GNU Linux-libre 4.4.288-gnu1
[releases.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/pagemap.h>
15 #include <linux/mpage.h>
16 #include <linux/vfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/parser.h>
19 #include <linux/uio.h>
20 #include <linux/blkdev.h>
21 #include <linux/backing-dev.h>
22 #include <asm/unaligned.h>
23 #include "fat.h"
24
25 #ifndef CONFIG_FAT_DEFAULT_IOCHARSET
26 /* if user don't select VFAT, this is undefined. */
27 #define CONFIG_FAT_DEFAULT_IOCHARSET    ""
28 #endif
29
30 #define KB_IN_SECTORS 2
31
32 /*
33  * A deserialized copy of the on-disk structure laid out in struct
34  * fat_boot_sector.
35  */
36 struct fat_bios_param_block {
37         u16     fat_sector_size;
38         u8      fat_sec_per_clus;
39         u16     fat_reserved;
40         u8      fat_fats;
41         u16     fat_dir_entries;
42         u16     fat_sectors;
43         u16     fat_fat_length;
44         u32     fat_total_sect;
45
46         u8      fat16_state;
47         u32     fat16_vol_id;
48
49         u32     fat32_length;
50         u32     fat32_root_cluster;
51         u16     fat32_info_sector;
52         u8      fat32_state;
53         u32     fat32_vol_id;
54 };
55
56 static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
57 static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
58
59 static struct fat_floppy_defaults {
60         unsigned nr_sectors;
61         unsigned sec_per_clus;
62         unsigned dir_entries;
63         unsigned media;
64         unsigned fat_length;
65 } floppy_defaults[] = {
66 {
67         .nr_sectors = 160 * KB_IN_SECTORS,
68         .sec_per_clus = 1,
69         .dir_entries = 64,
70         .media = 0xFE,
71         .fat_length = 1,
72 },
73 {
74         .nr_sectors = 180 * KB_IN_SECTORS,
75         .sec_per_clus = 1,
76         .dir_entries = 64,
77         .media = 0xFC,
78         .fat_length = 2,
79 },
80 {
81         .nr_sectors = 320 * KB_IN_SECTORS,
82         .sec_per_clus = 2,
83         .dir_entries = 112,
84         .media = 0xFF,
85         .fat_length = 1,
86 },
87 {
88         .nr_sectors = 360 * KB_IN_SECTORS,
89         .sec_per_clus = 2,
90         .dir_entries = 112,
91         .media = 0xFD,
92         .fat_length = 2,
93 },
94 };
95
96 static int fat_add_cluster(struct inode *inode)
97 {
98         int err, cluster;
99
100         err = fat_alloc_clusters(inode, &cluster, 1);
101         if (err)
102                 return err;
103         /* FIXME: this cluster should be added after data of this
104          * cluster is writed */
105         err = fat_chain_add(inode, cluster, 1);
106         if (err)
107                 fat_free_clusters(inode, cluster);
108         return err;
109 }
110
111 static inline int __fat_get_block(struct inode *inode, sector_t iblock,
112                                   unsigned long *max_blocks,
113                                   struct buffer_head *bh_result, int create)
114 {
115         struct super_block *sb = inode->i_sb;
116         struct msdos_sb_info *sbi = MSDOS_SB(sb);
117         unsigned long mapped_blocks;
118         sector_t phys;
119         int err, offset;
120
121         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
122         if (err)
123                 return err;
124         if (phys) {
125                 map_bh(bh_result, sb, phys);
126                 *max_blocks = min(mapped_blocks, *max_blocks);
127                 return 0;
128         }
129         if (!create)
130                 return 0;
131
132         if (iblock != MSDOS_I(inode)->mmu_private >> sb->s_blocksize_bits) {
133                 fat_fs_error(sb, "corrupted file size (i_pos %lld, %lld)",
134                         MSDOS_I(inode)->i_pos, MSDOS_I(inode)->mmu_private);
135                 return -EIO;
136         }
137
138         offset = (unsigned long)iblock & (sbi->sec_per_clus - 1);
139         if (!offset) {
140                 /* TODO: multiple cluster allocation would be desirable. */
141                 err = fat_add_cluster(inode);
142                 if (err)
143                         return err;
144         }
145         /* available blocks on this cluster */
146         mapped_blocks = sbi->sec_per_clus - offset;
147
148         *max_blocks = min(mapped_blocks, *max_blocks);
149         MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
150
151         err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
152         if (err)
153                 return err;
154
155         BUG_ON(!phys);
156         BUG_ON(*max_blocks != mapped_blocks);
157         set_buffer_new(bh_result);
158         map_bh(bh_result, sb, phys);
159
160         return 0;
161 }
162
163 static int fat_get_block(struct inode *inode, sector_t iblock,
164                          struct buffer_head *bh_result, int create)
165 {
166         struct super_block *sb = inode->i_sb;
167         unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
168         int err;
169
170         err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
171         if (err)
172                 return err;
173         bh_result->b_size = max_blocks << sb->s_blocksize_bits;
174         return 0;
175 }
176
177 static int fat_writepage(struct page *page, struct writeback_control *wbc)
178 {
179         return block_write_full_page(page, fat_get_block, wbc);
180 }
181
182 static int fat_writepages(struct address_space *mapping,
183                           struct writeback_control *wbc)
184 {
185         return mpage_writepages(mapping, wbc, fat_get_block);
186 }
187
188 static int fat_readpage(struct file *file, struct page *page)
189 {
190         return mpage_readpage(page, fat_get_block);
191 }
192
193 static int fat_readpages(struct file *file, struct address_space *mapping,
194                          struct list_head *pages, unsigned nr_pages)
195 {
196         return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
197 }
198
199 static void fat_write_failed(struct address_space *mapping, loff_t to)
200 {
201         struct inode *inode = mapping->host;
202
203         if (to > inode->i_size) {
204                 truncate_pagecache(inode, inode->i_size);
205                 fat_truncate_blocks(inode, inode->i_size);
206         }
207 }
208
209 static int fat_write_begin(struct file *file, struct address_space *mapping,
210                         loff_t pos, unsigned len, unsigned flags,
211                         struct page **pagep, void **fsdata)
212 {
213         int err;
214
215         *pagep = NULL;
216         err = cont_write_begin(file, mapping, pos, len, flags,
217                                 pagep, fsdata, fat_get_block,
218                                 &MSDOS_I(mapping->host)->mmu_private);
219         if (err < 0)
220                 fat_write_failed(mapping, pos + len);
221         return err;
222 }
223
224 static int fat_write_end(struct file *file, struct address_space *mapping,
225                         loff_t pos, unsigned len, unsigned copied,
226                         struct page *pagep, void *fsdata)
227 {
228         struct inode *inode = mapping->host;
229         int err;
230         err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
231         if (err < len)
232                 fat_write_failed(mapping, pos + len);
233         if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
234                 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
235                 MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
236                 mark_inode_dirty(inode);
237         }
238         return err;
239 }
240
241 static ssize_t fat_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
242                              loff_t offset)
243 {
244         struct file *file = iocb->ki_filp;
245         struct address_space *mapping = file->f_mapping;
246         struct inode *inode = mapping->host;
247         size_t count = iov_iter_count(iter);
248         ssize_t ret;
249
250         if (iov_iter_rw(iter) == WRITE) {
251                 /*
252                  * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
253                  * so we need to update the ->mmu_private to block boundary.
254                  *
255                  * But we must fill the remaining area or hole by nul for
256                  * updating ->mmu_private.
257                  *
258                  * Return 0, and fallback to normal buffered write.
259                  */
260                 loff_t size = offset + count;
261                 if (MSDOS_I(inode)->mmu_private < size)
262                         return 0;
263         }
264
265         /*
266          * FAT need to use the DIO_LOCKING for avoiding the race
267          * condition of fat_get_block() and ->truncate().
268          */
269         ret = blockdev_direct_IO(iocb, inode, iter, offset, fat_get_block);
270         if (ret < 0 && iov_iter_rw(iter) == WRITE)
271                 fat_write_failed(mapping, offset + count);
272
273         return ret;
274 }
275
276 static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
277 {
278         sector_t blocknr;
279
280         /* fat_get_cluster() assumes the requested blocknr isn't truncated. */
281         down_read(&MSDOS_I(mapping->host)->truncate_lock);
282         blocknr = generic_block_bmap(mapping, block, fat_get_block);
283         up_read(&MSDOS_I(mapping->host)->truncate_lock);
284
285         return blocknr;
286 }
287
288 /*
289  * fat_block_truncate_page() zeroes out a mapping from file offset `from'
290  * up to the end of the block which corresponds to `from'.
291  * This is required during truncate to physically zeroout the tail end
292  * of that block so it doesn't yield old data if the file is later grown.
293  * Also, avoid causing failure from fsx for cases of "data past EOF"
294  */
295 int fat_block_truncate_page(struct inode *inode, loff_t from)
296 {
297         return block_truncate_page(inode->i_mapping, from, fat_get_block);
298 }
299
300 static const struct address_space_operations fat_aops = {
301         .readpage       = fat_readpage,
302         .readpages      = fat_readpages,
303         .writepage      = fat_writepage,
304         .writepages     = fat_writepages,
305         .write_begin    = fat_write_begin,
306         .write_end      = fat_write_end,
307         .direct_IO      = fat_direct_IO,
308         .bmap           = _fat_bmap
309 };
310
311 /*
312  * New FAT inode stuff. We do the following:
313  *      a) i_ino is constant and has nothing with on-disk location.
314  *      b) FAT manages its own cache of directory entries.
315  *      c) *This* cache is indexed by on-disk location.
316  *      d) inode has an associated directory entry, all right, but
317  *              it may be unhashed.
318  *      e) currently entries are stored within struct inode. That should
319  *              change.
320  *      f) we deal with races in the following way:
321  *              1. readdir() and lookup() do FAT-dir-cache lookup.
322  *              2. rename() unhashes the F-d-c entry and rehashes it in
323  *                      a new place.
324  *              3. unlink() and rmdir() unhash F-d-c entry.
325  *              4. fat_write_inode() checks whether the thing is unhashed.
326  *                      If it is we silently return. If it isn't we do bread(),
327  *                      check if the location is still valid and retry if it
328  *                      isn't. Otherwise we do changes.
329  *              5. Spinlock is used to protect hash/unhash/location check/lookup
330  *              6. fat_evict_inode() unhashes the F-d-c entry.
331  *              7. lookup() and readdir() do igrab() if they find a F-d-c entry
332  *                      and consider negative result as cache miss.
333  */
334
335 static void fat_hash_init(struct super_block *sb)
336 {
337         struct msdos_sb_info *sbi = MSDOS_SB(sb);
338         int i;
339
340         spin_lock_init(&sbi->inode_hash_lock);
341         for (i = 0; i < FAT_HASH_SIZE; i++)
342                 INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
343 }
344
345 static inline unsigned long fat_hash(loff_t i_pos)
346 {
347         return hash_32(i_pos, FAT_HASH_BITS);
348 }
349
350 static void dir_hash_init(struct super_block *sb)
351 {
352         struct msdos_sb_info *sbi = MSDOS_SB(sb);
353         int i;
354
355         spin_lock_init(&sbi->dir_hash_lock);
356         for (i = 0; i < FAT_HASH_SIZE; i++)
357                 INIT_HLIST_HEAD(&sbi->dir_hashtable[i]);
358 }
359
360 void fat_attach(struct inode *inode, loff_t i_pos)
361 {
362         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
363
364         if (inode->i_ino != MSDOS_ROOT_INO) {
365                 struct hlist_head *head =   sbi->inode_hashtable
366                                           + fat_hash(i_pos);
367
368                 spin_lock(&sbi->inode_hash_lock);
369                 MSDOS_I(inode)->i_pos = i_pos;
370                 hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
371                 spin_unlock(&sbi->inode_hash_lock);
372         }
373
374         /* If NFS support is enabled, cache the mapping of start cluster
375          * to directory inode. This is used during reconnection of
376          * dentries to the filesystem root.
377          */
378         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
379                 struct hlist_head *d_head = sbi->dir_hashtable;
380                 d_head += fat_dir_hash(MSDOS_I(inode)->i_logstart);
381
382                 spin_lock(&sbi->dir_hash_lock);
383                 hlist_add_head(&MSDOS_I(inode)->i_dir_hash, d_head);
384                 spin_unlock(&sbi->dir_hash_lock);
385         }
386 }
387 EXPORT_SYMBOL_GPL(fat_attach);
388
389 void fat_detach(struct inode *inode)
390 {
391         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
392         spin_lock(&sbi->inode_hash_lock);
393         MSDOS_I(inode)->i_pos = 0;
394         hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
395         spin_unlock(&sbi->inode_hash_lock);
396
397         if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
398                 spin_lock(&sbi->dir_hash_lock);
399                 hlist_del_init(&MSDOS_I(inode)->i_dir_hash);
400                 spin_unlock(&sbi->dir_hash_lock);
401         }
402 }
403 EXPORT_SYMBOL_GPL(fat_detach);
404
405 struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
406 {
407         struct msdos_sb_info *sbi = MSDOS_SB(sb);
408         struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
409         struct msdos_inode_info *i;
410         struct inode *inode = NULL;
411
412         spin_lock(&sbi->inode_hash_lock);
413         hlist_for_each_entry(i, head, i_fat_hash) {
414                 BUG_ON(i->vfs_inode.i_sb != sb);
415                 if (i->i_pos != i_pos)
416                         continue;
417                 inode = igrab(&i->vfs_inode);
418                 if (inode)
419                         break;
420         }
421         spin_unlock(&sbi->inode_hash_lock);
422         return inode;
423 }
424
425 static int is_exec(unsigned char *extension)
426 {
427         unsigned char exe_extensions[] = "EXECOMBAT", *walk;
428
429         for (walk = exe_extensions; *walk; walk += 3)
430                 if (!strncmp(extension, walk, 3))
431                         return 1;
432         return 0;
433 }
434
435 static int fat_calc_dir_size(struct inode *inode)
436 {
437         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
438         int ret, fclus, dclus;
439
440         inode->i_size = 0;
441         if (MSDOS_I(inode)->i_start == 0)
442                 return 0;
443
444         ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
445         if (ret < 0)
446                 return ret;
447         inode->i_size = (fclus + 1) << sbi->cluster_bits;
448
449         return 0;
450 }
451
452 /* doesn't deal with root inode */
453 int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
454 {
455         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
456         int error;
457
458         MSDOS_I(inode)->i_pos = 0;
459         inode->i_uid = sbi->options.fs_uid;
460         inode->i_gid = sbi->options.fs_gid;
461         inode->i_version++;
462         inode->i_generation = get_seconds();
463
464         if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
465                 inode->i_generation &= ~1;
466                 inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
467                 inode->i_op = sbi->dir_ops;
468                 inode->i_fop = &fat_dir_operations;
469
470                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
471                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
472                 error = fat_calc_dir_size(inode);
473                 if (error < 0)
474                         return error;
475                 MSDOS_I(inode)->mmu_private = inode->i_size;
476
477                 set_nlink(inode, fat_subdirs(inode));
478         } else { /* not a directory */
479                 inode->i_generation |= 1;
480                 inode->i_mode = fat_make_mode(sbi, de->attr,
481                         ((sbi->options.showexec && !is_exec(de->name + 8))
482                          ? S_IRUGO|S_IWUGO : S_IRWXUGO));
483                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
484
485                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
486                 inode->i_size = le32_to_cpu(de->size);
487                 inode->i_op = &fat_file_inode_operations;
488                 inode->i_fop = &fat_file_operations;
489                 inode->i_mapping->a_ops = &fat_aops;
490                 MSDOS_I(inode)->mmu_private = inode->i_size;
491         }
492         if (de->attr & ATTR_SYS) {
493                 if (sbi->options.sys_immutable)
494                         inode->i_flags |= S_IMMUTABLE;
495         }
496         fat_save_attrs(inode, de->attr);
497
498         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
499                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
500
501         fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
502         if (sbi->options.isvfat) {
503                 fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
504                                   de->cdate, de->ctime_cs);
505                 fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
506         } else
507                 inode->i_ctime = inode->i_atime = inode->i_mtime;
508
509         return 0;
510 }
511
512 static inline void fat_lock_build_inode(struct msdos_sb_info *sbi)
513 {
514         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
515                 mutex_lock(&sbi->nfs_build_inode_lock);
516 }
517
518 static inline void fat_unlock_build_inode(struct msdos_sb_info *sbi)
519 {
520         if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
521                 mutex_unlock(&sbi->nfs_build_inode_lock);
522 }
523
524 struct inode *fat_build_inode(struct super_block *sb,
525                         struct msdos_dir_entry *de, loff_t i_pos)
526 {
527         struct inode *inode;
528         int err;
529
530         fat_lock_build_inode(MSDOS_SB(sb));
531         inode = fat_iget(sb, i_pos);
532         if (inode)
533                 goto out;
534         inode = new_inode(sb);
535         if (!inode) {
536                 inode = ERR_PTR(-ENOMEM);
537                 goto out;
538         }
539         inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
540         inode->i_version = 1;
541         err = fat_fill_inode(inode, de);
542         if (err) {
543                 iput(inode);
544                 inode = ERR_PTR(err);
545                 goto out;
546         }
547         fat_attach(inode, i_pos);
548         insert_inode_hash(inode);
549 out:
550         fat_unlock_build_inode(MSDOS_SB(sb));
551         return inode;
552 }
553
554 EXPORT_SYMBOL_GPL(fat_build_inode);
555
556 static void fat_evict_inode(struct inode *inode)
557 {
558         truncate_inode_pages_final(&inode->i_data);
559         if (!inode->i_nlink) {
560                 inode->i_size = 0;
561                 fat_truncate_blocks(inode, 0);
562         }
563         invalidate_inode_buffers(inode);
564         clear_inode(inode);
565         fat_cache_inval_inode(inode);
566         fat_detach(inode);
567 }
568
569 static void fat_set_state(struct super_block *sb,
570                         unsigned int set, unsigned int force)
571 {
572         struct buffer_head *bh;
573         struct fat_boot_sector *b;
574         struct msdos_sb_info *sbi = MSDOS_SB(sb);
575
576         /* do not change any thing if mounted read only */
577         if ((sb->s_flags & MS_RDONLY) && !force)
578                 return;
579
580         /* do not change state if fs was dirty */
581         if (sbi->dirty) {
582                 /* warn only on set (mount). */
583                 if (set)
584                         fat_msg(sb, KERN_WARNING, "Volume was not properly "
585                                 "unmounted. Some data may be corrupt. "
586                                 "Please run fsck.");
587                 return;
588         }
589
590         bh = sb_bread(sb, 0);
591         if (bh == NULL) {
592                 fat_msg(sb, KERN_ERR, "unable to read boot sector "
593                         "to mark fs as dirty");
594                 return;
595         }
596
597         b = (struct fat_boot_sector *) bh->b_data;
598
599         if (sbi->fat_bits == 32) {
600                 if (set)
601                         b->fat32.state |= FAT_STATE_DIRTY;
602                 else
603                         b->fat32.state &= ~FAT_STATE_DIRTY;
604         } else /* fat 16 and 12 */ {
605                 if (set)
606                         b->fat16.state |= FAT_STATE_DIRTY;
607                 else
608                         b->fat16.state &= ~FAT_STATE_DIRTY;
609         }
610
611         mark_buffer_dirty(bh);
612         sync_dirty_buffer(bh);
613         brelse(bh);
614 }
615
616 static void fat_reset_iocharset(struct fat_mount_options *opts)
617 {
618         if (opts->iocharset != fat_default_iocharset) {
619                 /* Note: opts->iocharset can be NULL here */
620                 kfree(opts->iocharset);
621                 opts->iocharset = fat_default_iocharset;
622         }
623 }
624
625 static void delayed_free(struct rcu_head *p)
626 {
627         struct msdos_sb_info *sbi = container_of(p, struct msdos_sb_info, rcu);
628         unload_nls(sbi->nls_disk);
629         unload_nls(sbi->nls_io);
630         fat_reset_iocharset(&sbi->options);
631         kfree(sbi);
632 }
633
634 static void fat_put_super(struct super_block *sb)
635 {
636         struct msdos_sb_info *sbi = MSDOS_SB(sb);
637
638         fat_set_state(sb, 0, 0);
639
640         iput(sbi->fsinfo_inode);
641         iput(sbi->fat_inode);
642
643         call_rcu(&sbi->rcu, delayed_free);
644 }
645
646 static struct kmem_cache *fat_inode_cachep;
647
648 static struct inode *fat_alloc_inode(struct super_block *sb)
649 {
650         struct msdos_inode_info *ei;
651         ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
652         if (!ei)
653                 return NULL;
654
655         init_rwsem(&ei->truncate_lock);
656         /* Zeroing to allow iput() even if partial initialized inode. */
657         ei->mmu_private = 0;
658         ei->i_start = 0;
659         ei->i_logstart = 0;
660         ei->i_attrs = 0;
661         ei->i_pos = 0;
662
663         return &ei->vfs_inode;
664 }
665
666 static void fat_i_callback(struct rcu_head *head)
667 {
668         struct inode *inode = container_of(head, struct inode, i_rcu);
669         kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
670 }
671
672 static void fat_destroy_inode(struct inode *inode)
673 {
674         call_rcu(&inode->i_rcu, fat_i_callback);
675 }
676
677 static void init_once(void *foo)
678 {
679         struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
680
681         spin_lock_init(&ei->cache_lru_lock);
682         ei->nr_caches = 0;
683         ei->cache_valid_id = FAT_CACHE_VALID + 1;
684         INIT_LIST_HEAD(&ei->cache_lru);
685         INIT_HLIST_NODE(&ei->i_fat_hash);
686         INIT_HLIST_NODE(&ei->i_dir_hash);
687         inode_init_once(&ei->vfs_inode);
688 }
689
690 static int __init fat_init_inodecache(void)
691 {
692         fat_inode_cachep = kmem_cache_create("fat_inode_cache",
693                                              sizeof(struct msdos_inode_info),
694                                              0, (SLAB_RECLAIM_ACCOUNT|
695                                                 SLAB_MEM_SPREAD),
696                                              init_once);
697         if (fat_inode_cachep == NULL)
698                 return -ENOMEM;
699         return 0;
700 }
701
702 static void __exit fat_destroy_inodecache(void)
703 {
704         /*
705          * Make sure all delayed rcu free inodes are flushed before we
706          * destroy cache.
707          */
708         rcu_barrier();
709         kmem_cache_destroy(fat_inode_cachep);
710 }
711
712 static int fat_remount(struct super_block *sb, int *flags, char *data)
713 {
714         int new_rdonly;
715         struct msdos_sb_info *sbi = MSDOS_SB(sb);
716         *flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
717
718         sync_filesystem(sb);
719
720         /* make sure we update state on remount. */
721         new_rdonly = *flags & MS_RDONLY;
722         if (new_rdonly != (sb->s_flags & MS_RDONLY)) {
723                 if (new_rdonly)
724                         fat_set_state(sb, 0, 0);
725                 else
726                         fat_set_state(sb, 1, 1);
727         }
728         return 0;
729 }
730
731 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
732 {
733         struct super_block *sb = dentry->d_sb;
734         struct msdos_sb_info *sbi = MSDOS_SB(sb);
735         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
736
737         /* If the count of free cluster is still unknown, counts it here. */
738         if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
739                 int err = fat_count_free_clusters(dentry->d_sb);
740                 if (err)
741                         return err;
742         }
743
744         buf->f_type = dentry->d_sb->s_magic;
745         buf->f_bsize = sbi->cluster_size;
746         buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
747         buf->f_bfree = sbi->free_clusters;
748         buf->f_bavail = sbi->free_clusters;
749         buf->f_fsid.val[0] = (u32)id;
750         buf->f_fsid.val[1] = (u32)(id >> 32);
751         buf->f_namelen =
752                 (sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
753
754         return 0;
755 }
756
757 static int __fat_write_inode(struct inode *inode, int wait)
758 {
759         struct super_block *sb = inode->i_sb;
760         struct msdos_sb_info *sbi = MSDOS_SB(sb);
761         struct buffer_head *bh;
762         struct msdos_dir_entry *raw_entry;
763         loff_t i_pos;
764         sector_t blocknr;
765         int err, offset;
766
767         if (inode->i_ino == MSDOS_ROOT_INO)
768                 return 0;
769
770 retry:
771         i_pos = fat_i_pos_read(sbi, inode);
772         if (!i_pos)
773                 return 0;
774
775         fat_get_blknr_offset(sbi, i_pos, &blocknr, &offset);
776         bh = sb_bread(sb, blocknr);
777         if (!bh) {
778                 fat_msg(sb, KERN_ERR, "unable to read inode block "
779                        "for updating (i_pos %lld)", i_pos);
780                 return -EIO;
781         }
782         spin_lock(&sbi->inode_hash_lock);
783         if (i_pos != MSDOS_I(inode)->i_pos) {
784                 spin_unlock(&sbi->inode_hash_lock);
785                 brelse(bh);
786                 goto retry;
787         }
788
789         raw_entry = &((struct msdos_dir_entry *) (bh->b_data))[offset];
790         if (S_ISDIR(inode->i_mode))
791                 raw_entry->size = 0;
792         else
793                 raw_entry->size = cpu_to_le32(inode->i_size);
794         raw_entry->attr = fat_make_attrs(inode);
795         fat_set_start(raw_entry, MSDOS_I(inode)->i_logstart);
796         fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
797                           &raw_entry->date, NULL);
798         if (sbi->options.isvfat) {
799                 __le16 atime;
800                 fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
801                                   &raw_entry->cdate, &raw_entry->ctime_cs);
802                 fat_time_unix2fat(sbi, &inode->i_atime, &atime,
803                                   &raw_entry->adate, NULL);
804         }
805         spin_unlock(&sbi->inode_hash_lock);
806         mark_buffer_dirty(bh);
807         err = 0;
808         if (wait)
809                 err = sync_dirty_buffer(bh);
810         brelse(bh);
811         return err;
812 }
813
814 static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
815 {
816         int err;
817
818         if (inode->i_ino == MSDOS_FSINFO_INO) {
819                 struct super_block *sb = inode->i_sb;
820
821                 mutex_lock(&MSDOS_SB(sb)->s_lock);
822                 err = fat_clusters_flush(sb);
823                 mutex_unlock(&MSDOS_SB(sb)->s_lock);
824         } else
825                 err = __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
826
827         return err;
828 }
829
830 int fat_sync_inode(struct inode *inode)
831 {
832         return __fat_write_inode(inode, 1);
833 }
834
835 EXPORT_SYMBOL_GPL(fat_sync_inode);
836
837 static int fat_show_options(struct seq_file *m, struct dentry *root);
838 static const struct super_operations fat_sops = {
839         .alloc_inode    = fat_alloc_inode,
840         .destroy_inode  = fat_destroy_inode,
841         .write_inode    = fat_write_inode,
842         .evict_inode    = fat_evict_inode,
843         .put_super      = fat_put_super,
844         .statfs         = fat_statfs,
845         .remount_fs     = fat_remount,
846
847         .show_options   = fat_show_options,
848 };
849
850 static int fat_show_options(struct seq_file *m, struct dentry *root)
851 {
852         struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
853         struct fat_mount_options *opts = &sbi->options;
854         int isvfat = opts->isvfat;
855
856         if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
857                 seq_printf(m, ",uid=%u",
858                                 from_kuid_munged(&init_user_ns, opts->fs_uid));
859         if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
860                 seq_printf(m, ",gid=%u",
861                                 from_kgid_munged(&init_user_ns, opts->fs_gid));
862         seq_printf(m, ",fmask=%04o", opts->fs_fmask);
863         seq_printf(m, ",dmask=%04o", opts->fs_dmask);
864         if (opts->allow_utime)
865                 seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
866         if (sbi->nls_disk)
867                 /* strip "cp" prefix from displayed option */
868                 seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
869         if (isvfat) {
870                 if (sbi->nls_io)
871                         seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
872
873                 switch (opts->shortname) {
874                 case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
875                         seq_puts(m, ",shortname=win95");
876                         break;
877                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
878                         seq_puts(m, ",shortname=winnt");
879                         break;
880                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
881                         seq_puts(m, ",shortname=mixed");
882                         break;
883                 case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
884                         seq_puts(m, ",shortname=lower");
885                         break;
886                 default:
887                         seq_puts(m, ",shortname=unknown");
888                         break;
889                 }
890         }
891         if (opts->name_check != 'n')
892                 seq_printf(m, ",check=%c", opts->name_check);
893         if (opts->usefree)
894                 seq_puts(m, ",usefree");
895         if (opts->quiet)
896                 seq_puts(m, ",quiet");
897         if (opts->showexec)
898                 seq_puts(m, ",showexec");
899         if (opts->sys_immutable)
900                 seq_puts(m, ",sys_immutable");
901         if (!isvfat) {
902                 if (opts->dotsOK)
903                         seq_puts(m, ",dotsOK=yes");
904                 if (opts->nocase)
905                         seq_puts(m, ",nocase");
906         } else {
907                 if (opts->utf8)
908                         seq_puts(m, ",utf8");
909                 if (opts->unicode_xlate)
910                         seq_puts(m, ",uni_xlate");
911                 if (!opts->numtail)
912                         seq_puts(m, ",nonumtail");
913                 if (opts->rodir)
914                         seq_puts(m, ",rodir");
915         }
916         if (opts->flush)
917                 seq_puts(m, ",flush");
918         if (opts->tz_set) {
919                 if (opts->time_offset)
920                         seq_printf(m, ",time_offset=%d", opts->time_offset);
921                 else
922                         seq_puts(m, ",tz=UTC");
923         }
924         if (opts->errors == FAT_ERRORS_CONT)
925                 seq_puts(m, ",errors=continue");
926         else if (opts->errors == FAT_ERRORS_PANIC)
927                 seq_puts(m, ",errors=panic");
928         else
929                 seq_puts(m, ",errors=remount-ro");
930         if (opts->nfs == FAT_NFS_NOSTALE_RO)
931                 seq_puts(m, ",nfs=nostale_ro");
932         else if (opts->nfs)
933                 seq_puts(m, ",nfs=stale_rw");
934         if (opts->discard)
935                 seq_puts(m, ",discard");
936         if (opts->dos1xfloppy)
937                 seq_puts(m, ",dos1xfloppy");
938
939         return 0;
940 }
941
942 enum {
943         Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
944         Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
945         Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
946         Opt_immutable, Opt_dots, Opt_nodots,
947         Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
948         Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
949         Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
950         Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
951         Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
952         Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
953 };
954
955 static const match_table_t fat_tokens = {
956         {Opt_check_r, "check=relaxed"},
957         {Opt_check_s, "check=strict"},
958         {Opt_check_n, "check=normal"},
959         {Opt_check_r, "check=r"},
960         {Opt_check_s, "check=s"},
961         {Opt_check_n, "check=n"},
962         {Opt_uid, "uid=%u"},
963         {Opt_gid, "gid=%u"},
964         {Opt_umask, "umask=%o"},
965         {Opt_dmask, "dmask=%o"},
966         {Opt_fmask, "fmask=%o"},
967         {Opt_allow_utime, "allow_utime=%o"},
968         {Opt_codepage, "codepage=%u"},
969         {Opt_usefree, "usefree"},
970         {Opt_nocase, "nocase"},
971         {Opt_quiet, "quiet"},
972         {Opt_showexec, "showexec"},
973         {Opt_debug, "debug"},
974         {Opt_immutable, "sys_immutable"},
975         {Opt_flush, "flush"},
976         {Opt_tz_utc, "tz=UTC"},
977         {Opt_time_offset, "time_offset=%d"},
978         {Opt_err_cont, "errors=continue"},
979         {Opt_err_panic, "errors=panic"},
980         {Opt_err_ro, "errors=remount-ro"},
981         {Opt_discard, "discard"},
982         {Opt_nfs_stale_rw, "nfs"},
983         {Opt_nfs_stale_rw, "nfs=stale_rw"},
984         {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
985         {Opt_dos1xfloppy, "dos1xfloppy"},
986         {Opt_obsolete, "conv=binary"},
987         {Opt_obsolete, "conv=text"},
988         {Opt_obsolete, "conv=auto"},
989         {Opt_obsolete, "conv=b"},
990         {Opt_obsolete, "conv=t"},
991         {Opt_obsolete, "conv=a"},
992         {Opt_obsolete, "fat=%u"},
993         {Opt_obsolete, "blocksize=%u"},
994         {Opt_obsolete, "cvf_format=%20s"},
995         {Opt_obsolete, "cvf_options=%100s"},
996         {Opt_obsolete, "posix"},
997         {Opt_err, NULL},
998 };
999 static const match_table_t msdos_tokens = {
1000         {Opt_nodots, "nodots"},
1001         {Opt_nodots, "dotsOK=no"},
1002         {Opt_dots, "dots"},
1003         {Opt_dots, "dotsOK=yes"},
1004         {Opt_err, NULL}
1005 };
1006 static const match_table_t vfat_tokens = {
1007         {Opt_charset, "iocharset=%s"},
1008         {Opt_shortname_lower, "shortname=lower"},
1009         {Opt_shortname_win95, "shortname=win95"},
1010         {Opt_shortname_winnt, "shortname=winnt"},
1011         {Opt_shortname_mixed, "shortname=mixed"},
1012         {Opt_utf8_no, "utf8=0"},                /* 0 or no or false */
1013         {Opt_utf8_no, "utf8=no"},
1014         {Opt_utf8_no, "utf8=false"},
1015         {Opt_utf8_yes, "utf8=1"},               /* empty or 1 or yes or true */
1016         {Opt_utf8_yes, "utf8=yes"},
1017         {Opt_utf8_yes, "utf8=true"},
1018         {Opt_utf8_yes, "utf8"},
1019         {Opt_uni_xl_no, "uni_xlate=0"},         /* 0 or no or false */
1020         {Opt_uni_xl_no, "uni_xlate=no"},
1021         {Opt_uni_xl_no, "uni_xlate=false"},
1022         {Opt_uni_xl_yes, "uni_xlate=1"},        /* empty or 1 or yes or true */
1023         {Opt_uni_xl_yes, "uni_xlate=yes"},
1024         {Opt_uni_xl_yes, "uni_xlate=true"},
1025         {Opt_uni_xl_yes, "uni_xlate"},
1026         {Opt_nonumtail_no, "nonumtail=0"},      /* 0 or no or false */
1027         {Opt_nonumtail_no, "nonumtail=no"},
1028         {Opt_nonumtail_no, "nonumtail=false"},
1029         {Opt_nonumtail_yes, "nonumtail=1"},     /* empty or 1 or yes or true */
1030         {Opt_nonumtail_yes, "nonumtail=yes"},
1031         {Opt_nonumtail_yes, "nonumtail=true"},
1032         {Opt_nonumtail_yes, "nonumtail"},
1033         {Opt_rodir, "rodir"},
1034         {Opt_err, NULL}
1035 };
1036
1037 static int parse_options(struct super_block *sb, char *options, int is_vfat,
1038                          int silent, int *debug, struct fat_mount_options *opts)
1039 {
1040         char *p;
1041         substring_t args[MAX_OPT_ARGS];
1042         int option;
1043         char *iocharset;
1044
1045         opts->isvfat = is_vfat;
1046
1047         opts->fs_uid = current_uid();
1048         opts->fs_gid = current_gid();
1049         opts->fs_fmask = opts->fs_dmask = current_umask();
1050         opts->allow_utime = -1;
1051         opts->codepage = fat_default_codepage;
1052         fat_reset_iocharset(opts);
1053         if (is_vfat) {
1054                 opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1055                 opts->rodir = 0;
1056         } else {
1057                 opts->shortname = 0;
1058                 opts->rodir = 1;
1059         }
1060         opts->name_check = 'n';
1061         opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
1062         opts->utf8 = opts->unicode_xlate = 0;
1063         opts->numtail = 1;
1064         opts->usefree = opts->nocase = 0;
1065         opts->tz_set = 0;
1066         opts->nfs = 0;
1067         opts->errors = FAT_ERRORS_RO;
1068         *debug = 0;
1069
1070         if (!options)
1071                 goto out;
1072
1073         while ((p = strsep(&options, ",")) != NULL) {
1074                 int token;
1075                 if (!*p)
1076                         continue;
1077
1078                 token = match_token(p, fat_tokens, args);
1079                 if (token == Opt_err) {
1080                         if (is_vfat)
1081                                 token = match_token(p, vfat_tokens, args);
1082                         else
1083                                 token = match_token(p, msdos_tokens, args);
1084                 }
1085                 switch (token) {
1086                 case Opt_check_s:
1087                         opts->name_check = 's';
1088                         break;
1089                 case Opt_check_r:
1090                         opts->name_check = 'r';
1091                         break;
1092                 case Opt_check_n:
1093                         opts->name_check = 'n';
1094                         break;
1095                 case Opt_usefree:
1096                         opts->usefree = 1;
1097                         break;
1098                 case Opt_nocase:
1099                         if (!is_vfat)
1100                                 opts->nocase = 1;
1101                         else {
1102                                 /* for backward compatibility */
1103                                 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1104                                         | VFAT_SFN_CREATE_WIN95;
1105                         }
1106                         break;
1107                 case Opt_quiet:
1108                         opts->quiet = 1;
1109                         break;
1110                 case Opt_showexec:
1111                         opts->showexec = 1;
1112                         break;
1113                 case Opt_debug:
1114                         *debug = 1;
1115                         break;
1116                 case Opt_immutable:
1117                         opts->sys_immutable = 1;
1118                         break;
1119                 case Opt_uid:
1120                         if (match_int(&args[0], &option))
1121                                 return -EINVAL;
1122                         opts->fs_uid = make_kuid(current_user_ns(), option);
1123                         if (!uid_valid(opts->fs_uid))
1124                                 return -EINVAL;
1125                         break;
1126                 case Opt_gid:
1127                         if (match_int(&args[0], &option))
1128                                 return -EINVAL;
1129                         opts->fs_gid = make_kgid(current_user_ns(), option);
1130                         if (!gid_valid(opts->fs_gid))
1131                                 return -EINVAL;
1132                         break;
1133                 case Opt_umask:
1134                         if (match_octal(&args[0], &option))
1135                                 return -EINVAL;
1136                         opts->fs_fmask = opts->fs_dmask = option;
1137                         break;
1138                 case Opt_dmask:
1139                         if (match_octal(&args[0], &option))
1140                                 return -EINVAL;
1141                         opts->fs_dmask = option;
1142                         break;
1143                 case Opt_fmask:
1144                         if (match_octal(&args[0], &option))
1145                                 return -EINVAL;
1146                         opts->fs_fmask = option;
1147                         break;
1148                 case Opt_allow_utime:
1149                         if (match_octal(&args[0], &option))
1150                                 return -EINVAL;
1151                         opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1152                         break;
1153                 case Opt_codepage:
1154                         if (match_int(&args[0], &option))
1155                                 return -EINVAL;
1156                         opts->codepage = option;
1157                         break;
1158                 case Opt_flush:
1159                         opts->flush = 1;
1160                         break;
1161                 case Opt_time_offset:
1162                         if (match_int(&args[0], &option))
1163                                 return -EINVAL;
1164                         if (option < -12 * 60 || option > 12 * 60)
1165                                 return -EINVAL;
1166                         opts->tz_set = 1;
1167                         opts->time_offset = option;
1168                         break;
1169                 case Opt_tz_utc:
1170                         opts->tz_set = 1;
1171                         opts->time_offset = 0;
1172                         break;
1173                 case Opt_err_cont:
1174                         opts->errors = FAT_ERRORS_CONT;
1175                         break;
1176                 case Opt_err_panic:
1177                         opts->errors = FAT_ERRORS_PANIC;
1178                         break;
1179                 case Opt_err_ro:
1180                         opts->errors = FAT_ERRORS_RO;
1181                         break;
1182                 case Opt_nfs_stale_rw:
1183                         opts->nfs = FAT_NFS_STALE_RW;
1184                         break;
1185                 case Opt_nfs_nostale_ro:
1186                         opts->nfs = FAT_NFS_NOSTALE_RO;
1187                         break;
1188                 case Opt_dos1xfloppy:
1189                         opts->dos1xfloppy = 1;
1190                         break;
1191
1192                 /* msdos specific */
1193                 case Opt_dots:
1194                         opts->dotsOK = 1;
1195                         break;
1196                 case Opt_nodots:
1197                         opts->dotsOK = 0;
1198                         break;
1199
1200                 /* vfat specific */
1201                 case Opt_charset:
1202                         fat_reset_iocharset(opts);
1203                         iocharset = match_strdup(&args[0]);
1204                         if (!iocharset)
1205                                 return -ENOMEM;
1206                         opts->iocharset = iocharset;
1207                         break;
1208                 case Opt_shortname_lower:
1209                         opts->shortname = VFAT_SFN_DISPLAY_LOWER
1210                                         | VFAT_SFN_CREATE_WIN95;
1211                         break;
1212                 case Opt_shortname_win95:
1213                         opts->shortname = VFAT_SFN_DISPLAY_WIN95
1214                                         | VFAT_SFN_CREATE_WIN95;
1215                         break;
1216                 case Opt_shortname_winnt:
1217                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1218                                         | VFAT_SFN_CREATE_WINNT;
1219                         break;
1220                 case Opt_shortname_mixed:
1221                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1222                                         | VFAT_SFN_CREATE_WIN95;
1223                         break;
1224                 case Opt_utf8_no:               /* 0 or no or false */
1225                         opts->utf8 = 0;
1226                         break;
1227                 case Opt_utf8_yes:              /* empty or 1 or yes or true */
1228                         opts->utf8 = 1;
1229                         break;
1230                 case Opt_uni_xl_no:             /* 0 or no or false */
1231                         opts->unicode_xlate = 0;
1232                         break;
1233                 case Opt_uni_xl_yes:            /* empty or 1 or yes or true */
1234                         opts->unicode_xlate = 1;
1235                         break;
1236                 case Opt_nonumtail_no:          /* 0 or no or false */
1237                         opts->numtail = 1;      /* negated option */
1238                         break;
1239                 case Opt_nonumtail_yes:         /* empty or 1 or yes or true */
1240                         opts->numtail = 0;      /* negated option */
1241                         break;
1242                 case Opt_rodir:
1243                         opts->rodir = 1;
1244                         break;
1245                 case Opt_discard:
1246                         opts->discard = 1;
1247                         break;
1248
1249                 /* obsolete mount options */
1250                 case Opt_obsolete:
1251                         fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1252                                "not supported now", p);
1253                         break;
1254                 /* unknown option */
1255                 default:
1256                         if (!silent) {
1257                                 fat_msg(sb, KERN_ERR,
1258                                        "Unrecognized mount option \"%s\" "
1259                                        "or missing value", p);
1260                         }
1261                         return -EINVAL;
1262                 }
1263         }
1264
1265 out:
1266         /* UTF-8 doesn't provide FAT semantics */
1267         if (!strcmp(opts->iocharset, "utf8")) {
1268                 fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1269                        " for FAT filesystems, filesystem will be "
1270                        "case sensitive!");
1271         }
1272
1273         /* If user doesn't specify allow_utime, it's initialized from dmask. */
1274         if (opts->allow_utime == (unsigned short)-1)
1275                 opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1276         if (opts->unicode_xlate)
1277                 opts->utf8 = 0;
1278         if (opts->nfs == FAT_NFS_NOSTALE_RO) {
1279                 sb->s_flags |= MS_RDONLY;
1280                 sb->s_export_op = &fat_export_ops_nostale;
1281         }
1282
1283         return 0;
1284 }
1285
1286 static int fat_read_root(struct inode *inode)
1287 {
1288         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
1289         int error;
1290
1291         MSDOS_I(inode)->i_pos = MSDOS_ROOT_INO;
1292         inode->i_uid = sbi->options.fs_uid;
1293         inode->i_gid = sbi->options.fs_gid;
1294         inode->i_version++;
1295         inode->i_generation = 0;
1296         inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1297         inode->i_op = sbi->dir_ops;
1298         inode->i_fop = &fat_dir_operations;
1299         if (sbi->fat_bits == 32) {
1300                 MSDOS_I(inode)->i_start = sbi->root_cluster;
1301                 error = fat_calc_dir_size(inode);
1302                 if (error < 0)
1303                         return error;
1304         } else {
1305                 MSDOS_I(inode)->i_start = 0;
1306                 inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1307         }
1308         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1309                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1310         MSDOS_I(inode)->i_logstart = 0;
1311         MSDOS_I(inode)->mmu_private = inode->i_size;
1312
1313         fat_save_attrs(inode, ATTR_DIR);
1314         inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1315         inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1316         set_nlink(inode, fat_subdirs(inode)+2);
1317
1318         return 0;
1319 }
1320
1321 static unsigned long calc_fat_clusters(struct super_block *sb)
1322 {
1323         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1324
1325         /* Divide first to avoid overflow */
1326         if (sbi->fat_bits != 12) {
1327                 unsigned long ent_per_sec = sb->s_blocksize * 8 / sbi->fat_bits;
1328                 return ent_per_sec * sbi->fat_length;
1329         }
1330
1331         return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1332 }
1333
1334 static bool fat_bpb_is_zero(struct fat_boot_sector *b)
1335 {
1336         if (get_unaligned_le16(&b->sector_size))
1337                 return false;
1338         if (b->sec_per_clus)
1339                 return false;
1340         if (b->reserved)
1341                 return false;
1342         if (b->fats)
1343                 return false;
1344         if (get_unaligned_le16(&b->dir_entries))
1345                 return false;
1346         if (get_unaligned_le16(&b->sectors))
1347                 return false;
1348         if (b->media)
1349                 return false;
1350         if (b->fat_length)
1351                 return false;
1352         if (b->secs_track)
1353                 return false;
1354         if (b->heads)
1355                 return false;
1356         return true;
1357 }
1358
1359 static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
1360         int silent, struct fat_bios_param_block *bpb)
1361 {
1362         int error = -EINVAL;
1363
1364         /* Read in BPB ... */
1365         memset(bpb, 0, sizeof(*bpb));
1366         bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
1367         bpb->fat_sec_per_clus = b->sec_per_clus;
1368         bpb->fat_reserved = le16_to_cpu(b->reserved);
1369         bpb->fat_fats = b->fats;
1370         bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
1371         bpb->fat_sectors = get_unaligned_le16(&b->sectors);
1372         bpb->fat_fat_length = le16_to_cpu(b->fat_length);
1373         bpb->fat_total_sect = le32_to_cpu(b->total_sect);
1374
1375         bpb->fat16_state = b->fat16.state;
1376         bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
1377
1378         bpb->fat32_length = le32_to_cpu(b->fat32.length);
1379         bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
1380         bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
1381         bpb->fat32_state = b->fat32.state;
1382         bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
1383
1384         /* Validate this looks like a FAT filesystem BPB */
1385         if (!bpb->fat_reserved) {
1386                 if (!silent)
1387                         fat_msg(sb, KERN_ERR,
1388                                 "bogus number of reserved sectors");
1389                 goto out;
1390         }
1391         if (!bpb->fat_fats) {
1392                 if (!silent)
1393                         fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1394                 goto out;
1395         }
1396
1397         /*
1398          * Earlier we checked here that b->secs_track and b->head are nonzero,
1399          * but it turns out valid FAT filesystems can have zero there.
1400          */
1401
1402         if (!fat_valid_media(b->media)) {
1403                 if (!silent)
1404                         fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1405                                 (unsigned)b->media);
1406                 goto out;
1407         }
1408
1409         if (!is_power_of_2(bpb->fat_sector_size)
1410             || (bpb->fat_sector_size < 512)
1411             || (bpb->fat_sector_size > 4096)) {
1412                 if (!silent)
1413                         fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1414                                (unsigned)bpb->fat_sector_size);
1415                 goto out;
1416         }
1417
1418         if (!is_power_of_2(bpb->fat_sec_per_clus)) {
1419                 if (!silent)
1420                         fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1421                                 (unsigned)bpb->fat_sec_per_clus);
1422                 goto out;
1423         }
1424
1425         if (bpb->fat_fat_length == 0 && bpb->fat32_length == 0) {
1426                 if (!silent)
1427                         fat_msg(sb, KERN_ERR, "bogus number of FAT sectors");
1428                 goto out;
1429         }
1430
1431         error = 0;
1432
1433 out:
1434         return error;
1435 }
1436
1437 static int fat_read_static_bpb(struct super_block *sb,
1438         struct fat_boot_sector *b, int silent,
1439         struct fat_bios_param_block *bpb)
1440 {
1441         static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
1442
1443         struct fat_floppy_defaults *fdefaults = NULL;
1444         int error = -EINVAL;
1445         sector_t bd_sects;
1446         unsigned i;
1447
1448         bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
1449
1450         /* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
1451         if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
1452                 if (!silent)
1453                         fat_msg(sb, KERN_ERR,
1454                                 "%s; no bootstrapping code", notdos1x);
1455                 goto out;
1456         }
1457
1458         /*
1459          * If any value in this region is non-zero, it isn't archaic
1460          * DOS.
1461          */
1462         if (!fat_bpb_is_zero(b)) {
1463                 if (!silent)
1464                         fat_msg(sb, KERN_ERR,
1465                                 "%s; DOS 2.x BPB is non-zero", notdos1x);
1466                 goto out;
1467         }
1468
1469         for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
1470                 if (floppy_defaults[i].nr_sectors == bd_sects) {
1471                         fdefaults = &floppy_defaults[i];
1472                         break;
1473                 }
1474         }
1475
1476         if (fdefaults == NULL) {
1477                 if (!silent)
1478                         fat_msg(sb, KERN_WARNING,
1479                                 "This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
1480                                 (u64)bd_sects);
1481                 goto out;
1482         }
1483
1484         if (!silent)
1485                 fat_msg(sb, KERN_INFO,
1486                         "This looks like a DOS 1.x volume; assuming default BPB values");
1487
1488         memset(bpb, 0, sizeof(*bpb));
1489         bpb->fat_sector_size = SECTOR_SIZE;
1490         bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
1491         bpb->fat_reserved = 1;
1492         bpb->fat_fats = 2;
1493         bpb->fat_dir_entries = fdefaults->dir_entries;
1494         bpb->fat_sectors = fdefaults->nr_sectors;
1495         bpb->fat_fat_length = fdefaults->fat_length;
1496
1497         error = 0;
1498
1499 out:
1500         return error;
1501 }
1502
1503 /*
1504  * Read the super block of an MS-DOS FS.
1505  */
1506 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1507                    void (*setup)(struct super_block *))
1508 {
1509         struct inode *root_inode = NULL, *fat_inode = NULL;
1510         struct inode *fsinfo_inode = NULL;
1511         struct buffer_head *bh;
1512         struct fat_bios_param_block bpb;
1513         struct msdos_sb_info *sbi;
1514         u16 logical_sector_size;
1515         u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1516         int debug;
1517         long error;
1518         char buf[50];
1519
1520         /*
1521          * GFP_KERNEL is ok here, because while we do hold the
1522          * supeblock lock, memory pressure can't call back into
1523          * the filesystem, since we're only just about to mount
1524          * it and have no inodes etc active!
1525          */
1526         sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1527         if (!sbi)
1528                 return -ENOMEM;
1529         sb->s_fs_info = sbi;
1530
1531         sb->s_flags |= MS_NODIRATIME;
1532         sb->s_magic = MSDOS_SUPER_MAGIC;
1533         sb->s_op = &fat_sops;
1534         sb->s_export_op = &fat_export_ops;
1535         mutex_init(&sbi->nfs_build_inode_lock);
1536         ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1537                              DEFAULT_RATELIMIT_BURST);
1538
1539         error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1540         if (error)
1541                 goto out_fail;
1542
1543         setup(sb); /* flavour-specific stuff that needs options */
1544
1545         error = -EIO;
1546         sb_min_blocksize(sb, 512);
1547         bh = sb_bread(sb, 0);
1548         if (bh == NULL) {
1549                 fat_msg(sb, KERN_ERR, "unable to read boot sector");
1550                 goto out_fail;
1551         }
1552
1553         error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
1554                 &bpb);
1555         if (error == -EINVAL && sbi->options.dos1xfloppy)
1556                 error = fat_read_static_bpb(sb,
1557                         (struct fat_boot_sector *)bh->b_data, silent, &bpb);
1558         brelse(bh);
1559
1560         if (error == -EINVAL)
1561                 goto out_invalid;
1562         else if (error)
1563                 goto out_fail;
1564
1565         logical_sector_size = bpb.fat_sector_size;
1566         sbi->sec_per_clus = bpb.fat_sec_per_clus;
1567
1568         error = -EIO;
1569         if (logical_sector_size < sb->s_blocksize) {
1570                 fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1571                        " (logical sector size = %u)", logical_sector_size);
1572                 goto out_fail;
1573         }
1574
1575         if (logical_sector_size > sb->s_blocksize) {
1576                 struct buffer_head *bh_resize;
1577
1578                 if (!sb_set_blocksize(sb, logical_sector_size)) {
1579                         fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1580                                logical_sector_size);
1581                         goto out_fail;
1582                 }
1583
1584                 /* Verify that the larger boot sector is fully readable */
1585                 bh_resize = sb_bread(sb, 0);
1586                 if (bh_resize == NULL) {
1587                         fat_msg(sb, KERN_ERR, "unable to read boot sector"
1588                                " (logical sector size = %lu)",
1589                                sb->s_blocksize);
1590                         goto out_fail;
1591                 }
1592                 brelse(bh_resize);
1593         }
1594
1595         mutex_init(&sbi->s_lock);
1596         sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1597         sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1598         sbi->fats = bpb.fat_fats;
1599         sbi->fat_bits = 0;              /* Don't know yet */
1600         sbi->fat_start = bpb.fat_reserved;
1601         sbi->fat_length = bpb.fat_fat_length;
1602         sbi->root_cluster = 0;
1603         sbi->free_clusters = -1;        /* Don't know yet */
1604         sbi->free_clus_valid = 0;
1605         sbi->prev_free = FAT_START_ENT;
1606         sb->s_maxbytes = 0xffffffff;
1607
1608         if (!sbi->fat_length && bpb.fat32_length) {
1609                 struct fat_boot_fsinfo *fsinfo;
1610                 struct buffer_head *fsinfo_bh;
1611
1612                 /* Must be FAT32 */
1613                 sbi->fat_bits = 32;
1614                 sbi->fat_length = bpb.fat32_length;
1615                 sbi->root_cluster = bpb.fat32_root_cluster;
1616
1617                 /* MC - if info_sector is 0, don't multiply by 0 */
1618                 sbi->fsinfo_sector = bpb.fat32_info_sector;
1619                 if (sbi->fsinfo_sector == 0)
1620                         sbi->fsinfo_sector = 1;
1621
1622                 fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1623                 if (fsinfo_bh == NULL) {
1624                         fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1625                                " (sector = %lu)", sbi->fsinfo_sector);
1626                         goto out_fail;
1627                 }
1628
1629                 fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1630                 if (!IS_FSINFO(fsinfo)) {
1631                         fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1632                                "0x%08x, 0x%08x (sector = %lu)",
1633                                le32_to_cpu(fsinfo->signature1),
1634                                le32_to_cpu(fsinfo->signature2),
1635                                sbi->fsinfo_sector);
1636                 } else {
1637                         if (sbi->options.usefree)
1638                                 sbi->free_clus_valid = 1;
1639                         sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1640                         sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1641                 }
1642
1643                 brelse(fsinfo_bh);
1644         }
1645
1646         /* interpret volume ID as a little endian 32 bit integer */
1647         if (sbi->fat_bits == 32)
1648                 sbi->vol_id = bpb.fat32_vol_id;
1649         else /* fat 16 or 12 */
1650                 sbi->vol_id = bpb.fat16_vol_id;
1651
1652         sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1653         sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1654
1655         sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1656         sbi->dir_entries = bpb.fat_dir_entries;
1657         if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1658                 if (!silent)
1659                         fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
1660                                " (%u)", sbi->dir_entries);
1661                 goto out_invalid;
1662         }
1663
1664         rootdir_sectors = sbi->dir_entries
1665                 * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1666         sbi->data_start = sbi->dir_start + rootdir_sectors;
1667         total_sectors = bpb.fat_sectors;
1668         if (total_sectors == 0)
1669                 total_sectors = bpb.fat_total_sect;
1670
1671         total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1672
1673         if (sbi->fat_bits != 32)
1674                 sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1675
1676         /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1677         if (sbi->fat_bits == 32)
1678                 sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
1679         else /* fat 16 or 12 */
1680                 sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
1681
1682         /* check that FAT table does not overflow */
1683         fat_clusters = calc_fat_clusters(sb);
1684         total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1685         if (total_clusters > MAX_FAT(sb)) {
1686                 if (!silent)
1687                         fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1688                                total_clusters);
1689                 goto out_invalid;
1690         }
1691
1692         sbi->max_cluster = total_clusters + FAT_START_ENT;
1693         /* check the free_clusters, it's not necessarily correct */
1694         if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1695                 sbi->free_clusters = -1;
1696         /* check the prev_free, it's not necessarily correct */
1697         sbi->prev_free %= sbi->max_cluster;
1698         if (sbi->prev_free < FAT_START_ENT)
1699                 sbi->prev_free = FAT_START_ENT;
1700
1701         /* set up enough so that it can read an inode */
1702         fat_hash_init(sb);
1703         dir_hash_init(sb);
1704         fat_ent_access_init(sb);
1705
1706         /*
1707          * The low byte of FAT's first entry must have same value with
1708          * media-field.  But in real world, too many devices is
1709          * writing wrong value.  So, removed that validity check.
1710          *
1711          * if (FAT_FIRST_ENT(sb, media) != first)
1712          */
1713
1714         error = -EINVAL;
1715         sprintf(buf, "cp%d", sbi->options.codepage);
1716         sbi->nls_disk = load_nls(buf);
1717         if (!sbi->nls_disk) {
1718                 fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1719                 goto out_fail;
1720         }
1721
1722         /* FIXME: utf8 is using iocharset for upper/lower conversion */
1723         if (sbi->options.isvfat) {
1724                 sbi->nls_io = load_nls(sbi->options.iocharset);
1725                 if (!sbi->nls_io) {
1726                         fat_msg(sb, KERN_ERR, "IO charset %s not found",
1727                                sbi->options.iocharset);
1728                         goto out_fail;
1729                 }
1730         }
1731
1732         error = -ENOMEM;
1733         fat_inode = new_inode(sb);
1734         if (!fat_inode)
1735                 goto out_fail;
1736         sbi->fat_inode = fat_inode;
1737
1738         fsinfo_inode = new_inode(sb);
1739         if (!fsinfo_inode)
1740                 goto out_fail;
1741         fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1742         sbi->fsinfo_inode = fsinfo_inode;
1743         insert_inode_hash(fsinfo_inode);
1744
1745         root_inode = new_inode(sb);
1746         if (!root_inode)
1747                 goto out_fail;
1748         root_inode->i_ino = MSDOS_ROOT_INO;
1749         root_inode->i_version = 1;
1750         error = fat_read_root(root_inode);
1751         if (error < 0) {
1752                 iput(root_inode);
1753                 goto out_fail;
1754         }
1755         error = -ENOMEM;
1756         insert_inode_hash(root_inode);
1757         fat_attach(root_inode, 0);
1758         sb->s_root = d_make_root(root_inode);
1759         if (!sb->s_root) {
1760                 fat_msg(sb, KERN_ERR, "get root inode failed");
1761                 goto out_fail;
1762         }
1763
1764         if (sbi->options.discard) {
1765                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1766                 if (!blk_queue_discard(q))
1767                         fat_msg(sb, KERN_WARNING,
1768                                         "mounting with \"discard\" option, but "
1769                                         "the device does not support discard");
1770         }
1771
1772         fat_set_state(sb, 1, 0);
1773         return 0;
1774
1775 out_invalid:
1776         error = -EINVAL;
1777         if (!silent)
1778                 fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1779
1780 out_fail:
1781         if (fsinfo_inode)
1782                 iput(fsinfo_inode);
1783         if (fat_inode)
1784                 iput(fat_inode);
1785         unload_nls(sbi->nls_io);
1786         unload_nls(sbi->nls_disk);
1787         fat_reset_iocharset(&sbi->options);
1788         sb->s_fs_info = NULL;
1789         kfree(sbi);
1790         return error;
1791 }
1792
1793 EXPORT_SYMBOL_GPL(fat_fill_super);
1794
1795 /*
1796  * helper function for fat_flush_inodes.  This writes both the inode
1797  * and the file data blocks, waiting for in flight data blocks before
1798  * the start of the call.  It does not wait for any io started
1799  * during the call
1800  */
1801 static int writeback_inode(struct inode *inode)
1802 {
1803
1804         int ret;
1805
1806         /* if we used wait=1, sync_inode_metadata waits for the io for the
1807         * inode to finish.  So wait=0 is sent down to sync_inode_metadata
1808         * and filemap_fdatawrite is used for the data blocks
1809         */
1810         ret = sync_inode_metadata(inode, 0);
1811         if (!ret)
1812                 ret = filemap_fdatawrite(inode->i_mapping);
1813         return ret;
1814 }
1815
1816 /*
1817  * write data and metadata corresponding to i1 and i2.  The io is
1818  * started but we do not wait for any of it to finish.
1819  *
1820  * filemap_flush is used for the block device, so if there is a dirty
1821  * page for a block already in flight, we will not wait and start the
1822  * io over again
1823  */
1824 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1825 {
1826         int ret = 0;
1827         if (!MSDOS_SB(sb)->options.flush)
1828                 return 0;
1829         if (i1)
1830                 ret = writeback_inode(i1);
1831         if (!ret && i2)
1832                 ret = writeback_inode(i2);
1833         if (!ret) {
1834                 struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1835                 ret = filemap_flush(mapping);
1836         }
1837         return ret;
1838 }
1839 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1840
1841 static int __init init_fat_fs(void)
1842 {
1843         int err;
1844
1845         err = fat_cache_init();
1846         if (err)
1847                 return err;
1848
1849         err = fat_init_inodecache();
1850         if (err)
1851                 goto failed;
1852
1853         return 0;
1854
1855 failed:
1856         fat_cache_destroy();
1857         return err;
1858 }
1859
1860 static void __exit exit_fat_fs(void)
1861 {
1862         fat_cache_destroy();
1863         fat_destroy_inodecache();
1864 }
1865
1866 module_init(init_fat_fs)
1867 module_exit(exit_fat_fs)
1868
1869 MODULE_LICENSE("GPL");