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 msdos_inode_info *i;
 345         struct inode *inode = NULL;
 346
 347         spin_lock(&sbi->inode_hash_lock);
 348         hlist_for_each_entry(i, head, i_fat_hash) {
 349                 BUG_ON(i->vfs_inode.i_sb != sb);
 350                 if (i->i_pos != i_pos)
 351                         continue;
 352                 inode = igrab(&i->vfs_inode);
 353                 if (inode)
 354                         break;
 355         }
 356         spin_unlock(&sbi->inode_hash_lock);
 357         return inode;
 358 }
 359
 360 static int is_exec(unsigned char *extension)
 361 {
 362         unsigned char *exe_extensions = "EXECOMBAT", *walk;
 363
 364         for (walk = exe_extensions; *walk; walk += 3)
 365                 if (!strncmp(extension, walk, 3))
 366                         return 1;
 367         return 0;
 368 }
 369
 370 static int fat_calc_dir_size(struct inode *inode)
 371 {
 372         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
 373         int ret, fclus, dclus;
 374
 375         inode->i_size = 0;
 376         if (MSDOS_I(inode)->i_start == 0)
 377                 return 0;
 378
 379         ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
 380         if (ret < 0)
 381                 return ret;
 382         inode->i_size = (fclus + 1) << sbi->cluster_bits;
 383
 384         return 0;
 385 }
 386
 387 /* doesn't deal with root inode */
 388 static int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
 389 {
 390         struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
 391         int error;
 392
 393         MSDOS_I(inode)->i_pos = 0;
 394         inode->i_uid = sbi->options.fs_uid;
 395         inode->i_gid = sbi->options.fs_gid;
 396         inode->i_version++;
 397         inode->i_generation = get_seconds();
 398
 399         if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
 400                 inode->i_generation &= ~1;
 401                 inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
 402                 inode->i_op = sbi->dir_ops;
 403                 inode->i_fop = &fat_dir_operations;
 404
 405                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
 406                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
 407                 error = fat_calc_dir_size(inode);
 408                 if (error < 0)
 409                         return error;
 410                 MSDOS_I(inode)->mmu_private = inode->i_size;
 411
 412                 set_nlink(inode, fat_subdirs(inode));
 413         } else { /* not a directory */
 414                 inode->i_generation |= 1;
 415                 inode->i_mode = fat_make_mode(sbi, de->attr,
 416                         ((sbi->options.showexec && !is_exec(de->name + 8))
 417                          ? S_IRUGO|S_IWUGO : S_IRWXUGO));
 418                 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
 419
 420                 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
 421                 inode->i_size = le32_to_cpu(de->size);
 422                 inode->i_op = &fat_file_inode_operations;
 423                 inode->i_fop = &fat_file_operations;
 424                 inode->i_mapping->a_ops = &fat_aops;
 425                 MSDOS_I(inode)->mmu_private = inode->i_size;
 426         }
 427         if (de->attr & ATTR_SYS) {
 428                 if (sbi->options.sys_immutable)
 429                         inode->i_flags |= S_IMMUTABLE;
 430         }
 431         fat_save_attrs(inode, de->attr);
 432
 433         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
 434                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
 435
 436         fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
 437         if (sbi->options.isvfat) {
 438                 fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
 439                                   de->cdate, de->ctime_cs);
 440                 fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
 441         } else
 442                 inode->i_ctime = inode->i_atime = inode->i_mtime;
 443
 444         return 0;
 445 }
 446
 447 struct inode *fat_build_inode(struct super_block *sb,
 448                         struct msdos_dir_entry *de, loff_t i_pos)
 449 {
 450         struct inode *inode;
 451         int err;
 452
 453         inode = fat_iget(sb, i_pos);
 454         if (inode)
 455                 goto out;
 456         inode = new_inode(sb);
 457         if (!inode) {
 458                 inode = ERR_PTR(-ENOMEM);
 459                 goto out;
 460         }
 461         inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
 462         inode->i_version = 1;
 463         err = fat_fill_inode(inode, de);
 464         if (err) {
 465                 iput(inode);
 466                 inode = ERR_PTR(err);
 467                 goto out;
 468         }
 469         fat_attach(inode, i_pos);
 470         insert_inode_hash(inode);
 471 out:
 472         return inode;
 473 }
 474
 475 EXPORT_SYMBOL_GPL(fat_build_inode);
 476
 477 static void fat_evict_inode(struct inode *inode)
 478 {
 479         truncate_inode_pages(&inode->i_data, 0);
 480         if (!inode->i_nlink) {
 481                 inode->i_size = 0;
 482                 fat_truncate_blocks(inode, 0);
 483         }
 484         invalidate_inode_buffers(inode);
 485         clear_inode(inode);
 486         fat_cache_inval_inode(inode);
 487         fat_detach(inode);
 488 }
 489
 490 static void fat_set_state(struct super_block *sb,
 491                         unsigned int set, unsigned int force)
 492 {
 493         struct buffer_head *bh;
 494         struct fat_boot_sector *b;
 495         struct msdos_sb_info *sbi = sb->s_fs_info;
 496
 497         /* do not change any thing if mounted read only */
 498         if ((sb->s_flags & MS_RDONLY) && !force)
 499                 return;
 500
 501         /* do not change state if fs was dirty */
 502         if (sbi->dirty) {
 503                 /* warn only on set (mount). */
 504                 if (set)
 505                         fat_msg(sb, KERN_WARNING, "Volume was not properly "
 506                                 "unmounted. Some data may be corrupt. "
 507                                 "Please run fsck.");
 508                 return;
 509         }
 510
 511         bh = sb_bread(sb, 0);
 512         if (bh == NULL) {
 513                 fat_msg(sb, KERN_ERR, "unable to read boot sector "
 514                         "to mark fs as dirty");
 515                 return;
 516         }
 517
 518         b = (struct fat_boot_sector *) bh->b_data;
 519
 520         if (sbi->fat_bits == 32) {
 521                 if (set)
 522                         b->fat32.state |= FAT_STATE_DIRTY;
 523                 else
 524                         b->fat32.state &= ~FAT_STATE_DIRTY;
 525         } else /* fat 16 and 12 */ {
 526                 if (set)
 527                         b->fat16.state |= FAT_STATE_DIRTY;
 528                 else
 529                         b->fat16.state &= ~FAT_STATE_DIRTY;
 530         }
 531
 532         mark_buffer_dirty(bh);
 533         sync_dirty_buffer(bh);
 534         brelse(bh);
 535 }
 536
 537 static void fat_put_super(struct super_block *sb)
 538 {
 539         struct msdos_sb_info *sbi = MSDOS_SB(sb);
 540
 541         fat_set_state(sb, 0, 0);
 542
 543         iput(sbi->fsinfo_inode);
 544         iput(sbi->fat_inode);
 545
 546         unload_nls(sbi->nls_disk);
 547         unload_nls(sbi->nls_io);
 548
 549         if (sbi->options.iocharset != fat_default_iocharset)
 550                 kfree(sbi->options.iocharset);
 551
 552         sb->s_fs_info = NULL;
 553         kfree(sbi);
 554 }
 555
 556 static struct kmem_cache *fat_inode_cachep;
 557
 558 static struct inode *fat_alloc_inode(struct super_block *sb)
 559 {
 560         struct msdos_inode_info *ei;
 561         ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
 562         if (!ei)
 563                 return NULL;
 564
 565         init_rwsem(&ei->truncate_lock);
 566         return &ei->vfs_inode;
 567 }
 568
 569 static void fat_i_callback(struct rcu_head *head)
 570 {
 571         struct inode *inode = container_of(head, struct inode, i_rcu);
 572         kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
 573 }
 574
 575 static void fat_destroy_inode(struct inode *inode)
 576 {
 577         call_rcu(&inode->i_rcu, fat_i_callback);
 578 }
 579
 580 static void init_once(void *foo)
 581 {
 582         struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
 583
 584         spin_lock_init(&ei->cache_lru_lock);
 585         ei->nr_caches = 0;
 586         ei->cache_valid_id = FAT_CACHE_VALID + 1;
 587         INIT_LIST_HEAD(&ei->cache_lru);
 588         INIT_HLIST_NODE(&ei->i_fat_hash);
 589         INIT_HLIST_NODE(&ei->i_dir_hash);
 590         inode_init_once(&ei->vfs_inode);
 591 }
 592
 593 static int __init fat_init_inodecache(void)
 594 {
 595         fat_inode_cachep = kmem_cache_create("fat_inode_cache",
 596                                              sizeof(struct msdos_inode_info),
 597                                              0, (SLAB_RECLAIM_ACCOUNT|
 598                                                 SLAB_MEM_SPREAD),
 599                                              init_once);
 600         if (fat_inode_cachep == NULL)
 601                 return -ENOMEM;
 602         return 0;
 603 }
 604
 605 static void __exit fat_destroy_inodecache(void)
 606 {
 607         /*
 608          * Make sure all delayed rcu free inodes are flushed before we
 609          * destroy cache.
 610          */
 611         rcu_barrier();
 612         kmem_cache_destroy(fat_inode_cachep);
 613 }
 614
 615 static int fat_remount(struct super_block *sb, int *flags, char *data)
 616 {
 617         int new_rdonly;
 618         struct msdos_sb_info *sbi = MSDOS_SB(sb);
 619         *flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
 620
 621         /* make sure we update state on remount. */
 622         new_rdonly = *flags & MS_RDONLY;
 623         if (new_rdonly != (sb->s_flags & MS_RDONLY)) {
 624                 if (new_rdonly)
 625                         fat_set_state(sb, 0, 0);
 626                 else
 627                         fat_set_state(sb, 1, 1);
 628         }
 629         return 0;
 630 }
 631
 632 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
 633 {
 634         struct super_block *sb = dentry->d_sb;
 635         struct msdos_sb_info *sbi = MSDOS_SB(sb);
 636         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 637
 638         /* If the count of free cluster is still unknown, counts it here. */
 639         if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
 640                 int err = fat_count_free_clusters(dentry->d_sb);
 641                 if (err)
 642                         return err;
 643         }
 644
 645         buf->f_type = dentry->d_sb->s_magic;
 646         buf->f_bsize = sbi->cluster_size;
 647         buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
 648         buf->f_bfree = sbi->free_clusters;
 649         buf->f_bavail = sbi->free_clusters;
 650         buf->f_fsid.val[0] = (u32)id;
 651         buf->f_fsid.val[1] = (u32)(id >> 32);
 652         buf->f_namelen =
 653                 (sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
 654
 655         return 0;
 656 }
 657
 658 static inline loff_t fat_i_pos_read(struct msdos_sb_info *sbi,
 659                                     struct inode *inode)
 660 {
 661         loff_t i_pos;
 662 #if BITS_PER_LONG == 32
 663         spin_lock(&sbi->inode_hash_lock);
 664 #endif
 665         i_pos = MSDOS_I(inode)->i_pos;
 666 #if BITS_PER_LONG == 32
 667         spin_unlock(&sbi->inode_hash_lock);
 668 #endif
 669         return i_pos;
 670 }
 671
 672 static int __fat_write_inode(struct inode *inode, int wait)
 673 {
 674         struct super_block *sb = inode->i_sb;
 675         struct msdos_sb_info *sbi = MSDOS_SB(sb);
 676         struct buffer_head *bh;
 677         struct msdos_dir_entry *raw_entry;
 678         loff_t i_pos;
 679         int err;
 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         bh = sb_bread(sb, i_pos >> sbi->dir_per_block_bits);
 690         if (!bh) {
 691                 fat_msg(sb, KERN_ERR, "unable to read inode block "
 692                        "for updating (i_pos %lld)", i_pos);
 693                 return -EIO;
 694         }
 695         spin_lock(&sbi->inode_hash_lock);
 696         if (i_pos != MSDOS_I(inode)->i_pos) {
 697                 spin_unlock(&sbi->inode_hash_lock);
 698                 brelse(bh);
 699                 goto retry;
 700         }
 701
 702         raw_entry = &((struct msdos_dir_entry *) (bh->b_data))
 703             [i_pos & (sbi->dir_per_block - 1)];
 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 const struct export_operations fat_export_ops = {
 765         .fh_to_dentry   = fat_fh_to_dentry,
 766         .fh_to_parent   = fat_fh_to_parent,
 767         .get_parent     = fat_get_parent,
 768 };
 769
 770 static int fat_show_options(struct seq_file *m, struct dentry *root)
 771 {
 772         struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
 773         struct fat_mount_options *opts = &sbi->options;
 774         int isvfat = opts->isvfat;
 775
 776         if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
 777                 seq_printf(m, ",uid=%u",
 778                                 from_kuid_munged(&init_user_ns, opts->fs_uid));
 779         if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
 780                 seq_printf(m, ",gid=%u",
 781                                 from_kgid_munged(&init_user_ns, opts->fs_gid));
 782         seq_printf(m, ",fmask=%04o", opts->fs_fmask);
 783         seq_printf(m, ",dmask=%04o", opts->fs_dmask);
 784         if (opts->allow_utime)
 785                 seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
 786         if (sbi->nls_disk)
 787                 /* strip "cp" prefix from displayed option */
 788                 seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
 789         if (isvfat) {
 790                 if (sbi->nls_io)
 791                         seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
 792
 793                 switch (opts->shortname) {
 794                 case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
 795                         seq_puts(m, ",shortname=win95");
 796                         break;
 797                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
 798                         seq_puts(m, ",shortname=winnt");
 799                         break;
 800                 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
 801                         seq_puts(m, ",shortname=mixed");
 802                         break;
 803                 case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
 804                         seq_puts(m, ",shortname=lower");
 805                         break;
 806                 default:
 807                         seq_puts(m, ",shortname=unknown");
 808                         break;
 809                 }
 810         }
 811         if (opts->name_check != 'n')
 812                 seq_printf(m, ",check=%c", opts->name_check);
 813         if (opts->usefree)
 814                 seq_puts(m, ",usefree");
 815         if (opts->quiet)
 816                 seq_puts(m, ",quiet");
 817         if (opts->showexec)
 818                 seq_puts(m, ",showexec");
 819         if (opts->sys_immutable)
 820                 seq_puts(m, ",sys_immutable");
 821         if (!isvfat) {
 822                 if (opts->dotsOK)
 823                         seq_puts(m, ",dotsOK=yes");
 824                 if (opts->nocase)
 825                         seq_puts(m, ",nocase");
 826         } else {
 827                 if (opts->utf8)
 828                         seq_puts(m, ",utf8");
 829                 if (opts->unicode_xlate)
 830                         seq_puts(m, ",uni_xlate");
 831                 if (!opts->numtail)
 832                         seq_puts(m, ",nonumtail");
 833                 if (opts->rodir)
 834                         seq_puts(m, ",rodir");
 835         }
 836         if (opts->flush)
 837                 seq_puts(m, ",flush");
 838         if (opts->tz_set) {
 839                 if (opts->time_offset)
 840                         seq_printf(m, ",time_offset=%d", opts->time_offset);
 841                 else
 842                         seq_puts(m, ",tz=UTC");
 843         }
 844         if (opts->errors == FAT_ERRORS_CONT)
 845                 seq_puts(m, ",errors=continue");
 846         else if (opts->errors == FAT_ERRORS_PANIC)
 847                 seq_puts(m, ",errors=panic");
 848         else
 849                 seq_puts(m, ",errors=remount-ro");
 850         if (opts->nfs == FAT_NFS_NOSTALE_RO)
 851                 seq_puts(m, ",nfs=nostale_ro");
 852         else if (opts->nfs)
 853                 seq_puts(m, ",nfs=stale_rw");
 854         if (opts->discard)
 855                 seq_puts(m, ",discard");
 856
 857         return 0;
 858 }
 859
 860 enum {
 861         Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
 862         Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
 863         Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
 864         Opt_immutable, Opt_dots, Opt_nodots,
 865         Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
 866         Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
 867         Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
 868         Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
 869         Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
 870         Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err,
 871 };
 872
 873 static const match_table_t fat_tokens = {
 874         {Opt_check_r, "check=relaxed"},
 875         {Opt_check_s, "check=strict"},
 876         {Opt_check_n, "check=normal"},
 877         {Opt_check_r, "check=r"},
 878         {Opt_check_s, "check=s"},
 879         {Opt_check_n, "check=n"},
 880         {Opt_uid, "uid=%u"},
 881         {Opt_gid, "gid=%u"},
 882         {Opt_umask, "umask=%o"},
 883         {Opt_dmask, "dmask=%o"},
 884         {Opt_fmask, "fmask=%o"},
 885         {Opt_allow_utime, "allow_utime=%o"},
 886         {Opt_codepage, "codepage=%u"},
 887         {Opt_usefree, "usefree"},
 888         {Opt_nocase, "nocase"},
 889         {Opt_quiet, "quiet"},
 890         {Opt_showexec, "showexec"},
 891         {Opt_debug, "debug"},
 892         {Opt_immutable, "sys_immutable"},
 893         {Opt_flush, "flush"},
 894         {Opt_tz_utc, "tz=UTC"},
 895         {Opt_time_offset, "time_offset=%d"},
 896         {Opt_err_cont, "errors=continue"},
 897         {Opt_err_panic, "errors=panic"},
 898         {Opt_err_ro, "errors=remount-ro"},
 899         {Opt_discard, "discard"},
 900         {Opt_nfs_stale_rw, "nfs"},
 901         {Opt_nfs_stale_rw, "nfs=stale_rw"},
 902         {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
 903         {Opt_obsolete, "conv=binary"},
 904         {Opt_obsolete, "conv=text"},
 905         {Opt_obsolete, "conv=auto"},
 906         {Opt_obsolete, "conv=b"},
 907         {Opt_obsolete, "conv=t"},
 908         {Opt_obsolete, "conv=a"},
 909         {Opt_obsolete, "fat=%u"},
 910         {Opt_obsolete, "blocksize=%u"},
 911         {Opt_obsolete, "cvf_format=%20s"},
 912         {Opt_obsolete, "cvf_options=%100s"},
 913         {Opt_obsolete, "posix"},
 914         {Opt_err, NULL},
 915 };
 916 static const match_table_t msdos_tokens = {
 917         {Opt_nodots, "nodots"},
 918         {Opt_nodots, "dotsOK=no"},
 919         {Opt_dots, "dots"},
 920         {Opt_dots, "dotsOK=yes"},
 921         {Opt_err, NULL}
 922 };
 923 static const match_table_t vfat_tokens = {
 924         {Opt_charset, "iocharset=%s"},
 925         {Opt_shortname_lower, "shortname=lower"},
 926         {Opt_shortname_win95, "shortname=win95"},
 927         {Opt_shortname_winnt, "shortname=winnt"},
 928         {Opt_shortname_mixed, "shortname=mixed"},
 929         {Opt_utf8_no, "utf8=0"},                /* 0 or no or false */
 930         {Opt_utf8_no, "utf8=no"},
 931         {Opt_utf8_no, "utf8=false"},
 932         {Opt_utf8_yes, "utf8=1"},               /* empty or 1 or yes or true */
 933         {Opt_utf8_yes, "utf8=yes"},
 934         {Opt_utf8_yes, "utf8=true"},
 935         {Opt_utf8_yes, "utf8"},
 936         {Opt_uni_xl_no, "uni_xlate=0"},         /* 0 or no or false */
 937         {Opt_uni_xl_no, "uni_xlate=no"},
 938         {Opt_uni_xl_no, "uni_xlate=false"},
 939         {Opt_uni_xl_yes, "uni_xlate=1"},        /* empty or 1 or yes or true */
 940         {Opt_uni_xl_yes, "uni_xlate=yes"},
 941         {Opt_uni_xl_yes, "uni_xlate=true"},
 942         {Opt_uni_xl_yes, "uni_xlate"},
 943         {Opt_nonumtail_no, "nonumtail=0"},      /* 0 or no or false */
 944         {Opt_nonumtail_no, "nonumtail=no"},
 945         {Opt_nonumtail_no, "nonumtail=false"},
 946         {Opt_nonumtail_yes, "nonumtail=1"},     /* empty or 1 or yes or true */
 947         {Opt_nonumtail_yes, "nonumtail=yes"},
 948         {Opt_nonumtail_yes, "nonumtail=true"},
 949         {Opt_nonumtail_yes, "nonumtail"},
 950         {Opt_rodir, "rodir"},
 951         {Opt_err, NULL}
 952 };
 953
 954 static int parse_options(struct super_block *sb, char *options, int is_vfat,
 955                          int silent, int *debug, struct fat_mount_options *opts)
 956 {
 957         char *p;
 958         substring_t args[MAX_OPT_ARGS];
 959         int option;
 960         char *iocharset;
 961
 962         opts->isvfat = is_vfat;
 963
 964         opts->fs_uid = current_uid();
 965         opts->fs_gid = current_gid();
 966         opts->fs_fmask = opts->fs_dmask = current_umask();
 967         opts->allow_utime = -1;
 968         opts->codepage = fat_default_codepage;
 969         opts->iocharset = fat_default_iocharset;
 970         if (is_vfat) {
 971                 opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
 972                 opts->rodir = 0;
 973         } else {
 974                 opts->shortname = 0;
 975                 opts->rodir = 1;
 976         }
 977         opts->name_check = 'n';
 978         opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK =  0;
 979         opts->utf8 = opts->unicode_xlate = 0;
 980         opts->numtail = 1;
 981         opts->usefree = opts->nocase = 0;
 982         opts->tz_set = 0;
 983         opts->nfs = 0;
 984         opts->errors = FAT_ERRORS_RO;
 985         *debug = 0;
 986
 987         if (!options)
 988                 goto out;
 989
 990         while ((p = strsep(&options, ",")) != NULL) {
 991                 int token;
 992                 if (!*p)
 993                         continue;
 994
 995                 token = match_token(p, fat_tokens, args);
 996                 if (token == Opt_err) {
 997                         if (is_vfat)
 998                                 token = match_token(p, vfat_tokens, args);
 999                         else
1000                                 token = match_token(p, msdos_tokens, args);
1001                 }
1002                 switch (token) {
1003                 case Opt_check_s:
1004                         opts->name_check = 's';
1005                         break;
1006                 case Opt_check_r:
1007                         opts->name_check = 'r';
1008                         break;
1009                 case Opt_check_n:
1010                         opts->name_check = 'n';
1011                         break;
1012                 case Opt_usefree:
1013                         opts->usefree = 1;
1014                         break;
1015                 case Opt_nocase:
1016                         if (!is_vfat)
1017                                 opts->nocase = 1;
1018                         else {
1019                                 /* for backward compatibility */
1020                                 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1021                                         | VFAT_SFN_CREATE_WIN95;
1022                         }
1023                         break;
1024                 case Opt_quiet:
1025                         opts->quiet = 1;
1026                         break;
1027                 case Opt_showexec:
1028                         opts->showexec = 1;
1029                         break;
1030                 case Opt_debug:
1031                         *debug = 1;
1032                         break;
1033                 case Opt_immutable:
1034                         opts->sys_immutable = 1;
1035                         break;
1036                 case Opt_uid:
1037                         if (match_int(&args[0], &option))
1038                                 return -EINVAL;
1039                         opts->fs_uid = make_kuid(current_user_ns(), option);
1040                         if (!uid_valid(opts->fs_uid))
1041                                 return -EINVAL;
1042                         break;
1043                 case Opt_gid:
1044                         if (match_int(&args[0], &option))
1045                                 return -EINVAL;
1046                         opts->fs_gid = make_kgid(current_user_ns(), option);
1047                         if (!gid_valid(opts->fs_gid))
1048                                 return -EINVAL;
1049                         break;
1050                 case Opt_umask:
1051                         if (match_octal(&args[0], &option))
1052                                 return -EINVAL;
1053                         opts->fs_fmask = opts->fs_dmask = option;
1054                         break;
1055                 case Opt_dmask:
1056                         if (match_octal(&args[0], &option))
1057                                 return -EINVAL;
1058                         opts->fs_dmask = option;
1059                         break;
1060                 case Opt_fmask:
1061                         if (match_octal(&args[0], &option))
1062                                 return -EINVAL;
1063                         opts->fs_fmask = option;
1064                         break;
1065                 case Opt_allow_utime:
1066                         if (match_octal(&args[0], &option))
1067                                 return -EINVAL;
1068                         opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1069                         break;
1070                 case Opt_codepage:
1071                         if (match_int(&args[0], &option))
1072                                 return -EINVAL;
1073                         opts->codepage = option;
1074                         break;
1075                 case Opt_flush:
1076                         opts->flush = 1;
1077                         break;
1078                 case Opt_time_offset:
1079                         if (match_int(&args[0], &option))
1080                                 return -EINVAL;
1081                         if (option < -12 * 60 || option > 12 * 60)
1082                                 return -EINVAL;
1083                         opts->tz_set = 1;
1084                         opts->time_offset = option;
1085                         break;
1086                 case Opt_tz_utc:
1087                         opts->tz_set = 1;
1088                         opts->time_offset = 0;
1089                         break;
1090                 case Opt_err_cont:
1091                         opts->errors = FAT_ERRORS_CONT;
1092                         break;
1093                 case Opt_err_panic:
1094                         opts->errors = FAT_ERRORS_PANIC;
1095                         break;
1096                 case Opt_err_ro:
1097                         opts->errors = FAT_ERRORS_RO;
1098                         break;
1099                 case Opt_nfs_stale_rw:
1100                         opts->nfs = FAT_NFS_STALE_RW;
1101                         break;
1102                 case Opt_nfs_nostale_ro:
1103                         opts->nfs = FAT_NFS_NOSTALE_RO;
1104                         break;
1105
1106                 /* msdos specific */
1107                 case Opt_dots:
1108                         opts->dotsOK = 1;
1109                         break;
1110                 case Opt_nodots:
1111                         opts->dotsOK = 0;
1112                         break;
1113
1114                 /* vfat specific */
1115                 case Opt_charset:
1116                         if (opts->iocharset != fat_default_iocharset)
1117                                 kfree(opts->iocharset);
1118                         iocharset = match_strdup(&args[0]);
1119                         if (!iocharset)
1120                                 return -ENOMEM;
1121                         opts->iocharset = iocharset;
1122                         break;
1123                 case Opt_shortname_lower:
1124                         opts->shortname = VFAT_SFN_DISPLAY_LOWER
1125                                         | VFAT_SFN_CREATE_WIN95;
1126                         break;
1127                 case Opt_shortname_win95:
1128                         opts->shortname = VFAT_SFN_DISPLAY_WIN95
1129                                         | VFAT_SFN_CREATE_WIN95;
1130                         break;
1131                 case Opt_shortname_winnt:
1132                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1133                                         | VFAT_SFN_CREATE_WINNT;
1134                         break;
1135                 case Opt_shortname_mixed:
1136                         opts->shortname = VFAT_SFN_DISPLAY_WINNT
1137                                         | VFAT_SFN_CREATE_WIN95;
1138                         break;
1139                 case Opt_utf8_no:               /* 0 or no or false */
1140                         opts->utf8 = 0;
1141                         break;
1142                 case Opt_utf8_yes:              /* empty or 1 or yes or true */
1143                         opts->utf8 = 1;
1144                         break;
1145                 case Opt_uni_xl_no:             /* 0 or no or false */
1146                         opts->unicode_xlate = 0;
1147                         break;
1148                 case Opt_uni_xl_yes:            /* empty or 1 or yes or true */
1149                         opts->unicode_xlate = 1;
1150                         break;
1151                 case Opt_nonumtail_no:          /* 0 or no or false */
1152                         opts->numtail = 1;      /* negated option */
1153                         break;
1154                 case Opt_nonumtail_yes:         /* empty or 1 or yes or true */
1155                         opts->numtail = 0;      /* negated option */
1156                         break;
1157                 case Opt_rodir:
1158                         opts->rodir = 1;
1159                         break;
1160                 case Opt_discard:
1161                         opts->discard = 1;
1162                         break;
1163
1164                 /* obsolete mount options */
1165                 case Opt_obsolete:
1166                         fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1167                                "not supported now", p);
1168                         break;
1169                 /* unknown option */
1170                 default:
1171                         if (!silent) {
1172                                 fat_msg(sb, KERN_ERR,
1173                                        "Unrecognized mount option \"%s\" "
1174                                        "or missing value", p);
1175                         }
1176                         return -EINVAL;
1177                 }
1178         }
1179
1180 out:
1181         /* UTF-8 doesn't provide FAT semantics */
1182         if (!strcmp(opts->iocharset, "utf8")) {
1183                 fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1184                        " for FAT filesystems, filesystem will be "
1185                        "case sensitive!");
1186         }
1187
1188         /* If user doesn't specify allow_utime, it's initialized from dmask. */
1189         if (opts->allow_utime == (unsigned short)-1)
1190                 opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1191         if (opts->unicode_xlate)
1192                 opts->utf8 = 0;
1193         if (opts->nfs == FAT_NFS_NOSTALE_RO)
1194                 sb->s_flags |= MS_RDONLY;
1195
1196         return 0;
1197 }
1198
1199 static int fat_read_root(struct inode *inode)
1200 {
1201         struct super_block *sb = inode->i_sb;
1202         struct msdos_sb_info *sbi = MSDOS_SB(sb);
1203         int error;
1204
1205         MSDOS_I(inode)->i_pos = 0;
1206         inode->i_uid = sbi->options.fs_uid;
1207         inode->i_gid = sbi->options.fs_gid;
1208         inode->i_version++;
1209         inode->i_generation = 0;
1210         inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1211         inode->i_op = sbi->dir_ops;
1212         inode->i_fop = &fat_dir_operations;
1213         if (sbi->fat_bits == 32) {
1214                 MSDOS_I(inode)->i_start = sbi->root_cluster;
1215                 error = fat_calc_dir_size(inode);
1216                 if (error < 0)
1217                         return error;
1218         } else {
1219                 MSDOS_I(inode)->i_start = 0;
1220                 inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1221         }
1222         inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1223                            & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1224         MSDOS_I(inode)->i_logstart = 0;
1225         MSDOS_I(inode)->mmu_private = inode->i_size;
1226
1227         fat_save_attrs(inode, ATTR_DIR);
1228         inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1229         inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1230         set_nlink(inode, fat_subdirs(inode)+2);
1231
1232         return 0;
1233 }
1234
1235 /*
1236  * Read the super block of an MS-DOS FS.
1237  */
1238 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1239                    void (*setup)(struct super_block *))
1240 {
1241         struct inode *root_inode = NULL, *fat_inode = NULL;
1242         struct inode *fsinfo_inode = NULL;
1243         struct buffer_head *bh;
1244         struct fat_boot_sector *b;
1245         struct msdos_sb_info *sbi;
1246         u16 logical_sector_size;
1247         u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1248         int debug;
1249         unsigned int media;
1250         long error;
1251         char buf[50];
1252
1253         /*
1254          * GFP_KERNEL is ok here, because while we do hold the
1255          * supeblock lock, memory pressure can't call back into
1256          * the filesystem, since we're only just about to mount
1257          * it and have no inodes etc active!
1258          */
1259         sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1260         if (!sbi)
1261                 return -ENOMEM;
1262         sb->s_fs_info = sbi;
1263
1264         sb->s_flags |= MS_NODIRATIME;
1265         sb->s_magic = MSDOS_SUPER_MAGIC;
1266         sb->s_op = &fat_sops;
1267         sb->s_export_op = &fat_export_ops;
1268         ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1269                              DEFAULT_RATELIMIT_BURST);
1270
1271         error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1272         if (error)
1273                 goto out_fail;
1274
1275         setup(sb); /* flavour-specific stuff that needs options */
1276
1277         error = -EIO;
1278         sb_min_blocksize(sb, 512);
1279         bh = sb_bread(sb, 0);
1280         if (bh == NULL) {
1281                 fat_msg(sb, KERN_ERR, "unable to read boot sector");
1282                 goto out_fail;
1283         }
1284
1285         b = (struct fat_boot_sector *) bh->b_data;
1286         if (!b->reserved) {
1287                 if (!silent)
1288                         fat_msg(sb, KERN_ERR, "bogus number of reserved sectors");
1289                 brelse(bh);
1290                 goto out_invalid;
1291         }
1292         if (!b->fats) {
1293                 if (!silent)
1294                         fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1295                 brelse(bh);
1296                 goto out_invalid;
1297         }
1298
1299         /*
1300          * Earlier we checked here that b->secs_track and b->head are nonzero,
1301          * but it turns out valid FAT filesystems can have zero there.
1302          */
1303
1304         media = b->media;
1305         if (!fat_valid_media(media)) {
1306                 if (!silent)
1307                         fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1308                                media);
1309                 brelse(bh);
1310                 goto out_invalid;
1311         }
1312         logical_sector_size = get_unaligned_le16(&b->sector_size);
1313         if (!is_power_of_2(logical_sector_size)
1314             || (logical_sector_size < 512)
1315             || (logical_sector_size > 4096)) {
1316                 if (!silent)
1317                         fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1318                                logical_sector_size);
1319                 brelse(bh);
1320                 goto out_invalid;
1321         }
1322         sbi->sec_per_clus = b->sec_per_clus;
1323         if (!is_power_of_2(sbi->sec_per_clus)) {
1324                 if (!silent)
1325                         fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1326                                sbi->sec_per_clus);
1327                 brelse(bh);
1328                 goto out_invalid;
1329         }
1330
1331         if (logical_sector_size < sb->s_blocksize) {
1332                 fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1333                        " (logical sector size = %u)", logical_sector_size);
1334                 brelse(bh);
1335                 goto out_fail;
1336         }
1337         if (logical_sector_size > sb->s_blocksize) {
1338                 brelse(bh);
1339
1340                 if (!sb_set_blocksize(sb, logical_sector_size)) {
1341                         fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1342                                logical_sector_size);
1343                         goto out_fail;
1344                 }
1345                 bh = sb_bread(sb, 0);
1346                 if (bh == NULL) {
1347                         fat_msg(sb, KERN_ERR, "unable to read boot sector"
1348                                " (logical sector size = %lu)",
1349                                sb->s_blocksize);
1350                         goto out_fail;
1351                 }
1352                 b = (struct fat_boot_sector *) bh->b_data;
1353         }
1354
1355         mutex_init(&sbi->s_lock);
1356         sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1357         sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1358         sbi->fats = b->fats;
1359         sbi->fat_bits = 0;              /* Don't know yet */
1360         sbi->fat_start = le16_to_cpu(b->reserved);
1361         sbi->fat_length = le16_to_cpu(b->fat_length);
1362         sbi->root_cluster = 0;
1363         sbi->free_clusters = -1;        /* Don't know yet */
1364         sbi->free_clus_valid = 0;
1365         sbi->prev_free = FAT_START_ENT;
1366         sb->s_maxbytes = 0xffffffff;
1367
1368         if (!sbi->fat_length && b->fat32.length) {
1369                 struct fat_boot_fsinfo *fsinfo;
1370                 struct buffer_head *fsinfo_bh;
1371
1372                 /* Must be FAT32 */
1373                 sbi->fat_bits = 32;
1374                 sbi->fat_length = le32_to_cpu(b->fat32.length);
1375                 sbi->root_cluster = le32_to_cpu(b->fat32.root_cluster);
1376
1377                 /* MC - if info_sector is 0, don't multiply by 0 */
1378                 sbi->fsinfo_sector = le16_to_cpu(b->fat32.info_sector);
1379                 if (sbi->fsinfo_sector == 0)
1380                         sbi->fsinfo_sector = 1;
1381
1382                 fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1383                 if (fsinfo_bh == NULL) {
1384                         fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1385                                " (sector = %lu)", sbi->fsinfo_sector);
1386                         brelse(bh);
1387                         goto out_fail;
1388                 }
1389
1390                 fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1391                 if (!IS_FSINFO(fsinfo)) {
1392                         fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1393                                "0x%08x, 0x%08x (sector = %lu)",
1394                                le32_to_cpu(fsinfo->signature1),
1395                                le32_to_cpu(fsinfo->signature2),
1396                                sbi->fsinfo_sector);
1397                 } else {
1398                         if (sbi->options.usefree)
1399                                 sbi->free_clus_valid = 1;
1400                         sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1401                         sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1402                 }
1403
1404                 brelse(fsinfo_bh);
1405         }
1406
1407         sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1408         sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1409
1410         sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1411         sbi->dir_entries = get_unaligned_le16(&b->dir_entries);
1412         if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1413                 if (!silent)
1414                         fat_msg(sb, KERN_ERR, "bogus directory-entries per block"
1415                                " (%u)", sbi->dir_entries);
1416                 brelse(bh);
1417                 goto out_invalid;
1418         }
1419
1420         rootdir_sectors = sbi->dir_entries
1421                 * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1422         sbi->data_start = sbi->dir_start + rootdir_sectors;
1423         total_sectors = get_unaligned_le16(&b->sectors);
1424         if (total_sectors == 0)
1425                 total_sectors = le32_to_cpu(b->total_sect);
1426
1427         total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1428
1429         if (sbi->fat_bits != 32)
1430                 sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1431
1432         /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1433         if (sbi->fat_bits == 32)
1434                 sbi->dirty = b->fat32.state & FAT_STATE_DIRTY;
1435         else /* fat 16 or 12 */
1436                 sbi->dirty = b->fat16.state & FAT_STATE_DIRTY;
1437
1438         /* check that FAT table does not overflow */
1439         fat_clusters = sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1440         total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1441         if (total_clusters > MAX_FAT(sb)) {
1442                 if (!silent)
1443                         fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1444                                total_clusters);
1445                 brelse(bh);
1446                 goto out_invalid;
1447         }
1448
1449         sbi->max_cluster = total_clusters + FAT_START_ENT;
1450         /* check the free_clusters, it's not necessarily correct */
1451         if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1452                 sbi->free_clusters = -1;
1453         /* check the prev_free, it's not necessarily correct */
1454         sbi->prev_free %= sbi->max_cluster;
1455         if (sbi->prev_free < FAT_START_ENT)
1456                 sbi->prev_free = FAT_START_ENT;
1457
1458         brelse(bh);
1459
1460         /* set up enough so that it can read an inode */
1461         fat_hash_init(sb);
1462         dir_hash_init(sb);
1463         fat_ent_access_init(sb);
1464
1465         /*
1466          * The low byte of FAT's first entry must have same value with
1467          * media-field.  But in real world, too many devices is
1468          * writing wrong value.  So, removed that validity check.
1469          *
1470          * if (FAT_FIRST_ENT(sb, media) != first)
1471          */
1472
1473         error = -EINVAL;
1474         sprintf(buf, "cp%d", sbi->options.codepage);
1475         sbi->nls_disk = load_nls(buf);
1476         if (!sbi->nls_disk) {
1477                 fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1478                 goto out_fail;
1479         }
1480
1481         /* FIXME: utf8 is using iocharset for upper/lower conversion */
1482         if (sbi->options.isvfat) {
1483                 sbi->nls_io = load_nls(sbi->options.iocharset);
1484                 if (!sbi->nls_io) {
1485                         fat_msg(sb, KERN_ERR, "IO charset %s not found",
1486                                sbi->options.iocharset);
1487                         goto out_fail;
1488                 }
1489         }
1490
1491         error = -ENOMEM;
1492         fat_inode = new_inode(sb);
1493         if (!fat_inode)
1494                 goto out_fail;
1495         MSDOS_I(fat_inode)->i_pos = 0;
1496         sbi->fat_inode = fat_inode;
1497
1498         fsinfo_inode = new_inode(sb);
1499         if (!fsinfo_inode)
1500                 goto out_fail;
1501         fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1502         sbi->fsinfo_inode = fsinfo_inode;
1503         insert_inode_hash(fsinfo_inode);
1504
1505         root_inode = new_inode(sb);
1506         if (!root_inode)
1507                 goto out_fail;
1508         root_inode->i_ino = MSDOS_ROOT_INO;
1509         root_inode->i_version = 1;
1510         error = fat_read_root(root_inode);
1511         if (error < 0) {
1512                 iput(root_inode);
1513                 goto out_fail;
1514         }
1515         error = -ENOMEM;
1516         insert_inode_hash(root_inode);
1517         fat_attach(root_inode, 0);
1518         sb->s_root = d_make_root(root_inode);
1519         if (!sb->s_root) {
1520                 fat_msg(sb, KERN_ERR, "get root inode failed");
1521                 goto out_fail;
1522         }
1523
1524         if (sbi->options.discard) {
1525                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1526                 if (!blk_queue_discard(q))
1527                         fat_msg(sb, KERN_WARNING,
1528                                         "mounting with \"discard\" option, but "
1529                                         "the device does not support discard");
1530         }
1531
1532         fat_set_state(sb, 1, 0);
1533         return 0;
1534
1535 out_invalid:
1536         error = -EINVAL;
1537         if (!silent)
1538                 fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1539
1540 out_fail:
1541         if (fsinfo_inode)
1542                 iput(fsinfo_inode);
1543         if (fat_inode)
1544                 iput(fat_inode);
1545         unload_nls(sbi->nls_io);
1546         unload_nls(sbi->nls_disk);
1547         if (sbi->options.iocharset != fat_default_iocharset)
1548                 kfree(sbi->options.iocharset);
1549         sb->s_fs_info = NULL;
1550         kfree(sbi);
1551         return error;
1552 }
1553
1554 EXPORT_SYMBOL_GPL(fat_fill_super);
1555
1556 /*
1557  * helper function for fat_flush_inodes.  This writes both the inode
1558  * and the file data blocks, waiting for in flight data blocks before
1559  * the start of the call.  It does not wait for any io started
1560  * during the call
1561  */
1562 static int writeback_inode(struct inode *inode)
1563 {
1564
1565         int ret;
1566
1567         /* if we used wait=1, sync_inode_metadata waits for the io for the
1568         * inode to finish.  So wait=0 is sent down to sync_inode_metadata
1569         * and filemap_fdatawrite is used for the data blocks
1570         */
1571         ret = sync_inode_metadata(inode, 0);
1572         if (!ret)
1573                 ret = filemap_fdatawrite(inode->i_mapping);
1574         return ret;
1575 }
1576
1577 /*
1578  * write data and metadata corresponding to i1 and i2.  The io is
1579  * started but we do not wait for any of it to finish.
1580  *
1581  * filemap_flush is used for the block device, so if there is a dirty
1582  * page for a block already in flight, we will not wait and start the
1583  * io over again
1584  */
1585 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1586 {
1587         int ret = 0;
1588         if (!MSDOS_SB(sb)->options.flush)
1589                 return 0;
1590         if (i1)
1591                 ret = writeback_inode(i1);
1592         if (!ret && i2)
1593                 ret = writeback_inode(i2);
1594         if (!ret) {
1595                 struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1596                 ret = filemap_flush(mapping);
1597         }
1598         return ret;
1599 }
1600 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1601
1602 static int __init init_fat_fs(void)
1603 {
1604         int err;
1605
1606         err = fat_cache_init();
1607         if (err)
1608                 return err;
1609
1610         err = fat_init_inodecache();
1611         if (err)
1612                 goto failed;
1613
1614         return 0;
1615
1616 failed:
1617         fat_cache_destroy();
1618         return err;
1619 }
1620
1621 static void __exit exit_fat_fs(void)
1622 {
1623         fat_cache_destroy();
1624         fat_destroy_inodecache();
1625 }
1626
1627 module_init(init_fat_fs)
1628 module_exit(exit_fat_fs)
1629
1630 MODULE_LICENSE("GPL");