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