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