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