fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/ext4_fs.h>
  25 #include <linux/ext4_jbd2.h>
  26 #include <linux/slab.h>
  27 #include <linux/init.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/parser.h>
  30 #include <linux/smp_lock.h>
  31 #include <linux/buffer_head.h>
  32 #include <linux/exportfs.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/seq_file.h>
  39
  40 #include <asm/uaccess.h>
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44 #include "namei.h"
  45
  46 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  47                              unsigned long journal_devnum);
  48 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
  49                                unsigned int);
  50 static void ext4_commit_super (struct super_block * sb,
  51                                struct ext4_super_block * es,
  52                                int sync);
  53 static void ext4_mark_recovery_complete(struct super_block * sb,
  54                                         struct ext4_super_block * es);
  55 static void ext4_clear_journal_err(struct super_block * sb,
  56                                    struct ext4_super_block * es);
  57 static int ext4_sync_fs(struct super_block *sb, int wait);
  58 static const char *ext4_decode_error(struct super_block * sb, int errno,
  59                                      char nbuf[16]);
  60 static int ext4_remount (struct super_block * sb, int * flags, char * data);
  61 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
  62 static void ext4_unlockfs(struct super_block *sb);
  63 static void ext4_write_super (struct super_block * sb);
  64 static void ext4_write_super_lockfs(struct super_block *sb);
  65
  66
  67 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  68                                struct ext4_group_desc *bg)
  69 {
  70         return le32_to_cpu(bg->bg_block_bitmap) |
  71                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  72                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  73 }
  74
  75 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  76                                struct ext4_group_desc *bg)
  77 {
  78         return le32_to_cpu(bg->bg_inode_bitmap) |
  79                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  80                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
  81 }
  82
  83 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
  84                               struct ext4_group_desc *bg)
  85 {
  86         return le32_to_cpu(bg->bg_inode_table) |
  87                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  88                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
  89 }
  90
  91 void ext4_block_bitmap_set(struct super_block *sb,
  92                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
  93 {
  94         bg->bg_block_bitmap = cpu_to_le32((u32)blk);
  95         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
  96                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
  97 }
  98
  99 void ext4_inode_bitmap_set(struct super_block *sb,
 100                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 101 {
 102         bg->bg_inode_bitmap  = cpu_to_le32((u32)blk);
 103         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 104                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 105 }
 106
 107 void ext4_inode_table_set(struct super_block *sb,
 108                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 109 {
 110         bg->bg_inode_table = cpu_to_le32((u32)blk);
 111         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 112                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 113 }
 114
 115 /*
 116  * Wrappers for jbd2_journal_start/end.
 117  *
 118  * The only special thing we need to do here is to make sure that all
 119  * journal_end calls result in the superblock being marked dirty, so
 120  * that sync() will call the filesystem's write_super callback if
 121  * appropriate.
 122  */
 123 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 124 {
 125         journal_t *journal;
 126
 127         if (sb->s_flags & MS_RDONLY)
 128                 return ERR_PTR(-EROFS);
 129
 130         /* Special case here: if the journal has aborted behind our
 131          * backs (eg. EIO in the commit thread), then we still need to
 132          * take the FS itself readonly cleanly. */
 133         journal = EXT4_SB(sb)->s_journal;
 134         if (is_journal_aborted(journal)) {
 135                 ext4_abort(sb, __FUNCTION__,
 136                            "Detected aborted journal");
 137                 return ERR_PTR(-EROFS);
 138         }
 139
 140         return jbd2_journal_start(journal, nblocks);
 141 }
 142
 143 /*
 144  * The only special thing we need to do here is to make sure that all
 145  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 146  * that sync() will call the filesystem's write_super callback if
 147  * appropriate.
 148  */
 149 int __ext4_journal_stop(const char *where, handle_t *handle)
 150 {
 151         struct super_block *sb;
 152         int err;
 153         int rc;
 154
 155         sb = handle->h_transaction->t_journal->j_private;
 156         err = handle->h_err;
 157         rc = jbd2_journal_stop(handle);
 158
 159         if (!err)
 160                 err = rc;
 161         if (err)
 162                 __ext4_std_error(sb, where, err);
 163         return err;
 164 }
 165
 166 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 167                 struct buffer_head *bh, handle_t *handle, int err)
 168 {
 169         char nbuf[16];
 170         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 171
 172         if (bh)
 173                 BUFFER_TRACE(bh, "abort");
 174
 175         if (!handle->h_err)
 176                 handle->h_err = err;
 177
 178         if (is_handle_aborted(handle))
 179                 return;
 180
 181         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 182                caller, errstr, err_fn);
 183
 184         jbd2_journal_abort_handle(handle);
 185 }
 186
 187 /* Deal with the reporting of failure conditions on a filesystem such as
 188  * inconsistencies detected or read IO failures.
 189  *
 190  * On ext2, we can store the error state of the filesystem in the
 191  * superblock.  That is not possible on ext4, because we may have other
 192  * write ordering constraints on the superblock which prevent us from
 193  * writing it out straight away; and given that the journal is about to
 194  * be aborted, we can't rely on the current, or future, transactions to
 195  * write out the superblock safely.
 196  *
 197  * We'll just use the jbd2_journal_abort() error code to record an error in
 198  * the journal instead.  On recovery, the journal will compain about
 199  * that error until we've noted it down and cleared it.
 200  */
 201
 202 static void ext4_handle_error(struct super_block *sb)
 203 {
 204         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 205
 206         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 207         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 208
 209         if (sb->s_flags & MS_RDONLY)
 210                 return;
 211
 212         if (!test_opt (sb, ERRORS_CONT)) {
 213                 journal_t *journal = EXT4_SB(sb)->s_journal;
 214
 215                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 216                 if (journal)
 217                         jbd2_journal_abort(journal, -EIO);
 218         }
 219         if (test_opt (sb, ERRORS_RO)) {
 220                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 221                 sb->s_flags |= MS_RDONLY;
 222         }
 223         ext4_commit_super(sb, es, 1);
 224         if (test_opt(sb, ERRORS_PANIC))
 225                 panic("EXT4-fs (device %s): panic forced after error\n",
 226                         sb->s_id);
 227 }
 228
 229 void ext4_error (struct super_block * sb, const char * function,
 230                  const char * fmt, ...)
 231 {
 232         va_list args;
 233
 234         va_start(args, fmt);
 235         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 236         vprintk(fmt, args);
 237         printk("\n");
 238         va_end(args);
 239
 240         ext4_handle_error(sb);
 241 }
 242
 243 static const char *ext4_decode_error(struct super_block * sb, int errno,
 244                                      char nbuf[16])
 245 {
 246         char *errstr = NULL;
 247
 248         switch (errno) {
 249         case -EIO:
 250                 errstr = "IO failure";
 251                 break;
 252         case -ENOMEM:
 253                 errstr = "Out of memory";
 254                 break;
 255         case -EROFS:
 256                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
 257                         errstr = "Journal has aborted";
 258                 else
 259                         errstr = "Readonly filesystem";
 260                 break;
 261         default:
 262                 /* If the caller passed in an extra buffer for unknown
 263                  * errors, textualise them now.  Else we just return
 264                  * NULL. */
 265                 if (nbuf) {
 266                         /* Check for truncated error codes... */
 267                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 268                                 errstr = nbuf;
 269                 }
 270                 break;
 271         }
 272
 273         return errstr;
 274 }
 275
 276 /* __ext4_std_error decodes expected errors from journaling functions
 277  * automatically and invokes the appropriate error response.  */
 278
 279 void __ext4_std_error (struct super_block * sb, const char * function,
 280                        int errno)
 281 {
 282         char nbuf[16];
 283         const char *errstr;
 284
 285         /* Special case: if the error is EROFS, and we're not already
 286          * inside a transaction, then there's really no point in logging
 287          * an error. */
 288         if (errno == -EROFS && journal_current_handle() == NULL &&
 289             (sb->s_flags & MS_RDONLY))
 290                 return;
 291
 292         errstr = ext4_decode_error(sb, errno, nbuf);
 293         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 294                 sb->s_id, function, errstr);
 295
 296         ext4_handle_error(sb);
 297 }
 298
 299 /*
 300  * ext4_abort is a much stronger failure handler than ext4_error.  The
 301  * abort function may be used to deal with unrecoverable failures such
 302  * as journal IO errors or ENOMEM at a critical moment in log management.
 303  *
 304  * We unconditionally force the filesystem into an ABORT|READONLY state,
 305  * unless the error response on the fs has been set to panic in which
 306  * case we take the easy way out and panic immediately.
 307  */
 308
 309 void ext4_abort (struct super_block * sb, const char * function,
 310                  const char * fmt, ...)
 311 {
 312         va_list args;
 313
 314         printk (KERN_CRIT "ext4_abort called.\n");
 315
 316         va_start(args, fmt);
 317         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
 318         vprintk(fmt, args);
 319         printk("\n");
 320         va_end(args);
 321
 322         if (test_opt(sb, ERRORS_PANIC))
 323                 panic("EXT4-fs panic from previous error\n");
 324
 325         if (sb->s_flags & MS_RDONLY)
 326                 return;
 327
 328         printk(KERN_CRIT "Remounting filesystem read-only\n");
 329         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 330         sb->s_flags |= MS_RDONLY;
 331         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
 332         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 333 }
 334
 335 void ext4_warning (struct super_block * sb, const char * function,
 336                    const char * fmt, ...)
 337 {
 338         va_list args;
 339
 340         va_start(args, fmt);
 341         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 342                sb->s_id, function);
 343         vprintk(fmt, args);
 344         printk("\n");
 345         va_end(args);
 346 }
 347
 348 void ext4_update_dynamic_rev(struct super_block *sb)
 349 {
 350         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 351
 352         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 353                 return;
 354
 355         ext4_warning(sb, __FUNCTION__,
 356                      "updating to rev %d because of new feature flag, "
 357                      "running e2fsck is recommended",
 358                      EXT4_DYNAMIC_REV);
 359
 360         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 361         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 362         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 363         /* leave es->s_feature_*compat flags alone */
 364         /* es->s_uuid will be set by e2fsck if empty */
 365
 366         /*
 367          * The rest of the superblock fields should be zero, and if not it
 368          * means they are likely already in use, so leave them alone.  We
 369          * can leave it up to e2fsck to clean up any inconsistencies there.
 370          */
 371 }
 372
 373 /*
 374  * Open the external journal device
 375  */
 376 static struct block_device *ext4_blkdev_get(dev_t dev)
 377 {
 378         struct block_device *bdev;
 379         char b[BDEVNAME_SIZE];
 380
 381         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 382         if (IS_ERR(bdev))
 383                 goto fail;
 384         return bdev;
 385
 386 fail:
 387         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
 388                         __bdevname(dev, b), PTR_ERR(bdev));
 389         return NULL;
 390 }
 391
 392 /*
 393  * Release the journal device
 394  */
 395 static int ext4_blkdev_put(struct block_device *bdev)
 396 {
 397         bd_release(bdev);
 398         return blkdev_put(bdev);
 399 }
 400
 401 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 402 {
 403         struct block_device *bdev;
 404         int ret = -ENODEV;
 405
 406         bdev = sbi->journal_bdev;
 407         if (bdev) {
 408                 ret = ext4_blkdev_put(bdev);
 409                 sbi->journal_bdev = NULL;
 410         }
 411         return ret;
 412 }
 413
 414 static inline struct inode *orphan_list_entry(struct list_head *l)
 415 {
 416         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 417 }
 418
 419 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 420 {
 421         struct list_head *l;
 422
 423         printk(KERN_ERR "sb orphan head is %d\n",
 424                le32_to_cpu(sbi->s_es->s_last_orphan));
 425
 426         printk(KERN_ERR "sb_info orphan list:\n");
 427         list_for_each(l, &sbi->s_orphan) {
 428                 struct inode *inode = orphan_list_entry(l);
 429                 printk(KERN_ERR "  "
 430                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 431                        inode->i_sb->s_id, inode->i_ino, inode,
 432                        inode->i_mode, inode->i_nlink,
 433                        NEXT_ORPHAN(inode));
 434         }
 435 }
 436
 437 static void ext4_put_super (struct super_block * sb)
 438 {
 439         struct ext4_sb_info *sbi = EXT4_SB(sb);
 440         struct ext4_super_block *es = sbi->s_es;
 441         int i;
 442
 443         ext4_ext_release(sb);
 444         ext4_xattr_put_super(sb);
 445         jbd2_journal_destroy(sbi->s_journal);
 446         if (!(sb->s_flags & MS_RDONLY)) {
 447                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 448                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 449                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 450                 mark_buffer_dirty(sbi->s_sbh);
 451                 ext4_commit_super(sb, es, 1);
 452         }
 453
 454         for (i = 0; i < sbi->s_gdb_count; i++)
 455                 brelse(sbi->s_group_desc[i]);
 456         kfree(sbi->s_group_desc);
 457         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 458         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 459         percpu_counter_destroy(&sbi->s_dirs_counter);
 460         brelse(sbi->s_sbh);
 461 #ifdef CONFIG_QUOTA
 462         for (i = 0; i < MAXQUOTAS; i++)
 463                 kfree(sbi->s_qf_names[i]);
 464 #endif
 465
 466         /* Debugging code just in case the in-memory inode orphan list
 467          * isn't empty.  The on-disk one can be non-empty if we've
 468          * detected an error and taken the fs readonly, but the
 469          * in-memory list had better be clean by this point. */
 470         if (!list_empty(&sbi->s_orphan))
 471                 dump_orphan_list(sb, sbi);
 472         J_ASSERT(list_empty(&sbi->s_orphan));
 473
 474         invalidate_bdev(sb->s_bdev);
 475         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 476                 /*
 477                  * Invalidate the journal device's buffers.  We don't want them
 478                  * floating about in memory - the physical journal device may
 479                  * hotswapped, and it breaks the `ro-after' testing code.
 480                  */
 481                 sync_blockdev(sbi->journal_bdev);
 482                 invalidate_bdev(sbi->journal_bdev);
 483                 ext4_blkdev_remove(sbi);
 484         }
 485         sb->s_fs_info = NULL;
 486         kfree(sbi);
 487         return;
 488 }
 489
 490 static struct kmem_cache *ext4_inode_cachep;
 491
 492 /*
 493  * Called inside transaction, so use GFP_NOFS
 494  */
 495 static struct inode *ext4_alloc_inode(struct super_block *sb)
 496 {
 497         struct ext4_inode_info *ei;
 498
 499         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 500         if (!ei)
 501                 return NULL;
 502 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 503         ei->i_acl = EXT4_ACL_NOT_CACHED;
 504         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 505 #endif
 506         ei->i_block_alloc_info = NULL;
 507         ei->vfs_inode.i_version = 1;
 508         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 509         return &ei->vfs_inode;
 510 }
 511
 512 static void ext4_destroy_inode(struct inode *inode)
 513 {
 514         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 515                 printk("EXT4 Inode %p: orphan list check failed!\n",
 516                         EXT4_I(inode));
 517                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 518                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 519                                 true);
 520                 dump_stack();
 521         }
 522         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 523 }
 524
 525 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
 526 {
 527         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 528
 529         INIT_LIST_HEAD(&ei->i_orphan);
 530 #ifdef CONFIG_EXT4DEV_FS_XATTR
 531         init_rwsem(&ei->xattr_sem);
 532 #endif
 533         mutex_init(&ei->truncate_mutex);
 534         inode_init_once(&ei->vfs_inode);
 535 }
 536
 537 static int init_inodecache(void)
 538 {
 539         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 540                                              sizeof(struct ext4_inode_info),
 541                                              0, (SLAB_RECLAIM_ACCOUNT|
 542                                                 SLAB_MEM_SPREAD),
 543                                              init_once, NULL);
 544         if (ext4_inode_cachep == NULL)
 545                 return -ENOMEM;
 546         return 0;
 547 }
 548
 549 static void destroy_inodecache(void)
 550 {
 551         kmem_cache_destroy(ext4_inode_cachep);
 552 }
 553
 554 static void ext4_clear_inode(struct inode *inode)
 555 {
 556         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
 557 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 558         if (EXT4_I(inode)->i_acl &&
 559                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
 560                 posix_acl_release(EXT4_I(inode)->i_acl);
 561                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
 562         }
 563         if (EXT4_I(inode)->i_default_acl &&
 564                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
 565                 posix_acl_release(EXT4_I(inode)->i_default_acl);
 566                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
 567         }
 568 #endif
 569         ext4_discard_reservation(inode);
 570         EXT4_I(inode)->i_block_alloc_info = NULL;
 571         if (unlikely(rsv))
 572                 kfree(rsv);
 573 }
 574
 575 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
 576 {
 577 #if defined(CONFIG_QUOTA)
 578         struct ext4_sb_info *sbi = EXT4_SB(sb);
 579
 580         if (sbi->s_jquota_fmt)
 581                 seq_printf(seq, ",jqfmt=%s",
 582                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
 583
 584         if (sbi->s_qf_names[USRQUOTA])
 585                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 586
 587         if (sbi->s_qf_names[GRPQUOTA])
 588                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 589
 590         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
 591                 seq_puts(seq, ",usrquota");
 592
 593         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
 594                 seq_puts(seq, ",grpquota");
 595 #endif
 596 }
 597
 598 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 599 {
 600         struct super_block *sb = vfs->mnt_sb;
 601
 602         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 603                 seq_puts(seq, ",data=journal");
 604         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 605                 seq_puts(seq, ",data=ordered");
 606         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 607                 seq_puts(seq, ",data=writeback");
 608
 609         ext4_show_quota_options(seq, sb);
 610
 611         return 0;
 612 }
 613
 614
 615 static struct dentry *ext4_get_dentry(struct super_block *sb, void *vobjp)
 616 {
 617         __u32 *objp = vobjp;
 618         unsigned long ino = objp[0];
 619         __u32 generation = objp[1];
 620         struct inode *inode;
 621         struct dentry *result;
 622
 623         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 624                 return ERR_PTR(-ESTALE);
 625         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 626                 return ERR_PTR(-ESTALE);
 627
 628         /* iget isn't really right if the inode is currently unallocated!!
 629          *
 630          * ext4_read_inode will return a bad_inode if the inode had been
 631          * deleted, so we should be safe.
 632          *
 633          * Currently we don't know the generation for parent directory, so
 634          * a generation of 0 means "accept any"
 635          */
 636         inode = iget(sb, ino);
 637         if (inode == NULL)
 638                 return ERR_PTR(-ENOMEM);
 639         if (is_bad_inode(inode) ||
 640             (generation && inode->i_generation != generation)) {
 641                 iput(inode);
 642                 return ERR_PTR(-ESTALE);
 643         }
 644         /* now to find a dentry.
 645          * If possible, get a well-connected one
 646          */
 647         result = d_alloc_anon(inode);
 648         if (!result) {
 649                 iput(inode);
 650                 return ERR_PTR(-ENOMEM);
 651         }
 652         return result;
 653 }
 654
 655 #ifdef CONFIG_QUOTA
 656 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 657 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 658
 659 static int ext4_dquot_initialize(struct inode *inode, int type);
 660 static int ext4_dquot_drop(struct inode *inode);
 661 static int ext4_write_dquot(struct dquot *dquot);
 662 static int ext4_acquire_dquot(struct dquot *dquot);
 663 static int ext4_release_dquot(struct dquot *dquot);
 664 static int ext4_mark_dquot_dirty(struct dquot *dquot);
 665 static int ext4_write_info(struct super_block *sb, int type);
 666 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
 667 static int ext4_quota_on_mount(struct super_block *sb, int type);
 668 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
 669                                size_t len, loff_t off);
 670 static ssize_t ext4_quota_write(struct super_block *sb, int type,
 671                                 const char *data, size_t len, loff_t off);
 672
 673 static struct dquot_operations ext4_quota_operations = {
 674         .initialize     = ext4_dquot_initialize,
 675         .drop           = ext4_dquot_drop,
 676         .alloc_space    = dquot_alloc_space,
 677         .alloc_inode    = dquot_alloc_inode,
 678         .free_space     = dquot_free_space,
 679         .free_inode     = dquot_free_inode,
 680         .transfer       = dquot_transfer,
 681         .write_dquot    = ext4_write_dquot,
 682         .acquire_dquot  = ext4_acquire_dquot,
 683         .release_dquot  = ext4_release_dquot,
 684         .mark_dirty     = ext4_mark_dquot_dirty,
 685         .write_info     = ext4_write_info
 686 };
 687
 688 static struct quotactl_ops ext4_qctl_operations = {
 689         .quota_on       = ext4_quota_on,
 690         .quota_off      = vfs_quota_off,
 691         .quota_sync     = vfs_quota_sync,
 692         .get_info       = vfs_get_dqinfo,
 693         .set_info       = vfs_set_dqinfo,
 694         .get_dqblk      = vfs_get_dqblk,
 695         .set_dqblk      = vfs_set_dqblk
 696 };
 697 #endif
 698
 699 static const struct super_operations ext4_sops = {
 700         .alloc_inode    = ext4_alloc_inode,
 701         .destroy_inode  = ext4_destroy_inode,
 702         .read_inode     = ext4_read_inode,
 703         .write_inode    = ext4_write_inode,
 704         .dirty_inode    = ext4_dirty_inode,
 705         .delete_inode   = ext4_delete_inode,
 706         .put_super      = ext4_put_super,
 707         .write_super    = ext4_write_super,
 708         .sync_fs        = ext4_sync_fs,
 709         .write_super_lockfs = ext4_write_super_lockfs,
 710         .unlockfs       = ext4_unlockfs,
 711         .statfs         = ext4_statfs,
 712         .remount_fs     = ext4_remount,
 713         .clear_inode    = ext4_clear_inode,
 714         .show_options   = ext4_show_options,
 715 #ifdef CONFIG_QUOTA
 716         .quota_read     = ext4_quota_read,
 717         .quota_write    = ext4_quota_write,
 718 #endif
 719 };
 720
 721 static struct export_operations ext4_export_ops = {
 722         .get_parent = ext4_get_parent,
 723         .get_dentry = ext4_get_dentry,
 724 };
 725
 726 enum {
 727         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 728         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 729         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 730         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 731         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
 732         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
 733         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 734         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 735         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 736         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 737         Opt_grpquota, Opt_extents,
 738 };
 739
 740 static match_table_t tokens = {
 741         {Opt_bsd_df, "bsddf"},
 742         {Opt_minix_df, "minixdf"},
 743         {Opt_grpid, "grpid"},
 744         {Opt_grpid, "bsdgroups"},
 745         {Opt_nogrpid, "nogrpid"},
 746         {Opt_nogrpid, "sysvgroups"},
 747         {Opt_resgid, "resgid=%u"},
 748         {Opt_resuid, "resuid=%u"},
 749         {Opt_sb, "sb=%u"},
 750         {Opt_err_cont, "errors=continue"},
 751         {Opt_err_panic, "errors=panic"},
 752         {Opt_err_ro, "errors=remount-ro"},
 753         {Opt_nouid32, "nouid32"},
 754         {Opt_nocheck, "nocheck"},
 755         {Opt_nocheck, "check=none"},
 756         {Opt_debug, "debug"},
 757         {Opt_oldalloc, "oldalloc"},
 758         {Opt_orlov, "orlov"},
 759         {Opt_user_xattr, "user_xattr"},
 760         {Opt_nouser_xattr, "nouser_xattr"},
 761         {Opt_acl, "acl"},
 762         {Opt_noacl, "noacl"},
 763         {Opt_reservation, "reservation"},
 764         {Opt_noreservation, "noreservation"},
 765         {Opt_noload, "noload"},
 766         {Opt_nobh, "nobh"},
 767         {Opt_bh, "bh"},
 768         {Opt_commit, "commit=%u"},
 769         {Opt_journal_update, "journal=update"},
 770         {Opt_journal_inum, "journal=%u"},
 771         {Opt_journal_dev, "journal_dev=%u"},
 772         {Opt_abort, "abort"},
 773         {Opt_data_journal, "data=journal"},
 774         {Opt_data_ordered, "data=ordered"},
 775         {Opt_data_writeback, "data=writeback"},
 776         {Opt_offusrjquota, "usrjquota="},
 777         {Opt_usrjquota, "usrjquota=%s"},
 778         {Opt_offgrpjquota, "grpjquota="},
 779         {Opt_grpjquota, "grpjquota=%s"},
 780         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 781         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 782         {Opt_grpquota, "grpquota"},
 783         {Opt_noquota, "noquota"},
 784         {Opt_quota, "quota"},
 785         {Opt_usrquota, "usrquota"},
 786         {Opt_barrier, "barrier=%u"},
 787         {Opt_extents, "extents"},
 788         {Opt_err, NULL},
 789         {Opt_resize, "resize"},
 790 };
 791
 792 static ext4_fsblk_t get_sb_block(void **data)
 793 {
 794         ext4_fsblk_t    sb_block;
 795         char            *options = (char *) *data;
 796
 797         if (!options || strncmp(options, "sb=", 3) != 0)
 798                 return 1;       /* Default location */
 799         options += 3;
 800         /*todo: use simple_strtoll with >32bit ext4 */
 801         sb_block = simple_strtoul(options, &options, 0);
 802         if (*options && *options != ',') {
 803                 printk("EXT4-fs: Invalid sb specification: %s\n",
 804                        (char *) *data);
 805                 return 1;
 806         }
 807         if (*options == ',')
 808                 options++;
 809         *data = (void *) options;
 810         return sb_block;
 811 }
 812
 813 static int parse_options (char *options, struct super_block *sb,
 814                           unsigned int *inum, unsigned long *journal_devnum,
 815                           ext4_fsblk_t *n_blocks_count, int is_remount)
 816 {
 817         struct ext4_sb_info *sbi = EXT4_SB(sb);
 818         char * p;
 819         substring_t args[MAX_OPT_ARGS];
 820         int data_opt = 0;
 821         int option;
 822 #ifdef CONFIG_QUOTA
 823         int qtype;
 824         char *qname;
 825 #endif
 826
 827         if (!options)
 828                 return 1;
 829
 830         while ((p = strsep (&options, ",")) != NULL) {
 831                 int token;
 832                 if (!*p)
 833                         continue;
 834
 835                 token = match_token(p, tokens, args);
 836                 switch (token) {
 837                 case Opt_bsd_df:
 838                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 839                         break;
 840                 case Opt_minix_df:
 841                         set_opt (sbi->s_mount_opt, MINIX_DF);
 842                         break;
 843                 case Opt_grpid:
 844                         set_opt (sbi->s_mount_opt, GRPID);
 845                         break;
 846                 case Opt_nogrpid:
 847                         clear_opt (sbi->s_mount_opt, GRPID);
 848                         break;
 849                 case Opt_resuid:
 850                         if (match_int(&args[0], &option))
 851                                 return 0;
 852                         sbi->s_resuid = option;
 853                         break;
 854                 case Opt_resgid:
 855                         if (match_int(&args[0], &option))
 856                                 return 0;
 857                         sbi->s_resgid = option;
 858                         break;
 859                 case Opt_sb:
 860                         /* handled by get_sb_block() instead of here */
 861                         /* *sb_block = match_int(&args[0]); */
 862                         break;
 863                 case Opt_err_panic:
 864                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 865                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 866                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 867                         break;
 868                 case Opt_err_ro:
 869                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 870                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 871                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 872                         break;
 873                 case Opt_err_cont:
 874                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 875                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 876                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 877                         break;
 878                 case Opt_nouid32:
 879                         set_opt (sbi->s_mount_opt, NO_UID32);
 880                         break;
 881                 case Opt_nocheck:
 882                         clear_opt (sbi->s_mount_opt, CHECK);
 883                         break;
 884                 case Opt_debug:
 885                         set_opt (sbi->s_mount_opt, DEBUG);
 886                         break;
 887                 case Opt_oldalloc:
 888                         set_opt (sbi->s_mount_opt, OLDALLOC);
 889                         break;
 890                 case Opt_orlov:
 891                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 892                         break;
 893 #ifdef CONFIG_EXT4DEV_FS_XATTR
 894                 case Opt_user_xattr:
 895                         set_opt (sbi->s_mount_opt, XATTR_USER);
 896                         break;
 897                 case Opt_nouser_xattr:
 898                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 899                         break;
 900 #else
 901                 case Opt_user_xattr:
 902                 case Opt_nouser_xattr:
 903                         printk("EXT4 (no)user_xattr options not supported\n");
 904                         break;
 905 #endif
 906 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
 907                 case Opt_acl:
 908                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 909                         break;
 910                 case Opt_noacl:
 911                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 912                         break;
 913 #else
 914                 case Opt_acl:
 915                 case Opt_noacl:
 916                         printk("EXT4 (no)acl options not supported\n");
 917                         break;
 918 #endif
 919                 case Opt_reservation:
 920                         set_opt(sbi->s_mount_opt, RESERVATION);
 921                         break;
 922                 case Opt_noreservation:
 923                         clear_opt(sbi->s_mount_opt, RESERVATION);
 924                         break;
 925                 case Opt_journal_update:
 926                         /* @@@ FIXME */
 927                         /* Eventually we will want to be able to create
 928                            a journal file here.  For now, only allow the
 929                            user to specify an existing inode to be the
 930                            journal file. */
 931                         if (is_remount) {
 932                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 933                                        "journal on remount\n");
 934                                 return 0;
 935                         }
 936                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 937                         break;
 938                 case Opt_journal_inum:
 939                         if (is_remount) {
 940                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 941                                        "journal on remount\n");
 942                                 return 0;
 943                         }
 944                         if (match_int(&args[0], &option))
 945                                 return 0;
 946                         *inum = option;
 947                         break;
 948                 case Opt_journal_dev:
 949                         if (is_remount) {
 950                                 printk(KERN_ERR "EXT4-fs: cannot specify "
 951                                        "journal on remount\n");
 952                                 return 0;
 953                         }
 954                         if (match_int(&args[0], &option))
 955                                 return 0;
 956                         *journal_devnum = option;
 957                         break;
 958                 case Opt_noload:
 959                         set_opt (sbi->s_mount_opt, NOLOAD);
 960                         break;
 961                 case Opt_commit:
 962                         if (match_int(&args[0], &option))
 963                                 return 0;
 964                         if (option < 0)
 965                                 return 0;
 966                         if (option == 0)
 967                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 968                         sbi->s_commit_interval = HZ * option;
 969                         break;
 970                 case Opt_data_journal:
 971                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
 972                         goto datacheck;
 973                 case Opt_data_ordered:
 974                         data_opt = EXT4_MOUNT_ORDERED_DATA;
 975                         goto datacheck;
 976                 case Opt_data_writeback:
 977                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
 978                 datacheck:
 979                         if (is_remount) {
 980                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
 981                                                 != data_opt) {
 982                                         printk(KERN_ERR
 983                                                 "EXT4-fs: cannot change data "
 984                                                 "mode on remount\n");
 985                                         return 0;
 986                                 }
 987                         } else {
 988                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
 989                                 sbi->s_mount_opt |= data_opt;
 990                         }
 991                         break;
 992 #ifdef CONFIG_QUOTA
 993                 case Opt_usrjquota:
 994                         qtype = USRQUOTA;
 995                         goto set_qf_name;
 996                 case Opt_grpjquota:
 997                         qtype = GRPQUOTA;
 998 set_qf_name:
 999                         if (sb_any_quota_enabled(sb)) {
1000                                 printk(KERN_ERR
1001                                         "EXT4-fs: Cannot change journalled "
1002                                         "quota options when quota turned on.\n");
1003                                 return 0;
1004                         }
1005                         qname = match_strdup(&args[0]);
1006                         if (!qname) {
1007                                 printk(KERN_ERR
1008                                         "EXT4-fs: not enough memory for "
1009                                         "storing quotafile name.\n");
1010                                 return 0;
1011                         }
1012                         if (sbi->s_qf_names[qtype] &&
1013                             strcmp(sbi->s_qf_names[qtype], qname)) {
1014                                 printk(KERN_ERR
1015                                         "EXT4-fs: %s quota file already "
1016                                         "specified.\n", QTYPE2NAME(qtype));
1017                                 kfree(qname);
1018                                 return 0;
1019                         }
1020                         sbi->s_qf_names[qtype] = qname;
1021                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1022                                 printk(KERN_ERR
1023                                         "EXT4-fs: quotafile must be on "
1024                                         "filesystem root.\n");
1025                                 kfree(sbi->s_qf_names[qtype]);
1026                                 sbi->s_qf_names[qtype] = NULL;
1027                                 return 0;
1028                         }
1029                         set_opt(sbi->s_mount_opt, QUOTA);
1030                         break;
1031                 case Opt_offusrjquota:
1032                         qtype = USRQUOTA;
1033                         goto clear_qf_name;
1034                 case Opt_offgrpjquota:
1035                         qtype = GRPQUOTA;
1036 clear_qf_name:
1037                         if (sb_any_quota_enabled(sb)) {
1038                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1039                                         "journalled quota options when "
1040                                         "quota turned on.\n");
1041                                 return 0;
1042                         }
1043                         /*
1044                          * The space will be released later when all options
1045                          * are confirmed to be correct
1046                          */
1047                         sbi->s_qf_names[qtype] = NULL;
1048                         break;
1049                 case Opt_jqfmt_vfsold:
1050                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1051                         break;
1052                 case Opt_jqfmt_vfsv0:
1053                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1054                         break;
1055                 case Opt_quota:
1056                 case Opt_usrquota:
1057                         set_opt(sbi->s_mount_opt, QUOTA);
1058                         set_opt(sbi->s_mount_opt, USRQUOTA);
1059                         break;
1060                 case Opt_grpquota:
1061                         set_opt(sbi->s_mount_opt, QUOTA);
1062                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1063                         break;
1064                 case Opt_noquota:
1065                         if (sb_any_quota_enabled(sb)) {
1066                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1067                                         "options when quota turned on.\n");
1068                                 return 0;
1069                         }
1070                         clear_opt(sbi->s_mount_opt, QUOTA);
1071                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1072                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1073                         break;
1074 #else
1075                 case Opt_quota:
1076                 case Opt_usrquota:
1077                 case Opt_grpquota:
1078                 case Opt_usrjquota:
1079                 case Opt_grpjquota:
1080                 case Opt_offusrjquota:
1081                 case Opt_offgrpjquota:
1082                 case Opt_jqfmt_vfsold:
1083                 case Opt_jqfmt_vfsv0:
1084                         printk(KERN_ERR
1085                                 "EXT4-fs: journalled quota options not "
1086                                 "supported.\n");
1087                         break;
1088                 case Opt_noquota:
1089                         break;
1090 #endif
1091                 case Opt_abort:
1092                         set_opt(sbi->s_mount_opt, ABORT);
1093                         break;
1094                 case Opt_barrier:
1095                         if (match_int(&args[0], &option))
1096                                 return 0;
1097                         if (option)
1098                                 set_opt(sbi->s_mount_opt, BARRIER);
1099                         else
1100                                 clear_opt(sbi->s_mount_opt, BARRIER);
1101                         break;
1102                 case Opt_ignore:
1103                         break;
1104                 case Opt_resize:
1105                         if (!is_remount) {
1106                                 printk("EXT4-fs: resize option only available "
1107                                         "for remount\n");
1108                                 return 0;
1109                         }
1110                         if (match_int(&args[0], &option) != 0)
1111                                 return 0;
1112                         *n_blocks_count = option;
1113                         break;
1114                 case Opt_nobh:
1115                         set_opt(sbi->s_mount_opt, NOBH);
1116                         break;
1117                 case Opt_bh:
1118                         clear_opt(sbi->s_mount_opt, NOBH);
1119                         break;
1120                 case Opt_extents:
1121                         set_opt (sbi->s_mount_opt, EXTENTS);
1122                         break;
1123                 default:
1124                         printk (KERN_ERR
1125                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1126                                 "or missing value\n", p);
1127                         return 0;
1128                 }
1129         }
1130 #ifdef CONFIG_QUOTA
1131         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1132                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1133                      sbi->s_qf_names[USRQUOTA])
1134                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1135
1136                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1137                      sbi->s_qf_names[GRPQUOTA])
1138                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1139
1140                 if ((sbi->s_qf_names[USRQUOTA] &&
1141                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1142                     (sbi->s_qf_names[GRPQUOTA] &&
1143                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1144                         printk(KERN_ERR "EXT4-fs: old and new quota "
1145                                         "format mixing.\n");
1146                         return 0;
1147                 }
1148
1149                 if (!sbi->s_jquota_fmt) {
1150                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1151                                         "not specified.\n");
1152                         return 0;
1153                 }
1154         } else {
1155                 if (sbi->s_jquota_fmt) {
1156                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1157                                         "specified with no journalling "
1158                                         "enabled.\n");
1159                         return 0;
1160                 }
1161         }
1162 #endif
1163         return 1;
1164 }
1165
1166 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1167                             int read_only)
1168 {
1169         struct ext4_sb_info *sbi = EXT4_SB(sb);
1170         int res = 0;
1171
1172         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1173                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1174                         "forcing read-only mode\n");
1175                 res = MS_RDONLY;
1176         }
1177         if (read_only)
1178                 return res;
1179         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1180                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1181                         "running e2fsck is recommended\n");
1182         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1183                 printk (KERN_WARNING
1184                         "EXT4-fs warning: mounting fs with errors, "
1185                         "running e2fsck is recommended\n");
1186         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1187                  le16_to_cpu(es->s_mnt_count) >=
1188                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1189                 printk (KERN_WARNING
1190                         "EXT4-fs warning: maximal mount count reached, "
1191                         "running e2fsck is recommended\n");
1192         else if (le32_to_cpu(es->s_checkinterval) &&
1193                 (le32_to_cpu(es->s_lastcheck) +
1194                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1195                 printk (KERN_WARNING
1196                         "EXT4-fs warning: checktime reached, "
1197                         "running e2fsck is recommended\n");
1198 #if 0
1199                 /* @@@ We _will_ want to clear the valid bit if we find
1200                  * inconsistencies, to force a fsck at reboot.  But for
1201                  * a plain journaled filesystem we can keep it set as
1202                  * valid forever! :)
1203                  */
1204         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1205 #endif
1206         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1207                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1208         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1209         es->s_mtime = cpu_to_le32(get_seconds());
1210         ext4_update_dynamic_rev(sb);
1211         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1212
1213         ext4_commit_super(sb, es, 1);
1214         if (test_opt(sb, DEBUG))
1215                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1216                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1217                         sb->s_blocksize,
1218                         sbi->s_groups_count,
1219                         EXT4_BLOCKS_PER_GROUP(sb),
1220                         EXT4_INODES_PER_GROUP(sb),
1221                         sbi->s_mount_opt);
1222
1223         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1224         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1225                 char b[BDEVNAME_SIZE];
1226
1227                 printk("external journal on %s\n",
1228                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1229         } else {
1230                 printk("internal journal\n");
1231         }
1232         return res;
1233 }
1234
1235 /* Called at mount-time, super-block is locked */
1236 static int ext4_check_descriptors (struct super_block * sb)
1237 {
1238         struct ext4_sb_info *sbi = EXT4_SB(sb);
1239         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1240         ext4_fsblk_t last_block;
1241         ext4_fsblk_t block_bitmap;
1242         ext4_fsblk_t inode_bitmap;
1243         ext4_fsblk_t inode_table;
1244         struct ext4_group_desc * gdp = NULL;
1245         int desc_block = 0;
1246         int i;
1247
1248         ext4_debug ("Checking group descriptors");
1249
1250         for (i = 0; i < sbi->s_groups_count; i++)
1251         {
1252                 if (i == sbi->s_groups_count - 1)
1253                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1254                 else
1255                         last_block = first_block +
1256                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1257
1258                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1259                         gdp = (struct ext4_group_desc *)
1260                                         sbi->s_group_desc[desc_block++]->b_data;
1261                 block_bitmap = ext4_block_bitmap(sb, gdp);
1262                 if (block_bitmap < first_block || block_bitmap > last_block)
1263                 {
1264                         ext4_error (sb, "ext4_check_descriptors",
1265                                     "Block bitmap for group %d"
1266                                     " not in group (block %llu)!",
1267                                     i, block_bitmap);
1268                         return 0;
1269                 }
1270                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1271                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1272                 {
1273                         ext4_error (sb, "ext4_check_descriptors",
1274                                     "Inode bitmap for group %d"
1275                                     " not in group (block %llu)!",
1276                                     i, inode_bitmap);
1277                         return 0;
1278                 }
1279                 inode_table = ext4_inode_table(sb, gdp);
1280                 if (inode_table < first_block ||
1281                     inode_table + sbi->s_itb_per_group > last_block)
1282                 {
1283                         ext4_error (sb, "ext4_check_descriptors",
1284                                     "Inode table for group %d"
1285                                     " not in group (block %llu)!",
1286                                     i, inode_table);
1287                         return 0;
1288                 }
1289                 first_block += EXT4_BLOCKS_PER_GROUP(sb);
1290                 gdp = (struct ext4_group_desc *)
1291                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1292         }
1293
1294         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1295         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1296         return 1;
1297 }
1298
1299
1300 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1301  * the superblock) which were deleted from all directories, but held open by
1302  * a process at the time of a crash.  We walk the list and try to delete these
1303  * inodes at recovery time (only with a read-write filesystem).
1304  *
1305  * In order to keep the orphan inode chain consistent during traversal (in
1306  * case of crash during recovery), we link each inode into the superblock
1307  * orphan list_head and handle it the same way as an inode deletion during
1308  * normal operation (which journals the operations for us).
1309  *
1310  * We only do an iget() and an iput() on each inode, which is very safe if we
1311  * accidentally point at an in-use or already deleted inode.  The worst that
1312  * can happen in this case is that we get a "bit already cleared" message from
1313  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1314  * e2fsck was run on this filesystem, and it must have already done the orphan
1315  * inode cleanup for us, so we can safely abort without any further action.
1316  */
1317 static void ext4_orphan_cleanup (struct super_block * sb,
1318                                  struct ext4_super_block * es)
1319 {
1320         unsigned int s_flags = sb->s_flags;
1321         int nr_orphans = 0, nr_truncates = 0;
1322 #ifdef CONFIG_QUOTA
1323         int i;
1324 #endif
1325         if (!es->s_last_orphan) {
1326                 jbd_debug(4, "no orphan inodes to clean up\n");
1327                 return;
1328         }
1329
1330         if (bdev_read_only(sb->s_bdev)) {
1331                 printk(KERN_ERR "EXT4-fs: write access "
1332                         "unavailable, skipping orphan cleanup.\n");
1333                 return;
1334         }
1335
1336         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1337                 if (es->s_last_orphan)
1338                         jbd_debug(1, "Errors on filesystem, "
1339                                   "clearing orphan list.\n");
1340                 es->s_last_orphan = 0;
1341                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1342                 return;
1343         }
1344
1345         if (s_flags & MS_RDONLY) {
1346                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1347                        sb->s_id);
1348                 sb->s_flags &= ~MS_RDONLY;
1349         }
1350 #ifdef CONFIG_QUOTA
1351         /* Needed for iput() to work correctly and not trash data */
1352         sb->s_flags |= MS_ACTIVE;
1353         /* Turn on quotas so that they are updated correctly */
1354         for (i = 0; i < MAXQUOTAS; i++) {
1355                 if (EXT4_SB(sb)->s_qf_names[i]) {
1356                         int ret = ext4_quota_on_mount(sb, i);
1357                         if (ret < 0)
1358                                 printk(KERN_ERR
1359                                         "EXT4-fs: Cannot turn on journalled "
1360                                         "quota: error %d\n", ret);
1361                 }
1362         }
1363 #endif
1364
1365         while (es->s_last_orphan) {
1366                 struct inode *inode;
1367
1368                 if (!(inode =
1369                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1370                         es->s_last_orphan = 0;
1371                         break;
1372                 }
1373
1374                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1375                 DQUOT_INIT(inode);
1376                 if (inode->i_nlink) {
1377                         printk(KERN_DEBUG
1378                                 "%s: truncating inode %lu to %Ld bytes\n",
1379                                 __FUNCTION__, inode->i_ino, inode->i_size);
1380                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1381                                   inode->i_ino, inode->i_size);
1382                         ext4_truncate(inode);
1383                         nr_truncates++;
1384                 } else {
1385                         printk(KERN_DEBUG
1386                                 "%s: deleting unreferenced inode %lu\n",
1387                                 __FUNCTION__, inode->i_ino);
1388                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1389                                   inode->i_ino);
1390                         nr_orphans++;
1391                 }
1392                 iput(inode);  /* The delete magic happens here! */
1393         }
1394
1395 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1396
1397         if (nr_orphans)
1398                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1399                        sb->s_id, PLURAL(nr_orphans));
1400         if (nr_truncates)
1401                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1402                        sb->s_id, PLURAL(nr_truncates));
1403 #ifdef CONFIG_QUOTA
1404         /* Turn quotas off */
1405         for (i = 0; i < MAXQUOTAS; i++) {
1406                 if (sb_dqopt(sb)->files[i])
1407                         vfs_quota_off(sb, i);
1408         }
1409 #endif
1410         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1411 }
1412
1413 #define log2(n) ffz(~(n))
1414
1415 /*
1416  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1417  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1418  * We need to be 1 filesystem block less than the 2^32 sector limit.
1419  */
1420 static loff_t ext4_max_size(int bits)
1421 {
1422         loff_t res = EXT4_NDIR_BLOCKS;
1423         /* This constant is calculated to be the largest file size for a
1424          * dense, 4k-blocksize file such that the total number of
1425          * sectors in the file, including data and all indirect blocks,
1426          * does not exceed 2^32. */
1427         const loff_t upper_limit = 0x1ff7fffd000LL;
1428
1429         res += 1LL << (bits-2);
1430         res += 1LL << (2*(bits-2));
1431         res += 1LL << (3*(bits-2));
1432         res <<= bits;
1433         if (res > upper_limit)
1434                 res = upper_limit;
1435         return res;
1436 }
1437
1438 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1439                                 ext4_fsblk_t logical_sb_block, int nr)
1440 {
1441         struct ext4_sb_info *sbi = EXT4_SB(sb);
1442         unsigned long bg, first_meta_bg;
1443         int has_super = 0;
1444
1445         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1446
1447         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1448             nr < first_meta_bg)
1449                 return logical_sb_block + nr + 1;
1450         bg = sbi->s_desc_per_block * nr;
1451         if (ext4_bg_has_super(sb, bg))
1452                 has_super = 1;
1453         return (has_super + ext4_group_first_block_no(sb, bg));
1454 }
1455
1456
1457 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1458 {
1459         struct buffer_head * bh;
1460         struct ext4_super_block *es = NULL;
1461         struct ext4_sb_info *sbi;
1462         ext4_fsblk_t block;
1463         ext4_fsblk_t sb_block = get_sb_block(&data);
1464         ext4_fsblk_t logical_sb_block;
1465         unsigned long offset = 0;
1466         unsigned int journal_inum = 0;
1467         unsigned long journal_devnum = 0;
1468         unsigned long def_mount_opts;
1469         struct inode *root;
1470         int blocksize;
1471         int hblock;
1472         int db_count;
1473         int i;
1474         int needs_recovery;
1475         __le32 features;
1476         __u64 blocks_count;
1477
1478         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1479         if (!sbi)
1480                 return -ENOMEM;
1481         sb->s_fs_info = sbi;
1482         sbi->s_mount_opt = 0;
1483         sbi->s_resuid = EXT4_DEF_RESUID;
1484         sbi->s_resgid = EXT4_DEF_RESGID;
1485
1486         unlock_kernel();
1487
1488         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1489         if (!blocksize) {
1490                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1491                 goto out_fail;
1492         }
1493
1494         /*
1495          * The ext4 superblock will not be buffer aligned for other than 1kB
1496          * block sizes.  We need to calculate the offset from buffer start.
1497          */
1498         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1499                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1500                 offset = do_div(logical_sb_block, blocksize);
1501         } else {
1502                 logical_sb_block = sb_block;
1503         }
1504
1505         if (!(bh = sb_bread(sb, logical_sb_block))) {
1506                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1507                 goto out_fail;
1508         }
1509         /*
1510          * Note: s_es must be initialized as soon as possible because
1511          *       some ext4 macro-instructions depend on its value
1512          */
1513         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1514         sbi->s_es = es;
1515         sb->s_magic = le16_to_cpu(es->s_magic);
1516         if (sb->s_magic != EXT4_SUPER_MAGIC)
1517                 goto cantfind_ext4;
1518
1519         /* Set defaults before we parse the mount options */
1520         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1521         if (def_mount_opts & EXT4_DEFM_DEBUG)
1522                 set_opt(sbi->s_mount_opt, DEBUG);
1523         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1524                 set_opt(sbi->s_mount_opt, GRPID);
1525         if (def_mount_opts & EXT4_DEFM_UID16)
1526                 set_opt(sbi->s_mount_opt, NO_UID32);
1527 #ifdef CONFIG_EXT4DEV_FS_XATTR
1528         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1529                 set_opt(sbi->s_mount_opt, XATTR_USER);
1530 #endif
1531 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
1532         if (def_mount_opts & EXT4_DEFM_ACL)
1533                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1534 #endif
1535         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1536                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1537         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1538                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1539         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1540                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1541
1542         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1543                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1544         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_RO)
1545                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1546         else
1547                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1548
1549         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1550         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1551
1552         set_opt(sbi->s_mount_opt, RESERVATION);
1553
1554         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1555                             NULL, 0))
1556                 goto failed_mount;
1557
1558         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1559                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1560
1561         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1562             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1563              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1564              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1565                 printk(KERN_WARNING
1566                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1567                        "running e2fsck is recommended\n");
1568         /*
1569          * Check feature flags regardless of the revision level, since we
1570          * previously didn't change the revision level when setting the flags,
1571          * so there is a chance incompat flags are set on a rev 0 filesystem.
1572          */
1573         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1574         if (features) {
1575                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1576                        "unsupported optional features (%x).\n",
1577                        sb->s_id, le32_to_cpu(features));
1578                 goto failed_mount;
1579         }
1580         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1581         if (!(sb->s_flags & MS_RDONLY) && features) {
1582                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1583                        "unsupported optional features (%x).\n",
1584                        sb->s_id, le32_to_cpu(features));
1585                 goto failed_mount;
1586         }
1587         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1588
1589         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1590             blocksize > EXT4_MAX_BLOCK_SIZE) {
1591                 printk(KERN_ERR
1592                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1593                        blocksize, sb->s_id);
1594                 goto failed_mount;
1595         }
1596
1597         hblock = bdev_hardsect_size(sb->s_bdev);
1598         if (sb->s_blocksize != blocksize) {
1599                 /*
1600                  * Make sure the blocksize for the filesystem is larger
1601                  * than the hardware sectorsize for the machine.
1602                  */
1603                 if (blocksize < hblock) {
1604                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1605                                "device blocksize %d.\n", blocksize, hblock);
1606                         goto failed_mount;
1607                 }
1608
1609                 brelse (bh);
1610                 sb_set_blocksize(sb, blocksize);
1611                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1612                 offset = do_div(logical_sb_block, blocksize);
1613                 bh = sb_bread(sb, logical_sb_block);
1614                 if (!bh) {
1615                         printk(KERN_ERR
1616                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1617                         goto failed_mount;
1618                 }
1619                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1620                 sbi->s_es = es;
1621                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1622                         printk (KERN_ERR
1623                                 "EXT4-fs: Magic mismatch, very weird !\n");
1624                         goto failed_mount;
1625                 }
1626         }
1627
1628         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1629
1630         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1631                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1632                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
1633         } else {
1634                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1635                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1636                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
1637                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1638                     (sbi->s_inode_size > blocksize)) {
1639                         printk (KERN_ERR
1640                                 "EXT4-fs: unsupported inode size: %d\n",
1641                                 sbi->s_inode_size);
1642                         goto failed_mount;
1643                 }
1644         }
1645         sbi->s_frag_size = EXT4_MIN_FRAG_SIZE <<
1646                                    le32_to_cpu(es->s_log_frag_size);
1647         if (blocksize != sbi->s_frag_size) {
1648                 printk(KERN_ERR
1649                        "EXT4-fs: fragsize %lu != blocksize %u (unsupported)\n",
1650                        sbi->s_frag_size, blocksize);
1651                 goto failed_mount;
1652         }
1653         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
1654         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
1655                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
1656                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
1657                     sbi->s_desc_size & (sbi->s_desc_size - 1)) {
1658                         printk(KERN_ERR
1659                                "EXT4-fs: unsupported descriptor size %lu\n",
1660                                sbi->s_desc_size);
1661                         goto failed_mount;
1662                 }
1663         } else
1664                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
1665         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1666         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1667         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1668         if (EXT4_INODE_SIZE(sb) == 0)
1669                 goto cantfind_ext4;
1670         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
1671         if (sbi->s_inodes_per_block == 0)
1672                 goto cantfind_ext4;
1673         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1674                                         sbi->s_inodes_per_block;
1675         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
1676         sbi->s_sbh = bh;
1677         sbi->s_mount_state = le16_to_cpu(es->s_state);
1678         sbi->s_addr_per_block_bits = log2(EXT4_ADDR_PER_BLOCK(sb));
1679         sbi->s_desc_per_block_bits = log2(EXT4_DESC_PER_BLOCK(sb));
1680         for (i=0; i < 4; i++)
1681                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1682         sbi->s_def_hash_version = es->s_def_hash_version;
1683
1684         if (sbi->s_blocks_per_group > blocksize * 8) {
1685                 printk (KERN_ERR
1686                         "EXT4-fs: #blocks per group too big: %lu\n",
1687                         sbi->s_blocks_per_group);
1688                 goto failed_mount;
1689         }
1690         if (sbi->s_frags_per_group > blocksize * 8) {
1691                 printk (KERN_ERR
1692                         "EXT4-fs: #fragments per group too big: %lu\n",
1693                         sbi->s_frags_per_group);
1694                 goto failed_mount;
1695         }
1696         if (sbi->s_inodes_per_group > blocksize * 8) {
1697                 printk (KERN_ERR
1698                         "EXT4-fs: #inodes per group too big: %lu\n",
1699                         sbi->s_inodes_per_group);
1700                 goto failed_mount;
1701         }
1702
1703         if (ext4_blocks_count(es) >
1704                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1705                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1706                         " too large to mount safely\n", sb->s_id);
1707                 if (sizeof(sector_t) < 8)
1708                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
1709                                         "enabled\n");
1710                 goto failed_mount;
1711         }
1712
1713         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
1714                 goto cantfind_ext4;
1715         blocks_count = (ext4_blocks_count(es) -
1716                         le32_to_cpu(es->s_first_data_block) +
1717                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
1718         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
1719         sbi->s_groups_count = blocks_count;
1720         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1721                    EXT4_DESC_PER_BLOCK(sb);
1722         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1723                                     GFP_KERNEL);
1724         if (sbi->s_group_desc == NULL) {
1725                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
1726                 goto failed_mount;
1727         }
1728
1729         bgl_lock_init(&sbi->s_blockgroup_lock);
1730
1731         for (i = 0; i < db_count; i++) {
1732                 block = descriptor_loc(sb, logical_sb_block, i);
1733                 sbi->s_group_desc[i] = sb_bread(sb, block);
1734                 if (!sbi->s_group_desc[i]) {
1735                         printk (KERN_ERR "EXT4-fs: "
1736                                 "can't read group descriptor %d\n", i);
1737                         db_count = i;
1738                         goto failed_mount2;
1739                 }
1740         }
1741         if (!ext4_check_descriptors (sb)) {
1742                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
1743                 goto failed_mount2;
1744         }
1745         sbi->s_gdb_count = db_count;
1746         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1747         spin_lock_init(&sbi->s_next_gen_lock);
1748
1749         percpu_counter_init(&sbi->s_freeblocks_counter,
1750                 ext4_count_free_blocks(sb));
1751         percpu_counter_init(&sbi->s_freeinodes_counter,
1752                 ext4_count_free_inodes(sb));
1753         percpu_counter_init(&sbi->s_dirs_counter,
1754                 ext4_count_dirs(sb));
1755
1756         /* per fileystem reservation list head & lock */
1757         spin_lock_init(&sbi->s_rsv_window_lock);
1758         sbi->s_rsv_window_root = RB_ROOT;
1759         /* Add a single, static dummy reservation to the start of the
1760          * reservation window list --- it gives us a placeholder for
1761          * append-at-start-of-list which makes the allocation logic
1762          * _much_ simpler. */
1763         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1764         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1765         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1766         sbi->s_rsv_window_head.rsv_goal_size = 0;
1767         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
1768
1769         /*
1770          * set up enough so that it can read an inode
1771          */
1772         sb->s_op = &ext4_sops;
1773         sb->s_export_op = &ext4_export_ops;
1774         sb->s_xattr = ext4_xattr_handlers;
1775 #ifdef CONFIG_QUOTA
1776         sb->s_qcop = &ext4_qctl_operations;
1777         sb->dq_op = &ext4_quota_operations;
1778 #endif
1779         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1780
1781         sb->s_root = NULL;
1782
1783         needs_recovery = (es->s_last_orphan != 0 ||
1784                           EXT4_HAS_INCOMPAT_FEATURE(sb,
1785                                     EXT4_FEATURE_INCOMPAT_RECOVER));
1786
1787         /*
1788          * The first inode we look at is the journal inode.  Don't try
1789          * root first: it may be modified in the journal!
1790          */
1791         if (!test_opt(sb, NOLOAD) &&
1792             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
1793                 if (ext4_load_journal(sb, es, journal_devnum))
1794                         goto failed_mount3;
1795         } else if (journal_inum) {
1796                 if (ext4_create_journal(sb, es, journal_inum))
1797                         goto failed_mount3;
1798         } else {
1799                 if (!silent)
1800                         printk (KERN_ERR
1801                                 "ext4: No journal on filesystem on %s\n",
1802                                 sb->s_id);
1803                 goto failed_mount3;
1804         }
1805
1806         /* We have now updated the journal if required, so we can
1807          * validate the data journaling mode. */
1808         switch (test_opt(sb, DATA_FLAGS)) {
1809         case 0:
1810                 /* No mode set, assume a default based on the journal
1811                  * capabilities: ORDERED_DATA if the journal can
1812                  * cope, else JOURNAL_DATA
1813                  */
1814                 if (jbd2_journal_check_available_features
1815                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
1816                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1817                 else
1818                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1819                 break;
1820
1821         case EXT4_MOUNT_ORDERED_DATA:
1822         case EXT4_MOUNT_WRITEBACK_DATA:
1823                 if (!jbd2_journal_check_available_features
1824                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
1825                         printk(KERN_ERR "EXT4-fs: Journal does not support "
1826                                "requested data journaling mode\n");
1827                         goto failed_mount4;
1828                 }
1829         default:
1830                 break;
1831         }
1832
1833         if (test_opt(sb, NOBH)) {
1834                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
1835                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
1836                                 "its supported only with writeback mode\n");
1837                         clear_opt(sbi->s_mount_opt, NOBH);
1838                 }
1839         }
1840         /*
1841          * The jbd2_journal_load will have done any necessary log recovery,
1842          * so we can safely mount the rest of the filesystem now.
1843          */
1844
1845         root = iget(sb, EXT4_ROOT_INO);
1846         sb->s_root = d_alloc_root(root);
1847         if (!sb->s_root) {
1848                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
1849                 iput(root);
1850                 goto failed_mount4;
1851         }
1852         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1853                 dput(sb->s_root);
1854                 sb->s_root = NULL;
1855                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
1856                 goto failed_mount4;
1857         }
1858
1859         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1860         /*
1861          * akpm: core read_super() calls in here with the superblock locked.
1862          * That deadlocks, because orphan cleanup needs to lock the superblock
1863          * in numerous places.  Here we just pop the lock - it's relatively
1864          * harmless, because we are now ready to accept write_super() requests,
1865          * and aviro says that's the only reason for hanging onto the
1866          * superblock lock.
1867          */
1868         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
1869         ext4_orphan_cleanup(sb, es);
1870         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
1871         if (needs_recovery)
1872                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
1873         ext4_mark_recovery_complete(sb, es);
1874         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
1875                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
1876                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
1877                 "writeback");
1878
1879         ext4_ext_init(sb);
1880
1881         lock_kernel();
1882         return 0;
1883
1884 cantfind_ext4:
1885         if (!silent)
1886                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
1887                        sb->s_id);
1888         goto failed_mount;
1889
1890 failed_mount4:
1891         jbd2_journal_destroy(sbi->s_journal);
1892 failed_mount3:
1893         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1894         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1895         percpu_counter_destroy(&sbi->s_dirs_counter);
1896 failed_mount2:
1897         for (i = 0; i < db_count; i++)
1898                 brelse(sbi->s_group_desc[i]);
1899         kfree(sbi->s_group_desc);
1900 failed_mount:
1901 #ifdef CONFIG_QUOTA
1902         for (i = 0; i < MAXQUOTAS; i++)
1903                 kfree(sbi->s_qf_names[i]);
1904 #endif
1905         ext4_blkdev_remove(sbi);
1906         brelse(bh);
1907 out_fail:
1908         sb->s_fs_info = NULL;
1909         kfree(sbi);
1910         lock_kernel();
1911         return -EINVAL;
1912 }
1913
1914 /*
1915  * Setup any per-fs journal parameters now.  We'll do this both on
1916  * initial mount, once the journal has been initialised but before we've
1917  * done any recovery; and again on any subsequent remount.
1918  */
1919 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
1920 {
1921         struct ext4_sb_info *sbi = EXT4_SB(sb);
1922
1923         if (sbi->s_commit_interval)
1924                 journal->j_commit_interval = sbi->s_commit_interval;
1925         /* We could also set up an ext4-specific default for the commit
1926          * interval here, but for now we'll just fall back to the jbd
1927          * default. */
1928
1929         spin_lock(&journal->j_state_lock);
1930         if (test_opt(sb, BARRIER))
1931                 journal->j_flags |= JBD2_BARRIER;
1932         else
1933                 journal->j_flags &= ~JBD2_BARRIER;
1934         spin_unlock(&journal->j_state_lock);
1935 }
1936
1937 static journal_t *ext4_get_journal(struct super_block *sb,
1938                                    unsigned int journal_inum)
1939 {
1940         struct inode *journal_inode;
1941         journal_t *journal;
1942
1943         /* First, test for the existence of a valid inode on disk.  Bad
1944          * things happen if we iget() an unused inode, as the subsequent
1945          * iput() will try to delete it. */
1946
1947         journal_inode = iget(sb, journal_inum);
1948         if (!journal_inode) {
1949                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
1950                 return NULL;
1951         }
1952         if (!journal_inode->i_nlink) {
1953                 make_bad_inode(journal_inode);
1954                 iput(journal_inode);
1955                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
1956                 return NULL;
1957         }
1958
1959         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1960                   journal_inode, journal_inode->i_size);
1961         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1962                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
1963                 iput(journal_inode);
1964                 return NULL;
1965         }
1966
1967         journal = jbd2_journal_init_inode(journal_inode);
1968         if (!journal) {
1969                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
1970                 iput(journal_inode);
1971                 return NULL;
1972         }
1973         journal->j_private = sb;
1974         ext4_init_journal_params(sb, journal);
1975         return journal;
1976 }
1977
1978 static journal_t *ext4_get_dev_journal(struct super_block *sb,
1979                                        dev_t j_dev)
1980 {
1981         struct buffer_head * bh;
1982         journal_t *journal;
1983         ext4_fsblk_t start;
1984         ext4_fsblk_t len;
1985         int hblock, blocksize;
1986         ext4_fsblk_t sb_block;
1987         unsigned long offset;
1988         struct ext4_super_block * es;
1989         struct block_device *bdev;
1990
1991         bdev = ext4_blkdev_get(j_dev);
1992         if (bdev == NULL)
1993                 return NULL;
1994
1995         if (bd_claim(bdev, sb)) {
1996                 printk(KERN_ERR
1997                         "EXT4: failed to claim external journal device.\n");
1998                 blkdev_put(bdev);
1999                 return NULL;
2000         }
2001
2002         blocksize = sb->s_blocksize;
2003         hblock = bdev_hardsect_size(bdev);
2004         if (blocksize < hblock) {
2005                 printk(KERN_ERR
2006                         "EXT4-fs: blocksize too small for journal device.\n");
2007                 goto out_bdev;
2008         }
2009
2010         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2011         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2012         set_blocksize(bdev, blocksize);
2013         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2014                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2015                        "external journal\n");
2016                 goto out_bdev;
2017         }
2018
2019         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2020         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2021             !(le32_to_cpu(es->s_feature_incompat) &
2022               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2023                 printk(KERN_ERR "EXT4-fs: external journal has "
2024                                         "bad superblock\n");
2025                 brelse(bh);
2026                 goto out_bdev;
2027         }
2028
2029         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2030                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2031                 brelse(bh);
2032                 goto out_bdev;
2033         }
2034
2035         len = ext4_blocks_count(es);
2036         start = sb_block + 1;
2037         brelse(bh);     /* we're done with the superblock */
2038
2039         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2040                                         start, len, blocksize);
2041         if (!journal) {
2042                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2043                 goto out_bdev;
2044         }
2045         journal->j_private = sb;
2046         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2047         wait_on_buffer(journal->j_sb_buffer);
2048         if (!buffer_uptodate(journal->j_sb_buffer)) {
2049                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2050                 goto out_journal;
2051         }
2052         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2053                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2054                                         "user (unsupported) - %d\n",
2055                         be32_to_cpu(journal->j_superblock->s_nr_users));
2056                 goto out_journal;
2057         }
2058         EXT4_SB(sb)->journal_bdev = bdev;
2059         ext4_init_journal_params(sb, journal);
2060         return journal;
2061 out_journal:
2062         jbd2_journal_destroy(journal);
2063 out_bdev:
2064         ext4_blkdev_put(bdev);
2065         return NULL;
2066 }
2067
2068 static int ext4_load_journal(struct super_block *sb,
2069                              struct ext4_super_block *es,
2070                              unsigned long journal_devnum)
2071 {
2072         journal_t *journal;
2073         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2074         dev_t journal_dev;
2075         int err = 0;
2076         int really_read_only;
2077
2078         if (journal_devnum &&
2079             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2080                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2081                         "numbers have changed\n");
2082                 journal_dev = new_decode_dev(journal_devnum);
2083         } else
2084                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2085
2086         really_read_only = bdev_read_only(sb->s_bdev);
2087
2088         /*
2089          * Are we loading a blank journal or performing recovery after a
2090          * crash?  For recovery, we need to check in advance whether we
2091          * can get read-write access to the device.
2092          */
2093
2094         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2095                 if (sb->s_flags & MS_RDONLY) {
2096                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2097                                         "required on readonly filesystem.\n");
2098                         if (really_read_only) {
2099                                 printk(KERN_ERR "EXT4-fs: write access "
2100                                         "unavailable, cannot proceed.\n");
2101                                 return -EROFS;
2102                         }
2103                         printk (KERN_INFO "EXT4-fs: write access will "
2104                                         "be enabled during recovery.\n");
2105                 }
2106         }
2107
2108         if (journal_inum && journal_dev) {
2109                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2110                        "and inode journals!\n");
2111                 return -EINVAL;
2112         }
2113
2114         if (journal_inum) {
2115                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2116                         return -EINVAL;
2117         } else {
2118                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2119                         return -EINVAL;
2120         }
2121
2122         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2123                 err = jbd2_journal_update_format(journal);
2124                 if (err)  {
2125                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2126                         jbd2_journal_destroy(journal);
2127                         return err;
2128                 }
2129         }
2130
2131         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2132                 err = jbd2_journal_wipe(journal, !really_read_only);
2133         if (!err)
2134                 err = jbd2_journal_load(journal);
2135
2136         if (err) {
2137                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2138                 jbd2_journal_destroy(journal);
2139                 return err;
2140         }
2141
2142         EXT4_SB(sb)->s_journal = journal;
2143         ext4_clear_journal_err(sb, es);
2144
2145         if (journal_devnum &&
2146             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2147                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2148                 sb->s_dirt = 1;
2149
2150                 /* Make sure we flush the recovery flag to disk. */
2151                 ext4_commit_super(sb, es, 1);
2152         }
2153
2154         return 0;
2155 }
2156
2157 static int ext4_create_journal(struct super_block * sb,
2158                                struct ext4_super_block * es,
2159                                unsigned int journal_inum)
2160 {
2161         journal_t *journal;
2162         int err;
2163
2164         if (sb->s_flags & MS_RDONLY) {
2165                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2166                                 "create journal.\n");
2167                 return -EROFS;
2168         }
2169
2170         journal = ext4_get_journal(sb, journal_inum);
2171         if (!journal)
2172                 return -EINVAL;
2173
2174         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2175                journal_inum);
2176
2177         err = jbd2_journal_create(journal);
2178         if (err) {
2179                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2180                 jbd2_journal_destroy(journal);
2181                 return -EIO;
2182         }
2183
2184         EXT4_SB(sb)->s_journal = journal;
2185
2186         ext4_update_dynamic_rev(sb);
2187         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2188         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2189
2190         es->s_journal_inum = cpu_to_le32(journal_inum);
2191         sb->s_dirt = 1;
2192
2193         /* Make sure we flush the recovery flag to disk. */
2194         ext4_commit_super(sb, es, 1);
2195
2196         return 0;
2197 }
2198
2199 static void ext4_commit_super (struct super_block * sb,
2200                                struct ext4_super_block * es,
2201                                int sync)
2202 {
2203         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2204
2205         if (!sbh)
2206                 return;
2207         es->s_wtime = cpu_to_le32(get_seconds());
2208         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2209         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2210         BUFFER_TRACE(sbh, "marking dirty");
2211         mark_buffer_dirty(sbh);
2212         if (sync)
2213                 sync_dirty_buffer(sbh);
2214 }
2215
2216
2217 /*
2218  * Have we just finished recovery?  If so, and if we are mounting (or
2219  * remounting) the filesystem readonly, then we will end up with a
2220  * consistent fs on disk.  Record that fact.
2221  */
2222 static void ext4_mark_recovery_complete(struct super_block * sb,
2223                                         struct ext4_super_block * es)
2224 {
2225         journal_t *journal = EXT4_SB(sb)->s_journal;
2226
2227         jbd2_journal_lock_updates(journal);
2228         jbd2_journal_flush(journal);
2229         lock_super(sb);
2230         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2231             sb->s_flags & MS_RDONLY) {
2232                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2233                 sb->s_dirt = 0;
2234                 ext4_commit_super(sb, es, 1);
2235         }
2236         unlock_super(sb);
2237         jbd2_journal_unlock_updates(journal);
2238 }
2239
2240 /*
2241  * If we are mounting (or read-write remounting) a filesystem whose journal
2242  * has recorded an error from a previous lifetime, move that error to the
2243  * main filesystem now.
2244  */
2245 static void ext4_clear_journal_err(struct super_block * sb,
2246                                    struct ext4_super_block * es)
2247 {
2248         journal_t *journal;
2249         int j_errno;
2250         const char *errstr;
2251
2252         journal = EXT4_SB(sb)->s_journal;
2253
2254         /*
2255          * Now check for any error status which may have been recorded in the
2256          * journal by a prior ext4_error() or ext4_abort()
2257          */
2258
2259         j_errno = jbd2_journal_errno(journal);
2260         if (j_errno) {
2261                 char nbuf[16];
2262
2263                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2264                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2265                              "from previous mount: %s", errstr);
2266                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2267                              "filesystem check.");
2268
2269                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2270                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2271                 ext4_commit_super (sb, es, 1);
2272
2273                 jbd2_journal_clear_err(journal);
2274         }
2275 }
2276
2277 /*
2278  * Force the running and committing transactions to commit,
2279  * and wait on the commit.
2280  */
2281 int ext4_force_commit(struct super_block *sb)
2282 {
2283         journal_t *journal;
2284         int ret;
2285
2286         if (sb->s_flags & MS_RDONLY)
2287                 return 0;
2288
2289         journal = EXT4_SB(sb)->s_journal;
2290         sb->s_dirt = 0;
2291         ret = ext4_journal_force_commit(journal);
2292         return ret;
2293 }
2294
2295 /*
2296  * Ext4 always journals updates to the superblock itself, so we don't
2297  * have to propagate any other updates to the superblock on disk at this
2298  * point.  Just start an async writeback to get the buffers on their way
2299  * to the disk.
2300  *
2301  * This implicitly triggers the writebehind on sync().
2302  */
2303
2304 static void ext4_write_super (struct super_block * sb)
2305 {
2306         if (mutex_trylock(&sb->s_lock) != 0)
2307                 BUG();
2308         sb->s_dirt = 0;
2309 }
2310
2311 static int ext4_sync_fs(struct super_block *sb, int wait)
2312 {
2313         tid_t target;
2314
2315         sb->s_dirt = 0;
2316         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2317                 if (wait)
2318                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2319         }
2320         return 0;
2321 }
2322
2323 /*
2324  * LVM calls this function before a (read-only) snapshot is created.  This
2325  * gives us a chance to flush the journal completely and mark the fs clean.
2326  */
2327 static void ext4_write_super_lockfs(struct super_block *sb)
2328 {
2329         sb->s_dirt = 0;
2330
2331         if (!(sb->s_flags & MS_RDONLY)) {
2332                 journal_t *journal = EXT4_SB(sb)->s_journal;
2333
2334                 /* Now we set up the journal barrier. */
2335                 jbd2_journal_lock_updates(journal);
2336                 jbd2_journal_flush(journal);
2337
2338                 /* Journal blocked and flushed, clear needs_recovery flag. */
2339                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2340                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2341         }
2342 }
2343
2344 /*
2345  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2346  * flag here, even though the filesystem is not technically dirty yet.
2347  */
2348 static void ext4_unlockfs(struct super_block *sb)
2349 {
2350         if (!(sb->s_flags & MS_RDONLY)) {
2351                 lock_super(sb);
2352                 /* Reser the needs_recovery flag before the fs is unlocked. */
2353                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2354                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2355                 unlock_super(sb);
2356                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2357         }
2358 }
2359
2360 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2361 {
2362         struct ext4_super_block * es;
2363         struct ext4_sb_info *sbi = EXT4_SB(sb);
2364         ext4_fsblk_t n_blocks_count = 0;
2365         unsigned long old_sb_flags;
2366         struct ext4_mount_options old_opts;
2367         int err;
2368 #ifdef CONFIG_QUOTA
2369         int i;
2370 #endif
2371
2372         /* Store the original options */
2373         old_sb_flags = sb->s_flags;
2374         old_opts.s_mount_opt = sbi->s_mount_opt;
2375         old_opts.s_resuid = sbi->s_resuid;
2376         old_opts.s_resgid = sbi->s_resgid;
2377         old_opts.s_commit_interval = sbi->s_commit_interval;
2378 #ifdef CONFIG_QUOTA
2379         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2380         for (i = 0; i < MAXQUOTAS; i++)
2381                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2382 #endif
2383
2384         /*
2385          * Allow the "check" option to be passed as a remount option.
2386          */
2387         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2388                 err = -EINVAL;
2389                 goto restore_opts;
2390         }
2391
2392         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2393                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2394
2395         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2396                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2397
2398         es = sbi->s_es;
2399
2400         ext4_init_journal_params(sb, sbi->s_journal);
2401
2402         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2403                 n_blocks_count > ext4_blocks_count(es)) {
2404                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2405                         err = -EROFS;
2406                         goto restore_opts;
2407                 }
2408
2409                 if (*flags & MS_RDONLY) {
2410                         /*
2411                          * First of all, the unconditional stuff we have to do
2412                          * to disable replay of the journal when we next remount
2413                          */
2414                         sb->s_flags |= MS_RDONLY;
2415
2416                         /*
2417                          * OK, test if we are remounting a valid rw partition
2418                          * readonly, and if so set the rdonly flag and then
2419                          * mark the partition as valid again.
2420                          */
2421                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2422                             (sbi->s_mount_state & EXT4_VALID_FS))
2423                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2424
2425                         /*
2426                          * We have to unlock super so that we can wait for
2427                          * transactions.
2428                          */
2429                         unlock_super(sb);
2430                         ext4_mark_recovery_complete(sb, es);
2431                         lock_super(sb);
2432                 } else {
2433                         __le32 ret;
2434                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2435                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2436                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2437                                        "remount RDWR because of unsupported "
2438                                        "optional features (%x).\n",
2439                                        sb->s_id, le32_to_cpu(ret));
2440                                 err = -EROFS;
2441                                 goto restore_opts;
2442                         }
2443
2444                         /*
2445                          * If we have an unprocessed orphan list hanging
2446                          * around from a previously readonly bdev mount,
2447                          * require a full umount/remount for now.
2448                          */
2449                         if (es->s_last_orphan) {
2450                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2451                                        "remount RDWR because of unprocessed "
2452                                        "orphan inode list.  Please "
2453                                        "umount/remount instead.\n",
2454                                        sb->s_id);
2455                                 err = -EINVAL;
2456                                 goto restore_opts;
2457                         }
2458
2459                         /*
2460                          * Mounting a RDONLY partition read-write, so reread
2461                          * and store the current valid flag.  (It may have
2462                          * been changed by e2fsck since we originally mounted
2463                          * the partition.)
2464                          */
2465                         ext4_clear_journal_err(sb, es);
2466                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2467                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2468                                 goto restore_opts;
2469                         if (!ext4_setup_super (sb, es, 0))
2470                                 sb->s_flags &= ~MS_RDONLY;
2471                 }
2472         }
2473 #ifdef CONFIG_QUOTA
2474         /* Release old quota file names */
2475         for (i = 0; i < MAXQUOTAS; i++)
2476                 if (old_opts.s_qf_names[i] &&
2477                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2478                         kfree(old_opts.s_qf_names[i]);
2479 #endif
2480         return 0;
2481 restore_opts:
2482         sb->s_flags = old_sb_flags;
2483         sbi->s_mount_opt = old_opts.s_mount_opt;
2484         sbi->s_resuid = old_opts.s_resuid;
2485         sbi->s_resgid = old_opts.s_resgid;
2486         sbi->s_commit_interval = old_opts.s_commit_interval;
2487 #ifdef CONFIG_QUOTA
2488         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2489         for (i = 0; i < MAXQUOTAS; i++) {
2490                 if (sbi->s_qf_names[i] &&
2491                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2492                         kfree(sbi->s_qf_names[i]);
2493                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2494         }
2495 #endif
2496         return err;
2497 }
2498
2499 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2500 {
2501         struct super_block *sb = dentry->d_sb;
2502         struct ext4_sb_info *sbi = EXT4_SB(sb);
2503         struct ext4_super_block *es = sbi->s_es;
2504         u64 fsid;
2505
2506         if (test_opt(sb, MINIX_DF)) {
2507                 sbi->s_overhead_last = 0;
2508         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2509                 unsigned long ngroups = sbi->s_groups_count, i;
2510                 ext4_fsblk_t overhead = 0;
2511                 smp_rmb();
2512
2513                 /*
2514                  * Compute the overhead (FS structures).  This is constant
2515                  * for a given filesystem unless the number of block groups
2516                  * changes so we cache the previous value until it does.
2517                  */
2518
2519                 /*
2520                  * All of the blocks before first_data_block are
2521                  * overhead
2522                  */
2523                 overhead = le32_to_cpu(es->s_first_data_block);
2524
2525                 /*
2526                  * Add the overhead attributed to the superblock and
2527                  * block group descriptors.  If the sparse superblocks
2528                  * feature is turned on, then not all groups have this.
2529                  */
2530                 for (i = 0; i < ngroups; i++) {
2531                         overhead += ext4_bg_has_super(sb, i) +
2532                                 ext4_bg_num_gdb(sb, i);
2533                         cond_resched();
2534                 }
2535
2536                 /*
2537                  * Every block group has an inode bitmap, a block
2538                  * bitmap, and an inode table.
2539                  */
2540                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2541                 sbi->s_overhead_last = overhead;
2542                 smp_wmb();
2543                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2544         }
2545
2546         buf->f_type = EXT4_SUPER_MAGIC;
2547         buf->f_bsize = sb->s_blocksize;
2548         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
2549         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2550         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2551         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2552         if (buf->f_bfree < ext4_r_blocks_count(es))
2553                 buf->f_bavail = 0;
2554         buf->f_files = le32_to_cpu(es->s_inodes_count);
2555         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2556         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2557         buf->f_namelen = EXT4_NAME_LEN;
2558         fsid = le64_to_cpup((void *)es->s_uuid) ^
2559                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2560         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2561         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2562         return 0;
2563 }
2564
2565 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2566  * is locked for write. Otherwise the are possible deadlocks:
2567  * Process 1                         Process 2
2568  * ext4_create()                     quota_sync()
2569  *   jbd2_journal_start()                   write_dquot()
2570  *   DQUOT_INIT()                        down(dqio_mutex)
2571  *     down(dqio_mutex)                    jbd2_journal_start()
2572  *
2573  */
2574
2575 #ifdef CONFIG_QUOTA
2576
2577 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2578 {
2579         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2580 }
2581
2582 static int ext4_dquot_initialize(struct inode *inode, int type)
2583 {
2584         handle_t *handle;
2585         int ret, err;
2586
2587         /* We may create quota structure so we need to reserve enough blocks */
2588         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
2589         if (IS_ERR(handle))
2590                 return PTR_ERR(handle);
2591         ret = dquot_initialize(inode, type);
2592         err = ext4_journal_stop(handle);
2593         if (!ret)
2594                 ret = err;
2595         return ret;
2596 }
2597
2598 static int ext4_dquot_drop(struct inode *inode)
2599 {
2600         handle_t *handle;
2601         int ret, err;
2602
2603         /* We may delete quota structure so we need to reserve enough blocks */
2604         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
2605         if (IS_ERR(handle))
2606                 return PTR_ERR(handle);
2607         ret = dquot_drop(inode);
2608         err = ext4_journal_stop(handle);
2609         if (!ret)
2610                 ret = err;
2611         return ret;
2612 }
2613
2614 static int ext4_write_dquot(struct dquot *dquot)
2615 {
2616         int ret, err;
2617         handle_t *handle;
2618         struct inode *inode;
2619
2620         inode = dquot_to_inode(dquot);
2621         handle = ext4_journal_start(inode,
2622                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2623         if (IS_ERR(handle))
2624                 return PTR_ERR(handle);
2625         ret = dquot_commit(dquot);
2626         err = ext4_journal_stop(handle);
2627         if (!ret)
2628                 ret = err;
2629         return ret;
2630 }
2631
2632 static int ext4_acquire_dquot(struct dquot *dquot)
2633 {
2634         int ret, err;
2635         handle_t *handle;
2636
2637         handle = ext4_journal_start(dquot_to_inode(dquot),
2638                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2639         if (IS_ERR(handle))
2640                 return PTR_ERR(handle);
2641         ret = dquot_acquire(dquot);
2642         err = ext4_journal_stop(handle);
2643         if (!ret)
2644                 ret = err;
2645         return ret;
2646 }
2647
2648 static int ext4_release_dquot(struct dquot *dquot)
2649 {
2650         int ret, err;
2651         handle_t *handle;
2652
2653         handle = ext4_journal_start(dquot_to_inode(dquot),
2654                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2655         if (IS_ERR(handle))
2656                 return PTR_ERR(handle);
2657         ret = dquot_release(dquot);
2658         err = ext4_journal_stop(handle);
2659         if (!ret)
2660                 ret = err;
2661         return ret;
2662 }
2663
2664 static int ext4_mark_dquot_dirty(struct dquot *dquot)
2665 {
2666         /* Are we journalling quotas? */
2667         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2668             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2669                 dquot_mark_dquot_dirty(dquot);
2670                 return ext4_write_dquot(dquot);
2671         } else {
2672                 return dquot_mark_dquot_dirty(dquot);
2673         }
2674 }
2675
2676 static int ext4_write_info(struct super_block *sb, int type)
2677 {
2678         int ret, err;
2679         handle_t *handle;
2680
2681         /* Data block + inode block */
2682         handle = ext4_journal_start(sb->s_root->d_inode, 2);
2683         if (IS_ERR(handle))
2684                 return PTR_ERR(handle);
2685         ret = dquot_commit_info(sb, type);
2686         err = ext4_journal_stop(handle);
2687         if (!ret)
2688                 ret = err;
2689         return ret;
2690 }
2691
2692 /*
2693  * Turn on quotas during mount time - we need to find
2694  * the quota file and such...
2695  */
2696 static int ext4_quota_on_mount(struct super_block *sb, int type)
2697 {
2698         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
2699                         EXT4_SB(sb)->s_jquota_fmt, type);
2700 }
2701
2702 /*
2703  * Standard function to be called on quota_on
2704  */
2705 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
2706                          char *path)
2707 {
2708         int err;
2709         struct nameidata nd;
2710
2711         if (!test_opt(sb, QUOTA))
2712                 return -EINVAL;
2713         /* Not journalling quota? */
2714         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
2715             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
2716                 return vfs_quota_on(sb, type, format_id, path);
2717         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2718         if (err)
2719                 return err;
2720         /* Quotafile not on the same filesystem? */
2721         if (nd.mnt->mnt_sb != sb) {
2722                 path_release(&nd);
2723                 return -EXDEV;
2724         }
2725         /* Quotafile not of fs root? */
2726         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2727                 printk(KERN_WARNING
2728                         "EXT4-fs: Quota file not on filesystem root. "
2729                         "Journalled quota will not work.\n");
2730         path_release(&nd);
2731         return vfs_quota_on(sb, type, format_id, path);
2732 }
2733
2734 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2735  * acquiring the locks... As quota files are never truncated and quota code
2736  * itself serializes the operations (and noone else should touch the files)
2737  * we don't have to be afraid of races */
2738 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
2739                                size_t len, loff_t off)
2740 {
2741         struct inode *inode = sb_dqopt(sb)->files[type];
2742         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2743         int err = 0;
2744         int offset = off & (sb->s_blocksize - 1);
2745         int tocopy;
2746         size_t toread;
2747         struct buffer_head *bh;
2748         loff_t i_size = i_size_read(inode);
2749
2750         if (off > i_size)
2751                 return 0;
2752         if (off+len > i_size)
2753                 len = i_size-off;
2754         toread = len;
2755         while (toread > 0) {
2756                 tocopy = sb->s_blocksize - offset < toread ?
2757                                 sb->s_blocksize - offset : toread;
2758                 bh = ext4_bread(NULL, inode, blk, 0, &err);
2759                 if (err)
2760                         return err;
2761                 if (!bh)        /* A hole? */
2762                         memset(data, 0, tocopy);
2763                 else
2764                         memcpy(data, bh->b_data+offset, tocopy);
2765                 brelse(bh);
2766                 offset = 0;
2767                 toread -= tocopy;
2768                 data += tocopy;
2769                 blk++;
2770         }
2771         return len;
2772 }
2773
2774 /* Write to quotafile (we know the transaction is already started and has
2775  * enough credits) */
2776 static ssize_t ext4_quota_write(struct super_block *sb, int type,
2777                                 const char *data, size_t len, loff_t off)
2778 {
2779         struct inode *inode = sb_dqopt(sb)->files[type];
2780         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2781         int err = 0;
2782         int offset = off & (sb->s_blocksize - 1);
2783         int tocopy;
2784         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
2785         size_t towrite = len;
2786         struct buffer_head *bh;
2787         handle_t *handle = journal_current_handle();
2788
2789         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2790         while (towrite > 0) {
2791                 tocopy = sb->s_blocksize - offset < towrite ?
2792                                 sb->s_blocksize - offset : towrite;
2793                 bh = ext4_bread(handle, inode, blk, 1, &err);
2794                 if (!bh)
2795                         goto out;
2796                 if (journal_quota) {
2797                         err = ext4_journal_get_write_access(handle, bh);
2798                         if (err) {
2799                                 brelse(bh);
2800                                 goto out;
2801                         }
2802                 }
2803                 lock_buffer(bh);
2804                 memcpy(bh->b_data+offset, data, tocopy);
2805                 flush_dcache_page(bh->b_page);
2806                 unlock_buffer(bh);
2807                 if (journal_quota)
2808                         err = ext4_journal_dirty_metadata(handle, bh);
2809                 else {
2810                         /* Always do at least ordered writes for quotas */
2811                         err = ext4_journal_dirty_data(handle, bh);
2812                         mark_buffer_dirty(bh);
2813                 }
2814                 brelse(bh);
2815                 if (err)
2816                         goto out;
2817                 offset = 0;
2818                 towrite -= tocopy;
2819                 data += tocopy;
2820                 blk++;
2821         }
2822 out:
2823         if (len == towrite)
2824                 return err;
2825         if (inode->i_size < off+len-towrite) {
2826                 i_size_write(inode, off+len-towrite);
2827                 EXT4_I(inode)->i_disksize = inode->i_size;
2828         }
2829         inode->i_version++;
2830         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2831         ext4_mark_inode_dirty(handle, inode);
2832         mutex_unlock(&inode->i_mutex);
2833         return len - towrite;
2834 }
2835
2836 #endif
2837
2838 static int ext4_get_sb(struct file_system_type *fs_type,
2839         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2840 {
2841         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
2842 }
2843
2844 static struct file_system_type ext4dev_fs_type = {
2845         .owner          = THIS_MODULE,
2846         .name           = "ext4dev",
2847         .get_sb         = ext4_get_sb,
2848         .kill_sb        = kill_block_super,
2849         .fs_flags       = FS_REQUIRES_DEV,
2850 };
2851
2852 static int __init init_ext4_fs(void)
2853 {
2854         int err = init_ext4_xattr();
2855         if (err)
2856                 return err;
2857         err = init_inodecache();
2858         if (err)
2859                 goto out1;
2860         err = register_filesystem(&ext4dev_fs_type);
2861         if (err)
2862                 goto out;
2863         return 0;
2864 out:
2865         destroy_inodecache();
2866 out1:
2867         exit_ext4_xattr();
2868         return err;
2869 }
2870
2871 static void __exit exit_ext4_fs(void)
2872 {
2873         unregister_filesystem(&ext4dev_fs_type);
2874         destroy_inodecache();
2875         exit_ext4_xattr();
2876 }
2877
2878 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2879 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
2880 MODULE_LICENSE("GPL");
2881 module_init(init_ext4_fs)
2882 module_exit(exit_ext4_fs)