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