fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/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/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.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 <asm/uaccess.h>
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
  43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  44                                int);
  45 static void ext3_commit_super (struct super_block * sb,
  46                                struct ext3_super_block * es,
  47                                int sync);
  48 static void ext3_mark_recovery_complete(struct super_block * sb,
  49                                         struct ext3_super_block * es);
  50 static void ext3_clear_journal_err(struct super_block * sb,
  51                                    struct ext3_super_block * es);
  52 static int ext3_sync_fs(struct super_block *sb, int wait);
  53
  54 /*
  55  * Wrappers for journal_start/end.
  56  *
  57  * The only special thing we need to do here is to make sure that all
  58  * journal_end calls result in the superblock being marked dirty, so
  59  * that sync() will call the filesystem's write_super callback if
  60  * appropriate.
  61  */
  62 handle_t *ext3_journal_start(struct inode *inode, int nblocks)
  63 {
  64         journal_t *journal;
  65
  66         if (inode->i_sb->s_flags & MS_RDONLY)
  67                 return ERR_PTR(-EROFS);
  68
  69         /* Special case here: if the journal has aborted behind our
  70          * backs (eg. EIO in the commit thread), then we still need to
  71          * take the FS itself readonly cleanly. */
  72         journal = EXT3_JOURNAL(inode);
  73         if (is_journal_aborted(journal)) {
  74                 ext3_abort(inode->i_sb, __FUNCTION__,
  75                            "Detected aborted journal");
  76                 return ERR_PTR(-EROFS);
  77         }
  78
  79         return journal_start(journal, nblocks);
  80 }
  81
  82 /*
  83  * The only special thing we need to do here is to make sure that all
  84  * journal_stop calls result in the superblock being marked dirty, so
  85  * that sync() will call the filesystem's write_super callback if
  86  * appropriate.
  87  */
  88 int __ext3_journal_stop(const char *where, handle_t *handle)
  89 {
  90         struct super_block *sb;
  91         int err;
  92         int rc;
  93
  94         sb = handle->h_transaction->t_journal->j_private;
  95         err = handle->h_err;
  96         rc = journal_stop(handle);
  97
  98         if (!err)
  99                 err = rc;
 100         if (err)
 101                 __ext3_std_error(sb, where, err);
 102         return err;
 103 }
 104
 105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 106                 struct buffer_head *bh, handle_t *handle, int err)
 107 {
 108         char nbuf[16];
 109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 110
 111         printk(KERN_ERR "%s: aborting transaction: %s in %s",
 112                caller, errstr, err_fn);
 113
 114         if (bh)
 115                 BUFFER_TRACE(bh, "abort");
 116         journal_abort_handle(handle);
 117         if (!handle->h_err)
 118                 handle->h_err = err;
 119 }
 120
 121 static char error_buf[1024];
 122
 123 /* Deal with the reporting of failure conditions on a filesystem such as
 124  * inconsistencies detected or read IO failures.
 125  *
 126  * On ext2, we can store the error state of the filesystem in the
 127  * superblock.  That is not possible on ext3, because we may have other
 128  * write ordering constraints on the superblock which prevent us from
 129  * writing it out straight away; and given that the journal is about to
 130  * be aborted, we can't rely on the current, or future, transactions to
 131  * write out the superblock safely.
 132  *
 133  * We'll just use the journal_abort() error code to record an error in
 134  * the journal instead.  On recovery, the journal will compain about
 135  * that error until we've noted it down and cleared it.
 136  */
 137
 138 static void ext3_handle_error(struct super_block *sb)
 139 {
 140         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 141
 142         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 143         es->s_state |= cpu_to_le32(EXT3_ERROR_FS);
 144
 145         if (sb->s_flags & MS_RDONLY)
 146                 return;
 147
 148         if (test_opt (sb, ERRORS_PANIC))
 149                 panic ("EXT3-fs (device %s): panic forced after error\n",
 150                        sb->s_id);
 151         if (test_opt (sb, ERRORS_RO)) {
 152                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 153                 sb->s_flags |= MS_RDONLY;
 154         } else {
 155                 journal_t *journal = EXT3_SB(sb)->s_journal;
 156
 157                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 158                 if (journal)
 159                         journal_abort(journal, -EIO);
 160         }
 161         ext3_commit_super(sb, es, 1);
 162 }
 163
 164 void ext3_error (struct super_block * sb, const char * function,
 165                  const char * fmt, ...)
 166 {
 167         va_list args;
 168
 169         va_start (args, fmt);
 170         vsprintf (error_buf, fmt, args);
 171         va_end (args);
 172
 173         printk (KERN_CRIT "EXT3-fs error (device %s): %s: %s\n",
 174                 sb->s_id, function, error_buf);
 175
 176         ext3_handle_error(sb);
 177 }
 178
 179 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
 180 {
 181         char *errstr = NULL;
 182
 183         switch (errno) {
 184         case -EIO:
 185                 errstr = "IO failure";
 186                 break;
 187         case -ENOMEM:
 188                 errstr = "Out of memory";
 189                 break;
 190         case -EROFS:
 191                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 192                         errstr = "Journal has aborted";
 193                 else
 194                         errstr = "Readonly filesystem";
 195                 break;
 196         default:
 197                 /* If the caller passed in an extra buffer for unknown
 198                  * errors, textualise them now.  Else we just return
 199                  * NULL. */
 200                 if (nbuf) {
 201                         /* Check for truncated error codes... */
 202                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 203                                 errstr = nbuf;
 204                 }
 205                 break;
 206         }
 207
 208         return errstr;
 209 }
 210
 211 /* __ext3_std_error decodes expected errors from journaling functions
 212  * automatically and invokes the appropriate error response.  */
 213
 214 void __ext3_std_error (struct super_block * sb, const char * function,
 215                        int errno)
 216 {
 217         char nbuf[16];
 218         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 219
 220         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 221                 sb->s_id, function, errstr);
 222
 223         ext3_handle_error(sb);
 224 }
 225
 226 /*
 227  * ext3_abort is a much stronger failure handler than ext3_error.  The
 228  * abort function may be used to deal with unrecoverable failures such
 229  * as journal IO errors or ENOMEM at a critical moment in log management.
 230  *
 231  * We unconditionally force the filesystem into an ABORT|READONLY state,
 232  * unless the error response on the fs has been set to panic in which
 233  * case we take the easy way out and panic immediately.
 234  */
 235
 236 void ext3_abort (struct super_block * sb, const char * function,
 237                  const char * fmt, ...)
 238 {
 239         va_list args;
 240
 241         printk (KERN_CRIT "ext3_abort called.\n");
 242
 243         va_start (args, fmt);
 244         vsprintf (error_buf, fmt, args);
 245         va_end (args);
 246
 247         if (test_opt (sb, ERRORS_PANIC))
 248                 panic ("EXT3-fs panic (device %s): %s: %s\n",
 249                        sb->s_id, function, error_buf);
 250
 251         printk (KERN_CRIT "EXT3-fs abort (device %s): %s: %s\n",
 252                 sb->s_id, function, error_buf);
 253
 254         if (sb->s_flags & MS_RDONLY)
 255                 return;
 256
 257         printk (KERN_CRIT "Remounting filesystem read-only\n");
 258         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 259         sb->s_flags |= MS_RDONLY;
 260         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 261         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 262 }
 263
 264 /* Deal with the reporting of failure conditions while running, such as
 265  * inconsistencies in operation or invalid system states.
 266  *
 267  * Use ext3_error() for cases of invalid filesystem states, as that will
 268  * record an error on disk and force a filesystem check on the next boot.
 269  */
 270 NORET_TYPE void ext3_panic (struct super_block * sb, const char * function,
 271                             const char * fmt, ...)
 272 {
 273         va_list args;
 274
 275         va_start (args, fmt);
 276         vsprintf (error_buf, fmt, args);
 277         va_end (args);
 278
 279         /* this is to prevent panic from syncing this filesystem */
 280         /* AKPM: is this sufficient? */
 281         sb->s_flags |= MS_RDONLY;
 282         panic ("EXT3-fs panic (device %s): %s: %s\n",
 283                sb->s_id, function, error_buf);
 284 }
 285
 286 void ext3_warning (struct super_block * sb, const char * function,
 287                    const char * fmt, ...)
 288 {
 289         va_list args;
 290
 291         va_start (args, fmt);
 292         vsprintf (error_buf, fmt, args);
 293         va_end (args);
 294         printk (KERN_WARNING "EXT3-fs warning (device %s): %s: %s\n",
 295                 sb->s_id, function, error_buf);
 296 }
 297
 298 void ext3_update_dynamic_rev(struct super_block *sb)
 299 {
 300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 301
 302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 303                 return;
 304
 305         ext3_warning(sb, __FUNCTION__,
 306                      "updating to rev %d because of new feature flag, "
 307                      "running e2fsck is recommended",
 308                      EXT3_DYNAMIC_REV);
 309
 310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 313         /* leave es->s_feature_*compat flags alone */
 314         /* es->s_uuid will be set by e2fsck if empty */
 315
 316         /*
 317          * The rest of the superblock fields should be zero, and if not it
 318          * means they are likely already in use, so leave them alone.  We
 319          * can leave it up to e2fsck to clean up any inconsistencies there.
 320          */
 321 }
 322
 323 /*
 324  * Open the external journal device
 325  */
 326 static struct block_device *ext3_blkdev_get(dev_t dev)
 327 {
 328         struct block_device *bdev;
 329         char b[BDEVNAME_SIZE];
 330
 331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 332         if (IS_ERR(bdev))
 333                 goto fail;
 334         return bdev;
 335
 336 fail:
 337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 338                         __bdevname(dev, b), PTR_ERR(bdev));
 339         return NULL;
 340 }
 341
 342 /*
 343  * Release the journal device
 344  */
 345 static int ext3_blkdev_put(struct block_device *bdev)
 346 {
 347         bd_release(bdev);
 348         return blkdev_put(bdev);
 349 }
 350
 351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 352 {
 353         struct block_device *bdev;
 354         int ret = -ENODEV;
 355
 356         bdev = sbi->journal_bdev;
 357         if (bdev) {
 358                 ret = ext3_blkdev_put(bdev);
 359                 sbi->journal_bdev = 0;
 360         }
 361         return ret;
 362 }
 363
 364 static inline struct inode *orphan_list_entry(struct list_head *l)
 365 {
 366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 367 }
 368
 369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 370 {
 371         struct list_head *l;
 372
 373         printk(KERN_ERR "sb orphan head is %d\n",
 374                le32_to_cpu(sbi->s_es->s_last_orphan));
 375
 376         printk(KERN_ERR "sb_info orphan list:\n");
 377         list_for_each(l, &sbi->s_orphan) {
 378                 struct inode *inode = orphan_list_entry(l);
 379                 printk(KERN_ERR "  "
 380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 381                        inode->i_sb->s_id, inode->i_ino, inode,
 382                        inode->i_mode, inode->i_nlink,
 383                        le32_to_cpu(NEXT_ORPHAN(inode)));
 384         }
 385 }
 386
 387 void ext3_put_super (struct super_block * sb)
 388 {
 389         struct ext3_sb_info *sbi = EXT3_SB(sb);
 390         struct ext3_super_block *es = sbi->s_es;
 391         int i;
 392
 393         ext3_xattr_put_super(sb);
 394         journal_destroy(sbi->s_journal);
 395         if (!(sb->s_flags & MS_RDONLY)) {
 396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 397                 es->s_state = le16_to_cpu(sbi->s_mount_state);
 398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 399                 mark_buffer_dirty(sbi->s_sbh);
 400                 ext3_commit_super(sb, es, 1);
 401         }
 402
 403         for (i = 0; i < sbi->s_gdb_count; i++)
 404                 brelse(sbi->s_group_desc[i]);
 405         kfree(sbi->s_group_desc);
 406         kfree(sbi->s_debts);
 407         brelse(sbi->s_sbh);
 408 #ifdef CONFIG_QUOTA
 409         for (i = 0; i < MAXQUOTAS; i++) {
 410                 if (sbi->s_qf_names[i])
 411                         kfree(sbi->s_qf_names[i]);
 412         }
 413 #endif
 414
 415         /* Debugging code just in case the in-memory inode orphan list
 416          * isn't empty.  The on-disk one can be non-empty if we've
 417          * detected an error and taken the fs readonly, but the
 418          * in-memory list had better be clean by this point. */
 419         if (!list_empty(&sbi->s_orphan))
 420                 dump_orphan_list(sb, sbi);
 421         J_ASSERT(list_empty(&sbi->s_orphan));
 422
 423         invalidate_bdev(sb->s_bdev, 0);
 424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 425                 /*
 426                  * Invalidate the journal device's buffers.  We don't want them
 427                  * floating about in memory - the physical journal device may
 428                  * hotswapped, and it breaks the `ro-after' testing code.
 429                  */
 430                 sync_blockdev(sbi->journal_bdev);
 431                 invalidate_bdev(sbi->journal_bdev, 0);
 432                 ext3_blkdev_remove(sbi);
 433         }
 434         sb->s_fs_info = NULL;
 435         kfree(sbi);
 436         return;
 437 }
 438
 439 static kmem_cache_t *ext3_inode_cachep;
 440
 441 /*
 442  * Called inside transaction, so use GFP_NOFS
 443  */
 444 static struct inode *ext3_alloc_inode(struct super_block *sb)
 445 {
 446         struct ext3_inode_info *ei;
 447
 448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 449         if (!ei)
 450                 return NULL;
 451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 452         ei->i_acl = EXT3_ACL_NOT_CACHED;
 453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 454 #endif
 455         ei->vfs_inode.i_version = 1;
 456         return &ei->vfs_inode;
 457 }
 458
 459 static void ext3_destroy_inode(struct inode *inode)
 460 {
 461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 462 }
 463
 464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 465 {
 466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 467
 468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 469             SLAB_CTOR_CONSTRUCTOR) {
 470                 INIT_LIST_HEAD(&ei->i_orphan);
 471 #ifdef CONFIG_EXT3_FS_XATTR
 472                 init_rwsem(&ei->xattr_sem);
 473 #endif
 474                 init_MUTEX(&ei->truncate_sem);
 475                 inode_init_once(&ei->vfs_inode);
 476         }
 477 }
 478
 479 static int init_inodecache(void)
 480 {
 481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 482                                              sizeof(struct ext3_inode_info),
 483                                              0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
 484                                              init_once, NULL);
 485         if (ext3_inode_cachep == NULL)
 486                 return -ENOMEM;
 487         return 0;
 488 }
 489
 490 static void destroy_inodecache(void)
 491 {
 492         if (kmem_cache_destroy(ext3_inode_cachep))
 493                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 494 }
 495
 496 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 497
 498 static void ext3_clear_inode(struct inode *inode)
 499 {
 500        if (EXT3_I(inode)->i_acl &&
 501            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 502                posix_acl_release(EXT3_I(inode)->i_acl);
 503                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 504        }
 505        if (EXT3_I(inode)->i_default_acl &&
 506            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 507                posix_acl_release(EXT3_I(inode)->i_default_acl);
 508                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 509        }
 510 }
 511
 512 #else
 513 # define ext3_clear_inode NULL
 514 #endif
 515
 516 #ifdef CONFIG_QUOTA
 517
 518 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 519 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 520
 521 static int ext3_dquot_initialize(struct inode *inode, int type);
 522 static int ext3_dquot_drop(struct inode *inode);
 523 static int ext3_write_dquot(struct dquot *dquot);
 524 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 525 static int ext3_write_info(struct super_block *sb, int type);
 526 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 527 static int ext3_quota_on_mount(struct super_block *sb, int type);
 528 static int ext3_quota_off_mount(struct super_block *sb, int type);
 529
 530 static struct dquot_operations ext3_quota_operations = {
 531         .initialize     = ext3_dquot_initialize,
 532         .drop           = ext3_dquot_drop,
 533         .alloc_space    = dquot_alloc_space,
 534         .alloc_inode    = dquot_alloc_inode,
 535         .free_space     = dquot_free_space,
 536         .free_inode     = dquot_free_inode,
 537         .transfer       = dquot_transfer,
 538         .write_dquot    = ext3_write_dquot,
 539         .mark_dirty     = ext3_mark_dquot_dirty,
 540         .write_info     = ext3_write_info
 541 };
 542
 543 static struct quotactl_ops ext3_qctl_operations = {
 544         .quota_on       = ext3_quota_on,
 545         .quota_off      = vfs_quota_off,
 546         .quota_sync     = vfs_quota_sync,
 547         .get_info       = vfs_get_dqinfo,
 548         .set_info       = vfs_set_dqinfo,
 549         .get_dqblk      = vfs_get_dqblk,
 550         .set_dqblk      = vfs_set_dqblk
 551 };
 552 #endif
 553
 554 static struct super_operations ext3_sops = {
 555         .alloc_inode    = ext3_alloc_inode,
 556         .destroy_inode  = ext3_destroy_inode,
 557         .read_inode     = ext3_read_inode,
 558         .write_inode    = ext3_write_inode,
 559         .dirty_inode    = ext3_dirty_inode,
 560         .put_inode      = ext3_put_inode,
 561         .delete_inode   = ext3_delete_inode,
 562         .put_super      = ext3_put_super,
 563         .write_super    = ext3_write_super,
 564         .sync_fs        = ext3_sync_fs,
 565         .write_super_lockfs = ext3_write_super_lockfs,
 566         .unlockfs       = ext3_unlockfs,
 567         .statfs         = ext3_statfs,
 568         .remount_fs     = ext3_remount,
 569         .clear_inode    = ext3_clear_inode,
 570 };
 571
 572 struct dentry *ext3_get_parent(struct dentry *child);
 573 static struct export_operations ext3_export_ops = {
 574         .get_parent = ext3_get_parent,
 575 };
 576
 577 enum {
 578         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 579         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 580         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 581         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, Opt_noload,
 582         Opt_commit, Opt_journal_update, Opt_journal_inum,
 583         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 584         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 585         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 586         Opt_tagxid, Opt_ignore, Opt_err,
 587 };
 588
 589 static match_table_t tokens = {
 590         {Opt_bsd_df, "bsddf"},
 591         {Opt_minix_df, "minixdf"},
 592         {Opt_grpid, "grpid"},
 593         {Opt_grpid, "bsdgroups"},
 594         {Opt_nogrpid, "nogrpid"},
 595         {Opt_nogrpid, "sysvgroups"},
 596         {Opt_resgid, "resgid=%u"},
 597         {Opt_resuid, "resuid=%u"},
 598         {Opt_sb, "sb=%u"},
 599         {Opt_err_cont, "errors=continue"},
 600         {Opt_err_panic, "errors=panic"},
 601         {Opt_err_ro, "errors=remount-ro"},
 602         {Opt_nouid32, "nouid32"},
 603         {Opt_nocheck, "nocheck"},
 604         {Opt_nocheck, "check=none"},
 605         {Opt_check, "check"},
 606         {Opt_debug, "debug"},
 607         {Opt_oldalloc, "oldalloc"},
 608         {Opt_orlov, "orlov"},
 609         {Opt_user_xattr, "user_xattr"},
 610         {Opt_nouser_xattr, "nouser_xattr"},
 611         {Opt_acl, "acl"},
 612         {Opt_noacl, "noacl"},
 613         {Opt_noload, "noload"},
 614         {Opt_commit, "commit=%u"},
 615         {Opt_journal_update, "journal=update"},
 616         {Opt_journal_inum, "journal=%u"},
 617         {Opt_abort, "abort"},
 618         {Opt_data_journal, "data=journal"},
 619         {Opt_data_ordered, "data=ordered"},
 620         {Opt_data_writeback, "data=writeback"},
 621         {Opt_offusrjquota, "usrjquota="},
 622         {Opt_usrjquota, "usrjquota=%s"},
 623         {Opt_offgrpjquota, "grpjquota="},
 624         {Opt_grpjquota, "grpjquota=%s"},
 625         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 626         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 627         {Opt_tagxid, "tagxid"},
 628         {Opt_ignore, "grpquota"},
 629         {Opt_ignore, "noquota"},
 630         {Opt_ignore, "quota"},
 631         {Opt_ignore, "usrquota"},
 632         {Opt_err, NULL}
 633 };
 634
 635 static unsigned long get_sb_block(void **data)
 636 {
 637         unsigned long   sb_block;
 638         char            *options = (char *) *data;
 639
 640         if (!options || strncmp(options, "sb=", 3) != 0)
 641                 return 1;       /* Default location */
 642         options += 3;
 643         sb_block = simple_strtoul(options, &options, 0);
 644         if (*options && *options != ',') {
 645                 printk("EXT3-fs: Invalid sb specification: %s\n",
 646                        (char *) *data);
 647                 return 1;
 648         }
 649         if (*options == ',')
 650                 options++;
 651         *data = (void *) options;
 652         return sb_block;
 653 }
 654
 655 static int parse_options (char * options, struct super_block *sb,
 656                           unsigned long * inum, int is_remount)
 657 {
 658         struct ext3_sb_info *sbi = EXT3_SB(sb);
 659         char * p;
 660         substring_t args[MAX_OPT_ARGS];
 661         int data_opt = 0;
 662         int option;
 663 #ifdef CONFIG_QUOTA
 664         int qtype;
 665 #endif
 666
 667         if (!options)
 668                 return 1;
 669
 670         while ((p = strsep (&options, ",")) != NULL) {
 671                 int token;
 672                 if (!*p)
 673                         continue;
 674
 675                 token = match_token(p, tokens, args);
 676                 switch (token) {
 677                 case Opt_bsd_df:
 678                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 679                         break;
 680                 case Opt_minix_df:
 681                         set_opt (sbi->s_mount_opt, MINIX_DF);
 682                         break;
 683                 case Opt_grpid:
 684                         set_opt (sbi->s_mount_opt, GRPID);
 685                         break;
 686                 case Opt_nogrpid:
 687                         clear_opt (sbi->s_mount_opt, GRPID);
 688                         break;
 689                 case Opt_resuid:
 690                         if (match_int(&args[0], &option))
 691                                 return 0;
 692                         sbi->s_resuid = option;
 693                         break;
 694                 case Opt_resgid:
 695                         if (match_int(&args[0], &option))
 696                                 return 0;
 697                         sbi->s_resgid = option;
 698                         break;
 699                 case Opt_sb:
 700                         /* handled by get_sb_block() instead of here */
 701                         /* *sb_block = match_int(&args[0]); */
 702                         break;
 703                 case Opt_err_panic:
 704                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 705                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 706                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 707                         break;
 708                 case Opt_err_ro:
 709                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 710                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 711                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 712                         break;
 713                 case Opt_err_cont:
 714                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 715                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 716                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 717                         break;
 718                 case Opt_nouid32:
 719                         set_opt (sbi->s_mount_opt, NO_UID32);
 720                         break;
 721 #ifndef CONFIG_INOXID_NONE
 722                 case Opt_tagxid:
 723                         set_opt (sbi->s_mount_opt, TAG_XID);
 724                         break;
 725 #endif
 726                 case Opt_check:
 727 #ifdef CONFIG_EXT3_CHECK
 728                         set_opt (sbi->s_mount_opt, CHECK);
 729 #else
 730                         printk(KERN_ERR
 731                                "EXT3 Check option not supported\n");
 732 #endif
 733                         break;
 734                 case Opt_nocheck:
 735                         clear_opt (sbi->s_mount_opt, CHECK);
 736                         break;
 737                 case Opt_debug:
 738                         set_opt (sbi->s_mount_opt, DEBUG);
 739                         break;
 740                 case Opt_oldalloc:
 741                         set_opt (sbi->s_mount_opt, OLDALLOC);
 742                         break;
 743                 case Opt_orlov:
 744                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 745                         break;
 746 #ifdef CONFIG_EXT3_FS_XATTR
 747                 case Opt_user_xattr:
 748                         set_opt (sbi->s_mount_opt, XATTR_USER);
 749                         break;
 750                 case Opt_nouser_xattr:
 751                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 752                         break;
 753 #else
 754                 case Opt_user_xattr:
 755                 case Opt_nouser_xattr:
 756                         printk("EXT3 (no)user_xattr options not supported\n");
 757                         break;
 758 #endif
 759 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 760                 case Opt_acl:
 761                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 762                         break;
 763                 case Opt_noacl:
 764                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 765                         break;
 766 #else
 767                 case Opt_acl:
 768                 case Opt_noacl:
 769                         printk("EXT3 (no)acl options not supported\n");
 770                         break;
 771 #endif
 772                 case Opt_journal_update:
 773                         /* @@@ FIXME */
 774                         /* Eventually we will want to be able to create
 775                            a journal file here.  For now, only allow the
 776                            user to specify an existing inode to be the
 777                            journal file. */
 778                         if (is_remount) {
 779                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 780                                        "journal on remount\n");
 781                                 return 0;
 782                         }
 783                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 784                         break;
 785                 case Opt_journal_inum:
 786                         if (is_remount) {
 787                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 788                                        "journal on remount\n");
 789                                 return 0;
 790                         }
 791                         if (match_int(&args[0], &option))
 792                                 return 0;
 793                         *inum = option;
 794                         break;
 795                 case Opt_noload:
 796                         set_opt (sbi->s_mount_opt, NOLOAD);
 797                         break;
 798                 case Opt_commit:
 799                         if (match_int(&args[0], &option))
 800                                 return 0;
 801                         if (option < 0)
 802                                 return 0;
 803                         if (option == 0)
 804                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 805                         sbi->s_commit_interval = HZ * option;
 806                         break;
 807                 case Opt_data_journal:
 808                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 809                         goto datacheck;
 810                 case Opt_data_ordered:
 811                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 812                         goto datacheck;
 813                 case Opt_data_writeback:
 814                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 815                 datacheck:
 816                         if (is_remount) {
 817                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 818                                                 != data_opt) {
 819                                         printk(KERN_ERR
 820                                                 "EXT3-fs: cannot change data "
 821                                                 "mode on remount\n");
 822                                         return 0;
 823                                 }
 824                         } else {
 825                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 826                                 sbi->s_mount_opt |= data_opt;
 827                         }
 828                         break;
 829 #ifdef CONFIG_QUOTA
 830                 case Opt_usrjquota:
 831                         qtype = USRQUOTA;
 832                         goto set_qf_name;
 833                 case Opt_grpjquota:
 834                         qtype = GRPQUOTA;
 835 set_qf_name:
 836                         if (sb_any_quota_enabled(sb)) {
 837                                 printk(KERN_ERR
 838                                         "EXT3-fs: Cannot change journalled "
 839                                         "quota options when quota turned on.\n");
 840                                 return 0;
 841                         }
 842                         if (sbi->s_qf_names[qtype]) {
 843                                 printk(KERN_ERR
 844                                         "EXT3-fs: %s quota file already "
 845                                         "specified.\n", QTYPE2NAME(qtype));
 846                                 return 0;
 847                         }
 848                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
 849                         if (!sbi->s_qf_names[qtype]) {
 850                                 printk(KERN_ERR
 851                                         "EXT3-fs: not enough memory for "
 852                                         "storing quotafile name.\n");
 853                                 return 0;
 854                         }
 855                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 856                                 printk(KERN_ERR
 857                                         "EXT3-fs: quotafile must be on "
 858                                         "filesystem root.\n");
 859                                 kfree(sbi->s_qf_names[qtype]);
 860                                 sbi->s_qf_names[qtype] = NULL;
 861                                 return 0;
 862                         }
 863                         break;
 864                 case Opt_offusrjquota:
 865                         qtype = USRQUOTA;
 866                         goto clear_qf_name;
 867                 case Opt_offgrpjquota:
 868                         qtype = GRPQUOTA;
 869 clear_qf_name:
 870                         if (sb_any_quota_enabled(sb)) {
 871                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 872                                         "journalled quota options when "
 873                                         "quota turned on.\n");
 874                                 return 0;
 875                         }
 876                         if (sbi->s_qf_names[qtype]) {
 877                                 kfree(sbi->s_qf_names[qtype]);
 878                                 sbi->s_qf_names[qtype] = NULL;
 879                         }
 880                         break;
 881                 case Opt_jqfmt_vfsold:
 882                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 883                         break;
 884                 case Opt_jqfmt_vfsv0:
 885                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 886                         break;
 887 #else
 888                 case Opt_usrjquota:
 889                 case Opt_grpjquota:
 890                 case Opt_offusrjquota:
 891                 case Opt_offgrpjquota:
 892                 case Opt_jqfmt_vfsold:
 893                 case Opt_jqfmt_vfsv0:
 894                         printk(KERN_ERR
 895                                 "EXT3-fs: journalled quota options not "
 896                                 "supported.\n");
 897                         break;
 898 #endif
 899                 case Opt_abort:
 900                         set_opt(sbi->s_mount_opt, ABORT);
 901                         break;
 902                 case Opt_ignore:
 903                         break;
 904                 default:
 905                         printk (KERN_ERR
 906                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 907                                 "or missing value\n", p);
 908                         return 0;
 909                 }
 910         }
 911 #ifdef CONFIG_QUOTA
 912         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 913             sbi->s_qf_names[GRPQUOTA])) {
 914                 printk(KERN_ERR
 915                         "EXT3-fs: journalled quota format not specified.\n");
 916                 return 0;
 917         }
 918 #endif
 919
 920         return 1;
 921 }
 922
 923 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 924                             int read_only)
 925 {
 926         struct ext3_sb_info *sbi = EXT3_SB(sb);
 927         int res = 0;
 928
 929         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 930                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 931                         "forcing read-only mode\n");
 932                 res = MS_RDONLY;
 933         }
 934         if (read_only)
 935                 return res;
 936         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 937                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 938                         "running e2fsck is recommended\n");
 939         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 940                 printk (KERN_WARNING
 941                         "EXT3-fs warning: mounting fs with errors, "
 942                         "running e2fsck is recommended\n");
 943         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 944                  le16_to_cpu(es->s_mnt_count) >=
 945                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 946                 printk (KERN_WARNING
 947                         "EXT3-fs warning: maximal mount count reached, "
 948                         "running e2fsck is recommended\n");
 949         else if (le32_to_cpu(es->s_checkinterval) &&
 950                 (le32_to_cpu(es->s_lastcheck) +
 951                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 952                 printk (KERN_WARNING
 953                         "EXT3-fs warning: checktime reached, "
 954                         "running e2fsck is recommended\n");
 955 #if 0
 956                 /* @@@ We _will_ want to clear the valid bit if we find
 957                    inconsistencies, to force a fsck at reboot.  But for
 958                    a plain journaled filesystem we can keep it set as
 959                    valid forever! :) */
 960         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 961 #endif
 962         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 963                 es->s_max_mnt_count =
 964                         (__s16) cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 965         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 966         es->s_mtime = cpu_to_le32(get_seconds());
 967         ext3_update_dynamic_rev(sb);
 968         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 969
 970         ext3_commit_super(sb, es, 1);
 971         if (test_opt(sb, DEBUG))
 972                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
 973                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
 974                         sb->s_blocksize,
 975                         sbi->s_groups_count,
 976                         EXT3_BLOCKS_PER_GROUP(sb),
 977                         EXT3_INODES_PER_GROUP(sb),
 978                         sbi->s_mount_opt);
 979
 980         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
 981         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
 982                 char b[BDEVNAME_SIZE];
 983
 984                 printk("external journal on %s\n",
 985                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
 986         } else {
 987                 printk("internal journal\n");
 988         }
 989 #ifdef CONFIG_EXT3_CHECK
 990         if (test_opt (sb, CHECK)) {
 991                 ext3_check_blocks_bitmap (sb);
 992                 ext3_check_inodes_bitmap (sb);
 993         }
 994 #endif
 995         return res;
 996 }
 997
 998 static int ext3_check_descriptors (struct super_block * sb)
 999 {
1000         struct ext3_sb_info *sbi = EXT3_SB(sb);
1001         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1002         struct ext3_group_desc * gdp = NULL;
1003         int desc_block = 0;
1004         int i;
1005
1006         ext3_debug ("Checking group descriptors");
1007
1008         for (i = 0; i < sbi->s_groups_count; i++)
1009         {
1010                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1011                         gdp = (struct ext3_group_desc *)
1012                                         sbi->s_group_desc[desc_block++]->b_data;
1013                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1014                     le32_to_cpu(gdp->bg_block_bitmap) >=
1015                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1016                 {
1017                         ext3_error (sb, "ext3_check_descriptors",
1018                                     "Block bitmap for group %d"
1019                                     " not in group (block %lu)!",
1020                                     i, (unsigned long)
1021                                         le32_to_cpu(gdp->bg_block_bitmap));
1022                         return 0;
1023                 }
1024                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1025                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1026                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1027                 {
1028                         ext3_error (sb, "ext3_check_descriptors",
1029                                     "Inode bitmap for group %d"
1030                                     " not in group (block %lu)!",
1031                                     i, (unsigned long)
1032                                         le32_to_cpu(gdp->bg_inode_bitmap));
1033                         return 0;
1034                 }
1035                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1036                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1037                     block + EXT3_BLOCKS_PER_GROUP(sb))
1038                 {
1039                         ext3_error (sb, "ext3_check_descriptors",
1040                                     "Inode table for group %d"
1041                                     " not in group (block %lu)!",
1042                                     i, (unsigned long)
1043                                         le32_to_cpu(gdp->bg_inode_table));
1044                         return 0;
1045                 }
1046                 block += EXT3_BLOCKS_PER_GROUP(sb);
1047                 gdp++;
1048         }
1049
1050         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1051         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1052         return 1;
1053 }
1054
1055
1056 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1057  * the superblock) which were deleted from all directories, but held open by
1058  * a process at the time of a crash.  We walk the list and try to delete these
1059  * inodes at recovery time (only with a read-write filesystem).
1060  *
1061  * In order to keep the orphan inode chain consistent during traversal (in
1062  * case of crash during recovery), we link each inode into the superblock
1063  * orphan list_head and handle it the same way as an inode deletion during
1064  * normal operation (which journals the operations for us).
1065  *
1066  * We only do an iget() and an iput() on each inode, which is very safe if we
1067  * accidentally point at an in-use or already deleted inode.  The worst that
1068  * can happen in this case is that we get a "bit already cleared" message from
1069  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1070  * e2fsck was run on this filesystem, and it must have already done the orphan
1071  * inode cleanup for us, so we can safely abort without any further action.
1072  */
1073 static void ext3_orphan_cleanup (struct super_block * sb,
1074                                  struct ext3_super_block * es)
1075 {
1076         unsigned int s_flags = sb->s_flags;
1077         int nr_orphans = 0, nr_truncates = 0;
1078 #ifdef CONFIG_QUOTA
1079         int i;
1080 #endif
1081         if (!es->s_last_orphan) {
1082                 jbd_debug(4, "no orphan inodes to clean up\n");
1083                 return;
1084         }
1085
1086         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1087                 if (es->s_last_orphan)
1088                         jbd_debug(1, "Errors on filesystem, "
1089                                   "clearing orphan list.\n");
1090                 es->s_last_orphan = 0;
1091                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1092                 return;
1093         }
1094
1095         if (s_flags & MS_RDONLY) {
1096                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1097                        sb->s_id);
1098                 sb->s_flags &= ~MS_RDONLY;
1099         }
1100 #ifdef CONFIG_QUOTA
1101         /* Needed for iput() to work correctly and not trash data */
1102         sb->s_flags |= MS_ACTIVE;
1103         /* Turn on quotas so that they are updated correctly */
1104         for (i = 0; i < MAXQUOTAS; i++) {
1105                 if (EXT3_SB(sb)->s_qf_names[i]) {
1106                         int ret = ext3_quota_on_mount(sb, i);
1107                         if (ret < 0)
1108                                 printk(KERN_ERR
1109                                         "EXT3-fs: Cannot turn on journalled "
1110                                         "quota: error %d\n", ret);
1111                 }
1112         }
1113 #endif
1114
1115         while (es->s_last_orphan) {
1116                 struct inode *inode;
1117
1118                 if (!(inode =
1119                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1120                         es->s_last_orphan = 0;
1121                         break;
1122                 }
1123
1124                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1125                 DQUOT_INIT(inode);
1126                 if (inode->i_nlink) {
1127                         printk(KERN_DEBUG
1128                                 "%s: truncating inode %ld to %Ld bytes\n",
1129                                 __FUNCTION__, inode->i_ino, inode->i_size);
1130                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1131                                   inode->i_ino, inode->i_size);
1132                         ext3_truncate(inode);
1133                         nr_truncates++;
1134                 } else {
1135                         printk(KERN_DEBUG
1136                                 "%s: deleting unreferenced inode %ld\n",
1137                                 __FUNCTION__, inode->i_ino);
1138                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1139                                   inode->i_ino);
1140                         nr_orphans++;
1141                 }
1142                 iput(inode);  /* The delete magic happens here! */
1143         }
1144
1145 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1146
1147         if (nr_orphans)
1148                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1149                        sb->s_id, PLURAL(nr_orphans));
1150         if (nr_truncates)
1151                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1152                        sb->s_id, PLURAL(nr_truncates));
1153 #ifdef CONFIG_QUOTA
1154         /* Turn quotas off */
1155         for (i = 0; i < MAXQUOTAS; i++) {
1156                 if (sb_dqopt(sb)->files[i])
1157                         ext3_quota_off_mount(sb, i);
1158         }
1159 #endif
1160         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1161 }
1162
1163 #define log2(n) ffz(~(n))
1164
1165 /*
1166  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1167  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1168  * We need to be 1 filesystem block less than the 2^32 sector limit.
1169  */
1170 static loff_t ext3_max_size(int bits)
1171 {
1172         loff_t res = EXT3_NDIR_BLOCKS;
1173         res += 1LL << (bits-2);
1174         res += 1LL << (2*(bits-2));
1175         res += 1LL << (3*(bits-2));
1176         res <<= bits;
1177         if (res > (512LL << 32) - (1 << bits))
1178                 res = (512LL << 32) - (1 << bits);
1179         return res;
1180 }
1181
1182 static unsigned long descriptor_loc(struct super_block *sb,
1183                                     unsigned long logic_sb_block,
1184                                     int nr)
1185 {
1186         struct ext3_sb_info *sbi = EXT3_SB(sb);
1187         unsigned long bg, first_data_block, first_meta_bg;
1188         int has_super = 0;
1189
1190         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1191         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1192
1193         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1194             nr < first_meta_bg)
1195                 return (logic_sb_block + nr + 1);
1196         bg = sbi->s_desc_per_block * nr;
1197         if (ext3_bg_has_super(sb, bg))
1198                 has_super = 1;
1199         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1200 }
1201
1202
1203 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1204 {
1205         struct buffer_head * bh;
1206         struct ext3_super_block *es = 0;
1207         struct ext3_sb_info *sbi;
1208         unsigned long block;
1209         unsigned long sb_block = get_sb_block(&data);
1210         unsigned long logic_sb_block;
1211         unsigned long offset = 0;
1212         unsigned long journal_inum = 0;
1213         unsigned long def_mount_opts;
1214         struct inode *root;
1215         int blocksize;
1216         int hblock;
1217         int db_count;
1218         int i;
1219         int needs_recovery;
1220
1221         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1222         if (!sbi)
1223                 return -ENOMEM;
1224         sb->s_fs_info = sbi;
1225         memset(sbi, 0, sizeof(*sbi));
1226         sbi->s_mount_opt = 0;
1227         sbi->s_resuid = EXT3_DEF_RESUID;
1228         sbi->s_resgid = EXT3_DEF_RESGID;
1229
1230         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1231         if (!blocksize) {
1232                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1233                 goto out_fail;
1234         }
1235
1236         /*
1237          * The ext3 superblock will not be buffer aligned for other than 1kB
1238          * block sizes.  We need to calculate the offset from buffer start.
1239          */
1240         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1241                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1242                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1243         } else {
1244                 logic_sb_block = sb_block;
1245         }
1246
1247         if (!(bh = sb_bread(sb, logic_sb_block))) {
1248                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1249                 goto out_fail;
1250         }
1251         /*
1252          * Note: s_es must be initialized as soon as possible because
1253          *       some ext3 macro-instructions depend on its value
1254          */
1255         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1256         sbi->s_es = es;
1257         sb->s_magic = le16_to_cpu(es->s_magic);
1258         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1259                 if (!silent)
1260                         printk(KERN_ERR
1261                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1262                                sb->s_id);
1263                 goto failed_mount;
1264         }
1265
1266         /* Set defaults before we parse the mount options */
1267         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1268         if (def_mount_opts & EXT3_DEFM_DEBUG)
1269                 set_opt(sbi->s_mount_opt, DEBUG);
1270         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1271                 set_opt(sbi->s_mount_opt, GRPID);
1272         if (def_mount_opts & EXT3_DEFM_UID16)
1273                 set_opt(sbi->s_mount_opt, NO_UID32);
1274         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1275                 set_opt(sbi->s_mount_opt, XATTR_USER);
1276         if (def_mount_opts & EXT3_DEFM_ACL)
1277                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1278         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1279                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1280         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1281                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1282         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1283                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1284
1285         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1286                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1287         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1288                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1289
1290         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1291         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1292
1293         if (!parse_options ((char *) data, sb, &journal_inum, 0))
1294                 goto failed_mount;
1295
1296         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1297                 sb->s_flags |= MS_TAGXID;
1298         sb->s_flags |= MS_ONE_SECOND;
1299         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1300                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1301
1302         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1303             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1304              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1305              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1306                 printk(KERN_WARNING
1307                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1308                        "running e2fsck is recommended\n");
1309         /*
1310          * Check feature flags regardless of the revision level, since we
1311          * previously didn't change the revision level when setting the flags,
1312          * so there is a chance incompat flags are set on a rev 0 filesystem.
1313          */
1314         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1315                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1316                        "unsupported optional features (%x).\n",
1317                        sb->s_id, i);
1318                 goto failed_mount;
1319         }
1320         if (!(sb->s_flags & MS_RDONLY) &&
1321             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1322                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1323                        "unsupported optional features (%x).\n",
1324                        sb->s_id, i);
1325                 goto failed_mount;
1326         }
1327         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1328
1329         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1330             blocksize > EXT3_MAX_BLOCK_SIZE) {
1331                 printk(KERN_ERR
1332                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1333                        blocksize, sb->s_id);
1334                 goto failed_mount;
1335         }
1336
1337         hblock = bdev_hardsect_size(sb->s_bdev);
1338         if (sb->s_blocksize != blocksize) {
1339                 /*
1340                  * Make sure the blocksize for the filesystem is larger
1341                  * than the hardware sectorsize for the machine.
1342                  */
1343                 if (blocksize < hblock) {
1344                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1345                                "device blocksize %d.\n", blocksize, hblock);
1346                         goto failed_mount;
1347                 }
1348
1349                 brelse (bh);
1350                 sb_set_blocksize(sb, blocksize);
1351                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1352                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1353                 bh = sb_bread(sb, logic_sb_block);
1354                 if (!bh) {
1355                         printk(KERN_ERR
1356                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1357                         goto failed_mount;
1358                 }
1359                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1360                 sbi->s_es = es;
1361                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1362                         printk (KERN_ERR
1363                                 "EXT3-fs: Magic mismatch, very weird !\n");
1364                         goto failed_mount;
1365                 }
1366         }
1367
1368         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1369
1370         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1371                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1372                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1373         } else {
1374                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1375                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1376                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1377                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1378                     (sbi->s_inode_size > blocksize)) {
1379                         printk (KERN_ERR
1380                                 "EXT3-fs: unsupported inode size: %d\n",
1381                                 sbi->s_inode_size);
1382                         goto failed_mount;
1383                 }
1384         }
1385         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1386                                    le32_to_cpu(es->s_log_frag_size);
1387         if (blocksize != sbi->s_frag_size) {
1388                 printk(KERN_ERR
1389                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1390                        sbi->s_frag_size, blocksize);
1391                 goto failed_mount;
1392         }
1393         sbi->s_frags_per_block = 1;
1394         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1395         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1396         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1397         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1398         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1399         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1400         sbi->s_sbh = bh;
1401         sbi->s_mount_state = le16_to_cpu(es->s_state);
1402         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1403         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1404         for (i=0; i < 4; i++)
1405                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1406         sbi->s_def_hash_version = es->s_def_hash_version;
1407
1408         if (sbi->s_blocks_per_group > blocksize * 8) {
1409                 printk (KERN_ERR
1410                         "EXT3-fs: #blocks per group too big: %lu\n",
1411                         sbi->s_blocks_per_group);
1412                 goto failed_mount;
1413         }
1414         if (sbi->s_frags_per_group > blocksize * 8) {
1415                 printk (KERN_ERR
1416                         "EXT3-fs: #fragments per group too big: %lu\n",
1417                         sbi->s_frags_per_group);
1418                 goto failed_mount;
1419         }
1420         if (sbi->s_inodes_per_group > blocksize * 8) {
1421                 printk (KERN_ERR
1422                         "EXT3-fs: #inodes per group too big: %lu\n",
1423                         sbi->s_inodes_per_group);
1424                 goto failed_mount;
1425         }
1426
1427         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1428                                le32_to_cpu(es->s_first_data_block) +
1429                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1430                               EXT3_BLOCKS_PER_GROUP(sb);
1431         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1432                    EXT3_DESC_PER_BLOCK(sb);
1433         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1434                                     GFP_KERNEL);
1435         if (sbi->s_group_desc == NULL) {
1436                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1437                 goto failed_mount;
1438         }
1439         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1440                         GFP_KERNEL);
1441         if (!sbi->s_debts) {
1442                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1443                 goto failed_mount2;
1444         }
1445         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1446
1447         percpu_counter_init(&sbi->s_freeblocks_counter);
1448         percpu_counter_init(&sbi->s_freeinodes_counter);
1449         percpu_counter_init(&sbi->s_dirs_counter);
1450         bgl_lock_init(&sbi->s_blockgroup_lock);
1451
1452         for (i = 0; i < db_count; i++) {
1453                 block = descriptor_loc(sb, logic_sb_block, i);
1454                 sbi->s_group_desc[i] = sb_bread(sb, block);
1455                 if (!sbi->s_group_desc[i]) {
1456                         printk (KERN_ERR "EXT3-fs: "
1457                                 "can't read group descriptor %d\n", i);
1458                         db_count = i;
1459                         goto failed_mount2;
1460                 }
1461         }
1462         if (!ext3_check_descriptors (sb)) {
1463                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1464                 goto failed_mount2;
1465         }
1466         sbi->s_gdb_count = db_count;
1467         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1468         spin_lock_init(&sbi->s_next_gen_lock);
1469         /*
1470          * set up enough so that it can read an inode
1471          */
1472         sb->s_op = &ext3_sops;
1473         sb->s_export_op = &ext3_export_ops;
1474 #ifdef CONFIG_QUOTA
1475         sb->s_qcop = &ext3_qctl_operations;
1476         sb->dq_op = &ext3_quota_operations;
1477 #endif
1478         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1479
1480         sb->s_root = 0;
1481
1482         needs_recovery = (es->s_last_orphan != 0 ||
1483                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1484                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1485
1486         /*
1487          * The first inode we look at is the journal inode.  Don't try
1488          * root first: it may be modified in the journal!
1489          */
1490         if (!test_opt(sb, NOLOAD) &&
1491             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1492                 if (ext3_load_journal(sb, es))
1493                         goto failed_mount2;
1494         } else if (journal_inum) {
1495                 if (ext3_create_journal(sb, es, journal_inum))
1496                         goto failed_mount2;
1497         } else {
1498                 if (!silent)
1499                         printk (KERN_ERR
1500                                 "ext3: No journal on filesystem on %s\n",
1501                                 sb->s_id);
1502                 goto failed_mount2;
1503         }
1504
1505         /* We have now updated the journal if required, so we can
1506          * validate the data journaling mode. */
1507         switch (test_opt(sb, DATA_FLAGS)) {
1508         case 0:
1509                 /* No mode set, assume a default based on the journal
1510                    capabilities: ORDERED_DATA if the journal can
1511                    cope, else JOURNAL_DATA */
1512                 if (journal_check_available_features
1513                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1514                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1515                 else
1516                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1517                 break;
1518
1519         case EXT3_MOUNT_ORDERED_DATA:
1520         case EXT3_MOUNT_WRITEBACK_DATA:
1521                 if (!journal_check_available_features
1522                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1523                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1524                                "requested data journaling mode\n");
1525                         goto failed_mount3;
1526                 }
1527         default:
1528                 break;
1529         }
1530
1531         /*
1532          * The journal_load will have done any necessary log recovery,
1533          * so we can safely mount the rest of the filesystem now.
1534          */
1535
1536         root = iget(sb, EXT3_ROOT_INO);
1537         sb->s_root = d_alloc_root(root);
1538         if (!sb->s_root) {
1539                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1540                 iput(root);
1541                 goto failed_mount3;
1542         }
1543         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1544                 dput(sb->s_root);
1545                 sb->s_root = NULL;
1546                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1547                 goto failed_mount3;
1548         }
1549
1550         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1551         /*
1552          * akpm: core read_super() calls in here with the superblock locked.
1553          * That deadlocks, because orphan cleanup needs to lock the superblock
1554          * in numerous places.  Here we just pop the lock - it's relatively
1555          * harmless, because we are now ready to accept write_super() requests,
1556          * and aviro says that's the only reason for hanging onto the
1557          * superblock lock.
1558          */
1559         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1560         ext3_orphan_cleanup(sb, es);
1561         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1562         if (needs_recovery)
1563                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1564         ext3_mark_recovery_complete(sb, es);
1565         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1566                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1567                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1568                 "writeback");
1569
1570         percpu_counter_mod(&sbi->s_freeblocks_counter,
1571                 ext3_count_free_blocks(sb));
1572         percpu_counter_mod(&sbi->s_freeinodes_counter,
1573                 ext3_count_free_inodes(sb));
1574         percpu_counter_mod(&sbi->s_dirs_counter,
1575                 ext3_count_dirs(sb));
1576
1577         return 0;
1578
1579 failed_mount3:
1580         journal_destroy(sbi->s_journal);
1581 failed_mount2:
1582         kfree(sbi->s_debts);
1583         for (i = 0; i < db_count; i++)
1584                 brelse(sbi->s_group_desc[i]);
1585         kfree(sbi->s_group_desc);
1586 failed_mount:
1587 #ifdef CONFIG_QUOTA
1588         for (i = 0; i < MAXQUOTAS; i++) {
1589                 if (sbi->s_qf_names[i])
1590                         kfree(sbi->s_qf_names[i]);
1591         }
1592 #endif
1593         ext3_blkdev_remove(sbi);
1594         brelse(bh);
1595 out_fail:
1596         sb->s_fs_info = NULL;
1597         kfree(sbi);
1598         return -EINVAL;
1599 }
1600
1601 /*
1602  * Setup any per-fs journal parameters now.  We'll do this both on
1603  * initial mount, once the journal has been initialised but before we've
1604  * done any recovery; and again on any subsequent remount.
1605  */
1606 static void ext3_init_journal_params(struct ext3_sb_info *sbi,
1607                                      journal_t *journal)
1608 {
1609         if (sbi->s_commit_interval)
1610                 journal->j_commit_interval = sbi->s_commit_interval;
1611         /* We could also set up an ext3-specific default for the commit
1612          * interval here, but for now we'll just fall back to the jbd
1613          * default. */
1614 }
1615
1616
1617 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1618 {
1619         struct inode *journal_inode;
1620         journal_t *journal;
1621
1622         /* First, test for the existence of a valid inode on disk.  Bad
1623          * things happen if we iget() an unused inode, as the subsequent
1624          * iput() will try to delete it. */
1625
1626         journal_inode = iget(sb, journal_inum);
1627         if (!journal_inode) {
1628                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1629                 return NULL;
1630         }
1631         if (!journal_inode->i_nlink) {
1632                 make_bad_inode(journal_inode);
1633                 iput(journal_inode);
1634                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1635                 return NULL;
1636         }
1637
1638         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1639                   journal_inode, journal_inode->i_size);
1640         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1641                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1642                 iput(journal_inode);
1643                 return NULL;
1644         }
1645
1646         journal = journal_init_inode(journal_inode);
1647         if (!journal) {
1648                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1649                 iput(journal_inode);
1650         }
1651         journal->j_private = sb;
1652         ext3_init_journal_params(EXT3_SB(sb), journal);
1653         return journal;
1654 }
1655
1656 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1657                                        dev_t j_dev)
1658 {
1659         struct buffer_head * bh;
1660         journal_t *journal;
1661         int start;
1662         int len;
1663         int hblock, blocksize;
1664         unsigned long sb_block;
1665         unsigned long offset;
1666         struct ext3_super_block * es;
1667         struct block_device *bdev;
1668
1669         bdev = ext3_blkdev_get(j_dev);
1670         if (bdev == NULL)
1671                 return NULL;
1672
1673         if (bd_claim(bdev, sb)) {
1674                 printk(KERN_ERR
1675                         "EXT3: failed to claim external journal device.\n");
1676                 blkdev_put(bdev);
1677                 return NULL;
1678         }
1679
1680         blocksize = sb->s_blocksize;
1681         hblock = bdev_hardsect_size(bdev);
1682         if (blocksize < hblock) {
1683                 printk(KERN_ERR
1684                         "EXT3-fs: blocksize too small for journal device.\n");
1685                 goto out_bdev;
1686         }
1687
1688         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1689         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1690         set_blocksize(bdev, blocksize);
1691         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1692                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1693                        "external journal\n");
1694                 goto out_bdev;
1695         }
1696
1697         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1698         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1699             !(le32_to_cpu(es->s_feature_incompat) &
1700               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1701                 printk(KERN_ERR "EXT3-fs: external journal has "
1702                                         "bad superblock\n");
1703                 brelse(bh);
1704                 goto out_bdev;
1705         }
1706
1707         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1708                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1709                 brelse(bh);
1710                 goto out_bdev;
1711         }
1712
1713         len = le32_to_cpu(es->s_blocks_count);
1714         start = sb_block + 1;
1715         brelse(bh);     /* we're done with the superblock */
1716
1717         journal = journal_init_dev(bdev, sb->s_bdev,
1718                                         start, len, blocksize);
1719         if (!journal) {
1720                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1721                 goto out_bdev;
1722         }
1723         journal->j_private = sb;
1724         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1725         wait_on_buffer(journal->j_sb_buffer);
1726         if (!buffer_uptodate(journal->j_sb_buffer)) {
1727                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1728                 goto out_journal;
1729         }
1730         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1731                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1732                                         "user (unsupported) - %d\n",
1733                         ntohl(journal->j_superblock->s_nr_users));
1734                 goto out_journal;
1735         }
1736         EXT3_SB(sb)->journal_bdev = bdev;
1737         ext3_init_journal_params(EXT3_SB(sb), journal);
1738         return journal;
1739 out_journal:
1740         journal_destroy(journal);
1741 out_bdev:
1742         ext3_blkdev_put(bdev);
1743         return NULL;
1744 }
1745
1746 static int ext3_load_journal(struct super_block * sb,
1747                              struct ext3_super_block * es)
1748 {
1749         journal_t *journal;
1750         int journal_inum = le32_to_cpu(es->s_journal_inum);
1751         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1752         int err = 0;
1753         int really_read_only;
1754
1755         really_read_only = bdev_read_only(sb->s_bdev);
1756
1757         /*
1758          * Are we loading a blank journal or performing recovery after a
1759          * crash?  For recovery, we need to check in advance whether we
1760          * can get read-write access to the device.
1761          */
1762
1763         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1764                 if (sb->s_flags & MS_RDONLY) {
1765                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1766                                         "required on readonly filesystem.\n");
1767                         if (really_read_only) {
1768                                 printk(KERN_ERR "EXT3-fs: write access "
1769                                         "unavailable, cannot proceed.\n");
1770                                 return -EROFS;
1771                         }
1772                         printk (KERN_INFO "EXT3-fs: write access will "
1773                                         "be enabled during recovery.\n");
1774                 }
1775         }
1776
1777         if (journal_inum && journal_dev) {
1778                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1779                        "and inode journals!\n");
1780                 return -EINVAL;
1781         }
1782
1783         if (journal_inum) {
1784                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1785                         return -EINVAL;
1786         } else {
1787                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1788                         return -EINVAL;
1789         }
1790
1791         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1792                 err = journal_update_format(journal);
1793                 if (err)  {
1794                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1795                         journal_destroy(journal);
1796                         return err;
1797                 }
1798         }
1799
1800         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1801                 err = journal_wipe(journal, !really_read_only);
1802         if (!err)
1803                 err = journal_load(journal);
1804
1805         if (err) {
1806                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1807                 journal_destroy(journal);
1808                 return err;
1809         }
1810
1811         EXT3_SB(sb)->s_journal = journal;
1812         ext3_clear_journal_err(sb, es);
1813         return 0;
1814 }
1815
1816 static int ext3_create_journal(struct super_block * sb,
1817                                struct ext3_super_block * es,
1818                                int journal_inum)
1819 {
1820         journal_t *journal;
1821
1822         if (sb->s_flags & MS_RDONLY) {
1823                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1824                                 "create journal.\n");
1825                 return -EROFS;
1826         }
1827
1828         if (!(journal = ext3_get_journal(sb, journal_inum)))
1829                 return -EINVAL;
1830
1831         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1832                journal_inum);
1833
1834         if (journal_create(journal)) {
1835                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1836                 journal_destroy(journal);
1837                 return -EIO;
1838         }
1839
1840         EXT3_SB(sb)->s_journal = journal;
1841
1842         ext3_update_dynamic_rev(sb);
1843         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1844         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1845
1846         es->s_journal_inum = cpu_to_le32(journal_inum);
1847         sb->s_dirt = 1;
1848
1849         /* Make sure we flush the recovery flag to disk. */
1850         ext3_commit_super(sb, es, 1);
1851
1852         return 0;
1853 }
1854
1855 static void ext3_commit_super (struct super_block * sb,
1856                                struct ext3_super_block * es,
1857                                int sync)
1858 {
1859         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1860
1861         if (!sbh)
1862                 return;
1863         es->s_wtime = cpu_to_le32(get_seconds());
1864         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1865         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1866         BUFFER_TRACE(sbh, "marking dirty");
1867         mark_buffer_dirty(sbh);
1868         if (sync)
1869                 sync_dirty_buffer(sbh);
1870 }
1871
1872
1873 /*
1874  * Have we just finished recovery?  If so, and if we are mounting (or
1875  * remounting) the filesystem readonly, then we will end up with a
1876  * consistent fs on disk.  Record that fact.
1877  */
1878 static void ext3_mark_recovery_complete(struct super_block * sb,
1879                                         struct ext3_super_block * es)
1880 {
1881         journal_flush(EXT3_SB(sb)->s_journal);
1882         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1883             sb->s_flags & MS_RDONLY) {
1884                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1885                 sb->s_dirt = 0;
1886                 ext3_commit_super(sb, es, 1);
1887         }
1888 }
1889
1890 /*
1891  * If we are mounting (or read-write remounting) a filesystem whose journal
1892  * has recorded an error from a previous lifetime, move that error to the
1893  * main filesystem now.
1894  */
1895 static void ext3_clear_journal_err(struct super_block * sb,
1896                                    struct ext3_super_block * es)
1897 {
1898         journal_t *journal;
1899         int j_errno;
1900         const char *errstr;
1901
1902         journal = EXT3_SB(sb)->s_journal;
1903
1904         /*
1905          * Now check for any error status which may have been recorded in the
1906          * journal by a prior ext3_error() or ext3_abort()
1907          */
1908
1909         j_errno = journal_errno(journal);
1910         if (j_errno) {
1911                 char nbuf[16];
1912
1913                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1914                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1915                              "from previous mount: %s", errstr);
1916                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1917                              "filesystem check.");
1918
1919                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1920                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1921                 ext3_commit_super (sb, es, 1);
1922
1923                 journal_clear_err(journal);
1924         }
1925 }
1926
1927 /*
1928  * Force the running and committing transactions to commit,
1929  * and wait on the commit.
1930  */
1931 int ext3_force_commit(struct super_block *sb)
1932 {
1933         journal_t *journal;
1934         int ret;
1935
1936         if (sb->s_flags & MS_RDONLY)
1937                 return 0;
1938
1939         journal = EXT3_SB(sb)->s_journal;
1940         sb->s_dirt = 0;
1941         ret = ext3_journal_force_commit(journal);
1942         return ret;
1943 }
1944
1945 /*
1946  * Ext3 always journals updates to the superblock itself, so we don't
1947  * have to propagate any other updates to the superblock on disk at this
1948  * point.  Just start an async writeback to get the buffers on their way
1949  * to the disk.
1950  *
1951  * This implicitly triggers the writebehind on sync().
1952  */
1953
1954 void ext3_write_super (struct super_block * sb)
1955 {
1956         if (down_trylock(&sb->s_lock) == 0)
1957                 BUG();
1958         sb->s_dirt = 0;
1959 }
1960
1961 static int ext3_sync_fs(struct super_block *sb, int wait)
1962 {
1963         tid_t target;
1964
1965         sb->s_dirt = 0;
1966         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1967                 if (wait)
1968                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1969         }
1970         return 0;
1971 }
1972
1973 /*
1974  * LVM calls this function before a (read-only) snapshot is created.  This
1975  * gives us a chance to flush the journal completely and mark the fs clean.
1976  */
1977 void ext3_write_super_lockfs(struct super_block *sb)
1978 {
1979         sb->s_dirt = 0;
1980
1981         if (!(sb->s_flags & MS_RDONLY)) {
1982                 journal_t *journal = EXT3_SB(sb)->s_journal;
1983
1984                 /* Now we set up the journal barrier. */
1985                 journal_lock_updates(journal);
1986                 journal_flush(journal);
1987
1988                 /* Journal blocked and flushed, clear needs_recovery flag. */
1989                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1990                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
1991         }
1992 }
1993
1994 /*
1995  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
1996  * flag here, even though the filesystem is not technically dirty yet.
1997  */
1998 void ext3_unlockfs(struct super_block *sb)
1999 {
2000         if (!(sb->s_flags & MS_RDONLY)) {
2001                 lock_super(sb);
2002                 /* Reser the needs_recovery flag before the fs is unlocked. */
2003                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2004                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2005                 unlock_super(sb);
2006                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2007         }
2008 }
2009
2010 int ext3_remount (struct super_block * sb, int * flags, char * data)
2011 {
2012         struct ext3_super_block * es;
2013         struct ext3_sb_info *sbi = EXT3_SB(sb);
2014         unsigned long tmp;
2015
2016         /*
2017          * Allow the "check" option to be passed as a remount option.
2018          */
2019         if (!parse_options(data, sb, &tmp, 1))
2020                 return -EINVAL;
2021
2022         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2023                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2024
2025         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2026                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2027
2028         es = sbi->s_es;
2029
2030         ext3_init_journal_params(sbi, sbi->s_journal);
2031
2032         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2033                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2034                         return -EROFS;
2035
2036                 if (*flags & MS_RDONLY) {
2037                         /*
2038                          * First of all, the unconditional stuff we have to do
2039                          * to disable replay of the journal when we next remount
2040                          */
2041                         sb->s_flags |= MS_RDONLY;
2042
2043                         /*
2044                          * OK, test if we are remounting a valid rw partition
2045                          * readonly, and if so set the rdonly flag and then
2046                          * mark the partition as valid again.
2047                          */
2048                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2049                             (sbi->s_mount_state & EXT3_VALID_FS))
2050                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2051
2052                         ext3_mark_recovery_complete(sb, es);
2053                 } else {
2054                         int ret;
2055                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2056                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2057                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2058                                        "remount RDWR because of unsupported "
2059                                        "optional features (%x).\n",
2060                                        sb->s_id, ret);
2061                                 return -EROFS;
2062                         }
2063                         /*
2064                          * Mounting a RDONLY partition read-write, so reread
2065                          * and store the current valid flag.  (It may have
2066                          * been changed by e2fsck since we originally mounted
2067                          * the partition.)
2068                          */
2069                         ext3_clear_journal_err(sb, es);
2070                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2071                         if (!ext3_setup_super (sb, es, 0))
2072                                 sb->s_flags &= ~MS_RDONLY;
2073                 }
2074         }
2075         return 0;
2076 }
2077
2078 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2079 {
2080         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2081         unsigned long overhead;
2082         int i;
2083
2084         if (test_opt (sb, MINIX_DF))
2085                 overhead = 0;
2086         else {
2087                 /*
2088                  * Compute the overhead (FS structures)
2089                  */
2090
2091                 /*
2092                  * All of the blocks before first_data_block are
2093                  * overhead
2094                  */
2095                 overhead = le32_to_cpu(es->s_first_data_block);
2096
2097                 /*
2098                  * Add the overhead attributed to the superblock and
2099                  * block group descriptors.  If the sparse superblocks
2100                  * feature is turned on, then not all groups have this.
2101                  */
2102                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2103                         overhead += ext3_bg_has_super(sb, i) +
2104                                 ext3_bg_num_gdb(sb, i);
2105
2106                 /*
2107                  * Every block group has an inode bitmap, a block
2108                  * bitmap, and an inode table.
2109                  */
2110                 overhead += (EXT3_SB(sb)->s_groups_count *
2111                              (2 + EXT3_SB(sb)->s_itb_per_group));
2112         }
2113
2114         buf->f_type = EXT3_SUPER_MAGIC;
2115         buf->f_bsize = sb->s_blocksize;
2116         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2117         buf->f_bfree = ext3_count_free_blocks (sb);
2118         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2119         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2120                 buf->f_bavail = 0;
2121         buf->f_files = le32_to_cpu(es->s_inodes_count);
2122         buf->f_ffree = ext3_count_free_inodes (sb);
2123         buf->f_namelen = EXT3_NAME_LEN;
2124         return 0;
2125 }
2126
2127 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2128  * is locked for write. Otherwise the are possible deadlocks:
2129  * Process 1                         Process 2
2130  * ext3_create()                     quota_sync()
2131  *   journal_start()                   write_dquot()
2132  *   DQUOT_INIT()                        down(dqio_sem)
2133  *     down(dqio_sem)                    journal_start()
2134  *
2135  */
2136
2137 #ifdef CONFIG_QUOTA
2138
2139 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2140 {
2141         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2142 }
2143
2144 static int ext3_dquot_initialize(struct inode *inode, int type)
2145 {
2146         handle_t *handle;
2147         int ret, err;
2148
2149         /* We may create quota structure so we need to reserve enough blocks */
2150         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2151         if (IS_ERR(handle))
2152                 return PTR_ERR(handle);
2153         ret = dquot_initialize(inode, type);
2154         err = ext3_journal_stop(handle);
2155         if (!ret)
2156                 ret = err;
2157         return ret;
2158 }
2159
2160 static int ext3_dquot_drop(struct inode *inode)
2161 {
2162         handle_t *handle;
2163         int ret, err;
2164
2165         /* We may delete quota structure so we need to reserve enough blocks */
2166         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2167         if (IS_ERR(handle))
2168                 return PTR_ERR(handle);
2169         ret = dquot_drop(inode);
2170         err = ext3_journal_stop(handle);
2171         if (!ret)
2172                 ret = err;
2173         return ret;
2174 }
2175
2176 static int ext3_write_dquot(struct dquot *dquot)
2177 {
2178         int ret, err;
2179         handle_t *handle;
2180
2181         handle = ext3_journal_start(dquot_to_inode(dquot),
2182                                         EXT3_QUOTA_TRANS_BLOCKS);
2183         if (IS_ERR(handle))
2184                 return PTR_ERR(handle);
2185         ret = dquot_commit(dquot);
2186         err = ext3_journal_stop(handle);
2187         if (!ret)
2188                 ret = err;
2189         return ret;
2190 }
2191
2192 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2193 {
2194         /* Are we journalling quotas? */
2195         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2196             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2197                 dquot_mark_dquot_dirty(dquot);
2198                 return ext3_write_dquot(dquot);
2199         } else {
2200                 return dquot_mark_dquot_dirty(dquot);
2201         }
2202 }
2203
2204 static int ext3_write_info(struct super_block *sb, int type)
2205 {
2206         int ret, err;
2207         handle_t *handle;
2208
2209         /* Data block + inode block */
2210         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2211         if (IS_ERR(handle))
2212                 return PTR_ERR(handle);
2213         ret = dquot_commit_info(sb, type);
2214         err = ext3_journal_stop(handle);
2215         if (!ret)
2216                 ret = err;
2217         return ret;
2218 }
2219
2220 /*
2221  * Turn on quotas during mount time - we need to find
2222  * the quota file and such...
2223  */
2224 static int ext3_quota_on_mount(struct super_block *sb, int type)
2225 {
2226         int err;
2227         struct dentry *dentry;
2228         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2229                              .hash = 0,
2230                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2231
2232         dentry = lookup_hash(&name, sb->s_root);
2233         if (IS_ERR(dentry))
2234                 return PTR_ERR(dentry);
2235         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2236         if (err)
2237                 dput(dentry);
2238         /* We keep the dentry reference if everything went ok - we drop it
2239          * on quota_off time */
2240         return err;
2241 }
2242
2243 /* Turn quotas off during mount time */
2244 static int ext3_quota_off_mount(struct super_block *sb, int type)
2245 {
2246         int err;
2247         struct dentry *dentry;
2248
2249         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2250         err = vfs_quota_off_mount(sb, type);
2251         /* We invalidate dentry - it has at least wrong hash... */
2252         d_invalidate(dentry);
2253         dput(dentry);
2254         return err;
2255 }
2256
2257 /*
2258  * Standard function to be called on quota_on
2259  */
2260 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2261                          char *path)
2262 {
2263         int err;
2264         struct nameidata nd;
2265
2266         /* Not journalling quota? */
2267         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2268             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2269                 return vfs_quota_on(sb, type, format_id, path);
2270         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2271         if (err)
2272                 return err;
2273         /* Quotafile not on the same filesystem? */
2274         if (nd.mnt->mnt_sb != sb)
2275                 return -EXDEV;
2276         /* Quotafile not of fs root? */
2277         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2278                 printk(KERN_WARNING
2279                         "EXT3-fs: Quota file not on filesystem root. "
2280                         "Journalled quota will not work.\n");
2281         if (!ext3_should_journal_data(nd.dentry->d_inode))
2282                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2283                         "data-journalling. Journalled quota will not work.\n");
2284         path_release(&nd);
2285         return vfs_quota_on(sb, type, format_id, path);
2286 }
2287
2288 #endif
2289
2290 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2291         int flags, const char *dev_name, void *data)
2292 {
2293         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2294 }
2295
2296 static struct file_system_type ext3_fs_type = {
2297         .owner          = THIS_MODULE,
2298         .name           = "ext3",
2299         .get_sb         = ext3_get_sb,
2300         .kill_sb        = kill_block_super,
2301         .fs_flags       = FS_REQUIRES_DEV,
2302 };
2303
2304 static int __init init_ext3_fs(void)
2305 {
2306         int err = init_ext3_xattr();
2307         if (err)
2308                 return err;
2309         err = init_inodecache();
2310         if (err)
2311                 goto out1;
2312         err = register_filesystem(&ext3_fs_type);
2313         if (err)
2314                 goto out;
2315         return 0;
2316 out:
2317         destroy_inodecache();
2318 out1:
2319         exit_ext3_xattr();
2320         return err;
2321 }
2322
2323 static void __exit exit_ext3_fs(void)
2324 {
2325         unregister_filesystem(&ext3_fs_type);
2326         destroy_inodecache();
2327         exit_ext3_xattr();
2328 }
2329
2330 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2331 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2332 MODULE_LICENSE("GPL");
2333 module_init(init_ext3_fs)
2334 module_exit(exit_ext3_fs)