807                 mutex_lock(&MSDOS_SB(sb)->s_lock);
808                 err = fat_clusters_flush(sb);
809                 mutex_unlock(&MSDOS_SB(sb)->s_lock);
810         } else
811                 err = __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
812
813         return err;
814 }
815
816 int fat_sync_inode(struct inode *inode)
817 {
818         return __fat_write_inode(inode, 1);
819 }
820
821 EXPORT_SYMBOL_GPL(fat_sync_inode);
822
823 static int fat_show_options(struct seq_file *m, struct dentry *root);
824 static const struct super_operations fat_sops = {
825         .alloc_inode    = fat_alloc_inode,
826         .destroy_inode  = fat_destroy_inode,
827         .write_inode    = fat_write_inode,
828         .evict_inode    = fat_evict_inode,
829         .put_super      = fat_put_super,
830         .statfs         = fat_statfs,
831         .remount_fs     = fat_remount,
832
833         .show_options   = fat_show_options,
834 };
835
836 static int fat_show_options(struct seq_file *m, struct dentry *root)
837 {
838         struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
839         struct fat_mount_options *opts = &sbi->options;
840         int isvfat = opts->isvfat;
841
842         if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
843                 seq_printf(m, ",uid=%u",
844                                 from_kuid_munged(&init_user_ns, opts->fs_uid));
845         if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
846                 seq_printf(m, ",gid=%u",
847                                 from_kgid_munged(&init_user_ns, opts->fs_gid));
848         seq_printf(m, ",fmask=%04o", opts->fs_fmask);
849         seq_printf(m, ",dmask=%04o", opts->fs_dmask);
850         if (opts->allow_utime)
851                 seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
852         if (sbi->nls_disk)
853                 /* strip "cp" prefix from displayed option */
854                 seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
855         if (isvfat) {
856                 if (sbi->nls_io)
857                         seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
858
859                 switch (opts->shortname) {
860                 case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
861                         seq_puts(m, ",shortname=win95");
862                         break;
863                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
864                         seq_puts(m, ",shortname=winnt");
865                         break;
866                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
867                         seq_puts(m, ",shortname=mixed");
868                         break;
869                 case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
870                         seq_puts(m, ",shortname=lower");
871                         break;
872                 default:
873                         seq_puts(m, ",shortname=unknown");
874                         break;
875                 }
876         }
877         if (opts->name_check != 'n')
878                 seq_printf(m, ",check=%c", opts->name_check);
879         if (opts->usefree)
880                 seq_puts(m, ",usefree");
881         if (opts->quiet)
882                 seq_puts(m, ",quiet");
883         if (opts->showexec)
884                 seq_puts(m, ",showexec");
885         if (opts->sys_immutable)
886                 seq_puts(m, ",sys_immutable");
887         if (!isvfat) {
888                 if (opts->dotsOK)
889                         seq_puts(m, ",dotsOK=yes");
890                 if (opts->nocase)
891                         seq_puts(m, ",nocase");
892         } else {
893                 if (opts->utf8)
894                         seq_puts(m, ",utf8");
895                 if (opts->unicode_xlate)
896                         seq_puts(m, ",uni_xlate");
897                 if (!opts->numtail)
898                         seq_puts(m, ",nonumtail");
899                 if (opts->rodir)
900                         seq_puts(m, ",rodir");
901         }
902         if (opts->flush)
903                 seq_puts(m, ",flush");
904         if (opts->tz_set) {
905                 if (opts->time_offset)
906                         seq_printf(m, ",time_offset=%d", opts->time_offset);
907                 else
908                         seq_puts(m, ",tz=UTC");
909         }
910         if (opts->errors == FAT_ERRORS_CONT)
911                 seq_puts(m, ",errors=continue");
912         else if (opts->errors == FAT_ERRORS_PANIC)
913                 seq_puts(m, ",errors=panic");
914         else
915                 seq_puts(m, ",errors=remount-ro");
916         if (opts->nfs == FAT_NFS_NOSTALE_RO)
917                 seq_puts(m, ",nfs=nostale_ro");
918         else if (opts->nfs)
919                 seq_puts(m, ",nfs=stale_rw");
920         if (opts->discard)
921                 seq_puts(m, ",discard");
922         if (opts->dos1xfloppy)
923                 seq_puts(m, ",dos1xfloppy");
924
925         return 0;
926 }
927
928 enum {
929         Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
930         Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
931         Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
932         Opt_immutable, Opt_dots, Opt_nodots,
933         Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
934         Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
935         Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
936         Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
937         Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
938         Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
939 };
940
941 static const match_table_t fat_tokens = {
942         {Opt_check_r, "check=relaxed"},
943         {Opt_check_s, "check=strict"},
944         {Opt_check_n, "check=normal"},
945         {Opt_check_r, "check=r"},
946         {Opt_check_s, "check=s"},
947         {Opt_check_n, "check=n"},
948         {Opt_uid, "uid=%u"},
949         {Opt_gid, "gid=%u"},
950         {Opt_umask, "umask=%o"},
951         {Opt_dmask, "dmask=%o"},
952         {Opt_fmask, "fmask=%o"},
953         {Opt_allow_utime, "allow_utime=%o"},
954         {Opt_codepage, "codepage=%u"},
955         {Opt_usefree, "usefree"},
956         {Opt_nocase, "nocase"},
957         {Opt_quiet, "quiet"},
958         {Opt_showexec, "showexec"},
959         {Opt_debug, "debug"},
960         {Opt_immutable, "sys_immutable"},
961         {Opt_flush, "flush"},
962         {Opt_tz_utc, "tz=UTC"},
963         {Opt_time_offset, "time_offset=%d"},
964         {Opt_err_cont, "errors=continue"},
965         {Opt_err_panic, "errors=panic"},
966         {Opt_err_ro, "errors=remount-ro"},
967         {Opt_discard, "discard"},
968         {Opt_nfs_stale_rw, "nfs"},
969         {Opt_nfs_stale_rw, "nfs=stale_rw"},
970         {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
971         {Opt_dos1xfloppy, "dos1xfloppy"},
972         {Opt_obsolete, "conv=binary"},
973         {Opt_obsolete, "conv=text"},
974         {Opt_obsolete, "conv=auto"},
975         {Opt_obsolete, "conv=b"},
976         {Opt_obsolete, "conv=t"},
977         {Opt_obsolete, "conv=a"},
978         {Opt_obsolete, "fat=%u"},
979         {Opt_obsolete, "blocksize=%u"},
980         {Opt_obsolete, "cvf_format=%20s"},
981         {Opt_obsolete, "cvf_options=%100s"},
982         {Opt_obsolete, "posix"},
983         {Opt_err, NULL},
984 };
985 static const match_table_t msdos_tokens = {
986         {Opt_nodots, "nodots"},
987         {Opt_nodots, "dotsOK=no"},
988         {Opt_dots, "dots"},
989         {Opt_dots, "dotsOK=yes"},
990         {Opt_err, NULL}
991 };
992 static const match_table_t vfat_tokens = {
993         {Opt_charset, "iocharset=%s"},
994         {Opt_shortname_lower, "shortname=lower"},
995         {Opt_shortname_win95, "shortname=win95"},
996         {Opt_shortname_winnt, "shortname=winnt"},
997         {Opt_shortname_mixed, "shortname=mixed"},
998         {Opt_utf8_no, "utf8=0"},                /* 0 or no or false */
999         {Opt_utf8_no, "utf8=no"},
1000         {Opt_utf8_no, "utf8=false"},
1001         {Opt_utf8_yes, "utf8=1"},               /* empty or 1 or yes or true */
1002         {Opt_utf8_yes, "utf8=yes"},
1003         {Opt_utf8_yes, "utf8=true"},
1004         {Opt_utf8_yes, "utf8"},
1005         {Opt_uni_xl_no, "uni_xlate=0"},         /* 0 or no or false */
1006         {Opt_uni_xl_no, "uni_xlate=no"},
1007         {Opt_uni_xl_no, "uni_xlate=false"},
1008         {Opt_uni_xl_yes, "uni_xlate=1"},        /* empty or 1 or yes or true */
1009         {Opt_uni_xl_yes, "uni_xlate=yes"},
1010         {Opt_uni_xl_yes, "uni_xlate=true"},
1011         {Opt_uni_xl_yes, "uni_xlate"},
1012         {Opt_nonumtail_no, "nonumtail=0"},      /* 0 or no or false */
1013         {Opt_nonumtail_no, "nonumtail=no"},
1014         {Opt_nonumtail_no, "nonumtail=false"},
1015         {Opt_nonumtail_yes, "nonumtail=1"},     /* empty or 1 or yes or true */
1016         {Opt_nonumtail_yes, "nonumtail=yes"},
1017         {Opt_nonumtail_yes, "nonumtail=true"},
1018         {Opt_nonumtail_yes, "nonumtail"},
1019         {Opt_rodir, "rodir"},
1020         {Opt_err, NULL}
1021 };
1022
1023 static int parse_options(struct super_block *sb, char *options, int is_vfat,
1024                          int silent, int *debug, struct fat_mount_options *opts)
1025 {
1026         char *p;
1027         substring_t args[MAX_OPT_ARGS];
1028         int option;
1029         char *iocharset;
1030
1031         opts->isvfat = is_vfat;
1032
1033         opts->fs_uid = current_uid();
1034         opts->fs_gid = current_gid();
1035         opts->fs_fmask = opts->fs_dmask = current_umask();
1036         opts->allow_utime = -1;
1037         opts->codepage = fat_default_codepage;
1038         opts->iocharset = fat_default_iocharset;
1039         if (is_vfat) {
1040                 opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1041                 opts->rodir = 0;
1042         } else {
1043                 opts->shortname = 0;
1044                 opts->rodir = 1;
1045         }
1046         opts->name_check = 'n';
1047         opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
1048         opts->utf8 = opts->unicode_xlate = 0;
1049         opts->numtail = 1;
1050         opts->usefree = opts->nocase = 0;
1051         opts->tz_set = 0;
1052         opts->nfs = 0;
1053         opts->errors = FAT_ERRORS_RO;
1054         *debug = 0;
1055
1056         if (!options)
1057                 goto out;
1058
1059         while ((p = strsep(&options, ",")) != NULL) {
1060                 int token;
1061                 if (!*p)
1062                         continue;
1063
1064                 token = match_token(p, fat_tokens, args);
1065                 if (token == Opt_err) {
1066                         if (is_vfat)
1067                                 token = match_token(p, vfat_tokens, args);
1068                         else
1069                                 token = match_token(p, msdos_tokens, args);
1070                 }
1071                 switch (token) {
1072                 case Opt_check_s:
1073                         opts->name_check = 's';
1074                         break;
1075                 case Opt_check_r:
1076                         opts->name_check = 'r';
1077                         break;
1078                 case Opt_check_n:
1079                         opts->name_check = 'n';
1080                         break;
1081                 case Opt_usefree:
1082                         opts->usefree = 1;
1083                         break;
1084                 case Opt_nocase:
1085                         if (!is_vfat)
1086                                 opts->nocase = 1;
1087                         else {
1088                                 /* for backward compatibility */
1089                                 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1090                                         | VFAT_SFN_CREATE_WIN95;
1091                         }
1092                         break;
1093                 case Opt_quiet:
1094                         opts->quiet = 1;
1095                         break;
1096                 case Opt_showexec:
1097                         opts->showexec = 1;
1098                         break;
1099                 case Opt_debug:
1100                         *debug = 1;
1101                         break;
1102                 case Opt_immutable:
1103                         opts->sys_immutable = 1;
1104                         break;
1105                 case Opt_uid:
1106                         if (match_int(&args[0], &option))
1107                                 return -EINVAL;
1108                         opts->fs_uid = make_kuid(current_user_ns(), option);
1109                         if (!uid_valid(opts->fs_uid))
1110                                 return -EINVAL;
1111                         break;
1112                 case Opt_gid:
1113                         if (match_int(&args[0], &option))
1114                                 return -EINVAL;
1115                         opts->fs_gid = make_kgid(current_user_ns(), option);
1116                         if (!gid_valid(opts->fs_gid))
1117                                 return -EINVAL;
1118                         break;
1119                 case Opt_umask:
1120                         if (match_octal(&args[0], &option))
1121                                 return -EINVAL;
1122                         opts->fs_fmask = opts->fs_dmask = option;
1123                         break;
1124                 case Opt_dmask:
1125                         if (match_octal(&args[0], &option))
1126                                 return -EINVAL;
1127                         opts->fs_dmask = option;
1128                         break;
1129                 case Opt_fmask:
1130                         if (match_octal(&args[0], &option))
1131                                 return -EINVAL;
1132                         opts->fs_fmask = option;
1133                         break;
1134                 case Opt_allow_utime:
1135                         if (match_octal(&args[0], &option))
1136                                 return -EINVAL;
1137                         opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1138                         break;
1139                 case Opt_codepage:
1140                         if (match_int(&args[0], &option))
1141                                 return -EINVAL;
1142                         opts->codepage = option;
1143                         break;
1144                 case Opt_flush:
1145                         opts->flush = 1;
1146                         break;
1147                 case Opt_time_offset:
1148                         if (match_int(&args[0], &option))
1149                                 return -EINVAL;
1150                         if (option < -12 * 60 || option > 12 * 60)
1151                                 return -EINVAL;
1152                         opts->tz_set = 1;
1153                         opts->time_offset = option;
1154                         break;
1155                 case Opt_tz_utc:
1156                         opts->tz_set = 1;
1157                         opts->time_offset = 0;
1158                         break;
1159                 case Opt_err_cont:
1160                         opts->errors = FAT_ERRORS_CONT;
1161                         break;
1162                 case Opt_err_panic:
1163                         opts->errors = FAT_ERRORS_PANIC;
1164                         break;
1165                 case Opt_err_ro:
1166                         opts->errors = FAT_ERRORS_RO;
1167                         break;
1168                 case Opt_nfs_stale_rw:
1169                         opts->nfs = FAT_NFS_STALE_RW;
1170                         break;
1171                 case Opt_nfs_nostale_ro:
1172                         opts->nfs = FAT_NFS_NOSTALE_RO;
1173                         break;
1174                 case Opt_dos1xfloppy:
1175                         opts->dos1xfloppy = 1;
1176                         break;
1177
1178                 /* msdos specific */
1179                 case Opt_dots:
1180                         opts->dotsOK = 1;
1181                         break;
1182                 case Opt_nodots:
1183                         opts->dotsOK = 0;
1184                         break;
1185
1186                 /* vfat specific */
1187                 case Opt_charset:
1188                         if (opts->iocharset != fat_default_iocharset)
1189                                 kfree(opts->iocharset);
1190                         iocharset = match_strdup(&args[0]);
1191                         if (!iocharset)
1192                                 return -ENOMEM;
1193                         opts->iocharset = iocharset;
1194                         break;
1195                 case Opt_shortname_lower:
1196                         opts->shortname = VFAT_SFN_DISPLAY_LOWER
1197                                         | VFAT_SFN_CREATE_WIN95;
1198                         break;
1199                 case Opt_shortname_win95:
1200                         opts->shortname = VFAT_SFN_DISPLAY_WIN95
1201                                         | VFAT_SFN_CREATE_WIN95;
1202                         break;
1203                 case Opt_shortname_winnt:
1204                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1205                                         | VFAT_SFN_CREATE_WINNT;
1206                         break;
1207                 case Opt_shortname_mixed:
1208                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1209                                         | VFAT_SFN_CREATE_WIN95;
1210                         break;
1211                 case Opt_utf8_no:               /* 0 or no or false */
1212                         opts->utf8 = 0;
1213                         break;
1214                 case Opt_utf8_yes:              /* empty or 1 or yes or true */
1215                         opts->utf8 = 1;
1216                         break;
1217                 case Opt_uni_xl_no:             /* 0 or no or false */
1218                         opts->unicode_xlate = 0;
1219                         break;
1220                 case Opt_uni_xl_yes:            /* empty or 1 or yes or true */
1221                         opts->unicode_xlate = 1;
1222                         break;
1223                 case Opt_nonumtail_no:          /* 0 or no or false */
1224                         opts->numtail = 1;      /* negated option */
1225                         break;
1226                 case Opt_nonumtail_yes:         /* empty or 1 or yes or true */
1227                         opts->numtail = 0;      /* negated option */
1228                         break;
1229                 case Opt_rodir:
1230                         opts->rodir = 1;
1231                         break;
1232                 case Opt_discard:
1233                         opts->discard = 1;
1234                         break;
1235
1236                 /* obsolete mount options */
1237                 case Opt_obsolete:
1238                         fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1239                                "not supported now", p);
1240                         break;
1241                 /* unknown option */
1242                 default:
1243                         if (!silent) {
1244                                 fat_msg(sb, KERN_ERR,
1245                                        "Unrecognized mount option \"%s\" "
1246                                        "or missing value", p);
1247                         }
1248                         return -EINVAL;
1249                 }
1250         }
1251
1252 out:
1253         /* UTF-8 doesn't provide FAT semantics */
1254         if (!strcmp(opts->iocharset, "utf8")) {
1255                 fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1256                        " for FAT filesystems, filesystem will be "
1257                        "case sensitive!");
1258         }
1259
1260         /* If user doesn't specify allow_utime, it's initialized from dmask. */
1261         if (opts->allow_utime == (unsigned short)-1)
1262                 opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1263         if (opts->unicode_xlate)
1264                 opts->utf8 = 0;
1265         if (opts->nfs == FAT_NFS_NOSTALE_RO) {
1266                 sb->s_flags |= MS_RDONLY;
1267                 sb->s_export_op = &fat_export_ops_nostale;
1268         }
1269
1270         return 0;
1271 }
1272
1273 static int fat_read_root(struct inode *inode)
1274 {
1275         struct super_block *sb = inode->i_sb;
1276         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1277         int error;
1278
1279         MSDOS_I(inode)->i_pos = MSDOS_ROOT_INO;
1280         inode->i_uid = sbi->options.fs_uid;
1281         inode->i_gid = sbi->options.fs_gid;
1282         inode->i_version++;
1283         inode->i_generation = 0;
1284         inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1285         inode->i_op = sbi->dir_ops;
1286         inode->i_fop = &fat_dir_operations;
1287         if (sbi->fat_bits == 32) {
1288                 MSDOS_I(inode)->i_start = sbi->root_cluster;
1289                 error = fat_calc_dir_size(inode);
1290                 if (error < 0)
1291                         return error;
1292         } else {
1293                 MSDOS_I(inode)->i_start = 0;
1294                 inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1295         }
1296         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1297                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1298         MSDOS_I(inode)->i_logstart = 0;
1299         MSDOS_I(inode)->mmu_private = inode->i_size;
1300         MSDOS_I(inode)->i_disksize = inode->i_size;
1301
1302         fat_save_attrs(inode, ATTR_DIR);
1303         inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1304         inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1305         set_nlink(inode, fat_subdirs(inode)+2);
1306
1307         return 0;
1308 }
1309
1310 static unsigned long calc_fat_clusters(struct super_block *sb)
1311 {
1312         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1313
1314         /* Divide first to avoid overflow */
1315         if (sbi->fat_bits != 12) {
1316                 unsigned long ent_per_sec = sb->s_blocksize * 8 / sbi->fat_bits;
1317                 return ent_per_sec * sbi->fat_length;
1318         }
1319
1320         return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1321 }
1322
1323 static bool fat_bpb_is_zero(struct fat_boot_sector *b)
1324 {
1325         if (get_unaligned_le16(&b->sector_size))
1326                 return false;
1327         if (b->sec_per_clus)
1328                 return false;
1329         if (b->reserved)
1330                 return false;
1331         if (b->fats)
1332                 return false;
1333         if (get_unaligned_le16(&b->dir_entries))
1334                 return false;
1335         if (get_unaligned_le16(&b->sectors))
1336                 return false;
1337         if (b->media)
1338                 return false;
1339         if (b->fat_length)
1340                 return false;
1341         if (b->secs_track)
1342                 return false;
1343         if (b->heads)
1344                 return false;
1345         return true;
1346 }
1347
1348 static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
1349         int silent, struct fat_bios_param_block *bpb)
1350 {
1351         int error = -EINVAL;
1352
1353         /* Read in BPB ... */
1354         memset(bpb, 0, sizeof(*bpb));
1355         bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
1356         bpb->fat_sec_per_clus = b->sec_per_clus;
1357         bpb->fat_reserved = le16_to_cpu(b->reserved);
1358         bpb->fat_fats = b->fats;
1359         bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
1360         bpb->fat_sectors = get_unaligned_le16(&b->sectors);
1361         bpb->fat_fat_length = le16_to_cpu(b->fat_length);
1362         bpb->fat_total_sect = le32_to_cpu(b->total_sect);
1363
1364         bpb->fat16_state = b->fat16.state;
1365         bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
1366
1367         bpb->fat32_length = le32_to_cpu(b->fat32.length);
1368         bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
1369         bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
1370         bpb->fat32_state = b->fat32.state;
1371         bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
1372
1373         /* Validate this looks like a FAT filesystem BPB */
1374         if (!bpb->fat_reserved) {
1375                 if (!silent)
1376                         fat_msg(sb, KERN_ERR,
1377                                 "bogus number of reserved sectors");
1378                 goto out;
1379         }
1380         if (!bpb->fat_fats) {
1381                 if (!silent)
1382                         fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1383                 goto out;
1384         }
1385
1386         /*
1387          * Earlier we checked here that b->secs_track and b->head are nonzero,
1388          * but it turns out valid FAT filesystems can have zero there.
1389          */
1390
1391         if (!fat_valid_media(b->media)) {
1392                 if (!silent)
1393                         fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1394                                 (unsigned)b->media);
1395                 goto out;
1396         }
1397
1398         if (!is_power_of_2(bpb->fat_sector_size)
1399             || (bpb->fat_sector_size < 512)
1400             || (bpb->fat_sector_size > 4096)) {
1401                 if (!silent)
1402                         fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1403                                (unsigned)bpb->fat_sector_size);
1404                 goto out;
1405         }
1406
1407         if (!is_power_of_2(bpb->fat_sec_per_clus)) {
1408                 if (!silent)
1409                         fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1410                                 (unsigned)bpb->fat_sec_per_clus);
1411                 goto out;
1412         }
1413
1414         error = 0;
1415
1416 out:
1417         return error;
1418 }
1419
1420 static int fat_read_static_bpb(struct super_block *sb,
1421         struct fat_boot_sector *b, int silent,
1422         struct fat_bios_param_block *bpb)
1423 {
1424         static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
1425
1426         struct fat_floppy_defaults *fdefaults = NULL;
1427         int error = -EINVAL;
1428         sector_t bd_sects;
1429         unsigned i;
1430
1431         bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
1432
1433         /* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
1434         if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
1435                 if (!silent)
1436                         fat_msg(sb, KERN_ERR,
1437                                 "%s; no bootstrapping code", notdos1x);
1438                 goto out;
1439         }
1440
1441         /*
1442          * If any value in this region is non-zero, it isn't archaic
1443          * DOS.
1444          */
1445         if (!fat_bpb_is_zero(b)) {
1446                 if (!silent)
1447                         fat_msg(sb, KERN_ERR,
1448                                 "%s; DOS 2.x BPB is non-zero", notdos1x);
1449                 goto out;
1450         }
1451
1452         for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
1453                 if (floppy_defaults[i].nr_sectors == bd_sects) {
1454                         fdefaults = &floppy_defaults[i];
1455                         break;
1456                 }
1457         }
1458
1459         if (fdefaults == NULL) {
1460                 if (!silent)
1461                         fat_msg(sb, KERN_WARNING,
1462                                 "This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
1463                                 (u64)bd_sects);
1464                 goto out;
1465         }
1466
1467         if (!silent)
1468                 fat_msg(sb, KERN_INFO,
1469                         "This looks like a DOS 1.x volume; assuming default BPB values");
1470
1471         memset(bpb, 0, sizeof(*bpb));
1472         bpb->fat_sector_size = SECTOR_SIZE;
1473         bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
1474         bpb->fat_reserved = 1;
1475         bpb->fat_fats = 2;
1476         bpb->fat_dir_entries = fdefaults->dir_entries;
1477         bpb->fat_sectors = fdefaults->nr_sectors;
1478         bpb->fat_fat_length = fdefaults->fat_length;
1479
1480         error = 0;
1481
1482 out:
1483         return error;
1484 }
1485
1486 /*
1487  * Read the super block of an MS-DOS FS.
1488  */
1489 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1490                    void (*setup)(struct super_block *))
1491 {
1492         struct inode *root_inode = NULL, *fat_inode = NULL;
1493         struct inode *fsinfo_inode = NULL;
1494         struct buffer_head *bh;
1495         struct fat_bios_param_block bpb;
1496         struct msdos_sb_info *sbi;
1497         u16 logical_sector_size;
1498         u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1499         int debug;
1500         long error;
1501         char buf[50];
1502
1503         /*
1504          * GFP_KERNEL is ok here, because while we do hold the
1505          * supeblock lock, memory pressure can't call back into
1506          * the filesystem, since we're only just about to mount
1507          * it and have no inodes etc active!
1508          */
1509         sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1510         if (!sbi)
1511                 return -ENOMEM;
1512         sb->s_fs_info = sbi;
1513
1514         sb->s_flags |= MS_NODIRATIME;
1515         sb->s_magic = MSDOS_SUPER_MAGIC;
1516         sb->s_op = &fat_sops;
1517         sb->s_export_op = &fat_export_ops;
1518         mutex_init(&sbi->nfs_build_inode_lock);
1519         ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1520                              DEFAULT_RATELIMIT_BURST);
1521
1522         error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1523         if (error)
1524                 goto out_fail;
1525
1526         setup(sb); /* flavour-specific stuff that needs options */
1527
1528         error = -EIO;
1529         sb_min_blocksize(sb, 512);
1530         bh = sb_bread(sb, 0);
1531         if (bh == NULL) {
1532                 fat_msg(sb, KERN_ERR, "unable to read boot sector");
1533                 goto out_fail;
1534         }
1535
1536         error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
1537                 &bpb);
1538         if (error == -EINVAL && sbi->options.dos1xfloppy)
1539                 error = fat_read_static_bpb(sb,
1540                         (struct fat_boot_sector *)bh->b_data, silent, &bpb);
1541         brelse(bh);
1542
1543         if (error == -EINVAL)
1544                 goto out_invalid;
1545         else if (error)
1546                 goto out_fail;
1547
1548         logical_sector_size = bpb.fat_sector_size;
1549         sbi->sec_per_clus = bpb.fat_sec_per_clus;
1550
1551         error = -EIO;
1552         if (logical_sector_size < sb->s_blocksize) {
1553                 fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1554                        " (logical sector size = %u)", logical_sector_size);
1555                 goto out_fail;
1556         }
1557
1558         if (logical_sector_size > sb->s_blocksize) {
1559                 struct buffer_head *bh_resize;
1560
1561                 if (!sb_set_blocksize(sb, logical_sector_size)) {
1562                         fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1563                                logical_sector_size);
1564                         goto out_fail;
1565                 }
1566
1567                 /* Verify that the larger boot sector is fully readable */
1568                 bh_resize = sb_bread(sb, 0);
1569                 if (bh_resize == NULL) {
1570                         fat_msg(sb, KERN_ERR, "unable to read boot sector"
1571                                " (logical sector size = %lu)",
1572                                sb->s_blocksize);
1573                         goto out_fail;
1574                 }
1575                 brelse(bh_resize);
1576         }
1577
1578         mutex_init(&sbi->s_lock);
1579         sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1580         sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1581         sbi->fats = bpb.fat_fats;
1582         sbi->fat_bits = 0;              /* Don't know yet */
1583         sbi->fat_start = bpb.fat_reserved;
1584         sbi->fat_length = bpb.fat_fat_length;
1585         sbi->root_cluster = 0;
1586         sbi->free_clusters = -1;        /* Don't know yet */
1587         sbi->free_clus_valid = 0;
1588         sbi->prev_free = FAT_START_ENT;
1589         sb->s_maxbytes = 0xffffffff;
1590
1591         if (!sbi->fat_length && bpb.fat32_length) {
1592                 struct fat_boot_fsinfo *fsinfo;
1593                 struct buffer_head *fsinfo_bh;
1594
1595                 /* Must be FAT32 */
1596                 sbi->fat_bits = 32;
1597                 sbi->fat_length = bpb.fat32_length;
1598                 sbi->root_cluster = bpb.fat32_root_cluster;
1599
1600                 /* MC - if info_sector is 0, don't multiply by 0 */
1601                 sbi->fsinfo_sector = bpb.fat32_info_sector;
1602                 if (sbi->fsinfo_sector == 0)
1603                         sbi->fsinfo_sector = 1;
1604
1605                 fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1606                 if (fsinfo_bh == NULL) {
1607                         fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1608                                " (sector = %lu)", sbi->fsinfo_sector);
1609                         goto out_fail;
1610                 }
1611
1612                 fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1613                 if (!IS_FSINFO(fsinfo)) {
1614                         fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1615                                "0x%08x, 0x%08x (sector = %lu)",
1616                                le32_to_cpu(fsinfo->signature1),
1617                                le32_to_cpu(fsinfo->signature2),
1618                                sbi->fsinfo_sector);
1619                 } else {
1620                         if (sbi->options.usefree)
1621                                 sbi->free_clus_valid = 1;
1622                         sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1623                         sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1624                 }
1625
1626                 brelse(fsinfo_bh);
1627         }
1628
1629         /* interpret volume ID as a little endian 32 bit integer */
1630         if (sbi->fat_bits == 32)
1631                 sbi->vol_id = bpb.fat32_vol_id;
1632         else /* fat 16 or 12 */
1633                 sbi->vol_id = bpb.fat16_vol_id;
1634
1635         sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1636         sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1637
1638         sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1639         sbi->dir_entries = bpb.fat_dir_entries;
1640         if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1641                 if (!silent)
1642                         fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
1643                                " (%u)", sbi->dir_entries);
1644                 goto out_invalid;
1645         }
1646
1647         rootdir_sectors = sbi->dir_entries
1648                 * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1649         sbi->data_start = sbi->dir_start + rootdir_sectors;
1650         total_sectors = bpb.fat_sectors;
1651         if (total_sectors == 0)
1652                 total_sectors = bpb.fat_total_sect;
1653
1654         total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1655
1656         if (sbi->fat_bits != 32)
1657                 sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1658
1659         /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1660         if (sbi->fat_bits == 32)
1661                 sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
1662         else /* fat 16 or 12 */
1663                 sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
1664
1665         /* check that FAT table does not overflow */
1666         fat_clusters = calc_fat_clusters(sb);
1667         total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1668         if (total_clusters > MAX_FAT(sb)) {
1669                 if (!silent)
1670                         fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1671                                total_clusters);
1672                 goto out_invalid;
1673         }
1674
1675         sbi->max_cluster = total_clusters + FAT_START_ENT;
1676         /* check the free_clusters, it's not necessarily correct */
1677         if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1678                 sbi->free_clusters = -1;
1679         /* check the prev_free, it's not necessarily correct */
1680         sbi->prev_free %= sbi->max_cluster;
1681         if (sbi->prev_free < FAT_START_ENT)
1682                 sbi->prev_free = FAT_START_ENT;
1683
1684         /* set up enough so that it can read an inode */
1685         fat_hash_init(sb);
1686         dir_hash_init(sb);
1687         fat_ent_access_init(sb);
1688
1689         /*
1690          * The low byte of FAT's first entry must have same value with
1691          * media-field.  But in real world, too many devices is
1692          * writing wrong value.  So, removed that validity check.
1693          *
1694          * if (FAT_FIRST_ENT(sb, media) != first)
1695          */
1696
1697         error = -EINVAL;
1698         sprintf(buf, "cp%d", sbi->options.codepage);
1699         sbi->nls_disk = load_nls(buf);
1700         if (!sbi->nls_disk) {
1701                 fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1702                 goto out_fail;
1703         }
1704
1705         /* FIXME: utf8 is using iocharset for upper/lower conversion */
1706         if (sbi->options.isvfat) {
1707                 sbi->nls_io = load_nls(sbi->options.iocharset);
1708                 if (!sbi->nls_io) {
1709                         fat_msg(sb, KERN_ERR, "IO charset %s not found",
1710                                sbi->options.iocharset);
1711                         goto out_fail;
1712                 }
1713         }
1714
1715         error = -ENOMEM;
1716         fat_inode = new_inode(sb);
1717         if (!fat_inode)
1718                 goto out_fail;
1719         MSDOS_I(fat_inode)->i_pos = 0;
1720         sbi->fat_inode = fat_inode;
1721
1722         fsinfo_inode = new_inode(sb);
1723         if (!fsinfo_inode)
1724                 goto out_fail;
1725         fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1726         sbi->fsinfo_inode = fsinfo_inode;
1727         insert_inode_hash(fsinfo_inode);
1728
1729         root_inode = new_inode(sb);
1730         if (!root_inode)
1731                 goto out_fail;
1732         root_inode->i_ino = MSDOS_ROOT_INO;
1733         root_inode->i_version = 1;
1734         error = fat_read_root(root_inode);
1735         if (error < 0) {
1736                 iput(root_inode);
1737                 goto out_fail;
1738         }
1739         error = -ENOMEM;
1740         insert_inode_hash(root_inode);
1741         fat_attach(root_inode, 0);
1742         sb->s_root = d_make_root(root_inode);
1743         if (!sb->s_root) {
1744                 fat_msg(sb, KERN_ERR, "get root inode failed");
1745                 goto out_fail;
1746         }
1747
1748         if (sbi->options.discard) {
1749                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1750                 if (!blk_queue_discard(q))
1751                         fat_msg(sb, KERN_WARNING,
1752                                         "mounting with \"discard\" option, but "
1753                                         "the device does not support discard");
1754         }
1755
1756         fat_set_state(sb, 1, 0);
1757         return 0;
1758
1759 out_invalid:
1760         error = -EINVAL;
1761         if (!silent)
1762                 fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1763
1764 out_fail:
1765         if (fsinfo_inode)
1766                 iput(fsinfo_inode);
1767         if (fat_inode)
1768                 iput(fat_inode);
1769         unload_nls(sbi->nls_io);
1770         unload_nls(sbi->nls_disk);
1771         if (sbi->options.iocharset != fat_default_iocharset)
1772                 kfree(sbi->options.iocharset);
1773         sb->s_fs_info = NULL;
1774         kfree(sbi);
1775         return error;
1776 }
1777
1778 EXPORT_SYMBOL_GPL(fat_fill_super);
1779
1780 /*
1781  * helper function for fat_flush_inodes.  This writes both the inode
1782  * and the file data blocks, waiting for in flight data blocks before
1783  * the start of the call.  It does not wait for any io started
1784  * during the call
1785  */
1786 static int writeback_inode(struct inode *inode)
1787 {
1788
1789         int ret;
1790
1791         /* if we used wait=1, sync_inode_metadata waits for the io for the
1792         * inode to finish.  So wait=0 is sent down to sync_inode_metadata
1793         * and filemap_fdatawrite is used for the data blocks
1794         */
1795         ret = sync_inode_metadata(inode, 0);
1796         if (!ret)
1797                 ret = filemap_fdatawrite(inode->i_mapping);
1798         return ret;
1799 }
1800
1801 /*
1802  * write data and metadata corresponding to i1 and i2.  The io is
1803  * started but we do not wait for any of it to finish.
1804  *
1805  * filemap_flush is used for the block device, so if there is a dirty
1806  * page for a block already in flight, we will not wait and start the
1807  * io over again
1808  */
1809 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1810 {
1811         int ret = 0;
1812         if (!MSDOS_SB(sb)->options.flush)
1813                 return 0;
1814         if (i1)
1815                 ret = writeback_inode(i1);
1816         if (!ret && i2)
1817                 ret = writeback_inode(i2);
1818         if (!ret) {
1819                 struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1820                 ret = filemap_flush(mapping);
1821         }
1822         return ret;
1823 }
1824 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1825
1826 static int __init init_fat_fs(void)
1827 {
1828         int err;
1829
1830         err = fat_cache_init();
1831         if (err)
1832                 return err;
1833
1834         err = fat_init_inodecache();
1835         if (err)
1836                 goto failed;
1837
1838         return 0;
1839
1840 failed:
1841         fat_cache_destroy();
1842         return err;
1843 }
1844
1845 static void __exit exit_fat_fs(void)
1846 {
1847         fat_cache_destroy();
1848         fat_destroy_inodecache();
1849 }
1850
1851 module_init(init_fat_fs)
1852 module_exit(exit_fat_fs)
1853
1854 MODULE_LICENSE("GPL");