]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/fat/inode.c
Merge remote-tracking branch 'livepatching/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/vfs.h>
23 #include <linux/parser.h>
24 #include <linux/uio.h>
25 #include <linux/writeback.h>
26 #include <linux/log2.h>
27 #include <linux/hash.h>
28 #include <linux/blkdev.h>
29 #include <asm/unaligned.h>
30 #include "fat.h"
31
32 #ifndef CONFIG_FAT_DEFAULT_IOCHARSET
33 /* if user don't select VFAT, this is undefined. */
34 #define CONFIG_FAT_DEFAULT_IOCHARSET    ""
35 #endif
36
37 #define KB_IN_SECTORS 2
38
39 /*
40  * A deserialized copy of the on-disk structure laid out in struct
41  * fat_boot_sector.
42  */
43 struct fat_bios_param_block {
44         u16     fat_sector_size;
45         u8      fat_sec_per_clus;
46         u16     fat_reserved;
47         u8      fat_fats;
48         u16     fat_dir_entries;
49         u16     fat_sectors;
50         u16     fat_fat_length;
51         u32     fat_total_sect;
52
53         u8      fat16_state;
54         u32     fat16_vol_id;
55
56         u32     fat32_length;
57         u32     fat32_root_cluster;
58         u16     fat32_info_sector;
59         u8      fat32_state;
60         u32     fat32_vol_id;
61 };
62
63 static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
64 static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
65
66 static struct fat_floppy_defaults {
67         unsigned nr_sectors;
68         unsigned sec_per_clus;
69         unsigned dir_entries;
70         unsigned media;
71         unsigned fat_length;
72 } floppy_defaults[] = {
73 {
74         .nr_sectors = 160 * KB_IN_SECTORS,
75         .sec_per_clus = 1,
76         .dir_entries = 64,
77         .media = 0xFE,
78         .fat_length = 1,
79 },
80 {
81         .nr_sectors = 180 * KB_IN_SECTORS,
82         .sec_per_clus = 1,
83         .dir_entries = 64,
84         .media = 0xFC,
85         .fat_length = 2,
86 },
87 {
88         .nr_sectors = 320 * KB_IN_SECTORS,
89         .sec_per_clus = 2,
90         .dir_entries = 112,
91         .media = 0xFF,
92         .fat_length = 1,
93 },
94 {
95         .nr_sectors = 360 * KB_IN_SECTORS,
96         .sec_per_clus = 2,
97         .dir_entries = 112,
98         .media = 0xFD,
99         .fat_length = 2,
100 },
101 };
102
103 static int fat_add_cluster(struct inode *inode)
104 {
105         int err, cluster;
106
107         err = fat_alloc_clusters(inode, &cluster, 1);
108         if (err)
109                 return err;
110         /* FIXME: this cluster should be added after data of this
111          * cluster is writed */
112         err = fat_chain_add(inode, cluster, 1);
113         if (err)
114                 fat_free_clusters(inode, cluster);
115         return err;
116 }
117
118 static inline int __fat_get_block(struct inode *inode, sector_t iblock,
119                                   unsigned long *max_blocks,
120                                   struct buffer_head *bh_result, int create)
121 {
122         struct super_block *sb = inode->i_sb;
123         struct msdos_sb_info *sbi = MSDOS_SB(sb);
124         unsigned long mapped_blocks;
125         sector_t phys;
126         int err, offset;
127
128         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
129         if (err)
130                 return err;
131         if (phys) {
132                 map_bh(bh_result, sb, phys);
133                 *max_blocks = min(mapped_blocks, *max_blocks);
134                 return 0;
135         }
136         if (!create)
137                 return 0;
138
139         if (iblock != MSDOS_I(inode)->mmu_private >> sb->s_blocksize_bits) {
140                 fat_fs_error(sb, "corrupted file size (i_pos %lld, %lld)",
141                         MSDOS_I(inode)->i_pos, MSDOS_I(inode)->mmu_private);
142                 return -EIO;
143         }
144
145         offset = (unsigned long)iblock & (sbi->sec_per_clus - 1);
146         if (!offset) {
147                 /* TODO: multiple cluster allocation would be desirable. */
148                 err = fat_add_cluster(inode);
149                 if (err)
150                         return err;
151         }
152         /* available blocks on this cluster */
153         mapped_blocks = sbi->sec_per_clus - offset;
154
155         *max_blocks = min(mapped_blocks, *max_blocks);
156         MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
157
158         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
159         if (err)
160                 return err;
161
162         BUG_ON(!phys);
163         BUG_ON(*max_blocks != mapped_blocks);
164         set_buffer_new(bh_result);
165         map_bh(bh_result, sb, phys);
166
167         return 0;
168 }
169
170 static int fat_get_block(struct inode *inode, sector_t iblock,
171                          struct buffer_head *bh_result, int create)
172 {
173         struct super_block *sb = inode->i_sb;
174         unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
175         int err;
176
177         err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
178         if (err)
179                 return err;
180         bh_result->b_size = max_blocks << sb->s_blocksize_bits;
181         return 0;
182 }
183
184 static int fat_writepage(struct page *page, struct writeback_control *wbc)
185 {
186         return block_write_full_page(page, fat_get_block, wbc);
187 }
188
189 static int fat_writepages(struct address_space *mapping,
190                           struct writeback_control *wbc)
191 {
192         return mpage_writepages(mapping, wbc, fat_get_block);
193 }
194
195 static int fat_readpage(struct file *file, struct page *page)
196 {
197         return mpage_readpage(page, fat_get_block);
198 }
199
200 static int fat_readpages(struct file *file, struct address_space *mapping,
201                          struct list_head *pages, unsigned nr_pages)
202 {
203         return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
204 }
205
206 static void fat_write_failed(struct address_space *mapping, loff_t to)
207 {
208         struct inode *inode = mapping->host;
209
210         if (to > inode->i_size) {
211                 truncate_pagecache(inode, inode->i_size);
212                 fat_truncate_blocks(inode, inode->i_size);
213         }
214 }
215
216 static int fat_write_begin(struct file *file, struct address_space *mapping,
217                         loff_t pos, unsigned len, unsigned flags,
218                         struct page **pagep, void **fsdata)
219 {
220         int err;
221
222         *pagep = NULL;
223         err = cont_write_begin(file, mapping, pos, len, flags,
224                                 pagep, fsdata, fat_get_block,
225                                 &MSDOS_I(mapping->host)->mmu_private);
226         if (err < 0)
227                 fat_write_failed(mapping, pos + len);
228         return err;
229 }
230
231 static int fat_write_end(struct file *file, struct address_space *mapping,
232                         loff_t pos, unsigned len, unsigned copied,
233                         struct page *pagep, void *fsdata)
234 {
235         struct inode *inode = mapping->host;
236         int err;
237         err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
238         if (err < len)
239                 fat_write_failed(mapping, pos + len);
240         if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
241                 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
242                 MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
243                 mark_inode_dirty(inode);
244         }
245         return err;
246 }
247
248 static ssize_t fat_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
249                              loff_t offset)
250 {
251         struct file *file = iocb->ki_filp;
252         struct address_space *mapping = file->f_mapping;
253         struct inode *inode = mapping->host;
254         size_t count = iov_iter_count(iter);
255         ssize_t ret;
256
257         if (iov_iter_rw(iter) == WRITE) {
258                 /*
259                  * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
260                  * so we need to update the ->mmu_private to block boundary.
261                  *
262                  * But we must fill the remaining area or hole by nul for
263                  * updating ->mmu_private.
264                  *
265                  * Return 0, and fallback to normal buffered write.
266                  */
267                 loff_t size = offset + count;
268                 if (MSDOS_I(inode)->mmu_private < size)
269                         return 0;
270         }
271
272         /*
273          * FAT need to use the DIO_LOCKING for avoiding the race
274          * condition of fat_get_block() and ->truncate().
275          */
276         ret = blockdev_direct_IO(iocb, inode, iter, offset, fat_get_block);
277         if (ret < 0 && iov_iter_rw(iter) == WRITE)
278                 fat_write_failed(mapping, offset + count);
279
280         return ret;
281 }
282
283 static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
284 {
285         sector_t blocknr;
286
287         /* fat_get_cluster() assumes the requested blocknr isn't truncated. */
288         down_read(&MSDOS_I(mapping->host)->truncate_lock);
289         blocknr = generic_block_bmap(mapping, block, fat_get_block);
290         up_read(&MSDOS_I(mapping->host)->truncate_lock);
291
292         return blocknr;
293 }
294
295 /*
296  * fat_block_truncate_page() zeroes out a mapping from file offset `from'
297  * up to the end of the block which corresponds to `from'.
298  * This is required during truncate to physically zeroout the tail end
299  * of that block so it doesn't yield old data if the file is later grown.
300  * Also, avoid causing failure from fsx for cases of "data past EOF"
301  */
302 int fat_block_truncate_page(struct inode *inode, loff_t from)
303 {
304         return block_truncate_page(inode->i_mapping, from, fat_get_block);
305 }
306
307 static const struct address_space_operations fat_aops = {
308         .readpage       = fat_readpage,
309         .readpages      = fat_readpages,
310         .writepage      = fat_writepage,
311         .writepages     = fat_writepages,
312         .write_begin    = fat_write_begin,
313         .write_end      = fat_write_end,
314         .direct_IO      = fat_direct_IO,
315         .bmap           = _fat_bmap
316 };
317
318 /*
319  * New FAT inode stuff. We do the following:
320  *      a) i_ino is constant and has nothing with on-disk location.
321  *      b) FAT manages its own cache of directory entries.
322  *      c) *This* cache is indexed by on-disk location.
323  *      d) inode has an associated directory entry, all right, but
324  *              it may be unhashed.
325  *      e) currently entries are stored within struct inode. That should
326  *              change.
327  *      f) we deal with races in the following way:
328  *              1. readdir() and lookup() do FAT-dir-cache lookup.
329  *              2. rename() unhashes the F-d-c entry and rehashes it in
330  *                      a new place.
331  *              3. unlink() and rmdir() unhash F-d-c entry.
332  *              4. fat_write_inode() checks whether the thing is unhashed.
333  *                      If it is we silently return. If it isn't we do bread(),
334  *                      check if the location is still valid and retry if it
335  *                      isn't. Otherwise we do changes.
336  *              5. Spinlock is used to protect hash/unhash/location check/lookup
337  *              6. fat_evict_inode() unhashes the F-d-c entry.
338  *              7. lookup() and readdir() do igrab() if they find a F-d-c entry
339  *                      and consider negative result as cache miss.
340  */
341
342 static void fat_hash_init(struct super_block *sb)
343 {
344         struct msdos_sb_info *sbi = MSDOS_SB(sb);
345         int i;
346
347         spin_lock_init(&sbi->inode_hash_lock);
348         for (i = 0; i < FAT_HASH_SIZE; i++)
349                 INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
350 }
351
352 static inline unsigned long fat_hash(loff_t i_pos)
353 {
354         return hash_32(i_pos, FAT_HASH_BITS);
355 }
356
357 static void dir_hash_init(struct super_block *sb)
358 {
359         struct msdos_sb_info *sbi = MSDOS_SB(sb);
360         int i;
361
362         spin_lock_init(&sbi->dir_hash_lock);
363         for (i = 0; i < FAT_HASH_SIZE; i++)
364                 INIT_HLIST_HEAD(&sbi->dir_hashtable[i]);
365 }
366
367 void fat_attach(struct inode *inode, loff_t i_pos)
368 {
369         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
370
371         if (inode->i_ino != MSDOS_ROOT_INO) {
372                 struct hlist_head *head =   sbi->inode_hashtable
373                                           + fat_hash(i_pos);
374
375                 spin_lock(&sbi->inode_hash_lock);
376                 MSDOS_I(inode)->i_pos = i_pos;
377                 hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
378                 spin_unlock(&sbi->inode_hash_lock);
379         }
380
381         /* If NFS support is enabled, cache the mapping of start cluster
382          * to directory inode. This is used during reconnection of
383          * dentries to the filesystem root.
384          */
385         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
386                 struct hlist_head *d_head = sbi->dir_hashtable;
387                 d_head += fat_dir_hash(MSDOS_I(inode)->i_logstart);
388
389                 spin_lock(&sbi->dir_hash_lock);
390                 hlist_add_head(&MSDOS_I(inode)->i_dir_hash, d_head);
391                 spin_unlock(&sbi->dir_hash_lock);
392         }
393 }
394 EXPORT_SYMBOL_GPL(fat_attach);
395
396 void fat_detach(struct inode *inode)
397 {
398         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
399         spin_lock(&sbi->inode_hash_lock);
400         MSDOS_I(inode)->i_pos = 0;
401         hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
402         spin_unlock(&sbi->inode_hash_lock);
403
404         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
405                 spin_lock(&sbi->dir_hash_lock);
406                 hlist_del_init(&MSDOS_I(inode)->i_dir_hash);
407                 spin_unlock(&sbi->dir_hash_lock);
408         }
409 }
410 EXPORT_SYMBOL_GPL(fat_detach);
411
412 struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
413 {
414         struct msdos_sb_info *sbi = MSDOS_SB(sb);
415         struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
416         struct msdos_inode_info *i;
417         struct inode *inode = NULL;
418
419         spin_lock(&sbi->inode_hash_lock);
420         hlist_for_each_entry(i, head, i_fat_hash) {
421                 BUG_ON(i->vfs_inode.i_sb != sb);
422                 if (i->i_pos != i_pos)
423                         continue;
424                 inode = igrab(&i->vfs_inode);
425                 if (inode)
426                         break;
427         }
428         spin_unlock(&sbi->inode_hash_lock);
429         return inode;
430 }
431
432 static int is_exec(unsigned char *extension)
433 {
434         unsigned char exe_extensions[] = "EXECOMBAT", *walk;
435
436         for (walk = exe_extensions; *walk; walk += 3)
437                 if (!strncmp(extension, walk, 3))
438                         return 1;
439         return 0;
440 }
441
442 static int fat_calc_dir_size(struct inode *inode)
443 {
444         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
445         int ret, fclus, dclus;
446
447         inode->i_size = 0;
448         if (MSDOS_I(inode)->i_start == 0)
449                 return 0;
450
451         ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
452         if (ret < 0)
453                 return ret;
454         inode->i_size = (fclus + 1) << sbi->cluster_bits;
455
456         return 0;
457 }
458
459 /* doesn't deal with root inode */
460 int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
461 {
462         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
463         int error;
464
465         MSDOS_I(inode)->i_pos = 0;
466         inode->i_uid = sbi->options.fs_uid;
467         inode->i_gid = sbi->options.fs_gid;
468         inode->i_version++;
469         inode->i_generation = get_seconds();
470
471         if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
472                 inode->i_generation &= ~1;
473                 inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
474                 inode->i_op = sbi->dir_ops;
475                 inode->i_fop = &fat_dir_operations;
476
477                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
478                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
479                 error = fat_calc_dir_size(inode);
480                 if (error < 0)
481                         return error;
482                 MSDOS_I(inode)->mmu_private = inode->i_size;
483
484                 set_nlink(inode, fat_subdirs(inode));
485         } else { /* not a directory */
486                 inode->i_generation |= 1;
487                 inode->i_mode = fat_make_mode(sbi, de->attr,
488                         ((sbi->options.showexec && !is_exec(de->name + 8))
489                          ? S_IRUGO|S_IWUGO : S_IRWXUGO));
490                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
491
492                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
493                 inode->i_size = le32_to_cpu(de->size);
494                 inode->i_op = &fat_file_inode_operations;
495                 inode->i_fop = &fat_file_operations;
496                 inode->i_mapping->a_ops = &fat_aops;
497                 MSDOS_I(inode)->mmu_private = inode->i_size;
498         }
499         if (de->attr & ATTR_SYS) {
500                 if (sbi->options.sys_immutable)
501                         inode->i_flags |= S_IMMUTABLE;
502         }
503         fat_save_attrs(inode, de->attr);
504
505         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
506                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
507
508         fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
509         if (sbi->options.isvfat) {
510                 fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
511                                   de->cdate, de->ctime_cs);
512                 fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
513         } else
514                 inode->i_ctime = inode->i_atime = inode->i_mtime;
515
516         return 0;
517 }
518
519 static inline void fat_lock_build_inode(struct msdos_sb_info *sbi)
520 {
521         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
522                 mutex_lock(&sbi->nfs_build_inode_lock);
523 }
524
525 static inline void fat_unlock_build_inode(struct msdos_sb_info *sbi)
526 {
527         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
528                 mutex_unlock(&sbi->nfs_build_inode_lock);
529 }
530
531 struct inode *fat_build_inode(struct super_block *sb,
532                         struct msdos_dir_entry *de, loff_t i_pos)
533 {
534         struct inode *inode;
535         int err;
536
537         fat_lock_build_inode(MSDOS_SB(sb));
538         inode = fat_iget(sb, i_pos);
539         if (inode)
540                 goto out;
541         inode = new_inode(sb);
542         if (!inode) {
543                 inode = ERR_PTR(-ENOMEM);
544                 goto out;
545         }
546         inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
547         inode->i_version = 1;
548         err = fat_fill_inode(inode, de);
549         if (err) {
550                 iput(inode);
551                 inode = ERR_PTR(err);
552                 goto out;
553         }
554         fat_attach(inode, i_pos);
555         insert_inode_hash(inode);
556 out:
557         fat_unlock_build_inode(MSDOS_SB(sb));
558         return inode;
559 }
560
561 EXPORT_SYMBOL_GPL(fat_build_inode);
562
563 static void fat_evict_inode(struct inode *inode)
564 {
565         truncate_inode_pages_final(&inode->i_data);
566         if (!inode->i_nlink) {
567                 inode->i_size = 0;
568                 fat_truncate_blocks(inode, 0);
569         }
570         invalidate_inode_buffers(inode);
571         clear_inode(inode);
572         fat_cache_inval_inode(inode);
573         fat_detach(inode);
574 }
575
576 static void fat_set_state(struct super_block *sb,
577                         unsigned int set, unsigned int force)
578 {
579         struct buffer_head *bh;
580         struct fat_boot_sector *b;
581         struct msdos_sb_info *sbi = MSDOS_SB(sb);
582
583         /* do not change any thing if mounted read only */
584         if ((sb->s_flags & MS_RDONLY) && !force)
585                 return;
586
587         /* do not change state if fs was dirty */
588         if (sbi->dirty) {
589                 /* warn only on set (mount). */
590                 if (set)
591                         fat_msg(sb, KERN_WARNING, "Volume was not properly "
592                                 "unmounted. Some data may be corrupt. "
593                                 "Please run fsck.");
594                 return;
595         }
596
597         bh = sb_bread(sb, 0);
598         if (bh == NULL) {
599                 fat_msg(sb, KERN_ERR, "unable to read boot sector "
600                         "to mark fs as dirty");
601                 return;
602         }
603
604         b = (struct fat_boot_sector *) bh->b_data;
605
606         if (sbi->fat_bits == 32) {
607                 if (set)
608                         b->fat32.state |= FAT_STATE_DIRTY;
609                 else
610                         b->fat32.state &= ~FAT_STATE_DIRTY;
611         } else /* fat 16 and 12 */ {
612                 if (set)
613                         b->fat16.state |= FAT_STATE_DIRTY;
614                 else
615                         b->fat16.state &= ~FAT_STATE_DIRTY;
616         }
617
618         mark_buffer_dirty(bh);
619         sync_dirty_buffer(bh);
620         brelse(bh);
621 }
622
623 static void delayed_free(struct rcu_head *p)
624 {
625         struct msdos_sb_info *sbi = container_of(p, struct msdos_sb_info, rcu);
626         unload_nls(sbi->nls_disk);
627         unload_nls(sbi->nls_io);
628         if (sbi->options.iocharset != fat_default_iocharset)
629                 kfree(sbi->options.iocharset);
630         kfree(sbi);
631 }
632
633 static void fat_put_super(struct super_block *sb)
634 {
635         struct msdos_sb_info *sbi = MSDOS_SB(sb);
636
637         fat_set_state(sb, 0, 0);
638
639         iput(sbi->fsinfo_inode);
640         iput(sbi->fat_inode);
641
642         call_rcu(&sbi->rcu, delayed_free);
643 }
644
645 static struct kmem_cache *fat_inode_cachep;
646
647 static struct inode *fat_alloc_inode(struct super_block *sb)
648 {
649         struct msdos_inode_info *ei;
650         ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
651         if (!ei)
652                 return NULL;
653
654         init_rwsem(&ei->truncate_lock);
655         return &ei->vfs_inode;
656 }
657
658 static void fat_i_callback(struct rcu_head *head)
659 {
660         struct inode *inode = container_of(head, struct inode, i_rcu);
661         kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
662 }
663
664 static void fat_destroy_inode(struct inode *inode)
665 {
666         call_rcu(&inode->i_rcu, fat_i_callback);
667 }
668
669 static void init_once(void *foo)
670 {
671         struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
672
673         spin_lock_init(&ei->cache_lru_lock);
674         ei->nr_caches = 0;
675         ei->cache_valid_id = FAT_CACHE_VALID + 1;
676         INIT_LIST_HEAD(&ei->cache_lru);
677         INIT_HLIST_NODE(&ei->i_fat_hash);
678         INIT_HLIST_NODE(&ei->i_dir_hash);
679         inode_init_once(&ei->vfs_inode);
680 }
681
682 static int __init fat_init_inodecache(void)
683 {
684         fat_inode_cachep = kmem_cache_create("fat_inode_cache",
685                                              sizeof(struct msdos_inode_info),
686                                              0, (SLAB_RECLAIM_ACCOUNT|
687                                                 SLAB_MEM_SPREAD),
688                                              init_once);
689         if (fat_inode_cachep == NULL)
690                 return -ENOMEM;
691         return 0;
692 }
693
694 static void __exit fat_destroy_inodecache(void)
695 {
696         /*
697          * Make sure all delayed rcu free inodes are flushed before we
698          * destroy cache.
699          */
700         rcu_barrier();
701         kmem_cache_destroy(fat_inode_cachep);
702 }
703
704 static int fat_remount(struct super_block *sb, int *flags, char *data)
705 {
706         int new_rdonly;
707         struct msdos_sb_info *sbi = MSDOS_SB(sb);
708         *flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
709
710         sync_filesystem(sb);
711
712         /* make sure we update state on remount. */
713         new_rdonly = *flags & MS_RDONLY;
714         if (new_rdonly != (sb->s_flags & MS_RDONLY)) {
715                 if (new_rdonly)
716                         fat_set_state(sb, 0, 0);
717                 else
718                         fat_set_state(sb, 1, 1);
719         }
720         return 0;
721 }
722
723 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
724 {
725         struct super_block *sb = dentry->d_sb;
726         struct msdos_sb_info *sbi = MSDOS_SB(sb);
727         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
728
729         /* If the count of free cluster is still unknown, counts it here. */
730         if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
731                 int err = fat_count_free_clusters(dentry->d_sb);
732                 if (err)
733                         return err;
734         }
735
736         buf->f_type = dentry->d_sb->s_magic;
737         buf->f_bsize = sbi->cluster_size;
738         buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
739         buf->f_bfree = sbi->free_clusters;
740         buf->f_bavail = sbi->free_clusters;
741         buf->f_fsid.val[0] = (u32)id;
742         buf->f_fsid.val[1] = (u32)(id >> 32);
743         buf->f_namelen =
744                 (sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
745
746         return 0;
747 }
748
749 static int __fat_write_inode(struct inode *inode, int wait)
750 {
751         struct super_block *sb = inode->i_sb;
752         struct msdos_sb_info *sbi = MSDOS_SB(sb);
753         struct buffer_head *bh;
754         struct msdos_dir_entry *raw_entry;
755         loff_t i_pos;
756         sector_t blocknr;
757         int err, offset;
758
759         if (inode->i_ino == MSDOS_ROOT_INO)
760                 return 0;
761
762 retry:
763         i_pos = fat_i_pos_read(sbi, inode);
764         if (!i_pos)
765                 return 0;
766
767         fat_get_blknr_offset(sbi, i_pos, &blocknr, &offset);
768         bh = sb_bread(sb, blocknr);
769         if (!bh) {
770                 fat_msg(sb, KERN_ERR, "unable to read inode block "
771                        "for updating (i_pos %lld)", i_pos);
772                 return -EIO;
773         }
774         spin_lock(&sbi->inode_hash_lock);
775         if (i_pos != MSDOS_I(inode)->i_pos) {
776                 spin_unlock(&sbi->inode_hash_lock);
777                 brelse(bh);
778                 goto retry;
779         }
780
781         raw_entry = &((struct msdos_dir_entry *) (bh->b_data))[offset];
782         if (S_ISDIR(inode->i_mode))
783                 raw_entry->size = 0;
784         else
785                 raw_entry->size = cpu_to_le32(inode->i_size);
786         raw_entry->attr = fat_make_attrs(inode);
787         fat_set_start(raw_entry, MSDOS_I(inode)->i_logstart);
788         fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
789                           &raw_entry->date, NULL);
790         if (sbi->options.isvfat) {
791                 __le16 atime;
792                 fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
793                                   &raw_entry->cdate, &raw_entry->ctime_cs);
794                 fat_time_unix2fat(sbi, &inode->i_atime, &atime,
795                                   &raw_entry->adate, NULL);
796         }
797         spin_unlock(&sbi->inode_hash_lock);
798         mark_buffer_dirty(bh);
799         err = 0;
800         if (wait)
801                 err = sync_dirty_buffer(bh);
802         brelse(bh);
803         return err;
804 }
805
806 static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
807 {
808         int err;
809
810         if (inode->i_ino == MSDOS_FSINFO_INO) {
811                 struct super_block *sb = inode->i_sb;
812
813                 mutex_lock(&MSDOS_SB(sb)->s_lock);
814                 err = fat_clusters_flush(sb);
815                 mutex_unlock(&MSDOS_SB(sb)->s_lock);
816         } else
817                 err = __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
818
819         return err;
820 }
821
822 int fat_sync_inode(struct inode *inode)
823 {
824         return __fat_write_inode(inode, 1);
825 }
826
827 EXPORT_SYMBOL_GPL(fat_sync_inode);
828
829 static int fat_show_options(struct seq_file *m, struct dentry *root);
830 static const struct super_operations fat_sops = {
831         .alloc_inode    = fat_alloc_inode,
832         .destroy_inode  = fat_destroy_inode,
833         .write_inode    = fat_write_inode,
834         .evict_inode    = fat_evict_inode,
835         .put_super      = fat_put_super,
836         .statfs         = fat_statfs,
837         .remount_fs     = fat_remount,
838
839         .show_options   = fat_show_options,
840 };
841
842 static int fat_show_options(struct seq_file *m, struct dentry *root)
843 {
844         struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
845         struct fat_mount_options *opts = &sbi->options;
846         int isvfat = opts->isvfat;
847
848         if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
849                 seq_printf(m, ",uid=%u",
850                                 from_kuid_munged(&init_user_ns, opts->fs_uid));
851         if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
852                 seq_printf(m, ",gid=%u",
853                                 from_kgid_munged(&init_user_ns, opts->fs_gid));
854         seq_printf(m, ",fmask=%04o", opts->fs_fmask);
855         seq_printf(m, ",dmask=%04o", opts->fs_dmask);
856         if (opts->allow_utime)
857                 seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
858         if (sbi->nls_disk)
859                 /* strip "cp" prefix from displayed option */
860                 seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
861         if (isvfat) {
862                 if (sbi->nls_io)
863                         seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
864
865                 switch (opts->shortname) {
866                 case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
867                         seq_puts(m, ",shortname=win95");
868                         break;
869                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
870                         seq_puts(m, ",shortname=winnt");
871                         break;
872                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
873                         seq_puts(m, ",shortname=mixed");
874                         break;
875                 case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
876                         seq_puts(m, ",shortname=lower");
877                         break;
878                 default:
879                         seq_puts(m, ",shortname=unknown");
880                         break;
881                 }
882         }
883         if (opts->name_check != 'n')
884                 seq_printf(m, ",check=%c", opts->name_check);
885         if (opts->usefree)
886                 seq_puts(m, ",usefree");
887         if (opts->quiet)
888                 seq_puts(m, ",quiet");
889         if (opts->showexec)
890                 seq_puts(m, ",showexec");
891         if (opts->sys_immutable)
892                 seq_puts(m, ",sys_immutable");
893         if (!isvfat) {
894                 if (opts->dotsOK)
895                         seq_puts(m, ",dotsOK=yes");
896                 if (opts->nocase)
897                         seq_puts(m, ",nocase");
898         } else {
899                 if (opts->utf8)
900                         seq_puts(m, ",utf8");
901                 if (opts->unicode_xlate)
902                         seq_puts(m, ",uni_xlate");
903                 if (!opts->numtail)
904                         seq_puts(m, ",nonumtail");
905                 if (opts->rodir)
906                         seq_puts(m, ",rodir");
907         }
908         if (opts->flush)
909                 seq_puts(m, ",flush");
910         if (opts->tz_set) {
911                 if (opts->time_offset)
912                         seq_printf(m, ",time_offset=%d", opts->time_offset);
913                 else
914                         seq_puts(m, ",tz=UTC");
915         }
916         if (opts->errors == FAT_ERRORS_CONT)
917                 seq_puts(m, ",errors=continue");
918         else if (opts->errors == FAT_ERRORS_PANIC)
919                 seq_puts(m, ",errors=panic");
920         else
921                 seq_puts(m, ",errors=remount-ro");
922         if (opts->nfs == FAT_NFS_NOSTALE_RO)
923                 seq_puts(m, ",nfs=nostale_ro");
924         else if (opts->nfs)
925                 seq_puts(m, ",nfs=stale_rw");
926         if (opts->discard)
927                 seq_puts(m, ",discard");
928         if (opts->dos1xfloppy)
929                 seq_puts(m, ",dos1xfloppy");
930
931         return 0;
932 }
933
934 enum {
935         Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
936         Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
937         Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
938         Opt_immutable, Opt_dots, Opt_nodots,
939         Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
940         Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
941         Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
942         Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
943         Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
944         Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
945 };
946
947 static const match_table_t fat_tokens = {
948         {Opt_check_r, "check=relaxed"},
949         {Opt_check_s, "check=strict"},
950         {Opt_check_n, "check=normal"},
951         {Opt_check_r, "check=r"},
952         {Opt_check_s, "check=s"},
953         {Opt_check_n, "check=n"},
954         {Opt_uid, "uid=%u"},
955         {Opt_gid, "gid=%u"},
956         {Opt_umask, "umask=%o"},
957         {Opt_dmask, "dmask=%o"},
958         {Opt_fmask, "fmask=%o"},
959         {Opt_allow_utime, "allow_utime=%o"},
960         {Opt_codepage, "codepage=%u"},
961         {Opt_usefree, "usefree"},
962         {Opt_nocase, "nocase"},
963         {Opt_quiet, "quiet"},
964         {Opt_showexec, "showexec"},
965         {Opt_debug, "debug"},
966         {Opt_immutable, "sys_immutable"},
967         {Opt_flush, "flush"},
968         {Opt_tz_utc, "tz=UTC"},
969         {Opt_time_offset, "time_offset=%d"},
970         {Opt_err_cont, "errors=continue"},
971         {Opt_err_panic, "errors=panic"},
972         {Opt_err_ro, "errors=remount-ro"},
973         {Opt_discard, "discard"},
974         {Opt_nfs_stale_rw, "nfs"},
975         {Opt_nfs_stale_rw, "nfs=stale_rw"},
976         {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
977         {Opt_dos1xfloppy, "dos1xfloppy"},
978         {Opt_obsolete, "conv=binary"},
979         {Opt_obsolete, "conv=text"},
980         {Opt_obsolete, "conv=auto"},
981         {Opt_obsolete, "conv=b"},
982         {Opt_obsolete, "conv=t"},
983         {Opt_obsolete, "conv=a"},
984         {Opt_obsolete, "fat=%u"},
985         {Opt_obsolete, "blocksize=%u"},
986         {Opt_obsolete, "cvf_format=%20s"},
987         {Opt_obsolete, "cvf_options=%100s"},
988         {Opt_obsolete, "posix"},
989         {Opt_err, NULL},
990 };
991 static const match_table_t msdos_tokens = {
992         {Opt_nodots, "nodots"},
993         {Opt_nodots, "dotsOK=no"},
994         {Opt_dots, "dots"},
995         {Opt_dots, "dotsOK=yes"},
996         {Opt_err, NULL}
997 };
998 static const match_table_t vfat_tokens = {
999         {Opt_charset, "iocharset=%s"},
1000         {Opt_shortname_lower, "shortname=lower"},
1001         {Opt_shortname_win95, "shortname=win95"},
1002         {Opt_shortname_winnt, "shortname=winnt"},
1003         {Opt_shortname_mixed, "shortname=mixed"},
1004         {Opt_utf8_no, "utf8=0"},                /* 0 or no or false */
1005         {Opt_utf8_no, "utf8=no"},
1006         {Opt_utf8_no, "utf8=false"},
1007         {Opt_utf8_yes, "utf8=1"},               /* empty or 1 or yes or true */
1008         {Opt_utf8_yes, "utf8=yes"},
1009         {Opt_utf8_yes, "utf8=true"},
1010         {Opt_utf8_yes, "utf8"},
1011         {Opt_uni_xl_no, "uni_xlate=0"},         /* 0 or no or false */
1012         {Opt_uni_xl_no, "uni_xlate=no"},
1013         {Opt_uni_xl_no, "uni_xlate=false"},
1014         {Opt_uni_xl_yes, "uni_xlate=1"},        /* empty or 1 or yes or true */
1015         {Opt_uni_xl_yes, "uni_xlate=yes"},
1016         {Opt_uni_xl_yes, "uni_xlate=true"},
1017         {Opt_uni_xl_yes, "uni_xlate"},
1018         {Opt_nonumtail_no, "nonumtail=0"},      /* 0 or no or false */
1019         {Opt_nonumtail_no, "nonumtail=no"},
1020         {Opt_nonumtail_no, "nonumtail=false"},
1021         {Opt_nonumtail_yes, "nonumtail=1"},     /* empty or 1 or yes or true */
1022         {Opt_nonumtail_yes, "nonumtail=yes"},
1023         {Opt_nonumtail_yes, "nonumtail=true"},
1024         {Opt_nonumtail_yes, "nonumtail"},
1025         {Opt_rodir, "rodir"},
1026         {Opt_err, NULL}
1027 };
1028
1029 static int parse_options(struct super_block *sb, char *options, int is_vfat,
1030                          int silent, int *debug, struct fat_mount_options *opts)
1031 {
1032         char *p;
1033         substring_t args[MAX_OPT_ARGS];
1034         int option;
1035         char *iocharset;
1036
1037         opts->isvfat = is_vfat;
1038
1039         opts->fs_uid = current_uid();
1040         opts->fs_gid = current_gid();
1041         opts->fs_fmask = opts->fs_dmask = current_umask();
1042         opts->allow_utime = -1;
1043         opts->codepage = fat_default_codepage;
1044         opts->iocharset = fat_default_iocharset;
1045         if (is_vfat) {
1046                 opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1047                 opts->rodir = 0;
1048         } else {
1049                 opts->shortname = 0;
1050                 opts->rodir = 1;
1051         }
1052         opts->name_check = 'n';
1053         opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
1054         opts->utf8 = opts->unicode_xlate = 0;
1055         opts->numtail = 1;
1056         opts->usefree = opts->nocase = 0;
1057         opts->tz_set = 0;
1058         opts->nfs = 0;
1059         opts->errors = FAT_ERRORS_RO;
1060         *debug = 0;
1061
1062         if (!options)
1063                 goto out;
1064
1065         while ((p = strsep(&options, ",")) != NULL) {
1066                 int token;
1067                 if (!*p)
1068                         continue;
1069
1070                 token = match_token(p, fat_tokens, args);
1071                 if (token == Opt_err) {
1072                         if (is_vfat)
1073                                 token = match_token(p, vfat_tokens, args);
1074                         else
1075                                 token = match_token(p, msdos_tokens, args);
1076                 }
1077                 switch (token) {
1078                 case Opt_check_s:
1079                         opts->name_check = 's';
1080                         break;
1081                 case Opt_check_r:
1082                         opts->name_check = 'r';
1083                         break;
1084                 case Opt_check_n:
1085                         opts->name_check = 'n';
1086                         break;
1087                 case Opt_usefree:
1088                         opts->usefree = 1;
1089                         break;
1090                 case Opt_nocase:
1091                         if (!is_vfat)
1092                                 opts->nocase = 1;
1093                         else {
1094                                 /* for backward compatibility */
1095                                 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1096                                         | VFAT_SFN_CREATE_WIN95;
1097                         }
1098                         break;
1099                 case Opt_quiet:
1100                         opts->quiet = 1;
1101                         break;
1102                 case Opt_showexec:
1103                         opts->showexec = 1;
1104                         break;
1105                 case Opt_debug:
1106                         *debug = 1;
1107                         break;
1108                 case Opt_immutable:
1109                         opts->sys_immutable = 1;
1110                         break;
1111                 case Opt_uid:
1112                         if (match_int(&args[0], &option))
1113                                 return -EINVAL;
1114                         opts->fs_uid = make_kuid(current_user_ns(), option);
1115                         if (!uid_valid(opts->fs_uid))
1116                                 return -EINVAL;
1117                         break;
1118                 case Opt_gid:
1119                         if (match_int(&args[0], &option))
1120                                 return -EINVAL;
1121                         opts->fs_gid = make_kgid(current_user_ns(), option);
1122                         if (!gid_valid(opts->fs_gid))
1123                                 return -EINVAL;
1124                         break;
1125                 case Opt_umask:
1126                         if (match_octal(&args[0], &option))
1127                                 return -EINVAL;
1128                         opts->fs_fmask = opts->fs_dmask = option;
1129                         break;
1130                 case Opt_dmask:
1131                         if (match_octal(&args[0], &option))
1132                                 return -EINVAL;
1133                         opts->fs_dmask = option;
1134                         break;
1135                 case Opt_fmask:
1136                         if (match_octal(&args[0], &option))
1137                                 return -EINVAL;
1138                         opts->fs_fmask = option;
1139                         break;
1140                 case Opt_allow_utime:
1141                         if (match_octal(&args[0], &option))
1142                                 return -EINVAL;
1143                         opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1144                         break;
1145                 case Opt_codepage:
1146                         if (match_int(&args[0], &option))
1147                                 return -EINVAL;
1148                         opts->codepage = option;
1149                         break;
1150                 case Opt_flush:
1151                         opts->flush = 1;
1152                         break;
1153                 case Opt_time_offset:
1154                         if (match_int(&args[0], &option))
1155                                 return -EINVAL;
1156                         if (option < -12 * 60 || option > 12 * 60)
1157                                 return -EINVAL;
1158                         opts->tz_set = 1;
1159                         opts->time_offset = option;
1160                         break;
1161                 case Opt_tz_utc:
1162                         opts->tz_set = 1;
1163                         opts->time_offset = 0;
1164                         break;
1165                 case Opt_err_cont:
1166                         opts->errors = FAT_ERRORS_CONT;
1167                         break;
1168                 case Opt_err_panic:
1169                         opts->errors = FAT_ERRORS_PANIC;
1170                         break;
1171                 case Opt_err_ro:
1172                         opts->errors = FAT_ERRORS_RO;
1173                         break;
1174                 case Opt_nfs_stale_rw:
1175                         opts->nfs = FAT_NFS_STALE_RW;
1176                         break;
1177                 case Opt_nfs_nostale_ro:
1178                         opts->nfs = FAT_NFS_NOSTALE_RO;
1179                         break;
1180                 case Opt_dos1xfloppy:
1181                         opts->dos1xfloppy = 1;
1182                         break;
1183
1184                 /* msdos specific */
1185                 case Opt_dots:
1186                         opts->dotsOK = 1;
1187                         break;
1188                 case Opt_nodots:
1189                         opts->dotsOK = 0;
1190                         break;
1191
1192                 /* vfat specific */
1193                 case Opt_charset:
1194                         if (opts->iocharset != fat_default_iocharset)
1195                                 kfree(opts->iocharset);
1196                         iocharset = match_strdup(&args[0]);
1197                         if (!iocharset)
1198                                 return -ENOMEM;
1199                         opts->iocharset = iocharset;
1200                         break;
1201                 case Opt_shortname_lower:
1202                         opts->shortname = VFAT_SFN_DISPLAY_LOWER
1203                                         | VFAT_SFN_CREATE_WIN95;
1204                         break;
1205                 case Opt_shortname_win95:
1206                         opts->shortname = VFAT_SFN_DISPLAY_WIN95
1207                                         | VFAT_SFN_CREATE_WIN95;
1208                         break;
1209                 case Opt_shortname_winnt:
1210                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1211                                         | VFAT_SFN_CREATE_WINNT;
1212                         break;
1213                 case Opt_shortname_mixed:
1214                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1215                                         | VFAT_SFN_CREATE_WIN95;
1216                         break;
1217                 case Opt_utf8_no:               /* 0 or no or false */
1218                         opts->utf8 = 0;
1219                         break;
1220                 case Opt_utf8_yes:              /* empty or 1 or yes or true */
1221                         opts->utf8 = 1;
1222                         break;
1223                 case Opt_uni_xl_no:             /* 0 or no or false */
1224                         opts->unicode_xlate = 0;
1225                         break;
1226                 case Opt_uni_xl_yes:            /* empty or 1 or yes or true */
1227                         opts->unicode_xlate = 1;
1228                         break;
1229                 case Opt_nonumtail_no:          /* 0 or no or false */
1230                         opts->numtail = 1;      /* negated option */
1231                         break;
1232                 case Opt_nonumtail_yes:         /* empty or 1 or yes or true */
1233                         opts->numtail = 0;      /* negated option */
1234                         break;
1235                 case Opt_rodir:
1236                         opts->rodir = 1;
1237                         break;
1238                 case Opt_discard:
1239                         opts->discard = 1;
1240                         break;
1241
1242                 /* obsolete mount options */
1243                 case Opt_obsolete:
1244                         fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1245                                "not supported now", p);
1246                         break;
1247                 /* unknown option */
1248                 default:
1249                         if (!silent) {
1250                                 fat_msg(sb, KERN_ERR,
1251                                        "Unrecognized mount option \"%s\" "
1252                                        "or missing value", p);
1253                         }
1254                         return -EINVAL;
1255                 }
1256         }
1257
1258 out:
1259         /* UTF-8 doesn't provide FAT semantics */
1260         if (!strcmp(opts->iocharset, "utf8")) {
1261                 fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1262                        " for FAT filesystems, filesystem will be "
1263                        "case sensitive!");
1264         }
1265
1266         /* If user doesn't specify allow_utime, it's initialized from dmask. */
1267         if (opts->allow_utime == (unsigned short)-1)
1268                 opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1269         if (opts->unicode_xlate)
1270                 opts->utf8 = 0;
1271         if (opts->nfs == FAT_NFS_NOSTALE_RO) {
1272                 sb->s_flags |= MS_RDONLY;
1273                 sb->s_export_op = &fat_export_ops_nostale;
1274         }
1275
1276         return 0;
1277 }
1278
1279 static int fat_read_root(struct inode *inode)
1280 {
1281         struct super_block *sb = inode->i_sb;
1282         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1283         int error;
1284
1285         MSDOS_I(inode)->i_pos = MSDOS_ROOT_INO;
1286         inode->i_uid = sbi->options.fs_uid;
1287         inode->i_gid = sbi->options.fs_gid;
1288         inode->i_version++;
1289         inode->i_generation = 0;
1290         inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1291         inode->i_op = sbi->dir_ops;
1292         inode->i_fop = &fat_dir_operations;
1293         if (sbi->fat_bits == 32) {
1294                 MSDOS_I(inode)->i_start = sbi->root_cluster;
1295                 error = fat_calc_dir_size(inode);
1296                 if (error < 0)
1297                         return error;
1298         } else {
1299                 MSDOS_I(inode)->i_start = 0;
1300                 inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1301         }
1302         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1303                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1304         MSDOS_I(inode)->i_logstart = 0;
1305         MSDOS_I(inode)->mmu_private = inode->i_size;
1306
1307         fat_save_attrs(inode, ATTR_DIR);
1308         inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1309         inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1310         set_nlink(inode, fat_subdirs(inode)+2);
1311
1312         return 0;
1313 }
1314
1315 static unsigned long calc_fat_clusters(struct super_block *sb)
1316 {
1317         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1318
1319         /* Divide first to avoid overflow */
1320         if (sbi->fat_bits != 12) {
1321                 unsigned long ent_per_sec = sb->s_blocksize * 8 / sbi->fat_bits;
1322                 return ent_per_sec * sbi->fat_length;
1323         }
1324
1325         return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1326 }
1327
1328 static bool fat_bpb_is_zero(struct fat_boot_sector *b)
1329 {
1330         if (get_unaligned_le16(&b->sector_size))
1331                 return false;
1332         if (b->sec_per_clus)
1333                 return false;
1334         if (b->reserved)
1335                 return false;
1336         if (b->fats)
1337                 return false;
1338         if (get_unaligned_le16(&b->dir_entries))
1339                 return false;
1340         if (get_unaligned_le16(&b->sectors))
1341                 return false;
1342         if (b->media)
1343                 return false;
1344         if (b->fat_length)
1345                 return false;
1346         if (b->secs_track)
1347                 return false;
1348         if (b->heads)
1349                 return false;
1350         return true;
1351 }
1352
1353 static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
1354         int silent, struct fat_bios_param_block *bpb)
1355 {
1356         int error = -EINVAL;
1357
1358         /* Read in BPB ... */
1359         memset(bpb, 0, sizeof(*bpb));
1360         bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
1361         bpb->fat_sec_per_clus = b->sec_per_clus;
1362         bpb->fat_reserved = le16_to_cpu(b->reserved);
1363         bpb->fat_fats = b->fats;
1364         bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
1365         bpb->fat_sectors = get_unaligned_le16(&b->sectors);
1366         bpb->fat_fat_length = le16_to_cpu(b->fat_length);
1367         bpb->fat_total_sect = le32_to_cpu(b->total_sect);
1368
1369         bpb->fat16_state = b->fat16.state;
1370         bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
1371
1372         bpb->fat32_length = le32_to_cpu(b->fat32.length);
1373         bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
1374         bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
1375         bpb->fat32_state = b->fat32.state;
1376         bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
1377
1378         /* Validate this looks like a FAT filesystem BPB */
1379         if (!bpb->fat_reserved) {
1380                 if (!silent)
1381                         fat_msg(sb, KERN_ERR,
1382                                 "bogus number of reserved sectors");
1383                 goto out;
1384         }
1385         if (!bpb->fat_fats) {
1386                 if (!silent)
1387                         fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1388                 goto out;
1389         }
1390
1391         /*
1392          * Earlier we checked here that b->secs_track and b->head are nonzero,
1393          * but it turns out valid FAT filesystems can have zero there.
1394          */
1395
1396         if (!fat_valid_media(b->media)) {
1397                 if (!silent)
1398                         fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1399                                 (unsigned)b->media);
1400                 goto out;
1401         }
1402
1403         if (!is_power_of_2(bpb->fat_sector_size)
1404             || (bpb->fat_sector_size < 512)
1405             || (bpb->fat_sector_size > 4096)) {
1406                 if (!silent)
1407                         fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1408                                (unsigned)bpb->fat_sector_size);
1409                 goto out;
1410         }
1411
1412         if (!is_power_of_2(bpb->fat_sec_per_clus)) {
1413                 if (!silent)
1414                         fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1415                                 (unsigned)bpb->fat_sec_per_clus);
1416                 goto out;
1417         }
1418
1419         error = 0;
1420
1421 out:
1422         return error;
1423 }
1424
1425 static int fat_read_static_bpb(struct super_block *sb,
1426         struct fat_boot_sector *b, int silent,
1427         struct fat_bios_param_block *bpb)
1428 {
1429         static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
1430
1431         struct fat_floppy_defaults *fdefaults = NULL;
1432         int error = -EINVAL;
1433         sector_t bd_sects;
1434         unsigned i;
1435
1436         bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
1437
1438         /* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
1439         if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
1440                 if (!silent)
1441                         fat_msg(sb, KERN_ERR,
1442                                 "%s; no bootstrapping code", notdos1x);
1443                 goto out;
1444         }
1445
1446         /*
1447          * If any value in this region is non-zero, it isn't archaic
1448          * DOS.
1449          */
1450         if (!fat_bpb_is_zero(b)) {
1451                 if (!silent)
1452                         fat_msg(sb, KERN_ERR,
1453                                 "%s; DOS 2.x BPB is non-zero", notdos1x);
1454                 goto out;
1455         }
1456
1457         for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
1458                 if (floppy_defaults[i].nr_sectors == bd_sects) {
1459                         fdefaults = &floppy_defaults[i];
1460                         break;
1461                 }
1462         }
1463
1464         if (fdefaults == NULL) {
1465                 if (!silent)
1466                         fat_msg(sb, KERN_WARNING,
1467                                 "This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
1468                                 (u64)bd_sects);
1469                 goto out;
1470         }
1471
1472         if (!silent)
1473                 fat_msg(sb, KERN_INFO,
1474                         "This looks like a DOS 1.x volume; assuming default BPB values");
1475
1476         memset(bpb, 0, sizeof(*bpb));
1477         bpb->fat_sector_size = SECTOR_SIZE;
1478         bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
1479         bpb->fat_reserved = 1;
1480         bpb->fat_fats = 2;
1481         bpb->fat_dir_entries = fdefaults->dir_entries;
1482         bpb->fat_sectors = fdefaults->nr_sectors;
1483         bpb->fat_fat_length = fdefaults->fat_length;
1484
1485         error = 0;
1486
1487 out:
1488         return error;
1489 }
1490
1491 /*
1492  * Read the super block of an MS-DOS FS.
1493  */
1494 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1495                    void (*setup)(struct super_block *))
1496 {
1497         struct inode *root_inode = NULL, *fat_inode = NULL;
1498         struct inode *fsinfo_inode = NULL;
1499         struct buffer_head *bh;
1500         struct fat_bios_param_block bpb;
1501         struct msdos_sb_info *sbi;
1502         u16 logical_sector_size;
1503         u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1504         int debug;
1505         long error;
1506         char buf[50];
1507
1508         /*
1509          * GFP_KERNEL is ok here, because while we do hold the
1510          * supeblock lock, memory pressure can't call back into
1511          * the filesystem, since we're only just about to mount
1512          * it and have no inodes etc active!
1513          */
1514         sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1515         if (!sbi)
1516                 return -ENOMEM;
1517         sb->s_fs_info = sbi;
1518
1519         sb->s_flags |= MS_NODIRATIME;
1520         sb->s_magic = MSDOS_SUPER_MAGIC;
1521         sb->s_op = &fat_sops;
1522         sb->s_export_op = &fat_export_ops;
1523         mutex_init(&sbi->nfs_build_inode_lock);
1524         ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1525                              DEFAULT_RATELIMIT_BURST);
1526
1527         error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1528         if (error)
1529                 goto out_fail;
1530
1531         setup(sb); /* flavour-specific stuff that needs options */
1532
1533         error = -EIO;
1534         sb_min_blocksize(sb, 512);
1535         bh = sb_bread(sb, 0);
1536         if (bh == NULL) {
1537                 fat_msg(sb, KERN_ERR, "unable to read boot sector");
1538                 goto out_fail;
1539         }
1540
1541         error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
1542                 &bpb);
1543         if (error == -EINVAL && sbi->options.dos1xfloppy)
1544                 error = fat_read_static_bpb(sb,
1545                         (struct fat_boot_sector *)bh->b_data, silent, &bpb);
1546         brelse(bh);
1547
1548         if (error == -EINVAL)
1549                 goto out_invalid;
1550         else if (error)
1551                 goto out_fail;
1552
1553         logical_sector_size = bpb.fat_sector_size;
1554         sbi->sec_per_clus = bpb.fat_sec_per_clus;
1555
1556         error = -EIO;
1557         if (logical_sector_size < sb->s_blocksize) {
1558                 fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1559                        " (logical sector size = %u)", logical_sector_size);
1560                 goto out_fail;
1561         }
1562
1563         if (logical_sector_size > sb->s_blocksize) {
1564                 struct buffer_head *bh_resize;
1565
1566                 if (!sb_set_blocksize(sb, logical_sector_size)) {
1567                         fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1568                                logical_sector_size);
1569                         goto out_fail;
1570                 }
1571
1572                 /* Verify that the larger boot sector is fully readable */
1573                 bh_resize = sb_bread(sb, 0);
1574                 if (bh_resize == NULL) {
1575                         fat_msg(sb, KERN_ERR, "unable to read boot sector"
1576                                " (logical sector size = %lu)",
1577                                sb->s_blocksize);
1578                         goto out_fail;
1579                 }
1580                 brelse(bh_resize);
1581         }
1582
1583         mutex_init(&sbi->s_lock);
1584         sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1585         sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1586         sbi->fats = bpb.fat_fats;
1587         sbi->fat_bits = 0;              /* Don't know yet */
1588         sbi->fat_start = bpb.fat_reserved;
1589         sbi->fat_length = bpb.fat_fat_length;
1590         sbi->root_cluster = 0;
1591         sbi->free_clusters = -1;        /* Don't know yet */
1592         sbi->free_clus_valid = 0;
1593         sbi->prev_free = FAT_START_ENT;
1594         sb->s_maxbytes = 0xffffffff;
1595
1596         if (!sbi->fat_length && bpb.fat32_length) {
1597                 struct fat_boot_fsinfo *fsinfo;
1598                 struct buffer_head *fsinfo_bh;
1599
1600                 /* Must be FAT32 */
1601                 sbi->fat_bits = 32;
1602                 sbi->fat_length = bpb.fat32_length;
1603                 sbi->root_cluster = bpb.fat32_root_cluster;
1604
1605                 /* MC - if info_sector is 0, don't multiply by 0 */
1606                 sbi->fsinfo_sector = bpb.fat32_info_sector;
1607                 if (sbi->fsinfo_sector == 0)
1608                         sbi->fsinfo_sector = 1;
1609
1610                 fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1611                 if (fsinfo_bh == NULL) {
1612                         fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1613                                " (sector = %lu)", sbi->fsinfo_sector);
1614                         goto out_fail;
1615                 }
1616
1617                 fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1618                 if (!IS_FSINFO(fsinfo)) {
1619                         fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1620                                "0x%08x, 0x%08x (sector = %lu)",
1621                                le32_to_cpu(fsinfo->signature1),
1622                                le32_to_cpu(fsinfo->signature2),
1623                                sbi->fsinfo_sector);
1624                 } else {
1625                         if (sbi->options.usefree)
1626                                 sbi->free_clus_valid = 1;
1627                         sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1628                         sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1629                 }
1630
1631                 brelse(fsinfo_bh);
1632         }
1633
1634         /* interpret volume ID as a little endian 32 bit integer */
1635         if (sbi->fat_bits == 32)
1636                 sbi->vol_id = bpb.fat32_vol_id;
1637         else /* fat 16 or 12 */
1638                 sbi->vol_id = bpb.fat16_vol_id;
1639
1640         sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1641         sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1642
1643         sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1644         sbi->dir_entries = bpb.fat_dir_entries;
1645         if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1646                 if (!silent)
1647                         fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
1648                                " (%u)", sbi->dir_entries);
1649                 goto out_invalid;
1650         }
1651
1652         rootdir_sectors = sbi->dir_entries
1653                 * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1654         sbi->data_start = sbi->dir_start + rootdir_sectors;
1655         total_sectors = bpb.fat_sectors;
1656         if (total_sectors == 0)
1657                 total_sectors = bpb.fat_total_sect;
1658
1659         total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1660
1661         if (sbi->fat_bits != 32)
1662                 sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1663
1664         /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1665         if (sbi->fat_bits == 32)
1666                 sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
1667         else /* fat 16 or 12 */
1668                 sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
1669
1670         /* check that FAT table does not overflow */
1671         fat_clusters = calc_fat_clusters(sb);
1672         total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1673         if (total_clusters > MAX_FAT(sb)) {
1674                 if (!silent)
1675                         fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1676                                total_clusters);
1677                 goto out_invalid;
1678         }
1679
1680         sbi->max_cluster = total_clusters + FAT_START_ENT;
1681         /* check the free_clusters, it's not necessarily correct */
1682         if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1683                 sbi->free_clusters = -1;
1684         /* check the prev_free, it's not necessarily correct */
1685         sbi->prev_free %= sbi->max_cluster;
1686         if (sbi->prev_free < FAT_START_ENT)
1687                 sbi->prev_free = FAT_START_ENT;
1688
1689         /* set up enough so that it can read an inode */
1690         fat_hash_init(sb);
1691         dir_hash_init(sb);
1692         fat_ent_access_init(sb);
1693
1694         /*
1695          * The low byte of FAT's first entry must have same value with
1696          * media-field.  But in real world, too many devices is
1697          * writing wrong value.  So, removed that validity check.
1698          *
1699          * if (FAT_FIRST_ENT(sb, media) != first)
1700          */
1701
1702         error = -EINVAL;
1703         sprintf(buf, "cp%d", sbi->options.codepage);
1704         sbi->nls_disk = load_nls(buf);
1705         if (!sbi->nls_disk) {
1706                 fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1707                 goto out_fail;
1708         }
1709
1710         /* FIXME: utf8 is using iocharset for upper/lower conversion */
1711         if (sbi->options.isvfat) {
1712                 sbi->nls_io = load_nls(sbi->options.iocharset);
1713                 if (!sbi->nls_io) {
1714                         fat_msg(sb, KERN_ERR, "IO charset %s not found",
1715                                sbi->options.iocharset);
1716                         goto out_fail;
1717                 }
1718         }
1719
1720         error = -ENOMEM;
1721         fat_inode = new_inode(sb);
1722         if (!fat_inode)
1723                 goto out_fail;
1724         MSDOS_I(fat_inode)->i_pos = 0;
1725         sbi->fat_inode = fat_inode;
1726
1727         fsinfo_inode = new_inode(sb);
1728         if (!fsinfo_inode)
1729                 goto out_fail;
1730         fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1731         sbi->fsinfo_inode = fsinfo_inode;
1732         insert_inode_hash(fsinfo_inode);
1733
1734         root_inode = new_inode(sb);
1735         if (!root_inode)
1736                 goto out_fail;
1737         root_inode->i_ino = MSDOS_ROOT_INO;
1738         root_inode->i_version = 1;
1739         error = fat_read_root(root_inode);
1740         if (error < 0) {
1741                 iput(root_inode);
1742                 goto out_fail;
1743         }
1744         error = -ENOMEM;
1745         insert_inode_hash(root_inode);
1746         fat_attach(root_inode, 0);
1747         sb->s_root = d_make_root(root_inode);
1748         if (!sb->s_root) {
1749                 fat_msg(sb, KERN_ERR, "get root inode failed");
1750                 goto out_fail;
1751         }
1752
1753         if (sbi->options.discard) {
1754                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1755                 if (!blk_queue_discard(q))
1756                         fat_msg(sb, KERN_WARNING,
1757                                         "mounting with \"discard\" option, but "
1758                                         "the device does not support discard");
1759         }
1760
1761         fat_set_state(sb, 1, 0);
1762         return 0;
1763
1764 out_invalid:
1765         error = -EINVAL;
1766         if (!silent)
1767                 fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1768
1769 out_fail:
1770         if (fsinfo_inode)
1771                 iput(fsinfo_inode);
1772         if (fat_inode)
1773                 iput(fat_inode);
1774         unload_nls(sbi->nls_io);
1775         unload_nls(sbi->nls_disk);
1776         if (sbi->options.iocharset != fat_default_iocharset)
1777                 kfree(sbi->options.iocharset);
1778         sb->s_fs_info = NULL;
1779         kfree(sbi);
1780         return error;
1781 }
1782
1783 EXPORT_SYMBOL_GPL(fat_fill_super);
1784
1785 /*
1786  * helper function for fat_flush_inodes.  This writes both the inode
1787  * and the file data blocks, waiting for in flight data blocks before
1788  * the start of the call.  It does not wait for any io started
1789  * during the call
1790  */
1791 static int writeback_inode(struct inode *inode)
1792 {
1793
1794         int ret;
1795
1796         /* if we used wait=1, sync_inode_metadata waits for the io for the
1797         * inode to finish.  So wait=0 is sent down to sync_inode_metadata
1798         * and filemap_fdatawrite is used for the data blocks
1799         */
1800         ret = sync_inode_metadata(inode, 0);
1801         if (!ret)
1802                 ret = filemap_fdatawrite(inode->i_mapping);
1803         return ret;
1804 }
1805
1806 /*
1807  * write data and metadata corresponding to i1 and i2.  The io is
1808  * started but we do not wait for any of it to finish.
1809  *
1810  * filemap_flush is used for the block device, so if there is a dirty
1811  * page for a block already in flight, we will not wait and start the
1812  * io over again
1813  */
1814 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1815 {
1816         int ret = 0;
1817         if (!MSDOS_SB(sb)->options.flush)
1818                 return 0;
1819         if (i1)
1820                 ret = writeback_inode(i1);
1821         if (!ret && i2)
1822                 ret = writeback_inode(i2);
1823         if (!ret) {
1824                 struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1825                 ret = filemap_flush(mapping);
1826         }
1827         return ret;
1828 }
1829 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1830
1831 static int __init init_fat_fs(void)
1832 {
1833         int err;
1834
1835         err = fat_cache_init();
1836         if (err)
1837                 return err;
1838
1839         err = fat_init_inodecache();
1840         if (err)
1841                 goto failed;
1842
1843         return 0;
1844
1845 failed:
1846         fat_cache_destroy();
1847         return err;
1848 }
1849
1850 static void __exit exit_fat_fs(void)
1851 {
1852         fat_cache_destroy();
1853         fat_destroy_inodecache();
1854 }
1855
1856 module_init(init_fat_fs)
1857 module_exit(exit_fat_fs)
1858
1859 MODULE_LICENSE("GPL");